git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/dirent.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "bkey_methods.h"
   5 #include "btree_update.h"
   6 #include "extents.h"
   7 #include "dirent.h"
   8 #include "fs.h"
   9 #include "keylist.h"
  10 #include "str_hash.h"
  11 #include "subvolume.h"
  12
  13 #include <linux/dcache.h>
  14
  15 unsigned bch2_dirent_name_bytes(struct bkey_s_c_dirent d)
  16 {
  17         unsigned len = bkey_val_bytes(d.k) -
  18                 offsetof(struct bch_dirent, d_name);
  19
  20         return strnlen(d.v->d_name, len);
  21 }
  22
  23 static u64 bch2_dirent_hash(const struct bch_hash_info *info,
  24                             const struct qstr *name)
  25 {
  26         struct bch_str_hash_ctx ctx;
  27
  28         bch2_str_hash_init(&ctx, info);
  29         bch2_str_hash_update(&ctx, info, name->name, name->len);
  30
  31         /* [0,2) reserved for dots */
  32         return max_t(u64, bch2_str_hash_end(&ctx, info), 2);
  33 }
  34
  35 static u64 dirent_hash_key(const struct bch_hash_info *info, const void *key)
  36 {
  37         return bch2_dirent_hash(info, key);
  38 }
  39
  40 static u64 dirent_hash_bkey(const struct bch_hash_info *info, struct bkey_s_c k)
  41 {
  42         struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
  43         struct qstr name = QSTR_INIT(d.v->d_name, bch2_dirent_name_bytes(d));
  44
  45         return bch2_dirent_hash(info, &name);
  46 }
  47
  48 static bool dirent_cmp_key(struct bkey_s_c _l, const void *_r)
  49 {
  50         struct bkey_s_c_dirent l = bkey_s_c_to_dirent(_l);
  51         int len = bch2_dirent_name_bytes(l);
  52         const struct qstr *r = _r;
  53
  54         return len - r->len ?: memcmp(l.v->d_name, r->name, len);
  55 }
  56
  57 static bool dirent_cmp_bkey(struct bkey_s_c _l, struct bkey_s_c _r)
  58 {
  59         struct bkey_s_c_dirent l = bkey_s_c_to_dirent(_l);
  60         struct bkey_s_c_dirent r = bkey_s_c_to_dirent(_r);
  61         int l_len = bch2_dirent_name_bytes(l);
  62         int r_len = bch2_dirent_name_bytes(r);
  63
  64         return l_len - r_len ?: memcmp(l.v->d_name, r.v->d_name, l_len);
  65 }
  66
  67 static bool dirent_is_visible(subvol_inum inum, struct bkey_s_c k)
  68 {
  69         struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
  70
  71         if (d.v->d_type == DT_SUBVOL)
  72                 return le32_to_cpu(d.v->d_parent_subvol) == inum.subvol;
  73         return true;
  74 }
  75
  76 const struct bch_hash_desc bch2_dirent_hash_desc = {
  77         .btree_id       = BTREE_ID_dirents,
  78         .key_type       = KEY_TYPE_dirent,
  79         .hash_key       = dirent_hash_key,
  80         .hash_bkey      = dirent_hash_bkey,
  81         .cmp_key        = dirent_cmp_key,
  82         .cmp_bkey       = dirent_cmp_bkey,
  83         .is_visible     = dirent_is_visible,
  84 };
  85
  86 const char *bch2_dirent_invalid(const struct bch_fs *c, struct bkey_s_c k)
  87 {
  88         struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
  89         unsigned len;
  90
  91         if (bkey_val_bytes(k.k) < sizeof(struct bch_dirent))
  92                 return "value too small";
  93
  94         len = bch2_dirent_name_bytes(d);
  95         if (!len)
  96                 return "empty name";
  97
  98         if (bkey_val_u64s(k.k) > dirent_val_u64s(len))
  99                 return "value too big";
 100
 101         if (len > BCH_NAME_MAX)
 102                 return "dirent name too big";
 103
 104         if (len == 1 && !memcmp(d.v->d_name, ".", 1))
 105                 return "invalid name";
 106
 107         if (len == 2 && !memcmp(d.v->d_name, "..", 2))
 108                 return "invalid name";
 109
 110         if (memchr(d.v->d_name, '/', len))
 111                 return "invalid name";
 112
 113         if (d.v->d_type != DT_SUBVOL &&
 114             le64_to_cpu(d.v->d_inum) == d.k->p.inode)
 115                 return "dirent points to own directory";
 116
 117         return NULL;
 118 }
 119
 120 void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c,
 121                          struct bkey_s_c k)
 122 {
 123         struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
 124
 125         pr_buf(out, "%.*s -> %llu type %s",
 126                bch2_dirent_name_bytes(d),
 127                d.v->d_name,
 128                d.v->d_type != DT_SUBVOL
 129                ? le64_to_cpu(d.v->d_inum)
 130                : le32_to_cpu(d.v->d_child_subvol),
 131                bch2_d_type_str(d.v->d_type));
 132 }
 133
 134 static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
 135                                 subvol_inum dir, u8 type,
 136                                 const struct qstr *name, u64 dst)
 137 {
 138         struct bkey_i_dirent *dirent;
 139         unsigned u64s = BKEY_U64s + dirent_val_u64s(name->len);
 140
 141         if (name->len > BCH_NAME_MAX)
 142                 return ERR_PTR(-ENAMETOOLONG);
 143
 144         BUG_ON(u64s > U8_MAX);
 145
 146         dirent = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
 147         if (IS_ERR(dirent))
 148                 return dirent;
 149
 150         bkey_dirent_init(&dirent->k_i);
 151         dirent->k.u64s = u64s;
 152
 153         if (type != DT_SUBVOL) {
 154                 dirent->v.d_inum = cpu_to_le64(dst);
 155         } else {
 156                 dirent->v.d_parent_subvol = cpu_to_le32(dir.subvol);
 157                 dirent->v.d_child_subvol = cpu_to_le32(dst);
 158         }
 159
 160         dirent->v.d_type = type;
 161
 162         memcpy(dirent->v.d_name, name->name, name->len);
 163         memset(dirent->v.d_name + name->len, 0,
 164                bkey_val_bytes(&dirent->k) -
 165                offsetof(struct bch_dirent, d_name) -
 166                name->len);
 167
 168         EBUG_ON(bch2_dirent_name_bytes(dirent_i_to_s_c(dirent)) != name->len);
 169
 170         return dirent;
 171 }
 172
 173 int bch2_dirent_create(struct btree_trans *trans, subvol_inum dir,
 174                        const struct bch_hash_info *hash_info,
 175                        u8 type, const struct qstr *name, u64 dst_inum,
 176                        u64 *dir_offset, int flags)
 177 {
 178         struct bkey_i_dirent *dirent;
 179         int ret;
 180
 181         dirent = dirent_create_key(trans, dir, type, name, dst_inum);
 182         ret = PTR_ERR_OR_ZERO(dirent);
 183         if (ret)
 184                 return ret;
 185
 186         ret = bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
 187                             dir, &dirent->k_i, flags);
 188         *dir_offset = dirent->k.p.offset;
 189
 190         return ret;
 191 }
 192
 193 static void dirent_copy_target(struct bkey_i_dirent *dst,
 194                                struct bkey_s_c_dirent src)
 195 {
 196         dst->v.d_inum = src.v->d_inum;
 197         dst->v.d_type = src.v->d_type;
 198 }
 199
 200 int bch2_dirent_read_target(struct btree_trans *trans, subvol_inum dir,
 201                             struct bkey_s_c_dirent d, subvol_inum *target)
 202 {
 203         struct bch_subvolume s;
 204         int ret = 0;
 205
 206         if (d.v->d_type == DT_SUBVOL &&
 207             d.v->d_parent_subvol != dir.subvol)
 208                 return 1;
 209
 210         if (likely(d.v->d_type != DT_SUBVOL)) {
 211                 target->subvol  = dir.subvol;
 212                 target->inum    = le64_to_cpu(d.v->d_inum);
 213         } else {
 214                 target->subvol  = le32_to_cpu(d.v->d_child_subvol);
 215
 216                 ret = bch2_subvolume_get(trans, target->subvol, true, BTREE_ITER_CACHED, &s);
 217
 218                 target->inum    = le64_to_cpu(s.inode);
 219         }
 220
 221         return ret;
 222 }
 223
 224 int bch2_dirent_rename(struct btree_trans *trans,
 225                 subvol_inum src_dir, struct bch_hash_info *src_hash,
 226                 subvol_inum dst_dir, struct bch_hash_info *dst_hash,
 227                 const struct qstr *src_name, subvol_inum *src_inum, u64 *src_offset,
 228                 const struct qstr *dst_name, subvol_inum *dst_inum, u64 *dst_offset,
 229                 enum bch_rename_mode mode)
 230 {
 231         struct btree_iter src_iter = { NULL };
 232         struct btree_iter dst_iter = { NULL };
 233         struct bkey_s_c old_src, old_dst = bkey_s_c_null;
 234         struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
 235         struct bpos dst_pos =
 236                 POS(dst_dir.inum, bch2_dirent_hash(dst_hash, dst_name));
 237         unsigned src_type = 0, dst_type = 0, src_update_flags = 0;
 238         int ret = 0;
 239
 240         if (src_dir.subvol != dst_dir.subvol)
 241                 return -EXDEV;
 242
 243         memset(src_inum, 0, sizeof(*src_inum));
 244         memset(dst_inum, 0, sizeof(*dst_inum));
 245
 246         /* Lookup src: */
 247         ret = bch2_hash_lookup(trans, &src_iter, bch2_dirent_hash_desc,
 248                                src_hash, src_dir, src_name,
 249                                BTREE_ITER_INTENT);
 250         if (ret)
 251                 goto out;
 252
 253         old_src = bch2_btree_iter_peek_slot(&src_iter);
 254         ret = bkey_err(old_src);
 255         if (ret)
 256                 goto out;
 257
 258         ret = bch2_dirent_read_target(trans, src_dir,
 259                         bkey_s_c_to_dirent(old_src), src_inum);
 260         if (ret)
 261                 goto out;
 262
 263         src_type = bkey_s_c_to_dirent(old_src).v->d_type;
 264
 265         if (src_type == DT_SUBVOL && mode == BCH_RENAME_EXCHANGE)
 266                 return -EOPNOTSUPP;
 267
 268
 269         /* Lookup dst: */
 270         if (mode == BCH_RENAME) {
 271                 /*
 272                  * Note that we're _not_ checking if the target already exists -
 273                  * we're relying on the VFS to do that check for us for
 274                  * correctness:
 275                  */
 276                 ret = bch2_hash_hole(trans, &dst_iter, bch2_dirent_hash_desc,
 277                                      dst_hash, dst_dir, dst_name);
 278                 if (ret)
 279                         goto out;
 280         } else {
 281                 ret = bch2_hash_lookup(trans, &dst_iter, bch2_dirent_hash_desc,
 282                                        dst_hash, dst_dir, dst_name,
 283                                        BTREE_ITER_INTENT);
 284                 if (ret)
 285                         goto out;
 286
 287                 old_dst = bch2_btree_iter_peek_slot(&dst_iter);
 288                 ret = bkey_err(old_dst);
 289                 if (ret)
 290                         goto out;
 291
 292                 ret = bch2_dirent_read_target(trans, dst_dir,
 293                                 bkey_s_c_to_dirent(old_dst), dst_inum);
 294                 if (ret)
 295                         goto out;
 296
 297                 dst_type = bkey_s_c_to_dirent(old_dst).v->d_type;
 298
 299                 if (dst_type == DT_SUBVOL)
 300                         return -EOPNOTSUPP;
 301         }
 302
 303         if (mode != BCH_RENAME_EXCHANGE)
 304                 *src_offset = dst_iter.pos.offset;
 305
 306         /* Create new dst key: */
 307         new_dst = dirent_create_key(trans, dst_dir, 0, dst_name, 0);
 308         ret = PTR_ERR_OR_ZERO(new_dst);
 309         if (ret)
 310                 goto out;
 311
 312         dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
 313         new_dst->k.p = dst_iter.pos;
 314
 315         /* Create new src key: */
 316         if (mode == BCH_RENAME_EXCHANGE) {
 317                 new_src = dirent_create_key(trans, src_dir, 0, src_name, 0);
 318                 ret = PTR_ERR_OR_ZERO(new_src);
 319                 if (ret)
 320                         goto out;
 321
 322                 dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
 323                 new_src->k.p = src_iter.pos;
 324         } else {
 325                 new_src = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
 326                 ret = PTR_ERR_OR_ZERO(new_src);
 327                 if (ret)
 328                         goto out;
 329
 330                 bkey_init(&new_src->k);
 331                 new_src->k.p = src_iter.pos;
 332
 333                 if (bkey_cmp(dst_pos, src_iter.pos) <= 0 &&
 334                     bkey_cmp(src_iter.pos, dst_iter.pos) < 0) {
 335                         /*
 336                          * We have a hash collision for the new dst key,
 337                          * and new_src - the key we're deleting - is between
 338                          * new_dst's hashed slot and the slot we're going to be
 339                          * inserting it into - oops.  This will break the hash
 340                          * table if we don't deal with it:
 341                          */
 342                         if (mode == BCH_RENAME) {
 343                                 /*
 344                                  * If we're not overwriting, we can just insert
 345                                  * new_dst at the src position:
 346                                  */
 347                                 new_src = new_dst;
 348                                 new_src->k.p = src_iter.pos;
 349                                 goto out_set_src;
 350                         } else {
 351                                 /* If we're overwriting, we can't insert new_dst
 352                                  * at a different slot because it has to
 353                                  * overwrite old_dst - just make sure to use a
 354                                  * whiteout when deleting src:
 355                                  */
 356                                 new_src->k.type = KEY_TYPE_hash_whiteout;
 357                         }
 358                 } else {
 359                         /* Check if we need a whiteout to delete src: */
 360                         ret = bch2_hash_needs_whiteout(trans, bch2_dirent_hash_desc,
 361                                                        src_hash, &src_iter);
 362                         if (ret < 0)
 363                                 goto out;
 364
 365                         if (ret)
 366                                 new_src->k.type = KEY_TYPE_hash_whiteout;
 367                 }
 368         }
 369
 370         ret = bch2_trans_update(trans, &dst_iter, &new_dst->k_i, 0);
 371         if (ret)
 372                 goto out;
 373 out_set_src:
 374
 375         /*
 376          * If we're deleting a subvolume, we need to really delete the dirent,
 377          * not just emit a whiteout in the current snapshot:
 378          */
 379         if (src_type == DT_SUBVOL) {
 380                 bch2_btree_iter_set_snapshot(&src_iter, old_src.k->p.snapshot);
 381                 ret = bch2_btree_iter_traverse(&src_iter);
 382                 if (ret)
 383                         goto out;
 384
 385                 new_src->k.p = src_iter.pos;
 386                 src_update_flags |= BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE;
 387         }
 388
 389         ret = bch2_trans_update(trans, &src_iter, &new_src->k_i, src_update_flags);
 390         if (ret)
 391                 goto out;
 392
 393         if (mode == BCH_RENAME_EXCHANGE)
 394                 *src_offset = new_src->k.p.offset;
 395         *dst_offset = new_dst->k.p.offset;
 396 out:
 397         bch2_trans_iter_exit(trans, &src_iter);
 398         bch2_trans_iter_exit(trans, &dst_iter);
 399         return ret;
 400 }
 401
 402 int __bch2_dirent_lookup_trans(struct btree_trans *trans,
 403                                struct btree_iter *iter,
 404                                subvol_inum dir,
 405                                const struct bch_hash_info *hash_info,
 406                                const struct qstr *name, subvol_inum *inum,
 407                                unsigned flags)
 408 {
 409         struct bkey_s_c k;
 410         struct bkey_s_c_dirent d;
 411         u32 snapshot;
 412         int ret;
 413
 414         ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
 415         if (ret)
 416                 return ret;
 417
 418         ret = bch2_hash_lookup(trans, iter, bch2_dirent_hash_desc,
 419                                hash_info, dir, name, flags);
 420         if (ret)
 421                 return ret;
 422
 423         k = bch2_btree_iter_peek_slot(iter);
 424         ret = bkey_err(k);
 425         if (ret)
 426                 goto err;
 427
 428         d = bkey_s_c_to_dirent(k);
 429
 430         ret = bch2_dirent_read_target(trans, dir, d, inum);
 431         if (ret > 0)
 432                 ret = -ENOENT;
 433 err:
 434         if (ret)
 435                 bch2_trans_iter_exit(trans, iter);
 436
 437         return ret;
 438 }
 439
 440 u64 bch2_dirent_lookup(struct bch_fs *c, subvol_inum dir,
 441                        const struct bch_hash_info *hash_info,
 442                        const struct qstr *name, subvol_inum *inum)
 443 {
 444         struct btree_trans trans;
 445         struct btree_iter iter;
 446         int ret;
 447
 448         bch2_trans_init(&trans, c, 0, 0);
 449 retry:
 450         bch2_trans_begin(&trans);
 451
 452         ret = __bch2_dirent_lookup_trans(&trans, &iter, dir, hash_info,
 453                                           name, inum, 0);
 454         if (ret == -EINTR)
 455                 goto retry;
 456         if (!ret)
 457                 bch2_trans_iter_exit(&trans, &iter);
 458         bch2_trans_exit(&trans);
 459         return ret;
 460 }
 461
 462 int bch2_empty_dir_trans(struct btree_trans *trans, subvol_inum dir)
 463 {
 464         struct btree_iter iter;
 465         struct bkey_s_c k;
 466         u32 snapshot;
 467         int ret;
 468
 469         ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
 470         if (ret)
 471                 return ret;
 472
 473         for_each_btree_key_norestart(trans, iter, BTREE_ID_dirents,
 474                            SPOS(dir.inum, 0, snapshot), 0, k, ret) {
 475                 if (k.k->p.inode > dir.inum)
 476                         break;
 477
 478                 if (k.k->type == KEY_TYPE_dirent) {
 479                         ret = -ENOTEMPTY;
 480                         break;
 481                 }
 482         }
 483         bch2_trans_iter_exit(trans, &iter);
 484
 485         return ret;
 486 }
 487
 488 int bch2_readdir(struct bch_fs *c, subvol_inum inum, struct dir_context *ctx)
 489 {
 490         struct btree_trans trans;
 491         struct btree_iter iter;
 492         struct bkey_s_c k;
 493         struct bkey_s_c_dirent dirent;
 494         subvol_inum target;
 495         u32 snapshot;
 496         int ret;
 497
 498         bch2_trans_init(&trans, c, 0, 0);
 499 retry:
 500         bch2_trans_begin(&trans);
 501
 502         ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot);
 503         if (ret)
 504                 goto err;
 505
 506         for_each_btree_key_norestart(&trans, iter, BTREE_ID_dirents,
 507                            SPOS(inum.inum, ctx->pos, snapshot), 0, k, ret) {
 508                 if (k.k->p.inode > inum.inum)
 509                         break;
 510
 511                 if (k.k->type != KEY_TYPE_dirent)
 512                         continue;
 513
 514                 dirent = bkey_s_c_to_dirent(k);
 515
 516                 ret = bch2_dirent_read_target(&trans, inum, dirent, &target);
 517                 if (ret < 0)
 518                         break;
 519                 if (ret)
 520                         continue;
 521
 522                 /*
 523                  * XXX: dir_emit() can fault and block, while we're holding
 524                  * locks
 525                  */
 526                 ctx->pos = dirent.k->p.offset;
 527                 if (!dir_emit(ctx, dirent.v->d_name,
 528                               bch2_dirent_name_bytes(dirent),
 529                               target.inum,
 530                               vfs_d_type(dirent.v->d_type)))
 531                         break;
 532                 ctx->pos = dirent.k->p.offset + 1;
 533
 534                 /*
 535                  * read_target looks up subvolumes, we can overflow paths if the
 536                  * directory has many subvolumes in it
 537                  */
 538                 ret = btree_trans_too_many_iters(&trans);
 539                 if (ret)
 540                         break;
 541         }
 542         bch2_trans_iter_exit(&trans, &iter);
 543 err:
 544         if (ret == -EINTR)
 545                 goto retry;
 546
 547         bch2_trans_exit(&trans);
 548
 549         return ret;
 550 }