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 !qstr_eq(l_name, *r_name);
  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 !qstr_eq(l_name, r_name);
  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, struct bkey_s_c k)
 148 {
 149         struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
 150         struct qstr d_name = bch2_dirent_get_name(d);
 151
 152         prt_printf(out, "%.*s -> ", d_name.len, d_name.name);
 153
 154         if (d.v->d_type != DT_SUBVOL)
 155                 prt_printf(out, "%llu", le64_to_cpu(d.v->d_inum));
 156         else
 157                 prt_printf(out, "%u -> %u",
 158                            le32_to_cpu(d.v->d_parent_subvol),
 159                            le32_to_cpu(d.v->d_child_subvol));
 160
 161         prt_printf(out, " type %s", bch2_d_type_str(d.v->d_type));
 162 }
 163
 164 static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
 165                                 subvol_inum dir, u8 type,
 166                                 const struct qstr *name, u64 dst)
 167 {
 168         struct bkey_i_dirent *dirent;
 169         unsigned u64s = BKEY_U64s + dirent_val_u64s(name->len);
 170
 171         if (name->len > BCH_NAME_MAX)
 172                 return ERR_PTR(-ENAMETOOLONG);
 173
 174         BUG_ON(u64s > U8_MAX);
 175
 176         dirent = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
 177         if (IS_ERR(dirent))
 178                 return dirent;
 179
 180         bkey_dirent_init(&dirent->k_i);
 181         dirent->k.u64s = u64s;
 182
 183         if (type != DT_SUBVOL) {
 184                 dirent->v.d_inum = cpu_to_le64(dst);
 185         } else {
 186                 dirent->v.d_parent_subvol = cpu_to_le32(dir.subvol);
 187                 dirent->v.d_child_subvol = cpu_to_le32(dst);
 188         }
 189
 190         dirent->v.d_type = type;
 191
 192         memcpy(dirent->v.d_name, name->name, name->len);
 193         memset(dirent->v.d_name + name->len, 0,
 194                bkey_val_bytes(&dirent->k) -
 195                offsetof(struct bch_dirent, d_name) -
 196                name->len);
 197
 198         EBUG_ON(bch2_dirent_name_bytes(dirent_i_to_s_c(dirent)) != name->len);
 199
 200         return dirent;
 201 }
 202
 203 int bch2_dirent_create_snapshot(struct btree_trans *trans,
 204                         u64 dir, u32 snapshot,
 205                         const struct bch_hash_info *hash_info,
 206                         u8 type, const struct qstr *name, u64 dst_inum,
 207                         u64 *dir_offset,
 208                         bch_str_hash_flags_t str_hash_flags)
 209 {
 210         subvol_inum zero_inum = { 0 };
 211         struct bkey_i_dirent *dirent;
 212         int ret;
 213
 214         dirent = dirent_create_key(trans, zero_inum, type, name, dst_inum);
 215         ret = PTR_ERR_OR_ZERO(dirent);
 216         if (ret)
 217                 return ret;
 218
 219         dirent->k.p.inode       = dir;
 220         dirent->k.p.snapshot    = snapshot;
 221
 222         ret = bch2_hash_set_snapshot(trans, bch2_dirent_hash_desc, hash_info,
 223                                      zero_inum, snapshot,
 224                                      &dirent->k_i, str_hash_flags,
 225                                      BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 226         *dir_offset = dirent->k.p.offset;
 227
 228         return ret;
 229 }
 230
 231 int bch2_dirent_create(struct btree_trans *trans, subvol_inum dir,
 232                        const struct bch_hash_info *hash_info,
 233                        u8 type, const struct qstr *name, u64 dst_inum,
 234                        u64 *dir_offset,
 235                        bch_str_hash_flags_t str_hash_flags)
 236 {
 237         struct bkey_i_dirent *dirent;
 238         int ret;
 239
 240         dirent = dirent_create_key(trans, dir, type, name, dst_inum);
 241         ret = PTR_ERR_OR_ZERO(dirent);
 242         if (ret)
 243                 return ret;
 244
 245         ret = bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
 246                             dir, &dirent->k_i, str_hash_flags);
 247         *dir_offset = dirent->k.p.offset;
 248
 249         return ret;
 250 }
 251
 252 static void dirent_copy_target(struct bkey_i_dirent *dst,
 253                                struct bkey_s_c_dirent src)
 254 {
 255         dst->v.d_inum = src.v->d_inum;
 256         dst->v.d_type = src.v->d_type;
 257 }
 258
 259 int bch2_dirent_read_target(struct btree_trans *trans, subvol_inum dir,
 260                             struct bkey_s_c_dirent d, subvol_inum *target)
 261 {
 262         struct bch_subvolume s;
 263         int ret = 0;
 264
 265         if (d.v->d_type == DT_SUBVOL &&
 266             le32_to_cpu(d.v->d_parent_subvol) != dir.subvol)
 267                 return 1;
 268
 269         if (likely(d.v->d_type != DT_SUBVOL)) {
 270                 target->subvol  = dir.subvol;
 271                 target->inum    = le64_to_cpu(d.v->d_inum);
 272         } else {
 273                 target->subvol  = le32_to_cpu(d.v->d_child_subvol);
 274
 275                 ret = bch2_subvolume_get(trans, target->subvol, true, BTREE_ITER_CACHED, &s);
 276
 277                 target->inum    = le64_to_cpu(s.inode);
 278         }
 279
 280         return ret;
 281 }
 282
 283 int bch2_dirent_rename(struct btree_trans *trans,
 284                 subvol_inum src_dir, struct bch_hash_info *src_hash,
 285                 subvol_inum dst_dir, struct bch_hash_info *dst_hash,
 286                 const struct qstr *src_name, subvol_inum *src_inum, u64 *src_offset,
 287                 const struct qstr *dst_name, subvol_inum *dst_inum, u64 *dst_offset,
 288                 enum bch_rename_mode mode)
 289 {
 290         struct btree_iter src_iter = { NULL };
 291         struct btree_iter dst_iter = { NULL };
 292         struct bkey_s_c old_src, old_dst = bkey_s_c_null;
 293         struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
 294         struct bpos dst_pos =
 295                 POS(dst_dir.inum, bch2_dirent_hash(dst_hash, dst_name));
 296         unsigned src_type = 0, dst_type = 0, src_update_flags = 0;
 297         int ret = 0;
 298
 299         if (src_dir.subvol != dst_dir.subvol)
 300                 return -EXDEV;
 301
 302         memset(src_inum, 0, sizeof(*src_inum));
 303         memset(dst_inum, 0, sizeof(*dst_inum));
 304
 305         /* Lookup src: */
 306         ret = bch2_hash_lookup(trans, &src_iter, bch2_dirent_hash_desc,
 307                                src_hash, src_dir, src_name,
 308                                BTREE_ITER_INTENT);
 309         if (ret)
 310                 goto out;
 311
 312         old_src = bch2_btree_iter_peek_slot(&src_iter);
 313         ret = bkey_err(old_src);
 314         if (ret)
 315                 goto out;
 316
 317         ret = bch2_dirent_read_target(trans, src_dir,
 318                         bkey_s_c_to_dirent(old_src), src_inum);
 319         if (ret)
 320                 goto out;
 321
 322         src_type = bkey_s_c_to_dirent(old_src).v->d_type;
 323
 324         if (src_type == DT_SUBVOL && mode == BCH_RENAME_EXCHANGE)
 325                 return -EOPNOTSUPP;
 326
 327
 328         /* Lookup dst: */
 329         if (mode == BCH_RENAME) {
 330                 /*
 331                  * Note that we're _not_ checking if the target already exists -
 332                  * we're relying on the VFS to do that check for us for
 333                  * correctness:
 334                  */
 335                 ret = bch2_hash_hole(trans, &dst_iter, bch2_dirent_hash_desc,
 336                                      dst_hash, dst_dir, dst_name);
 337                 if (ret)
 338                         goto out;
 339         } else {
 340                 ret = bch2_hash_lookup(trans, &dst_iter, bch2_dirent_hash_desc,
 341                                        dst_hash, dst_dir, dst_name,
 342                                        BTREE_ITER_INTENT);
 343                 if (ret)
 344                         goto out;
 345
 346                 old_dst = bch2_btree_iter_peek_slot(&dst_iter);
 347                 ret = bkey_err(old_dst);
 348                 if (ret)
 349                         goto out;
 350
 351                 ret = bch2_dirent_read_target(trans, dst_dir,
 352                                 bkey_s_c_to_dirent(old_dst), dst_inum);
 353                 if (ret)
 354                         goto out;
 355
 356                 dst_type = bkey_s_c_to_dirent(old_dst).v->d_type;
 357
 358                 if (dst_type == DT_SUBVOL)
 359                         return -EOPNOTSUPP;
 360         }
 361
 362         if (mode != BCH_RENAME_EXCHANGE)
 363                 *src_offset = dst_iter.pos.offset;
 364
 365         /* Create new dst key: */
 366         new_dst = dirent_create_key(trans, dst_dir, 0, dst_name, 0);
 367         ret = PTR_ERR_OR_ZERO(new_dst);
 368         if (ret)
 369                 goto out;
 370
 371         dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
 372         new_dst->k.p = dst_iter.pos;
 373
 374         /* Create new src key: */
 375         if (mode == BCH_RENAME_EXCHANGE) {
 376                 new_src = dirent_create_key(trans, src_dir, 0, src_name, 0);
 377                 ret = PTR_ERR_OR_ZERO(new_src);
 378                 if (ret)
 379                         goto out;
 380
 381                 dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
 382                 new_src->k.p = src_iter.pos;
 383         } else {
 384                 new_src = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
 385                 ret = PTR_ERR_OR_ZERO(new_src);
 386                 if (ret)
 387                         goto out;
 388
 389                 bkey_init(&new_src->k);
 390                 new_src->k.p = src_iter.pos;
 391
 392                 if (bkey_le(dst_pos, src_iter.pos) &&
 393                     bkey_lt(src_iter.pos, dst_iter.pos)) {
 394                         /*
 395                          * We have a hash collision for the new dst key,
 396                          * and new_src - the key we're deleting - is between
 397                          * new_dst's hashed slot and the slot we're going to be
 398                          * inserting it into - oops.  This will break the hash
 399                          * table if we don't deal with it:
 400                          */
 401                         if (mode == BCH_RENAME) {
 402                                 /*
 403                                  * If we're not overwriting, we can just insert
 404                                  * new_dst at the src position:
 405                                  */
 406                                 new_src = new_dst;
 407                                 new_src->k.p = src_iter.pos;
 408                                 goto out_set_src;
 409                         } else {
 410                                 /* If we're overwriting, we can't insert new_dst
 411                                  * at a different slot because it has to
 412                                  * overwrite old_dst - just make sure to use a
 413                                  * whiteout when deleting src:
 414                                  */
 415                                 new_src->k.type = KEY_TYPE_hash_whiteout;
 416                         }
 417                 } else {
 418                         /* Check if we need a whiteout to delete src: */
 419                         ret = bch2_hash_needs_whiteout(trans, bch2_dirent_hash_desc,
 420                                                        src_hash, &src_iter);
 421                         if (ret < 0)
 422                                 goto out;
 423
 424                         if (ret)
 425                                 new_src->k.type = KEY_TYPE_hash_whiteout;
 426                 }
 427         }
 428
 429         ret = bch2_trans_update(trans, &dst_iter, &new_dst->k_i, 0);
 430         if (ret)
 431                 goto out;
 432 out_set_src:
 433
 434         /*
 435          * If we're deleting a subvolume, we need to really delete the dirent,
 436          * not just emit a whiteout in the current snapshot:
 437          */
 438         if (src_type == DT_SUBVOL) {
 439                 bch2_btree_iter_set_snapshot(&src_iter, old_src.k->p.snapshot);
 440                 ret = bch2_btree_iter_traverse(&src_iter);
 441                 if (ret)
 442                         goto out;
 443
 444                 new_src->k.p = src_iter.pos;
 445                 src_update_flags |= BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE;
 446         }
 447
 448         ret = bch2_trans_update(trans, &src_iter, &new_src->k_i, src_update_flags);
 449         if (ret)
 450                 goto out;
 451
 452         if (mode == BCH_RENAME_EXCHANGE)
 453                 *src_offset = new_src->k.p.offset;
 454         *dst_offset = new_dst->k.p.offset;
 455 out:
 456         bch2_trans_iter_exit(trans, &src_iter);
 457         bch2_trans_iter_exit(trans, &dst_iter);
 458         return ret;
 459 }
 460
 461 int __bch2_dirent_lookup_trans(struct btree_trans *trans,
 462                                struct btree_iter *iter,
 463                                subvol_inum dir,
 464                                const struct bch_hash_info *hash_info,
 465                                const struct qstr *name, subvol_inum *inum,
 466                                unsigned flags)
 467 {
 468         struct bkey_s_c k;
 469         struct bkey_s_c_dirent d;
 470         u32 snapshot;
 471         int ret;
 472
 473         ret = bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot);
 474         if (ret)
 475                 return ret;
 476
 477         ret = bch2_hash_lookup(trans, iter, bch2_dirent_hash_desc,
 478                                hash_info, dir, name, flags);
 479         if (ret)
 480                 return ret;
 481
 482         k = bch2_btree_iter_peek_slot(iter);
 483         ret = bkey_err(k);
 484         if (ret)
 485                 goto err;
 486
 487         d = bkey_s_c_to_dirent(k);
 488
 489         ret = bch2_dirent_read_target(trans, dir, d, inum);
 490         if (ret > 0)
 491                 ret = -ENOENT;
 492 err:
 493         if (ret)
 494                 bch2_trans_iter_exit(trans, iter);
 495
 496         return ret;
 497 }
 498
 499 u64 bch2_dirent_lookup(struct bch_fs *c, subvol_inum dir,
 500                        const struct bch_hash_info *hash_info,
 501                        const struct qstr *name, subvol_inum *inum)
 502 {
 503         struct btree_trans *trans = bch2_trans_get(c);
 504         struct btree_iter iter = { NULL };
 505
 506         int ret = lockrestart_do(trans,
 507                 __bch2_dirent_lookup_trans(trans, &iter, dir, hash_info, name, inum, 0));
 508         bch2_trans_iter_exit(trans, &iter);
 509         bch2_trans_put(trans);
 510         return ret;
 511 }
 512
 513 int bch2_empty_dir_snapshot(struct btree_trans *trans, u64 dir, u32 snapshot)
 514 {
 515         struct btree_iter iter;
 516         struct bkey_s_c k;
 517         int ret;
 518
 519         for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
 520                            SPOS(dir, 0, snapshot),
 521                            POS(dir, U64_MAX), 0, k, ret)
 522                 if (k.k->type == KEY_TYPE_dirent) {
 523                         ret = -ENOTEMPTY;
 524                         break;
 525                 }
 526         bch2_trans_iter_exit(trans, &iter);
 527
 528         return ret;
 529 }
 530
 531 int bch2_empty_dir_trans(struct btree_trans *trans, subvol_inum dir)
 532 {
 533         u32 snapshot;
 534
 535         return bch2_subvolume_get_snapshot(trans, dir.subvol, &snapshot) ?:
 536                 bch2_empty_dir_snapshot(trans, dir.inum, snapshot);
 537 }
 538
 539 int bch2_readdir(struct bch_fs *c, subvol_inum inum, struct dir_context *ctx)
 540 {
 541         struct btree_trans *trans = bch2_trans_get(c);
 542         struct btree_iter iter;
 543         struct bkey_s_c k;
 544         struct bkey_s_c_dirent dirent;
 545         subvol_inum target;
 546         u32 snapshot;
 547         struct bkey_buf sk;
 548         struct qstr name;
 549         int ret;
 550
 551         bch2_bkey_buf_init(&sk);
 552 retry:
 553         bch2_trans_begin(trans);
 554
 555         ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
 556         if (ret)
 557                 goto err;
 558
 559         for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
 560                            SPOS(inum.inum, ctx->pos, snapshot),
 561                            POS(inum.inum, U64_MAX), 0, k, ret) {
 562                 if (k.k->type != KEY_TYPE_dirent)
 563                         continue;
 564
 565                 dirent = bkey_s_c_to_dirent(k);
 566
 567                 ret = bch2_dirent_read_target(trans, inum, dirent, &target);
 568                 if (ret < 0)
 569                         break;
 570                 if (ret)
 571                         continue;
 572
 573                 /* dir_emit() can fault and block: */
 574                 bch2_bkey_buf_reassemble(&sk, c, k);
 575                 dirent = bkey_i_to_s_c_dirent(sk.k);
 576                 bch2_trans_unlock(trans);
 577
 578                 name = bch2_dirent_get_name(dirent);
 579
 580                 ctx->pos = dirent.k->p.offset;
 581                 if (!dir_emit(ctx, name.name,
 582                               name.len,
 583                               target.inum,
 584                               vfs_d_type(dirent.v->d_type)))
 585                         break;
 586                 ctx->pos = dirent.k->p.offset + 1;
 587
 588                 /*
 589                  * read_target looks up subvolumes, we can overflow paths if the
 590                  * directory has many subvolumes in it
 591                  */
 592                 ret = btree_trans_too_many_iters(trans);
 593                 if (ret)
 594                         break;
 595         }
 596         bch2_trans_iter_exit(trans, &iter);
 597 err:
 598         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 599                 goto retry;
 600
 601         bch2_trans_put(trans);
 602         bch2_bkey_buf_exit(&sk, c);
 603
 604         return ret;
 605 }