linux/fs/btrfs/ref-verify.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 2014 Facebook.  All rights reserved.
   4 */
   5
   6#include <linux/sched.h>
   7#include <linux/stacktrace.h>
   8#include "ctree.h"
   9#include "disk-io.h"
  10#include "locking.h"
  11#include "delayed-ref.h"
  12#include "ref-verify.h"
  13
  14/*
  15 * Used to keep track the roots and number of refs each root has for a given
  16 * bytenr.  This just tracks the number of direct references, no shared
  17 * references.
  18 */
  19struct root_entry {
  20        u64 root_objectid;
  21        u64 num_refs;
  22        struct rb_node node;
  23};
  24
  25/*
  26 * These are meant to represent what should exist in the extent tree, these can
  27 * be used to verify the extent tree is consistent as these should all match
  28 * what the extent tree says.
  29 */
  30struct ref_entry {
  31        u64 root_objectid;
  32        u64 parent;
  33        u64 owner;
  34        u64 offset;
  35        u64 num_refs;
  36        struct rb_node node;
  37};
  38
  39#define MAX_TRACE       16
  40
  41/*
  42 * Whenever we add/remove a reference we record the action.  The action maps
  43 * back to the delayed ref action.  We hold the ref we are changing in the
  44 * action so we can account for the history properly, and we record the root we
  45 * were called with since it could be different from ref_root.  We also store
  46 * stack traces because that's how I roll.
  47 */
  48struct ref_action {
  49        int action;
  50        u64 root;
  51        struct ref_entry ref;
  52        struct list_head list;
  53        unsigned long trace[MAX_TRACE];
  54        unsigned int trace_len;
  55};
  56
  57/*
  58 * One of these for every block we reference, it holds the roots and references
  59 * to it as well as all of the ref actions that have occurred to it.  We never
  60 * free it until we unmount the file system in order to make sure re-allocations
  61 * are happening properly.
  62 */
  63struct block_entry {
  64        u64 bytenr;
  65        u64 len;
  66        u64 num_refs;
  67        int metadata;
  68        int from_disk;
  69        struct rb_root roots;
  70        struct rb_root refs;
  71        struct rb_node node;
  72        struct list_head actions;
  73};
  74
  75static struct block_entry *insert_block_entry(struct rb_root *root,
  76                                              struct block_entry *be)
  77{
  78        struct rb_node **p = &root->rb_node;
  79        struct rb_node *parent_node = NULL;
  80        struct block_entry *entry;
  81
  82        while (*p) {
  83                parent_node = *p;
  84                entry = rb_entry(parent_node, struct block_entry, node);
  85                if (entry->bytenr > be->bytenr)
  86                        p = &(*p)->rb_left;
  87                else if (entry->bytenr < be->bytenr)
  88                        p = &(*p)->rb_right;
  89                else
  90                        return entry;
  91        }
  92
  93        rb_link_node(&be->node, parent_node, p);
  94        rb_insert_color(&be->node, root);
  95        return NULL;
  96}
  97
  98static struct block_entry *lookup_block_entry(struct rb_root *root, u64 bytenr)
  99{
 100        struct rb_node *n;
 101        struct block_entry *entry = NULL;
 102
 103        n = root->rb_node;
 104        while (n) {
 105                entry = rb_entry(n, struct block_entry, node);
 106                if (entry->bytenr < bytenr)
 107                        n = n->rb_right;
 108                else if (entry->bytenr > bytenr)
 109                        n = n->rb_left;
 110                else
 111                        return entry;
 112        }
 113        return NULL;
 114}
 115
 116static struct root_entry *insert_root_entry(struct rb_root *root,
 117                                            struct root_entry *re)
 118{
 119        struct rb_node **p = &root->rb_node;
 120        struct rb_node *parent_node = NULL;
 121        struct root_entry *entry;
 122
 123        while (*p) {
 124                parent_node = *p;
 125                entry = rb_entry(parent_node, struct root_entry, node);
 126                if (entry->root_objectid > re->root_objectid)
 127                        p = &(*p)->rb_left;
 128                else if (entry->root_objectid < re->root_objectid)
 129                        p = &(*p)->rb_right;
 130                else
 131                        return entry;
 132        }
 133
 134        rb_link_node(&re->node, parent_node, p);
 135        rb_insert_color(&re->node, root);
 136        return NULL;
 137
 138}
 139
 140static int comp_refs(struct ref_entry *ref1, struct ref_entry *ref2)
 141{
 142        if (ref1->root_objectid < ref2->root_objectid)
 143                return -1;
 144        if (ref1->root_objectid > ref2->root_objectid)
 145                return 1;
 146        if (ref1->parent < ref2->parent)
 147                return -1;
 148        if (ref1->parent > ref2->parent)
 149                return 1;
 150        if (ref1->owner < ref2->owner)
 151                return -1;
 152        if (ref1->owner > ref2->owner)
 153                return 1;
 154        if (ref1->offset < ref2->offset)
 155                return -1;
 156        if (ref1->offset > ref2->offset)
 157                return 1;
 158        return 0;
 159}
 160
 161static struct ref_entry *insert_ref_entry(struct rb_root *root,
 162                                          struct ref_entry *ref)
 163{
 164        struct rb_node **p = &root->rb_node;
 165        struct rb_node *parent_node = NULL;
 166        struct ref_entry *entry;
 167        int cmp;
 168
 169        while (*p) {
 170                parent_node = *p;
 171                entry = rb_entry(parent_node, struct ref_entry, node);
 172                cmp = comp_refs(entry, ref);
 173                if (cmp > 0)
 174                        p = &(*p)->rb_left;
 175                else if (cmp < 0)
 176                        p = &(*p)->rb_right;
 177                else
 178                        return entry;
 179        }
 180
 181        rb_link_node(&ref->node, parent_node, p);
 182        rb_insert_color(&ref->node, root);
 183        return NULL;
 184
 185}
 186
 187static struct root_entry *lookup_root_entry(struct rb_root *root, u64 objectid)
 188{
 189        struct rb_node *n;
 190        struct root_entry *entry = NULL;
 191
 192        n = root->rb_node;
 193        while (n) {
 194                entry = rb_entry(n, struct root_entry, node);
 195                if (entry->root_objectid < objectid)
 196                        n = n->rb_right;
 197                else if (entry->root_objectid > objectid)
 198                        n = n->rb_left;
 199                else
 200                        return entry;
 201        }
 202        return NULL;
 203}
 204
 205#ifdef CONFIG_STACKTRACE
 206static void __save_stack_trace(struct ref_action *ra)
 207{
 208        ra->trace_len = stack_trace_save(ra->trace, MAX_TRACE, 2);
 209}
 210
 211static void __print_stack_trace(struct btrfs_fs_info *fs_info,
 212                                struct ref_action *ra)
 213{
 214        if (ra->trace_len == 0) {
 215                btrfs_err(fs_info, "  ref-verify: no stacktrace");
 216                return;
 217        }
 218        stack_trace_print(ra->trace, ra->trace_len, 2);
 219}
 220#else
 221static inline void __save_stack_trace(struct ref_action *ra)
 222{
 223}
 224
 225static inline void __print_stack_trace(struct btrfs_fs_info *fs_info,
 226                                       struct ref_action *ra)
 227{
 228        btrfs_err(fs_info, "  ref-verify: no stacktrace support");
 229}
 230#endif
 231
 232static void free_block_entry(struct block_entry *be)
 233{
 234        struct root_entry *re;
 235        struct ref_entry *ref;
 236        struct ref_action *ra;
 237        struct rb_node *n;
 238
 239        while ((n = rb_first(&be->roots))) {
 240                re = rb_entry(n, struct root_entry, node);
 241                rb_erase(&re->node, &be->roots);
 242                kfree(re);
 243        }
 244
 245        while((n = rb_first(&be->refs))) {
 246                ref = rb_entry(n, struct ref_entry, node);
 247                rb_erase(&ref->node, &be->refs);
 248                kfree(ref);
 249        }
 250
 251        while (!list_empty(&be->actions)) {
 252                ra = list_first_entry(&be->actions, struct ref_action,
 253                                      list);
 254                list_del(&ra->list);
 255                kfree(ra);
 256        }
 257        kfree(be);
 258}
 259
 260static struct block_entry *add_block_entry(struct btrfs_fs_info *fs_info,
 261                                           u64 bytenr, u64 len,
 262                                           u64 root_objectid)
 263{
 264        struct block_entry *be = NULL, *exist;
 265        struct root_entry *re = NULL;
 266
 267        re = kzalloc(sizeof(struct root_entry), GFP_NOFS);
 268        be = kzalloc(sizeof(struct block_entry), GFP_NOFS);
 269        if (!be || !re) {
 270                kfree(re);
 271                kfree(be);
 272                return ERR_PTR(-ENOMEM);
 273        }
 274        be->bytenr = bytenr;
 275        be->len = len;
 276
 277        re->root_objectid = root_objectid;
 278        re->num_refs = 0;
 279
 280        spin_lock(&fs_info->ref_verify_lock);
 281        exist = insert_block_entry(&fs_info->block_tree, be);
 282        if (exist) {
 283                if (root_objectid) {
 284                        struct root_entry *exist_re;
 285
 286                        exist_re = insert_root_entry(&exist->roots, re);
 287                        if (exist_re)
 288                                kfree(re);
 289                } else {
 290                        kfree(re);
 291                }
 292                kfree(be);
 293                return exist;
 294        }
 295
 296        be->num_refs = 0;
 297        be->metadata = 0;
 298        be->from_disk = 0;
 299        be->roots = RB_ROOT;
 300        be->refs = RB_ROOT;
 301        INIT_LIST_HEAD(&be->actions);
 302        if (root_objectid)
 303                insert_root_entry(&be->roots, re);
 304        else
 305                kfree(re);
 306        return be;
 307}
 308
 309static int add_tree_block(struct btrfs_fs_info *fs_info, u64 ref_root,
 310                          u64 parent, u64 bytenr, int level)
 311{
 312        struct block_entry *be;
 313        struct root_entry *re;
 314        struct ref_entry *ref = NULL, *exist;
 315
 316        ref = kmalloc(sizeof(struct ref_entry), GFP_NOFS);
 317        if (!ref)
 318                return -ENOMEM;
 319
 320        if (parent)
 321                ref->root_objectid = 0;
 322        else
 323                ref->root_objectid = ref_root;
 324        ref->parent = parent;
 325        ref->owner = level;
 326        ref->offset = 0;
 327        ref->num_refs = 1;
 328
 329        be = add_block_entry(fs_info, bytenr, fs_info->nodesize, ref_root);
 330        if (IS_ERR(be)) {
 331                kfree(ref);
 332                return PTR_ERR(be);
 333        }
 334        be->num_refs++;
 335        be->from_disk = 1;
 336        be->metadata = 1;
 337
 338        if (!parent) {
 339                ASSERT(ref_root);
 340                re = lookup_root_entry(&be->roots, ref_root);
 341                ASSERT(re);
 342                re->num_refs++;
 343        }
 344        exist = insert_ref_entry(&be->refs, ref);
 345        if (exist) {
 346                exist->num_refs++;
 347                kfree(ref);
 348        }
 349        spin_unlock(&fs_info->ref_verify_lock);
 350
 351        return 0;
 352}
 353
 354static int add_shared_data_ref(struct btrfs_fs_info *fs_info,
 355                               u64 parent, u32 num_refs, u64 bytenr,
 356                               u64 num_bytes)
 357{
 358        struct block_entry *be;
 359        struct ref_entry *ref;
 360
 361        ref = kzalloc(sizeof(struct ref_entry), GFP_NOFS);
 362        if (!ref)
 363                return -ENOMEM;
 364        be = add_block_entry(fs_info, bytenr, num_bytes, 0);
 365        if (IS_ERR(be)) {
 366                kfree(ref);
 367                return PTR_ERR(be);
 368        }
 369        be->num_refs += num_refs;
 370
 371        ref->parent = parent;
 372        ref->num_refs = num_refs;
 373        if (insert_ref_entry(&be->refs, ref)) {
 374                spin_unlock(&fs_info->ref_verify_lock);
 375                btrfs_err(fs_info, "existing shared ref when reading from disk?");
 376                kfree(ref);
 377                return -EINVAL;
 378        }
 379        spin_unlock(&fs_info->ref_verify_lock);
 380        return 0;
 381}
 382
 383static int add_extent_data_ref(struct btrfs_fs_info *fs_info,
 384                               struct extent_buffer *leaf,
 385                               struct btrfs_extent_data_ref *dref,
 386                               u64 bytenr, u64 num_bytes)
 387{
 388        struct block_entry *be;
 389        struct ref_entry *ref;
 390        struct root_entry *re;
 391        u64 ref_root = btrfs_extent_data_ref_root(leaf, dref);
 392        u64 owner = btrfs_extent_data_ref_objectid(leaf, dref);
 393        u64 offset = btrfs_extent_data_ref_offset(leaf, dref);
 394        u32 num_refs = btrfs_extent_data_ref_count(leaf, dref);
 395
 396        ref = kzalloc(sizeof(struct ref_entry), GFP_NOFS);
 397        if (!ref)
 398                return -ENOMEM;
 399        be = add_block_entry(fs_info, bytenr, num_bytes, ref_root);
 400        if (IS_ERR(be)) {
 401                kfree(ref);
 402                return PTR_ERR(be);
 403        }
 404        be->num_refs += num_refs;
 405
 406        ref->parent = 0;
 407        ref->owner = owner;
 408        ref->root_objectid = ref_root;
 409        ref->offset = offset;
 410        ref->num_refs = num_refs;
 411        if (insert_ref_entry(&be->refs, ref)) {
 412                spin_unlock(&fs_info->ref_verify_lock);
 413                btrfs_err(fs_info, "existing ref when reading from disk?");
 414                kfree(ref);
 415                return -EINVAL;
 416        }
 417
 418        re = lookup_root_entry(&be->roots, ref_root);
 419        if (!re) {
 420                spin_unlock(&fs_info->ref_verify_lock);
 421                btrfs_err(fs_info, "missing root in new block entry?");
 422                return -EINVAL;
 423        }
 424        re->num_refs += num_refs;
 425        spin_unlock(&fs_info->ref_verify_lock);
 426        return 0;
 427}
 428
 429static int process_extent_item(struct btrfs_fs_info *fs_info,
 430                               struct btrfs_path *path, struct btrfs_key *key,
 431                               int slot, int *tree_block_level)
 432{
 433        struct btrfs_extent_item *ei;
 434        struct btrfs_extent_inline_ref *iref;
 435        struct btrfs_extent_data_ref *dref;
 436        struct btrfs_shared_data_ref *sref;
 437        struct extent_buffer *leaf = path->nodes[0];
 438        u32 item_size = btrfs_item_size(leaf, slot);
 439        unsigned long end, ptr;
 440        u64 offset, flags, count;
 441        int type, ret;
 442
 443        ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
 444        flags = btrfs_extent_flags(leaf, ei);
 445
 446        if ((key->type == BTRFS_EXTENT_ITEM_KEY) &&
 447            flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 448                struct btrfs_tree_block_info *info;
 449
 450                info = (struct btrfs_tree_block_info *)(ei + 1);
 451                *tree_block_level = btrfs_tree_block_level(leaf, info);
 452                iref = (struct btrfs_extent_inline_ref *)(info + 1);
 453        } else {
 454                if (key->type == BTRFS_METADATA_ITEM_KEY)
 455                        *tree_block_level = key->offset;
 456                iref = (struct btrfs_extent_inline_ref *)(ei + 1);
 457        }
 458
 459        ptr = (unsigned long)iref;
 460        end = (unsigned long)ei + item_size;
 461        while (ptr < end) {
 462                iref = (struct btrfs_extent_inline_ref *)ptr;
 463                type = btrfs_extent_inline_ref_type(leaf, iref);
 464                offset = btrfs_extent_inline_ref_offset(leaf, iref);
 465                switch (type) {
 466                case BTRFS_TREE_BLOCK_REF_KEY:
 467                        ret = add_tree_block(fs_info, offset, 0, key->objectid,
 468                                             *tree_block_level);
 469                        break;
 470                case BTRFS_SHARED_BLOCK_REF_KEY:
 471                        ret = add_tree_block(fs_info, 0, offset, key->objectid,
 472                                             *tree_block_level);
 473                        break;
 474                case BTRFS_EXTENT_DATA_REF_KEY:
 475                        dref = (struct btrfs_extent_data_ref *)(&iref->offset);
 476                        ret = add_extent_data_ref(fs_info, leaf, dref,
 477                                                  key->objectid, key->offset);
 478                        break;
 479                case BTRFS_SHARED_DATA_REF_KEY:
 480                        sref = (struct btrfs_shared_data_ref *)(iref + 1);
 481                        count = btrfs_shared_data_ref_count(leaf, sref);
 482                        ret = add_shared_data_ref(fs_info, offset, count,
 483                                                  key->objectid, key->offset);
 484                        break;
 485                default:
 486                        btrfs_err(fs_info, "invalid key type in iref");
 487                        ret = -EINVAL;
 488                        break;
 489                }
 490                if (ret)
 491                        break;
 492                ptr += btrfs_extent_inline_ref_size(type);
 493        }
 494        return ret;
 495}
 496
 497static int process_leaf(struct btrfs_root *root,
 498                        struct btrfs_path *path, u64 *bytenr, u64 *num_bytes,
 499                        int *tree_block_level)
 500{
 501        struct btrfs_fs_info *fs_info = root->fs_info;
 502        struct extent_buffer *leaf = path->nodes[0];
 503        struct btrfs_extent_data_ref *dref;
 504        struct btrfs_shared_data_ref *sref;
 505        u32 count;
 506        int i = 0, ret = 0;
 507        struct btrfs_key key;
 508        int nritems = btrfs_header_nritems(leaf);
 509
 510        for (i = 0; i < nritems; i++) {
 511                btrfs_item_key_to_cpu(leaf, &key, i);
 512                switch (key.type) {
 513                case BTRFS_EXTENT_ITEM_KEY:
 514                        *num_bytes = key.offset;
 515                        fallthrough;
 516                case BTRFS_METADATA_ITEM_KEY:
 517                        *bytenr = key.objectid;
 518                        ret = process_extent_item(fs_info, path, &key, i,
 519                                                  tree_block_level);
 520                        break;
 521                case BTRFS_TREE_BLOCK_REF_KEY:
 522                        ret = add_tree_block(fs_info, key.offset, 0,
 523                                             key.objectid, *tree_block_level);
 524                        break;
 525                case BTRFS_SHARED_BLOCK_REF_KEY:
 526                        ret = add_tree_block(fs_info, 0, key.offset,
 527                                             key.objectid, *tree_block_level);
 528                        break;
 529                case BTRFS_EXTENT_DATA_REF_KEY:
 530                        dref = btrfs_item_ptr(leaf, i,
 531                                              struct btrfs_extent_data_ref);
 532                        ret = add_extent_data_ref(fs_info, leaf, dref, *bytenr,
 533                                                  *num_bytes);
 534                        break;
 535                case BTRFS_SHARED_DATA_REF_KEY:
 536                        sref = btrfs_item_ptr(leaf, i,
 537                                              struct btrfs_shared_data_ref);
 538                        count = btrfs_shared_data_ref_count(leaf, sref);
 539                        ret = add_shared_data_ref(fs_info, key.offset, count,
 540                                                  *bytenr, *num_bytes);
 541                        break;
 542                default:
 543                        break;
 544                }
 545                if (ret)
 546                        break;
 547        }
 548        return ret;
 549}
 550
 551/* Walk down to the leaf from the given level */
 552static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
 553                          int level, u64 *bytenr, u64 *num_bytes,
 554                          int *tree_block_level)
 555{
 556        struct extent_buffer *eb;
 557        int ret = 0;
 558
 559        while (level >= 0) {
 560                if (level) {
 561                        eb = btrfs_read_node_slot(path->nodes[level],
 562                                                  path->slots[level]);
 563                        if (IS_ERR(eb))
 564                                return PTR_ERR(eb);
 565                        btrfs_tree_read_lock(eb);
 566                        path->nodes[level-1] = eb;
 567                        path->slots[level-1] = 0;
 568                        path->locks[level-1] = BTRFS_READ_LOCK;
 569                } else {
 570                        ret = process_leaf(root, path, bytenr, num_bytes,
 571                                           tree_block_level);
 572                        if (ret)
 573                                break;
 574                }
 575                level--;
 576        }
 577        return ret;
 578}
 579
 580/* Walk up to the next node that needs to be processed */
 581static int walk_up_tree(struct btrfs_path *path, int *level)
 582{
 583        int l;
 584
 585        for (l = 0; l < BTRFS_MAX_LEVEL; l++) {
 586                if (!path->nodes[l])
 587                        continue;
 588                if (l) {
 589                        path->slots[l]++;
 590                        if (path->slots[l] <
 591                            btrfs_header_nritems(path->nodes[l])) {
 592                                *level = l;
 593                                return 0;
 594                        }
 595                }
 596                btrfs_tree_unlock_rw(path->nodes[l], path->locks[l]);
 597                free_extent_buffer(path->nodes[l]);
 598                path->nodes[l] = NULL;
 599                path->slots[l] = 0;
 600                path->locks[l] = 0;
 601        }
 602
 603        return 1;
 604}
 605
 606static void dump_ref_action(struct btrfs_fs_info *fs_info,
 607                            struct ref_action *ra)
 608{
 609        btrfs_err(fs_info,
 610"  Ref action %d, root %llu, ref_root %llu, parent %llu, owner %llu, offset %llu, num_refs %llu",
 611                  ra->action, ra->root, ra->ref.root_objectid, ra->ref.parent,
 612                  ra->ref.owner, ra->ref.offset, ra->ref.num_refs);
 613        __print_stack_trace(fs_info, ra);
 614}
 615
 616/*
 617 * Dumps all the information from the block entry to printk, it's going to be
 618 * awesome.
 619 */
 620static void dump_block_entry(struct btrfs_fs_info *fs_info,
 621                             struct block_entry *be)
 622{
 623        struct ref_entry *ref;
 624        struct root_entry *re;
 625        struct ref_action *ra;
 626        struct rb_node *n;
 627
 628        btrfs_err(fs_info,
 629"dumping block entry [%llu %llu], num_refs %llu, metadata %d, from disk %d",
 630                  be->bytenr, be->len, be->num_refs, be->metadata,
 631                  be->from_disk);
 632
 633        for (n = rb_first(&be->refs); n; n = rb_next(n)) {
 634                ref = rb_entry(n, struct ref_entry, node);
 635                btrfs_err(fs_info,
 636"  ref root %llu, parent %llu, owner %llu, offset %llu, num_refs %llu",
 637                          ref->root_objectid, ref->parent, ref->owner,
 638                          ref->offset, ref->num_refs);
 639        }
 640
 641        for (n = rb_first(&be->roots); n; n = rb_next(n)) {
 642                re = rb_entry(n, struct root_entry, node);
 643                btrfs_err(fs_info, "  root entry %llu, num_refs %llu",
 644                          re->root_objectid, re->num_refs);
 645        }
 646
 647        list_for_each_entry(ra, &be->actions, list)
 648                dump_ref_action(fs_info, ra);
 649}
 650
 651/*
 652 * btrfs_ref_tree_mod: called when we modify a ref for a bytenr
 653 *
 654 * This will add an action item to the given bytenr and do sanity checks to make
 655 * sure we haven't messed something up.  If we are making a new allocation and
 656 * this block entry has history we will delete all previous actions as long as
 657 * our sanity checks pass as they are no longer needed.
 658 */
 659int btrfs_ref_tree_mod(struct btrfs_fs_info *fs_info,
 660                       struct btrfs_ref *generic_ref)
 661{
 662        struct ref_entry *ref = NULL, *exist;
 663        struct ref_action *ra = NULL;
 664        struct block_entry *be = NULL;
 665        struct root_entry *re = NULL;
 666        int action = generic_ref->action;
 667        int ret = 0;
 668        bool metadata;
 669        u64 bytenr = generic_ref->bytenr;
 670        u64 num_bytes = generic_ref->len;
 671        u64 parent = generic_ref->parent;
 672        u64 ref_root = 0;
 673        u64 owner = 0;
 674        u64 offset = 0;
 675
 676        if (!btrfs_test_opt(fs_info, REF_VERIFY))
 677                return 0;
 678
 679        if (generic_ref->type == BTRFS_REF_METADATA) {
 680                if (!parent)
 681                        ref_root = generic_ref->tree_ref.owning_root;
 682                owner = generic_ref->tree_ref.level;
 683        } else if (!parent) {
 684                ref_root = generic_ref->data_ref.owning_root;
 685                owner = generic_ref->data_ref.ino;
 686                offset = generic_ref->data_ref.offset;
 687        }
 688        metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
 689
 690        ref = kzalloc(sizeof(struct ref_entry), GFP_NOFS);
 691        ra = kmalloc(sizeof(struct ref_action), GFP_NOFS);
 692        if (!ra || !ref) {
 693                kfree(ref);
 694                kfree(ra);
 695                ret = -ENOMEM;
 696                goto out;
 697        }
 698
 699        ref->parent = parent;
 700        ref->owner = owner;
 701        ref->root_objectid = ref_root;
 702        ref->offset = offset;
 703        ref->num_refs = (action == BTRFS_DROP_DELAYED_REF) ? -1 : 1;
 704
 705        memcpy(&ra->ref, ref, sizeof(struct ref_entry));
 706        /*
 707         * Save the extra info from the delayed ref in the ref action to make it
 708         * easier to figure out what is happening.  The real ref's we add to the
 709         * ref tree need to reflect what we save on disk so it matches any
 710         * on-disk refs we pre-loaded.
 711         */
 712        ra->ref.owner = owner;
 713        ra->ref.offset = offset;
 714        ra->ref.root_objectid = ref_root;
 715        __save_stack_trace(ra);
 716
 717        INIT_LIST_HEAD(&ra->list);
 718        ra->action = action;
 719        ra->root = generic_ref->real_root;
 720
 721        /*
 722         * This is an allocation, preallocate the block_entry in case we haven't
 723         * used it before.
 724         */
 725        ret = -EINVAL;
 726        if (action == BTRFS_ADD_DELAYED_EXTENT) {
 727                /*
 728                 * For subvol_create we'll just pass in whatever the parent root
 729                 * is and the new root objectid, so let's not treat the passed
 730                 * in root as if it really has a ref for this bytenr.
 731                 */
 732                be = add_block_entry(fs_info, bytenr, num_bytes, ref_root);
 733                if (IS_ERR(be)) {
 734                        kfree(ref);
 735                        kfree(ra);
 736                        ret = PTR_ERR(be);
 737                        goto out;
 738                }
 739                be->num_refs++;
 740                if (metadata)
 741                        be->metadata = 1;
 742
 743                if (be->num_refs != 1) {
 744                        btrfs_err(fs_info,
 745                        "re-allocated a block that still has references to it!");
 746                        dump_block_entry(fs_info, be);
 747                        dump_ref_action(fs_info, ra);
 748                        kfree(ref);
 749                        kfree(ra);
 750                        goto out_unlock;
 751                }
 752
 753                while (!list_empty(&be->actions)) {
 754                        struct ref_action *tmp;
 755
 756                        tmp = list_first_entry(&be->actions, struct ref_action,
 757                                               list);
 758                        list_del(&tmp->list);
 759                        kfree(tmp);
 760                }
 761        } else {
 762                struct root_entry *tmp;
 763
 764                if (!parent) {
 765                        re = kmalloc(sizeof(struct root_entry), GFP_NOFS);
 766                        if (!re) {
 767                                kfree(ref);
 768                                kfree(ra);
 769                                ret = -ENOMEM;
 770                                goto out;
 771                        }
 772                        /*
 773                         * This is the root that is modifying us, so it's the
 774                         * one we want to lookup below when we modify the
 775                         * re->num_refs.
 776                         */
 777                        ref_root = generic_ref->real_root;
 778                        re->root_objectid = generic_ref->real_root;
 779                        re->num_refs = 0;
 780                }
 781
 782                spin_lock(&fs_info->ref_verify_lock);
 783                be = lookup_block_entry(&fs_info->block_tree, bytenr);
 784                if (!be) {
 785                        btrfs_err(fs_info,
 786"trying to do action %d to bytenr %llu num_bytes %llu but there is no existing entry!",
 787                                  action, bytenr, num_bytes);
 788                        dump_ref_action(fs_info, ra);
 789                        kfree(ref);
 790                        kfree(ra);
 791                        goto out_unlock;
 792                } else if (be->num_refs == 0) {
 793                        btrfs_err(fs_info,
 794                "trying to do action %d for a bytenr that has 0 total references",
 795                                action);
 796                        dump_block_entry(fs_info, be);
 797                        dump_ref_action(fs_info, ra);
 798                        kfree(ref);
 799                        kfree(ra);
 800                        goto out_unlock;
 801                }
 802
 803                if (!parent) {
 804                        tmp = insert_root_entry(&be->roots, re);
 805                        if (tmp) {
 806                                kfree(re);
 807                                re = tmp;
 808                        }
 809                }
 810        }
 811
 812        exist = insert_ref_entry(&be->refs, ref);
 813        if (exist) {
 814                if (action == BTRFS_DROP_DELAYED_REF) {
 815                        if (exist->num_refs == 0) {
 816                                btrfs_err(fs_info,
 817"dropping a ref for a existing root that doesn't have a ref on the block");
 818                                dump_block_entry(fs_info, be);
 819                                dump_ref_action(fs_info, ra);
 820                                kfree(ref);
 821                                kfree(ra);
 822                                goto out_unlock;
 823                        }
 824                        exist->num_refs--;
 825                        if (exist->num_refs == 0) {
 826                                rb_erase(&exist->node, &be->refs);
 827                                kfree(exist);
 828                        }
 829                } else if (!be->metadata) {
 830                        exist->num_refs++;
 831                } else {
 832                        btrfs_err(fs_info,
 833"attempting to add another ref for an existing ref on a tree block");
 834                        dump_block_entry(fs_info, be);
 835                        dump_ref_action(fs_info, ra);
 836                        kfree(ref);
 837                        kfree(ra);
 838                        goto out_unlock;
 839                }
 840                kfree(ref);
 841        } else {
 842                if (action == BTRFS_DROP_DELAYED_REF) {
 843                        btrfs_err(fs_info,
 844"dropping a ref for a root that doesn't have a ref on the block");
 845                        dump_block_entry(fs_info, be);
 846                        dump_ref_action(fs_info, ra);
 847                        kfree(ref);
 848                        kfree(ra);
 849                        goto out_unlock;
 850                }
 851        }
 852
 853        if (!parent && !re) {
 854                re = lookup_root_entry(&be->roots, ref_root);
 855                if (!re) {
 856                        /*
 857                         * This shouldn't happen because we will add our re
 858                         * above when we lookup the be with !parent, but just in
 859                         * case catch this case so we don't panic because I
 860                         * didn't think of some other corner case.
 861                         */
 862                        btrfs_err(fs_info, "failed to find root %llu for %llu",
 863                                  generic_ref->real_root, be->bytenr);
 864                        dump_block_entry(fs_info, be);
 865                        dump_ref_action(fs_info, ra);
 866                        kfree(ra);
 867                        goto out_unlock;
 868                }
 869        }
 870        if (action == BTRFS_DROP_DELAYED_REF) {
 871                if (re)
 872                        re->num_refs--;
 873                be->num_refs--;
 874        } else if (action == BTRFS_ADD_DELAYED_REF) {
 875                be->num_refs++;
 876                if (re)
 877                        re->num_refs++;
 878        }
 879        list_add_tail(&ra->list, &be->actions);
 880        ret = 0;
 881out_unlock:
 882        spin_unlock(&fs_info->ref_verify_lock);
 883out:
 884        if (ret)
 885                btrfs_clear_opt(fs_info->mount_opt, REF_VERIFY);
 886        return ret;
 887}
 888
 889/* Free up the ref cache */
 890void btrfs_free_ref_cache(struct btrfs_fs_info *fs_info)
 891{
 892        struct block_entry *be;
 893        struct rb_node *n;
 894
 895        if (!btrfs_test_opt(fs_info, REF_VERIFY))
 896                return;
 897
 898        spin_lock(&fs_info->ref_verify_lock);
 899        while ((n = rb_first(&fs_info->block_tree))) {
 900                be = rb_entry(n, struct block_entry, node);
 901                rb_erase(&be->node, &fs_info->block_tree);
 902                free_block_entry(be);
 903                cond_resched_lock(&fs_info->ref_verify_lock);
 904        }
 905        spin_unlock(&fs_info->ref_verify_lock);
 906}
 907
 908void btrfs_free_ref_tree_range(struct btrfs_fs_info *fs_info, u64 start,
 909                               u64 len)
 910{
 911        struct block_entry *be = NULL, *entry;
 912        struct rb_node *n;
 913
 914        if (!btrfs_test_opt(fs_info, REF_VERIFY))
 915                return;
 916
 917        spin_lock(&fs_info->ref_verify_lock);
 918        n = fs_info->block_tree.rb_node;
 919        while (n) {
 920                entry = rb_entry(n, struct block_entry, node);
 921                if (entry->bytenr < start) {
 922                        n = n->rb_right;
 923                } else if (entry->bytenr > start) {
 924                        n = n->rb_left;
 925                } else {
 926                        be = entry;
 927                        break;
 928                }
 929                /* We want to get as close to start as possible */
 930                if (be == NULL ||
 931                    (entry->bytenr < start && be->bytenr > start) ||
 932                    (entry->bytenr < start && entry->bytenr > be->bytenr))
 933                        be = entry;
 934        }
 935
 936        /*
 937         * Could have an empty block group, maybe have something to check for
 938         * this case to verify we were actually empty?
 939         */
 940        if (!be) {
 941                spin_unlock(&fs_info->ref_verify_lock);
 942                return;
 943        }
 944
 945        n = &be->node;
 946        while (n) {
 947                be = rb_entry(n, struct block_entry, node);
 948                n = rb_next(n);
 949                if (be->bytenr < start && be->bytenr + be->len > start) {
 950                        btrfs_err(fs_info,
 951                                "block entry overlaps a block group [%llu,%llu]!",
 952                                start, len);
 953                        dump_block_entry(fs_info, be);
 954                        continue;
 955                }
 956                if (be->bytenr < start)
 957                        continue;
 958                if (be->bytenr >= start + len)
 959                        break;
 960                if (be->bytenr + be->len > start + len) {
 961                        btrfs_err(fs_info,
 962                                "block entry overlaps a block group [%llu,%llu]!",
 963                                start, len);
 964                        dump_block_entry(fs_info, be);
 965                }
 966                rb_erase(&be->node, &fs_info->block_tree);
 967                free_block_entry(be);
 968        }
 969        spin_unlock(&fs_info->ref_verify_lock);
 970}
 971
 972/* Walk down all roots and build the ref tree, meant to be called at mount */
 973int btrfs_build_ref_tree(struct btrfs_fs_info *fs_info)
 974{
 975        struct btrfs_root *extent_root;
 976        struct btrfs_path *path;
 977        struct extent_buffer *eb;
 978        int tree_block_level = 0;
 979        u64 bytenr = 0, num_bytes = 0;
 980        int ret, level;
 981
 982        if (!btrfs_test_opt(fs_info, REF_VERIFY))
 983                return 0;
 984
 985        path = btrfs_alloc_path();
 986        if (!path)
 987                return -ENOMEM;
 988
 989        extent_root = btrfs_extent_root(fs_info, 0);
 990        eb = btrfs_read_lock_root_node(extent_root);
 991        level = btrfs_header_level(eb);
 992        path->nodes[level] = eb;
 993        path->slots[level] = 0;
 994        path->locks[level] = BTRFS_READ_LOCK;
 995
 996        while (1) {
 997                /*
 998                 * We have to keep track of the bytenr/num_bytes we last hit
 999                 * because we could have run out of space for an inline ref, and
1000                 * would have had to added a ref key item which may appear on a
1001                 * different leaf from the original extent item.
1002                 */
1003                ret = walk_down_tree(extent_root, path, level,
1004                                     &bytenr, &num_bytes, &tree_block_level);
1005                if (ret)
1006                        break;
1007                ret = walk_up_tree(path, &level);
1008                if (ret < 0)
1009                        break;
1010                if (ret > 0) {
1011                        ret = 0;
1012                        break;
1013                }
1014        }
1015        if (ret) {
1016                btrfs_clear_opt(fs_info->mount_opt, REF_VERIFY);
1017                btrfs_free_ref_cache(fs_info);
1018        }
1019        btrfs_free_path(path);
1020        return ret;
1021}
1022