fs/btrfs/ref-verify.c

   1 /*
   2  * Copyright (C) 2014 Facebook.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/sched.h>
  20 #include <linux/stacktrace.h>
  21 #include "ctree.h"
  22 #include "disk-io.h"
  23 #include "locking.h"
  24 #include "delayed-ref.h"
  25 #include "ref-verify.h"
  26
  27 /*
  28  * Used to keep track the roots and number of refs each root has for a given
  29  * bytenr.  This just tracks the number of direct references, no shared
  30  * references.
  31  */
  32 struct root_entry {
  33         u64 root_objectid;
  34         u64 num_refs;
  35         struct rb_node node;
  36 };
  37
  38 /*
  39  * These are meant to represent what should exist in the extent tree, these can
  40  * be used to verify the extent tree is consistent as these should all match
  41  * what the extent tree says.
  42  */
  43 struct ref_entry {
  44         u64 root_objectid;
  45         u64 parent;
  46         u64 owner;
  47         u64 offset;
  48         u64 num_refs;
  49         struct rb_node node;
  50 };
  51
  52 #define MAX_TRACE       16
  53
  54 /*
  55  * Whenever we add/remove a reference we record the action.  The action maps
  56  * back to the delayed ref action.  We hold the ref we are changing in the
  57  * action so we can account for the history properly, and we record the root we
  58  * were called with since it could be different from ref_root.  We also store
  59  * stack traces because thats how I roll.
  60  */
  61 struct ref_action {
  62         int action;
  63         u64 root;
  64         struct ref_entry ref;
  65         struct list_head list;
  66         unsigned long trace[MAX_TRACE];
  67         unsigned int trace_len;
  68 };
  69
  70 /*
  71  * One of these for every block we reference, it holds the roots and references
  72  * to it as well as all of the ref actions that have occured to it.  We never
  73  * free it until we unmount the file system in order to make sure re-allocations
  74  * are happening properly.
  75  */
  76 struct block_entry {
  77         u64 bytenr;
  78         u64 len;
  79         u64 num_refs;
  80         int metadata;
  81         int from_disk;
  82         struct rb_root roots;
  83         struct rb_root refs;
  84         struct rb_node node;
  85         struct list_head actions;
  86 };
  87
  88 static struct block_entry *insert_block_entry(struct rb_root *root,
  89                                               struct block_entry *be)
  90 {
  91         struct rb_node **p = &root->rb_node;
  92         struct rb_node *parent_node = NULL;
  93         struct block_entry *entry;
  94
  95         while (*p) {
  96                 parent_node = *p;
  97                 entry = rb_entry(parent_node, struct block_entry, node);
  98                 if (entry->bytenr > be->bytenr)
  99                         p = &(*p)->rb_left;
 100                 else if (entry->bytenr < be->bytenr)
 101                         p = &(*p)->rb_right;
 102                 else
 103                         return entry;
 104         }
 105
 106         rb_link_node(&be->node, parent_node, p);
 107         rb_insert_color(&be->node, root);
 108         return NULL;
 109 }
 110
 111 static struct block_entry *lookup_block_entry(struct rb_root *root, u64 bytenr)
 112 {
 113         struct rb_node *n;
 114         struct block_entry *entry = NULL;
 115
 116         n = root->rb_node;
 117         while (n) {
 118                 entry = rb_entry(n, struct block_entry, node);
 119                 if (entry->bytenr < bytenr)
 120                         n = n->rb_right;
 121                 else if (entry->bytenr > bytenr)
 122                         n = n->rb_left;
 123                 else
 124                         return entry;
 125         }
 126         return NULL;
 127 }
 128
 129 static struct root_entry *insert_root_entry(struct rb_root *root,
 130                                             struct root_entry *re)
 131 {
 132         struct rb_node **p = &root->rb_node;
 133         struct rb_node *parent_node = NULL;
 134         struct root_entry *entry;
 135
 136         while (*p) {
 137                 parent_node = *p;
 138                 entry = rb_entry(parent_node, struct root_entry, node);
 139                 if (entry->root_objectid > re->root_objectid)
 140                         p = &(*p)->rb_left;
 141                 else if (entry->root_objectid < re->root_objectid)
 142                         p = &(*p)->rb_right;
 143                 else
 144                         return entry;
 145         }
 146
 147         rb_link_node(&re->node, parent_node, p);
 148         rb_insert_color(&re->node, root);
 149         return NULL;
 150
 151 }
 152
 153 static int comp_refs(struct ref_entry *ref1, struct ref_entry *ref2)
 154 {
 155         if (ref1->root_objectid < ref2->root_objectid)
 156                 return -1;
 157         if (ref1->root_objectid > ref2->root_objectid)
 158                 return 1;
 159         if (ref1->parent < ref2->parent)
 160                 return -1;
 161         if (ref1->parent > ref2->parent)
 162                 return 1;
 163         if (ref1->owner < ref2->owner)
 164                 return -1;
 165         if (ref1->owner > ref2->owner)
 166                 return 1;
 167         if (ref1->offset < ref2->offset)
 168                 return -1;
 169         if (ref1->offset > ref2->offset)
 170                 return 1;
 171         return 0;
 172 }
 173
 174 static struct ref_entry *insert_ref_entry(struct rb_root *root,
 175                                           struct ref_entry *ref)
 176 {
 177         struct rb_node **p = &root->rb_node;
 178         struct rb_node *parent_node = NULL;
 179         struct ref_entry *entry;
 180         int cmp;
 181
 182         while (*p) {
 183                 parent_node = *p;
 184                 entry = rb_entry(parent_node, struct ref_entry, node);
 185                 cmp = comp_refs(entry, ref);
 186                 if (cmp > 0)
 187                         p = &(*p)->rb_left;
 188                 else if (cmp < 0)
 189                         p = &(*p)->rb_right;
 190                 else
 191                         return entry;
 192         }
 193
 194         rb_link_node(&ref->node, parent_node, p);
 195         rb_insert_color(&ref->node, root);
 196         return NULL;
 197
 198 }
 199
 200 static struct root_entry *lookup_root_entry(struct rb_root *root, u64 objectid)
 201 {
 202         struct rb_node *n;
 203         struct root_entry *entry = NULL;
 204
 205         n = root->rb_node;
 206         while (n) {
 207                 entry = rb_entry(n, struct root_entry, node);
 208                 if (entry->root_objectid < objectid)
 209                         n = n->rb_right;
 210                 else if (entry->root_objectid > objectid)
 211                         n = n->rb_left;
 212                 else
 213                         return entry;
 214         }
 215         return NULL;
 216 }
 217
 218 #ifdef CONFIG_STACKTRACE
 219 static void __save_stack_trace(struct ref_action *ra)
 220 {
 221         struct stack_trace stack_trace;
 222
 223         stack_trace.max_entries = MAX_TRACE;
 224         stack_trace.nr_entries = 0;
 225         stack_trace.entries = ra->trace;
 226         stack_trace.skip = 2;
 227         save_stack_trace(&stack_trace);
 228         ra->trace_len = stack_trace.nr_entries;
 229 }
 230
 231 static void __print_stack_trace(struct btrfs_fs_info *fs_info,
 232                                 struct ref_action *ra)
 233 {
 234         struct stack_trace trace;
 235
 236         if (ra->trace_len == 0) {
 237                 btrfs_err(fs_info, "  ref-verify: no stacktrace");
 238                 return;
 239         }
 240         trace.nr_entries = ra->trace_len;
 241         trace.entries = ra->trace;
 242         print_stack_trace(&trace, 2);
 243 }
 244 #else
 245 static void inline __save_stack_trace(struct ref_action *ra)
 246 {
 247 }
 248
 249 static void inline __print_stack_trace(struct btrfs_fs_info *fs_info,
 250                                        struct ref_action *ra)
 251 {
 252         btrfs_err(fs_info, "  ref-verify: no stacktrace support");
 253 }
 254 #endif
 255
 256 static void free_block_entry(struct block_entry *be)
 257 {
 258         struct root_entry *re;
 259         struct ref_entry *ref;
 260         struct ref_action *ra;
 261         struct rb_node *n;
 262
 263         while ((n = rb_first(&be->roots))) {
 264                 re = rb_entry(n, struct root_entry, node);
 265                 rb_erase(&re->node, &be->roots);
 266                 kfree(re);
 267         }
 268
 269         while((n = rb_first(&be->refs))) {
 270                 ref = rb_entry(n, struct ref_entry, node);
 271                 rb_erase(&ref->node, &be->refs);
 272                 kfree(ref);
 273         }
 274
 275         while (!list_empty(&be->actions)) {
 276                 ra = list_first_entry(&be->actions, struct ref_action,
 277                                       list);
 278                 list_del(&ra->list);
 279                 kfree(ra);
 280         }
 281         kfree(be);
 282 }
 283
 284 static struct block_entry *add_block_entry(struct btrfs_fs_info *fs_info,
 285                                            u64 bytenr, u64 len,
 286                                            u64 root_objectid)
 287 {
 288         struct block_entry *be = NULL, *exist;
 289         struct root_entry *re = NULL;
 290
 291         re = kzalloc(sizeof(struct root_entry), GFP_KERNEL);
 292         be = kzalloc(sizeof(struct block_entry), GFP_KERNEL);
 293         if (!be || !re) {
 294                 kfree(re);
 295                 kfree(be);
 296                 return ERR_PTR(-ENOMEM);
 297         }
 298         be->bytenr = bytenr;
 299         be->len = len;
 300
 301         re->root_objectid = root_objectid;
 302         re->num_refs = 0;
 303
 304         spin_lock(&fs_info->ref_verify_lock);
 305         exist = insert_block_entry(&fs_info->block_tree, be);
 306         if (exist) {
 307                 if (root_objectid) {
 308                         struct root_entry *exist_re;
 309
 310                         exist_re = insert_root_entry(&exist->roots, re);
 311                         if (exist_re)
 312                                 kfree(re);
 313                 }
 314                 kfree(be);
 315                 return exist;
 316         }
 317
 318         be->num_refs = 0;
 319         be->metadata = 0;
 320         be->from_disk = 0;
 321         be->roots = RB_ROOT;
 322         be->refs = RB_ROOT;
 323         INIT_LIST_HEAD(&be->actions);
 324         if (root_objectid)
 325                 insert_root_entry(&be->roots, re);
 326         else
 327                 kfree(re);
 328         return be;
 329 }
 330
 331 static int add_tree_block(struct btrfs_fs_info *fs_info, u64 ref_root,
 332                           u64 parent, u64 bytenr, int level)
 333 {
 334         struct block_entry *be;
 335         struct root_entry *re;
 336         struct ref_entry *ref = NULL, *exist;
 337
 338         ref = kmalloc(sizeof(struct ref_entry), GFP_KERNEL);
 339         if (!ref)
 340                 return -ENOMEM;
 341
 342         if (parent)
 343                 ref->root_objectid = 0;
 344         else
 345                 ref->root_objectid = ref_root;
 346         ref->parent = parent;
 347         ref->owner = level;
 348         ref->offset = 0;
 349         ref->num_refs = 1;
 350
 351         be = add_block_entry(fs_info, bytenr, fs_info->nodesize, ref_root);
 352         if (IS_ERR(be)) {
 353                 kfree(ref);
 354                 return PTR_ERR(be);
 355         }
 356         be->num_refs++;
 357         be->from_disk = 1;
 358         be->metadata = 1;
 359
 360         if (!parent) {
 361                 ASSERT(ref_root);
 362                 re = lookup_root_entry(&be->roots, ref_root);
 363                 ASSERT(re);
 364                 re->num_refs++;
 365         }
 366         exist = insert_ref_entry(&be->refs, ref);
 367         if (exist) {
 368                 exist->num_refs++;
 369                 kfree(ref);
 370         }
 371         spin_unlock(&fs_info->ref_verify_lock);
 372
 373         return 0;
 374 }
 375
 376 static int add_shared_data_ref(struct btrfs_fs_info *fs_info,
 377                                u64 parent, u32 num_refs, u64 bytenr,
 378                                u64 num_bytes)
 379 {
 380         struct block_entry *be;
 381         struct ref_entry *ref;
 382
 383         ref = kzalloc(sizeof(struct ref_entry), GFP_KERNEL);
 384         if (!ref)
 385                 return -ENOMEM;
 386         be = add_block_entry(fs_info, bytenr, num_bytes, 0);
 387         if (IS_ERR(be)) {
 388                 kfree(ref);
 389                 return PTR_ERR(be);
 390         }
 391         be->num_refs += num_refs;
 392
 393         ref->parent = parent;
 394         ref->num_refs = num_refs;
 395         if (insert_ref_entry(&be->refs, ref)) {
 396                 spin_unlock(&fs_info->ref_verify_lock);
 397                 btrfs_err(fs_info, "existing shared ref when reading from disk?");
 398                 kfree(ref);
 399                 return -EINVAL;
 400         }
 401         spin_unlock(&fs_info->ref_verify_lock);
 402         return 0;
 403 }
 404
 405 static int add_extent_data_ref(struct btrfs_fs_info *fs_info,
 406                                struct extent_buffer *leaf,
 407                                struct btrfs_extent_data_ref *dref,
 408                                u64 bytenr, u64 num_bytes)
 409 {
 410         struct block_entry *be;
 411         struct ref_entry *ref;
 412         struct root_entry *re;
 413         u64 ref_root = btrfs_extent_data_ref_root(leaf, dref);
 414         u64 owner = btrfs_extent_data_ref_objectid(leaf, dref);
 415         u64 offset = btrfs_extent_data_ref_offset(leaf, dref);
 416         u32 num_refs = btrfs_extent_data_ref_count(leaf, dref);
 417
 418         ref = kzalloc(sizeof(struct ref_entry), GFP_KERNEL);
 419         if (!ref)
 420                 return -ENOMEM;
 421         be = add_block_entry(fs_info, bytenr, num_bytes, ref_root);
 422         if (IS_ERR(be)) {
 423                 kfree(ref);
 424                 return PTR_ERR(be);
 425         }
 426         be->num_refs += num_refs;
 427
 428         ref->parent = 0;
 429         ref->owner = owner;
 430         ref->root_objectid = ref_root;
 431         ref->offset = offset;
 432         ref->num_refs = num_refs;
 433         if (insert_ref_entry(&be->refs, ref)) {
 434                 spin_unlock(&fs_info->ref_verify_lock);
 435                 btrfs_err(fs_info, "existing ref when reading from disk?");
 436                 kfree(ref);
 437                 return -EINVAL;
 438         }
 439
 440         re = lookup_root_entry(&be->roots, ref_root);
 441         if (!re) {
 442                 spin_unlock(&fs_info->ref_verify_lock);
 443                 btrfs_err(fs_info, "missing root in new block entry?");
 444                 return -EINVAL;
 445         }
 446         re->num_refs += num_refs;
 447         spin_unlock(&fs_info->ref_verify_lock);
 448         return 0;
 449 }
 450
 451 static int process_extent_item(struct btrfs_fs_info *fs_info,
 452                                struct btrfs_path *path, struct btrfs_key *key,
 453                                int slot, int *tree_block_level)
 454 {
 455         struct btrfs_extent_item *ei;
 456         struct btrfs_extent_inline_ref *iref;
 457         struct btrfs_extent_data_ref *dref;
 458         struct btrfs_shared_data_ref *sref;
 459         struct extent_buffer *leaf = path->nodes[0];
 460         u32 item_size = btrfs_item_size_nr(leaf, slot);
 461         unsigned long end, ptr;
 462         u64 offset, flags, count;
 463         int type, ret;
 464
 465         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
 466         flags = btrfs_extent_flags(leaf, ei);
 467
 468         if ((key->type == BTRFS_EXTENT_ITEM_KEY) &&
 469             flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 470                 struct btrfs_tree_block_info *info;
 471
 472                 info = (struct btrfs_tree_block_info *)(ei + 1);
 473                 *tree_block_level = btrfs_tree_block_level(leaf, info);
 474                 iref = (struct btrfs_extent_inline_ref *)(info + 1);
 475         } else {
 476                 if (key->type == BTRFS_METADATA_ITEM_KEY)
 477                         *tree_block_level = key->offset;
 478                 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
 479         }
 480
 481         ptr = (unsigned long)iref;
 482         end = (unsigned long)ei + item_size;
 483         while (ptr < end) {
 484                 iref = (struct btrfs_extent_inline_ref *)ptr;
 485                 type = btrfs_extent_inline_ref_type(leaf, iref);
 486                 offset = btrfs_extent_inline_ref_offset(leaf, iref);
 487                 switch (type) {
 488                 case BTRFS_TREE_BLOCK_REF_KEY:
 489                         ret = add_tree_block(fs_info, offset, 0, key->objectid,
 490                                              *tree_block_level);
 491                         break;
 492                 case BTRFS_SHARED_BLOCK_REF_KEY:
 493                         ret = add_tree_block(fs_info, 0, offset, key->objectid,
 494                                              *tree_block_level);
 495                         break;
 496                 case BTRFS_EXTENT_DATA_REF_KEY:
 497                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
 498                         ret = add_extent_data_ref(fs_info, leaf, dref,
 499                                                   key->objectid, key->offset);
 500                         break;
 501                 case BTRFS_SHARED_DATA_REF_KEY:
 502                         sref = (struct btrfs_shared_data_ref *)(iref + 1);
 503                         count = btrfs_shared_data_ref_count(leaf, sref);
 504                         ret = add_shared_data_ref(fs_info, offset, count,
 505                                                   key->objectid, key->offset);
 506                         break;
 507                 default:
 508                         btrfs_err(fs_info, "invalid key type in iref");
 509                         ret = -EINVAL;
 510                         break;
 511                 }
 512                 if (ret)
 513                         break;
 514                 ptr += btrfs_extent_inline_ref_size(type);
 515         }
 516         return ret;
 517 }
 518
 519 static int process_leaf(struct btrfs_root *root,
 520                         struct btrfs_path *path, u64 *bytenr, u64 *num_bytes)
 521 {
 522         struct btrfs_fs_info *fs_info = root->fs_info;
 523         struct extent_buffer *leaf = path->nodes[0];
 524         struct btrfs_extent_data_ref *dref;
 525         struct btrfs_shared_data_ref *sref;
 526         u32 count;
 527         int i = 0, tree_block_level = 0, ret;
 528         struct btrfs_key key;
 529         int nritems = btrfs_header_nritems(leaf);
 530
 531         for (i = 0; i < nritems; i++) {
 532                 btrfs_item_key_to_cpu(leaf, &key, i);
 533                 switch (key.type) {
 534                 case BTRFS_EXTENT_ITEM_KEY:
 535                         *num_bytes = key.offset;
 536                 case BTRFS_METADATA_ITEM_KEY:
 537                         *bytenr = key.objectid;
 538                         ret = process_extent_item(fs_info, path, &key, i,
 539                                                   &tree_block_level);
 540                         break;
 541                 case BTRFS_TREE_BLOCK_REF_KEY:
 542                         ret = add_tree_block(fs_info, key.offset, 0,
 543                                              key.objectid, tree_block_level);
 544                         break;
 545                 case BTRFS_SHARED_BLOCK_REF_KEY:
 546                         ret = add_tree_block(fs_info, 0, key.offset,
 547                                              key.objectid, tree_block_level);
 548                         break;
 549                 case BTRFS_EXTENT_DATA_REF_KEY:
 550                         dref = btrfs_item_ptr(leaf, i,
 551                                               struct btrfs_extent_data_ref);
 552                         ret = add_extent_data_ref(fs_info, leaf, dref, *bytenr,
 553                                                   *num_bytes);
 554                         break;
 555                 case BTRFS_SHARED_DATA_REF_KEY:
 556                         sref = btrfs_item_ptr(leaf, i,
 557                                               struct btrfs_shared_data_ref);
 558                         count = btrfs_shared_data_ref_count(leaf, sref);
 559                         ret = add_shared_data_ref(fs_info, key.offset, count,
 560                                                   *bytenr, *num_bytes);
 561                         break;
 562                 default:
 563                         break;
 564                 }
 565                 if (ret)
 566                         break;
 567         }
 568         return ret;
 569 }
 570
 571 /* Walk down to the leaf from the given level */
 572 static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
 573                           int level, u64 *bytenr, u64 *num_bytes)
 574 {
 575         struct btrfs_fs_info *fs_info = root->fs_info;
 576         struct extent_buffer *eb;
 577         u64 block_bytenr, gen;
 578         int ret = 0;
 579
 580         while (level >= 0) {
 581                 if (level) {
 582                         block_bytenr = btrfs_node_blockptr(path->nodes[level],
 583                                                            path->slots[level]);
 584                         gen = btrfs_node_ptr_generation(path->nodes[level],
 585                                                         path->slots[level]);
 586                         eb = read_tree_block(fs_info, block_bytenr, gen);
 587                         if (IS_ERR(eb))
 588                                 return PTR_ERR(eb);
 589                         if (!extent_buffer_uptodate(eb)) {
 590                                 free_extent_buffer(eb);
 591                                 return -EIO;
 592                         }
 593                         btrfs_tree_read_lock(eb);
 594                         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 595                         path->nodes[level-1] = eb;
 596                         path->slots[level-1] = 0;
 597                         path->locks[level-1] = BTRFS_READ_LOCK_BLOCKING;
 598                 } else {
 599                         ret = process_leaf(root, path, bytenr, num_bytes);
 600                         if (ret)
 601                                 break;
 602                 }
 603                 level--;
 604         }
 605         return ret;
 606 }
 607
 608 /* Walk up to the next node that needs to be processed */
 609 static int walk_up_tree(struct btrfs_path *path, int *level)
 610 {
 611         int l;
 612
 613         for (l = 0; l < BTRFS_MAX_LEVEL; l++) {
 614                 if (!path->nodes[l])
 615                         continue;
 616                 if (l) {
 617                         path->slots[l]++;
 618                         if (path->slots[l] <
 619                             btrfs_header_nritems(path->nodes[l])) {
 620                                 *level = l;
 621                                 return 0;
 622                         }
 623                 }
 624                 btrfs_tree_unlock_rw(path->nodes[l], path->locks[l]);
 625                 free_extent_buffer(path->nodes[l]);
 626                 path->nodes[l] = NULL;
 627                 path->slots[l] = 0;
 628                 path->locks[l] = 0;
 629         }
 630
 631         return 1;
 632 }
 633
 634 static void dump_ref_action(struct btrfs_fs_info *fs_info,
 635                             struct ref_action *ra)
 636 {
 637         btrfs_err(fs_info,
 638 "  Ref action %d, root %llu, ref_root %llu, parent %llu, owner %llu, offset %llu, num_refs %llu",
 639                   ra->action, ra->root, ra->ref.root_objectid, ra->ref.parent,
 640                   ra->ref.owner, ra->ref.offset, ra->ref.num_refs);
 641         __print_stack_trace(fs_info, ra);
 642 }
 643
 644 /*
 645  * Dumps all the information from the block entry to printk, it's going to be
 646  * awesome.
 647  */
 648 static void dump_block_entry(struct btrfs_fs_info *fs_info,
 649                              struct block_entry *be)
 650 {
 651         struct ref_entry *ref;
 652         struct root_entry *re;
 653         struct ref_action *ra;
 654         struct rb_node *n;
 655
 656         btrfs_err(fs_info,
 657 "dumping block entry [%llu %llu], num_refs %llu, metadata %d, from disk %d",
 658                   be->bytenr, be->len, be->num_refs, be->metadata,
 659                   be->from_disk);
 660
 661         for (n = rb_first(&be->refs); n; n = rb_next(n)) {
 662                 ref = rb_entry(n, struct ref_entry, node);
 663                 btrfs_err(fs_info,
 664 "  ref root %llu, parent %llu, owner %llu, offset %llu, num_refs %llu",
 665                           ref->root_objectid, ref->parent, ref->owner,
 666                           ref->offset, ref->num_refs);
 667         }
 668
 669         for (n = rb_first(&be->roots); n; n = rb_next(n)) {
 670                 re = rb_entry(n, struct root_entry, node);
 671                 btrfs_err(fs_info, "  root entry %llu, num_refs %llu",
 672                           re->root_objectid, re->num_refs);
 673         }
 674
 675         list_for_each_entry(ra, &be->actions, list)
 676                 dump_ref_action(fs_info, ra);
 677 }
 678
 679 /*
 680  * btrfs_ref_tree_mod: called when we modify a ref for a bytenr
 681  * @root: the root we are making this modification from.
 682  * @bytenr: the bytenr we are modifying.
 683  * @num_bytes: number of bytes.
 684  * @parent: the parent bytenr.
 685  * @ref_root: the original root owner of the bytenr.
 686  * @owner: level in the case of metadata, inode in the case of data.
 687  * @offset: 0 for metadata, file offset for data.
 688  * @action: the action that we are doing, this is the same as the delayed ref
 689  *      action.
 690  *
 691  * This will add an action item to the given bytenr and do sanity checks to make
 692  * sure we haven't messed something up.  If we are making a new allocation and
 693  * this block entry has history we will delete all previous actions as long as
 694  * our sanity checks pass as they are no longer needed.
 695  */
 696 int btrfs_ref_tree_mod(struct btrfs_root *root, u64 bytenr, u64 num_bytes,
 697                        u64 parent, u64 ref_root, u64 owner, u64 offset,
 698                        int action)
 699 {
 700         struct btrfs_fs_info *fs_info = root->fs_info;
 701         struct ref_entry *ref = NULL, *exist;
 702         struct ref_action *ra = NULL;
 703         struct block_entry *be = NULL;
 704         struct root_entry *re = NULL;
 705         int ret = 0;
 706         bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
 707
 708         if (!btrfs_test_opt(root->fs_info, REF_VERIFY))
 709                 return 0;
 710
 711         ref = kzalloc(sizeof(struct ref_entry), GFP_NOFS);
 712         ra = kmalloc(sizeof(struct ref_action), GFP_NOFS);
 713         if (!ra || !ref) {
 714                 kfree(ref);
 715                 kfree(ra);
 716                 ret = -ENOMEM;
 717                 goto out;
 718         }
 719
 720         if (parent) {
 721                 ref->parent = parent;
 722         } else {
 723                 ref->root_objectid = ref_root;
 724                 ref->owner = owner;
 725                 ref->offset = offset;
 726         }
 727         ref->num_refs = (action == BTRFS_DROP_DELAYED_REF) ? -1 : 1;
 728
 729         memcpy(&ra->ref, ref, sizeof(struct ref_entry));
 730         /*
 731          * Save the extra info from the delayed ref in the ref action to make it
 732          * easier to figure out what is happening.  The real ref's we add to the
 733          * ref tree need to reflect what we save on disk so it matches any
 734          * on-disk refs we pre-loaded.
 735          */
 736         ra->ref.owner = owner;
 737         ra->ref.offset = offset;
 738         ra->ref.root_objectid = ref_root;
 739         __save_stack_trace(ra);
 740
 741         INIT_LIST_HEAD(&ra->list);
 742         ra->action = action;
 743         ra->root = root->objectid;
 744
 745         /*
 746          * This is an allocation, preallocate the block_entry in case we haven't
 747          * used it before.
 748          */
 749         ret = -EINVAL;
 750         if (action == BTRFS_ADD_DELAYED_EXTENT) {
 751                 /*
 752                  * For subvol_create we'll just pass in whatever the parent root
 753                  * is and the new root objectid, so let's not treat the passed
 754                  * in root as if it really has a ref for this bytenr.
 755                  */
 756                 be = add_block_entry(root->fs_info, bytenr, num_bytes, ref_root);
 757                 if (IS_ERR(be)) {
 758                         kfree(ra);
 759                         ret = PTR_ERR(be);
 760                         goto out;
 761                 }
 762                 be->num_refs++;
 763                 if (metadata)
 764                         be->metadata = 1;
 765
 766                 if (be->num_refs != 1) {
 767                         btrfs_err(fs_info,
 768                         "re-allocated a block that still has references to it!");
 769                         dump_block_entry(fs_info, be);
 770                         dump_ref_action(fs_info, ra);
 771                         goto out_unlock;
 772                 }
 773
 774                 while (!list_empty(&be->actions)) {
 775                         struct ref_action *tmp;
 776
 777                         tmp = list_first_entry(&be->actions, struct ref_action,
 778                                                list);
 779                         list_del(&tmp->list);
 780                         kfree(tmp);
 781                 }
 782         } else {
 783                 struct root_entry *tmp;
 784
 785                 if (!parent) {
 786                         re = kmalloc(sizeof(struct root_entry), GFP_NOFS);
 787                         if (!re) {
 788                                 kfree(ref);
 789                                 kfree(ra);
 790                                 ret = -ENOMEM;
 791                                 goto out;
 792                         }
 793                         /*
 794                          * This is the root that is modifying us, so it's the
 795                          * one we want to lookup below when we modify the
 796                          * re->num_refs.
 797                          */
 798                         ref_root = root->objectid;
 799                         re->root_objectid = root->objectid;
 800                         re->num_refs = 0;
 801                 }
 802
 803                 spin_lock(&root->fs_info->ref_verify_lock);
 804                 be = lookup_block_entry(&root->fs_info->block_tree, bytenr);
 805                 if (!be) {
 806                         btrfs_err(fs_info,
 807 "trying to do action %d to bytenr %llu num_bytes %llu but there is no existing entry!",
 808                                   action, (unsigned long long)bytenr,
 809                                   (unsigned long long)num_bytes);
 810                         dump_ref_action(fs_info, ra);
 811                         kfree(ref);
 812                         kfree(ra);
 813                         goto out_unlock;
 814                 }
 815
 816                 if (!parent) {
 817                         tmp = insert_root_entry(&be->roots, re);
 818                         if (tmp) {
 819                                 kfree(re);
 820                                 re = tmp;
 821                         }
 822                 }
 823         }
 824
 825         exist = insert_ref_entry(&be->refs, ref);
 826         if (exist) {
 827                 if (action == BTRFS_DROP_DELAYED_REF) {
 828                         if (exist->num_refs == 0) {
 829                                 btrfs_err(fs_info,
 830 "dropping a ref for a existing root that doesn't have a ref on the block");
 831                                 dump_block_entry(fs_info, be);
 832                                 dump_ref_action(fs_info, ra);
 833                                 kfree(ra);
 834                                 goto out_unlock;
 835                         }
 836                         exist->num_refs--;
 837                         if (exist->num_refs == 0) {
 838                                 rb_erase(&exist->node, &be->refs);
 839                                 kfree(exist);
 840                         }
 841                 } else if (!be->metadata) {
 842                         exist->num_refs++;
 843                 } else {
 844                         btrfs_err(fs_info,
 845 "attempting to add another ref for an existing ref on a tree block");
 846                         dump_block_entry(fs_info, be);
 847                         dump_ref_action(fs_info, ra);
 848                         kfree(ra);
 849                         goto out_unlock;
 850                 }
 851                 kfree(ref);
 852         } else {
 853                 if (action == BTRFS_DROP_DELAYED_REF) {
 854                         btrfs_err(fs_info,
 855 "dropping a ref for a root that doesn't have a ref on the block");
 856                         dump_block_entry(fs_info, be);
 857                         dump_ref_action(fs_info, ra);
 858                         kfree(ra);
 859                         goto out_unlock;
 860                 }
 861         }
 862
 863         if (!parent && !re) {
 864                 re = lookup_root_entry(&be->roots, ref_root);
 865                 if (!re) {
 866                         /*
 867                          * This shouldn't happen because we will add our re
 868                          * above when we lookup the be with !parent, but just in
 869                          * case catch this case so we don't panic because I
 870                          * didn't thik of some other corner case.
 871                          */
 872                         btrfs_err(fs_info, "failed to find root %llu for %llu",
 873                                   root->objectid, be->bytenr);
 874                         dump_block_entry(fs_info, be);
 875                         dump_ref_action(fs_info, ra);
 876                         kfree(ra);
 877                         goto out_unlock;
 878                 }
 879         }
 880         if (action == BTRFS_DROP_DELAYED_REF) {
 881                 if (re)
 882                         re->num_refs--;
 883                 be->num_refs--;
 884         } else if (action == BTRFS_ADD_DELAYED_REF) {
 885                 be->num_refs++;
 886                 if (re)
 887                         re->num_refs++;
 888         }
 889         list_add_tail(&ra->list, &be->actions);
 890         ret = 0;
 891 out_unlock:
 892         spin_unlock(&root->fs_info->ref_verify_lock);
 893 out:
 894         if (ret)
 895                 btrfs_clear_opt(fs_info->mount_opt, REF_VERIFY);
 896         return ret;
 897 }
 898
 899 /* Free up the ref cache */
 900 void btrfs_free_ref_cache(struct btrfs_fs_info *fs_info)
 901 {
 902         struct block_entry *be;
 903         struct rb_node *n;
 904
 905         if (!btrfs_test_opt(fs_info, REF_VERIFY))
 906                 return;
 907
 908         spin_lock(&fs_info->ref_verify_lock);
 909         while ((n = rb_first(&fs_info->block_tree))) {
 910                 be = rb_entry(n, struct block_entry, node);
 911                 rb_erase(&be->node, &fs_info->block_tree);
 912                 free_block_entry(be);
 913                 cond_resched_lock(&fs_info->ref_verify_lock);
 914         }
 915         spin_unlock(&fs_info->ref_verify_lock);
 916 }
 917
 918 void btrfs_free_ref_tree_range(struct btrfs_fs_info *fs_info, u64 start,
 919                                u64 len)
 920 {
 921         struct block_entry *be = NULL, *entry;
 922         struct rb_node *n;
 923
 924         if (!btrfs_test_opt(fs_info, REF_VERIFY))
 925                 return;
 926
 927         spin_lock(&fs_info->ref_verify_lock);
 928         n = fs_info->block_tree.rb_node;
 929         while (n) {
 930                 entry = rb_entry(n, struct block_entry, node);
 931                 if (entry->bytenr < start) {
 932                         n = n->rb_right;
 933                 } else if (entry->bytenr > start) {
 934                         n = n->rb_left;
 935                 } else {
 936                         be = entry;
 937                         break;
 938                 }
 939                 /* We want to get as close to start as possible */
 940                 if (be == NULL ||
 941                     (entry->bytenr < start && be->bytenr > start) ||
 942                     (entry->bytenr < start && entry->bytenr > be->bytenr))
 943                         be = entry;
 944         }
 945
 946         /*
 947          * Could have an empty block group, maybe have something to check for
 948          * this case to verify we were actually empty?
 949          */
 950         if (!be) {
 951                 spin_unlock(&fs_info->ref_verify_lock);
 952                 return;
 953         }
 954
 955         n = &be->node;
 956         while (n) {
 957                 be = rb_entry(n, struct block_entry, node);
 958                 n = rb_next(n);
 959                 if (be->bytenr < start && be->bytenr + be->len > start) {
 960                         btrfs_err(fs_info,
 961                                 "block entry overlaps a block group [%llu,%llu]!",
 962                                 start, len);
 963                         dump_block_entry(fs_info, be);
 964                         continue;
 965                 }
 966                 if (be->bytenr < start)
 967                         continue;
 968                 if (be->bytenr >= start + len)
 969                         break;
 970                 if (be->bytenr + be->len > start + len) {
 971                         btrfs_err(fs_info,
 972                                 "block entry overlaps a block group [%llu,%llu]!",
 973                                 start, len);
 974                         dump_block_entry(fs_info, be);
 975                 }
 976                 rb_erase(&be->node, &fs_info->block_tree);
 977                 free_block_entry(be);
 978         }
 979         spin_unlock(&fs_info->ref_verify_lock);
 980 }
 981
 982 /* Walk down all roots and build the ref tree, meant to be called at mount */
 983 int btrfs_build_ref_tree(struct btrfs_fs_info *fs_info)
 984 {
 985         struct btrfs_path *path;
 986         struct extent_buffer *eb;
 987         u64 bytenr = 0, num_bytes = 0;
 988         int ret, level;
 989
 990         if (!btrfs_test_opt(fs_info, REF_VERIFY))
 991                 return 0;
 992
 993         path = btrfs_alloc_path();
 994         if (!path)
 995                 return -ENOMEM;
 996
 997         eb = btrfs_read_lock_root_node(fs_info->extent_root);
 998         btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 999         level = btrfs_header_level(eb);
1000         path->nodes[level] = eb;
1001         path->slots[level] = 0;
1002         path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
1003
1004         while (1) {
1005                 /*
1006                  * We have to keep track of the bytenr/num_bytes we last hit
1007                  * because we could have run out of space for an inline ref, and
1008                  * would have had to added a ref key item which may appear on a
1009                  * different leaf from the original extent item.
1010                  */
1011                 ret = walk_down_tree(fs_info->extent_root, path, level,
1012                                      &bytenr, &num_bytes);
1013                 if (ret)
1014                         break;
1015                 ret = walk_up_tree(path, &level);
1016                 if (ret < 0)
1017                         break;
1018                 if (ret > 0) {
1019                         ret = 0;
1020                         break;
1021                 }
1022         }
1023         if (ret) {
1024                 btrfs_clear_opt(fs_info->mount_opt, REF_VERIFY);
1025                 btrfs_free_ref_cache(fs_info);
1026         }
1027         btrfs_free_path(path);
1028         return ret;
1029 }