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
  13 /*
  14  * Read a root item from the tree. In case we detect a root item smaller then
  15  * sizeof(root_item), we know it's an old version of the root structure and
  16  * initialize all new fields to zero. The same happens if we detect mismatching
  17  * generation numbers as then we know the root was once mounted with an older
  18  * kernel that was not aware of the root item structure change.
  19  */
  20 static void btrfs_read_root_item(struct extent_buffer *eb, int slot,
  21                                 struct btrfs_root_item *item)
  22 {
  23         uuid_le uuid;
  24         u32 len;
  25         int need_reset = 0;
  26
  27         len = btrfs_item_size_nr(eb, slot);
  28         read_extent_buffer(eb, item, btrfs_item_ptr_offset(eb, slot),
  29                            min_t(u32, len, sizeof(*item)));
  30         if (len < sizeof(*item))
  31                 need_reset = 1;
  32         if (!need_reset && btrfs_root_generation(item)
  33                 != btrfs_root_generation_v2(item)) {
  34                 if (btrfs_root_generation_v2(item) != 0) {
  35                         btrfs_warn(eb->fs_info,
  36                                         "mismatching generation and generation_v2 found in root item. This root was probably mounted with an older kernel. Resetting all new fields.");
  37                 }
  38                 need_reset = 1;
  39         }
  40         if (need_reset) {
  41                 memset(&item->generation_v2, 0,
  42                         sizeof(*item) - offsetof(struct btrfs_root_item,
  43                                         generation_v2));
  44
  45                 uuid_le_gen(&uuid);
  46                 memcpy(item->uuid, uuid.b, BTRFS_UUID_SIZE);
  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_key root_key;
 214         struct btrfs_root *root;
 215         int err = 0;
 216         int ret;
 217
 218         path = btrfs_alloc_path();
 219         if (!path)
 220                 return -ENOMEM;
 221
 222         key.objectid = BTRFS_ORPHAN_OBJECTID;
 223         key.type = BTRFS_ORPHAN_ITEM_KEY;
 224         key.offset = 0;
 225
 226         root_key.type = BTRFS_ROOT_ITEM_KEY;
 227         root_key.offset = (u64)-1;
 228
 229         while (1) {
 230                 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
 231                 if (ret < 0) {
 232                         err = ret;
 233                         break;
 234                 }
 235
 236                 leaf = path->nodes[0];
 237                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 238                         ret = btrfs_next_leaf(tree_root, path);
 239                         if (ret < 0)
 240                                 err = ret;
 241                         if (ret != 0)
 242                                 break;
 243                         leaf = path->nodes[0];
 244                 }
 245
 246                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 247                 btrfs_release_path(path);
 248
 249                 if (key.objectid != BTRFS_ORPHAN_OBJECTID ||
 250                     key.type != BTRFS_ORPHAN_ITEM_KEY)
 251                         break;
 252
 253                 root_key.objectid = key.offset;
 254                 key.offset++;
 255
 256                 /*
 257                  * The root might have been inserted already, as before we look
 258                  * for orphan roots, log replay might have happened, which
 259                  * triggers a transaction commit and qgroup accounting, which
 260                  * in turn reads and inserts fs roots while doing backref
 261                  * walking.
 262                  */
 263                 root = btrfs_lookup_fs_root(fs_info, root_key.objectid);
 264                 if (root) {
 265                         WARN_ON(!test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
 266                                           &root->state));
 267                         if (btrfs_root_refs(&root->root_item) == 0) {
 268                                 set_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
 269                                 btrfs_add_dead_root(root);
 270                         }
 271                         continue;
 272                 }
 273
 274                 root = btrfs_read_fs_root(tree_root, &root_key);
 275                 err = PTR_ERR_OR_ZERO(root);
 276                 if (err && err != -ENOENT) {
 277                         break;
 278                 } else if (err == -ENOENT) {
 279                         struct btrfs_trans_handle *trans;
 280
 281                         btrfs_release_path(path);
 282
 283                         trans = btrfs_join_transaction(tree_root);
 284                         if (IS_ERR(trans)) {
 285                                 err = PTR_ERR(trans);
 286                                 btrfs_handle_fs_error(fs_info, err,
 287                                             "Failed to start trans to delete orphan item");
 288                                 break;
 289                         }
 290                         err = btrfs_del_orphan_item(trans, tree_root,
 291                                                     root_key.objectid);
 292                         btrfs_end_transaction(trans);
 293                         if (err) {
 294                                 btrfs_handle_fs_error(fs_info, err,
 295                                             "Failed to delete root orphan item");
 296                                 break;
 297                         }
 298                         continue;
 299                 }
 300
 301                 err = btrfs_init_fs_root(root);
 302                 if (err) {
 303                         btrfs_free_fs_root(root);
 304                         break;
 305                 }
 306
 307                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
 308
 309                 err = btrfs_insert_fs_root(fs_info, root);
 310                 if (err) {
 311                         BUG_ON(err == -EEXIST);
 312                         btrfs_free_fs_root(root);
 313                         break;
 314                 }
 315
 316                 if (btrfs_root_refs(&root->root_item) == 0) {
 317                         set_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
 318                         btrfs_add_dead_root(root);
 319                 }
 320         }
 321
 322         btrfs_free_path(path);
 323         return err;
 324 }
 325
 326 /* drop the root item for 'key' from the tree root */
 327 int btrfs_del_root(struct btrfs_trans_handle *trans,
 328                    const struct btrfs_key *key)
 329 {
 330         struct btrfs_root *root = trans->fs_info->tree_root;
 331         struct btrfs_path *path;
 332         int ret;
 333
 334         path = btrfs_alloc_path();
 335         if (!path)
 336                 return -ENOMEM;
 337         ret = btrfs_search_slot(trans, root, key, path, -1, 1);
 338         if (ret < 0)
 339                 goto out;
 340
 341         BUG_ON(ret != 0);
 342
 343         ret = btrfs_del_item(trans, root, path);
 344 out:
 345         btrfs_free_path(path);
 346         return ret;
 347 }
 348
 349 int btrfs_del_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
 350                        u64 ref_id, u64 dirid, u64 *sequence, const char *name,
 351                        int name_len)
 352
 353 {
 354         struct btrfs_root *tree_root = trans->fs_info->tree_root;
 355         struct btrfs_path *path;
 356         struct btrfs_root_ref *ref;
 357         struct extent_buffer *leaf;
 358         struct btrfs_key key;
 359         unsigned long ptr;
 360         int err = 0;
 361         int ret;
 362
 363         path = btrfs_alloc_path();
 364         if (!path)
 365                 return -ENOMEM;
 366
 367         key.objectid = root_id;
 368         key.type = BTRFS_ROOT_BACKREF_KEY;
 369         key.offset = ref_id;
 370 again:
 371         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
 372         BUG_ON(ret < 0);
 373         if (ret == 0) {
 374                 leaf = path->nodes[0];
 375                 ref = btrfs_item_ptr(leaf, path->slots[0],
 376                                      struct btrfs_root_ref);
 377
 378                 WARN_ON(btrfs_root_ref_dirid(leaf, ref) != dirid);
 379                 WARN_ON(btrfs_root_ref_name_len(leaf, ref) != name_len);
 380                 ptr = (unsigned long)(ref + 1);
 381                 WARN_ON(memcmp_extent_buffer(leaf, name, ptr, name_len));
 382                 *sequence = btrfs_root_ref_sequence(leaf, ref);
 383
 384                 ret = btrfs_del_item(trans, tree_root, path);
 385                 if (ret) {
 386                         err = ret;
 387                         goto out;
 388                 }
 389         } else
 390                 err = -ENOENT;
 391
 392         if (key.type == BTRFS_ROOT_BACKREF_KEY) {
 393                 btrfs_release_path(path);
 394                 key.objectid = ref_id;
 395                 key.type = BTRFS_ROOT_REF_KEY;
 396                 key.offset = root_id;
 397                 goto again;
 398         }
 399
 400 out:
 401         btrfs_free_path(path);
 402         return err;
 403 }
 404
 405 /*
 406  * add a btrfs_root_ref item.  type is either BTRFS_ROOT_REF_KEY
 407  * or BTRFS_ROOT_BACKREF_KEY.
 408  *
 409  * The dirid, sequence, name and name_len refer to the directory entry
 410  * that is referencing the root.
 411  *
 412  * For a forward ref, the root_id is the id of the tree referencing
 413  * the root and ref_id is the id of the subvol  or snapshot.
 414  *
 415  * For a back ref the root_id is the id of the subvol or snapshot and
 416  * ref_id is the id of the tree referencing it.
 417  *
 418  * Will return 0, -ENOMEM, or anything from the CoW path
 419  */
 420 int btrfs_add_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
 421                        u64 ref_id, u64 dirid, u64 sequence, const char *name,
 422                        int name_len)
 423 {
 424         struct btrfs_root *tree_root = trans->fs_info->tree_root;
 425         struct btrfs_key key;
 426         int ret;
 427         struct btrfs_path *path;
 428         struct btrfs_root_ref *ref;
 429         struct extent_buffer *leaf;
 430         unsigned long ptr;
 431
 432         path = btrfs_alloc_path();
 433         if (!path)
 434                 return -ENOMEM;
 435
 436         key.objectid = root_id;
 437         key.type = BTRFS_ROOT_BACKREF_KEY;
 438         key.offset = ref_id;
 439 again:
 440         ret = btrfs_insert_empty_item(trans, tree_root, path, &key,
 441                                       sizeof(*ref) + name_len);
 442         if (ret) {
 443                 btrfs_abort_transaction(trans, ret);
 444                 btrfs_free_path(path);
 445                 return ret;
 446         }
 447
 448         leaf = path->nodes[0];
 449         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
 450         btrfs_set_root_ref_dirid(leaf, ref, dirid);
 451         btrfs_set_root_ref_sequence(leaf, ref, sequence);
 452         btrfs_set_root_ref_name_len(leaf, ref, name_len);
 453         ptr = (unsigned long)(ref + 1);
 454         write_extent_buffer(leaf, name, ptr, name_len);
 455         btrfs_mark_buffer_dirty(leaf);
 456
 457         if (key.type == BTRFS_ROOT_BACKREF_KEY) {
 458                 btrfs_release_path(path);
 459                 key.objectid = ref_id;
 460                 key.type = BTRFS_ROOT_REF_KEY;
 461                 key.offset = root_id;
 462                 goto again;
 463         }
 464
 465         btrfs_free_path(path);
 466         return 0;
 467 }
 468
 469 /*
 470  * Old btrfs forgets to init root_item->flags and root_item->byte_limit
 471  * for subvolumes. To work around this problem, we steal a bit from
 472  * root_item->inode_item->flags, and use it to indicate if those fields
 473  * have been properly initialized.
 474  */
 475 void btrfs_check_and_init_root_item(struct btrfs_root_item *root_item)
 476 {
 477         u64 inode_flags = btrfs_stack_inode_flags(&root_item->inode);
 478
 479         if (!(inode_flags & BTRFS_INODE_ROOT_ITEM_INIT)) {
 480                 inode_flags |= BTRFS_INODE_ROOT_ITEM_INIT;
 481                 btrfs_set_stack_inode_flags(&root_item->inode, inode_flags);
 482                 btrfs_set_root_flags(root_item, 0);
 483                 btrfs_set_root_limit(root_item, 0);
 484         }
 485 }
 486
 487 void btrfs_update_root_times(struct btrfs_trans_handle *trans,
 488                              struct btrfs_root *root)
 489 {
 490         struct btrfs_root_item *item = &root->root_item;
 491         struct timespec64 ct;
 492
 493         ktime_get_real_ts64(&ct);
 494         spin_lock(&root->root_item_lock);
 495         btrfs_set_root_ctransid(item, trans->transid);
 496         btrfs_set_stack_timespec_sec(&item->ctime, ct.tv_sec);
 497         btrfs_set_stack_timespec_nsec(&item->ctime, ct.tv_nsec);
 498         spin_unlock(&root->root_item_lock);
 499 }