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 static const char *bch2_inode_invalid(const struct bch_fs *c,
 179                                       struct bkey_s_c k)
 180 {
 181         if (k.k->p.offset)
 182                 return "nonzero offset";
 183
 184         switch (k.k->type) {
 185         case BCH_INODE_FS: {
 186                 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
 187                 struct bch_inode_unpacked unpacked;
 188
 189                 if (bkey_val_bytes(k.k) < sizeof(struct bch_inode))
 190                         return "incorrect value size";
 191
 192                 if (k.k->p.inode < BLOCKDEV_INODE_MAX)
 193                         return "fs inode in blockdev range";
 194
 195                 if (INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR)
 196                         return "invalid str hash type";
 197
 198                 if (bch2_inode_unpack(inode, &unpacked))
 199                         return "invalid variable length fields";
 200
 201                 if (unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1)
 202                         return "invalid data checksum type";
 203
 204                 if (unpacked.bi_compression >= BCH_COMPRESSION_OPT_NR + 1)
 205                         return "invalid data checksum type";
 206
 207                 return NULL;
 208         }
 209         case BCH_INODE_BLOCKDEV:
 210                 if (bkey_val_bytes(k.k) != sizeof(struct bch_inode_blockdev))
 211                         return "incorrect value size";
 212
 213                 if (k.k->p.inode >= BLOCKDEV_INODE_MAX)
 214                         return "blockdev inode in fs range";
 215
 216                 return NULL;
 217         case BCH_INODE_GENERATION:
 218                 if (bkey_val_bytes(k.k) != sizeof(struct bch_inode_generation))
 219                         return "incorrect value size";
 220
 221                 return NULL;
 222         default:
 223                 return "invalid type";
 224         }
 225 }
 226
 227 static void bch2_inode_to_text(struct bch_fs *c, char *buf,
 228                                size_t size, struct bkey_s_c k)
 229 {
 230         char *out = buf, *end = out + size;
 231         struct bkey_s_c_inode inode;
 232         struct bch_inode_unpacked unpacked;
 233
 234         switch (k.k->type) {
 235         case BCH_INODE_FS:
 236                 inode = bkey_s_c_to_inode(k);
 237                 if (bch2_inode_unpack(inode, &unpacked)) {
 238                         out += scnprintf(out, end - out, "(unpack error)");
 239                         break;
 240                 }
 241
 242 #define BCH_INODE_FIELD(_name, _bits)                                           \
 243                 out += scnprintf(out, end - out, #_name ": %llu ", (u64) unpacked._name);
 244                 BCH_INODE_FIELDS()
 245 #undef  BCH_INODE_FIELD
 246                 break;
 247         }
 248 }
 249
 250 const struct bkey_ops bch2_bkey_inode_ops = {
 251         .key_invalid    = bch2_inode_invalid,
 252         .val_to_text    = bch2_inode_to_text,
 253 };
 254
 255 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
 256                      uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
 257                      struct bch_inode_unpacked *parent)
 258 {
 259         s64 now = timespec_to_bch2_time(c,
 260                 timespec_trunc(current_kernel_time(),
 261                                c->sb.time_precision));
 262
 263         memset(inode_u, 0, sizeof(*inode_u));
 264
 265         /* ick */
 266         inode_u->bi_flags |= c->opts.str_hash << INODE_STR_HASH_OFFSET;
 267         get_random_bytes(&inode_u->bi_hash_seed, sizeof(inode_u->bi_hash_seed));
 268
 269         inode_u->bi_mode        = mode;
 270         inode_u->bi_uid         = uid;
 271         inode_u->bi_gid         = gid;
 272         inode_u->bi_dev         = rdev;
 273         inode_u->bi_atime       = now;
 274         inode_u->bi_mtime       = now;
 275         inode_u->bi_ctime       = now;
 276         inode_u->bi_otime       = now;
 277
 278         if (parent) {
 279 #define BCH_INODE_FIELD(_name)  inode_u->_name = parent->_name;
 280                 BCH_INODE_FIELDS_INHERIT()
 281 #undef BCH_INODE_FIELD
 282         }
 283 }
 284
 285 int bch2_inode_create(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
 286                       u64 min, u64 max, u64 *hint)
 287 {
 288         struct bkey_inode_buf inode_p;
 289         struct btree_iter iter;
 290         bool searched_from_start = false;
 291         int ret;
 292
 293         if (!max)
 294                 max = ULLONG_MAX;
 295
 296         if (c->opts.inodes_32bit)
 297                 max = min_t(u64, max, U32_MAX);
 298
 299         if (*hint >= max || *hint < min)
 300                 *hint = min;
 301
 302         if (*hint == min)
 303                 searched_from_start = true;
 304 again:
 305         bch2_btree_iter_init(&iter, c, BTREE_ID_INODES, POS(*hint, 0),
 306                              BTREE_ITER_INTENT);
 307
 308         while (1) {
 309                 struct bkey_s_c k = bch2_btree_iter_peek_with_holes(&iter);
 310                 u32 bi_generation = 0;
 311
 312                 ret = btree_iter_err(k);
 313                 if (ret) {
 314                         bch2_btree_iter_unlock(&iter);
 315                         return ret;
 316                 }
 317
 318                 switch (k.k->type) {
 319                 case BCH_INODE_BLOCKDEV:
 320                 case BCH_INODE_FS:
 321                         /* slot used */
 322                         if (iter.pos.inode == max)
 323                                 goto out;
 324
 325                         bch2_btree_iter_advance_pos(&iter);
 326                         break;
 327
 328                 case BCH_INODE_GENERATION: {
 329                         struct bkey_s_c_inode_generation g =
 330                                 bkey_s_c_to_inode_generation(k);
 331                         bi_generation = le32_to_cpu(g.v->bi_generation);
 332                         /* fallthrough: */
 333                 }
 334                 default:
 335                         inode_u->bi_generation = bi_generation;
 336
 337                         bch2_inode_pack(&inode_p, inode_u);
 338                         inode_p.inode.k.p = k.k->p;
 339
 340                         ret = bch2_btree_insert_at(c, NULL, NULL, NULL,
 341                                         BTREE_INSERT_ATOMIC,
 342                                         BTREE_INSERT_ENTRY(&iter,
 343                                                            &inode_p.inode.k_i));
 344
 345                         if (ret != -EINTR) {
 346                                 bch2_btree_iter_unlock(&iter);
 347
 348                                 if (!ret) {
 349                                         inode_u->bi_inum =
 350                                                 inode_p.inode.k.p.inode;
 351                                         *hint = inode_p.inode.k.p.inode + 1;
 352                                 }
 353
 354                                 return ret;
 355                         }
 356
 357                         if (ret == -EINTR)
 358                                 continue;
 359
 360                 }
 361         }
 362 out:
 363         bch2_btree_iter_unlock(&iter);
 364
 365         if (!searched_from_start) {
 366                 /* Retry from start */
 367                 *hint = min;
 368                 searched_from_start = true;
 369                 goto again;
 370         }
 371
 372         return -ENOSPC;
 373 }
 374
 375 int bch2_inode_truncate(struct bch_fs *c, u64 inode_nr, u64 new_size,
 376                         struct extent_insert_hook *hook, u64 *journal_seq)
 377 {
 378         return bch2_btree_delete_range(c, BTREE_ID_EXTENTS,
 379                                        POS(inode_nr, new_size),
 380                                        POS(inode_nr + 1, 0),
 381                                        ZERO_VERSION, NULL, hook,
 382                                        journal_seq);
 383 }
 384
 385 int bch2_inode_rm(struct bch_fs *c, u64 inode_nr)
 386 {
 387         struct btree_iter iter;
 388         struct bkey_i_inode_generation delete;
 389         int ret;
 390
 391         ret = bch2_inode_truncate(c, inode_nr, 0, NULL, NULL);
 392         if (ret < 0)
 393                 return ret;
 394
 395         ret = bch2_btree_delete_range(c, BTREE_ID_XATTRS,
 396                                      POS(inode_nr, 0),
 397                                      POS(inode_nr + 1, 0),
 398                                      ZERO_VERSION, NULL, NULL, NULL);
 399         if (ret < 0)
 400                 return ret;
 401
 402         /*
 403          * If this was a directory, there shouldn't be any real dirents left -
 404          * but there could be whiteouts (from hash collisions) that we should
 405          * delete:
 406          *
 407          * XXX: the dirent could ideally would delete whiteouts when they're no
 408          * longer needed
 409          */
 410         ret = bch2_btree_delete_range(c, BTREE_ID_DIRENTS,
 411                                      POS(inode_nr, 0),
 412                                      POS(inode_nr + 1, 0),
 413                                      ZERO_VERSION, NULL, NULL, NULL);
 414         if (ret < 0)
 415                 return ret;
 416
 417         bch2_btree_iter_init(&iter, c, BTREE_ID_INODES, POS(inode_nr, 0),
 418                              BTREE_ITER_INTENT);
 419         do {
 420                 struct bkey_s_c k = bch2_btree_iter_peek_with_holes(&iter);
 421                 u32 bi_generation = 0;
 422
 423                 ret = btree_iter_err(k);
 424                 if (ret) {
 425                         bch2_btree_iter_unlock(&iter);
 426                         return ret;
 427                 }
 428
 429                 bch2_fs_inconsistent_on(k.k->type != BCH_INODE_FS, c,
 430                                         "inode %llu not found when deleting",
 431                                         inode_nr);
 432
 433                 switch (k.k->type) {
 434                 case BCH_INODE_FS: {
 435                         struct bch_inode_unpacked inode_u;
 436
 437                         if (!bch2_inode_unpack(bkey_s_c_to_inode(k), &inode_u))
 438                                 bi_generation = inode_u.bi_generation + 1;
 439                         break;
 440                 }
 441                 case BCH_INODE_GENERATION: {
 442                         struct bkey_s_c_inode_generation g =
 443                                 bkey_s_c_to_inode_generation(k);
 444                         bi_generation = le32_to_cpu(g.v->bi_generation);
 445                         break;
 446                 }
 447                 }
 448
 449                 if (!bi_generation) {
 450                         bkey_init(&delete.k);
 451                         delete.k.p.inode = inode_nr;
 452                 } else {
 453                         bkey_inode_generation_init(&delete.k_i);
 454                         delete.k.p.inode = inode_nr;
 455                         delete.v.bi_generation = cpu_to_le32(bi_generation);
 456                 }
 457
 458                 ret = bch2_btree_insert_at(c, NULL, NULL, NULL,
 459                                 BTREE_INSERT_ATOMIC|
 460                                 BTREE_INSERT_NOFAIL,
 461                                 BTREE_INSERT_ENTRY(&iter, &delete.k_i));
 462         } while (ret == -EINTR);
 463
 464         bch2_btree_iter_unlock(&iter);
 465         return ret;
 466 }
 467
 468 int bch2_inode_find_by_inum(struct bch_fs *c, u64 inode_nr,
 469                             struct bch_inode_unpacked *inode)
 470 {
 471         struct btree_iter iter;
 472         struct bkey_s_c k;
 473         int ret = -ENOENT;
 474
 475         for_each_btree_key(&iter, c, BTREE_ID_INODES,
 476                            POS(inode_nr, 0),
 477                            BTREE_ITER_WITH_HOLES, k) {
 478                 switch (k.k->type) {
 479                 case BCH_INODE_FS:
 480                         ret = bch2_inode_unpack(bkey_s_c_to_inode(k), inode);
 481                         break;
 482                 default:
 483                         /* hole, not found */
 484                         break;
 485                 }
 486
 487                 break;
 488
 489         }
 490
 491         return bch2_btree_iter_unlock(&iter) ?: ret;
 492 }
 493
 494 int bch2_cached_dev_inode_find_by_uuid(struct bch_fs *c, uuid_le *uuid,
 495                                        struct bkey_i_inode_blockdev *ret)
 496 {
 497         struct btree_iter iter;
 498         struct bkey_s_c k;
 499
 500         for_each_btree_key(&iter, c, BTREE_ID_INODES, POS(0, 0), 0, k) {
 501                 if (k.k->p.inode >= BLOCKDEV_INODE_MAX)
 502                         break;
 503
 504                 if (k.k->type == BCH_INODE_BLOCKDEV) {
 505                         struct bkey_s_c_inode_blockdev inode =
 506                                 bkey_s_c_to_inode_blockdev(k);
 507
 508                         pr_debug("found inode %llu: %pU (u64s %u)",
 509                                  inode.k->p.inode, inode.v->i_uuid.b,
 510                                  inode.k->u64s);
 511
 512                         if (CACHED_DEV(inode.v) &&
 513                             !memcmp(uuid, &inode.v->i_uuid, 16)) {
 514                                 bkey_reassemble(&ret->k_i, k);
 515                                 bch2_btree_iter_unlock(&iter);
 516                                 return 0;
 517                         }
 518                 }
 519
 520                 bch2_btree_iter_cond_resched(&iter);
 521         }
 522         bch2_btree_iter_unlock(&iter);
 523         return -ENOENT;
 524 }
 525
 526 #ifdef CONFIG_BCACHEFS_DEBUG
 527 void bch2_inode_pack_test(void)
 528 {
 529         struct bch_inode_unpacked *u, test_inodes[] = {
 530                 {
 531                         .bi_atime       = U64_MAX,
 532                         .bi_ctime       = U64_MAX,
 533                         .bi_mtime       = U64_MAX,
 534                         .bi_otime       = U64_MAX,
 535                         .bi_size        = U64_MAX,
 536                         .bi_sectors     = U64_MAX,
 537                         .bi_uid         = U32_MAX,
 538                         .bi_gid         = U32_MAX,
 539                         .bi_nlink       = U32_MAX,
 540                         .bi_generation  = U32_MAX,
 541                         .bi_dev         = U32_MAX,
 542                 },
 543         };
 544
 545         for (u = test_inodes;
 546              u < test_inodes + ARRAY_SIZE(test_inodes);
 547              u++) {
 548                 struct bkey_inode_buf p;
 549
 550                 bch2_inode_pack(&p, u);
 551         }
 552 }
 553 #endif