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
  12 #include <linux/dcache.h>
  13
  14 unsigned bch2_dirent_name_bytes(struct bkey_s_c_dirent d)
  15 {
  16         unsigned len = bkey_val_bytes(d.k) -
  17                 offsetof(struct bch_dirent, d_name);
  18
  19         return strnlen(d.v->d_name, len);
  20 }
  21
  22 static u64 bch2_dirent_hash(const struct bch_hash_info *info,
  23                             const struct qstr *name)
  24 {
  25         struct bch_str_hash_ctx ctx;
  26
  27         bch2_str_hash_init(&ctx, info);
  28         bch2_str_hash_update(&ctx, info, name->name, name->len);
  29
  30         /* [0,2) reserved for dots */
  31         return max_t(u64, bch2_str_hash_end(&ctx, info), 2);
  32 }
  33
  34 static u64 dirent_hash_key(const struct bch_hash_info *info, const void *key)
  35 {
  36         return bch2_dirent_hash(info, key);
  37 }
  38
  39 static u64 dirent_hash_bkey(const struct bch_hash_info *info, struct bkey_s_c k)
  40 {
  41         struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
  42         struct qstr name = QSTR_INIT(d.v->d_name, bch2_dirent_name_bytes(d));
  43
  44         return bch2_dirent_hash(info, &name);
  45 }
  46
  47 static bool dirent_cmp_key(struct bkey_s_c _l, const void *_r)
  48 {
  49         struct bkey_s_c_dirent l = bkey_s_c_to_dirent(_l);
  50         int len = bch2_dirent_name_bytes(l);
  51         const struct qstr *r = _r;
  52
  53         return len - r->len ?: memcmp(l.v->d_name, r->name, len);
  54 }
  55
  56 static bool dirent_cmp_bkey(struct bkey_s_c _l, struct bkey_s_c _r)
  57 {
  58         struct bkey_s_c_dirent l = bkey_s_c_to_dirent(_l);
  59         struct bkey_s_c_dirent r = bkey_s_c_to_dirent(_r);
  60         int l_len = bch2_dirent_name_bytes(l);
  61         int r_len = bch2_dirent_name_bytes(r);
  62
  63         return l_len - r_len ?: memcmp(l.v->d_name, r.v->d_name, l_len);
  64 }
  65
  66 const struct bch_hash_desc bch2_dirent_hash_desc = {
  67         .btree_id       = BTREE_ID_DIRENTS,
  68         .key_type       = KEY_TYPE_dirent,
  69         .hash_key       = dirent_hash_key,
  70         .hash_bkey      = dirent_hash_bkey,
  71         .cmp_key        = dirent_cmp_key,
  72         .cmp_bkey       = dirent_cmp_bkey,
  73 };
  74
  75 const char *bch2_dirent_invalid(const struct bch_fs *c, struct bkey_s_c k)
  76 {
  77         struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
  78         unsigned len;
  79
  80         if (bkey_val_bytes(k.k) < sizeof(struct bch_dirent))
  81                 return "value too small";
  82
  83         len = bch2_dirent_name_bytes(d);
  84         if (!len)
  85                 return "empty name";
  86
  87         /*
  88          * older versions of bcachefs were buggy and creating dirent
  89          * keys that were bigger than necessary:
  90          */
  91         if (bkey_val_u64s(k.k) > dirent_val_u64s(len + 7))
  92                 return "value too big";
  93
  94         if (len > BCH_NAME_MAX)
  95                 return "dirent name too big";
  96
  97         return NULL;
  98 }
  99
 100 void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c,
 101                          struct bkey_s_c k)
 102 {
 103         struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
 104
 105         bch_scnmemcpy(out, d.v->d_name,
 106                       bch2_dirent_name_bytes(d));
 107         pr_buf(out, " -> %llu", d.v->d_inum);
 108 }
 109
 110 static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
 111                                 u8 type, const struct qstr *name, u64 dst)
 112 {
 113         struct bkey_i_dirent *dirent;
 114         unsigned u64s = BKEY_U64s + dirent_val_u64s(name->len);
 115
 116         if (name->len > BCH_NAME_MAX)
 117                 return ERR_PTR(-ENAMETOOLONG);
 118
 119         BUG_ON(u64s > U8_MAX);
 120
 121         dirent = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
 122         if (IS_ERR(dirent))
 123                 return dirent;
 124
 125         bkey_dirent_init(&dirent->k_i);
 126         dirent->k.u64s = u64s;
 127         dirent->v.d_inum = cpu_to_le64(dst);
 128         dirent->v.d_type = type;
 129
 130         memcpy(dirent->v.d_name, name->name, name->len);
 131         memset(dirent->v.d_name + name->len, 0,
 132                bkey_val_bytes(&dirent->k) -
 133                offsetof(struct bch_dirent, d_name) -
 134                name->len);
 135
 136         EBUG_ON(bch2_dirent_name_bytes(dirent_i_to_s_c(dirent)) != name->len);
 137
 138         return dirent;
 139 }
 140
 141 int __bch2_dirent_create(struct btree_trans *trans,
 142                          u64 dir_inum, const struct bch_hash_info *hash_info,
 143                          u8 type, const struct qstr *name, u64 dst_inum,
 144                          int flags)
 145 {
 146         struct bkey_i_dirent *dirent;
 147         int ret;
 148
 149         dirent = dirent_create_key(trans, type, name, dst_inum);
 150         ret = PTR_ERR_OR_ZERO(dirent);
 151         if (ret)
 152                 return ret;
 153
 154         return bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
 155                              dir_inum, &dirent->k_i, flags);
 156 }
 157
 158 int bch2_dirent_create(struct bch_fs *c, u64 dir_inum,
 159                        const struct bch_hash_info *hash_info,
 160                        u8 type, const struct qstr *name, u64 dst_inum,
 161                        u64 *journal_seq, int flags)
 162 {
 163         return bch2_trans_do(c, journal_seq, flags,
 164                 __bch2_dirent_create(&trans, dir_inum, hash_info,
 165                                      type, name, dst_inum, flags));
 166 }
 167
 168 static void dirent_copy_target(struct bkey_i_dirent *dst,
 169                                struct bkey_s_c_dirent src)
 170 {
 171         dst->v.d_inum = src.v->d_inum;
 172         dst->v.d_type = src.v->d_type;
 173 }
 174
 175 static struct bpos bch2_dirent_pos(struct bch_inode_info *inode,
 176                                    const struct qstr *name)
 177 {
 178         return POS(inode->v.i_ino, bch2_dirent_hash(&inode->ei_str_hash, name));
 179 }
 180
 181 int bch2_dirent_rename(struct btree_trans *trans,
 182                 struct bch_inode_info *src_dir, const struct qstr *src_name,
 183                 struct bch_inode_info *dst_dir, const struct qstr *dst_name,
 184                 enum bch_rename_mode mode)
 185 {
 186         struct btree_iter *src_iter, *dst_iter;
 187         struct bkey_s_c old_src, old_dst;
 188         struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
 189         struct bpos dst_pos = bch2_dirent_pos(dst_dir, dst_name);
 190         int ret;
 191
 192         /*
 193          * Lookup dst:
 194          *
 195          * Note that in BCH_RENAME mode, we're _not_ checking if
 196          * the target already exists - we're relying on the VFS
 197          * to do that check for us for correctness:
 198          */
 199         dst_iter = mode == BCH_RENAME
 200                 ? bch2_hash_hole(trans, bch2_dirent_hash_desc,
 201                                  &dst_dir->ei_str_hash,
 202                                  dst_dir->v.i_ino, dst_name)
 203                 : bch2_hash_lookup(trans, bch2_dirent_hash_desc,
 204                                    &dst_dir->ei_str_hash,
 205                                    dst_dir->v.i_ino, dst_name,
 206                                    BTREE_ITER_INTENT);
 207         if (IS_ERR(dst_iter))
 208                 return PTR_ERR(dst_iter);
 209         old_dst = bch2_btree_iter_peek_slot(dst_iter);
 210
 211         /* Lookup src: */
 212         src_iter = bch2_hash_lookup(trans, bch2_dirent_hash_desc,
 213                                     &src_dir->ei_str_hash,
 214                                     src_dir->v.i_ino, src_name,
 215                                     BTREE_ITER_INTENT);
 216         if (IS_ERR(src_iter))
 217                 return PTR_ERR(src_iter);
 218         old_src = bch2_btree_iter_peek_slot(src_iter);
 219
 220         /* Create new dst key: */
 221         new_dst = dirent_create_key(trans, 0, dst_name, 0);
 222         if (IS_ERR(new_dst))
 223                 return PTR_ERR(new_dst);
 224
 225         dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
 226         new_dst->k.p = dst_iter->pos;
 227
 228         /* Create new src key: */
 229         if (mode == BCH_RENAME_EXCHANGE) {
 230                 new_src = dirent_create_key(trans, 0, src_name, 0);
 231                 if (IS_ERR(new_src))
 232                         return PTR_ERR(new_src);
 233
 234                 dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
 235                 new_src->k.p = src_iter->pos;
 236         } else {
 237                 new_src = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
 238                 if (IS_ERR(new_src))
 239                         return PTR_ERR(new_src);
 240                 bkey_init(&new_src->k);
 241                 new_src->k.p = src_iter->pos;
 242
 243                 if (bkey_cmp(dst_pos, src_iter->pos) <= 0 &&
 244                     bkey_cmp(src_iter->pos, dst_iter->pos) < 0) {
 245                         /*
 246                          * We have a hash collision for the new dst key,
 247                          * and new_src - the key we're deleting - is between
 248                          * new_dst's hashed slot and the slot we're going to be
 249                          * inserting it into - oops.  This will break the hash
 250                          * table if we don't deal with it:
 251                          */
 252                         if (mode == BCH_RENAME) {
 253                                 /*
 254                                  * If we're not overwriting, we can just insert
 255                                  * new_dst at the src position:
 256                                  */
 257                                 new_dst->k.p = src_iter->pos;
 258                                 bch2_trans_update(trans,
 259                                         BTREE_INSERT_ENTRY(src_iter,
 260                                                            &new_dst->k_i));
 261                                 return 0;
 262                         } else {
 263                                 /* If we're overwriting, we can't insert new_dst
 264                                  * at a different slot because it has to
 265                                  * overwrite old_dst - just make sure to use a
 266                                  * whiteout when deleting src:
 267                                  */
 268                                 new_src->k.type = KEY_TYPE_whiteout;
 269                         }
 270                 } else {
 271                         /* Check if we need a whiteout to delete src: */
 272                         ret = bch2_hash_needs_whiteout(trans, bch2_dirent_hash_desc,
 273                                                        &src_dir->ei_str_hash,
 274                                                        src_iter);
 275                         if (ret < 0)
 276                                 return ret;
 277
 278                         if (ret)
 279                                 new_src->k.type = KEY_TYPE_whiteout;
 280                 }
 281         }
 282
 283         bch2_trans_update(trans, BTREE_INSERT_ENTRY(src_iter, &new_src->k_i));
 284         bch2_trans_update(trans, BTREE_INSERT_ENTRY(dst_iter, &new_dst->k_i));
 285         return 0;
 286 }
 287
 288 int __bch2_dirent_delete(struct btree_trans *trans, u64 dir_inum,
 289                          const struct bch_hash_info *hash_info,
 290                          const struct qstr *name)
 291 {
 292         return bch2_hash_delete(trans, bch2_dirent_hash_desc, hash_info,
 293                                 dir_inum, name);
 294 }
 295
 296 int bch2_dirent_delete(struct bch_fs *c, u64 dir_inum,
 297                        const struct bch_hash_info *hash_info,
 298                        const struct qstr *name,
 299                        u64 *journal_seq)
 300 {
 301         return bch2_trans_do(c, journal_seq,
 302                              BTREE_INSERT_ATOMIC|
 303                              BTREE_INSERT_NOFAIL,
 304                 __bch2_dirent_delete(&trans, dir_inum, hash_info, name));
 305 }
 306
 307 u64 bch2_dirent_lookup(struct bch_fs *c, u64 dir_inum,
 308                        const struct bch_hash_info *hash_info,
 309                        const struct qstr *name)
 310 {
 311         struct btree_trans trans;
 312         struct btree_iter *iter;
 313         struct bkey_s_c k;
 314         u64 inum = 0;
 315
 316         bch2_trans_init(&trans, c, 0, 0);
 317
 318         iter = bch2_hash_lookup(&trans, bch2_dirent_hash_desc,
 319                                 hash_info, dir_inum, name, 0);
 320         if (IS_ERR(iter)) {
 321                 BUG_ON(PTR_ERR(iter) == -EINTR);
 322                 goto out;
 323         }
 324
 325         k = bch2_btree_iter_peek_slot(iter);
 326         inum = le64_to_cpu(bkey_s_c_to_dirent(k).v->d_inum);
 327 out:
 328         bch2_trans_exit(&trans);
 329         return inum;
 330 }
 331
 332 int bch2_empty_dir_trans(struct btree_trans *trans, u64 dir_inum)
 333 {
 334         struct btree_iter *iter;
 335         struct bkey_s_c k;
 336         int ret;
 337
 338         for_each_btree_key(trans, iter, BTREE_ID_DIRENTS,
 339                            POS(dir_inum, 0), 0, k, ret) {
 340                 if (k.k->p.inode > dir_inum)
 341                         break;
 342
 343                 if (k.k->type == KEY_TYPE_dirent) {
 344                         ret = -ENOTEMPTY;
 345                         break;
 346                 }
 347         }
 348         bch2_trans_iter_put(trans, iter);
 349
 350         return ret;
 351 }
 352
 353 int bch2_empty_dir(struct bch_fs *c, u64 dir_inum)
 354 {
 355         return bch2_trans_do(c, NULL, 0,
 356                 bch2_empty_dir_trans(&trans, dir_inum));
 357 }
 358
 359 int bch2_readdir(struct bch_fs *c, struct file *file,
 360                  struct dir_context *ctx)
 361 {
 362         struct bch_inode_info *inode = file_bch_inode(file);
 363         struct btree_trans trans;
 364         struct btree_iter *iter;
 365         struct bkey_s_c k;
 366         struct bkey_s_c_dirent dirent;
 367         unsigned len;
 368         int ret;
 369
 370         if (!dir_emit_dots(file, ctx))
 371                 return 0;
 372
 373         bch2_trans_init(&trans, c, 0, 0);
 374
 375         for_each_btree_key(&trans, iter, BTREE_ID_DIRENTS,
 376                            POS(inode->v.i_ino, ctx->pos), 0, k, ret) {
 377                 if (k.k->type != KEY_TYPE_dirent)
 378                         continue;
 379
 380                 dirent = bkey_s_c_to_dirent(k);
 381
 382                 if (bkey_cmp(k.k->p, POS(inode->v.i_ino, ctx->pos)) < 0)
 383                         continue;
 384
 385                 if (k.k->p.inode > inode->v.i_ino)
 386                         break;
 387
 388                 len = bch2_dirent_name_bytes(dirent);
 389
 390                 /*
 391                  * XXX: dir_emit() can fault and block, while we're holding
 392                  * locks
 393                  */
 394                 if (!dir_emit(ctx, dirent.v->d_name, len,
 395                               le64_to_cpu(dirent.v->d_inum),
 396                               dirent.v->d_type))
 397                         break;
 398
 399                 ctx->pos = k.k->p.offset + 1;
 400         }
 401         ret = bch2_trans_exit(&trans) ?: ret;
 402
 403         return ret;
 404 }