1 // SPDX-License-Identifier: GPL-2.0 2 3 #include "bcachefs.h" 4 #include "btree_key_cache.h" 5 #include "btree_update.h" 6 #include "errcode.h" 7 #include "error.h" 8 #include "fs.h" 9 #include "snapshot.h" 10 #include "subvolume.h" 11 12 #include <linux/random.h> 13 14 static int bch2_subvolume_delete(struct btree_trans *, u32); 15 16 static int check_subvol(struct btree_trans *trans, 17 struct btree_iter *iter, 18 struct bkey_s_c k) 19 { 20 struct bch_fs *c = trans->c; 21 struct bkey_s_c_subvolume subvol; 22 struct bch_snapshot snapshot; 23 unsigned snapid; 24 int ret = 0; 25 26 if (k.k->type != KEY_TYPE_subvolume) 27 return 0; 28 29 subvol = bkey_s_c_to_subvolume(k); 30 snapid = le32_to_cpu(subvol.v->snapshot); 31 ret = bch2_snapshot_lookup(trans, snapid, &snapshot); 32 33 if (bch2_err_matches(ret, ENOENT)) 34 bch_err(c, "subvolume %llu points to nonexistent snapshot %u", 35 k.k->p.offset, snapid); 36 if (ret) 37 return ret; 38 39 if (BCH_SUBVOLUME_UNLINKED(subvol.v)) { 40 bch2_fs_lazy_rw(c); 41 42 ret = bch2_subvolume_delete(trans, iter->pos.offset); 43 if (ret) 44 bch_err_msg(c, ret, "deleting subvolume %llu", iter->pos.offset); 45 return ret ?: -BCH_ERR_transaction_restart_nested; 46 } 47 48 if (!BCH_SUBVOLUME_SNAP(subvol.v)) { 49 u32 snapshot_root = bch2_snapshot_root(c, le32_to_cpu(subvol.v->snapshot)); 50 u32 snapshot_tree; 51 struct bch_snapshot_tree st; 52 53 rcu_read_lock(); 54 snapshot_tree = snapshot_t(c, snapshot_root)->tree; 55 rcu_read_unlock(); 56 57 ret = bch2_snapshot_tree_lookup(trans, snapshot_tree, &st); 58 59 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c, 60 "%s: snapshot tree %u not found", __func__, snapshot_tree); 61 62 if (ret) 63 return ret; 64 65 if (fsck_err_on(le32_to_cpu(st.master_subvol) != subvol.k->p.offset, 66 c, subvol_not_master_and_not_snapshot, 67 "subvolume %llu is not set as snapshot but is not master subvolume", 68 k.k->p.offset)) { 69 struct bkey_i_subvolume *s = 70 bch2_bkey_make_mut_typed(trans, iter, &subvol.s_c, 0, subvolume); 71 ret = PTR_ERR_OR_ZERO(s); 72 if (ret) 73 return ret; 74 75 SET_BCH_SUBVOLUME_SNAP(&s->v, true); 76 } 77 } 78 79 fsck_err: 80 return ret; 81 } 82 83 int bch2_check_subvols(struct bch_fs *c) 84 { 85 struct btree_iter iter; 86 struct bkey_s_c k; 87 int ret; 88 89 ret = bch2_trans_run(c, 90 for_each_btree_key_commit(trans, iter, 91 BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k, 92 NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL, 93 check_subvol(trans, &iter, k))); 94 if (ret) 95 bch_err_fn(c, ret); 96 return ret; 97 } 98 99 /* Subvolumes: */ 100 101 int bch2_subvolume_invalid(struct bch_fs *c, struct bkey_s_c k, 102 enum bkey_invalid_flags flags, struct printbuf *err) 103 { 104 int ret = 0; 105 106 bkey_fsck_err_on(bkey_lt(k.k->p, SUBVOL_POS_MIN) || 107 bkey_gt(k.k->p, SUBVOL_POS_MAX), c, err, 108 subvol_pos_bad, 109 "invalid pos"); 110 fsck_err: 111 return ret; 112 } 113 114 void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c, 115 struct bkey_s_c k) 116 { 117 struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k); 118 119 prt_printf(out, "root %llu snapshot id %u", 120 le64_to_cpu(s.v->inode), 121 le32_to_cpu(s.v->snapshot)); 122 123 if (bkey_val_bytes(s.k) > offsetof(struct bch_subvolume, parent)) 124 prt_printf(out, " parent %u", le32_to_cpu(s.v->parent)); 125 } 126 127 static __always_inline int 128 bch2_subvolume_get_inlined(struct btree_trans *trans, unsigned subvol, 129 bool inconsistent_if_not_found, 130 int iter_flags, 131 struct bch_subvolume *s) 132 { 133 int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_subvolumes, POS(0, subvol), 134 iter_flags, subvolume, s); 135 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT) && 136 inconsistent_if_not_found, 137 trans->c, "missing subvolume %u", subvol); 138 return ret; 139 } 140 141 int bch2_subvolume_get(struct btree_trans *trans, unsigned subvol, 142 bool inconsistent_if_not_found, 143 int iter_flags, 144 struct bch_subvolume *s) 145 { 146 return bch2_subvolume_get_inlined(trans, subvol, inconsistent_if_not_found, iter_flags, s); 147 } 148 149 int bch2_snapshot_get_subvol(struct btree_trans *trans, u32 snapshot, 150 struct bch_subvolume *subvol) 151 { 152 struct bch_snapshot snap; 153 154 return bch2_snapshot_lookup(trans, snapshot, &snap) ?: 155 bch2_subvolume_get(trans, le32_to_cpu(snap.subvol), true, 0, subvol); 156 } 157 158 int bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvolid, 159 u32 *snapid) 160 { 161 struct btree_iter iter; 162 struct bkey_s_c_subvolume subvol; 163 int ret; 164 165 subvol = bch2_bkey_get_iter_typed(trans, &iter, 166 BTREE_ID_subvolumes, POS(0, subvolid), 167 BTREE_ITER_CACHED|BTREE_ITER_WITH_UPDATES, 168 subvolume); 169 ret = bkey_err(subvol); 170 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c, 171 "missing subvolume %u", subvolid); 172 173 if (likely(!ret)) 174 *snapid = le32_to_cpu(subvol.v->snapshot); 175 bch2_trans_iter_exit(trans, &iter); 176 return ret; 177 } 178 179 static int bch2_subvolume_reparent(struct btree_trans *trans, 180 struct btree_iter *iter, 181 struct bkey_s_c k, 182 u32 old_parent, u32 new_parent) 183 { 184 struct bkey_i_subvolume *s; 185 int ret; 186 187 if (k.k->type != KEY_TYPE_subvolume) 188 return 0; 189 190 if (bkey_val_bytes(k.k) > offsetof(struct bch_subvolume, parent) && 191 le32_to_cpu(bkey_s_c_to_subvolume(k).v->parent) != old_parent) 192 return 0; 193 194 s = bch2_bkey_make_mut_typed(trans, iter, &k, 0, subvolume); 195 ret = PTR_ERR_OR_ZERO(s); 196 if (ret) 197 return ret; 198 199 s->v.parent = cpu_to_le32(new_parent); 200 return 0; 201 } 202 203 /* 204 * Separate from the snapshot tree in the snapshots btree, we record the tree 205 * structure of how snapshot subvolumes were created - the parent subvolume of 206 * each snapshot subvolume. 207 * 208 * When a subvolume is deleted, we scan for child subvolumes and reparant them, 209 * to avoid dangling references: 210 */ 211 static int bch2_subvolumes_reparent(struct btree_trans *trans, u32 subvolid_to_delete) 212 { 213 struct btree_iter iter; 214 struct bkey_s_c k; 215 struct bch_subvolume s; 216 217 return lockrestart_do(trans, 218 bch2_subvolume_get(trans, subvolid_to_delete, true, 219 BTREE_ITER_CACHED, &s)) ?: 220 for_each_btree_key_commit(trans, iter, 221 BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k, 222 NULL, NULL, BTREE_INSERT_NOFAIL, 223 bch2_subvolume_reparent(trans, &iter, k, 224 subvolid_to_delete, le32_to_cpu(s.parent))); 225 } 226 227 /* 228 * Delete subvolume, mark snapshot ID as deleted, queue up snapshot 229 * deletion/cleanup: 230 */ 231 static int __bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid) 232 { 233 struct btree_iter iter; 234 struct bkey_s_c_subvolume subvol; 235 u32 snapid; 236 int ret = 0; 237 238 subvol = bch2_bkey_get_iter_typed(trans, &iter, 239 BTREE_ID_subvolumes, POS(0, subvolid), 240 BTREE_ITER_CACHED|BTREE_ITER_INTENT, 241 subvolume); 242 ret = bkey_err(subvol); 243 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c, 244 "missing subvolume %u", subvolid); 245 if (ret) 246 return ret; 247 248 snapid = le32_to_cpu(subvol.v->snapshot); 249 250 ret = bch2_btree_delete_at(trans, &iter, 0) ?: 251 bch2_snapshot_node_set_deleted(trans, snapid); 252 bch2_trans_iter_exit(trans, &iter); 253 return ret; 254 } 255 256 static int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid) 257 { 258 return bch2_subvolumes_reparent(trans, subvolid) ?: 259 commit_do(trans, NULL, NULL, BTREE_INSERT_NOFAIL, 260 __bch2_subvolume_delete(trans, subvolid)); 261 } 262 263 static void bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct *work) 264 { 265 struct bch_fs *c = container_of(work, struct bch_fs, 266 snapshot_wait_for_pagecache_and_delete_work); 267 snapshot_id_list s; 268 u32 *id; 269 int ret = 0; 270 271 while (!ret) { 272 mutex_lock(&c->snapshots_unlinked_lock); 273 s = c->snapshots_unlinked; 274 darray_init(&c->snapshots_unlinked); 275 mutex_unlock(&c->snapshots_unlinked_lock); 276 277 if (!s.nr) 278 break; 279 280 bch2_evict_subvolume_inodes(c, &s); 281 282 for (id = s.data; id < s.data + s.nr; id++) { 283 ret = bch2_trans_run(c, bch2_subvolume_delete(trans, *id)); 284 if (ret) { 285 bch_err_msg(c, ret, "deleting subvolume %u", *id); 286 break; 287 } 288 } 289 290 darray_exit(&s); 291 } 292 293 bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache); 294 } 295 296 struct subvolume_unlink_hook { 297 struct btree_trans_commit_hook h; 298 u32 subvol; 299 }; 300 301 static int bch2_subvolume_wait_for_pagecache_and_delete_hook(struct btree_trans *trans, 302 struct btree_trans_commit_hook *_h) 303 { 304 struct subvolume_unlink_hook *h = container_of(_h, struct subvolume_unlink_hook, h); 305 struct bch_fs *c = trans->c; 306 int ret = 0; 307 308 mutex_lock(&c->snapshots_unlinked_lock); 309 if (!snapshot_list_has_id(&c->snapshots_unlinked, h->subvol)) 310 ret = snapshot_list_add(c, &c->snapshots_unlinked, h->subvol); 311 mutex_unlock(&c->snapshots_unlinked_lock); 312 313 if (ret) 314 return ret; 315 316 if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_snapshot_delete_pagecache)) 317 return -EROFS; 318 319 if (!queue_work(c->write_ref_wq, &c->snapshot_wait_for_pagecache_and_delete_work)) 320 bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache); 321 return 0; 322 } 323 324 int bch2_subvolume_unlink(struct btree_trans *trans, u32 subvolid) 325 { 326 struct btree_iter iter; 327 struct bkey_i_subvolume *n; 328 struct subvolume_unlink_hook *h; 329 int ret = 0; 330 331 h = bch2_trans_kmalloc(trans, sizeof(*h)); 332 ret = PTR_ERR_OR_ZERO(h); 333 if (ret) 334 return ret; 335 336 h->h.fn = bch2_subvolume_wait_for_pagecache_and_delete_hook; 337 h->subvol = subvolid; 338 bch2_trans_commit_hook(trans, &h->h); 339 340 n = bch2_bkey_get_mut_typed(trans, &iter, 341 BTREE_ID_subvolumes, POS(0, subvolid), 342 BTREE_ITER_CACHED, subvolume); 343 ret = PTR_ERR_OR_ZERO(n); 344 if (unlikely(ret)) { 345 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c, 346 "missing subvolume %u", subvolid); 347 return ret; 348 } 349 350 SET_BCH_SUBVOLUME_UNLINKED(&n->v, true); 351 bch2_trans_iter_exit(trans, &iter); 352 return ret; 353 } 354 355 int bch2_subvolume_create(struct btree_trans *trans, u64 inode, 356 u32 src_subvolid, 357 u32 *new_subvolid, 358 u32 *new_snapshotid, 359 bool ro) 360 { 361 struct bch_fs *c = trans->c; 362 struct btree_iter dst_iter, src_iter = (struct btree_iter) { NULL }; 363 struct bkey_i_subvolume *new_subvol = NULL; 364 struct bkey_i_subvolume *src_subvol = NULL; 365 u32 parent = 0, new_nodes[2], snapshot_subvols[2]; 366 int ret = 0; 367 368 ret = bch2_bkey_get_empty_slot(trans, &dst_iter, 369 BTREE_ID_subvolumes, POS(0, U32_MAX)); 370 if (ret == -BCH_ERR_ENOSPC_btree_slot) 371 ret = -BCH_ERR_ENOSPC_subvolume_create; 372 if (ret) 373 return ret; 374 375 snapshot_subvols[0] = dst_iter.pos.offset; 376 snapshot_subvols[1] = src_subvolid; 377 378 if (src_subvolid) { 379 /* Creating a snapshot: */ 380 381 src_subvol = bch2_bkey_get_mut_typed(trans, &src_iter, 382 BTREE_ID_subvolumes, POS(0, src_subvolid), 383 BTREE_ITER_CACHED, subvolume); 384 ret = PTR_ERR_OR_ZERO(src_subvol); 385 if (unlikely(ret)) { 386 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c, 387 "subvolume %u not found", src_subvolid); 388 goto err; 389 } 390 391 parent = le32_to_cpu(src_subvol->v.snapshot); 392 } 393 394 ret = bch2_snapshot_node_create(trans, parent, new_nodes, 395 snapshot_subvols, 396 src_subvolid ? 2 : 1); 397 if (ret) 398 goto err; 399 400 if (src_subvolid) { 401 src_subvol->v.snapshot = cpu_to_le32(new_nodes[1]); 402 ret = bch2_trans_update(trans, &src_iter, &src_subvol->k_i, 0); 403 if (ret) 404 goto err; 405 } 406 407 new_subvol = bch2_bkey_alloc(trans, &dst_iter, 0, subvolume); 408 ret = PTR_ERR_OR_ZERO(new_subvol); 409 if (ret) 410 goto err; 411 412 new_subvol->v.flags = 0; 413 new_subvol->v.snapshot = cpu_to_le32(new_nodes[0]); 414 new_subvol->v.inode = cpu_to_le64(inode); 415 new_subvol->v.parent = cpu_to_le32(src_subvolid); 416 new_subvol->v.otime.lo = cpu_to_le64(bch2_current_time(c)); 417 new_subvol->v.otime.hi = 0; 418 419 SET_BCH_SUBVOLUME_RO(&new_subvol->v, ro); 420 SET_BCH_SUBVOLUME_SNAP(&new_subvol->v, src_subvolid != 0); 421 422 *new_subvolid = new_subvol->k.p.offset; 423 *new_snapshotid = new_nodes[0]; 424 err: 425 bch2_trans_iter_exit(trans, &src_iter); 426 bch2_trans_iter_exit(trans, &dst_iter); 427 return ret; 428 } 429 430 int bch2_fs_subvolumes_init(struct bch_fs *c) 431 { 432 INIT_WORK(&c->snapshot_delete_work, bch2_delete_dead_snapshots_work); 433 INIT_WORK(&c->snapshot_wait_for_pagecache_and_delete_work, 434 bch2_subvolume_wait_for_pagecache_and_delete); 435 mutex_init(&c->snapshots_unlinked_lock); 436 return 0; 437 } 438