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_buf.h"
   5 #include "bkey_methods.h"
   6 #include "btree_update.h"
   7 #include "extents.h"
   8 #include "dirent.h"
   9 #include "fs.h"
  10 #include "keylist.h"
  11 #include "str_hash.h"
  12 #include "subvolume.h"
  13
  14 #include <linux/dcache.h>
  15
  16 static unsigned bch2_dirent_name_bytes(struct bkey_s_c_dirent d)
  17 {
  18         unsigned bkey_u64s = bkey_val_u64s(d.k);
  19         unsigned bkey_bytes = bkey_u64s * sizeof(u64);
  20         u64 last_u64 = ((u64*)d.v)[bkey_u64s - 1];
  21 #if CPU_BIG_ENDIAN
  22         unsigned trailing_nuls = last_u64 ? __builtin_ctzll(last_u64) / 8 : 64 / 8;
  23 #else
  24         unsigned trailing_nuls = last_u64 ? __builtin_clzll(last_u64) / 8 : 64 / 8;
  25 #endif
  26
  27         return bkey_bytes -
  28                 offsetof(struct bch_dirent, d_name) -
  29                 trailing_nuls;
  30 }
  31
  32 struct qstr bch2_dirent_get_name(struct bkey_s_c_dirent d)
  33 {
  34         return (struct qstr) QSTR_INIT(d.v->d_name, bch2_dirent_name_bytes(d));
  35 }
  36
  37 static u64 bch2_dirent_hash(const struct bch_hash_info *info,
  38                             const struct qstr *name)
  39 {
  40         struct bch_str_hash_ctx ctx;
  41
  42         bch2_str_hash_init(&ctx, info);
  43         bch2_str_hash_update(&ctx, info, name->name, name->len);
  44
  45         /* [0,2) reserved for dots */
  46         return max_t(u64, bch2_str_hash_end(&ctx, info), 2);
  47 }
  48
  49 static u64 dirent_hash_key(const struct bch_hash_info *info, const void *key)
  50 {
  51         return bch2_dirent_hash(info, key);
  52 }
  53
  54 static u64 dirent_hash_bkey(const struct bch_hash_info *info, struct bkey_s_c k)
  55 {
  56         struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
  57         struct qstr name = bch2_dirent_get_name(d);
  58
  59         return bch2_dirent_hash(info, &name);
  60 }
  61
  62 static bool dirent_cmp_key(struct bkey_s_c _l, const void *_r)
  63 {
  64         struct bkey_s_c_dirent l = bkey_s_c_to_dirent(_l);
  65         const struct qstr l_name = bch2_dirent_get_name(l);
  66         const struct qstr *r_name = _r;
  67
  68         return l_name.len - r_name->len ?: memcmp(l_name.name, r_name->name, l_name.len);
  69 }
  70
  71 static bool dirent_cmp_bkey(struct bkey_s_c _l, struct bkey_s_c _r)
  72 {
  73         struct bkey_s_c_dirent l = bkey_s_c_to_dirent(_l);
  74         struct bkey_s_c_dirent r = bkey_s_c_to_dirent(_r);
  75         const struct qstr l_name = bch2_dirent_get_name(l);
  76         const struct qstr r_name = bch2_dirent_get_name(r);
  77
  78         return l_name.len - r_name.len ?: memcmp(l_name.name, r_name.name, l_name.len);
  79 }
  80
  81 static bool dirent_is_visible(subvol_inum inum, struct bkey_s_c k)
  82 {
  83         struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
  84
  85         if (d.v->d_type == DT_SUBVOL)
  86                 return le32_to_cpu(d.v->d_parent_subvol) == inum.subvol;
  87         return true;
  88 }
  89
  90 const struct bch_hash_desc bch2_dirent_hash_desc = {
  91         .btree_id       = BTREE_ID_dirents,
  92         .key_type       = KEY_TYPE_dirent,
  93         .hash_key       = dirent_hash_key,
  94         .hash_bkey      = dirent_hash_bkey,
  95         .cmp_key        = dirent_cmp_key,
  96         .cmp_bkey       = dirent_cmp_bkey,
  97         .is_visible     = dirent_is_visible,
  98 };
  99
 100 int bch2_dirent_invalid(struct bch_fs *c, struct bkey_s_c k,
 101                         enum bkey_invalid_flags flags,
 102                         struct printbuf *err)
 103 {
 104         struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
 105         struct qstr d_name = bch2_dirent_get_name(d);
 106         int ret = 0;
 107
 108         bkey_fsck_err_on(!d_name.len, c, err,
 109                          dirent_empty_name,
 110                          "empty name");
 111
 112         bkey_fsck_err_on(bkey_val_u64s(k.k) > dirent_val_u64s(d_name.len), c, err,
 113                          dirent_val_too_big,
 114                          "value too big (%zu > %u)",
 115                          bkey_val_u64s(k.k), dirent_val_u64s(d_name.len));
 116
 117         /*
 118          * Check new keys don't exceed the max length
 119          * (older keys may be larger.)
 120          */
 121         bkey_fsck_err_on((flags & BKEY_INVALID_COMMIT) && d_name.len > BCH_NAME_MAX, c, err,
 122                          dirent_name_too_long,
 123                          "dirent name too big (%u > %u)",
 124                          d_name.len, BCH_NAME_MAX);
 125
 126         bkey_fsck_err_on(d_name.len != strnlen(d_name.name, d_name.len), c, err,
 127                          dirent_name_embedded_nul,
 128                          "dirent has stray data after name's NUL");
 129
 130         bkey_fsck_err_on((d_name.len == 1 && !memcmp(d_name.name, ".", 1)) ||
 131                          (d_name.len == 2 && !memcmp(d_name.name, "..", 2)), c, err,
 132                          dirent_name_dot_or_dotdot,
 133                          "invalid name");
 134
 135         bkey_fsck_err_on(memchr(d_name.name, '/', d_name.len), c, err,
 136                          dirent_name_has_slash,
 137                          "name with /");
 138
 139         bkey_fsck_err_on(d.v->d_type != DT_SUBVOL &&
 140                          le64_to_cpu(d.v->d_inum) == d.k->p.inode, c, err,
 141                          dirent_to_itself,
 142                          "dirent points to own directory");
 143 fsck_err:
 144         return ret;
 145 }
 146
 147 void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c,
 148                          struct bkey_s_c k)
 149 {
 150         struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
 151         struct qstr d_name = bch2_dirent_get_name(d);
 152
 153         prt_printf(out, "%.*s -> %llu type %s",
 154                d_name.len,
 155                d_name.name,
 156                d.v->d_type != DT_SUBVOL
 157                ? le64_to_cpu(d.v->d_inum)
 158                : le32_to_cpu(d.v->d_child_subvol),
 159                bch2_d_type_str(d.v->d_type));
 160 }
 161
 162 static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
 163                                 subvol_inum dir, u8 type,
 164                                 const struct qstr *name, u64 dst)
 165 {
 166         struct bkey_i_dirent *dirent;
 167         unsigned u64s = BKEY_U64s + dirent_val_u64s(name->len);
 168
 169         if (name->len > BCH_NAME_MAX)
 170                 return ERR_PTR(-ENAMETOOLONG);
 171
 172         BUG_ON(u64s > U8_MAX);
 173
 174         dirent = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
 175         if (IS_ERR(dirent))
 176                 return dirent;
 177
 178         bkey_dirent_init(&dirent->k_i);
 179         dirent->k.u64s = u64s;
 180
 181         if (type != DT_SUBVOL) {
 182                 dirent->v.d_inum = cpu_to_le64(dst);
 183         } else {
 184                 dirent->v.d_parent_subvol = cpu_to_le32(dir.subvol);
 185                 dirent->v.d_child_subvol = cpu_to_le32(dst);
 186         }
 187
 188         dirent->v.d_type = type;
 189
 190         memcpy(dirent->v.d_name, name->name, name->len);
 191         memset(dirent->v.d_name + name->len, 0,
 192                bkey_val_bytes(&dirent->k) -
 193                offsetof(struct bch_dirent, d_name) -
 194                name->len);
 195
 196         EBUG_ON(bch2_dirent_name_bytes(dirent_i_to_s_c(dirent)) != name->len);
 197
 198         return dirent;
 199 }
 200
 201 int bch2_dirent_create(struct btree_trans *trans, subvol_inum dir,
 202                        const struct bch_hash_info *hash_info,
 203                        u8 type, const struct qstr *name, u64 dst_inum,
 204                        u64 *dir_offset,
 205                        bch_str_hash_flags_t str_hash_flags)
 206 {
 207         struct bkey_i_dirent *dirent;
 208         int ret;
 209
 210         dirent = dirent_create_key(trans, dir, type, name, dst_inum);
 211         ret = PTR_ERR_OR_ZERO(dirent);
 212         if (ret)
 213                 return ret;
 214
 215         ret = bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
 216                             dir, &dirent->k_i, str_hash_flags);
 217         *dir_offset = dirent->k.p.offset;
 218
 219         return ret;
 220 }
 221
 222 static void dirent_copy_target(struct bkey_i_dirent *dst,
 223                                struct bkey_s_c_dirent src)
 224 {
 225         dst->v.d_inum = src.v->d_inum;
 226         dst->v.d_type = src.v->d_type;
 227 }
 228
 229 int bch2_dirent_read_target(struct btree_trans *trans, subvol_inum dir,
 230                             struct bkey_s_c_dirent d, subvol_inum *target)
 231 {
 232         struct bch_subvolume s;
 233         int ret = 0;
 234
 235         if (d.v->d_type == DT_SUBVOL &&
 236             le32_to_cpu(d.v->d_parent_subvol) != dir.subvol)
 237                 return 1;
 238
 239         if (likely(d.v->d_type != DT_SUBVOL)) {
 240                 target->subvol  = dir.subvol;
 241                 target->inum    = le64_to_cpu(d.v->d_inum);
 242         } else {
 243                 target->subvol  = le32_to_cpu(d.v->d_child_subvol);
 244
 245                 ret = bch2_subvolume_get(trans, target->subvol, true, BTREE_ITER_CACHED, &s);
 246
 247                 target->inum    = le64_to_cpu(s.inode);
 248         }
 249
 250         return ret;
 251 }
 252
 253 int bch2_dirent_rename(struct btree_trans *trans,
 254                 subvol_inum src_dir, struct bch_hash_info *src_hash,
 255                 subvol_inum dst_dir, struct bch_hash_info *dst_hash,
 256                 const struct qstr *src_name, subvol_inum *src_inum, u64 *src_offset,
 257                 const struct qstr *dst_name, subvol_inum *dst_inum, u64 *dst_offset,
 258                 enum bch_rename_mode mode)
 259 {
 260         struct btree_iter src_iter = { NULL };
 261         struct btree_iter dst_iter = { NULL };
 262         struct bkey_s_c old_src, old_dst = bkey_s_c_null;
 263         struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
 264         struct bpos dst_pos =
 265                 POS(dst_dir.inum, bch2_dirent_hash(dst_hash, dst_name));
 266         unsigned src_type = 0, dst_type = 0, src_update_flags = 0;
 267         int ret = 0;
 268
 269         if (src_dir.subvol != dst_dir.subvol)
 270                 return -EXDEV;
 271
 272         memset(src_inum, 0, sizeof(*src_inum));
 273         memset(dst_inum, 0, sizeof(*dst_inum));
 274
 275         /* Lookup src: */
 276         ret = bch2_hash_lookup(trans, &src_iter, bch2_dirent_hash_desc,
 277                                src_hash, src_dir, src_name,
 278                                BTREE_ITER_INTENT);
 279         if (ret)
 280                 goto out;
 281
 282         old_src = bch2_btree_iter_peek_slot(&src_iter);
 283         ret = bkey_err(old_src);
 284         if (ret)
 285                 goto out;
 286
 287         ret = bch2_dirent_read_target(trans, src_dir,
 288                         bkey_s_c_to_dirent(old_src), src_inum);
 289         if (ret)
 290                 goto out;
 291
 292         src_type = bkey_s_c_to_dirent(old_src).v->d_type;
 293
 294         if (src_type == DT_SUBVOL && mode == BCH_RENAME_EXCHANGE)
 295                 return -EOPNOTSUPP;
 296
 297
 298         /* Lookup dst: */
 299         if (mode == BCH_RENAME) {
 300                 /*
 301                  * Note that we're _not_ checking if the target already exists -
 302                  * we're relying on the VFS to do that check for us for
 303                  * correctness:
 304                  */
 305                 ret = bch2_hash_hole(trans, &dst_iter, bch2_dirent_hash_desc,
 306                                      dst_hash, dst_dir, dst_name);
 307                 if (ret)
 308                         goto out;
 309         } else {
 310                 ret = bch2_hash_lookup(trans, &dst_iter, bch2_dirent_hash_desc,
 311                                        dst_hash, dst_dir, dst_name,
 312                                        BTREE_ITER_INTENT);
 313                 if (ret)
 314                         goto out;
 315
 316                 old_dst = bch2_btree_iter_peek_slot(&dst_iter);
 317                 ret = bkey_err(old_dst);
 318                 if (ret)
 319                         goto out;
 320
 321                 ret = bch2_dirent_read_target(trans, dst_dir,
 322                                 bkey_s_c_to_dirent(old_dst), dst_inum);
 323                 if (ret)
 324                         goto out;
 325
 326                 dst_type = bkey_s_c_to_dirent(old_dst).v->d_type;
 327
 328                 if (dst_type == DT_SUBVOL)
 329                         return -EOPNOTSUPP;
 330         }
 331
 332         if (mode != BCH_RENAME_EXCHANGE)
 333                 *src_offset = dst_iter.pos.offset;
 334
 335         /* Create new dst key: */
 336         new_dst = dirent_create_key(trans, dst_dir, 0, dst_name, 0);
 337         ret = PTR_ERR_OR_ZERO(new_dst);
 338         if (ret)
 339                 goto out;
 340
 341         dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
 342         new_dst->k.p = dst_iter.pos;
 343
 344         /* Create new src key: */
 345         if (mode == BCH_RENAME_EXCHANGE) {
 346                 new_src = dirent_create_key(trans, src_dir, 0, src_name, 0);
 347                 ret = PTR_ERR_OR_ZERO(new_src);
 348                 if (ret)
 349                         goto out;
 350
 351                 dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
 352                 new_src->k.p = src_iter.pos;
 353         } else {
 354                 new_src = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
 355                 ret = PTR_ERR_OR_ZERO(new_src);
 356                 if (ret)
 357                         goto out;
 358
 359                 bkey_init(&new_src->k);
 360                 new_src->k.p = src_iter.pos;
 361
 362                 if (bkey_le(dst_pos, src_iter.pos) &&
 363                     bkey_lt(src_iter.pos, dst_iter.pos)) {
 364                         /*
 365                          * We have a hash collision for the new dst key,
 366                          * and new_src - the key we're deleting - is between
 367                          * new_dst's hashed slot and the slot we're going to be
 368                          * inserting it into - oops.  This will break the hash
 369                          * table if we don't deal with it:
 370                          */
 371                         if (mode == BCH_RENAME) {
 372                                 /*
 373                                  * If we're not overwriting, we can just insert
 374                                  * new_dst at the src position:
 375                                  */
 376                                 new_src = new_dst;
 377                                 new_src->k.p = src_iter.pos;
 378                                 goto out_set_src;
 379                         } else {
 380                                 /* If we're overwriting, we can't insert new_dst
 381                                  * at a different slot because it has to
 382                                  * overwrite old_dst - just make sure to use a
 383                                  * whiteout when deleting src:
 384                                  */
 385                                 new_src->k.type = KEY_TYPE_hash_whiteout;
 386                         }
 387                 } else {
 388                         /* Check if we need a whiteout to delete src: */
 389                         ret = bch2_hash_needs_whiteout(trans, bch2_dirent_hash_desc,
 390                                                        src_hash, &src_iter);
 391                         if (ret < 0)
 392                                 goto out;
 393
 394                         if (ret)
 395                                 new_src->k.type = KEY_TYPE_hash_whiteout;
 396                 }
 397         }
 398
 399         ret = bch2_trans_update(trans, &dst_iter, &new_dst->k_i, 0);
 400         if (ret)
 401                 goto out;
 402 out_set_src:
 403
 404         /*
 405          * If we're deleting a subvolume, we need to really delete the dirent,
 406          * not just emit a whiteout in the current snapshot:
 407          */
 408         if (src_type == DT_SUBVOL) {
 409                 bch2_btree_iter_set_snapshot(&src_iter, old_src.k->p.snapshot);
 410                 ret = bch2_btree_iter_traverse(&src_iter);
 411                 if (ret)
 412                         goto out;
 413
 414                 new_src->k.p = src_iter.pos;
 415                 src_update_flags |= BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE;
 416         }
 417
 418         ret = bch2_trans_update(trans, &src_iter, &new_src->k_i, src_update_flags);
 419         if (ret)
 420                 goto out;
 421
 422         if (mode == BCH_RENAME_EXCHANGE)
 423                 *src_offset = new_src->k.p.offset;
 424         *dst_offset = new_dst->k.p.offset;
 425 out:
 426         bch2_trans_iter_exit(trans, &src_iter);
 427         bch2_trans_iter_exit(trans, &dst_iter);
 428         return ret;
 429 }
 430
 431 int __bch2_dirent_lookup_trans(struct btree_trans *trans,
 432                                struct btree_iter *iter,
 433                                subvol_inum dir,
 434                                const struct bch_hash_info *hash_info,
 435                                const struct qstr *name, subvol_inum *inum,
 436                                unsigned flags)
 437 {
 438         struct bkey_s_c k;
 439         struct bkey_s_c_dirent d;
 440         u32 snapshot;
 441         int ret;
 442
 443         ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
 444         if (ret)
 445                 return ret;
 446
 447         ret = bch2_hash_lookup(trans, iter, bch2_dirent_hash_desc,
 448                                hash_info, dir, name, flags);
 449         if (ret)
 450                 return ret;
 451
 452         k = bch2_btree_iter_peek_slot(iter);
 453         ret = bkey_err(k);
 454         if (ret)
 455                 goto err;
 456
 457         d = bkey_s_c_to_dirent(k);
 458
 459         ret = bch2_dirent_read_target(trans, dir, d, inum);
 460         if (ret > 0)
 461                 ret = -ENOENT;
 462 err:
 463         if (ret)
 464                 bch2_trans_iter_exit(trans, iter);
 465
 466         return ret;
 467 }
 468
 469 u64 bch2_dirent_lookup(struct bch_fs *c, subvol_inum dir,
 470                        const struct bch_hash_info *hash_info,
 471                        const struct qstr *name, subvol_inum *inum)
 472 {
 473         struct btree_trans *trans = bch2_trans_get(c);
 474         struct btree_iter iter;
 475         int ret;
 476 retry:
 477         bch2_trans_begin(trans);
 478
 479         ret = __bch2_dirent_lookup_trans(trans, &iter, dir, hash_info,
 480                                           name, inum, 0);
 481         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 482                 goto retry;
 483         if (!ret)
 484                 bch2_trans_iter_exit(trans, &iter);
 485         bch2_trans_put(trans);
 486         return ret;
 487 }
 488
 489 int bch2_empty_dir_trans(struct btree_trans *trans, subvol_inum dir)
 490 {
 491         struct btree_iter iter;
 492         struct bkey_s_c k;
 493         u32 snapshot;
 494         int ret;
 495
 496         ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
 497         if (ret)
 498                 return ret;
 499
 500         for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
 501                            SPOS(dir.inum, 0, snapshot),
 502                            POS(dir.inum, U64_MAX), 0, k, ret)
 503                 if (k.k->type == KEY_TYPE_dirent) {
 504                         ret = -ENOTEMPTY;
 505                         break;
 506                 }
 507         bch2_trans_iter_exit(trans, &iter);
 508
 509         return ret;
 510 }
 511
 512 int bch2_readdir(struct bch_fs *c, subvol_inum inum, struct dir_context *ctx)
 513 {
 514         struct btree_trans *trans = bch2_trans_get(c);
 515         struct btree_iter iter;
 516         struct bkey_s_c k;
 517         struct bkey_s_c_dirent dirent;
 518         subvol_inum target;
 519         u32 snapshot;
 520         struct bkey_buf sk;
 521         struct qstr name;
 522         int ret;
 523
 524         bch2_bkey_buf_init(&sk);
 525 retry:
 526         bch2_trans_begin(trans);
 527
 528         ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
 529         if (ret)
 530                 goto err;
 531
 532         for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
 533                            SPOS(inum.inum, ctx->pos, snapshot),
 534                            POS(inum.inum, U64_MAX), 0, k, ret) {
 535                 if (k.k->type != KEY_TYPE_dirent)
 536                         continue;
 537
 538                 dirent = bkey_s_c_to_dirent(k);
 539
 540                 ret = bch2_dirent_read_target(trans, inum, dirent, &target);
 541                 if (ret < 0)
 542                         break;
 543                 if (ret)
 544                         continue;
 545
 546                 /* dir_emit() can fault and block: */
 547                 bch2_bkey_buf_reassemble(&sk, c, k);
 548                 dirent = bkey_i_to_s_c_dirent(sk.k);
 549                 bch2_trans_unlock(trans);
 550
 551                 name = bch2_dirent_get_name(dirent);
 552
 553                 ctx->pos = dirent.k->p.offset;
 554                 if (!dir_emit(ctx, name.name,
 555                               name.len,
 556                               target.inum,
 557                               vfs_d_type(dirent.v->d_type)))
 558                         break;
 559                 ctx->pos = dirent.k->p.offset + 1;
 560
 561                 /*
 562                  * read_target looks up subvolumes, we can overflow paths if the
 563                  * directory has many subvolumes in it
 564                  */
 565                 ret = btree_trans_too_many_iters(trans);
 566                 if (ret)
 567                         break;
 568         }
 569         bch2_trans_iter_exit(trans, &iter);
 570 err:
 571         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 572                 goto retry;
 573
 574         bch2_trans_put(trans);
 575         bch2_bkey_buf_exit(&sk, c);
 576
 577         return ret;
 578 }