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         return strnlen(d.v->d_name, len);
  19 }
  20
  21 static u64 bch2_dirent_hash(const struct bch_hash_info *info,
  22                             const struct qstr *name)
  23 {
  24         struct bch_str_hash_ctx ctx;
  25
  26         bch2_str_hash_init(&ctx, info);
  27         bch2_str_hash_update(&ctx, info, name->name, name->len);
  28
  29         /* [0,2) reserved for dots */
  30         return max_t(u64, bch2_str_hash_end(&ctx, info), 2);
  31 }
  32
  33 static u64 dirent_hash_key(const struct bch_hash_info *info, const void *key)
  34 {
  35         return bch2_dirent_hash(info, key);
  36 }
  37
  38 static u64 dirent_hash_bkey(const struct bch_hash_info *info, struct bkey_s_c k)
  39 {
  40         struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
  41         struct qstr name = QSTR_INIT(d.v->d_name, bch2_dirent_name_bytes(d));
  42
  43         return bch2_dirent_hash(info, &name);
  44 }
  45
  46 static bool dirent_cmp_key(struct bkey_s_c _l, const void *_r)
  47 {
  48         struct bkey_s_c_dirent l = bkey_s_c_to_dirent(_l);
  49         int len = bch2_dirent_name_bytes(l);
  50         const struct qstr *r = _r;
  51
  52         return len - r->len ?: memcmp(l.v->d_name, r->name, len);
  53 }
  54
  55 static bool dirent_cmp_bkey(struct bkey_s_c _l, struct bkey_s_c _r)
  56 {
  57         struct bkey_s_c_dirent l = bkey_s_c_to_dirent(_l);
  58         struct bkey_s_c_dirent r = bkey_s_c_to_dirent(_r);
  59         int l_len = bch2_dirent_name_bytes(l);
  60         int r_len = bch2_dirent_name_bytes(r);
  61
  62         return l_len - r_len ?: memcmp(l.v->d_name, r.v->d_name, l_len);
  63 }
  64
  65 const struct bch_hash_desc bch2_dirent_hash_desc = {
  66         .btree_id       = BTREE_ID_DIRENTS,
  67         .key_type       = KEY_TYPE_dirent,
  68         .hash_key       = dirent_hash_key,
  69         .hash_bkey      = dirent_hash_bkey,
  70         .cmp_key        = dirent_cmp_key,
  71         .cmp_bkey       = dirent_cmp_bkey,
  72 };
  73
  74 const char *bch2_dirent_invalid(const struct bch_fs *c, struct bkey_s_c k)
  75 {
  76         struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
  77         unsigned len;
  78
  79         if (bkey_val_bytes(k.k) < sizeof(struct bch_dirent))
  80                 return "value too small";
  81
  82         len = bch2_dirent_name_bytes(d);
  83         if (!len)
  84                 return "empty name";
  85
  86         /*
  87          * older versions of bcachefs were buggy and creating dirent
  88          * keys that were bigger than necessary:
  89          */
  90         if (bkey_val_u64s(k.k) > dirent_val_u64s(len + 7))
  91                 return "value too big";
  92
  93         if (len > BCH_NAME_MAX)
  94                 return "dirent name too big";
  95
  96         return NULL;
  97 }
  98
  99 void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c,
 100                          struct bkey_s_c k)
 101 {
 102         struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
 103
 104         bch_scnmemcpy(out, d.v->d_name,
 105                       bch2_dirent_name_bytes(d));
 106         pr_buf(out, " -> %llu", d.v->d_inum);
 107 }
 108
 109 static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
 110                                 u8 type, const struct qstr *name, u64 dst)
 111 {
 112         struct bkey_i_dirent *dirent;
 113         unsigned u64s = BKEY_U64s + dirent_val_u64s(name->len);
 114
 115         if (name->len > BCH_NAME_MAX)
 116                 return ERR_PTR(-ENAMETOOLONG);
 117
 118         BUG_ON(u64s > U8_MAX);
 119
 120         dirent = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
 121         if (IS_ERR(dirent))
 122                 return dirent;
 123
 124         bkey_dirent_init(&dirent->k_i);
 125         dirent->k.u64s = u64s;
 126         dirent->v.d_inum = cpu_to_le64(dst);
 127         dirent->v.d_type = type;
 128
 129         memcpy(dirent->v.d_name, name->name, name->len);
 130         memset(dirent->v.d_name + name->len, 0,
 131                bkey_val_bytes(&dirent->k) -
 132                offsetof(struct bch_dirent, d_name) -
 133                name->len);
 134
 135         EBUG_ON(bch2_dirent_name_bytes(dirent_i_to_s_c(dirent)) != name->len);
 136
 137         return dirent;
 138 }
 139
 140 int __bch2_dirent_create(struct btree_trans *trans,
 141                          u64 dir_inum, const struct bch_hash_info *hash_info,
 142                          u8 type, const struct qstr *name, u64 dst_inum,
 143                          int flags)
 144 {
 145         struct bkey_i_dirent *dirent;
 146         int ret;
 147
 148         dirent = dirent_create_key(trans, type, name, dst_inum);
 149         ret = PTR_ERR_OR_ZERO(dirent);
 150         if (ret)
 151                 return ret;
 152
 153         return bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
 154                              dir_inum, &dirent->k_i, flags);
 155 }
 156
 157 int bch2_dirent_create(struct bch_fs *c, u64 dir_inum,
 158                        const struct bch_hash_info *hash_info,
 159                        u8 type, const struct qstr *name, u64 dst_inum,
 160                        u64 *journal_seq, int flags)
 161 {
 162         return bch2_trans_do(c, journal_seq, flags,
 163                 __bch2_dirent_create(&trans, dir_inum, hash_info,
 164                                      type, name, dst_inum, flags));
 165 }
 166
 167 static void dirent_copy_target(struct bkey_i_dirent *dst,
 168                                struct bkey_s_c_dirent src)
 169 {
 170         dst->v.d_inum = src.v->d_inum;
 171         dst->v.d_type = src.v->d_type;
 172 }
 173
 174 static struct bpos bch2_dirent_pos(struct bch_inode_info *inode,
 175                                    const struct qstr *name)
 176 {
 177         return POS(inode->v.i_ino, bch2_dirent_hash(&inode->ei_str_hash, name));
 178 }
 179
 180 int bch2_dirent_rename(struct btree_trans *trans,
 181                 struct bch_inode_info *src_dir, const struct qstr *src_name,
 182                 struct bch_inode_info *dst_dir, const struct qstr *dst_name,
 183                 enum bch_rename_mode mode)
 184 {
 185         struct btree_iter *src_iter, *dst_iter;
 186         struct bkey_s_c old_src, old_dst;
 187         struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
 188         struct bpos dst_pos = bch2_dirent_pos(dst_dir, dst_name);
 189         int ret;
 190
 191         /*
 192          * Lookup dst:
 193          *
 194          * Note that in BCH_RENAME mode, we're _not_ checking if
 195          * the target already exists - we're relying on the VFS
 196          * to do that check for us for correctness:
 197          */
 198         dst_iter = mode == BCH_RENAME
 199                 ? bch2_hash_hole(trans, bch2_dirent_hash_desc,
 200                                  &dst_dir->ei_str_hash,
 201                                  dst_dir->v.i_ino, dst_name)
 202                 : bch2_hash_lookup(trans, bch2_dirent_hash_desc,
 203                                    &dst_dir->ei_str_hash,
 204                                    dst_dir->v.i_ino, dst_name,
 205                                    BTREE_ITER_INTENT);
 206         if (IS_ERR(dst_iter))
 207                 return PTR_ERR(dst_iter);
 208         old_dst = bch2_btree_iter_peek_slot(dst_iter);
 209
 210         /* Lookup src: */
 211         src_iter = bch2_hash_lookup(trans, bch2_dirent_hash_desc,
 212                                     &src_dir->ei_str_hash,
 213                                     src_dir->v.i_ino, src_name,
 214                                     BTREE_ITER_INTENT);
 215         if (IS_ERR(src_iter))
 216                 return PTR_ERR(src_iter);
 217         old_src = bch2_btree_iter_peek_slot(src_iter);
 218
 219         /* Create new dst key: */
 220         new_dst = dirent_create_key(trans, 0, dst_name, 0);
 221         if (IS_ERR(new_dst))
 222                 return PTR_ERR(new_dst);
 223
 224         dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
 225         new_dst->k.p = dst_iter->pos;
 226
 227         /* Create new src key: */
 228         if (mode == BCH_RENAME_EXCHANGE) {
 229                 new_src = dirent_create_key(trans, 0, src_name, 0);
 230                 if (IS_ERR(new_src))
 231                         return PTR_ERR(new_src);
 232
 233                 dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
 234                 new_src->k.p = src_iter->pos;
 235         } else {
 236                 new_src = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
 237                 if (IS_ERR(new_src))
 238                         return PTR_ERR(new_src);
 239                 bkey_init(&new_src->k);
 240                 new_src->k.p = src_iter->pos;
 241
 242                 if (bkey_cmp(dst_pos, src_iter->pos) <= 0 &&
 243                     bkey_cmp(src_iter->pos, dst_iter->pos) < 0) {
 244                         /*
 245                          * We have a hash collision for the new dst key,
 246                          * and new_src - the key we're deleting - is between
 247                          * new_dst's hashed slot and the slot we're going to be
 248                          * inserting it into - oops.  This will break the hash
 249                          * table if we don't deal with it:
 250                          */
 251                         if (mode == BCH_RENAME) {
 252                                 /*
 253                                  * If we're not overwriting, we can just insert
 254                                  * new_dst at the src position:
 255                                  */
 256                                 new_dst->k.p = src_iter->pos;
 257                                 bch2_trans_update(trans,
 258                                         BTREE_INSERT_ENTRY(src_iter,
 259                                                            &new_dst->k_i));
 260                                 return 0;
 261                         } else {
 262                                 /* If we're overwriting, we can't insert new_dst
 263                                  * at a different slot because it has to
 264                                  * overwrite old_dst - just make sure to use a
 265                                  * whiteout when deleting src:
 266                                  */
 267                                 new_src->k.type = KEY_TYPE_whiteout;
 268                         }
 269                 } else {
 270                         /* Check if we need a whiteout to delete src: */
 271                         ret = bch2_hash_needs_whiteout(trans, bch2_dirent_hash_desc,
 272                                                        &src_dir->ei_str_hash,
 273                                                        src_iter);
 274                         if (ret < 0)
 275                                 return ret;
 276
 277                         if (ret)
 278                                 new_src->k.type = KEY_TYPE_whiteout;
 279                 }
 280         }
 281
 282         bch2_trans_update(trans, BTREE_INSERT_ENTRY(src_iter, &new_src->k_i));
 283         bch2_trans_update(trans, BTREE_INSERT_ENTRY(dst_iter, &new_dst->k_i));
 284         return 0;
 285 }
 286
 287 int __bch2_dirent_delete(struct btree_trans *trans, u64 dir_inum,
 288                          const struct bch_hash_info *hash_info,
 289                          const struct qstr *name)
 290 {
 291         return bch2_hash_delete(trans, bch2_dirent_hash_desc, hash_info,
 292                                 dir_inum, name);
 293 }
 294
 295 int bch2_dirent_delete(struct bch_fs *c, u64 dir_inum,
 296                        const struct bch_hash_info *hash_info,
 297                        const struct qstr *name,
 298                        u64 *journal_seq)
 299 {
 300         return bch2_trans_do(c, journal_seq,
 301                              BTREE_INSERT_ATOMIC|
 302                              BTREE_INSERT_NOFAIL,
 303                 __bch2_dirent_delete(&trans, dir_inum, hash_info, name));
 304 }
 305
 306 u64 bch2_dirent_lookup(struct bch_fs *c, u64 dir_inum,
 307                        const struct bch_hash_info *hash_info,
 308                        const struct qstr *name)
 309 {
 310         struct btree_trans trans;
 311         struct btree_iter *iter;
 312         struct bkey_s_c k;
 313         u64 inum = 0;
 314
 315         bch2_trans_init(&trans, c);
 316
 317         iter = bch2_hash_lookup(&trans, bch2_dirent_hash_desc,
 318                                 hash_info, dir_inum, name, 0);
 319         if (IS_ERR(iter)) {
 320                 BUG_ON(PTR_ERR(iter) == -EINTR);
 321                 goto out;
 322         }
 323
 324         k = bch2_btree_iter_peek_slot(iter);
 325         inum = le64_to_cpu(bkey_s_c_to_dirent(k).v->d_inum);
 326 out:
 327         bch2_trans_exit(&trans);
 328         return inum;
 329 }
 330
 331 int bch2_empty_dir_trans(struct btree_trans *trans, u64 dir_inum)
 332 {
 333         struct btree_iter *iter;
 334         struct bkey_s_c k;
 335         int ret;
 336
 337         for_each_btree_key(trans, iter, BTREE_ID_DIRENTS,
 338                            POS(dir_inum, 0), 0, k, ret) {
 339                 if (k.k->p.inode > dir_inum)
 340                         break;
 341
 342                 if (k.k->type == KEY_TYPE_dirent) {
 343                         ret = -ENOTEMPTY;
 344                         break;
 345                 }
 346         }
 347         bch2_trans_iter_put(trans, iter);
 348
 349         return ret;
 350 }
 351
 352 int bch2_empty_dir(struct bch_fs *c, u64 dir_inum)
 353 {
 354         return bch2_trans_do(c, NULL, 0,
 355                 bch2_empty_dir_trans(&trans, dir_inum));
 356 }
 357
 358 int bch2_readdir(struct bch_fs *c, struct file *file,
 359                  struct dir_context *ctx)
 360 {
 361         struct bch_inode_info *inode = file_bch_inode(file);
 362         struct btree_trans trans;
 363         struct btree_iter *iter;
 364         struct bkey_s_c k;
 365         struct bkey_s_c_dirent dirent;
 366         unsigned len;
 367         int ret;
 368
 369         if (!dir_emit_dots(file, ctx))
 370                 return 0;
 371
 372         bch2_trans_init(&trans, c);
 373
 374         for_each_btree_key(&trans, iter, BTREE_ID_DIRENTS,
 375                            POS(inode->v.i_ino, ctx->pos), 0, k, ret) {
 376                 if (k.k->type != KEY_TYPE_dirent)
 377                         continue;
 378
 379                 dirent = bkey_s_c_to_dirent(k);
 380
 381                 if (bkey_cmp(k.k->p, POS(inode->v.i_ino, ctx->pos)) < 0)
 382                         continue;
 383
 384                 if (k.k->p.inode > inode->v.i_ino)
 385                         break;
 386
 387                 len = bch2_dirent_name_bytes(dirent);
 388
 389                 /*
 390                  * XXX: dir_emit() can fault and block, while we're holding
 391                  * locks
 392                  */
 393                 if (!dir_emit(ctx, dirent.v->d_name, len,
 394                               le64_to_cpu(dirent.v->d_inum),
 395                               dirent.v->d_type))
 396                         break;
 397
 398                 ctx->pos = k.k->p.offset + 1;
 399         }
 400         ret = bch2_trans_exit(&trans) ?: ret;
 401
 402         return ret;
 403 }