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

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "btree_key_cache.h"
   5 #include "bkey_methods.h"
   6 #include "btree_update.h"
   7 #include "buckets.h"
   8 #include "error.h"
   9 #include "extents.h"
  10 #include "extent_update.h"
  11 #include "inode.h"
  12 #include "str_hash.h"
  13 #include "subvolume.h"
  14 #include "varint.h"
  15
  16 #include <linux/random.h>
  17
  18 #include <asm/unaligned.h>
  19
  20 const char * const bch2_inode_opts[] = {
  21 #define x(name, ...)    #name,
  22         BCH_INODE_OPTS()
  23 #undef  x
  24         NULL,
  25 };
  26
  27 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
  28
  29 static int inode_decode_field(const u8 *in, const u8 *end,
  30                               u64 out[2], unsigned *out_bits)
  31 {
  32         __be64 be[2] = { 0, 0 };
  33         unsigned bytes, shift;
  34         u8 *p;
  35
  36         if (in >= end)
  37                 return -1;
  38
  39         if (!*in)
  40                 return -1;
  41
  42         /*
  43          * position of highest set bit indicates number of bytes:
  44          * shift = number of bits to remove in high byte:
  45          */
  46         shift   = 8 - __fls(*in); /* 1 <= shift <= 8 */
  47         bytes   = byte_table[shift - 1];
  48
  49         if (in + bytes > end)
  50                 return -1;
  51
  52         p = (u8 *) be + 16 - bytes;
  53         memcpy(p, in, bytes);
  54         *p ^= (1 << 8) >> shift;
  55
  56         out[0] = be64_to_cpu(be[0]);
  57         out[1] = be64_to_cpu(be[1]);
  58         *out_bits = out[0] ? 64 + fls64(out[0]) : fls64(out[1]);
  59
  60         return bytes;
  61 }
  62
  63 static inline void bch2_inode_pack_inlined(struct bkey_inode_buf *packed,
  64                                            const struct bch_inode_unpacked *inode)
  65 {
  66         struct bkey_i_inode_v3 *k = &packed->inode;
  67         u8 *out = k->v.fields;
  68         u8 *end = (void *) &packed[1];
  69         u8 *last_nonzero_field = out;
  70         unsigned nr_fields = 0, last_nonzero_fieldnr = 0;
  71         unsigned bytes;
  72         int ret;
  73
  74         bkey_inode_v3_init(&packed->inode.k_i);
  75         packed->inode.k.p.offset        = inode->bi_inum;
  76         packed->inode.v.bi_journal_seq  = cpu_to_le64(inode->bi_journal_seq);
  77         packed->inode.v.bi_hash_seed    = inode->bi_hash_seed;
  78         packed->inode.v.bi_flags        = cpu_to_le64(inode->bi_flags);
  79         packed->inode.v.bi_sectors      = cpu_to_le64(inode->bi_sectors);
  80         packed->inode.v.bi_size         = cpu_to_le64(inode->bi_size);
  81         packed->inode.v.bi_version      = cpu_to_le64(inode->bi_version);
  82         SET_INODEv3_MODE(&packed->inode.v, inode->bi_mode);
  83         SET_INODEv3_FIELDS_START(&packed->inode.v, INODEv3_FIELDS_START_CUR);
  84
  85
  86 #define x(_name, _bits)                                                 \
  87         nr_fields++;                                                    \
  88                                                                         \
  89         if (inode->_name) {                                             \
  90                 ret = bch2_varint_encode_fast(out, inode->_name);       \
  91                 out += ret;                                             \
  92                                                                         \
  93                 if (_bits > 64)                                         \
  94                         *out++ = 0;                                     \
  95                                                                         \
  96                 last_nonzero_field = out;                               \
  97                 last_nonzero_fieldnr = nr_fields;                       \
  98         } else {                                                        \
  99                 *out++ = 0;                                             \
 100                                                                         \
 101                 if (_bits > 64)                                         \
 102                         *out++ = 0;                                     \
 103         }
 104
 105         BCH_INODE_FIELDS_v3()
 106 #undef  x
 107         BUG_ON(out > end);
 108
 109         out = last_nonzero_field;
 110         nr_fields = last_nonzero_fieldnr;
 111
 112         bytes = out - (u8 *) &packed->inode.v;
 113         set_bkey_val_bytes(&packed->inode.k, bytes);
 114         memset_u64s_tail(&packed->inode.v, 0, bytes);
 115
 116         SET_INODEv3_NR_FIELDS(&k->v, nr_fields);
 117
 118         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
 119                 struct bch_inode_unpacked unpacked;
 120
 121                 int ret = bch2_inode_unpack(bkey_i_to_s_c(&packed->inode.k_i),
 122                                            &unpacked);
 123                 BUG_ON(ret);
 124                 BUG_ON(unpacked.bi_inum         != inode->bi_inum);
 125                 BUG_ON(unpacked.bi_hash_seed    != inode->bi_hash_seed);
 126                 BUG_ON(unpacked.bi_sectors      != inode->bi_sectors);
 127                 BUG_ON(unpacked.bi_size         != inode->bi_size);
 128                 BUG_ON(unpacked.bi_version      != inode->bi_version);
 129                 BUG_ON(unpacked.bi_mode         != inode->bi_mode);
 130
 131 #define x(_name, _bits) if (unpacked._name != inode->_name)             \
 132                         panic("unpacked %llu should be %llu",           \
 133                               (u64) unpacked._name, (u64) inode->_name);
 134                 BCH_INODE_FIELDS_v3()
 135 #undef  x
 136         }
 137 }
 138
 139 void bch2_inode_pack(struct bkey_inode_buf *packed,
 140                      const struct bch_inode_unpacked *inode)
 141 {
 142         bch2_inode_pack_inlined(packed, inode);
 143 }
 144
 145 static noinline int bch2_inode_unpack_v1(struct bkey_s_c_inode inode,
 146                                 struct bch_inode_unpacked *unpacked)
 147 {
 148         const u8 *in = inode.v->fields;
 149         const u8 *end = bkey_val_end(inode);
 150         u64 field[2];
 151         unsigned fieldnr = 0, field_bits;
 152         int ret;
 153
 154 #define x(_name, _bits)                                 \
 155         if (fieldnr++ == INODE_NR_FIELDS(inode.v)) {                    \
 156                 unsigned offset = offsetof(struct bch_inode_unpacked, _name);\
 157                 memset((void *) unpacked + offset, 0,                   \
 158                        sizeof(*unpacked) - offset);                     \
 159                 return 0;                                               \
 160         }                                                               \
 161                                                                         \
 162         ret = inode_decode_field(in, end, field, &field_bits);          \
 163         if (ret < 0)                                                    \
 164                 return ret;                                             \
 165                                                                         \
 166         if (field_bits > sizeof(unpacked->_name) * 8)                   \
 167                 return -1;                                              \
 168                                                                         \
 169         unpacked->_name = field[1];                                     \
 170         in += ret;
 171
 172         BCH_INODE_FIELDS_v2()
 173 #undef  x
 174
 175         /* XXX: signal if there were more fields than expected? */
 176         return 0;
 177 }
 178
 179 static int bch2_inode_unpack_v2(struct bch_inode_unpacked *unpacked,
 180                                 const u8 *in, const u8 *end,
 181                                 unsigned nr_fields)
 182 {
 183         unsigned fieldnr = 0;
 184         int ret;
 185         u64 v[2];
 186
 187 #define x(_name, _bits)                                                 \
 188         if (fieldnr < nr_fields) {                                      \
 189                 ret = bch2_varint_decode_fast(in, end, &v[0]);          \
 190                 if (ret < 0)                                            \
 191                         return ret;                                     \
 192                 in += ret;                                              \
 193                                                                         \
 194                 if (_bits > 64) {                                       \
 195                         ret = bch2_varint_decode_fast(in, end, &v[1]);  \
 196                         if (ret < 0)                                    \
 197                                 return ret;                             \
 198                         in += ret;                                      \
 199                 } else {                                                \
 200                         v[1] = 0;                                       \
 201                 }                                                       \
 202         } else {                                                        \
 203                 v[0] = v[1] = 0;                                        \
 204         }                                                               \
 205                                                                         \
 206         unpacked->_name = v[0];                                         \
 207         if (v[1] || v[0] != unpacked->_name)                            \
 208                 return -1;                                              \
 209         fieldnr++;
 210
 211         BCH_INODE_FIELDS_v2()
 212 #undef  x
 213
 214         /* XXX: signal if there were more fields than expected? */
 215         return 0;
 216 }
 217
 218 static int bch2_inode_unpack_v3(struct bkey_s_c k,
 219                                 struct bch_inode_unpacked *unpacked)
 220 {
 221         struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
 222         const u8 *in = inode.v->fields;
 223         const u8 *end = bkey_val_end(inode);
 224         unsigned nr_fields = INODEv3_NR_FIELDS(inode.v);
 225         unsigned fieldnr = 0;
 226         int ret;
 227         u64 v[2];
 228
 229         unpacked->bi_inum       = inode.k->p.offset;
 230         unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
 231         unpacked->bi_hash_seed  = inode.v->bi_hash_seed;
 232         unpacked->bi_flags      = le64_to_cpu(inode.v->bi_flags);
 233         unpacked->bi_sectors    = le64_to_cpu(inode.v->bi_sectors);
 234         unpacked->bi_size       = le64_to_cpu(inode.v->bi_size);
 235         unpacked->bi_version    = le64_to_cpu(inode.v->bi_version);
 236         unpacked->bi_mode       = INODEv3_MODE(inode.v);
 237
 238 #define x(_name, _bits)                                                 \
 239         if (fieldnr < nr_fields) {                                      \
 240                 ret = bch2_varint_decode_fast(in, end, &v[0]);          \
 241                 if (ret < 0)                                            \
 242                         return ret;                                     \
 243                 in += ret;                                              \
 244                                                                         \
 245                 if (_bits > 64) {                                       \
 246                         ret = bch2_varint_decode_fast(in, end, &v[1]);  \
 247                         if (ret < 0)                                    \
 248                                 return ret;                             \
 249                         in += ret;                                      \
 250                 } else {                                                \
 251                         v[1] = 0;                                       \
 252                 }                                                       \
 253         } else {                                                        \
 254                 v[0] = v[1] = 0;                                        \
 255         }                                                               \
 256                                                                         \
 257         unpacked->_name = v[0];                                         \
 258         if (v[1] || v[0] != unpacked->_name)                            \
 259                 return -1;                                              \
 260         fieldnr++;
 261
 262         BCH_INODE_FIELDS_v3()
 263 #undef  x
 264
 265         /* XXX: signal if there were more fields than expected? */
 266         return 0;
 267 }
 268
 269 static noinline int bch2_inode_unpack_slowpath(struct bkey_s_c k,
 270                                                struct bch_inode_unpacked *unpacked)
 271 {
 272         memset(unpacked, 0, sizeof(*unpacked));
 273
 274         switch (k.k->type) {
 275         case KEY_TYPE_inode: {
 276                 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
 277
 278                 unpacked->bi_inum       = inode.k->p.offset;
 279                 unpacked->bi_journal_seq= 0;
 280                 unpacked->bi_hash_seed  = inode.v->bi_hash_seed;
 281                 unpacked->bi_flags      = le32_to_cpu(inode.v->bi_flags);
 282                 unpacked->bi_mode       = le16_to_cpu(inode.v->bi_mode);
 283
 284                 if (INODE_NEW_VARINT(inode.v)) {
 285                         return bch2_inode_unpack_v2(unpacked, inode.v->fields,
 286                                                     bkey_val_end(inode),
 287                                                     INODE_NR_FIELDS(inode.v));
 288                 } else {
 289                         return bch2_inode_unpack_v1(inode, unpacked);
 290                 }
 291                 break;
 292         }
 293         case KEY_TYPE_inode_v2: {
 294                 struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
 295
 296                 unpacked->bi_inum       = inode.k->p.offset;
 297                 unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
 298                 unpacked->bi_hash_seed  = inode.v->bi_hash_seed;
 299                 unpacked->bi_flags      = le64_to_cpu(inode.v->bi_flags);
 300                 unpacked->bi_mode       = le16_to_cpu(inode.v->bi_mode);
 301
 302                 return bch2_inode_unpack_v2(unpacked, inode.v->fields,
 303                                             bkey_val_end(inode),
 304                                             INODEv2_NR_FIELDS(inode.v));
 305         }
 306         default:
 307                 BUG();
 308         }
 309 }
 310
 311 int bch2_inode_unpack(struct bkey_s_c k,
 312                       struct bch_inode_unpacked *unpacked)
 313 {
 314         if (likely(k.k->type == KEY_TYPE_inode_v3))
 315                 return bch2_inode_unpack_v3(k, unpacked);
 316         return bch2_inode_unpack_slowpath(k, unpacked);
 317 }
 318
 319 int bch2_inode_peek(struct btree_trans *trans,
 320                     struct btree_iter *iter,
 321                     struct bch_inode_unpacked *inode,
 322                     subvol_inum inum, unsigned flags)
 323 {
 324         struct bkey_s_c k;
 325         u32 snapshot;
 326         int ret;
 327
 328         ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
 329         if (ret)
 330                 return ret;
 331
 332         k = bch2_bkey_get_iter(trans, iter, BTREE_ID_inodes,
 333                                SPOS(0, inum.inum, snapshot),
 334                                flags|BTREE_ITER_CACHED);
 335         ret = bkey_err(k);
 336         if (ret)
 337                 return ret;
 338
 339         ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
 340         if (ret)
 341                 goto err;
 342
 343         ret = bch2_inode_unpack(k, inode);
 344         if (ret)
 345                 goto err;
 346
 347         return 0;
 348 err:
 349         bch2_trans_iter_exit(trans, iter);
 350         return ret;
 351 }
 352
 353 int bch2_inode_write(struct btree_trans *trans,
 354                      struct btree_iter *iter,
 355                      struct bch_inode_unpacked *inode)
 356 {
 357         struct bkey_inode_buf *inode_p;
 358
 359         inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
 360         if (IS_ERR(inode_p))
 361                 return PTR_ERR(inode_p);
 362
 363         bch2_inode_pack_inlined(inode_p, inode);
 364         inode_p->inode.k.p.snapshot = iter->snapshot;
 365         return bch2_trans_update(trans, iter, &inode_p->inode.k_i, 0);
 366 }
 367
 368 struct bkey_i *bch2_inode_to_v3(struct btree_trans *trans, struct bkey_i *k)
 369 {
 370         struct bch_inode_unpacked u;
 371         struct bkey_inode_buf *inode_p;
 372         int ret;
 373
 374         if (!bkey_is_inode(&k->k))
 375                 return ERR_PTR(-ENOENT);
 376
 377         inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
 378         if (IS_ERR(inode_p))
 379                 return ERR_CAST(inode_p);
 380
 381         ret = bch2_inode_unpack(bkey_i_to_s_c(k), &u);
 382         if (ret)
 383                 return ERR_PTR(ret);
 384
 385         bch2_inode_pack(inode_p, &u);
 386         return &inode_p->inode.k_i;
 387 }
 388
 389 static int __bch2_inode_invalid(struct bkey_s_c k, struct printbuf *err)
 390 {
 391         struct bch_inode_unpacked unpacked;
 392
 393         if (k.k->p.inode) {
 394                 prt_printf(err, "nonzero k.p.inode");
 395                 return -BCH_ERR_invalid_bkey;
 396         }
 397
 398         if (k.k->p.offset < BLOCKDEV_INODE_MAX) {
 399                 prt_printf(err, "fs inode in blockdev range");
 400                 return -BCH_ERR_invalid_bkey;
 401         }
 402
 403         if (bch2_inode_unpack(k, &unpacked)) {
 404                 prt_printf(err, "invalid variable length fields");
 405                 return -BCH_ERR_invalid_bkey;
 406         }
 407
 408         if (unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1) {
 409                 prt_printf(err, "invalid data checksum type (%u >= %u",
 410                         unpacked.bi_data_checksum, BCH_CSUM_OPT_NR + 1);
 411                 return -BCH_ERR_invalid_bkey;
 412         }
 413
 414         if (unpacked.bi_compression >= BCH_COMPRESSION_OPT_NR + 1) {
 415                 prt_printf(err, "invalid data checksum type (%u >= %u)",
 416                        unpacked.bi_compression, BCH_COMPRESSION_OPT_NR + 1);
 417                 return -BCH_ERR_invalid_bkey;
 418         }
 419
 420         if ((unpacked.bi_flags & BCH_INODE_UNLINKED) &&
 421             unpacked.bi_nlink != 0) {
 422                 prt_printf(err, "flagged as unlinked but bi_nlink != 0");
 423                 return -BCH_ERR_invalid_bkey;
 424         }
 425
 426         if (unpacked.bi_subvol && !S_ISDIR(unpacked.bi_mode)) {
 427                 prt_printf(err, "subvolume root but not a directory");
 428                 return -BCH_ERR_invalid_bkey;
 429         }
 430
 431         return 0;
 432 }
 433
 434 int bch2_inode_invalid(const struct bch_fs *c, struct bkey_s_c k,
 435                        enum bkey_invalid_flags flags,
 436                        struct printbuf *err)
 437 {
 438         struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
 439
 440         if (INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR) {
 441                 prt_printf(err, "invalid str hash type (%llu >= %u)",
 442                        INODE_STR_HASH(inode.v), BCH_STR_HASH_NR);
 443                 return -BCH_ERR_invalid_bkey;
 444         }
 445
 446         return __bch2_inode_invalid(k, err);
 447 }
 448
 449 int bch2_inode_v2_invalid(const struct bch_fs *c, struct bkey_s_c k,
 450                           enum bkey_invalid_flags flags,
 451                           struct printbuf *err)
 452 {
 453         struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
 454
 455         if (INODEv2_STR_HASH(inode.v) >= BCH_STR_HASH_NR) {
 456                 prt_printf(err, "invalid str hash type (%llu >= %u)",
 457                        INODEv2_STR_HASH(inode.v), BCH_STR_HASH_NR);
 458                 return -BCH_ERR_invalid_bkey;
 459         }
 460
 461         return __bch2_inode_invalid(k, err);
 462 }
 463
 464 int bch2_inode_v3_invalid(const struct bch_fs *c, struct bkey_s_c k,
 465                           enum bkey_invalid_flags flags,
 466                           struct printbuf *err)
 467 {
 468         struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
 469
 470         if (INODEv3_FIELDS_START(inode.v) < INODEv3_FIELDS_START_INITIAL ||
 471             INODEv3_FIELDS_START(inode.v) > bkey_val_u64s(inode.k)) {
 472                 prt_printf(err, "invalid fields_start (got %llu, min %u max %zu)",
 473                        INODEv3_FIELDS_START(inode.v),
 474                        INODEv3_FIELDS_START_INITIAL,
 475                        bkey_val_u64s(inode.k));
 476                 return -BCH_ERR_invalid_bkey;
 477         }
 478
 479         if (INODEv3_STR_HASH(inode.v) >= BCH_STR_HASH_NR) {
 480                 prt_printf(err, "invalid str hash type (%llu >= %u)",
 481                        INODEv3_STR_HASH(inode.v), BCH_STR_HASH_NR);
 482                 return -BCH_ERR_invalid_bkey;
 483         }
 484
 485         return __bch2_inode_invalid(k, err);
 486 }
 487
 488 static void __bch2_inode_unpacked_to_text(struct printbuf *out,
 489                                           struct bch_inode_unpacked *inode)
 490 {
 491         prt_printf(out, "mode %o flags %x journal_seq %llu bi_size %llu bi_sectors %llu bi_version %llu",
 492                inode->bi_mode, inode->bi_flags,
 493                inode->bi_journal_seq,
 494                inode->bi_size,
 495                inode->bi_sectors,
 496                inode->bi_version);
 497
 498 #define x(_name, _bits)                                         \
 499         prt_printf(out, " "#_name " %llu", (u64) inode->_name);
 500         BCH_INODE_FIELDS_v3()
 501 #undef  x
 502 }
 503
 504 void bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked *inode)
 505 {
 506         prt_printf(out, "inum: %llu ", inode->bi_inum);
 507         __bch2_inode_unpacked_to_text(out, inode);
 508 }
 509
 510 void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
 511 {
 512         struct bch_inode_unpacked inode;
 513
 514         if (bch2_inode_unpack(k, &inode)) {
 515                 prt_printf(out, "(unpack error)");
 516                 return;
 517         }
 518
 519         __bch2_inode_unpacked_to_text(out, &inode);
 520 }
 521
 522 int bch2_trans_mark_inode(struct btree_trans *trans,
 523                           enum btree_id btree_id, unsigned level,
 524                           struct bkey_s_c old,
 525                           struct bkey_i *new,
 526                           unsigned flags)
 527 {
 528         int nr = bkey_is_inode(&new->k) - bkey_is_inode(old.k);
 529
 530         if (nr) {
 531                 int ret = bch2_replicas_deltas_realloc(trans, 0);
 532                 struct replicas_delta_list *d = trans->fs_usage_deltas;
 533
 534                 if (ret)
 535                         return ret;
 536
 537                 d->nr_inodes += nr;
 538         }
 539
 540         return 0;
 541 }
 542
 543 int bch2_mark_inode(struct btree_trans *trans,
 544                     enum btree_id btree_id, unsigned level,
 545                     struct bkey_s_c old, struct bkey_s_c new,
 546                     unsigned flags)
 547 {
 548         struct bch_fs *c = trans->c;
 549         struct bch_fs_usage *fs_usage;
 550         u64 journal_seq = trans->journal_res.seq;
 551
 552         if (flags & BTREE_TRIGGER_INSERT) {
 553                 struct bch_inode_v3 *v = (struct bch_inode_v3 *) new.v;
 554
 555                 BUG_ON(!journal_seq);
 556                 BUG_ON(new.k->type != KEY_TYPE_inode_v3);
 557
 558                 v->bi_journal_seq = cpu_to_le64(journal_seq);
 559         }
 560
 561         if (flags & BTREE_TRIGGER_GC) {
 562                 percpu_down_read(&c->mark_lock);
 563                 preempt_disable();
 564
 565                 fs_usage = fs_usage_ptr(c, journal_seq, flags & BTREE_TRIGGER_GC);
 566                 fs_usage->nr_inodes += bkey_is_inode(new.k);
 567                 fs_usage->nr_inodes -= bkey_is_inode(old.k);
 568
 569                 preempt_enable();
 570                 percpu_up_read(&c->mark_lock);
 571         }
 572         return 0;
 573 }
 574
 575 int bch2_inode_generation_invalid(const struct bch_fs *c, struct bkey_s_c k,
 576                                   enum bkey_invalid_flags flags,
 577                                   struct printbuf *err)
 578 {
 579         if (k.k->p.inode) {
 580                 prt_printf(err, "nonzero k.p.inode");
 581                 return -BCH_ERR_invalid_bkey;
 582         }
 583
 584         return 0;
 585 }
 586
 587 void bch2_inode_generation_to_text(struct printbuf *out, struct bch_fs *c,
 588                                    struct bkey_s_c k)
 589 {
 590         struct bkey_s_c_inode_generation gen = bkey_s_c_to_inode_generation(k);
 591
 592         prt_printf(out, "generation: %u", le32_to_cpu(gen.v->bi_generation));
 593 }
 594
 595 void bch2_inode_init_early(struct bch_fs *c,
 596                            struct bch_inode_unpacked *inode_u)
 597 {
 598         enum bch_str_hash_type str_hash =
 599                 bch2_str_hash_opt_to_type(c, c->opts.str_hash);
 600
 601         memset(inode_u, 0, sizeof(*inode_u));
 602
 603         /* ick */
 604         inode_u->bi_flags |= str_hash << INODE_STR_HASH_OFFSET;
 605         get_random_bytes(&inode_u->bi_hash_seed,
 606                          sizeof(inode_u->bi_hash_seed));
 607 }
 608
 609 void bch2_inode_init_late(struct bch_inode_unpacked *inode_u, u64 now,
 610                           uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
 611                           struct bch_inode_unpacked *parent)
 612 {
 613         inode_u->bi_mode        = mode;
 614         inode_u->bi_uid         = uid;
 615         inode_u->bi_gid         = gid;
 616         inode_u->bi_dev         = rdev;
 617         inode_u->bi_atime       = now;
 618         inode_u->bi_mtime       = now;
 619         inode_u->bi_ctime       = now;
 620         inode_u->bi_otime       = now;
 621
 622         if (parent && parent->bi_mode & S_ISGID) {
 623                 inode_u->bi_gid = parent->bi_gid;
 624                 if (S_ISDIR(mode))
 625                         inode_u->bi_mode |= S_ISGID;
 626         }
 627
 628         if (parent) {
 629 #define x(_name, ...)   inode_u->bi_##_name = parent->bi_##_name;
 630                 BCH_INODE_OPTS()
 631 #undef x
 632         }
 633 }
 634
 635 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
 636                      uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
 637                      struct bch_inode_unpacked *parent)
 638 {
 639         bch2_inode_init_early(c, inode_u);
 640         bch2_inode_init_late(inode_u, bch2_current_time(c),
 641                              uid, gid, mode, rdev, parent);
 642 }
 643
 644 static inline u32 bkey_generation(struct bkey_s_c k)
 645 {
 646         switch (k.k->type) {
 647         case KEY_TYPE_inode:
 648         case KEY_TYPE_inode_v2:
 649                 BUG();
 650         case KEY_TYPE_inode_generation:
 651                 return le32_to_cpu(bkey_s_c_to_inode_generation(k).v->bi_generation);
 652         default:
 653                 return 0;
 654         }
 655 }
 656
 657 /*
 658  * This just finds an empty slot:
 659  */
 660 int bch2_inode_create(struct btree_trans *trans,
 661                       struct btree_iter *iter,
 662                       struct bch_inode_unpacked *inode_u,
 663                       u32 snapshot, u64 cpu)
 664 {
 665         struct bch_fs *c = trans->c;
 666         struct bkey_s_c k;
 667         u64 min, max, start, pos, *hint;
 668         int ret = 0;
 669         unsigned bits = (c->opts.inodes_32bit ? 31 : 63);
 670
 671         if (c->opts.shard_inode_numbers) {
 672                 bits -= c->inode_shard_bits;
 673
 674                 min = (cpu << bits);
 675                 max = (cpu << bits) | ~(ULLONG_MAX << bits);
 676
 677                 min = max_t(u64, min, BLOCKDEV_INODE_MAX);
 678                 hint = c->unused_inode_hints + cpu;
 679         } else {
 680                 min = BLOCKDEV_INODE_MAX;
 681                 max = ~(ULLONG_MAX << bits);
 682                 hint = c->unused_inode_hints;
 683         }
 684
 685         start = READ_ONCE(*hint);
 686
 687         if (start >= max || start < min)
 688                 start = min;
 689
 690         pos = start;
 691         bch2_trans_iter_init(trans, iter, BTREE_ID_inodes, POS(0, pos),
 692                              BTREE_ITER_ALL_SNAPSHOTS|
 693                              BTREE_ITER_INTENT);
 694 again:
 695         while ((k = bch2_btree_iter_peek(iter)).k &&
 696                !(ret = bkey_err(k)) &&
 697                bkey_lt(k.k->p, POS(0, max))) {
 698                 if (pos < iter->pos.offset)
 699                         goto found_slot;
 700
 701                 /*
 702                  * We don't need to iterate over keys in every snapshot once
 703                  * we've found just one:
 704                  */
 705                 pos = iter->pos.offset + 1;
 706                 bch2_btree_iter_set_pos(iter, POS(0, pos));
 707         }
 708
 709         if (!ret && pos < max)
 710                 goto found_slot;
 711
 712         if (!ret && start == min)
 713                 ret = -BCH_ERR_ENOSPC_inode_create;
 714
 715         if (ret) {
 716                 bch2_trans_iter_exit(trans, iter);
 717                 return ret;
 718         }
 719
 720         /* Retry from start */
 721         pos = start = min;
 722         bch2_btree_iter_set_pos(iter, POS(0, pos));
 723         goto again;
 724 found_slot:
 725         bch2_btree_iter_set_pos(iter, SPOS(0, pos, snapshot));
 726         k = bch2_btree_iter_peek_slot(iter);
 727         ret = bkey_err(k);
 728         if (ret) {
 729                 bch2_trans_iter_exit(trans, iter);
 730                 return ret;
 731         }
 732
 733         *hint                   = k.k->p.offset;
 734         inode_u->bi_inum        = k.k->p.offset;
 735         inode_u->bi_generation  = bkey_generation(k);
 736         return 0;
 737 }
 738
 739 static int bch2_inode_delete_keys(struct btree_trans *trans,
 740                                   subvol_inum inum, enum btree_id id)
 741 {
 742         struct btree_iter iter;
 743         struct bkey_s_c k;
 744         struct bkey_i delete;
 745         u32 snapshot;
 746         int ret = 0;
 747
 748         /*
 749          * We're never going to be deleting partial extents, no need to use an
 750          * extent iterator:
 751          */
 752         bch2_trans_iter_init(trans, &iter, id, POS(inum.inum, 0),
 753                              BTREE_ITER_INTENT|BTREE_ITER_NOT_EXTENTS);
 754
 755         while (1) {
 756                 bch2_trans_begin(trans);
 757
 758                 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
 759                 if (ret)
 760                         goto err;
 761
 762                 bch2_btree_iter_set_snapshot(&iter, snapshot);
 763
 764                 k = bch2_btree_iter_peek_upto(&iter, POS(inum.inum, U64_MAX));
 765                 ret = bkey_err(k);
 766                 if (ret)
 767                         goto err;
 768
 769                 if (!k.k)
 770                         break;
 771
 772                 bkey_init(&delete.k);
 773                 delete.k.p = iter.pos;
 774
 775                 ret = bch2_trans_update(trans, &iter, &delete, 0) ?:
 776                       bch2_trans_commit(trans, NULL, NULL,
 777                                         BTREE_INSERT_NOFAIL);
 778 err:
 779                 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
 780                         break;
 781         }
 782
 783         bch2_trans_iter_exit(trans, &iter);
 784         return ret;
 785 }
 786
 787 int bch2_inode_rm(struct bch_fs *c, subvol_inum inum)
 788 {
 789         struct btree_trans trans;
 790         struct btree_iter iter = { NULL };
 791         struct bkey_i_inode_generation delete;
 792         struct bch_inode_unpacked inode_u;
 793         struct bkey_s_c k;
 794         u32 snapshot;
 795         int ret;
 796
 797         bch2_trans_init(&trans, c, 0, 1024);
 798
 799         /*
 800          * If this was a directory, there shouldn't be any real dirents left -
 801          * but there could be whiteouts (from hash collisions) that we should
 802          * delete:
 803          *
 804          * XXX: the dirent could ideally would delete whiteouts when they're no
 805          * longer needed
 806          */
 807         ret   = bch2_inode_delete_keys(&trans, inum, BTREE_ID_extents) ?:
 808                 bch2_inode_delete_keys(&trans, inum, BTREE_ID_xattrs) ?:
 809                 bch2_inode_delete_keys(&trans, inum, BTREE_ID_dirents);
 810         if (ret)
 811                 goto err;
 812 retry:
 813         bch2_trans_begin(&trans);
 814
 815         ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot);
 816         if (ret)
 817                 goto err;
 818
 819         k = bch2_bkey_get_iter(&trans, &iter, BTREE_ID_inodes,
 820                                SPOS(0, inum.inum, snapshot),
 821                                BTREE_ITER_INTENT|BTREE_ITER_CACHED);
 822         ret = bkey_err(k);
 823         if (ret)
 824                 goto err;
 825
 826         if (!bkey_is_inode(k.k)) {
 827                 bch2_fs_inconsistent(trans.c,
 828                                      "inode %llu:%u not found when deleting",
 829                                      inum.inum, snapshot);
 830                 ret = -EIO;
 831                 goto err;
 832         }
 833
 834         bch2_inode_unpack(k, &inode_u);
 835
 836         bkey_inode_generation_init(&delete.k_i);
 837         delete.k.p = iter.pos;
 838         delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
 839
 840         ret   = bch2_trans_update(&trans, &iter, &delete.k_i, 0) ?:
 841                 bch2_trans_commit(&trans, NULL, NULL,
 842                                 BTREE_INSERT_NOFAIL);
 843 err:
 844         bch2_trans_iter_exit(&trans, &iter);
 845         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 846                 goto retry;
 847
 848         bch2_trans_exit(&trans);
 849         return ret;
 850 }
 851
 852 int bch2_inode_find_by_inum_trans(struct btree_trans *trans,
 853                                   subvol_inum inum,
 854                                   struct bch_inode_unpacked *inode)
 855 {
 856         struct btree_iter iter;
 857         int ret;
 858
 859         ret = bch2_inode_peek(trans, &iter, inode, inum, 0);
 860         if (!ret)
 861                 bch2_trans_iter_exit(trans, &iter);
 862         return ret;
 863 }
 864
 865 int bch2_inode_find_by_inum(struct bch_fs *c, subvol_inum inum,
 866                             struct bch_inode_unpacked *inode)
 867 {
 868         return bch2_trans_do(c, NULL, NULL, 0,
 869                 bch2_inode_find_by_inum_trans(&trans, inum, inode));
 870 }
 871
 872 int bch2_inode_nlink_inc(struct bch_inode_unpacked *bi)
 873 {
 874         if (bi->bi_flags & BCH_INODE_UNLINKED)
 875                 bi->bi_flags &= ~BCH_INODE_UNLINKED;
 876         else {
 877                 if (bi->bi_nlink == U32_MAX)
 878                         return -EINVAL;
 879
 880                 bi->bi_nlink++;
 881         }
 882
 883         return 0;
 884 }
 885
 886 void bch2_inode_nlink_dec(struct btree_trans *trans, struct bch_inode_unpacked *bi)
 887 {
 888         if (bi->bi_nlink && (bi->bi_flags & BCH_INODE_UNLINKED)) {
 889                 bch2_trans_inconsistent(trans, "inode %llu unlinked but link count nonzero",
 890                                         bi->bi_inum);
 891                 return;
 892         }
 893
 894         if (bi->bi_flags & BCH_INODE_UNLINKED) {
 895                 bch2_trans_inconsistent(trans, "inode %llu link count underflow", bi->bi_inum);
 896                 return;
 897         }
 898
 899         if (bi->bi_nlink)
 900                 bi->bi_nlink--;
 901         else
 902                 bi->bi_flags |= BCH_INODE_UNLINKED;
 903 }
 904
 905 struct bch_opts bch2_inode_opts_to_opts(struct bch_inode_unpacked *inode)
 906 {
 907         struct bch_opts ret = { 0 };
 908 #define x(_name, _bits)                                                 \
 909         if (inode->bi_##_name)                                          \
 910                 opt_set(ret, _name, inode->bi_##_name - 1);
 911         BCH_INODE_OPTS()
 912 #undef x
 913         return ret;
 914 }
 915
 916 void bch2_inode_opts_get(struct bch_io_opts *opts, struct bch_fs *c,
 917                          struct bch_inode_unpacked *inode)
 918 {
 919 #define x(_name, _bits)         opts->_name = inode_opt_get(c, inode, _name);
 920         BCH_INODE_OPTS()
 921 #undef x
 922
 923         if (opts->nocow)
 924                 opts->compression = opts->background_compression = opts->data_checksum = opts->erasure_code = 0;
 925 }
 926
 927 int bch2_inode_rm_snapshot(struct btree_trans *trans, u64 inum, u32 snapshot)
 928 {
 929         struct bch_fs *c = trans->c;
 930         struct btree_iter iter = { NULL };
 931         struct bkey_i_inode_generation delete;
 932         struct bch_inode_unpacked inode_u;
 933         struct bkey_s_c k;
 934         int ret;
 935
 936         do {
 937                 ret   = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
 938                                                       SPOS(inum, 0, snapshot),
 939                                                       SPOS(inum, U64_MAX, snapshot),
 940                                                       0, NULL) ?:
 941                         bch2_btree_delete_range_trans(trans, BTREE_ID_dirents,
 942                                                       SPOS(inum, 0, snapshot),
 943                                                       SPOS(inum, U64_MAX, snapshot),
 944                                                       0, NULL) ?:
 945                         bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs,
 946                                                       SPOS(inum, 0, snapshot),
 947                                                       SPOS(inum, U64_MAX, snapshot),
 948                                                       0, NULL);
 949         } while (ret == -BCH_ERR_transaction_restart_nested);
 950         if (ret)
 951                 goto err;
 952 retry:
 953         bch2_trans_begin(trans);
 954
 955         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
 956                                SPOS(0, inum, snapshot), BTREE_ITER_INTENT);
 957         ret = bkey_err(k);
 958         if (ret)
 959                 goto err;
 960
 961         if (!bkey_is_inode(k.k)) {
 962                 bch2_fs_inconsistent(c,
 963                                      "inode %llu:%u not found when deleting",
 964                                      inum, snapshot);
 965                 ret = -EIO;
 966                 goto err;
 967         }
 968
 969         bch2_inode_unpack(k, &inode_u);
 970
 971         /* Subvolume root? */
 972         if (inode_u.bi_subvol)
 973                 bch_warn(c, "deleting inode %llu marked as unlinked, but also a subvolume root!?", inode_u.bi_inum);
 974
 975         bkey_inode_generation_init(&delete.k_i);
 976         delete.k.p = iter.pos;
 977         delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
 978
 979         ret   = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
 980                 bch2_trans_commit(trans, NULL, NULL,
 981                                 BTREE_INSERT_NOFAIL);
 982 err:
 983         bch2_trans_iter_exit(trans, &iter);
 984         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 985                 goto retry;
 986
 987         return ret ?: -BCH_ERR_transaction_restart_nested;
 988 }