~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/btrfs/relocation.c

Version: ~ [ linux-5.17-rc1 ] ~ [ linux-5.16.2 ] ~ [ linux-5.15.16 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.93 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.173 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.225 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.262 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.297 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.299 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0
  2 /*
  3  * Copyright (C) 2009 Oracle.  All rights reserved.
  4  */
  5 
  6 #include <linux/sched.h>
  7 #include <linux/pagemap.h>
  8 #include <linux/writeback.h>
  9 #include <linux/blkdev.h>
 10 #include <linux/rbtree.h>
 11 #include <linux/slab.h>
 12 #include "ctree.h"
 13 #include "disk-io.h"
 14 #include "transaction.h"
 15 #include "volumes.h"
 16 #include "locking.h"
 17 #include "btrfs_inode.h"
 18 #include "async-thread.h"
 19 #include "free-space-cache.h"
 20 #include "inode-map.h"
 21 #include "qgroup.h"
 22 #include "print-tree.h"
 23 
 24 /*
 25  * backref_node, mapping_node and tree_block start with this
 26  */
 27 struct tree_entry {
 28         struct rb_node rb_node;
 29         u64 bytenr;
 30 };
 31 
 32 /*
 33  * present a tree block in the backref cache
 34  */
 35 struct backref_node {
 36         struct rb_node rb_node;
 37         u64 bytenr;
 38 
 39         u64 new_bytenr;
 40         /* objectid of tree block owner, can be not uptodate */
 41         u64 owner;
 42         /* link to pending, changed or detached list */
 43         struct list_head list;
 44         /* list of upper level blocks reference this block */
 45         struct list_head upper;
 46         /* list of child blocks in the cache */
 47         struct list_head lower;
 48         /* NULL if this node is not tree root */
 49         struct btrfs_root *root;
 50         /* extent buffer got by COW the block */
 51         struct extent_buffer *eb;
 52         /* level of tree block */
 53         unsigned int level:8;
 54         /* is the block in non-reference counted tree */
 55         unsigned int cowonly:1;
 56         /* 1 if no child node in the cache */
 57         unsigned int lowest:1;
 58         /* is the extent buffer locked */
 59         unsigned int locked:1;
 60         /* has the block been processed */
 61         unsigned int processed:1;
 62         /* have backrefs of this block been checked */
 63         unsigned int checked:1;
 64         /*
 65          * 1 if corresponding block has been cowed but some upper
 66          * level block pointers may not point to the new location
 67          */
 68         unsigned int pending:1;
 69         /*
 70          * 1 if the backref node isn't connected to any other
 71          * backref node.
 72          */
 73         unsigned int detached:1;
 74 };
 75 
 76 /*
 77  * present a block pointer in the backref cache
 78  */
 79 struct backref_edge {
 80         struct list_head list[2];
 81         struct backref_node *node[2];
 82 };
 83 
 84 #define LOWER   0
 85 #define UPPER   1
 86 #define RELOCATION_RESERVED_NODES       256
 87 
 88 struct backref_cache {
 89         /* red black tree of all backref nodes in the cache */
 90         struct rb_root rb_root;
 91         /* for passing backref nodes to btrfs_reloc_cow_block */
 92         struct backref_node *path[BTRFS_MAX_LEVEL];
 93         /*
 94          * list of blocks that have been cowed but some block
 95          * pointers in upper level blocks may not reflect the
 96          * new location
 97          */
 98         struct list_head pending[BTRFS_MAX_LEVEL];
 99         /* list of backref nodes with no child node */
100         struct list_head leaves;
101         /* list of blocks that have been cowed in current transaction */
102         struct list_head changed;
103         /* list of detached backref node. */
104         struct list_head detached;
105 
106         u64 last_trans;
107 
108         int nr_nodes;
109         int nr_edges;
110 };
111 
112 /*
113  * map address of tree root to tree
114  */
115 struct mapping_node {
116         struct rb_node rb_node;
117         u64 bytenr;
118         void *data;
119 };
120 
121 struct mapping_tree {
122         struct rb_root rb_root;
123         spinlock_t lock;
124 };
125 
126 /*
127  * present a tree block to process
128  */
129 struct tree_block {
130         struct rb_node rb_node;
131         u64 bytenr;
132         struct btrfs_key key;
133         unsigned int level:8;
134         unsigned int key_ready:1;
135 };
136 
137 #define MAX_EXTENTS 128
138 
139 struct file_extent_cluster {
140         u64 start;
141         u64 end;
142         u64 boundary[MAX_EXTENTS];
143         unsigned int nr;
144 };
145 
146 struct reloc_control {
147         /* block group to relocate */
148         struct btrfs_block_group_cache *block_group;
149         /* extent tree */
150         struct btrfs_root *extent_root;
151         /* inode for moving data */
152         struct inode *data_inode;
153 
154         struct btrfs_block_rsv *block_rsv;
155 
156         struct backref_cache backref_cache;
157 
158         struct file_extent_cluster cluster;
159         /* tree blocks have been processed */
160         struct extent_io_tree processed_blocks;
161         /* map start of tree root to corresponding reloc tree */
162         struct mapping_tree reloc_root_tree;
163         /* list of reloc trees */
164         struct list_head reloc_roots;
165         /* list of subvolume trees that get relocated */
166         struct list_head dirty_subvol_roots;
167         /* size of metadata reservation for merging reloc trees */
168         u64 merging_rsv_size;
169         /* size of relocated tree nodes */
170         u64 nodes_relocated;
171         /* reserved size for block group relocation*/
172         u64 reserved_bytes;
173 
174         u64 search_start;
175         u64 extents_found;
176 
177         unsigned int stage:8;
178         unsigned int create_reloc_tree:1;
179         unsigned int merge_reloc_tree:1;
180         unsigned int found_file_extent:1;
181 };
182 
183 /* stages of data relocation */
184 #define MOVE_DATA_EXTENTS       0
185 #define UPDATE_DATA_PTRS        1
186 
187 static void remove_backref_node(struct backref_cache *cache,
188                                 struct backref_node *node);
189 static void __mark_block_processed(struct reloc_control *rc,
190                                    struct backref_node *node);
191 
192 static void mapping_tree_init(struct mapping_tree *tree)
193 {
194         tree->rb_root = RB_ROOT;
195         spin_lock_init(&tree->lock);
196 }
197 
198 static void backref_cache_init(struct backref_cache *cache)
199 {
200         int i;
201         cache->rb_root = RB_ROOT;
202         for (i = 0; i < BTRFS_MAX_LEVEL; i++)
203                 INIT_LIST_HEAD(&cache->pending[i]);
204         INIT_LIST_HEAD(&cache->changed);
205         INIT_LIST_HEAD(&cache->detached);
206         INIT_LIST_HEAD(&cache->leaves);
207 }
208 
209 static void backref_cache_cleanup(struct backref_cache *cache)
210 {
211         struct backref_node *node;
212         int i;
213 
214         while (!list_empty(&cache->detached)) {
215                 node = list_entry(cache->detached.next,
216                                   struct backref_node, list);
217                 remove_backref_node(cache, node);
218         }
219 
220         while (!list_empty(&cache->leaves)) {
221                 node = list_entry(cache->leaves.next,
222                                   struct backref_node, lower);
223                 remove_backref_node(cache, node);
224         }
225 
226         cache->last_trans = 0;
227 
228         for (i = 0; i < BTRFS_MAX_LEVEL; i++)
229                 ASSERT(list_empty(&cache->pending[i]));
230         ASSERT(list_empty(&cache->changed));
231         ASSERT(list_empty(&cache->detached));
232         ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
233         ASSERT(!cache->nr_nodes);
234         ASSERT(!cache->nr_edges);
235 }
236 
237 static struct backref_node *alloc_backref_node(struct backref_cache *cache)
238 {
239         struct backref_node *node;
240 
241         node = kzalloc(sizeof(*node), GFP_NOFS);
242         if (node) {
243                 INIT_LIST_HEAD(&node->list);
244                 INIT_LIST_HEAD(&node->upper);
245                 INIT_LIST_HEAD(&node->lower);
246                 RB_CLEAR_NODE(&node->rb_node);
247                 cache->nr_nodes++;
248         }
249         return node;
250 }
251 
252 static void free_backref_node(struct backref_cache *cache,
253                               struct backref_node *node)
254 {
255         if (node) {
256                 cache->nr_nodes--;
257                 kfree(node);
258         }
259 }
260 
261 static struct backref_edge *alloc_backref_edge(struct backref_cache *cache)
262 {
263         struct backref_edge *edge;
264 
265         edge = kzalloc(sizeof(*edge), GFP_NOFS);
266         if (edge)
267                 cache->nr_edges++;
268         return edge;
269 }
270 
271 static void free_backref_edge(struct backref_cache *cache,
272                               struct backref_edge *edge)
273 {
274         if (edge) {
275                 cache->nr_edges--;
276                 kfree(edge);
277         }
278 }
279 
280 static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
281                                    struct rb_node *node)
282 {
283         struct rb_node **p = &root->rb_node;
284         struct rb_node *parent = NULL;
285         struct tree_entry *entry;
286 
287         while (*p) {
288                 parent = *p;
289                 entry = rb_entry(parent, struct tree_entry, rb_node);
290 
291                 if (bytenr < entry->bytenr)
292                         p = &(*p)->rb_left;
293                 else if (bytenr > entry->bytenr)
294                         p = &(*p)->rb_right;
295                 else
296                         return parent;
297         }
298 
299         rb_link_node(node, parent, p);
300         rb_insert_color(node, root);
301         return NULL;
302 }
303 
304 static struct rb_node *tree_search(struct rb_root *root, u64 bytenr)
305 {
306         struct rb_node *n = root->rb_node;
307         struct tree_entry *entry;
308 
309         while (n) {
310                 entry = rb_entry(n, struct tree_entry, rb_node);
311 
312                 if (bytenr < entry->bytenr)
313                         n = n->rb_left;
314                 else if (bytenr > entry->bytenr)
315                         n = n->rb_right;
316                 else
317                         return n;
318         }
319         return NULL;
320 }
321 
322 static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr)
323 {
324 
325         struct btrfs_fs_info *fs_info = NULL;
326         struct backref_node *bnode = rb_entry(rb_node, struct backref_node,
327                                               rb_node);
328         if (bnode->root)
329                 fs_info = bnode->root->fs_info;
330         btrfs_panic(fs_info, errno,
331                     "Inconsistency in backref cache found at offset %llu",
332                     bytenr);
333 }
334 
335 /*
336  * walk up backref nodes until reach node presents tree root
337  */
338 static struct backref_node *walk_up_backref(struct backref_node *node,
339                                             struct backref_edge *edges[],
340                                             int *index)
341 {
342         struct backref_edge *edge;
343         int idx = *index;
344 
345         while (!list_empty(&node->upper)) {
346                 edge = list_entry(node->upper.next,
347                                   struct backref_edge, list[LOWER]);
348                 edges[idx++] = edge;
349                 node = edge->node[UPPER];
350         }
351         BUG_ON(node->detached);
352         *index = idx;
353         return node;
354 }
355 
356 /*
357  * walk down backref nodes to find start of next reference path
358  */
359 static struct backref_node *walk_down_backref(struct backref_edge *edges[],
360                                               int *index)
361 {
362         struct backref_edge *edge;
363         struct backref_node *lower;
364         int idx = *index;
365 
366         while (idx > 0) {
367                 edge = edges[idx - 1];
368                 lower = edge->node[LOWER];
369                 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
370                         idx--;
371                         continue;
372                 }
373                 edge = list_entry(edge->list[LOWER].next,
374                                   struct backref_edge, list[LOWER]);
375                 edges[idx - 1] = edge;
376                 *index = idx;
377                 return edge->node[UPPER];
378         }
379         *index = 0;
380         return NULL;
381 }
382 
383 static void unlock_node_buffer(struct backref_node *node)
384 {
385         if (node->locked) {
386                 btrfs_tree_unlock(node->eb);
387                 node->locked = 0;
388         }
389 }
390 
391 static void drop_node_buffer(struct backref_node *node)
392 {
393         if (node->eb) {
394                 unlock_node_buffer(node);
395                 free_extent_buffer(node->eb);
396                 node->eb = NULL;
397         }
398 }
399 
400 static void drop_backref_node(struct backref_cache *tree,
401                               struct backref_node *node)
402 {
403         BUG_ON(!list_empty(&node->upper));
404 
405         drop_node_buffer(node);
406         list_del(&node->list);
407         list_del(&node->lower);
408         if (!RB_EMPTY_NODE(&node->rb_node))
409                 rb_erase(&node->rb_node, &tree->rb_root);
410         free_backref_node(tree, node);
411 }
412 
413 /*
414  * remove a backref node from the backref cache
415  */
416 static void remove_backref_node(struct backref_cache *cache,
417                                 struct backref_node *node)
418 {
419         struct backref_node *upper;
420         struct backref_edge *edge;
421 
422         if (!node)
423                 return;
424 
425         BUG_ON(!node->lowest && !node->detached);
426         while (!list_empty(&node->upper)) {
427                 edge = list_entry(node->upper.next, struct backref_edge,
428                                   list[LOWER]);
429                 upper = edge->node[UPPER];
430                 list_del(&edge->list[LOWER]);
431                 list_del(&edge->list[UPPER]);
432                 free_backref_edge(cache, edge);
433 
434                 if (RB_EMPTY_NODE(&upper->rb_node)) {
435                         BUG_ON(!list_empty(&node->upper));
436                         drop_backref_node(cache, node);
437                         node = upper;
438                         node->lowest = 1;
439                         continue;
440                 }
441                 /*
442                  * add the node to leaf node list if no other
443                  * child block cached.
444                  */
445                 if (list_empty(&upper->lower)) {
446                         list_add_tail(&upper->lower, &cache->leaves);
447                         upper->lowest = 1;
448                 }
449         }
450 
451         drop_backref_node(cache, node);
452 }
453 
454 static void update_backref_node(struct backref_cache *cache,
455                                 struct backref_node *node, u64 bytenr)
456 {
457         struct rb_node *rb_node;
458         rb_erase(&node->rb_node, &cache->rb_root);
459         node->bytenr = bytenr;
460         rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
461         if (rb_node)
462                 backref_tree_panic(rb_node, -EEXIST, bytenr);
463 }
464 
465 /*
466  * update backref cache after a transaction commit
467  */
468 static int update_backref_cache(struct btrfs_trans_handle *trans,
469                                 struct backref_cache *cache)
470 {
471         struct backref_node *node;
472         int level = 0;
473 
474         if (cache->last_trans == 0) {
475                 cache->last_trans = trans->transid;
476                 return 0;
477         }
478 
479         if (cache->last_trans == trans->transid)
480                 return 0;
481 
482         /*
483          * detached nodes are used to avoid unnecessary backref
484          * lookup. transaction commit changes the extent tree.
485          * so the detached nodes are no longer useful.
486          */
487         while (!list_empty(&cache->detached)) {
488                 node = list_entry(cache->detached.next,
489                                   struct backref_node, list);
490                 remove_backref_node(cache, node);
491         }
492 
493         while (!list_empty(&cache->changed)) {
494                 node = list_entry(cache->changed.next,
495                                   struct backref_node, list);
496                 list_del_init(&node->list);
497                 BUG_ON(node->pending);
498                 update_backref_node(cache, node, node->new_bytenr);
499         }
500 
501         /*
502          * some nodes can be left in the pending list if there were
503          * errors during processing the pending nodes.
504          */
505         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
506                 list_for_each_entry(node, &cache->pending[level], list) {
507                         BUG_ON(!node->pending);
508                         if (node->bytenr == node->new_bytenr)
509                                 continue;
510                         update_backref_node(cache, node, node->new_bytenr);
511                 }
512         }
513 
514         cache->last_trans = 0;
515         return 1;
516 }
517 
518 
519 static int should_ignore_root(struct btrfs_root *root)
520 {
521         struct btrfs_root *reloc_root;
522 
523         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
524                 return 0;
525 
526         reloc_root = root->reloc_root;
527         if (!reloc_root)
528                 return 0;
529 
530         if (btrfs_root_last_snapshot(&reloc_root->root_item) ==
531             root->fs_info->running_transaction->transid - 1)
532                 return 0;
533         /*
534          * if there is reloc tree and it was created in previous
535          * transaction backref lookup can find the reloc tree,
536          * so backref node for the fs tree root is useless for
537          * relocation.
538          */
539         return 1;
540 }
541 /*
542  * find reloc tree by address of tree root
543  */
544 static struct btrfs_root *find_reloc_root(struct reloc_control *rc,
545                                           u64 bytenr)
546 {
547         struct rb_node *rb_node;
548         struct mapping_node *node;
549         struct btrfs_root *root = NULL;
550 
551         spin_lock(&rc->reloc_root_tree.lock);
552         rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr);
553         if (rb_node) {
554                 node = rb_entry(rb_node, struct mapping_node, rb_node);
555                 root = (struct btrfs_root *)node->data;
556         }
557         spin_unlock(&rc->reloc_root_tree.lock);
558         return root;
559 }
560 
561 static int is_cowonly_root(u64 root_objectid)
562 {
563         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
564             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
565             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
566             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
567             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
568             root_objectid == BTRFS_CSUM_TREE_OBJECTID ||
569             root_objectid == BTRFS_UUID_TREE_OBJECTID ||
570             root_objectid == BTRFS_QUOTA_TREE_OBJECTID ||
571             root_objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
572                 return 1;
573         return 0;
574 }
575 
576 static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info,
577                                         u64 root_objectid)
578 {
579         struct btrfs_key key;
580 
581         key.objectid = root_objectid;
582         key.type = BTRFS_ROOT_ITEM_KEY;
583         if (is_cowonly_root(root_objectid))
584                 key.offset = 0;
585         else
586                 key.offset = (u64)-1;
587 
588         return btrfs_get_fs_root(fs_info, &key, false);
589 }
590 
591 static noinline_for_stack
592 int find_inline_backref(struct extent_buffer *leaf, int slot,
593                         unsigned long *ptr, unsigned long *end)
594 {
595         struct btrfs_key key;
596         struct btrfs_extent_item *ei;
597         struct btrfs_tree_block_info *bi;
598         u32 item_size;
599 
600         btrfs_item_key_to_cpu(leaf, &key, slot);
601 
602         item_size = btrfs_item_size_nr(leaf, slot);
603         if (item_size < sizeof(*ei)) {
604                 btrfs_print_v0_err(leaf->fs_info);
605                 btrfs_handle_fs_error(leaf->fs_info, -EINVAL, NULL);
606                 return 1;
607         }
608         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
609         WARN_ON(!(btrfs_extent_flags(leaf, ei) &
610                   BTRFS_EXTENT_FLAG_TREE_BLOCK));
611 
612         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
613             item_size <= sizeof(*ei) + sizeof(*bi)) {
614                 WARN_ON(item_size < sizeof(*ei) + sizeof(*bi));
615                 return 1;
616         }
617         if (key.type == BTRFS_METADATA_ITEM_KEY &&
618             item_size <= sizeof(*ei)) {
619                 WARN_ON(item_size < sizeof(*ei));
620                 return 1;
621         }
622 
623         if (key.type == BTRFS_EXTENT_ITEM_KEY) {
624                 bi = (struct btrfs_tree_block_info *)(ei + 1);
625                 *ptr = (unsigned long)(bi + 1);
626         } else {
627                 *ptr = (unsigned long)(ei + 1);
628         }
629         *end = (unsigned long)ei + item_size;
630         return 0;
631 }
632 
633 /*
634  * build backref tree for a given tree block. root of the backref tree
635  * corresponds the tree block, leaves of the backref tree correspond
636  * roots of b-trees that reference the tree block.
637  *
638  * the basic idea of this function is check backrefs of a given block
639  * to find upper level blocks that reference the block, and then check
640  * backrefs of these upper level blocks recursively. the recursion stop
641  * when tree root is reached or backrefs for the block is cached.
642  *
643  * NOTE: if we find backrefs for a block are cached, we know backrefs
644  * for all upper level blocks that directly/indirectly reference the
645  * block are also cached.
646  */
647 static noinline_for_stack
648 struct backref_node *build_backref_tree(struct reloc_control *rc,
649                                         struct btrfs_key *node_key,
650                                         int level, u64 bytenr)
651 {
652         struct backref_cache *cache = &rc->backref_cache;
653         struct btrfs_path *path1; /* For searching extent root */
654         struct btrfs_path *path2; /* For searching parent of TREE_BLOCK_REF */
655         struct extent_buffer *eb;
656         struct btrfs_root *root;
657         struct backref_node *cur;
658         struct backref_node *upper;
659         struct backref_node *lower;
660         struct backref_node *node = NULL;
661         struct backref_node *exist = NULL;
662         struct backref_edge *edge;
663         struct rb_node *rb_node;
664         struct btrfs_key key;
665         unsigned long end;
666         unsigned long ptr;
667         LIST_HEAD(list); /* Pending edge list, upper node needs to be checked */
668         LIST_HEAD(useless);
669         int cowonly;
670         int ret;
671         int err = 0;
672         bool need_check = true;
673 
674         path1 = btrfs_alloc_path();
675         path2 = btrfs_alloc_path();
676         if (!path1 || !path2) {
677                 err = -ENOMEM;
678                 goto out;
679         }
680         path1->reada = READA_FORWARD;
681         path2->reada = READA_FORWARD;
682 
683         node = alloc_backref_node(cache);
684         if (!node) {
685                 err = -ENOMEM;
686                 goto out;
687         }
688 
689         node->bytenr = bytenr;
690         node->level = level;
691         node->lowest = 1;
692         cur = node;
693 again:
694         end = 0;
695         ptr = 0;
696         key.objectid = cur->bytenr;
697         key.type = BTRFS_METADATA_ITEM_KEY;
698         key.offset = (u64)-1;
699 
700         path1->search_commit_root = 1;
701         path1->skip_locking = 1;
702         ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1,
703                                 0, 0);
704         if (ret < 0) {
705                 err = ret;
706                 goto out;
707         }
708         ASSERT(ret);
709         ASSERT(path1->slots[0]);
710 
711         path1->slots[0]--;
712 
713         WARN_ON(cur->checked);
714         if (!list_empty(&cur->upper)) {
715                 /*
716                  * the backref was added previously when processing
717                  * backref of type BTRFS_TREE_BLOCK_REF_KEY
718                  */
719                 ASSERT(list_is_singular(&cur->upper));
720                 edge = list_entry(cur->upper.next, struct backref_edge,
721                                   list[LOWER]);
722                 ASSERT(list_empty(&edge->list[UPPER]));
723                 exist = edge->node[UPPER];
724                 /*
725                  * add the upper level block to pending list if we need
726                  * check its backrefs
727                  */
728                 if (!exist->checked)
729                         list_add_tail(&edge->list[UPPER], &list);
730         } else {
731                 exist = NULL;
732         }
733 
734         while (1) {
735                 cond_resched();
736                 eb = path1->nodes[0];
737 
738                 if (ptr >= end) {
739                         if (path1->slots[0] >= btrfs_header_nritems(eb)) {
740                                 ret = btrfs_next_leaf(rc->extent_root, path1);
741                                 if (ret < 0) {
742                                         err = ret;
743                                         goto out;
744                                 }
745                                 if (ret > 0)
746                                         break;
747                                 eb = path1->nodes[0];
748                         }
749 
750                         btrfs_item_key_to_cpu(eb, &key, path1->slots[0]);
751                         if (key.objectid != cur->bytenr) {
752                                 WARN_ON(exist);
753                                 break;
754                         }
755 
756                         if (key.type == BTRFS_EXTENT_ITEM_KEY ||
757                             key.type == BTRFS_METADATA_ITEM_KEY) {
758                                 ret = find_inline_backref(eb, path1->slots[0],
759                                                           &ptr, &end);
760                                 if (ret)
761                                         goto next;
762                         }
763                 }
764 
765                 if (ptr < end) {
766                         /* update key for inline back ref */
767                         struct btrfs_extent_inline_ref *iref;
768                         int type;
769                         iref = (struct btrfs_extent_inline_ref *)ptr;
770                         type = btrfs_get_extent_inline_ref_type(eb, iref,
771                                                         BTRFS_REF_TYPE_BLOCK);
772                         if (type == BTRFS_REF_TYPE_INVALID) {
773                                 err = -EUCLEAN;
774                                 goto out;
775                         }
776                         key.type = type;
777                         key.offset = btrfs_extent_inline_ref_offset(eb, iref);
778 
779                         WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY &&
780                                 key.type != BTRFS_SHARED_BLOCK_REF_KEY);
781                 }
782 
783                 /*
784                  * Parent node found and matches current inline ref, no need to
785                  * rebuild this node for this inline ref.
786                  */
787                 if (exist &&
788                     ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
789                       exist->owner == key.offset) ||
790                      (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
791                       exist->bytenr == key.offset))) {
792                         exist = NULL;
793                         goto next;
794                 }
795 
796                 /* SHARED_BLOCK_REF means key.offset is the parent bytenr */
797                 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
798                         if (key.objectid == key.offset) {
799                                 /*
800                                  * Only root blocks of reloc trees use backref
801                                  * pointing to itself.
802                                  */
803                                 root = find_reloc_root(rc, cur->bytenr);
804                                 ASSERT(root);
805                                 cur->root = root;
806                                 break;
807                         }
808 
809                         edge = alloc_backref_edge(cache);
810                         if (!edge) {
811                                 err = -ENOMEM;
812                                 goto out;
813                         }
814                         rb_node = tree_search(&cache->rb_root, key.offset);
815                         if (!rb_node) {
816                                 upper = alloc_backref_node(cache);
817                                 if (!upper) {
818                                         free_backref_edge(cache, edge);
819                                         err = -ENOMEM;
820                                         goto out;
821                                 }
822                                 upper->bytenr = key.offset;
823                                 upper->level = cur->level + 1;
824                                 /*
825                                  *  backrefs for the upper level block isn't
826                                  *  cached, add the block to pending list
827                                  */
828                                 list_add_tail(&edge->list[UPPER], &list);
829                         } else {
830                                 upper = rb_entry(rb_node, struct backref_node,
831                                                  rb_node);
832                                 ASSERT(upper->checked);
833                                 INIT_LIST_HEAD(&edge->list[UPPER]);
834                         }
835                         list_add_tail(&edge->list[LOWER], &cur->upper);
836                         edge->node[LOWER] = cur;
837                         edge->node[UPPER] = upper;
838 
839                         goto next;
840                 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
841                         err = -EINVAL;
842                         btrfs_print_v0_err(rc->extent_root->fs_info);
843                         btrfs_handle_fs_error(rc->extent_root->fs_info, err,
844                                               NULL);
845                         goto out;
846                 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
847                         goto next;
848                 }
849 
850                 /*
851                  * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset
852                  * means the root objectid. We need to search the tree to get
853                  * its parent bytenr.
854                  */
855                 root = read_fs_root(rc->extent_root->fs_info, key.offset);
856                 if (IS_ERR(root)) {
857                         err = PTR_ERR(root);
858                         goto out;
859                 }
860 
861                 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
862                         cur->cowonly = 1;
863 
864                 if (btrfs_root_level(&root->root_item) == cur->level) {
865                         /* tree root */
866                         ASSERT(btrfs_root_bytenr(&root->root_item) ==
867                                cur->bytenr);
868                         if (should_ignore_root(root))
869                                 list_add(&cur->list, &useless);
870                         else
871                                 cur->root = root;
872                         break;
873                 }
874 
875                 level = cur->level + 1;
876 
877                 /* Search the tree to find parent blocks referring the block. */
878                 path2->search_commit_root = 1;
879                 path2->skip_locking = 1;
880                 path2->lowest_level = level;
881                 ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0);
882                 path2->lowest_level = 0;
883                 if (ret < 0) {
884                         err = ret;
885                         goto out;
886                 }
887                 if (ret > 0 && path2->slots[level] > 0)
888                         path2->slots[level]--;
889 
890                 eb = path2->nodes[level];
891                 if (btrfs_node_blockptr(eb, path2->slots[level]) !=
892                     cur->bytenr) {
893                         btrfs_err(root->fs_info,
894         "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
895                                   cur->bytenr, level - 1,
896                                   root->root_key.objectid,
897                                   node_key->objectid, node_key->type,
898                                   node_key->offset);
899                         err = -ENOENT;
900                         goto out;
901                 }
902                 lower = cur;
903                 need_check = true;
904 
905                 /* Add all nodes and edges in the path */
906                 for (; level < BTRFS_MAX_LEVEL; level++) {
907                         if (!path2->nodes[level]) {
908                                 ASSERT(btrfs_root_bytenr(&root->root_item) ==
909                                        lower->bytenr);
910                                 if (should_ignore_root(root))
911                                         list_add(&lower->list, &useless);
912                                 else
913                                         lower->root = root;
914                                 break;
915                         }
916 
917                         edge = alloc_backref_edge(cache);
918                         if (!edge) {
919                                 err = -ENOMEM;
920                                 goto out;
921                         }
922 
923                         eb = path2->nodes[level];
924                         rb_node = tree_search(&cache->rb_root, eb->start);
925                         if (!rb_node) {
926                                 upper = alloc_backref_node(cache);
927                                 if (!upper) {
928                                         free_backref_edge(cache, edge);
929                                         err = -ENOMEM;
930                                         goto out;
931                                 }
932                                 upper->bytenr = eb->start;
933                                 upper->owner = btrfs_header_owner(eb);
934                                 upper->level = lower->level + 1;
935                                 if (!test_bit(BTRFS_ROOT_REF_COWS,
936                                               &root->state))
937                                         upper->cowonly = 1;
938 
939                                 /*
940                                  * if we know the block isn't shared
941                                  * we can void checking its backrefs.
942                                  */
943                                 if (btrfs_block_can_be_shared(root, eb))
944                                         upper->checked = 0;
945                                 else
946                                         upper->checked = 1;
947 
948                                 /*
949                                  * add the block to pending list if we
950                                  * need check its backrefs, we only do this once
951                                  * while walking up a tree as we will catch
952                                  * anything else later on.
953                                  */
954                                 if (!upper->checked && need_check) {
955                                         need_check = false;
956                                         list_add_tail(&edge->list[UPPER],
957                                                       &list);
958                                 } else {
959                                         if (upper->checked)
960                                                 need_check = true;
961                                         INIT_LIST_HEAD(&edge->list[UPPER]);
962                                 }
963                         } else {
964                                 upper = rb_entry(rb_node, struct backref_node,
965                                                  rb_node);
966                                 ASSERT(upper->checked);
967                                 INIT_LIST_HEAD(&edge->list[UPPER]);
968                                 if (!upper->owner)
969                                         upper->owner = btrfs_header_owner(eb);
970                         }
971                         list_add_tail(&edge->list[LOWER], &lower->upper);
972                         edge->node[LOWER] = lower;
973                         edge->node[UPPER] = upper;
974 
975                         if (rb_node)
976                                 break;
977                         lower = upper;
978                         upper = NULL;
979                 }
980                 btrfs_release_path(path2);
981 next:
982                 if (ptr < end) {
983                         ptr += btrfs_extent_inline_ref_size(key.type);
984                         if (ptr >= end) {
985                                 WARN_ON(ptr > end);
986                                 ptr = 0;
987                                 end = 0;
988                         }
989                 }
990                 if (ptr >= end)
991                         path1->slots[0]++;
992         }
993         btrfs_release_path(path1);
994 
995         cur->checked = 1;
996         WARN_ON(exist);
997 
998         /* the pending list isn't empty, take the first block to process */
999         if (!list_empty(&list)) {
1000                 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1001                 list_del_init(&edge->list[UPPER]);
1002                 cur = edge->node[UPPER];
1003                 goto again;
1004         }
1005 
1006         /*
1007          * everything goes well, connect backref nodes and insert backref nodes
1008          * into the cache.
1009          */
1010         ASSERT(node->checked);
1011         cowonly = node->cowonly;
1012         if (!cowonly) {
1013                 rb_node = tree_insert(&cache->rb_root, node->bytenr,
1014                                       &node->rb_node);
1015                 if (rb_node)
1016                         backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1017                 list_add_tail(&node->lower, &cache->leaves);
1018         }
1019 
1020         list_for_each_entry(edge, &node->upper, list[LOWER])
1021                 list_add_tail(&edge->list[UPPER], &list);
1022 
1023         while (!list_empty(&list)) {
1024                 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1025                 list_del_init(&edge->list[UPPER]);
1026                 upper = edge->node[UPPER];
1027                 if (upper->detached) {
1028                         list_del(&edge->list[LOWER]);
1029                         lower = edge->node[LOWER];
1030                         free_backref_edge(cache, edge);
1031                         if (list_empty(&lower->upper))
1032                                 list_add(&lower->list, &useless);
1033                         continue;
1034                 }
1035 
1036                 if (!RB_EMPTY_NODE(&upper->rb_node)) {
1037                         if (upper->lowest) {
1038                                 list_del_init(&upper->lower);
1039                                 upper->lowest = 0;
1040                         }
1041 
1042                         list_add_tail(&edge->list[UPPER], &upper->lower);
1043                         continue;
1044                 }
1045 
1046                 if (!upper->checked) {
1047                         /*
1048                          * Still want to blow up for developers since this is a
1049                          * logic bug.
1050                          */
1051                         ASSERT(0);
1052                         err = -EINVAL;
1053                         goto out;
1054                 }
1055                 if (cowonly != upper->cowonly) {
1056                         ASSERT(0);
1057                         err = -EINVAL;
1058                         goto out;
1059                 }
1060 
1061                 if (!cowonly) {
1062                         rb_node = tree_insert(&cache->rb_root, upper->bytenr,
1063                                               &upper->rb_node);
1064                         if (rb_node)
1065                                 backref_tree_panic(rb_node, -EEXIST,
1066                                                    upper->bytenr);
1067                 }
1068 
1069                 list_add_tail(&edge->list[UPPER], &upper->lower);
1070 
1071                 list_for_each_entry(edge, &upper->upper, list[LOWER])
1072                         list_add_tail(&edge->list[UPPER], &list);
1073         }
1074         /*
1075          * process useless backref nodes. backref nodes for tree leaves
1076          * are deleted from the cache. backref nodes for upper level
1077          * tree blocks are left in the cache to avoid unnecessary backref
1078          * lookup.
1079          */
1080         while (!list_empty(&useless)) {
1081                 upper = list_entry(useless.next, struct backref_node, list);
1082                 list_del_init(&upper->list);
1083                 ASSERT(list_empty(&upper->upper));
1084                 if (upper == node)
1085                         node = NULL;
1086                 if (upper->lowest) {
1087                         list_del_init(&upper->lower);
1088                         upper->lowest = 0;
1089                 }
1090                 while (!list_empty(&upper->lower)) {
1091                         edge = list_entry(upper->lower.next,
1092                                           struct backref_edge, list[UPPER]);
1093                         list_del(&edge->list[UPPER]);
1094                         list_del(&edge->list[LOWER]);
1095                         lower = edge->node[LOWER];
1096                         free_backref_edge(cache, edge);
1097 
1098                         if (list_empty(&lower->upper))
1099                                 list_add(&lower->list, &useless);
1100                 }
1101                 __mark_block_processed(rc, upper);
1102                 if (upper->level > 0) {
1103                         list_add(&upper->list, &cache->detached);
1104                         upper->detached = 1;
1105                 } else {
1106                         rb_erase(&upper->rb_node, &cache->rb_root);
1107                         free_backref_node(cache, upper);
1108                 }
1109         }
1110 out:
1111         btrfs_free_path(path1);
1112         btrfs_free_path(path2);
1113         if (err) {
1114                 while (!list_empty(&useless)) {
1115                         lower = list_entry(useless.next,
1116                                            struct backref_node, list);
1117                         list_del_init(&lower->list);
1118                 }
1119                 while (!list_empty(&list)) {
1120                         edge = list_first_entry(&list, struct backref_edge,
1121                                                 list[UPPER]);
1122                         list_del(&edge->list[UPPER]);
1123                         list_del(&edge->list[LOWER]);
1124                         lower = edge->node[LOWER];
1125                         upper = edge->node[UPPER];
1126                         free_backref_edge(cache, edge);
1127 
1128                         /*
1129                          * Lower is no longer linked to any upper backref nodes
1130                          * and isn't in the cache, we can free it ourselves.
1131                          */
1132                         if (list_empty(&lower->upper) &&
1133                             RB_EMPTY_NODE(&lower->rb_node))
1134                                 list_add(&lower->list, &useless);
1135 
1136                         if (!RB_EMPTY_NODE(&upper->rb_node))
1137                                 continue;
1138 
1139                         /* Add this guy's upper edges to the list to process */
1140                         list_for_each_entry(edge, &upper->upper, list[LOWER])
1141                                 list_add_tail(&edge->list[UPPER], &list);
1142                         if (list_empty(&upper->upper))
1143                                 list_add(&upper->list, &useless);
1144                 }
1145 
1146                 while (!list_empty(&useless)) {
1147                         lower = list_entry(useless.next,
1148                                            struct backref_node, list);
1149                         list_del_init(&lower->list);
1150                         if (lower == node)
1151                                 node = NULL;
1152                         free_backref_node(cache, lower);
1153                 }
1154 
1155                 free_backref_node(cache, node);
1156                 return ERR_PTR(err);
1157         }
1158         ASSERT(!node || !node->detached);
1159         return node;
1160 }
1161 
1162 /*
1163  * helper to add backref node for the newly created snapshot.
1164  * the backref node is created by cloning backref node that
1165  * corresponds to root of source tree
1166  */
1167 static int clone_backref_node(struct btrfs_trans_handle *trans,
1168                               struct reloc_control *rc,
1169                               struct btrfs_root *src,
1170                               struct btrfs_root *dest)
1171 {
1172         struct btrfs_root *reloc_root = src->reloc_root;
1173         struct backref_cache *cache = &rc->backref_cache;
1174         struct backref_node *node = NULL;
1175         struct backref_node *new_node;
1176         struct backref_edge *edge;
1177         struct backref_edge *new_edge;
1178         struct rb_node *rb_node;
1179 
1180         if (cache->last_trans > 0)
1181                 update_backref_cache(trans, cache);
1182 
1183         rb_node = tree_search(&cache->rb_root, src->commit_root->start);
1184         if (rb_node) {
1185                 node = rb_entry(rb_node, struct backref_node, rb_node);
1186                 if (node->detached)
1187                         node = NULL;
1188                 else
1189                         BUG_ON(node->new_bytenr != reloc_root->node->start);
1190         }
1191 
1192         if (!node) {
1193                 rb_node = tree_search(&cache->rb_root,
1194                                       reloc_root->commit_root->start);
1195                 if (rb_node) {
1196                         node = rb_entry(rb_node, struct backref_node,
1197                                         rb_node);
1198                         BUG_ON(node->detached);
1199                 }
1200         }
1201 
1202         if (!node)
1203                 return 0;
1204 
1205         new_node = alloc_backref_node(cache);
1206         if (!new_node)
1207                 return -ENOMEM;
1208 
1209         new_node->bytenr = dest->node->start;
1210         new_node->level = node->level;
1211         new_node->lowest = node->lowest;
1212         new_node->checked = 1;
1213         new_node->root = dest;
1214 
1215         if (!node->lowest) {
1216                 list_for_each_entry(edge, &node->lower, list[UPPER]) {
1217                         new_edge = alloc_backref_edge(cache);
1218                         if (!new_edge)
1219                                 goto fail;
1220 
1221                         new_edge->node[UPPER] = new_node;
1222                         new_edge->node[LOWER] = edge->node[LOWER];
1223                         list_add_tail(&new_edge->list[UPPER],
1224                                       &new_node->lower);
1225                 }
1226         } else {
1227                 list_add_tail(&new_node->lower, &cache->leaves);
1228         }
1229 
1230         rb_node = tree_insert(&cache->rb_root, new_node->bytenr,
1231                               &new_node->rb_node);
1232         if (rb_node)
1233                 backref_tree_panic(rb_node, -EEXIST, new_node->bytenr);
1234 
1235         if (!new_node->lowest) {
1236                 list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
1237                         list_add_tail(&new_edge->list[LOWER],
1238                                       &new_edge->node[LOWER]->upper);
1239                 }
1240         }
1241         return 0;
1242 fail:
1243         while (!list_empty(&new_node->lower)) {
1244                 new_edge = list_entry(new_node->lower.next,
1245                                       struct backref_edge, list[UPPER]);
1246                 list_del(&new_edge->list[UPPER]);
1247                 free_backref_edge(cache, new_edge);
1248         }
1249         free_backref_node(cache, new_node);
1250         return -ENOMEM;
1251 }
1252 
1253 /*
1254  * helper to add 'address of tree root -> reloc tree' mapping
1255  */
1256 static int __must_check __add_reloc_root(struct btrfs_root *root)
1257 {
1258         struct btrfs_fs_info *fs_info = root->fs_info;
1259         struct rb_node *rb_node;
1260         struct mapping_node *node;
1261         struct reloc_control *rc = fs_info->reloc_ctl;
1262 
1263         node = kmalloc(sizeof(*node), GFP_NOFS);
1264         if (!node)
1265                 return -ENOMEM;
1266 
1267         node->bytenr = root->node->start;
1268         node->data = root;
1269 
1270         spin_lock(&rc->reloc_root_tree.lock);
1271         rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1272                               node->bytenr, &node->rb_node);
1273         spin_unlock(&rc->reloc_root_tree.lock);
1274         if (rb_node) {
1275                 btrfs_panic(fs_info, -EEXIST,
1276                             "Duplicate root found for start=%llu while inserting into relocation tree",
1277                             node->bytenr);
1278         }
1279 
1280         list_add_tail(&root->root_list, &rc->reloc_roots);
1281         return 0;
1282 }
1283 
1284 /*
1285  * helper to delete the 'address of tree root -> reloc tree'
1286  * mapping
1287  */
1288 static void __del_reloc_root(struct btrfs_root *root)
1289 {
1290         struct btrfs_fs_info *fs_info = root->fs_info;
1291         struct rb_node *rb_node;
1292         struct mapping_node *node = NULL;
1293         struct reloc_control *rc = fs_info->reloc_ctl;
1294 
1295         if (rc && root->node) {
1296                 spin_lock(&rc->reloc_root_tree.lock);
1297                 rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1298                                       root->node->start);
1299                 if (rb_node) {
1300                         node = rb_entry(rb_node, struct mapping_node, rb_node);
1301                         rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1302                 }
1303                 spin_unlock(&rc->reloc_root_tree.lock);
1304                 if (!node)
1305                         return;
1306                 BUG_ON((struct btrfs_root *)node->data != root);
1307         }
1308 
1309         spin_lock(&fs_info->trans_lock);
1310         list_del_init(&root->root_list);
1311         spin_unlock(&fs_info->trans_lock);
1312         kfree(node);
1313 }
1314 
1315 /*
1316  * helper to update the 'address of tree root -> reloc tree'
1317  * mapping
1318  */
1319 static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr)
1320 {
1321         struct btrfs_fs_info *fs_info = root->fs_info;
1322         struct rb_node *rb_node;
1323         struct mapping_node *node = NULL;
1324         struct reloc_control *rc = fs_info->reloc_ctl;
1325 
1326         spin_lock(&rc->reloc_root_tree.lock);
1327         rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1328                               root->node->start);
1329         if (rb_node) {
1330                 node = rb_entry(rb_node, struct mapping_node, rb_node);
1331                 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1332         }
1333         spin_unlock(&rc->reloc_root_tree.lock);
1334 
1335         if (!node)
1336                 return 0;
1337         BUG_ON((struct btrfs_root *)node->data != root);
1338 
1339         spin_lock(&rc->reloc_root_tree.lock);
1340         node->bytenr = new_bytenr;
1341         rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1342                               node->bytenr, &node->rb_node);
1343         spin_unlock(&rc->reloc_root_tree.lock);
1344         if (rb_node)
1345                 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1346         return 0;
1347 }
1348 
1349 static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
1350                                         struct btrfs_root *root, u64 objectid)
1351 {
1352         struct btrfs_fs_info *fs_info = root->fs_info;
1353         struct btrfs_root *reloc_root;
1354         struct extent_buffer *eb;
1355         struct btrfs_root_item *root_item;
1356         struct btrfs_key root_key;
1357         int ret;
1358 
1359         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
1360         BUG_ON(!root_item);
1361 
1362         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
1363         root_key.type = BTRFS_ROOT_ITEM_KEY;
1364         root_key.offset = objectid;
1365 
1366         if (root->root_key.objectid == objectid) {
1367                 u64 commit_root_gen;
1368 
1369                 /* called by btrfs_init_reloc_root */
1370                 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
1371                                       BTRFS_TREE_RELOC_OBJECTID);
1372                 BUG_ON(ret);
1373                 /*
1374                  * Set the last_snapshot field to the generation of the commit
1375                  * root - like this ctree.c:btrfs_block_can_be_shared() behaves
1376                  * correctly (returns true) when the relocation root is created
1377                  * either inside the critical section of a transaction commit
1378                  * (through transaction.c:qgroup_account_snapshot()) and when
1379                  * it's created before the transaction commit is started.
1380                  */
1381                 commit_root_gen = btrfs_header_generation(root->commit_root);
1382                 btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
1383         } else {
1384                 /*
1385                  * called by btrfs_reloc_post_snapshot_hook.
1386                  * the source tree is a reloc tree, all tree blocks
1387                  * modified after it was created have RELOC flag
1388                  * set in their headers. so it's OK to not update
1389                  * the 'last_snapshot'.
1390                  */
1391                 ret = btrfs_copy_root(trans, root, root->node, &eb,
1392                                       BTRFS_TREE_RELOC_OBJECTID);
1393                 BUG_ON(ret);
1394         }
1395 
1396         memcpy(root_item, &root->root_item, sizeof(*root_item));
1397         btrfs_set_root_bytenr(root_item, eb->start);
1398         btrfs_set_root_level(root_item, btrfs_header_level(eb));
1399         btrfs_set_root_generation(root_item, trans->transid);
1400 
1401         if (root->root_key.objectid == objectid) {
1402                 btrfs_set_root_refs(root_item, 0);
1403                 memset(&root_item->drop_progress, 0,
1404                        sizeof(struct btrfs_disk_key));
1405                 root_item->drop_level = 0;
1406         }
1407 
1408         btrfs_tree_unlock(eb);
1409         free_extent_buffer(eb);
1410 
1411         ret = btrfs_insert_root(trans, fs_info->tree_root,
1412                                 &root_key, root_item);
1413         BUG_ON(ret);
1414         kfree(root_item);
1415 
1416         reloc_root = btrfs_read_fs_root(fs_info->tree_root, &root_key);
1417         BUG_ON(IS_ERR(reloc_root));
1418         reloc_root->last_trans = trans->transid;
1419         return reloc_root;
1420 }
1421 
1422 /*
1423  * create reloc tree for a given fs tree. reloc tree is just a
1424  * snapshot of the fs tree with special root objectid.
1425  */
1426 int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
1427                           struct btrfs_root *root)
1428 {
1429         struct btrfs_fs_info *fs_info = root->fs_info;
1430         struct btrfs_root *reloc_root;
1431         struct reloc_control *rc = fs_info->reloc_ctl;
1432         struct btrfs_block_rsv *rsv;
1433         int clear_rsv = 0;
1434         int ret;
1435 
1436         if (root->reloc_root) {
1437                 reloc_root = root->reloc_root;
1438                 reloc_root->last_trans = trans->transid;
1439                 return 0;
1440         }
1441 
1442         if (!rc || !rc->create_reloc_tree ||
1443             root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1444                 return 0;
1445 
1446         if (!trans->reloc_reserved) {
1447                 rsv = trans->block_rsv;
1448                 trans->block_rsv = rc->block_rsv;
1449                 clear_rsv = 1;
1450         }
1451         reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
1452         if (clear_rsv)
1453                 trans->block_rsv = rsv;
1454 
1455         ret = __add_reloc_root(reloc_root);
1456         BUG_ON(ret < 0);
1457         root->reloc_root = reloc_root;
1458         return 0;
1459 }
1460 
1461 /*
1462  * update root item of reloc tree
1463  */
1464 int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
1465                             struct btrfs_root *root)
1466 {
1467         struct btrfs_fs_info *fs_info = root->fs_info;
1468         struct btrfs_root *reloc_root;
1469         struct btrfs_root_item *root_item;
1470         int ret;
1471 
1472         if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state) ||
1473             !root->reloc_root)
1474                 goto out;
1475 
1476         reloc_root = root->reloc_root;
1477         root_item = &reloc_root->root_item;
1478 
1479         /* root->reloc_root will stay until current relocation finished */
1480         if (fs_info->reloc_ctl->merge_reloc_tree &&
1481             btrfs_root_refs(root_item) == 0) {
1482                 set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
1483                 __del_reloc_root(reloc_root);
1484         }
1485 
1486         if (reloc_root->commit_root != reloc_root->node) {
1487                 btrfs_set_root_node(root_item, reloc_root->node);
1488                 free_extent_buffer(reloc_root->commit_root);
1489                 reloc_root->commit_root = btrfs_root_node(reloc_root);
1490         }
1491 
1492         ret = btrfs_update_root(trans, fs_info->tree_root,
1493                                 &reloc_root->root_key, root_item);
1494         BUG_ON(ret);
1495 
1496 out:
1497         return 0;
1498 }
1499 
1500 /*
1501  * helper to find first cached inode with inode number >= objectid
1502  * in a subvolume
1503  */
1504 static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
1505 {
1506         struct rb_node *node;
1507         struct rb_node *prev;
1508         struct btrfs_inode *entry;
1509         struct inode *inode;
1510 
1511         spin_lock(&root->inode_lock);
1512 again:
1513         node = root->inode_tree.rb_node;
1514         prev = NULL;
1515         while (node) {
1516                 prev = node;
1517                 entry = rb_entry(node, struct btrfs_inode, rb_node);
1518 
1519                 if (objectid < btrfs_ino(entry))
1520                         node = node->rb_left;
1521                 else if (objectid > btrfs_ino(entry))
1522                         node = node->rb_right;
1523                 else
1524                         break;
1525         }
1526         if (!node) {
1527                 while (prev) {
1528                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
1529                         if (objectid <= btrfs_ino(entry)) {
1530                                 node = prev;
1531                                 break;
1532                         }
1533                         prev = rb_next(prev);
1534                 }
1535         }
1536         while (node) {
1537                 entry = rb_entry(node, struct btrfs_inode, rb_node);
1538                 inode = igrab(&entry->vfs_inode);
1539                 if (inode) {
1540                         spin_unlock(&root->inode_lock);
1541                         return inode;
1542                 }
1543 
1544                 objectid = btrfs_ino(entry) + 1;
1545                 if (cond_resched_lock(&root->inode_lock))
1546                         goto again;
1547 
1548                 node = rb_next(node);
1549         }
1550         spin_unlock(&root->inode_lock);
1551         return NULL;
1552 }
1553 
1554 static int in_block_group(u64 bytenr,
1555                           struct btrfs_block_group_cache *block_group)
1556 {
1557         if (bytenr >= block_group->key.objectid &&
1558             bytenr < block_group->key.objectid + block_group->key.offset)
1559                 return 1;
1560         return 0;
1561 }
1562 
1563 /*
1564  * get new location of data
1565  */
1566 static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1567                             u64 bytenr, u64 num_bytes)
1568 {
1569         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1570         struct btrfs_path *path;
1571         struct btrfs_file_extent_item *fi;
1572         struct extent_buffer *leaf;
1573         int ret;
1574 
1575         path = btrfs_alloc_path();
1576         if (!path)
1577                 return -ENOMEM;
1578 
1579         bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1580         ret = btrfs_lookup_file_extent(NULL, root, path,
1581                         btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
1582         if (ret < 0)
1583                 goto out;
1584         if (ret > 0) {
1585                 ret = -ENOENT;
1586                 goto out;
1587         }
1588 
1589         leaf = path->nodes[0];
1590         fi = btrfs_item_ptr(leaf, path->slots[0],
1591                             struct btrfs_file_extent_item);
1592 
1593         BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1594                btrfs_file_extent_compression(leaf, fi) ||
1595                btrfs_file_extent_encryption(leaf, fi) ||
1596                btrfs_file_extent_other_encoding(leaf, fi));
1597 
1598         if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1599                 ret = -EINVAL;
1600                 goto out;
1601         }
1602 
1603         *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1604         ret = 0;
1605 out:
1606         btrfs_free_path(path);
1607         return ret;
1608 }
1609 
1610 /*
1611  * update file extent items in the tree leaf to point to
1612  * the new locations.
1613  */
1614 static noinline_for_stack
1615 int replace_file_extents(struct btrfs_trans_handle *trans,
1616                          struct reloc_control *rc,
1617                          struct btrfs_root *root,
1618                          struct extent_buffer *leaf)
1619 {
1620         struct btrfs_fs_info *fs_info = root->fs_info;
1621         struct btrfs_key key;
1622         struct btrfs_file_extent_item *fi;
1623         struct inode *inode = NULL;
1624         u64 parent;
1625         u64 bytenr;
1626         u64 new_bytenr = 0;
1627         u64 num_bytes;
1628         u64 end;
1629         u32 nritems;
1630         u32 i;
1631         int ret = 0;
1632         int first = 1;
1633         int dirty = 0;
1634 
1635         if (rc->stage != UPDATE_DATA_PTRS)
1636                 return 0;
1637 
1638         /* reloc trees always use full backref */
1639         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1640                 parent = leaf->start;
1641         else
1642                 parent = 0;
1643 
1644         nritems = btrfs_header_nritems(leaf);
1645         for (i = 0; i < nritems; i++) {
1646                 cond_resched();
1647                 btrfs_item_key_to_cpu(leaf, &key, i);
1648                 if (key.type != BTRFS_EXTENT_DATA_KEY)
1649                         continue;
1650                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1651                 if (btrfs_file_extent_type(leaf, fi) ==
1652                     BTRFS_FILE_EXTENT_INLINE)
1653                         continue;
1654                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1655                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1656                 if (bytenr == 0)
1657                         continue;
1658                 if (!in_block_group(bytenr, rc->block_group))
1659                         continue;
1660 
1661                 /*
1662                  * if we are modifying block in fs tree, wait for readpage
1663                  * to complete and drop the extent cache
1664                  */
1665                 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1666                         if (first) {
1667                                 inode = find_next_inode(root, key.objectid);
1668                                 first = 0;
1669                         } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) {
1670                                 btrfs_add_delayed_iput(inode);
1671                                 inode = find_next_inode(root, key.objectid);
1672                         }
1673                         if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
1674                                 end = key.offset +
1675                                       btrfs_file_extent_num_bytes(leaf, fi);
1676                                 WARN_ON(!IS_ALIGNED(key.offset,
1677                                                     fs_info->sectorsize));
1678                                 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1679                                 end--;
1680                                 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1681                                                       key.offset, end);
1682                                 if (!ret)
1683                                         continue;
1684 
1685                                 btrfs_drop_extent_cache(BTRFS_I(inode),
1686                                                 key.offset,     end, 1);
1687                                 unlock_extent(&BTRFS_I(inode)->io_tree,
1688                                               key.offset, end);
1689                         }
1690                 }
1691 
1692                 ret = get_new_location(rc->data_inode, &new_bytenr,
1693                                        bytenr, num_bytes);
1694                 if (ret) {
1695                         /*
1696                          * Don't have to abort since we've not changed anything
1697                          * in the file extent yet.
1698                          */
1699                         break;
1700                 }
1701 
1702                 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1703                 dirty = 1;
1704 
1705                 key.offset -= btrfs_file_extent_offset(leaf, fi);
1706                 ret = btrfs_inc_extent_ref(trans, root, new_bytenr,
1707                                            num_bytes, parent,
1708                                            btrfs_header_owner(leaf),
1709                                            key.objectid, key.offset);
1710                 if (ret) {
1711                         btrfs_abort_transaction(trans, ret);
1712                         break;
1713                 }
1714 
1715                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1716                                         parent, btrfs_header_owner(leaf),
1717                                         key.objectid, key.offset);
1718                 if (ret) {
1719                         btrfs_abort_transaction(trans, ret);
1720                         break;
1721                 }
1722         }
1723         if (dirty)
1724                 btrfs_mark_buffer_dirty(leaf);
1725         if (inode)
1726                 btrfs_add_delayed_iput(inode);
1727         return ret;
1728 }
1729 
1730 static noinline_for_stack
1731 int memcmp_node_keys(struct extent_buffer *eb, int slot,
1732                      struct btrfs_path *path, int level)
1733 {
1734         struct btrfs_disk_key key1;
1735         struct btrfs_disk_key key2;
1736         btrfs_node_key(eb, &key1, slot);
1737         btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1738         return memcmp(&key1, &key2, sizeof(key1));
1739 }
1740 
1741 /*
1742  * try to replace tree blocks in fs tree with the new blocks
1743  * in reloc tree. tree blocks haven't been modified since the
1744  * reloc tree was create can be replaced.
1745  *
1746  * if a block was replaced, level of the block + 1 is returned.
1747  * if no block got replaced, 0 is returned. if there are other
1748  * errors, a negative error number is returned.
1749  */
1750 static noinline_for_stack
1751 int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
1752                  struct btrfs_root *dest, struct btrfs_root *src,
1753                  struct btrfs_path *path, struct btrfs_key *next_key,
1754                  int lowest_level, int max_level)
1755 {
1756         struct btrfs_fs_info *fs_info = dest->fs_info;
1757         struct extent_buffer *eb;
1758         struct extent_buffer *parent;
1759         struct btrfs_key key;
1760         u64 old_bytenr;
1761         u64 new_bytenr;
1762         u64 old_ptr_gen;
1763         u64 new_ptr_gen;
1764         u64 last_snapshot;
1765         u32 blocksize;
1766         int cow = 0;
1767         int level;
1768         int ret;
1769         int slot;
1770 
1771         BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1772         BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1773 
1774         last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1775 again:
1776         slot = path->slots[lowest_level];
1777         btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1778 
1779         eb = btrfs_lock_root_node(dest);
1780         btrfs_set_lock_blocking_write(eb);
1781         level = btrfs_header_level(eb);
1782 
1783         if (level < lowest_level) {
1784                 btrfs_tree_unlock(eb);
1785                 free_extent_buffer(eb);
1786                 return 0;
1787         }
1788 
1789         if (cow) {
1790                 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1791                 BUG_ON(ret);
1792         }
1793         btrfs_set_lock_blocking_write(eb);
1794 
1795         if (next_key) {
1796                 next_key->objectid = (u64)-1;
1797                 next_key->type = (u8)-1;
1798                 next_key->offset = (u64)-1;
1799         }
1800 
1801         parent = eb;
1802         while (1) {
1803                 struct btrfs_key first_key;
1804 
1805                 level = btrfs_header_level(parent);
1806                 BUG_ON(level < lowest_level);
1807 
1808                 ret = btrfs_bin_search(parent, &key, level, &slot);
1809                 if (ret < 0)
1810                         break;
1811                 if (ret && slot > 0)
1812                         slot--;
1813 
1814                 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1815                         btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1816 
1817                 old_bytenr = btrfs_node_blockptr(parent, slot);
1818                 blocksize = fs_info->nodesize;
1819                 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1820                 btrfs_node_key_to_cpu(parent, &first_key, slot);
1821 
1822                 if (level <= max_level) {
1823                         eb = path->nodes[level];
1824                         new_bytenr = btrfs_node_blockptr(eb,
1825                                                         path->slots[level]);
1826                         new_ptr_gen = btrfs_node_ptr_generation(eb,
1827                                                         path->slots[level]);
1828                 } else {
1829                         new_bytenr = 0;
1830                         new_ptr_gen = 0;
1831                 }
1832 
1833                 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1834                         ret = level;
1835                         break;
1836                 }
1837 
1838                 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1839                     memcmp_node_keys(parent, slot, path, level)) {
1840                         if (level <= lowest_level) {
1841                                 ret = 0;
1842                                 break;
1843                         }
1844 
1845                         eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen,
1846                                              level - 1, &first_key);
1847                         if (IS_ERR(eb)) {
1848                                 ret = PTR_ERR(eb);
1849                                 break;
1850                         } else if (!extent_buffer_uptodate(eb)) {
1851                                 ret = -EIO;
1852                                 free_extent_buffer(eb);
1853                                 break;
1854                         }
1855                         btrfs_tree_lock(eb);
1856                         if (cow) {
1857                                 ret = btrfs_cow_block(trans, dest, eb, parent,
1858                                                       slot, &eb);
1859                                 BUG_ON(ret);
1860                         }
1861                         btrfs_set_lock_blocking_write(eb);
1862 
1863                         btrfs_tree_unlock(parent);
1864                         free_extent_buffer(parent);
1865 
1866                         parent = eb;
1867                         continue;
1868                 }
1869 
1870                 if (!cow) {
1871                         btrfs_tree_unlock(parent);
1872                         free_extent_buffer(parent);
1873                         cow = 1;
1874                         goto again;
1875                 }
1876 
1877                 btrfs_node_key_to_cpu(path->nodes[level], &key,
1878                                       path->slots[level]);
1879                 btrfs_release_path(path);
1880 
1881                 path->lowest_level = level;
1882                 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1883                 path->lowest_level = 0;
1884                 BUG_ON(ret);
1885 
1886                 /*
1887                  * Info qgroup to trace both subtrees.
1888                  *
1889                  * We must trace both trees.
1890                  * 1) Tree reloc subtree
1891                  *    If not traced, we will leak data numbers
1892                  * 2) Fs subtree
1893                  *    If not traced, we will double count old data
1894                  *
1895                  * We don't scan the subtree right now, but only record
1896                  * the swapped tree blocks.
1897                  * The real subtree rescan is delayed until we have new
1898                  * CoW on the subtree root node before transaction commit.
1899                  */
1900                 ret = btrfs_qgroup_add_swapped_blocks(trans, dest,
1901                                 rc->block_group, parent, slot,
1902                                 path->nodes[level], path->slots[level],
1903                                 last_snapshot);
1904                 if (ret < 0)
1905                         break;
1906                 /*
1907                  * swap blocks in fs tree and reloc tree.
1908                  */
1909                 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1910                 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1911                 btrfs_mark_buffer_dirty(parent);
1912 
1913                 btrfs_set_node_blockptr(path->nodes[level],
1914                                         path->slots[level], old_bytenr);
1915                 btrfs_set_node_ptr_generation(path->nodes[level],
1916                                               path->slots[level], old_ptr_gen);
1917                 btrfs_mark_buffer_dirty(path->nodes[level]);
1918 
1919                 ret = btrfs_inc_extent_ref(trans, src, old_bytenr,
1920                                         blocksize, path->nodes[level]->start,
1921                                         src->root_key.objectid, level - 1, 0);
1922                 BUG_ON(ret);
1923                 ret = btrfs_inc_extent_ref(trans, dest, new_bytenr,
1924                                         blocksize, 0, dest->root_key.objectid,
1925                                         level - 1, 0);
1926                 BUG_ON(ret);
1927 
1928                 ret = btrfs_free_extent(trans, src, new_bytenr, blocksize,
1929                                         path->nodes[level]->start,
1930                                         src->root_key.objectid, level - 1, 0);
1931                 BUG_ON(ret);
1932 
1933                 ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize,
1934                                         0, dest->root_key.objectid, level - 1,
1935                                         0);
1936                 BUG_ON(ret);
1937 
1938                 btrfs_unlock_up_safe(path, 0);
1939 
1940                 ret = level;
1941                 break;
1942         }
1943         btrfs_tree_unlock(parent);
1944         free_extent_buffer(parent);
1945         return ret;
1946 }
1947 
1948 /*
1949  * helper to find next relocated block in reloc tree
1950  */
1951 static noinline_for_stack
1952 int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1953                        int *level)
1954 {
1955         struct extent_buffer *eb;
1956         int i;
1957         u64 last_snapshot;
1958         u32 nritems;
1959 
1960         last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1961 
1962         for (i = 0; i < *level; i++) {
1963                 free_extent_buffer(path->nodes[i]);
1964                 path->nodes[i] = NULL;
1965         }
1966 
1967         for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1968                 eb = path->nodes[i];
1969                 nritems = btrfs_header_nritems(eb);
1970                 while (path->slots[i] + 1 < nritems) {
1971                         path->slots[i]++;
1972                         if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1973                             last_snapshot)
1974                                 continue;
1975 
1976                         *level = i;
1977                         return 0;
1978                 }
1979                 free_extent_buffer(path->nodes[i]);
1980                 path->nodes[i] = NULL;
1981         }
1982         return 1;
1983 }
1984 
1985 /*
1986  * walk down reloc tree to find relocated block of lowest level
1987  */
1988 static noinline_for_stack
1989 int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1990                          int *level)
1991 {
1992         struct btrfs_fs_info *fs_info = root->fs_info;
1993         struct extent_buffer *eb = NULL;
1994         int i;
1995         u64 bytenr;
1996         u64 ptr_gen = 0;
1997         u64 last_snapshot;
1998         u32 nritems;
1999 
2000         last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2001 
2002         for (i = *level; i > 0; i--) {
2003                 struct btrfs_key first_key;
2004 
2005                 eb = path->nodes[i];
2006                 nritems = btrfs_header_nritems(eb);
2007                 while (path->slots[i] < nritems) {
2008                         ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
2009                         if (ptr_gen > last_snapshot)
2010                                 break;
2011                         path->slots[i]++;
2012                 }
2013                 if (path->slots[i] >= nritems) {
2014                         if (i == *level)
2015                                 break;
2016                         *level = i + 1;
2017                         return 0;
2018                 }
2019                 if (i == 1) {
2020                         *level = i;
2021                         return 0;
2022                 }
2023 
2024                 bytenr = btrfs_node_blockptr(eb, path->slots[i]);
2025                 btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]);
2026                 eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1,
2027                                      &first_key);
2028                 if (IS_ERR(eb)) {
2029                         return PTR_ERR(eb);
2030                 } else if (!extent_buffer_uptodate(eb)) {
2031                         free_extent_buffer(eb);
2032                         return -EIO;
2033                 }
2034                 BUG_ON(btrfs_header_level(eb) != i - 1);
2035                 path->nodes[i - 1] = eb;
2036                 path->slots[i - 1] = 0;
2037         }
2038         return 1;
2039 }
2040 
2041 /*
2042  * invalidate extent cache for file extents whose key in range of
2043  * [min_key, max_key)
2044  */
2045 static int invalidate_extent_cache(struct btrfs_root *root,
2046                                    struct btrfs_key *min_key,
2047                                    struct btrfs_key *max_key)
2048 {
2049         struct btrfs_fs_info *fs_info = root->fs_info;
2050         struct inode *inode = NULL;
2051         u64 objectid;
2052         u64 start, end;
2053         u64 ino;
2054 
2055         objectid = min_key->objectid;
2056         while (1) {
2057                 cond_resched();
2058                 iput(inode);
2059 
2060                 if (objectid > max_key->objectid)
2061                         break;
2062 
2063                 inode = find_next_inode(root, objectid);
2064                 if (!inode)
2065                         break;
2066                 ino = btrfs_ino(BTRFS_I(inode));
2067 
2068                 if (ino > max_key->objectid) {
2069                         iput(inode);
2070                         break;
2071                 }
2072 
2073                 objectid = ino + 1;
2074                 if (!S_ISREG(inode->i_mode))
2075                         continue;
2076 
2077                 if (unlikely(min_key->objectid == ino)) {
2078                         if (min_key->type > BTRFS_EXTENT_DATA_KEY)
2079                                 continue;
2080                         if (min_key->type < BTRFS_EXTENT_DATA_KEY)
2081                                 start = 0;
2082                         else {
2083                                 start = min_key->offset;
2084                                 WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
2085                         }
2086                 } else {
2087                         start = 0;
2088                 }
2089 
2090                 if (unlikely(max_key->objectid == ino)) {
2091                         if (max_key->type < BTRFS_EXTENT_DATA_KEY)
2092                                 continue;
2093                         if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
2094                                 end = (u64)-1;
2095                         } else {
2096                                 if (max_key->offset == 0)
2097                                         continue;
2098                                 end = max_key->offset;
2099                                 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
2100                                 end--;
2101                         }
2102                 } else {
2103                         end = (u64)-1;
2104                 }
2105 
2106                 /* the lock_extent waits for readpage to complete */
2107                 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2108                 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1);
2109                 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2110         }
2111         return 0;
2112 }
2113 
2114 static int find_next_key(struct btrfs_path *path, int level,
2115                          struct btrfs_key *key)
2116 
2117 {
2118         while (level < BTRFS_MAX_LEVEL) {
2119                 if (!path->nodes[level])
2120                         break;
2121                 if (path->slots[level] + 1 <
2122                     btrfs_header_nritems(path->nodes[level])) {
2123                         btrfs_node_key_to_cpu(path->nodes[level], key,
2124                                               path->slots[level] + 1);
2125                         return 0;
2126                 }
2127                 level++;
2128         }
2129         return 1;
2130 }
2131 
2132 /*
2133  * Insert current subvolume into reloc_control::dirty_subvol_roots
2134  */
2135 static void insert_dirty_subvol(struct btrfs_trans_handle *trans,
2136                                 struct reloc_control *rc,
2137                                 struct btrfs_root *root)
2138 {
2139         struct btrfs_root *reloc_root = root->reloc_root;
2140         struct btrfs_root_item *reloc_root_item;
2141 
2142         /* @root must be a subvolume tree root with a valid reloc tree */
2143         ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
2144         ASSERT(reloc_root);
2145 
2146         reloc_root_item = &reloc_root->root_item;
2147         memset(&reloc_root_item->drop_progress, 0,
2148                 sizeof(reloc_root_item->drop_progress));
2149         reloc_root_item->drop_level = 0;
2150         btrfs_set_root_refs(reloc_root_item, 0);
2151         btrfs_update_reloc_root(trans, root);
2152 
2153         if (list_empty(&root->reloc_dirty_list)) {
2154                 btrfs_grab_fs_root(root);
2155                 list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots);
2156         }
2157 }
2158 
2159 static int clean_dirty_subvols(struct reloc_control *rc)
2160 {
2161         struct btrfs_root *root;
2162         struct btrfs_root *next;
2163         int ret = 0;
2164         int ret2;
2165 
2166         list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots,
2167                                  reloc_dirty_list) {
2168                 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2169                         /* Merged subvolume, cleanup its reloc root */
2170                         struct btrfs_root *reloc_root = root->reloc_root;
2171 
2172                         clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
2173                         list_del_init(&root->reloc_dirty_list);
2174                         root->reloc_root = NULL;
2175                         if (reloc_root) {
2176 
2177                                 ret2 = btrfs_drop_snapshot(reloc_root, NULL, 0, 1);
2178                                 if (ret2 < 0 && !ret)
2179                                         ret = ret2;
2180                         }
2181                         btrfs_put_fs_root(root);
2182                 } else {
2183                         /* Orphan reloc tree, just clean it up */
2184                         ret2 = btrfs_drop_snapshot(root, NULL, 0, 1);
2185                         if (ret2 < 0 && !ret)
2186                                 ret = ret2;
2187                 }
2188         }
2189         return ret;
2190 }
2191 
2192 /*
2193  * merge the relocated tree blocks in reloc tree with corresponding
2194  * fs tree.
2195  */
2196 static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
2197                                                struct btrfs_root *root)
2198 {
2199         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2200         struct btrfs_key key;
2201         struct btrfs_key next_key;
2202         struct btrfs_trans_handle *trans = NULL;
2203         struct btrfs_root *reloc_root;
2204         struct btrfs_root_item *root_item;
2205         struct btrfs_path *path;
2206         struct extent_buffer *leaf;
2207         int level;
2208         int max_level;
2209         int replaced = 0;
2210         int ret;
2211         int err = 0;
2212         u32 min_reserved;
2213 
2214         path = btrfs_alloc_path();
2215         if (!path)
2216                 return -ENOMEM;
2217         path->reada = READA_FORWARD;
2218 
2219         reloc_root = root->reloc_root;
2220         root_item = &reloc_root->root_item;
2221 
2222         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2223                 level = btrfs_root_level(root_item);
2224                 extent_buffer_get(reloc_root->node);
2225                 path->nodes[level] = reloc_root->node;
2226                 path->slots[level] = 0;
2227         } else {
2228                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2229 
2230                 level = root_item->drop_level;
2231                 BUG_ON(level == 0);
2232                 path->lowest_level = level;
2233                 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
2234                 path->lowest_level = 0;
2235                 if (ret < 0) {
2236                         btrfs_free_path(path);
2237                         return ret;
2238                 }
2239 
2240                 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
2241                                       path->slots[level]);
2242                 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
2243 
2244                 btrfs_unlock_up_safe(path, 0);
2245         }
2246 
2247         min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2248         memset(&next_key, 0, sizeof(next_key));
2249 
2250         while (1) {
2251                 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
2252                                              BTRFS_RESERVE_FLUSH_ALL);
2253                 if (ret) {
2254                         err = ret;
2255                         goto out;
2256                 }
2257                 trans = btrfs_start_transaction(root, 0);
2258                 if (IS_ERR(trans)) {
2259                         err = PTR_ERR(trans);
2260                         trans = NULL;
2261                         goto out;
2262                 }
2263                 trans->block_rsv = rc->block_rsv;
2264 
2265                 replaced = 0;
2266                 max_level = level;
2267 
2268                 ret = walk_down_reloc_tree(reloc_root, path, &level);
2269                 if (ret < 0) {
2270                         err = ret;
2271                         goto out;
2272                 }
2273                 if (ret > 0)
2274                         break;
2275 
2276                 if (!find_next_key(path, level, &key) &&
2277                     btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
2278                         ret = 0;
2279                 } else {
2280                         ret = replace_path(trans, rc, root, reloc_root, path,
2281                                            &next_key, level, max_level);
2282                 }
2283                 if (ret < 0) {
2284                         err = ret;
2285                         goto out;
2286                 }
2287 
2288                 if (ret > 0) {
2289                         level = ret;
2290                         btrfs_node_key_to_cpu(path->nodes[level], &key,
2291                                               path->slots[level]);
2292                         replaced = 1;
2293                 }
2294 
2295                 ret = walk_up_reloc_tree(reloc_root, path, &level);
2296                 if (ret > 0)
2297                         break;
2298 
2299                 BUG_ON(level == 0);
2300                 /*
2301                  * save the merging progress in the drop_progress.
2302                  * this is OK since root refs == 1 in this case.
2303                  */
2304                 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
2305                                path->slots[level]);
2306                 root_item->drop_level = level;
2307 
2308                 btrfs_end_transaction_throttle(trans);
2309                 trans = NULL;
2310 
2311                 btrfs_btree_balance_dirty(fs_info);
2312 
2313                 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2314                         invalidate_extent_cache(root, &key, &next_key);
2315         }
2316 
2317         /*
2318          * handle the case only one block in the fs tree need to be
2319          * relocated and the block is tree root.
2320          */
2321         leaf = btrfs_lock_root_node(root);
2322         ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
2323         btrfs_tree_unlock(leaf);
2324         free_extent_buffer(leaf);
2325         if (ret < 0)
2326                 err = ret;
2327 out:
2328         btrfs_free_path(path);
2329 
2330         if (err == 0)
2331                 insert_dirty_subvol(trans, rc, root);
2332 
2333         if (trans)
2334                 btrfs_end_transaction_throttle(trans);
2335 
2336         btrfs_btree_balance_dirty(fs_info);
2337 
2338         if (replaced && rc->stage == UPDATE_DATA_PTRS)
2339                 invalidate_extent_cache(root, &key, &next_key);
2340 
2341         return err;
2342 }
2343 
2344 static noinline_for_stack
2345 int prepare_to_merge(struct reloc_control *rc, int err)
2346 {
2347         struct btrfs_root *root = rc->extent_root;
2348         struct btrfs_fs_info *fs_info = root->fs_info;
2349         struct btrfs_root *reloc_root;
2350         struct btrfs_trans_handle *trans;
2351         LIST_HEAD(reloc_roots);
2352         u64 num_bytes = 0;
2353         int ret;
2354 
2355         mutex_lock(&fs_info->reloc_mutex);
2356         rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2357         rc->merging_rsv_size += rc->nodes_relocated * 2;
2358         mutex_unlock(&fs_info->reloc_mutex);
2359 
2360 again:
2361         if (!err) {
2362                 num_bytes = rc->merging_rsv_size;
2363                 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
2364                                           BTRFS_RESERVE_FLUSH_ALL);
2365                 if (ret)
2366                         err = ret;
2367         }
2368 
2369         trans = btrfs_join_transaction(rc->extent_root);
2370         if (IS_ERR(trans)) {
2371                 if (!err)
2372                         btrfs_block_rsv_release(fs_info, rc->block_rsv,
2373                                                 num_bytes);
2374                 return PTR_ERR(trans);
2375         }
2376 
2377         if (!err) {
2378                 if (num_bytes != rc->merging_rsv_size) {
2379                         btrfs_end_transaction(trans);
2380                         btrfs_block_rsv_release(fs_info, rc->block_rsv,
2381                                                 num_bytes);
2382                         goto again;
2383                 }
2384         }
2385 
2386         rc->merge_reloc_tree = 1;
2387 
2388         while (!list_empty(&rc->reloc_roots)) {
2389                 reloc_root = list_entry(rc->reloc_roots.next,
2390                                         struct btrfs_root, root_list);
2391                 list_del_init(&reloc_root->root_list);
2392 
2393                 root = read_fs_root(fs_info, reloc_root->root_key.offset);
2394                 BUG_ON(IS_ERR(root));
2395                 BUG_ON(root->reloc_root != reloc_root);
2396 
2397                 /*
2398                  * set reference count to 1, so btrfs_recover_relocation
2399                  * knows it should resumes merging
2400                  */
2401                 if (!err)
2402                         btrfs_set_root_refs(&reloc_root->root_item, 1);
2403                 btrfs_update_reloc_root(trans, root);
2404 
2405                 list_add(&reloc_root->root_list, &reloc_roots);
2406         }
2407 
2408         list_splice(&reloc_roots, &rc->reloc_roots);
2409 
2410         if (!err)
2411                 btrfs_commit_transaction(trans);
2412         else
2413                 btrfs_end_transaction(trans);
2414         return err;
2415 }
2416 
2417 static noinline_for_stack
2418 void free_reloc_roots(struct list_head *list)
2419 {
2420         struct btrfs_root *reloc_root;
2421 
2422         while (!list_empty(list)) {
2423                 reloc_root = list_entry(list->next, struct btrfs_root,
2424                                         root_list);
2425                 __del_reloc_root(reloc_root);
2426                 free_extent_buffer(reloc_root->node);
2427                 free_extent_buffer(reloc_root->commit_root);
2428                 reloc_root->node = NULL;
2429                 reloc_root->commit_root = NULL;
2430         }
2431 }
2432 
2433 static noinline_for_stack
2434 void merge_reloc_roots(struct reloc_control *rc)
2435 {
2436         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2437         struct btrfs_root *root;
2438         struct btrfs_root *reloc_root;
2439         LIST_HEAD(reloc_roots);
2440         int found = 0;
2441         int ret = 0;
2442 again:
2443         root = rc->extent_root;
2444 
2445         /*
2446          * this serializes us with btrfs_record_root_in_transaction,
2447          * we have to make sure nobody is in the middle of
2448          * adding their roots to the list while we are
2449          * doing this splice
2450          */
2451         mutex_lock(&fs_info->reloc_mutex);
2452         list_splice_init(&rc->reloc_roots, &reloc_roots);
2453         mutex_unlock(&fs_info->reloc_mutex);
2454 
2455         while (!list_empty(&reloc_roots)) {
2456                 found = 1;
2457                 reloc_root = list_entry(reloc_roots.next,
2458                                         struct btrfs_root, root_list);
2459 
2460                 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
2461                         root = read_fs_root(fs_info,
2462                                             reloc_root->root_key.offset);
2463                         BUG_ON(IS_ERR(root));
2464                         BUG_ON(root->reloc_root != reloc_root);
2465 
2466                         ret = merge_reloc_root(rc, root);
2467                         if (ret) {
2468                                 if (list_empty(&reloc_root->root_list))
2469                                         list_add_tail(&reloc_root->root_list,
2470                                                       &reloc_roots);
2471                                 goto out;
2472                         }
2473                 } else {
2474                         list_del_init(&reloc_root->root_list);
2475                         /* Don't forget to queue this reloc root for cleanup */
2476                         list_add_tail(&reloc_root->reloc_dirty_list,
2477                                       &rc->dirty_subvol_roots);
2478                 }
2479         }
2480 
2481         if (found) {
2482                 found = 0;
2483                 goto again;
2484         }
2485 out:
2486         if (ret) {
2487                 btrfs_handle_fs_error(fs_info, ret, NULL);
2488                 if (!list_empty(&reloc_roots))
2489                         free_reloc_roots(&reloc_roots);
2490 
2491                 /* new reloc root may be added */
2492                 mutex_lock(&fs_info->reloc_mutex);
2493                 list_splice_init(&rc->reloc_roots, &reloc_roots);
2494                 mutex_unlock(&fs_info->reloc_mutex);
2495                 if (!list_empty(&reloc_roots))
2496                         free_reloc_roots(&reloc_roots);
2497         }
2498 
2499         BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
2500 }
2501 
2502 static void free_block_list(struct rb_root *blocks)
2503 {
2504         struct tree_block *block;
2505         struct rb_node *rb_node;
2506         while ((rb_node = rb_first(blocks))) {
2507                 block = rb_entry(rb_node, struct tree_block, rb_node);
2508                 rb_erase(rb_node, blocks);
2509                 kfree(block);
2510         }
2511 }
2512 
2513 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2514                                       struct btrfs_root *reloc_root)
2515 {
2516         struct btrfs_fs_info *fs_info = reloc_root->fs_info;
2517         struct btrfs_root *root;
2518 
2519         if (reloc_root->last_trans == trans->transid)
2520                 return 0;
2521 
2522         root = read_fs_root(fs_info, reloc_root->root_key.offset);
2523         BUG_ON(IS_ERR(root));
2524         BUG_ON(root->reloc_root != reloc_root);
2525 
2526         return btrfs_record_root_in_trans(trans, root);
2527 }
2528 
2529 static noinline_for_stack
2530 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2531                                      struct reloc_control *rc,
2532                                      struct backref_node *node,
2533                                      struct backref_edge *edges[])
2534 {
2535         struct backref_node *next;
2536         struct btrfs_root *root;
2537         int index = 0;
2538 
2539         next = node;
2540         while (1) {
2541                 cond_resched();
2542                 next = walk_up_backref(next, edges, &index);
2543                 root = next->root;
2544                 BUG_ON(!root);
2545                 BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state));
2546 
2547                 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2548                         record_reloc_root_in_trans(trans, root);
2549                         break;
2550                 }
2551 
2552                 btrfs_record_root_in_trans(trans, root);
2553                 root = root->reloc_root;
2554 
2555                 if (next->new_bytenr != root->node->start) {
2556                         BUG_ON(next->new_bytenr);
2557                         BUG_ON(!list_empty(&next->list));
2558                         next->new_bytenr = root->node->start;
2559                         next->root = root;
2560                         list_add_tail(&next->list,
2561                                       &rc->backref_cache.changed);
2562                         __mark_block_processed(rc, next);
2563                         break;
2564                 }
2565 
2566                 WARN_ON(1);
2567                 root = NULL;
2568                 next = walk_down_backref(edges, &index);
2569                 if (!next || next->level <= node->level)
2570                         break;
2571         }
2572         if (!root)
2573                 return NULL;
2574 
2575         next = node;
2576         /* setup backref node path for btrfs_reloc_cow_block */
2577         while (1) {
2578                 rc->backref_cache.path[next->level] = next;
2579                 if (--index < 0)
2580                         break;
2581                 next = edges[index]->node[UPPER];
2582         }
2583         return root;
2584 }
2585 
2586 /*
2587  * select a tree root for relocation. return NULL if the block
2588  * is reference counted. we should use do_relocation() in this
2589  * case. return a tree root pointer if the block isn't reference
2590  * counted. return -ENOENT if the block is root of reloc tree.
2591  */
2592 static noinline_for_stack
2593 struct btrfs_root *select_one_root(struct backref_node *node)
2594 {
2595         struct backref_node *next;
2596         struct btrfs_root *root;
2597         struct btrfs_root *fs_root = NULL;
2598         struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2599         int index = 0;
2600 
2601         next = node;
2602         while (1) {
2603                 cond_resched();
2604                 next = walk_up_backref(next, edges, &index);
2605                 root = next->root;
2606                 BUG_ON(!root);
2607 
2608                 /* no other choice for non-references counted tree */
2609                 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
2610                         return root;
2611 
2612                 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2613                         fs_root = root;
2614 
2615                 if (next != node)
2616                         return NULL;
2617 
2618                 next = walk_down_backref(edges, &index);
2619                 if (!next || next->level <= node->level)
2620                         break;
2621         }
2622 
2623         if (!fs_root)
2624                 return ERR_PTR(-ENOENT);
2625         return fs_root;
2626 }
2627 
2628 static noinline_for_stack
2629 u64 calcu_metadata_size(struct reloc_control *rc,
2630                         struct backref_node *node, int reserve)
2631 {
2632         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2633         struct backref_node *next = node;
2634         struct backref_edge *edge;
2635         struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2636         u64 num_bytes = 0;
2637         int index = 0;
2638 
2639         BUG_ON(reserve && node->processed);
2640 
2641         while (next) {
2642                 cond_resched();
2643                 while (1) {
2644                         if (next->processed && (reserve || next != node))
2645                                 break;
2646 
2647                         num_bytes += fs_info->nodesize;
2648 
2649                         if (list_empty(&next->upper))
2650                                 break;
2651 
2652                         edge = list_entry(next->upper.next,
2653                                           struct backref_edge, list[LOWER]);
2654                         edges[index++] = edge;
2655                         next = edge->node[UPPER];
2656                 }
2657                 next = walk_down_backref(edges, &index);
2658         }
2659         return num_bytes;
2660 }
2661 
2662 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2663                                   struct reloc_control *rc,
2664                                   struct backref_node *node)
2665 {
2666         struct btrfs_root *root = rc->extent_root;
2667         struct btrfs_fs_info *fs_info = root->fs_info;
2668         u64 num_bytes;
2669         int ret;
2670         u64 tmp;
2671 
2672         num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2673 
2674         trans->block_rsv = rc->block_rsv;
2675         rc->reserved_bytes += num_bytes;
2676 
2677         /*
2678          * We are under a transaction here so we can only do limited flushing.
2679          * If we get an enospc just kick back -EAGAIN so we know to drop the
2680          * transaction and try to refill when we can flush all the things.
2681          */
2682         ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2683                                 BTRFS_RESERVE_FLUSH_LIMIT);
2684         if (ret) {
2685                 tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2686                 while (tmp <= rc->reserved_bytes)
2687                         tmp <<= 1;
2688                 /*
2689                  * only one thread can access block_rsv at this point,
2690                  * so we don't need hold lock to protect block_rsv.
2691                  * we expand more reservation size here to allow enough
2692                  * space for relocation and we will return earlier in
2693                  * enospc case.
2694                  */
2695                 rc->block_rsv->size = tmp + fs_info->nodesize *
2696                                       RELOCATION_RESERVED_NODES;
2697                 return -EAGAIN;
2698         }
2699 
2700         return 0;
2701 }
2702 
2703 /*
2704  * relocate a block tree, and then update pointers in upper level
2705  * blocks that reference the block to point to the new location.
2706  *
2707  * if called by link_to_upper, the block has already been relocated.
2708  * in that case this function just updates pointers.
2709  */
2710 static int do_relocation(struct btrfs_trans_handle *trans,
2711                          struct reloc_control *rc,
2712                          struct backref_node *node,
2713                          struct btrfs_key *key,
2714                          struct btrfs_path *path, int lowest)
2715 {
2716         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2717         struct backref_node *upper;
2718         struct backref_edge *edge;
2719         struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2720         struct btrfs_root *root;
2721         struct extent_buffer *eb;
2722         u32 blocksize;
2723         u64 bytenr;
2724         u64 generation;
2725         int slot;
2726         int ret;
2727         int err = 0;
2728 
2729         BUG_ON(lowest && node->eb);
2730 
2731         path->lowest_level = node->level + 1;
2732         rc->backref_cache.path[node->level] = node;
2733         list_for_each_entry(edge, &node->upper, list[LOWER]) {
2734                 struct btrfs_key first_key;
2735 
2736                 cond_resched();
2737 
2738                 upper = edge->node[UPPER];
2739                 root = select_reloc_root(trans, rc, upper, edges);
2740                 BUG_ON(!root);
2741 
2742                 if (upper->eb && !upper->locked) {
2743                         if (!lowest) {
2744                                 ret = btrfs_bin_search(upper->eb, key,
2745                                                        upper->level, &slot);
2746                                 if (ret < 0) {
2747                                         err = ret;
2748                                         goto next;
2749                                 }
2750                                 BUG_ON(ret);
2751                                 bytenr = btrfs_node_blockptr(upper->eb, slot);
2752                                 if (node->eb->start == bytenr)
2753                                         goto next;
2754                         }
2755                         drop_node_buffer(upper);
2756                 }
2757 
2758                 if (!upper->eb) {
2759                         ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2760                         if (ret) {
2761                                 if (ret < 0)
2762                                         err = ret;
2763                                 else
2764                                         err = -ENOENT;
2765 
2766                                 btrfs_release_path(path);
2767                                 break;
2768                         }
2769 
2770                         if (!upper->eb) {
2771                                 upper->eb = path->nodes[upper->level];
2772                                 path->nodes[upper->level] = NULL;
2773                         } else {
2774                                 BUG_ON(upper->eb != path->nodes[upper->level]);
2775                         }
2776 
2777                         upper->locked = 1;
2778                         path->locks[upper->level] = 0;
2779 
2780                         slot = path->slots[upper->level];
2781                         btrfs_release_path(path);
2782                 } else {
2783                         ret = btrfs_bin_search(upper->eb, key, upper->level,
2784                                                &slot);
2785                         if (ret < 0) {
2786                                 err = ret;
2787                                 goto next;
2788                         }
2789                         BUG_ON(ret);
2790                 }
2791 
2792                 bytenr = btrfs_node_blockptr(upper->eb, slot);
2793                 if (lowest) {
2794                         if (bytenr != node->bytenr) {
2795                                 btrfs_err(root->fs_info,
2796                 "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2797                                           bytenr, node->bytenr, slot,
2798                                           upper->eb->start);
2799                                 err = -EIO;
2800                                 goto next;
2801                         }
2802                 } else {
2803                         if (node->eb->start == bytenr)
2804                                 goto next;
2805                 }
2806 
2807                 blocksize = root->fs_info->nodesize;
2808                 generation = btrfs_node_ptr_generation(upper->eb, slot);
2809                 btrfs_node_key_to_cpu(upper->eb, &first_key, slot);
2810                 eb = read_tree_block(fs_info, bytenr, generation,
2811                                      upper->level - 1, &first_key);
2812                 if (IS_ERR(eb)) {
2813                         err = PTR_ERR(eb);
2814                         goto next;
2815                 } else if (!extent_buffer_uptodate(eb)) {
2816                         free_extent_buffer(eb);
2817                         err = -EIO;
2818                         goto next;
2819                 }
2820                 btrfs_tree_lock(eb);
2821                 btrfs_set_lock_blocking_write(eb);
2822 
2823                 if (!node->eb) {
2824                         ret = btrfs_cow_block(trans, root, eb, upper->eb,
2825                                               slot, &eb);
2826                         btrfs_tree_unlock(eb);
2827                         free_extent_buffer(eb);
2828                         if (ret < 0) {
2829                                 err = ret;
2830                                 goto next;
2831                         }
2832                         BUG_ON(node->eb != eb);
2833                 } else {
2834                         btrfs_set_node_blockptr(upper->eb, slot,
2835                                                 node->eb->start);
2836                         btrfs_set_node_ptr_generation(upper->eb, slot,
2837                                                       trans->transid);
2838                         btrfs_mark_buffer_dirty(upper->eb);
2839 
2840                         ret = btrfs_inc_extent_ref(trans, root,
2841                                                 node->eb->start, blocksize,
2842                                                 upper->eb->start,
2843                                                 btrfs_header_owner(upper->eb),
2844                                                 node->level, 0);
2845                         BUG_ON(ret);
2846 
2847                         ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2848                         BUG_ON(ret);
2849                 }
2850 next:
2851                 if (!upper->pending)
2852                         drop_node_buffer(upper);
2853                 else
2854                         unlock_node_buffer(upper);
2855                 if (err)
2856                         break;
2857         }
2858 
2859         if (!err && node->pending) {
2860                 drop_node_buffer(node);
2861                 list_move_tail(&node->list, &rc->backref_cache.changed);
2862                 node->pending = 0;
2863         }
2864 
2865         path->lowest_level = 0;
2866         BUG_ON(err == -ENOSPC);
2867         return err;
2868 }
2869 
2870 static int link_to_upper(struct btrfs_trans_handle *trans,
2871                          struct reloc_control *rc,
2872                          struct backref_node *node,
2873                          struct btrfs_path *path)
2874 {
2875         struct btrfs_key key;
2876 
2877         btrfs_node_key_to_cpu(node->eb, &key, 0);
2878         return do_relocation(trans, rc, node, &key, path, 0);
2879 }
2880 
2881 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2882                                 struct reloc_control *rc,
2883                                 struct btrfs_path *path, int err)
2884 {
2885         LIST_HEAD(list);
2886         struct backref_cache *cache = &rc->backref_cache;
2887         struct backref_node *node;
2888         int level;
2889         int ret;
2890 
2891         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2892                 while (!list_empty(&cache->pending[level])) {
2893                         node = list_entry(cache->pending[level].next,
2894                                           struct backref_node, list);
2895                         list_move_tail(&node->list, &list);
2896                         BUG_ON(!node->pending);
2897 
2898                         if (!err) {
2899                                 ret = link_to_upper(trans, rc, node, path);
2900                                 if (ret < 0)
2901                                         err = ret;
2902                         }
2903                 }
2904                 list_splice_init(&list, &cache->pending[level]);
2905         }
2906         return err;
2907 }
2908 
2909 static void mark_block_processed(struct reloc_control *rc,
2910                                  u64 bytenr, u32 blocksize)
2911 {
2912         set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1,
2913                         EXTENT_DIRTY);
2914 }
2915 
2916 static void __mark_block_processed(struct reloc_control *rc,
2917                                    struct backref_node *node)
2918 {
2919         u32 blocksize;
2920         if (node->level == 0 ||
2921             in_block_group(node->bytenr, rc->block_group)) {
2922                 blocksize = rc->extent_root->fs_info->nodesize;
2923                 mark_block_processed(rc, node->bytenr, blocksize);
2924         }
2925         node->processed = 1;
2926 }
2927 
2928 /*
2929  * mark a block and all blocks directly/indirectly reference the block
2930  * as processed.
2931  */
2932 static void update_processed_blocks(struct reloc_control *rc,
2933                                     struct backref_node *node)
2934 {
2935         struct backref_node *next = node;
2936         struct backref_edge *edge;
2937         struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2938         int index = 0;
2939 
2940         while (next) {
2941                 cond_resched();
2942                 while (1) {
2943                         if (next->processed)
2944                                 break;
2945 
2946                         __mark_block_processed(rc, next);
2947 
2948                         if (list_empty(&next->upper))
2949                                 break;
2950 
2951                         edge = list_entry(next->upper.next,
2952                                           struct backref_edge, list[LOWER]);
2953                         edges[index++] = edge;
2954                         next = edge->node[UPPER];
2955                 }
2956                 next = walk_down_backref(edges, &index);
2957         }
2958 }
2959 
2960 static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2961 {
2962         u32 blocksize = rc->extent_root->fs_info->nodesize;
2963 
2964         if (test_range_bit(&rc->processed_blocks, bytenr,
2965                            bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2966                 return 1;
2967         return 0;
2968 }
2969 
2970 static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2971                               struct tree_block *block)
2972 {
2973         struct extent_buffer *eb;
2974 
2975         BUG_ON(block->key_ready);
2976         eb = read_tree_block(fs_info, block->bytenr, block->key.offset,
2977                              block->level, NULL);
2978         if (IS_ERR(eb)) {
2979                 return PTR_ERR(eb);
2980         } else if (!extent_buffer_uptodate(eb)) {
2981                 free_extent_buffer(eb);
2982                 return -EIO;
2983         }
2984         if (block->level == 0)
2985                 btrfs_item_key_to_cpu(eb, &block->key, 0);
2986         else
2987                 btrfs_node_key_to_cpu(eb, &block->key, 0);
2988         free_extent_buffer(eb);
2989         block->key_ready = 1;
2990         return 0;
2991 }
2992 
2993 /*
2994  * helper function to relocate a tree block
2995  */
2996 static int relocate_tree_block(struct btrfs_trans_handle *trans,
2997                                 struct reloc_control *rc,
2998                                 struct backref_node *node,
2999                                 struct btrfs_key *key,
3000                                 struct btrfs_path *path)
3001 {
3002         struct btrfs_root *root;
3003         int ret = 0;
3004 
3005         if (!node)
3006                 return 0;
3007 
3008         BUG_ON(node->processed);
3009         root = select_one_root(node);
3010         if (root == ERR_PTR(-ENOENT)) {
3011                 update_processed_blocks(rc, node);
3012                 goto out;
3013         }
3014 
3015         if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
3016                 ret = reserve_metadata_space(trans, rc, node);
3017                 if (ret)
3018                         goto out;
3019         }
3020 
3021         if (root) {
3022                 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
3023                         BUG_ON(node->new_bytenr);
3024                         BUG_ON(!list_empty(&node->list));
3025                         btrfs_record_root_in_trans(trans, root);
3026                         root = root->reloc_root;
3027                         node->new_bytenr = root->node->start;
3028                         node->root = root;
3029                         list_add_tail(&node->list, &rc->backref_cache.changed);
3030                 } else {
3031                         path->lowest_level = node->level;
3032                         ret = btrfs_search_slot(trans, root, key, path, 0, 1);
3033                         btrfs_release_path(path);
3034                         if (ret > 0)
3035                                 ret = 0;
3036                 }
3037                 if (!ret)
3038                         update_processed_blocks(rc, node);
3039         } else {
3040                 ret = do_relocation(trans, rc, node, key, path, 1);
3041         }
3042 out:
3043         if (ret || node->level == 0 || node->cowonly)
3044                 remove_backref_node(&rc->backref_cache, node);
3045         return ret;
3046 }
3047 
3048 /*
3049  * relocate a list of blocks
3050  */
3051 static noinline_for_stack
3052 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
3053                          struct reloc_control *rc, struct rb_root *blocks)
3054 {
3055         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3056         struct backref_node *node;
3057         struct btrfs_path *path;
3058         struct tree_block *block;
3059         struct tree_block *next;
3060         int ret;
3061         int err = 0;
3062 
3063         path = btrfs_alloc_path();
3064         if (!path) {
3065                 err = -ENOMEM;
3066                 goto out_free_blocks;
3067         }
3068 
3069         /* Kick in readahead for tree blocks with missing keys */
3070         rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
3071                 if (!block->key_ready)
3072                         readahead_tree_block(fs_info, block->bytenr);
3073         }
3074 
3075         /* Get first keys */
3076         rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
3077                 if (!block->key_ready) {
3078                         err = get_tree_block_key(fs_info, block);
3079                         if (err)
3080                                 goto out_free_path;
3081                 }
3082         }
3083 
3084         /* Do tree relocation */
3085         rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
3086                 node = build_backref_tree(rc, &block->key,
3087                                           block->level, block->bytenr);
3088                 if (IS_ERR(node)) {
3089                         err = PTR_ERR(node);
3090                         goto out;
3091                 }
3092 
3093                 ret = relocate_tree_block(trans, rc, node, &block->key,
3094                                           path);
3095                 if (ret < 0) {
3096                         if (ret != -EAGAIN || &block->rb_node == rb_first(blocks))
3097                                 err = ret;
3098                         goto out;
3099                 }
3100         }
3101 out:
3102         err = finish_pending_nodes(trans, rc, path, err);
3103 
3104 out_free_path:
3105         btrfs_free_path(path);
3106 out_free_blocks:
3107         free_block_list(blocks);
3108         return err;
3109 }
3110 
3111 static noinline_for_stack
3112 int prealloc_file_extent_cluster(struct inode *inode,
3113                                  struct file_extent_cluster *cluster)
3114 {
3115         u64 alloc_hint = 0;
3116         u64 start;
3117         u64 end;
3118         u64 offset = BTRFS_I(inode)->index_cnt;
3119         u64 num_bytes;
3120         int nr = 0;
3121         int ret = 0;
3122         u64 prealloc_start = cluster->start - offset;
3123         u64 prealloc_end = cluster->end - offset;
3124         u64 cur_offset;
3125         struct extent_changeset *data_reserved = NULL;
3126 
3127         BUG_ON(cluster->start != cluster->boundary[0]);
3128         inode_lock(inode);
3129 
3130         ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start,
3131                                           prealloc_end + 1 - prealloc_start);
3132         if (ret)
3133                 goto out;
3134 
3135         cur_offset = prealloc_start;
3136         while (nr < cluster->nr) {
3137                 start = cluster->boundary[nr] - offset;
3138                 if (nr + 1 < cluster->nr)
3139                         end = cluster->boundary[nr + 1] - 1 - offset;
3140                 else
3141                         end = cluster->end - offset;
3142 
3143                 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3144                 num_bytes = end + 1 - start;
3145                 if (cur_offset < start)
3146                         btrfs_free_reserved_data_space(inode, data_reserved,
3147                                         cur_offset, start - cur_offset);
3148                 ret = btrfs_prealloc_file_range(inode, 0, start,
3149                                                 num_bytes, num_bytes,
3150                                                 end + 1, &alloc_hint);
3151                 cur_offset = end + 1;
3152                 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3153                 if (ret)
3154                         break;
3155                 nr++;
3156         }
3157         if (cur_offset < prealloc_end)
3158                 btrfs_free_reserved_data_space(inode, data_reserved,
3159                                 cur_offset, prealloc_end + 1 - cur_offset);
3160 out:
3161         inode_unlock(inode);
3162         extent_changeset_free(data_reserved);
3163         return ret;
3164 }
3165 
3166 static noinline_for_stack
3167 int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
3168                          u64 block_start)
3169 {
3170         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3171         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3172         struct extent_map *em;
3173         int ret = 0;
3174 
3175         em = alloc_extent_map();
3176         if (!em)
3177                 return -ENOMEM;
3178 
3179         em->start = start;
3180         em->len = end + 1 - start;
3181         em->block_len = em->len;
3182         em->block_start = block_start;
3183         em->bdev = fs_info->fs_devices->latest_bdev;
3184         set_bit(EXTENT_FLAG_PINNED, &em->flags);
3185 
3186         lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3187         while (1) {
3188                 write_lock(&em_tree->lock);
3189                 ret = add_extent_mapping(em_tree, em, 0);
3190                 write_unlock(&em_tree->lock);
3191                 if (ret != -EEXIST) {
3192                         free_extent_map(em);
3193                         break;
3194                 }
3195                 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3196         }
3197         unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3198         return ret;
3199 }
3200 
3201 static int relocate_file_extent_cluster(struct inode *inode,
3202                                         struct file_extent_cluster *cluster)
3203 {
3204         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3205         u64 page_start;
3206         u64 page_end;
3207         u64 offset = BTRFS_I(inode)->index_cnt;
3208         unsigned long index;
3209         unsigned long last_index;
3210         struct page *page;
3211         struct file_ra_state *ra;
3212         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
3213         int nr = 0;
3214         int ret = 0;
3215 
3216         if (!cluster->nr)
3217                 return 0;
3218 
3219         ra = kzalloc(sizeof(*ra), GFP_NOFS);
3220         if (!ra)
3221                 return -ENOMEM;
3222 
3223         ret = prealloc_file_extent_cluster(inode, cluster);
3224         if (ret)
3225                 goto out;
3226 
3227         file_ra_state_init(ra, inode->i_mapping);
3228 
3229         ret = setup_extent_mapping(inode, cluster->start - offset,
3230                                    cluster->end - offset, cluster->start);
3231         if (ret)
3232                 goto out;
3233 
3234         index = (cluster->start - offset) >> PAGE_SHIFT;
3235         last_index = (cluster->end - offset) >> PAGE_SHIFT;
3236         while (index <= last_index) {
3237                 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
3238                                 PAGE_SIZE);
3239                 if (ret)
3240                         goto out;
3241 
3242                 page = find_lock_page(inode->i_mapping, index);
3243                 if (!page) {
3244                         page_cache_sync_readahead(inode->i_mapping,
3245                                                   ra, NULL, index,
3246                                                   last_index + 1 - index);
3247                         page = find_or_create_page(inode->i_mapping, index,
3248                                                    mask);
3249                         if (!page) {
3250                                 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3251                                                         PAGE_SIZE, true);
3252                                 ret = -ENOMEM;
3253                                 goto out;
3254                         }
3255                 }
3256 
3257                 if (PageReadahead(page)) {
3258                         page_cache_async_readahead(inode->i_mapping,
3259                                                    ra, NULL, page, index,
3260                                                    last_index + 1 - index);
3261                 }
3262 
3263                 if (!PageUptodate(page)) {
3264                         btrfs_readpage(NULL, page);
3265                         lock_page(page);
3266                         if (!PageUptodate(page)) {
3267                                 unlock_page(page);
3268                                 put_page(page);
3269                                 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3270                                                         PAGE_SIZE, true);
3271                                 btrfs_delalloc_release_extents(BTRFS_I(inode),
3272                                                                PAGE_SIZE, true);
3273                                 ret = -EIO;
3274                                 goto out;
3275                         }
3276                 }
3277 
3278                 page_start = page_offset(page);
3279                 page_end = page_start + PAGE_SIZE - 1;
3280 
3281                 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
3282 
3283                 set_page_extent_mapped(page);
3284 
3285                 if (nr < cluster->nr &&
3286                     page_start + offset == cluster->boundary[nr]) {
3287                         set_extent_bits(&BTRFS_I(inode)->io_tree,
3288                                         page_start, page_end,
3289                                         EXTENT_BOUNDARY);
3290                         nr++;
3291                 }
3292 
3293                 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
3294                                                 NULL, 0);
3295                 if (ret) {
3296                         unlock_page(page);
3297                         put_page(page);
3298                         btrfs_delalloc_release_metadata(BTRFS_I(inode),
3299                                                          PAGE_SIZE, true);
3300                         btrfs_delalloc_release_extents(BTRFS_I(inode),
3301                                                        PAGE_SIZE, true);
3302 
3303                         clear_extent_bits(&BTRFS_I(inode)->io_tree,
3304                                           page_start, page_end,
3305                                           EXTENT_LOCKED | EXTENT_BOUNDARY);
3306                         goto out;
3307 
3308                 }
3309                 set_page_dirty(page);
3310 
3311                 unlock_extent(&BTRFS_I(inode)->io_tree,
3312                               page_start, page_end);
3313                 unlock_page(page);
3314                 put_page(page);
3315 
3316                 index++;
3317                 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE,
3318                                                false);
3319                 balance_dirty_pages_ratelimited(inode->i_mapping);
3320                 btrfs_throttle(fs_info);
3321         }
3322         WARN_ON(nr != cluster->nr);
3323 out:
3324         kfree(ra);
3325         return ret;
3326 }
3327 
3328 static noinline_for_stack
3329 int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
3330                          struct file_extent_cluster *cluster)
3331 {
3332         int ret;
3333 
3334         if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3335                 ret = relocate_file_extent_cluster(inode, cluster);
3336                 if (ret)
3337                         return ret;
3338                 cluster->nr = 0;
3339         }
3340 
3341         if (!cluster->nr)
3342                 cluster->start = extent_key->objectid;
3343         else
3344                 BUG_ON(cluster->nr >= MAX_EXTENTS);
3345         cluster->end = extent_key->objectid + extent_key->offset - 1;
3346         cluster->boundary[cluster->nr] = extent_key->objectid;
3347         cluster->nr++;
3348 
3349         if (cluster->nr >= MAX_EXTENTS) {
3350                 ret = relocate_file_extent_cluster(inode, cluster);
3351                 if (ret)
3352                         return ret;
3353                 cluster->nr = 0;
3354         }
3355         return 0;
3356 }
3357 
3358 /*
3359  * helper to add a tree block to the list.
3360  * the major work is getting the generation and level of the block
3361  */
3362 static int add_tree_block(struct reloc_control *rc,
3363                           struct btrfs_key *extent_key,
3364                           struct btrfs_path *path,
3365                           struct rb_root *blocks)
3366 {
3367         struct extent_buffer *eb;
3368         struct btrfs_extent_item *ei;
3369         struct btrfs_tree_block_info *bi;
3370         struct tree_block *block;
3371         struct rb_node *rb_node;
3372         u32 item_size;
3373         int level = -1;
3374         u64 generation;
3375 
3376         eb =  path->nodes[0];
3377         item_size = btrfs_item_size_nr(eb, path->slots[0]);
3378 
3379         if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
3380             item_size >= sizeof(*ei) + sizeof(*bi)) {
3381                 ei = btrfs_item_ptr(eb, path->slots[0],
3382                                 struct btrfs_extent_item);
3383                 if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
3384                         bi = (struct btrfs_tree_block_info *)(ei + 1);
3385                         level = btrfs_tree_block_level(eb, bi);
3386                 } else {
3387                         level = (int)extent_key->offset;
3388                 }
3389                 generation = btrfs_extent_generation(eb, ei);
3390         } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
3391                 btrfs_print_v0_err(eb->fs_info);
3392                 btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
3393                 return -EINVAL;
3394         } else {
3395                 BUG();
3396         }
3397 
3398         btrfs_release_path(path);
3399 
3400         BUG_ON(level == -1);
3401 
3402         block = kmalloc(sizeof(*block), GFP_NOFS);
3403         if (!block)
3404                 return -ENOMEM;
3405 
3406         block->bytenr = extent_key->objectid;
3407         block->key.objectid = rc->extent_root->fs_info->nodesize;
3408         block->key.offset = generation;
3409         block->level = level;
3410         block->key_ready = 0;
3411 
3412         rb_node = tree_insert(blocks, block->bytenr, &block->rb_node);
3413         if (rb_node)
3414                 backref_tree_panic(rb_node, -EEXIST, block->bytenr);
3415 
3416         return 0;
3417 }
3418 
3419 /*
3420  * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3421  */
3422 static int __add_tree_block(struct reloc_control *rc,
3423                             u64 bytenr, u32 blocksize,
3424                             struct rb_root *blocks)
3425 {
3426         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3427         struct btrfs_path *path;
3428         struct btrfs_key key;
3429         int ret;
3430         bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
3431 
3432         if (tree_block_processed(bytenr, rc))
3433                 return 0;
3434 
3435         if (tree_search(blocks, bytenr))
3436                 return 0;
3437 
3438         path = btrfs_alloc_path();
3439         if (!path)
3440                 return -ENOMEM;
3441 again:
3442         key.objectid = bytenr;
3443         if (skinny) {
3444                 key.type = BTRFS_METADATA_ITEM_KEY;
3445                 key.offset = (u64)-1;
3446         } else {
3447                 key.type = BTRFS_EXTENT_ITEM_KEY;
3448                 key.offset = blocksize;
3449         }
3450 
3451         path->search_commit_root = 1;
3452         path->skip_locking = 1;
3453         ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3454         if (ret < 0)
3455                 goto out;
3456 
3457         if (ret > 0 && skinny) {
3458                 if (path->slots[0]) {
3459                         path->slots[0]--;
3460                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3461                                               path->slots[0]);
3462                         if (key.objectid == bytenr &&
3463                             (key.type == BTRFS_METADATA_ITEM_KEY ||
3464                              (key.type == BTRFS_EXTENT_ITEM_KEY &&
3465                               key.offset == blocksize)))
3466                                 ret = 0;
3467                 }
3468 
3469                 if (ret) {
3470                         skinny = false;
3471                         btrfs_release_path(path);
3472                         goto again;
3473                 }
3474         }
3475         if (ret) {
3476                 ASSERT(ret == 1);
3477                 btrfs_print_leaf(path->nodes[0]);
3478                 btrfs_err(fs_info,
3479              "tree block extent item (%llu) is not found in extent tree",
3480                      bytenr);
3481                 WARN_ON(1);
3482                 ret = -EINVAL;
3483                 goto out;
3484         }
3485 
3486         ret = add_tree_block(rc, &key, path, blocks);
3487 out:
3488         btrfs_free_path(path);
3489         return ret;
3490 }
3491 
3492 /*
3493  * helper to check if the block use full backrefs for pointers in it
3494  */
3495 static int block_use_full_backref(struct reloc_control *rc,
3496                                   struct extent_buffer *eb)
3497 {
3498         u64 flags;
3499         int ret;
3500 
3501         if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) ||
3502             btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV)
3503                 return 1;
3504 
3505         ret = btrfs_lookup_extent_info(NULL, rc->extent_root->fs_info,
3506                                        eb->start, btrfs_header_level(eb), 1,
3507                                        NULL, &flags);
3508         BUG_ON(ret);
3509 
3510         if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3511                 ret = 1;
3512         else
3513                 ret = 0;
3514         return ret;
3515 }
3516 
3517 static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3518                                     struct btrfs_block_group_cache *block_group,
3519                                     struct inode *inode,
3520                                     u64 ino)
3521 {
3522         struct btrfs_key key;
3523         struct btrfs_root *root = fs_info->tree_root;
3524         struct btrfs_trans_handle *trans;
3525         int ret = 0;
3526 
3527         if (inode)
3528                 goto truncate;
3529 
3530         key.objectid = ino;
3531         key.type = BTRFS_INODE_ITEM_KEY;
3532         key.offset = 0;
3533 
3534         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3535         if (IS_ERR(inode))
3536                 return -ENOENT;
3537 
3538 truncate:
3539         ret = btrfs_check_trunc_cache_free_space(fs_info,
3540                                                  &fs_info->global_block_rsv);
3541         if (ret)
3542                 goto out;
3543 
3544         trans = btrfs_join_transaction(root);
3545         if (IS_ERR(trans)) {
3546                 ret = PTR_ERR(trans);
3547                 goto out;
3548         }
3549 
3550         ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
3551 
3552         btrfs_end_transaction(trans);
3553         btrfs_btree_balance_dirty(fs_info);
3554 out:
3555         iput(inode);
3556         return ret;
3557 }
3558 
3559 /*
3560  * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY
3561  * this function scans fs tree to find blocks reference the data extent
3562  */
3563 static int find_data_references(struct reloc_control *rc,
3564                                 struct btrfs_key *extent_key,
3565                                 struct extent_buffer *leaf,
3566                                 struct btrfs_extent_data_ref *ref,
3567                                 struct rb_root *blocks)
3568 {
3569         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3570         struct btrfs_path *path;
3571         struct tree_block *block;
3572         struct btrfs_root *root;
3573         struct btrfs_file_extent_item *fi;
3574         struct rb_node *rb_node;
3575         struct btrfs_key key;
3576         u64 ref_root;
3577         u64 ref_objectid;
3578         u64 ref_offset;
3579         u32 ref_count;
3580         u32 nritems;
3581         int err = 0;
3582         int added = 0;
3583         int counted;
3584         int ret;
3585 
3586         ref_root = btrfs_extent_data_ref_root(leaf, ref);
3587         ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref);
3588         ref_offset = btrfs_extent_data_ref_offset(leaf, ref);
3589         ref_count = btrfs_extent_data_ref_count(leaf, ref);
3590 
3591         /*
3592          * This is an extent belonging to the free space cache, lets just delete
3593          * it and redo the search.
3594          */
3595         if (ref_root == BTRFS_ROOT_TREE_OBJECTID) {
3596                 ret = delete_block_group_cache(fs_info, rc->block_group,
3597                                                NULL, ref_objectid);
3598                 if (ret != -ENOENT)
3599                         return ret;
3600                 ret = 0;
3601         }
3602 
3603         path = btrfs_alloc_path();
3604         if (!path)
3605                 return -ENOMEM;
3606         path->reada = READA_FORWARD;
3607 
3608         root = read_fs_root(fs_info, ref_root);
3609         if (IS_ERR(root)) {
3610                 err = PTR_ERR(root);
3611                 goto out;
3612         }
3613 
3614         key.objectid = ref_objectid;
3615         key.type = BTRFS_EXTENT_DATA_KEY;
3616         if (ref_offset > ((u64)-1 << 32))
3617                 key.offset = 0;
3618         else
3619                 key.offset = ref_offset;
3620 
3621         path->search_commit_root = 1;
3622         path->skip_locking = 1;
3623         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3624         if (ret < 0) {
3625                 err = ret;
3626                 goto out;
3627         }
3628 
3629         leaf = path->nodes[0];
3630         nritems = btrfs_header_nritems(leaf);
3631         /*
3632          * the references in tree blocks that use full backrefs
3633          * are not counted in
3634          */
3635         if (block_use_full_backref(rc, leaf))
3636                 counted = 0;
3637         else
3638                 counted = 1;
3639         rb_node = tree_search(blocks, leaf->start);
3640         if (rb_node) {
3641                 if (counted)
3642                         added = 1;
3643                 else
3644                         path->slots[0] = nritems;
3645         }
3646 
3647         while (ref_count > 0) {
3648                 while (path->slots[0] >= nritems) {
3649                         ret = btrfs_next_leaf(root, path);
3650                         if (ret < 0) {
3651                                 err = ret;
3652                                 goto out;
3653                         }
3654                         if (WARN_ON(ret > 0))
3655                                 goto out;
3656 
3657                         leaf = path->nodes[0];
3658                         nritems = btrfs_header_nritems(leaf);
3659                         added = 0;
3660 
3661                         if (block_use_full_backref(rc, leaf))
3662                                 counted = 0;
3663                         else
3664                                 counted = 1;
3665                         rb_node = tree_search(blocks, leaf->start);
3666                         if (rb_node) {
3667                                 if (counted)
3668                                         added = 1;
3669                                 else
3670                                         path->slots[0] = nritems;
3671                         }
3672                 }
3673 
3674                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3675                 if (WARN_ON(key.objectid != ref_objectid ||
3676                     key.type != BTRFS_EXTENT_DATA_KEY))
3677                         break;
3678 
3679                 fi = btrfs_item_ptr(leaf, path->slots[0],
3680                                     struct btrfs_file_extent_item);
3681 
3682                 if (btrfs_file_extent_type(leaf, fi) ==
3683                     BTRFS_FILE_EXTENT_INLINE)
3684                         goto next;
3685 
3686                 if (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3687                     extent_key->objectid)
3688                         goto next;
3689 
3690                 key.offset -= btrfs_file_extent_offset(leaf, fi);
3691                 if (key.offset != ref_offset)
3692                         goto next;
3693 
3694                 if (counted)
3695                         ref_count--;
3696                 if (added)
3697                         goto next;
3698 
3699                 if (!tree_block_processed(leaf->start, rc)) {
3700                         block = kmalloc(sizeof(*block), GFP_NOFS);
3701                         if (!block) {
3702                                 err = -ENOMEM;
3703                                 break;
3704                         }
3705                         block->bytenr = leaf->start;
3706                         btrfs_item_key_to_cpu(leaf, &block->key, 0);
3707                         block->level = 0;
3708                         block->key_ready = 1;
3709                         rb_node = tree_insert(blocks, block->bytenr,
3710                                               &block->rb_node);
3711                         if (rb_node)
3712                                 backref_tree_panic(rb_node, -EEXIST,
3713                                                    block->bytenr);
3714                 }
3715                 if (counted)
3716                         added = 1;
3717                 else
3718                         path->slots[0] = nritems;
3719 next:
3720                 path->slots[0]++;
3721 
3722         }
3723 out:
3724         btrfs_free_path(path);
3725         return err;
3726 }
3727 
3728 /*
3729  * helper to find all tree blocks that reference a given data extent
3730  */
3731 static noinline_for_stack
3732 int add_data_references(struct reloc_control *rc,
3733                         struct btrfs_key *extent_key,
3734                         struct btrfs_path *path,
3735                         struct rb_root *blocks)
3736 {
3737         struct btrfs_key key;
3738         struct extent_buffer *eb;
3739         struct btrfs_extent_data_ref *dref;
3740         struct btrfs_extent_inline_ref *iref;
3741         unsigned long ptr;
3742         unsigned long end;
3743         u32 blocksize = rc->extent_root->fs_info->nodesize;
3744         int ret = 0;
3745         int err = 0;
3746 
3747         eb = path->nodes[0];
3748         ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
3749         end = ptr + btrfs_item_size_nr(eb, path->slots[0]);
3750         ptr += sizeof(struct btrfs_extent_item);
3751 
3752         while (ptr < end) {
3753                 iref = (struct btrfs_extent_inline_ref *)ptr;
3754                 key.type = btrfs_get_extent_inline_ref_type(eb, iref,
3755                                                         BTRFS_REF_TYPE_DATA);
3756                 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3757                         key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3758                         ret = __add_tree_block(rc, key.offset, blocksize,
3759                                                blocks);
3760                 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3761                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3762                         ret = find_data_references(rc, extent_key,
3763                                                    eb, dref, blocks);
3764                 } else {
3765                         ret = -EUCLEAN;
3766                         btrfs_err(rc->extent_root->fs_info,
3767                      "extent %llu slot %d has an invalid inline ref type",
3768                              eb->start, path->slots[0]);
3769                 }
3770                 if (ret) {
3771                         err = ret;
3772                         goto out;
3773                 }
3774                 ptr += btrfs_extent_inline_ref_size(key.type);
3775         }
3776         WARN_ON(ptr > end);
3777 
3778         while (1) {
3779                 cond_resched();
3780                 eb = path->nodes[0];
3781                 if (path->slots[0] >= btrfs_header_nritems(eb)) {
3782                         ret = btrfs_next_leaf(rc->extent_root, path);
3783                         if (ret < 0) {
3784                                 err = ret;
3785                                 break;
3786                         }
3787                         if (ret > 0)
3788                                 break;
3789                         eb = path->nodes[0];
3790                 }
3791 
3792                 btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
3793                 if (key.objectid != extent_key->objectid)
3794                         break;
3795 
3796                 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3797                         ret = __add_tree_block(rc, key.offset, blocksize,
3798                                                blocks);
3799                 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3800                         dref = btrfs_item_ptr(eb, path->slots[0],
3801                                               struct btrfs_extent_data_ref);
3802                         ret = find_data_references(rc, extent_key,
3803                                                    eb, dref, blocks);
3804                 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
3805                         btrfs_print_v0_err(eb->fs_info);
3806                         btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
3807                         ret = -EINVAL;
3808                 } else {
3809                         ret = 0;
3810                 }
3811                 if (ret) {
3812                         err = ret;
3813                         break;
3814                 }
3815                 path->slots[0]++;
3816         }
3817 out:
3818         btrfs_release_path(path);
3819         if (err)
3820                 free_block_list(blocks);
3821         return err;
3822 }
3823 
3824 /*
3825  * helper to find next unprocessed extent
3826  */
3827 static noinline_for_stack
3828 int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3829                      struct btrfs_key *extent_key)
3830 {
3831         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3832         struct btrfs_key key;
3833         struct extent_buffer *leaf;
3834         u64 start, end, last;
3835         int ret;
3836 
3837         last = rc->block_group->key.objectid + rc->block_group->key.offset;
3838         while (1) {
3839                 cond_resched();
3840                 if (rc->search_start >= last) {
3841                         ret = 1;
3842                         break;
3843                 }
3844 
3845                 key.objectid = rc->search_start;
3846                 key.type = BTRFS_EXTENT_ITEM_KEY;
3847                 key.offset = 0;
3848 
3849                 path->search_commit_root = 1;
3850                 path->skip_locking = 1;
3851                 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3852                                         0, 0);
3853                 if (ret < 0)
3854                         break;
3855 next:
3856                 leaf = path->nodes[0];
3857                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3858                         ret = btrfs_next_leaf(rc->extent_root, path);
3859                         if (ret != 0)
3860                                 break;
3861                         leaf = path->nodes[0];
3862                 }
3863 
3864                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3865                 if (key.objectid >= last) {
3866                         ret = 1;
3867                         break;
3868                 }
3869 
3870                 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3871                     key.type != BTRFS_METADATA_ITEM_KEY) {
3872                         path->slots[0]++;
3873                         goto next;
3874                 }
3875 
3876                 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3877                     key.objectid + key.offset <= rc->search_start) {
3878                         path->slots[0]++;
3879                         goto next;
3880                 }
3881 
3882                 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3883                     key.objectid + fs_info->nodesize <=
3884                     rc->search_start) {
3885                         path->slots[0]++;
3886                         goto next;
3887                 }
3888 
3889                 ret = find_first_extent_bit(&rc->processed_blocks,
3890                                             key.objectid, &start, &end,
3891                                             EXTENT_DIRTY, NULL);
3892 
3893                 if (ret == 0 && start <= key.objectid) {
3894                         btrfs_release_path(path);
3895                         rc->search_start = end + 1;
3896                 } else {
3897                         if (key.type == BTRFS_EXTENT_ITEM_KEY)
3898                                 rc->search_start = key.objectid + key.offset;
3899                         else
3900                                 rc->search_start = key.objectid +
3901                                         fs_info->nodesize;
3902                         memcpy(extent_key, &key, sizeof(key));
3903                         return 0;
3904                 }
3905         }
3906         btrfs_release_path(path);
3907         return ret;
3908 }
3909 
3910 static void set_reloc_control(struct reloc_control *rc)
3911 {
3912         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3913 
3914         mutex_lock(&fs_info->reloc_mutex);
3915         fs_info->reloc_ctl = rc;
3916         mutex_unlock(&fs_info->reloc_mutex);
3917 }
3918 
3919 static void unset_reloc_control(struct reloc_control *rc)
3920 {
3921         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3922 
3923         mutex_lock(&fs_info->reloc_mutex);
3924         fs_info->reloc_ctl = NULL;
3925         mutex_unlock(&fs_info->reloc_mutex);
3926 }
3927 
3928 static int check_extent_flags(u64 flags)
3929 {
3930         if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3931             (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3932                 return 1;
3933         if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3934             !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3935                 return 1;
3936         if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3937             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3938                 return 1;
3939         return 0;
3940 }
3941 
3942 static noinline_for_stack
3943 int prepare_to_relocate(struct reloc_control *rc)
3944 {
3945         struct btrfs_trans_handle *trans;
3946         int ret;
3947 
3948         rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3949                                               BTRFS_BLOCK_RSV_TEMP);
3950         if (!rc->block_rsv)
3951                 return -ENOMEM;
3952 
3953         memset(&rc->cluster, 0, sizeof(rc->cluster));
3954         rc->search_start = rc->block_group->key.objectid;
3955         rc->extents_found = 0;
3956         rc->nodes_relocated = 0;
3957         rc->merging_rsv_size = 0;
3958         rc->reserved_bytes = 0;
3959         rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3960                               RELOCATION_RESERVED_NODES;
3961         ret = btrfs_block_rsv_refill(rc->extent_root,
3962                                      rc->block_rsv, rc->block_rsv->size,
3963                                      BTRFS_RESERVE_FLUSH_ALL);
3964         if (ret)
3965                 return ret;
3966 
3967         rc->create_reloc_tree = 1;
3968         set_reloc_control(rc);
3969 
3970         trans = btrfs_join_transaction(rc->extent_root);
3971         if (IS_ERR(trans)) {
3972                 unset_reloc_control(rc);
3973                 /*
3974                  * extent tree is not a ref_cow tree and has no reloc_root to
3975                  * cleanup.  And callers are responsible to free the above
3976                  * block rsv.
3977                  */
3978                 return PTR_ERR(trans);
3979         }
3980         btrfs_commit_transaction(trans);
3981         return 0;
3982 }
3983 
3984 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3985 {
3986         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3987         struct rb_root blocks = RB_ROOT;
3988         struct btrfs_key key;
3989         struct btrfs_trans_handle *trans = NULL;
3990         struct btrfs_path *path;
3991         struct btrfs_extent_item *ei;
3992         u64 flags;
3993         u32 item_size;
3994         int ret;
3995         int err = 0;
3996         int progress = 0;
3997 
3998         path = btrfs_alloc_path();
3999         if (!path)
4000                 return -ENOMEM;
4001         path->reada = READA_FORWARD;
4002 
4003         ret = prepare_to_relocate(rc);
4004         if (ret) {
4005                 err = ret;
4006                 goto out_free;
4007         }
4008 
4009         while (1) {
4010                 rc->reserved_bytes = 0;
4011                 ret = btrfs_block_rsv_refill(rc->extent_root,
4012                                         rc->block_rsv, rc->block_rsv->size,
4013                                         BTRFS_RESERVE_FLUSH_ALL);
4014                 if (ret) {
4015                         err = ret;
4016                         break;
4017                 }
4018                 progress++;
4019                 trans = btrfs_start_transaction(rc->extent_root, 0);
4020                 if (IS_ERR(trans)) {
4021                         err = PTR_ERR(trans);
4022                         trans = NULL;
4023                         break;
4024                 }
4025 restart:
4026                 if (update_backref_cache(trans, &rc->backref_cache)) {
4027                         btrfs_end_transaction(trans);
4028                         trans = NULL;
4029                         continue;
4030                 }
4031 
4032                 ret = find_next_extent(rc, path, &key);
4033                 if (ret < 0)
4034                         err = ret;
4035                 if (ret != 0)
4036                         break;
4037 
4038                 rc->extents_found++;
4039 
4040                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4041                                     struct btrfs_extent_item);
4042                 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
4043                 if (item_size >= sizeof(*ei)) {
4044                         flags = btrfs_extent_flags(path->nodes[0], ei);
4045                         ret = check_extent_flags(flags);
4046                         BUG_ON(ret);
4047                 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
4048                         err = -EINVAL;
4049                         btrfs_print_v0_err(trans->fs_info);
4050                         btrfs_abort_transaction(trans, err);
4051                         break;
4052                 } else {
4053                         BUG();
4054                 }
4055 
4056                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
4057                         ret = add_tree_block(rc, &key, path, &blocks);
4058                 } else if (rc->stage == UPDATE_DATA_PTRS &&
4059                            (flags & BTRFS_EXTENT_FLAG_DATA)) {
4060                         ret = add_data_references(rc, &key, path, &blocks);
4061                 } else {
4062                         btrfs_release_path(path);
4063                         ret = 0;
4064                 }
4065                 if (ret < 0) {
4066                         err = ret;
4067                         break;
4068                 }
4069 
4070                 if (!RB_EMPTY_ROOT(&blocks)) {
4071                         ret = relocate_tree_blocks(trans, rc, &blocks);
4072                         if (ret < 0) {
4073                                 /*
4074                                  * if we fail to relocate tree blocks, force to update
4075                                  * backref cache when committing transaction.
4076                                  */
4077                                 rc->backref_cache.last_trans = trans->transid - 1;
4078 
4079                                 if (ret != -EAGAIN) {
4080                                         err = ret;
4081                                         break;
4082                                 }
4083                                 rc->extents_found--;
4084                                 rc->search_start = key.objectid;
4085                         }
4086                 }
4087 
4088                 btrfs_end_transaction_throttle(trans);
4089                 btrfs_btree_balance_dirty(fs_info);
4090                 trans = NULL;
4091 
4092                 if (rc->stage == MOVE_DATA_EXTENTS &&
4093                     (flags & BTRFS_EXTENT_FLAG_DATA)) {
4094                         rc->found_file_extent = 1;
4095                         ret = relocate_data_extent(rc->data_inode,
4096                                                    &key, &rc->cluster);
4097                         if (ret < 0) {
4098                                 err = ret;
4099                                 break;
4100                         }
4101                 }
4102         }
4103         if (trans && progress && err == -ENOSPC) {
4104                 ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags);
4105                 if (ret == 1) {
4106                         err = 0;
4107                         progress = 0;
4108                         goto restart;
4109                 }
4110         }
4111 
4112         btrfs_release_path(path);
4113         clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
4114 
4115         if (trans) {
4116                 btrfs_end_transaction_throttle(trans);
4117                 btrfs_btree_balance_dirty(fs_info);
4118         }
4119 
4120         if (!err) {
4121                 ret = relocate_file_extent_cluster(rc->data_inode,
4122                                                    &rc->cluster);
4123                 if (ret < 0)
4124                         err = ret;
4125         }
4126 
4127         rc->create_reloc_tree = 0;
4128         set_reloc_control(rc);
4129 
4130         backref_cache_cleanup(&rc->backref_cache);
4131         btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1);
4132 
4133         err = prepare_to_merge(rc, err);
4134 
4135         merge_reloc_roots(rc);
4136 
4137         rc->merge_reloc_tree = 0;
4138         unset_reloc_control(rc);
4139         btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1);
4140 
4141         /* get rid of pinned extents */
4142         trans = btrfs_join_transaction(rc->extent_root);
4143         if (IS_ERR(trans)) {
4144                 err = PTR_ERR(trans);
4145                 goto out_free;
4146         }
4147         btrfs_commit_transaction(trans);
4148         ret = clean_dirty_subvols(rc);
4149         if (ret < 0 && !err)
4150                 err = ret;
4151 out_free:
4152         btrfs_free_block_rsv(fs_info, rc->block_rsv);
4153         btrfs_free_path(path);
4154         return err;
4155 }
4156 
4157 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4158                                  struct btrfs_root *root, u64 objectid)
4159 {
4160         struct btrfs_path *path;
4161         struct btrfs_inode_item *item;
4162         struct extent_buffer *leaf;
4163         int ret;
4164 
4165         path = btrfs_alloc_path();
4166         if (!path)
4167                 return -ENOMEM;
4168 
4169         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4170         if (ret)
4171                 goto out;
4172 
4173         leaf = path->nodes[0];
4174         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4175         memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
4176         btrfs_set_inode_generation(leaf, item, 1);
4177         btrfs_set_inode_size(leaf, item, 0);
4178         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4179         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
4180                                           BTRFS_INODE_PREALLOC);
4181         btrfs_mark_buffer_dirty(leaf);
4182 out:
4183         btrfs_free_path(path);
4184         return ret;
4185 }
4186 
4187 /*
4188  * helper to create inode for data relocation.
4189  * the inode is in data relocation tree and its link count is 0
4190  */
4191 static noinline_for_stack
4192 struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
4193                                  struct btrfs_block_group_cache *group)
4194 {
4195         struct inode *inode = NULL;
4196         struct btrfs_trans_handle *trans;
4197         struct btrfs_root *root;
4198         struct btrfs_key key;
4199         u64 objectid;
4200         int err = 0;
4201 
4202         root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4203         if (IS_ERR(root))
4204                 return ERR_CAST(root);
4205 
4206         trans = btrfs_start_transaction(root, 6);
4207         if (IS_ERR(trans))
4208                 return ERR_CAST(trans);
4209 
4210         err = btrfs_find_free_objectid(root, &objectid);
4211         if (err)
4212                 goto out;
4213 
4214         err = __insert_orphan_inode(trans, root, objectid);
4215         BUG_ON(err);
4216 
4217         key.objectid = objectid;
4218         key.type = BTRFS_INODE_ITEM_KEY;
4219         key.offset = 0;
4220         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4221         BUG_ON(IS_ERR(inode));
4222         BTRFS_I(inode)->index_cnt = group->key.objectid;
4223 
4224         err = btrfs_orphan_add(trans, BTRFS_I(inode));
4225 out:
4226         btrfs_end_transaction(trans);
4227         btrfs_btree_balance_dirty(fs_info);
4228         if (err) {
4229                 if (inode)
4230                         iput(inode);
4231                 inode = ERR_PTR(err);
4232         }
4233         return inode;
4234 }
4235 
4236 static struct reloc_control *alloc_reloc_control(void)
4237 {
4238         struct reloc_control *rc;
4239 
4240         rc = kzalloc(sizeof(*rc), GFP_NOFS);
4241         if (!rc)
4242                 return NULL;
4243 
4244         INIT_LIST_HEAD(&rc->reloc_roots);
4245         INIT_LIST_HEAD(&rc->dirty_subvol_roots);
4246         backref_cache_init(&rc->backref_cache);
4247         mapping_tree_init(&rc->reloc_root_tree);
4248         extent_io_tree_init(&rc->processed_blocks, NULL);
4249         return rc;
4250 }
4251 
4252 /*
4253  * Print the block group being relocated
4254  */
4255 static void describe_relocation(struct btrfs_fs_info *fs_info,
4256                                 struct btrfs_block_group_cache *block_group)
4257 {
4258         char buf[128] = {'\0'};
4259 
4260         btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf));
4261 
4262         btrfs_info(fs_info,
4263                    "relocating block group %llu flags %s",
4264                    block_group->key.objectid, buf);
4265 }
4266 
4267 /*
4268  * function to relocate all extents in a block group.
4269  */
4270 int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
4271 {
4272         struct btrfs_block_group_cache *bg;
4273         struct btrfs_root *extent_root = fs_info->extent_root;
4274         struct reloc_control *rc;
4275         struct inode *inode;
4276         struct btrfs_path *path;
4277         int ret;
4278         int rw = 0;
4279         int err = 0;
4280 
4281         bg = btrfs_lookup_block_group(fs_info, group_start);
4282         if (!bg)
4283                 return -ENOENT;
4284 
4285         if (btrfs_pinned_by_swapfile(fs_info, bg)) {
4286                 btrfs_put_block_group(bg);
4287                 return -ETXTBSY;
4288         }
4289 
4290         rc = alloc_reloc_control();
4291         if (!rc) {
4292                 btrfs_put_block_group(bg);
4293                 return -ENOMEM;
4294         }
4295 
4296         rc->extent_root = extent_root;
4297         rc->block_group = bg;
4298 
4299         ret = btrfs_inc_block_group_ro(rc->block_group);
4300         if (ret) {
4301                 err = ret;
4302                 goto out;
4303         }
4304         rw = 1;
4305 
4306         path = btrfs_alloc_path();
4307         if (!path) {
4308                 err = -ENOMEM;
4309                 goto out;
4310         }
4311 
4312         inode = lookup_free_space_inode(fs_info, rc->block_group, path);
4313         btrfs_free_path(path);
4314 
4315         if (!IS_ERR(inode))
4316                 ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
4317         else
4318                 ret = PTR_ERR(inode);
4319 
4320         if (ret && ret != -ENOENT) {
4321                 err = ret;
4322                 goto out;
4323         }
4324 
4325         rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
4326         if (IS_ERR(rc->data_inode)) {
4327                 err = PTR_ERR(rc->data_inode);
4328                 rc->data_inode = NULL;
4329                 goto out;
4330         }
4331 
4332         describe_relocation(fs_info, rc->block_group);
4333 
4334         btrfs_wait_block_group_reservations(rc->block_group);
4335         btrfs_wait_nocow_writers(rc->block_group);
4336         btrfs_wait_ordered_roots(fs_info, U64_MAX,
4337                                  rc->block_group->key.objectid,
4338                                  rc->block_group->key.offset);
4339 
4340         while (1) {
4341                 mutex_lock(&fs_info->cleaner_mutex);
4342                 ret = relocate_block_group(rc);
4343                 mutex_unlock(&fs_info->cleaner_mutex);
4344                 if (ret < 0)
4345                         err = ret;
4346 
4347                 /*
4348                  * We may have gotten ENOSPC after we already dirtied some
4349                  * extents.  If writeout happens while we're relocating a
4350                  * different block group we could end up hitting the
4351                  * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in
4352                  * btrfs_reloc_cow_block.  Make sure we write everything out
4353                  * properly so we don't trip over this problem, and then break
4354                  * out of the loop if we hit an error.
4355                  */
4356                 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
4357                         ret = btrfs_wait_ordered_range(rc->data_inode, 0,
4358                                                        (u64)-1);
4359                         if (ret)
4360                                 err = ret;
4361                         invalidate_mapping_pages(rc->data_inode->i_mapping,
4362                                                  0, -1);
4363                         rc->stage = UPDATE_DATA_PTRS;
4364                 }
4365 
4366                 if (err < 0)
4367                         goto out;
4368 
4369                 if (rc->extents_found == 0)
4370                         break;
4371 
4372                 btrfs_info(fs_info, "found %llu extents", rc->extents_found);
4373 
4374         }
4375 
4376         WARN_ON(rc->block_group->pinned > 0);
4377         WARN_ON(rc->block_group->reserved > 0);
4378         WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0);
4379 out:
4380         if (err && rw)
4381                 btrfs_dec_block_group_ro(rc->block_group);
4382         iput(rc->data_inode);
4383         btrfs_put_block_group(rc->block_group);
4384         kfree(rc);
4385         return err;
4386 }
4387 
4388 static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4389 {
4390         struct btrfs_fs_info *fs_info = root->fs_info;
4391         struct btrfs_trans_handle *trans;
4392         int ret, err;
4393 
4394         trans = btrfs_start_transaction(fs_info->tree_root, 0);
4395         if (IS_ERR(trans))
4396                 return PTR_ERR(trans);
4397 
4398         memset(&root->root_item.drop_progress, 0,
4399                 sizeof(root->root_item.drop_progress));
4400         root->root_item.drop_level = 0;
4401         btrfs_set_root_refs(&root->root_item, 0);
4402         ret = btrfs_update_root(trans, fs_info->tree_root,
4403                                 &root->root_key, &root->root_item);
4404 
4405         err = btrfs_end_transaction(trans);
4406         if (err)
4407                 return err;
4408         return ret;
4409 }
4410 
4411 /*
4412  * recover relocation interrupted by system crash.
4413  *
4414  * this function resumes merging reloc trees with corresponding fs trees.
4415  * this is important for keeping the sharing of tree blocks
4416  */
4417 int btrfs_recover_relocation(struct btrfs_root *root)
4418 {
4419         struct btrfs_fs_info *fs_info = root->fs_info;
4420         LIST_HEAD(reloc_roots);
4421         struct btrfs_key key;
4422         struct btrfs_root *fs_root;
4423         struct btrfs_root *reloc_root;
4424         struct btrfs_path *path;
4425         struct extent_buffer *leaf;
4426         struct reloc_control *rc = NULL;
4427         struct btrfs_trans_handle *trans;
4428         int ret;
4429         int err = 0;
4430 
4431         path = btrfs_alloc_path();
4432         if (!path)
4433                 return -ENOMEM;
4434         path->reada = READA_BACK;
4435 
4436         key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4437         key.type = BTRFS_ROOT_ITEM_KEY;
4438         key.offset = (u64)-1;
4439 
4440         while (1) {
4441                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
4442                                         path, 0, 0);
4443                 if (ret < 0) {
4444                         err = ret;
4445                         goto out;
4446                 }
4447                 if (ret > 0) {
4448                         if (path->slots[0] == 0)
4449                                 break;
4450                         path->slots[0]--;
4451                 }
4452                 leaf = path->nodes[0];
4453                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4454                 btrfs_release_path(path);
4455 
4456                 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4457                     key.type != BTRFS_ROOT_ITEM_KEY)
4458                         break;
4459 
4460                 reloc_root = btrfs_read_fs_root(root, &key);
4461                 if (IS_ERR(reloc_root)) {
4462                         err = PTR_ERR(reloc_root);
4463                         goto out;
4464                 }
4465 
4466                 list_add(&reloc_root->root_list, &reloc_roots);
4467 
4468                 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4469                         fs_root = read_fs_root(fs_info,
4470                                                reloc_root->root_key.offset);
4471                         if (IS_ERR(fs_root)) {
4472                                 ret = PTR_ERR(fs_root);
4473                                 if (ret != -ENOENT) {
4474                                         err = ret;
4475                                         goto out;
4476                                 }
4477                                 ret = mark_garbage_root(reloc_root);
4478                                 if (ret < 0) {
4479                                         err = ret;
4480                                         goto out;
4481                                 }
4482                         }
4483                 }
4484 
4485                 if (key.offset == 0)
4486                         break;
4487 
4488                 key.offset--;
4489         }
4490         btrfs_release_path(path);
4491 
4492         if (list_empty(&reloc_roots))
4493                 goto out;
4494 
4495         rc = alloc_reloc_control();
4496         if (!rc) {
4497                 err = -ENOMEM;
4498                 goto out;
4499         }
4500 
4501         rc->extent_root = fs_info->extent_root;
4502 
4503         set_reloc_control(rc);
4504 
4505         trans = btrfs_join_transaction(rc->extent_root);
4506         if (IS_ERR(trans)) {
4507                 unset_reloc_control(rc);
4508                 err = PTR_ERR(trans);
4509                 goto out_free;
4510         }
4511 
4512         rc->merge_reloc_tree = 1;
4513 
4514         while (!list_empty(&reloc_roots)) {
4515                 reloc_root = list_entry(reloc_roots.next,
4516                                         struct btrfs_root, root_list);
4517                 list_del(&reloc_root->root_list);
4518 
4519                 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4520                         list_add_tail(&reloc_root->root_list,
4521                                       &rc->reloc_roots);
4522                         continue;
4523                 }
4524 
4525                 fs_root = read_fs_root(fs_info, reloc_root->root_key.offset);
4526                 if (IS_ERR(fs_root)) {
4527                         err = PTR_ERR(fs_root);
4528                         goto out_free;
4529                 }
4530 
4531                 err = __add_reloc_root(reloc_root);
4532                 BUG_ON(err < 0); /* -ENOMEM or logic error */
4533                 fs_root->reloc_root = reloc_root;
4534         }
4535 
4536         err = btrfs_commit_transaction(trans);
4537         if (err)
4538                 goto out_free;
4539 
4540         merge_reloc_roots(rc);
4541 
4542         unset_reloc_control(rc);
4543 
4544         trans = btrfs_join_transaction(rc->extent_root);
4545         if (IS_ERR(trans)) {
4546                 err = PTR_ERR(trans);
4547                 goto out_free;
4548         }
4549         err = btrfs_commit_transaction(trans);
4550 
4551         ret = clean_dirty_subvols(rc);
4552         if (ret < 0 && !err)
4553                 err = ret;
4554 out_free:
4555         kfree(rc);
4556 out:
4557         if (!list_empty(&reloc_roots))
4558                 free_reloc_roots(&reloc_roots);
4559 
4560         btrfs_free_path(path);
4561 
4562         if (err == 0) {
4563                 /* cleanup orphan inode in data relocation tree */
4564                 fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4565                 if (IS_ERR(fs_root))
4566                         err = PTR_ERR(fs_root);
4567                 else
4568                         err = btrfs_orphan_cleanup(fs_root);
4569         }
4570         return err;
4571 }
4572 
4573 /*
4574  * helper to add ordered checksum for data relocation.
4575  *
4576  * cloning checksum properly handles the nodatasum extents.
4577  * it also saves CPU time to re-calculate the checksum.
4578  */
4579 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
4580 {
4581         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4582         struct btrfs_ordered_sum *sums;
4583         struct btrfs_ordered_extent *ordered;
4584         int ret;
4585         u64 disk_bytenr;
4586         u64 new_bytenr;
4587         LIST_HEAD(list);
4588 
4589         ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4590         BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
4591 
4592         disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
4593         ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr,
4594                                        disk_bytenr + len - 1, &list, 0);
4595         if (ret)
4596                 goto out;
4597 
4598         while (!list_empty(&list)) {
4599                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
4600                 list_del_init(&sums->list);
4601 
4602                 /*
4603                  * We need to offset the new_bytenr based on where the csum is.
4604                  * We need to do this because we will read in entire prealloc
4605                  * extents but we may have written to say the middle of the
4606                  * prealloc extent, so we need to make sure the csum goes with
4607                  * the right disk offset.
4608                  *
4609                  * We can do this because the data reloc inode refers strictly
4610                  * to the on disk bytes, so we don't have to worry about
4611                  * disk_len vs real len like with real inodes since it's all
4612                  * disk length.
4613                  */
4614                 new_bytenr = ordered->start + (sums->bytenr - disk_bytenr);
4615                 sums->bytenr = new_bytenr;
4616 
4617                 btrfs_add_ordered_sum(inode, ordered, sums);
4618         }
4619 out:
4620         btrfs_put_ordered_extent(ordered);
4621         return ret;
4622 }
4623 
4624 int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4625                           struct btrfs_root *root, struct extent_buffer *buf,
4626                           struct extent_buffer *cow)
4627 {
4628         struct btrfs_fs_info *fs_info = root->fs_info;
4629         struct reloc_control *rc;
4630         struct backref_node *node;
4631         int first_cow = 0;
4632         int level;
4633         int ret = 0;
4634 
4635         rc = fs_info->reloc_ctl;
4636         if (!rc)
4637                 return 0;
4638 
4639         BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
4640                root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
4641 
4642         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
4643                 if (buf == root->node)
4644                         __update_reloc_root(root, cow->start);
4645         }
4646 
4647         level = btrfs_header_level(buf);
4648         if (btrfs_header_generation(buf) <=
4649             btrfs_root_last_snapshot(&root->root_item))
4650                 first_cow = 1;
4651 
4652         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4653             rc->create_reloc_tree) {
4654                 WARN_ON(!first_cow && level == 0);
4655 
4656                 node = rc->backref_cache.path[level];
4657                 BUG_ON(node->bytenr != buf->start &&
4658                        node->new_bytenr != buf->start);
4659 
4660                 drop_node_buffer(node);
4661                 extent_buffer_get(cow);
4662                 node->eb = cow;
4663                 node->new_bytenr = cow->start;
4664 
4665                 if (!node->pending) {
4666                         list_move_tail(&node->list,
4667                                        &rc->backref_cache.pending[level]);
4668                         node->pending = 1;
4669                 }
4670 
4671                 if (first_cow)
4672                         __mark_block_processed(rc, node);
4673 
4674                 if (first_cow && level > 0)
4675                         rc->nodes_relocated += buf->len;
4676         }
4677 
4678         if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4679                 ret = replace_file_extents(trans, rc, root, cow);
4680         return ret;
4681 }
4682 
4683 /*
4684  * called before creating snapshot. it calculates metadata reservation
4685  * required for relocating tree blocks in the snapshot
4686  */
4687 void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4688                               u64 *bytes_to_reserve)
4689 {
4690         struct btrfs_root *root = pending->root;
4691         struct reloc_control *rc = root->fs_info->reloc_ctl;
4692 
4693         if (!root->reloc_root || !rc)
4694                 return;
4695 
4696         if (!rc->merge_reloc_tree)
4697                 return;
4698 
4699         root = root->reloc_root;
4700         BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4701         /*
4702          * relocation is in the stage of merging trees. the space
4703          * used by merging a reloc tree is twice the size of
4704          * relocated tree nodes in the worst case. half for cowing
4705          * the reloc tree, half for cowing the fs tree. the space
4706          * used by cowing the reloc tree will be freed after the
4707          * tree is dropped. if we create snapshot, cowing the fs
4708          * tree may use more space than it frees. so we need
4709          * reserve extra space.
4710          */
4711         *bytes_to_reserve += rc->nodes_relocated;
4712 }
4713 
4714 /*
4715  * called after snapshot is created. migrate block reservation
4716  * and create reloc root for the newly created snapshot
4717  */
4718 int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4719                                struct btrfs_pending_snapshot *pending)
4720 {
4721         struct btrfs_root *root = pending->root;
4722         struct btrfs_root *reloc_root;
4723         struct btrfs_root *new_root;
4724         struct reloc_control *rc = root->fs_info->reloc_ctl;
4725         int ret;
4726 
4727         if (!root->reloc_root || !rc)
4728                 return 0;
4729 
4730         rc = root->fs_info->reloc_ctl;
4731         rc->merging_rsv_size += rc->nodes_relocated;
4732 
4733         if (rc->merge_reloc_tree) {
4734                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4735                                               rc->block_rsv,
4736                                               rc->nodes_relocated, true);
4737                 if (ret)
4738                         return ret;
4739         }
4740 
4741         new_root = pending->snap;
4742         reloc_root = create_reloc_root(trans, root->reloc_root,
4743                                        new_root->root_key.objectid);
4744         if (IS_ERR(reloc_root))
4745                 return PTR_ERR(reloc_root);
4746 
4747         ret = __add_reloc_root(reloc_root);
4748         BUG_ON(ret < 0);
4749         new_root->reloc_root = reloc_root;
4750 
4751         if (rc->create_reloc_tree)
4752                 ret = clone_backref_node(trans, rc, root, reloc_root);
4753         return ret;
4754 }
4755 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp