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

   1
   2 #include "bcachefs.h"
   3 #include "bkey_methods.h"
   4 #include "btree_update.h"
   5 #include "error.h"
   6 #include "extents.h"
   7 #include "inode.h"
   8 #include "io.h"
   9 #include "keylist.h"
  10
  11 #include <linux/random.h>
  12
  13 #include <asm/unaligned.h>
  14
  15 #define FIELD_BYTES()                                           \
  16
  17 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
  18 static const u8 bits_table[8] = {
  19         1  * 8 - 1,
  20         2  * 8 - 2,
  21         3  * 8 - 3,
  22         4  * 8 - 4,
  23         6  * 8 - 5,
  24         8  * 8 - 6,
  25         10 * 8 - 7,
  26         13 * 8 - 8,
  27 };
  28
  29 static int inode_encode_field(u8 *out, u8 *end, u64 hi, u64 lo)
  30 {
  31         __be64 in[2] = { cpu_to_be64(hi), cpu_to_be64(lo), };
  32         unsigned shift, bytes, bits = likely(!hi)
  33                 ? fls64(lo)
  34                 : fls64(hi) + 64;
  35
  36         for (shift = 1; shift <= 8; shift++)
  37                 if (bits < bits_table[shift - 1])
  38                         goto got_shift;
  39
  40         BUG();
  41 got_shift:
  42         bytes = byte_table[shift - 1];
  43
  44         BUG_ON(out + bytes > end);
  45
  46         memcpy(out, (u8 *) in + 16 - bytes, bytes);
  47         *out |= (1 << 8) >> shift;
  48
  49         return bytes;
  50 }
  51
  52 static int inode_decode_field(const u8 *in, const u8 *end,
  53                               u64 out[2], unsigned *out_bits)
  54 {
  55         __be64 be[2] = { 0, 0 };
  56         unsigned bytes, shift;
  57         u8 *p;
  58
  59         if (in >= end)
  60                 return -1;
  61
  62         if (!*in)
  63                 return -1;
  64
  65         /*
  66          * position of highest set bit indicates number of bytes:
  67          * shift = number of bits to remove in high byte:
  68          */
  69         shift   = 8 - __fls(*in); /* 1 <= shift <= 8 */
  70         bytes   = byte_table[shift - 1];
  71
  72         if (in + bytes > end)
  73                 return -1;
  74
  75         p = (u8 *) be + 16 - bytes;
  76         memcpy(p, in, bytes);
  77         *p ^= (1 << 8) >> shift;
  78
  79         out[0] = be64_to_cpu(be[0]);
  80         out[1] = be64_to_cpu(be[1]);
  81         *out_bits = out[0] ? 64 + fls64(out[0]) : fls64(out[1]);
  82
  83         return bytes;
  84 }
  85
  86 void bch2_inode_pack(struct bkey_inode_buf *packed,
  87                      const struct bch_inode_unpacked *inode)
  88 {
  89         u8 *out = packed->inode.v.fields;
  90         u8 *end = (void *) &packed[1];
  91         u8 *last_nonzero_field = out;
  92         unsigned nr_fields = 0, last_nonzero_fieldnr = 0;
  93
  94         bkey_inode_init(&packed->inode.k_i);
  95         packed->inode.k.p.inode         = inode->bi_inum;
  96         packed->inode.v.bi_hash_seed    = inode->bi_hash_seed;
  97         packed->inode.v.bi_flags        = cpu_to_le32(inode->bi_flags);
  98         packed->inode.v.bi_mode         = cpu_to_le16(inode->bi_mode);
  99
 100 #define BCH_INODE_FIELD(_name, _bits)                                   \
 101         out += inode_encode_field(out, end, 0, inode->_name);           \
 102         nr_fields++;                                                    \
 103                                                                         \
 104         if (inode->_name) {                                             \
 105                 last_nonzero_field = out;                               \
 106                 last_nonzero_fieldnr = nr_fields;                       \
 107         }
 108
 109         BCH_INODE_FIELDS()
 110 #undef  BCH_INODE_FIELD
 111
 112         out = last_nonzero_field;
 113         nr_fields = last_nonzero_fieldnr;
 114
 115         set_bkey_val_bytes(&packed->inode.k, out - (u8 *) &packed->inode.v);
 116         memset(out, 0,
 117                (u8 *) &packed->inode.v +
 118                bkey_val_bytes(&packed->inode.k) - out);
 119
 120         SET_INODE_NR_FIELDS(&packed->inode.v, nr_fields);
 121
 122         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
 123                 struct bch_inode_unpacked unpacked;
 124
 125                 int ret = bch2_inode_unpack(inode_i_to_s_c(&packed->inode),
 126                                            &unpacked);
 127                 BUG_ON(ret);
 128                 BUG_ON(unpacked.bi_inum         != inode->bi_inum);
 129                 BUG_ON(unpacked.bi_hash_seed    != inode->bi_hash_seed);
 130                 BUG_ON(unpacked.bi_mode         != inode->bi_mode);
 131
 132 #define BCH_INODE_FIELD(_name, _bits)   BUG_ON(unpacked._name != inode->_name);
 133                 BCH_INODE_FIELDS()
 134 #undef  BCH_INODE_FIELD
 135         }
 136 }
 137
 138 int bch2_inode_unpack(struct bkey_s_c_inode inode,
 139                       struct bch_inode_unpacked *unpacked)
 140 {
 141         const u8 *in = inode.v->fields;
 142         const u8 *end = (void *) inode.v + bkey_val_bytes(inode.k);
 143         u64 field[2];
 144         unsigned fieldnr = 0, field_bits;
 145         int ret;
 146
 147         unpacked->bi_inum       = inode.k->p.inode;
 148         unpacked->bi_hash_seed  = inode.v->bi_hash_seed;
 149         unpacked->bi_flags      = le32_to_cpu(inode.v->bi_flags);
 150         unpacked->bi_mode       = le16_to_cpu(inode.v->bi_mode);
 151
 152 #define BCH_INODE_FIELD(_name, _bits)                                   \
 153         if (fieldnr++ == INODE_NR_FIELDS(inode.v)) {                    \
 154                 memset(&unpacked->_name, 0,                             \
 155                        sizeof(*unpacked) -                              \
 156                        offsetof(struct bch_inode_unpacked, _name));     \
 157                 return 0;                                               \
 158         }                                                               \
 159                                                                         \
 160         ret = inode_decode_field(in, end, field, &field_bits);          \
 161         if (ret < 0)                                                    \
 162                 return ret;                                             \
 163                                                                         \
 164         if (field_bits > sizeof(unpacked->_name) * 8)                   \
 165                 return -1;                                              \
 166                                                                         \
 167         unpacked->_name = field[1];                                     \
 168         in += ret;
 169
 170         BCH_INODE_FIELDS()
 171 #undef  BCH_INODE_FIELD
 172
 173         /* XXX: signal if there were more fields than expected? */
 174
 175         return 0;
 176 }
 177
 178 const char *bch2_inode_invalid(const struct bch_fs *c, struct bkey_s_c k)
 179 {
 180         if (k.k->p.offset)
 181                 return "nonzero offset";
 182
 183         switch (k.k->type) {
 184         case BCH_INODE_FS: {
 185                 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
 186                 struct bch_inode_unpacked unpacked;
 187
 188                 if (bkey_val_bytes(k.k) < sizeof(struct bch_inode))
 189                         return "incorrect value size";
 190
 191                 if (k.k->p.inode < BLOCKDEV_INODE_MAX)
 192                         return "fs inode in blockdev range";
 193
 194                 if (INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR)
 195                         return "invalid str hash type";
 196
 197                 if (bch2_inode_unpack(inode, &unpacked))
 198                         return "invalid variable length fields";
 199
 200                 if (unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1)
 201                         return "invalid data checksum type";
 202
 203                 if (unpacked.bi_compression >= BCH_COMPRESSION_OPT_NR + 1)
 204                         return "invalid data checksum type";
 205
 206                 if ((unpacked.bi_flags & BCH_INODE_UNLINKED) &&
 207                     unpacked.bi_nlink != 0)
 208                         return "flagged as unlinked but bi_nlink != 0";
 209
 210                 return NULL;
 211         }
 212         case BCH_INODE_BLOCKDEV:
 213                 if (bkey_val_bytes(k.k) != sizeof(struct bch_inode_blockdev))
 214                         return "incorrect value size";
 215
 216                 if (k.k->p.inode >= BLOCKDEV_INODE_MAX)
 217                         return "blockdev inode in fs range";
 218
 219                 return NULL;
 220         case BCH_INODE_GENERATION:
 221                 if (bkey_val_bytes(k.k) != sizeof(struct bch_inode_generation))
 222                         return "incorrect value size";
 223
 224                 return NULL;
 225         default:
 226                 return "invalid type";
 227         }
 228 }
 229
 230 void bch2_inode_to_text(struct bch_fs *c, char *buf,
 231                         size_t size, struct bkey_s_c k)
 232 {
 233         char *out = buf, *end = out + size;
 234         struct bkey_s_c_inode inode;
 235         struct bch_inode_unpacked unpacked;
 236
 237         switch (k.k->type) {
 238         case BCH_INODE_FS:
 239                 inode = bkey_s_c_to_inode(k);
 240                 if (bch2_inode_unpack(inode, &unpacked)) {
 241                         out += scnprintf(out, end - out, "(unpack error)");
 242                         break;
 243                 }
 244
 245 #define BCH_INODE_FIELD(_name, _bits)                                           \
 246                 out += scnprintf(out, end - out, #_name ": %llu ", (u64) unpacked._name);
 247                 BCH_INODE_FIELDS()
 248 #undef  BCH_INODE_FIELD
 249                 break;
 250         }
 251 }
 252
 253 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
 254                      uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
 255                      struct bch_inode_unpacked *parent)
 256 {
 257         s64 now = timespec_to_bch2_time(c,
 258                 timespec_trunc(current_kernel_time(),
 259                                c->sb.time_precision));
 260
 261         memset(inode_u, 0, sizeof(*inode_u));
 262
 263         /* ick */
 264         inode_u->bi_flags |= c->opts.str_hash << INODE_STR_HASH_OFFSET;
 265         get_random_bytes(&inode_u->bi_hash_seed, sizeof(inode_u->bi_hash_seed));
 266
 267         inode_u->bi_mode        = mode;
 268         inode_u->bi_uid         = uid;
 269         inode_u->bi_gid         = gid;
 270         inode_u->bi_dev         = rdev;
 271         inode_u->bi_atime       = now;
 272         inode_u->bi_mtime       = now;
 273         inode_u->bi_ctime       = now;
 274         inode_u->bi_otime       = now;
 275
 276         if (parent) {
 277 #define BCH_INODE_FIELD(_name)  inode_u->_name = parent->_name;
 278                 BCH_INODE_FIELDS_INHERIT()
 279 #undef BCH_INODE_FIELD
 280         }
 281 }
 282
 283 static inline u32 bkey_generation(struct bkey_s_c k)
 284 {
 285         switch (k.k->type) {
 286         case BCH_INODE_BLOCKDEV:
 287         case BCH_INODE_FS:
 288                 BUG();
 289         case BCH_INODE_GENERATION:
 290                 return le32_to_cpu(bkey_s_c_to_inode_generation(k).v->bi_generation);
 291         default:
 292                 return 0;
 293         }
 294 }
 295
 296 int __bch2_inode_create(struct btree_trans *trans,
 297                         struct bch_inode_unpacked *inode_u,
 298                         u64 min, u64 max, u64 *hint)
 299 {
 300         struct bch_fs *c = trans->c;
 301         struct bkey_inode_buf *inode_p;
 302         struct btree_iter *iter;
 303         u64 start;
 304         int ret;
 305
 306         if (!max)
 307                 max = ULLONG_MAX;
 308
 309         if (c->opts.inodes_32bit)
 310                 max = min_t(u64, max, U32_MAX);
 311
 312         start = READ_ONCE(*hint);
 313
 314         if (start >= max || start < min)
 315                 start = min;
 316
 317         inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
 318         if (IS_ERR(inode_p))
 319                 return PTR_ERR(inode_p);
 320
 321         iter = bch2_trans_get_iter(trans,
 322                         BTREE_ID_INODES, POS(start, 0),
 323                         BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 324         if (IS_ERR(iter))
 325                 return PTR_ERR(iter);
 326 again:
 327         while (1) {
 328                 struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
 329
 330                 ret = btree_iter_err(k);
 331                 if (ret)
 332                         return ret;
 333
 334                 switch (k.k->type) {
 335                 case BCH_INODE_BLOCKDEV:
 336                 case BCH_INODE_FS:
 337                         /* slot used */
 338                         if (iter->pos.inode >= max)
 339                                 goto out;
 340
 341                         bch2_btree_iter_next_slot(iter);
 342                         break;
 343
 344                 default:
 345                         *hint                   = k.k->p.inode;
 346                         inode_u->bi_inum        = k.k->p.inode;
 347                         inode_u->bi_generation  = bkey_generation(k);
 348
 349                         bch2_inode_pack(inode_p, inode_u);
 350                         bch2_trans_update(trans, iter, &inode_p->inode.k_i, 0);
 351                         return 0;
 352                 }
 353         }
 354 out:
 355         if (start != min) {
 356                 /* Retry from start */
 357                 start = min;
 358                 bch2_btree_iter_set_pos(iter, POS(start, 0));
 359                 goto again;
 360         }
 361
 362         return -ENOSPC;
 363 }
 364
 365 int bch2_inode_create(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
 366                       u64 min, u64 max, u64 *hint)
 367 {
 368         return bch2_trans_do(c, NULL, BTREE_INSERT_ATOMIC,
 369                         __bch2_inode_create(&trans, inode_u, min, max, hint));
 370 }
 371
 372 int bch2_inode_truncate(struct bch_fs *c, u64 inode_nr, u64 new_size,
 373                         struct extent_insert_hook *hook, u64 *journal_seq)
 374 {
 375         return bch2_btree_delete_range(c, BTREE_ID_EXTENTS,
 376                                        POS(inode_nr, new_size),
 377                                        POS(inode_nr + 1, 0),
 378                                        ZERO_VERSION, NULL, hook,
 379                                        journal_seq);
 380 }
 381
 382 int bch2_inode_rm(struct bch_fs *c, u64 inode_nr)
 383 {
 384         struct btree_iter iter;
 385         struct bkey_i_inode_generation delete;
 386         int ret;
 387
 388         ret = bch2_inode_truncate(c, inode_nr, 0, NULL, NULL);
 389         if (ret < 0)
 390                 return ret;
 391
 392         ret = bch2_btree_delete_range(c, BTREE_ID_XATTRS,
 393                                      POS(inode_nr, 0),
 394                                      POS(inode_nr + 1, 0),
 395                                      ZERO_VERSION, NULL, NULL, NULL);
 396         if (ret < 0)
 397                 return ret;
 398
 399         /*
 400          * If this was a directory, there shouldn't be any real dirents left -
 401          * but there could be whiteouts (from hash collisions) that we should
 402          * delete:
 403          *
 404          * XXX: the dirent could ideally would delete whiteouts when they're no
 405          * longer needed
 406          */
 407         ret = bch2_btree_delete_range(c, BTREE_ID_DIRENTS,
 408                                      POS(inode_nr, 0),
 409                                      POS(inode_nr + 1, 0),
 410                                      ZERO_VERSION, NULL, NULL, NULL);
 411         if (ret < 0)
 412                 return ret;
 413
 414         bch2_btree_iter_init(&iter, c, BTREE_ID_INODES, POS(inode_nr, 0),
 415                              BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 416         do {
 417                 struct bkey_s_c k = bch2_btree_iter_peek_slot(&iter);
 418                 u32 bi_generation = 0;
 419
 420                 ret = btree_iter_err(k);
 421                 if (ret) {
 422                         bch2_btree_iter_unlock(&iter);
 423                         return ret;
 424                 }
 425
 426                 bch2_fs_inconsistent_on(k.k->type != BCH_INODE_FS, c,
 427                                         "inode %llu not found when deleting",
 428                                         inode_nr);
 429
 430                 switch (k.k->type) {
 431                 case BCH_INODE_FS: {
 432                         struct bch_inode_unpacked inode_u;
 433
 434                         if (!bch2_inode_unpack(bkey_s_c_to_inode(k), &inode_u))
 435                                 bi_generation = inode_u.bi_generation + 1;
 436                         break;
 437                 }
 438                 case BCH_INODE_GENERATION: {
 439                         struct bkey_s_c_inode_generation g =
 440                                 bkey_s_c_to_inode_generation(k);
 441                         bi_generation = le32_to_cpu(g.v->bi_generation);
 442                         break;
 443                 }
 444                 }
 445
 446                 if (!bi_generation) {
 447                         bkey_init(&delete.k);
 448                         delete.k.p.inode = inode_nr;
 449                 } else {
 450                         bkey_inode_generation_init(&delete.k_i);
 451                         delete.k.p.inode = inode_nr;
 452                         delete.v.bi_generation = cpu_to_le32(bi_generation);
 453                 }
 454
 455                 ret = bch2_btree_insert_at(c, NULL, NULL, NULL,
 456                                 BTREE_INSERT_ATOMIC|
 457                                 BTREE_INSERT_NOFAIL,
 458                                 BTREE_INSERT_ENTRY(&iter, &delete.k_i));
 459         } while (ret == -EINTR);
 460
 461         bch2_btree_iter_unlock(&iter);
 462         return ret;
 463 }
 464
 465 int bch2_inode_find_by_inum(struct bch_fs *c, u64 inode_nr,
 466                             struct bch_inode_unpacked *inode)
 467 {
 468         struct btree_iter iter;
 469         struct bkey_s_c k;
 470         int ret = -ENOENT;
 471
 472         for_each_btree_key(&iter, c, BTREE_ID_INODES,
 473                            POS(inode_nr, 0),
 474                            BTREE_ITER_SLOTS, k) {
 475                 switch (k.k->type) {
 476                 case BCH_INODE_FS:
 477                         ret = bch2_inode_unpack(bkey_s_c_to_inode(k), inode);
 478                         break;
 479                 default:
 480                         /* hole, not found */
 481                         break;
 482                 }
 483
 484                 break;
 485
 486         }
 487
 488         return bch2_btree_iter_unlock(&iter) ?: ret;
 489 }
 490
 491 #ifdef CONFIG_BCACHEFS_DEBUG
 492 void bch2_inode_pack_test(void)
 493 {
 494         struct bch_inode_unpacked *u, test_inodes[] = {
 495                 {
 496                         .bi_atime       = U64_MAX,
 497                         .bi_ctime       = U64_MAX,
 498                         .bi_mtime       = U64_MAX,
 499                         .bi_otime       = U64_MAX,
 500                         .bi_size        = U64_MAX,
 501                         .bi_sectors     = U64_MAX,
 502                         .bi_uid         = U32_MAX,
 503                         .bi_gid         = U32_MAX,
 504                         .bi_nlink       = U32_MAX,
 505                         .bi_generation  = U32_MAX,
 506                         .bi_dev         = U32_MAX,
 507                 },
 508         };
 509
 510         for (u = test_inodes;
 511              u < test_inodes + ARRAY_SIZE(test_inodes);
 512              u++) {
 513                 struct bkey_inode_buf p;
 514
 515                 bch2_inode_pack(&p, u);
 516         }
 517 }
 518 #endif