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

   1
   2 #include "bcachefs.h"
   3 #include "bkey_methods.h"
   4 #include "btree_update.h"
   5 #include "extents.h"
   6 #include "dirent.h"
   7 #include "fs.h"
   8 #include "keylist.h"
   9 #include "str_hash.h"
  10
  11 #include <linux/dcache.h>
  12
  13 unsigned bch2_dirent_name_bytes(struct bkey_s_c_dirent d)
  14 {
  15         unsigned len = bkey_val_bytes(d.k) -
  16                 offsetof(struct bch_dirent, d_name);
  17
  18         while (len && !d.v->d_name[len - 1])
  19                 --len;
  20
  21         return len;
  22 }
  23
  24 static unsigned dirent_val_u64s(unsigned len)
  25 {
  26         return DIV_ROUND_UP(offsetof(struct bch_dirent, d_name) + len,
  27                             sizeof(u64));
  28 }
  29
  30 static u64 bch2_dirent_hash(const struct bch_hash_info *info,
  31                             const struct qstr *name)
  32 {
  33         struct bch_str_hash_ctx ctx;
  34
  35         bch2_str_hash_init(&ctx, info);
  36         bch2_str_hash_update(&ctx, info, name->name, name->len);
  37
  38         /* [0,2) reserved for dots */
  39         return max_t(u64, bch2_str_hash_end(&ctx, info), 2);
  40 }
  41
  42 static u64 dirent_hash_key(const struct bch_hash_info *info, const void *key)
  43 {
  44         return bch2_dirent_hash(info, key);
  45 }
  46
  47 static u64 dirent_hash_bkey(const struct bch_hash_info *info, struct bkey_s_c k)
  48 {
  49         struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
  50         struct qstr name = QSTR_INIT(d.v->d_name, bch2_dirent_name_bytes(d));
  51
  52         return bch2_dirent_hash(info, &name);
  53 }
  54
  55 static bool dirent_cmp_key(struct bkey_s_c _l, const void *_r)
  56 {
  57         struct bkey_s_c_dirent l = bkey_s_c_to_dirent(_l);
  58         int len = bch2_dirent_name_bytes(l);
  59         const struct qstr *r = _r;
  60
  61         return len - r->len ?: memcmp(l.v->d_name, r->name, len);
  62 }
  63
  64 static bool dirent_cmp_bkey(struct bkey_s_c _l, struct bkey_s_c _r)
  65 {
  66         struct bkey_s_c_dirent l = bkey_s_c_to_dirent(_l);
  67         struct bkey_s_c_dirent r = bkey_s_c_to_dirent(_r);
  68         int l_len = bch2_dirent_name_bytes(l);
  69         int r_len = bch2_dirent_name_bytes(r);
  70
  71         return l_len - r_len ?: memcmp(l.v->d_name, r.v->d_name, l_len);
  72 }
  73
  74 const struct bch_hash_desc bch2_dirent_hash_desc = {
  75         .btree_id       = BTREE_ID_DIRENTS,
  76         .key_type       = BCH_DIRENT,
  77         .whiteout_type  = BCH_DIRENT_WHITEOUT,
  78         .hash_key       = dirent_hash_key,
  79         .hash_bkey      = dirent_hash_bkey,
  80         .cmp_key        = dirent_cmp_key,
  81         .cmp_bkey       = dirent_cmp_bkey,
  82 };
  83
  84 const char *bch2_dirent_invalid(const struct bch_fs *c, struct bkey_s_c k)
  85 {
  86         struct bkey_s_c_dirent d;
  87         unsigned len;
  88
  89         switch (k.k->type) {
  90         case BCH_DIRENT:
  91                 if (bkey_val_bytes(k.k) < sizeof(struct bch_dirent))
  92                         return "value too small";
  93
  94                 d = bkey_s_c_to_dirent(k);
  95                 len = bch2_dirent_name_bytes(d);
  96
  97                 if (!len)
  98                         return "empty name";
  99
 100                 /*
 101                  * older versions of bcachefs were buggy and creating dirent
 102                  * keys that were bigger than necessary:
 103                  */
 104                 if (bkey_val_u64s(k.k) > dirent_val_u64s(len + 7))
 105                         return "value too big";
 106
 107                 if (len > BCH_NAME_MAX)
 108                         return "dirent name too big";
 109
 110                 if (memchr(d.v->d_name, '/', len))
 111                         return "dirent name has invalid characters";
 112
 113                 return NULL;
 114         case BCH_DIRENT_WHITEOUT:
 115                 return bkey_val_bytes(k.k) != 0
 116                         ? "value size should be zero"
 117                         : NULL;
 118
 119         default:
 120                 return "invalid type";
 121         }
 122 }
 123
 124 void bch2_dirent_to_text(struct bch_fs *c, char *buf,
 125                          size_t size, struct bkey_s_c k)
 126 {
 127         struct bkey_s_c_dirent d;
 128         size_t n = 0;
 129
 130         switch (k.k->type) {
 131         case BCH_DIRENT:
 132                 d = bkey_s_c_to_dirent(k);
 133
 134                 n += bch_scnmemcpy(buf + n, size - n, d.v->d_name,
 135                                    bch2_dirent_name_bytes(d));
 136                 n += scnprintf(buf + n, size - n, " -> %llu", d.v->d_inum);
 137                 break;
 138         case BCH_DIRENT_WHITEOUT:
 139                 scnprintf(buf, size, "whiteout");
 140                 break;
 141         }
 142 }
 143
 144 static struct bkey_i_dirent *dirent_create_key(u8 type,
 145                                 const struct qstr *name, u64 dst)
 146 {
 147         struct bkey_i_dirent *dirent;
 148         unsigned u64s = BKEY_U64s + dirent_val_u64s(name->len);
 149
 150         if (name->len > BCH_NAME_MAX)
 151                 return ERR_PTR(-ENAMETOOLONG);
 152
 153         BUG_ON(u64s > U8_MAX);
 154
 155         dirent = kmalloc(u64s * sizeof(u64), GFP_NOFS);
 156         if (!dirent)
 157                 return ERR_PTR(-ENOMEM);
 158
 159         bkey_dirent_init(&dirent->k_i);
 160         dirent->k.u64s = u64s;
 161         dirent->v.d_inum = cpu_to_le64(dst);
 162         dirent->v.d_type = type;
 163
 164         memcpy(dirent->v.d_name, name->name, name->len);
 165         memset(dirent->v.d_name + name->len, 0,
 166                bkey_val_bytes(&dirent->k) -
 167                offsetof(struct bch_dirent, d_name) -
 168                name->len);
 169
 170         EBUG_ON(bch2_dirent_name_bytes(dirent_i_to_s_c(dirent)) != name->len);
 171
 172         return dirent;
 173 }
 174
 175 int bch2_dirent_create(struct bch_fs *c, u64 dir_inum,
 176                        const struct bch_hash_info *hash_info,
 177                        u8 type, const struct qstr *name, u64 dst_inum,
 178                        u64 *journal_seq, int flags)
 179 {
 180         struct bkey_i_dirent *dirent;
 181         int ret;
 182
 183         dirent = dirent_create_key(type, name, dst_inum);
 184         if (IS_ERR(dirent))
 185                 return PTR_ERR(dirent);
 186
 187         ret = bch2_hash_set(bch2_dirent_hash_desc, hash_info, c, dir_inum,
 188                            journal_seq, &dirent->k_i, flags);
 189         kfree(dirent);
 190
 191         return ret;
 192 }
 193
 194 static void dirent_copy_target(struct bkey_i_dirent *dst,
 195                                struct bkey_s_c_dirent src)
 196 {
 197         dst->v.d_inum = src.v->d_inum;
 198         dst->v.d_type = src.v->d_type;
 199 }
 200
 201 static struct bpos bch2_dirent_pos(struct bch_inode_info *inode,
 202                                    const struct qstr *name)
 203 {
 204         return POS(inode->v.i_ino, bch2_dirent_hash(&inode->ei_str_hash, name));
 205 }
 206
 207 int bch2_dirent_rename(struct bch_fs *c,
 208                 struct bch_inode_info *src_dir, const struct qstr *src_name,
 209                 struct bch_inode_info *dst_dir, const struct qstr *dst_name,
 210                 u64 *journal_seq, enum bch_rename_mode mode)
 211 {
 212         struct btree_iter src_iter, dst_iter, whiteout_iter;
 213         struct bkey_s_c old_src, old_dst;
 214         struct bkey delete;
 215         struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
 216         struct bpos src_pos = bch2_dirent_pos(src_dir, src_name);
 217         struct bpos dst_pos = bch2_dirent_pos(dst_dir, dst_name);
 218         bool need_whiteout;
 219         int ret;
 220
 221         bch2_btree_iter_init(&src_iter, c, BTREE_ID_DIRENTS, src_pos,
 222                              BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 223         bch2_btree_iter_init(&dst_iter, c, BTREE_ID_DIRENTS, dst_pos,
 224                              BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 225         bch2_btree_iter_link(&src_iter, &dst_iter);
 226
 227         bch2_btree_iter_init(&whiteout_iter, c, BTREE_ID_DIRENTS, src_pos,
 228                              BTREE_ITER_SLOTS);
 229         bch2_btree_iter_link(&src_iter, &whiteout_iter);
 230
 231         if (mode == BCH_RENAME_EXCHANGE) {
 232                 new_src = dirent_create_key(0, src_name, 0);
 233                 if (IS_ERR(new_src)) {
 234                         ret = PTR_ERR(new_src);
 235                         goto err;
 236                 }
 237         } else {
 238                 new_src = (void *) &delete;
 239         }
 240
 241         new_dst = dirent_create_key(0, dst_name, 0);
 242         if (IS_ERR(new_dst)) {
 243                 ret = PTR_ERR(new_dst);
 244                 goto err;
 245         }
 246 retry:
 247         /*
 248          * Note that on -EINTR/dropped locks we're not restarting the lookup
 249          * from the original hashed position (like we do when creating dirents,
 250          * in bch_hash_set) -  we never move existing dirents to different slot:
 251          */
 252         old_src = bch2_hash_lookup_at(bch2_dirent_hash_desc,
 253                                      &src_dir->ei_str_hash,
 254                                      &src_iter, src_name);
 255         if ((ret = btree_iter_err(old_src)))
 256                 goto err;
 257
 258         ret = bch2_hash_needs_whiteout(bch2_dirent_hash_desc,
 259                                 &src_dir->ei_str_hash,
 260                                 &whiteout_iter, &src_iter);
 261         if (ret < 0)
 262                 goto err;
 263         need_whiteout = ret;
 264
 265         /*
 266          * Note that in BCH_RENAME mode, we're _not_ checking if
 267          * the target already exists - we're relying on the VFS
 268          * to do that check for us for correctness:
 269          */
 270         old_dst = mode == BCH_RENAME
 271                 ? bch2_hash_hole_at(bch2_dirent_hash_desc, &dst_iter)
 272                 : bch2_hash_lookup_at(bch2_dirent_hash_desc,
 273                                      &dst_dir->ei_str_hash,
 274                                      &dst_iter, dst_name);
 275         if ((ret = btree_iter_err(old_dst)))
 276                 goto err;
 277
 278         switch (mode) {
 279         case BCH_RENAME:
 280                 bkey_init(&new_src->k);
 281                 dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
 282
 283                 if (bkey_cmp(dst_pos, src_iter.pos) <= 0 &&
 284                     bkey_cmp(src_iter.pos, dst_iter.pos) < 0) {
 285                         /*
 286                          * If we couldn't insert new_dst at its hashed
 287                          * position (dst_pos) due to a hash collision,
 288                          * and we're going to be deleting in
 289                          * between the hashed position and first empty
 290                          * slot we found - just overwrite the pos we
 291                          * were going to delete:
 292                          *
 293                          * Note: this is a correctness issue, in this
 294                          * situation bch2_hash_needs_whiteout() could
 295                          * return false when the whiteout would have
 296                          * been needed if we inserted at the pos
 297                          * __dirent_find_hole() found
 298                          */
 299                         new_dst->k.p = src_iter.pos;
 300                         ret = bch2_btree_insert_at(c, NULL, NULL,
 301                                         journal_seq,
 302                                         BTREE_INSERT_ATOMIC,
 303                                         BTREE_INSERT_ENTRY(&src_iter,
 304                                                            &new_dst->k_i));
 305                         goto err;
 306                 }
 307
 308                 if (need_whiteout)
 309                         new_src->k.type = BCH_DIRENT_WHITEOUT;
 310                 break;
 311         case BCH_RENAME_OVERWRITE:
 312                 bkey_init(&new_src->k);
 313                 dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
 314
 315                 if (bkey_cmp(dst_pos, src_iter.pos) <= 0 &&
 316                     bkey_cmp(src_iter.pos, dst_iter.pos) < 0) {
 317                         /*
 318                          * Same case described above -
 319                          * bch_hash_needs_whiteout could spuriously
 320                          * return false, but we have to insert at
 321                          * dst_iter.pos because we're overwriting
 322                          * another dirent:
 323                          */
 324                         new_src->k.type = BCH_DIRENT_WHITEOUT;
 325                 } else if (need_whiteout)
 326                         new_src->k.type = BCH_DIRENT_WHITEOUT;
 327                 break;
 328         case BCH_RENAME_EXCHANGE:
 329                 dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
 330                 dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
 331                 break;
 332         }
 333
 334         new_src->k.p = src_iter.pos;
 335         new_dst->k.p = dst_iter.pos;
 336         ret = bch2_btree_insert_at(c, NULL, NULL, journal_seq,
 337                         BTREE_INSERT_ATOMIC,
 338                         BTREE_INSERT_ENTRY(&src_iter, &new_src->k_i),
 339                         BTREE_INSERT_ENTRY(&dst_iter, &new_dst->k_i));
 340 err:
 341         if (ret == -EINTR)
 342                 goto retry;
 343
 344         bch2_btree_iter_unlock(&whiteout_iter);
 345         bch2_btree_iter_unlock(&dst_iter);
 346         bch2_btree_iter_unlock(&src_iter);
 347
 348         if (new_src != (void *) &delete)
 349                 kfree(new_src);
 350         kfree(new_dst);
 351         return ret;
 352 }
 353
 354 int bch2_dirent_delete(struct bch_fs *c, u64 dir_inum,
 355                        const struct bch_hash_info *hash_info,
 356                        const struct qstr *name,
 357                        u64 *journal_seq)
 358 {
 359         return bch2_hash_delete(bch2_dirent_hash_desc, hash_info,
 360                                c, dir_inum, journal_seq, name);
 361 }
 362
 363 u64 bch2_dirent_lookup(struct bch_fs *c, u64 dir_inum,
 364                        const struct bch_hash_info *hash_info,
 365                        const struct qstr *name)
 366 {
 367         struct btree_iter iter;
 368         struct bkey_s_c k;
 369         u64 inum;
 370
 371         k = bch2_hash_lookup(bch2_dirent_hash_desc, hash_info, c,
 372                             dir_inum, &iter, name);
 373         if (IS_ERR(k.k)) {
 374                 bch2_btree_iter_unlock(&iter);
 375                 return 0;
 376         }
 377
 378         inum = le64_to_cpu(bkey_s_c_to_dirent(k).v->d_inum);
 379         bch2_btree_iter_unlock(&iter);
 380
 381         return inum;
 382 }
 383
 384 int bch2_empty_dir(struct bch_fs *c, u64 dir_inum)
 385 {
 386         struct btree_iter iter;
 387         struct bkey_s_c k;
 388         int ret = 0;
 389
 390         for_each_btree_key(&iter, c, BTREE_ID_DIRENTS, POS(dir_inum, 0), 0, k) {
 391                 if (k.k->p.inode > dir_inum)
 392                         break;
 393
 394                 if (k.k->type == BCH_DIRENT) {
 395                         ret = -ENOTEMPTY;
 396                         break;
 397                 }
 398         }
 399         bch2_btree_iter_unlock(&iter);
 400
 401         return ret;
 402 }
 403
 404 int bch2_readdir(struct bch_fs *c, struct file *file,
 405                  struct dir_context *ctx)
 406 {
 407         struct bch_inode_info *inode = file_bch_inode(file);
 408         struct btree_iter iter;
 409         struct bkey_s_c k;
 410         struct bkey_s_c_dirent dirent;
 411         unsigned len;
 412
 413         if (!dir_emit_dots(file, ctx))
 414                 return 0;
 415
 416         for_each_btree_key(&iter, c, BTREE_ID_DIRENTS,
 417                            POS(inode->v.i_ino, ctx->pos), 0, k) {
 418                 if (k.k->type != BCH_DIRENT)
 419                         continue;
 420
 421                 dirent = bkey_s_c_to_dirent(k);
 422
 423                 if (bkey_cmp(k.k->p, POS(inode->v.i_ino, ctx->pos)) < 0)
 424                         continue;
 425
 426                 if (k.k->p.inode > inode->v.i_ino)
 427                         break;
 428
 429                 len = bch2_dirent_name_bytes(dirent);
 430
 431                 /*
 432                  * XXX: dir_emit() can fault and block, while we're holding
 433                  * locks
 434                  */
 435                 if (!dir_emit(ctx, dirent.v->d_name, len,
 436                               le64_to_cpu(dirent.v->d_inum),
 437                               dirent.v->d_type))
 438                         break;
 439
 440                 ctx->pos = k.k->p.offset + 1;
 441         }
 442         bch2_btree_iter_unlock(&iter);
 443
 444         return 0;
 445 }