fs/btrfs/root-tree.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/err.h>
   7 #include <linux/uuid.h>
   8 #include "ctree.h"
   9 #include "transaction.h"
  10 #include "disk-io.h"
  11 #include "print-tree.h"
  12 #include "qgroup.h"
  13 #include "space-info.h"
  14
  15 /*
  16  * Read a root item from the tree. In case we detect a root item smaller then
  17  * sizeof(root_item), we know it's an old version of the root structure and
  18  * initialize all new fields to zero. The same happens if we detect mismatching
  19  * generation numbers as then we know the root was once mounted with an older
  20  * kernel that was not aware of the root item structure change.
  21  */
  22 static void btrfs_read_root_item(struct extent_buffer *eb, int slot,
  23                                 struct btrfs_root_item *item)
  24 {
  25         u32 len;
  26         int need_reset = 0;
  27
  28         len = btrfs_item_size_nr(eb, slot);
  29         read_extent_buffer(eb, item, btrfs_item_ptr_offset(eb, slot),
  30                            min_t(u32, len, sizeof(*item)));
  31         if (len < sizeof(*item))
  32                 need_reset = 1;
  33         if (!need_reset && btrfs_root_generation(item)
  34                 != btrfs_root_generation_v2(item)) {
  35                 if (btrfs_root_generation_v2(item) != 0) {
  36                         btrfs_warn(eb->fs_info,
  37                                         "mismatching generation and generation_v2 found in root item. This root was probably mounted with an older kernel. Resetting all new fields.");
  38                 }
  39                 need_reset = 1;
  40         }
  41         if (need_reset) {
  42                 memset(&item->generation_v2, 0,
  43                         sizeof(*item) - offsetof(struct btrfs_root_item,
  44                                         generation_v2));
  45
  46                 generate_random_guid(item->uuid);
  47         }
  48 }
  49
  50 /*
  51  * btrfs_find_root - lookup the root by the key.
  52  * root: the root of the root tree
  53  * search_key: the key to search
  54  * path: the path we search
  55  * root_item: the root item of the tree we look for
  56  * root_key: the root key of the tree we look for
  57  *
  58  * If ->offset of 'search_key' is -1ULL, it means we are not sure the offset
  59  * of the search key, just lookup the root with the highest offset for a
  60  * given objectid.
  61  *
  62  * If we find something return 0, otherwise > 0, < 0 on error.
  63  */
  64 int btrfs_find_root(struct btrfs_root *root, const struct btrfs_key *search_key,
  65                     struct btrfs_path *path, struct btrfs_root_item *root_item,
  66                     struct btrfs_key *root_key)
  67 {
  68         struct btrfs_key found_key;
  69         struct extent_buffer *l;
  70         int ret;
  71         int slot;
  72
  73         ret = btrfs_search_slot(NULL, root, search_key, path, 0, 0);
  74         if (ret < 0)
  75                 return ret;
  76
  77         if (search_key->offset != -1ULL) {      /* the search key is exact */
  78                 if (ret > 0)
  79                         goto out;
  80         } else {
  81                 BUG_ON(ret == 0);               /* Logical error */
  82                 if (path->slots[0] == 0)
  83                         goto out;
  84                 path->slots[0]--;
  85                 ret = 0;
  86         }
  87
  88         l = path->nodes[0];
  89         slot = path->slots[0];
  90
  91         btrfs_item_key_to_cpu(l, &found_key, slot);
  92         if (found_key.objectid != search_key->objectid ||
  93             found_key.type != BTRFS_ROOT_ITEM_KEY) {
  94                 ret = 1;
  95                 goto out;
  96         }
  97
  98         if (root_item)
  99                 btrfs_read_root_item(l, slot, root_item);
 100         if (root_key)
 101                 memcpy(root_key, &found_key, sizeof(found_key));
 102 out:
 103         btrfs_release_path(path);
 104         return ret;
 105 }
 106
 107 void btrfs_set_root_node(struct btrfs_root_item *item,
 108                          struct extent_buffer *node)
 109 {
 110         btrfs_set_root_bytenr(item, node->start);
 111         btrfs_set_root_level(item, btrfs_header_level(node));
 112         btrfs_set_root_generation(item, btrfs_header_generation(node));
 113 }
 114
 115 /*
 116  * copy the data in 'item' into the btree
 117  */
 118 int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root
 119                       *root, struct btrfs_key *key, struct btrfs_root_item
 120                       *item)
 121 {
 122         struct btrfs_fs_info *fs_info = root->fs_info;
 123         struct btrfs_path *path;
 124         struct extent_buffer *l;
 125         int ret;
 126         int slot;
 127         unsigned long ptr;
 128         u32 old_len;
 129
 130         path = btrfs_alloc_path();
 131         if (!path)
 132                 return -ENOMEM;
 133
 134         ret = btrfs_search_slot(trans, root, key, path, 0, 1);
 135         if (ret < 0)
 136                 goto out;
 137
 138         if (ret > 0) {
 139                 btrfs_crit(fs_info,
 140                         "unable to find root key (%llu %u %llu) in tree %llu",
 141                         key->objectid, key->type, key->offset,
 142                         root->root_key.objectid);
 143                 ret = -EUCLEAN;
 144                 btrfs_abort_transaction(trans, ret);
 145                 goto out;
 146         }
 147
 148         l = path->nodes[0];
 149         slot = path->slots[0];
 150         ptr = btrfs_item_ptr_offset(l, slot);
 151         old_len = btrfs_item_size_nr(l, slot);
 152
 153         /*
 154          * If this is the first time we update the root item which originated
 155          * from an older kernel, we need to enlarge the item size to make room
 156          * for the added fields.
 157          */
 158         if (old_len < sizeof(*item)) {
 159                 btrfs_release_path(path);
 160                 ret = btrfs_search_slot(trans, root, key, path,
 161                                 -1, 1);
 162                 if (ret < 0) {
 163                         btrfs_abort_transaction(trans, ret);
 164                         goto out;
 165                 }
 166
 167                 ret = btrfs_del_item(trans, root, path);
 168                 if (ret < 0) {
 169                         btrfs_abort_transaction(trans, ret);
 170                         goto out;
 171                 }
 172                 btrfs_release_path(path);
 173                 ret = btrfs_insert_empty_item(trans, root, path,
 174                                 key, sizeof(*item));
 175                 if (ret < 0) {
 176                         btrfs_abort_transaction(trans, ret);
 177                         goto out;
 178                 }
 179                 l = path->nodes[0];
 180                 slot = path->slots[0];
 181                 ptr = btrfs_item_ptr_offset(l, slot);
 182         }
 183
 184         /*
 185          * Update generation_v2 so at the next mount we know the new root
 186          * fields are valid.
 187          */
 188         btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
 189
 190         write_extent_buffer(l, item, ptr, sizeof(*item));
 191         btrfs_mark_buffer_dirty(path->nodes[0]);
 192 out:
 193         btrfs_free_path(path);
 194         return ret;
 195 }
 196
 197 int btrfs_insert_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 198                       const struct btrfs_key *key, struct btrfs_root_item *item)
 199 {
 200         /*
 201          * Make sure generation v1 and v2 match. See update_root for details.
 202          */
 203         btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
 204         return btrfs_insert_item(trans, root, key, item, sizeof(*item));
 205 }
 206
 207 int btrfs_find_orphan_roots(struct btrfs_fs_info *fs_info)
 208 {
 209         struct btrfs_root *tree_root = fs_info->tree_root;
 210         struct extent_buffer *leaf;
 211         struct btrfs_path *path;
 212         struct btrfs_key key;
 213         struct btrfs_root *root;
 214         int err = 0;
 215         int ret;
 216
 217         path = btrfs_alloc_path();
 218         if (!path)
 219                 return -ENOMEM;
 220
 221         key.objectid = BTRFS_ORPHAN_OBJECTID;
 222         key.type = BTRFS_ORPHAN_ITEM_KEY;
 223         key.offset = 0;
 224
 225         while (1) {
 226                 u64 root_objectid;
 227
 228                 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
 229                 if (ret < 0) {
 230                         err = ret;
 231                         break;
 232                 }
 233
 234                 leaf = path->nodes[0];
 235                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 236                         ret = btrfs_next_leaf(tree_root, path);
 237                         if (ret < 0)
 238                                 err = ret;
 239                         if (ret != 0)
 240                                 break;
 241                         leaf = path->nodes[0];
 242                 }
 243
 244                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 245                 btrfs_release_path(path);
 246
 247                 if (key.objectid != BTRFS_ORPHAN_OBJECTID ||
 248                     key.type != BTRFS_ORPHAN_ITEM_KEY)
 249                         break;
 250
 251                 root_objectid = key.offset;
 252                 key.offset++;
 253
 254                 root = btrfs_get_fs_root(fs_info, root_objectid, false);
 255                 err = PTR_ERR_OR_ZERO(root);
 256                 if (err && err != -ENOENT) {
 257                         break;
 258                 } else if (err == -ENOENT) {
 259                         struct btrfs_trans_handle *trans;
 260
 261                         btrfs_release_path(path);
 262
 263                         trans = btrfs_join_transaction(tree_root);
 264                         if (IS_ERR(trans)) {
 265                                 err = PTR_ERR(trans);
 266                                 btrfs_handle_fs_error(fs_info, err,
 267                                             "Failed to start trans to delete orphan item");
 268                                 break;
 269                         }
 270                         err = btrfs_del_orphan_item(trans, tree_root,
 271                                                     root_objectid);
 272                         btrfs_end_transaction(trans);
 273                         if (err) {
 274                                 btrfs_handle_fs_error(fs_info, err,
 275                                             "Failed to delete root orphan item");
 276                                 break;
 277                         }
 278                         continue;
 279                 }
 280
 281                 WARN_ON(!test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state));
 282                 if (btrfs_root_refs(&root->root_item) == 0) {
 283                         set_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
 284                         btrfs_add_dead_root(root);
 285                 }
 286                 btrfs_put_root(root);
 287         }
 288
 289         btrfs_free_path(path);
 290         return err;
 291 }
 292
 293 /* drop the root item for 'key' from the tree root */
 294 int btrfs_del_root(struct btrfs_trans_handle *trans,
 295                    const struct btrfs_key *key)
 296 {
 297         struct btrfs_root *root = trans->fs_info->tree_root;
 298         struct btrfs_path *path;
 299         int ret;
 300
 301         path = btrfs_alloc_path();
 302         if (!path)
 303                 return -ENOMEM;
 304         ret = btrfs_search_slot(trans, root, key, path, -1, 1);
 305         if (ret < 0)
 306                 goto out;
 307
 308         BUG_ON(ret != 0);
 309
 310         ret = btrfs_del_item(trans, root, path);
 311 out:
 312         btrfs_free_path(path);
 313         return ret;
 314 }
 315
 316 int btrfs_del_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
 317                        u64 ref_id, u64 dirid, u64 *sequence, const char *name,
 318                        int name_len)
 319
 320 {
 321         struct btrfs_root *tree_root = trans->fs_info->tree_root;
 322         struct btrfs_path *path;
 323         struct btrfs_root_ref *ref;
 324         struct extent_buffer *leaf;
 325         struct btrfs_key key;
 326         unsigned long ptr;
 327         int err = 0;
 328         int ret;
 329
 330         path = btrfs_alloc_path();
 331         if (!path)
 332                 return -ENOMEM;
 333
 334         key.objectid = root_id;
 335         key.type = BTRFS_ROOT_BACKREF_KEY;
 336         key.offset = ref_id;
 337 again:
 338         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
 339         BUG_ON(ret < 0);
 340         if (ret == 0) {
 341                 leaf = path->nodes[0];
 342                 ref = btrfs_item_ptr(leaf, path->slots[0],
 343                                      struct btrfs_root_ref);
 344                 ptr = (unsigned long)(ref + 1);
 345                 if ((btrfs_root_ref_dirid(leaf, ref) != dirid) ||
 346                     (btrfs_root_ref_name_len(leaf, ref) != name_len) ||
 347                     memcmp_extent_buffer(leaf, name, ptr, name_len)) {
 348                         err = -ENOENT;
 349                         goto out;
 350                 }
 351                 *sequence = btrfs_root_ref_sequence(leaf, ref);
 352
 353                 ret = btrfs_del_item(trans, tree_root, path);
 354                 if (ret) {
 355                         err = ret;
 356                         goto out;
 357                 }
 358         } else
 359                 err = -ENOENT;
 360
 361         if (key.type == BTRFS_ROOT_BACKREF_KEY) {
 362                 btrfs_release_path(path);
 363                 key.objectid = ref_id;
 364                 key.type = BTRFS_ROOT_REF_KEY;
 365                 key.offset = root_id;
 366                 goto again;
 367         }
 368
 369 out:
 370         btrfs_free_path(path);
 371         return err;
 372 }
 373
 374 /*
 375  * add a btrfs_root_ref item.  type is either BTRFS_ROOT_REF_KEY
 376  * or BTRFS_ROOT_BACKREF_KEY.
 377  *
 378  * The dirid, sequence, name and name_len refer to the directory entry
 379  * that is referencing the root.
 380  *
 381  * For a forward ref, the root_id is the id of the tree referencing
 382  * the root and ref_id is the id of the subvol  or snapshot.
 383  *
 384  * For a back ref the root_id is the id of the subvol or snapshot and
 385  * ref_id is the id of the tree referencing it.
 386  *
 387  * Will return 0, -ENOMEM, or anything from the CoW path
 388  */
 389 int btrfs_add_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
 390                        u64 ref_id, u64 dirid, u64 sequence, const char *name,
 391                        int name_len)
 392 {
 393         struct btrfs_root *tree_root = trans->fs_info->tree_root;
 394         struct btrfs_key key;
 395         int ret;
 396         struct btrfs_path *path;
 397         struct btrfs_root_ref *ref;
 398         struct extent_buffer *leaf;
 399         unsigned long ptr;
 400
 401         path = btrfs_alloc_path();
 402         if (!path)
 403                 return -ENOMEM;
 404
 405         key.objectid = root_id;
 406         key.type = BTRFS_ROOT_BACKREF_KEY;
 407         key.offset = ref_id;
 408 again:
 409         ret = btrfs_insert_empty_item(trans, tree_root, path, &key,
 410                                       sizeof(*ref) + name_len);
 411         if (ret) {
 412                 btrfs_abort_transaction(trans, ret);
 413                 btrfs_free_path(path);
 414                 return ret;
 415         }
 416
 417         leaf = path->nodes[0];
 418         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
 419         btrfs_set_root_ref_dirid(leaf, ref, dirid);
 420         btrfs_set_root_ref_sequence(leaf, ref, sequence);
 421         btrfs_set_root_ref_name_len(leaf, ref, name_len);
 422         ptr = (unsigned long)(ref + 1);
 423         write_extent_buffer(leaf, name, ptr, name_len);
 424         btrfs_mark_buffer_dirty(leaf);
 425
 426         if (key.type == BTRFS_ROOT_BACKREF_KEY) {
 427                 btrfs_release_path(path);
 428                 key.objectid = ref_id;
 429                 key.type = BTRFS_ROOT_REF_KEY;
 430                 key.offset = root_id;
 431                 goto again;
 432         }
 433
 434         btrfs_free_path(path);
 435         return 0;
 436 }
 437
 438 /*
 439  * Old btrfs forgets to init root_item->flags and root_item->byte_limit
 440  * for subvolumes. To work around this problem, we steal a bit from
 441  * root_item->inode_item->flags, and use it to indicate if those fields
 442  * have been properly initialized.
 443  */
 444 void btrfs_check_and_init_root_item(struct btrfs_root_item *root_item)
 445 {
 446         u64 inode_flags = btrfs_stack_inode_flags(&root_item->inode);
 447
 448         if (!(inode_flags & BTRFS_INODE_ROOT_ITEM_INIT)) {
 449                 inode_flags |= BTRFS_INODE_ROOT_ITEM_INIT;
 450                 btrfs_set_stack_inode_flags(&root_item->inode, inode_flags);
 451                 btrfs_set_root_flags(root_item, 0);
 452                 btrfs_set_root_limit(root_item, 0);
 453         }
 454 }
 455
 456 void btrfs_update_root_times(struct btrfs_trans_handle *trans,
 457                              struct btrfs_root *root)
 458 {
 459         struct btrfs_root_item *item = &root->root_item;
 460         struct timespec64 ct;
 461
 462         ktime_get_real_ts64(&ct);
 463         spin_lock(&root->root_item_lock);
 464         btrfs_set_root_ctransid(item, trans->transid);
 465         btrfs_set_stack_timespec_sec(&item->ctime, ct.tv_sec);
 466         btrfs_set_stack_timespec_nsec(&item->ctime, ct.tv_nsec);
 467         spin_unlock(&root->root_item_lock);
 468 }
 469
 470 /*
 471  * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
 472  * root: the root of the parent directory
 473  * rsv: block reservation
 474  * items: the number of items that we need do reservation
 475  * use_global_rsv: allow fallback to the global block reservation
 476  *
 477  * This function is used to reserve the space for snapshot/subvolume
 478  * creation and deletion. Those operations are different with the
 479  * common file/directory operations, they change two fs/file trees
 480  * and root tree, the number of items that the qgroup reserves is
 481  * different with the free space reservation. So we can not use
 482  * the space reservation mechanism in start_transaction().
 483  */
 484 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
 485                                      struct btrfs_block_rsv *rsv, int items,
 486                                      bool use_global_rsv)
 487 {
 488         u64 qgroup_num_bytes = 0;
 489         u64 num_bytes;
 490         int ret;
 491         struct btrfs_fs_info *fs_info = root->fs_info;
 492         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
 493
 494         if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
 495                 /* One for parent inode, two for dir entries */
 496                 qgroup_num_bytes = 3 * fs_info->nodesize;
 497                 ret = btrfs_qgroup_reserve_meta_prealloc(root,
 498                                 qgroup_num_bytes, true);
 499                 if (ret)
 500                         return ret;
 501         }
 502
 503         num_bytes = btrfs_calc_insert_metadata_size(fs_info, items);
 504         rsv->space_info = btrfs_find_space_info(fs_info,
 505                                             BTRFS_BLOCK_GROUP_METADATA);
 506         ret = btrfs_block_rsv_add(root, rsv, num_bytes,
 507                                   BTRFS_RESERVE_FLUSH_ALL);
 508
 509         if (ret == -ENOSPC && use_global_rsv)
 510                 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true);
 511
 512         if (ret && qgroup_num_bytes)
 513                 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
 514
 515         if (!ret) {
 516                 spin_lock(&rsv->lock);
 517                 rsv->qgroup_rsv_reserved += qgroup_num_bytes;
 518                 spin_unlock(&rsv->lock);
 519         }
 520         return ret;
 521 }
 522
 523 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
 524                                       struct btrfs_block_rsv *rsv)
 525 {
 526         struct btrfs_fs_info *fs_info = root->fs_info;
 527         u64 qgroup_to_release;
 528
 529         btrfs_block_rsv_release(fs_info, rsv, (u64)-1, &qgroup_to_release);
 530         btrfs_qgroup_convert_reserved_meta(root, qgroup_to_release);
 531 }