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 "btree_write_buffer.h"
   6 #include "bkey_methods.h"
   7 #include "btree_update.h"
   8 #include "buckets.h"
   9 #include "compress.h"
  10 #include "error.h"
  11 #include "extents.h"
  12 #include "extent_update.h"
  13 #include "inode.h"
  14 #include "str_hash.h"
  15 #include "snapshot.h"
  16 #include "subvolume.h"
  17 #include "varint.h"
  18
  19 #include <linux/random.h>
  20
  21 #include <asm/unaligned.h>
  22
  23 #define x(name, ...)    #name,
  24 const char * const bch2_inode_opts[] = {
  25         BCH_INODE_OPTS()
  26         NULL,
  27 };
  28
  29 static const char * const bch2_inode_flag_strs[] = {
  30         BCH_INODE_FLAGS()
  31         NULL
  32 };
  33 #undef  x
  34
  35 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
  36
  37 static int inode_decode_field(const u8 *in, const u8 *end,
  38                               u64 out[2], unsigned *out_bits)
  39 {
  40         __be64 be[2] = { 0, 0 };
  41         unsigned bytes, shift;
  42         u8 *p;
  43
  44         if (in >= end)
  45                 return -1;
  46
  47         if (!*in)
  48                 return -1;
  49
  50         /*
  51          * position of highest set bit indicates number of bytes:
  52          * shift = number of bits to remove in high byte:
  53          */
  54         shift   = 8 - __fls(*in); /* 1 <= shift <= 8 */
  55         bytes   = byte_table[shift - 1];
  56
  57         if (in + bytes > end)
  58                 return -1;
  59
  60         p = (u8 *) be + 16 - bytes;
  61         memcpy(p, in, bytes);
  62         *p ^= (1 << 8) >> shift;
  63
  64         out[0] = be64_to_cpu(be[0]);
  65         out[1] = be64_to_cpu(be[1]);
  66         *out_bits = out[0] ? 64 + fls64(out[0]) : fls64(out[1]);
  67
  68         return bytes;
  69 }
  70
  71 static inline void bch2_inode_pack_inlined(struct bkey_inode_buf *packed,
  72                                            const struct bch_inode_unpacked *inode)
  73 {
  74         struct bkey_i_inode_v3 *k = &packed->inode;
  75         u8 *out = k->v.fields;
  76         u8 *end = (void *) &packed[1];
  77         u8 *last_nonzero_field = out;
  78         unsigned nr_fields = 0, last_nonzero_fieldnr = 0;
  79         unsigned bytes;
  80         int ret;
  81
  82         bkey_inode_v3_init(&packed->inode.k_i);
  83         packed->inode.k.p.offset        = inode->bi_inum;
  84         packed->inode.v.bi_journal_seq  = cpu_to_le64(inode->bi_journal_seq);
  85         packed->inode.v.bi_hash_seed    = inode->bi_hash_seed;
  86         packed->inode.v.bi_flags        = cpu_to_le64(inode->bi_flags);
  87         packed->inode.v.bi_sectors      = cpu_to_le64(inode->bi_sectors);
  88         packed->inode.v.bi_size         = cpu_to_le64(inode->bi_size);
  89         packed->inode.v.bi_version      = cpu_to_le64(inode->bi_version);
  90         SET_INODEv3_MODE(&packed->inode.v, inode->bi_mode);
  91         SET_INODEv3_FIELDS_START(&packed->inode.v, INODEv3_FIELDS_START_CUR);
  92
  93
  94 #define x(_name, _bits)                                                 \
  95         nr_fields++;                                                    \
  96                                                                         \
  97         if (inode->_name) {                                             \
  98                 ret = bch2_varint_encode_fast(out, inode->_name);       \
  99                 out += ret;                                             \
 100                                                                         \
 101                 if (_bits > 64)                                         \
 102                         *out++ = 0;                                     \
 103                                                                         \
 104                 last_nonzero_field = out;                               \
 105                 last_nonzero_fieldnr = nr_fields;                       \
 106         } else {                                                        \
 107                 *out++ = 0;                                             \
 108                                                                         \
 109                 if (_bits > 64)                                         \
 110                         *out++ = 0;                                     \
 111         }
 112
 113         BCH_INODE_FIELDS_v3()
 114 #undef  x
 115         BUG_ON(out > end);
 116
 117         out = last_nonzero_field;
 118         nr_fields = last_nonzero_fieldnr;
 119
 120         bytes = out - (u8 *) &packed->inode.v;
 121         set_bkey_val_bytes(&packed->inode.k, bytes);
 122         memset_u64s_tail(&packed->inode.v, 0, bytes);
 123
 124         SET_INODEv3_NR_FIELDS(&k->v, nr_fields);
 125
 126         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
 127                 struct bch_inode_unpacked unpacked;
 128
 129                 ret = bch2_inode_unpack(bkey_i_to_s_c(&packed->inode.k_i), &unpacked);
 130                 BUG_ON(ret);
 131                 BUG_ON(unpacked.bi_inum         != inode->bi_inum);
 132                 BUG_ON(unpacked.bi_hash_seed    != inode->bi_hash_seed);
 133                 BUG_ON(unpacked.bi_sectors      != inode->bi_sectors);
 134                 BUG_ON(unpacked.bi_size         != inode->bi_size);
 135                 BUG_ON(unpacked.bi_version      != inode->bi_version);
 136                 BUG_ON(unpacked.bi_mode         != inode->bi_mode);
 137
 138 #define x(_name, _bits) if (unpacked._name != inode->_name)             \
 139                         panic("unpacked %llu should be %llu",           \
 140                               (u64) unpacked._name, (u64) inode->_name);
 141                 BCH_INODE_FIELDS_v3()
 142 #undef  x
 143         }
 144 }
 145
 146 void bch2_inode_pack(struct bkey_inode_buf *packed,
 147                      const struct bch_inode_unpacked *inode)
 148 {
 149         bch2_inode_pack_inlined(packed, inode);
 150 }
 151
 152 static noinline int bch2_inode_unpack_v1(struct bkey_s_c_inode inode,
 153                                 struct bch_inode_unpacked *unpacked)
 154 {
 155         const u8 *in = inode.v->fields;
 156         const u8 *end = bkey_val_end(inode);
 157         u64 field[2];
 158         unsigned fieldnr = 0, field_bits;
 159         int ret;
 160
 161 #define x(_name, _bits)                                 \
 162         if (fieldnr++ == INODE_NR_FIELDS(inode.v)) {                    \
 163                 unsigned offset = offsetof(struct bch_inode_unpacked, _name);\
 164                 memset((void *) unpacked + offset, 0,                   \
 165                        sizeof(*unpacked) - offset);                     \
 166                 return 0;                                               \
 167         }                                                               \
 168                                                                         \
 169         ret = inode_decode_field(in, end, field, &field_bits);          \
 170         if (ret < 0)                                                    \
 171                 return ret;                                             \
 172                                                                         \
 173         if (field_bits > sizeof(unpacked->_name) * 8)                   \
 174                 return -1;                                              \
 175                                                                         \
 176         unpacked->_name = field[1];                                     \
 177         in += ret;
 178
 179         BCH_INODE_FIELDS_v2()
 180 #undef  x
 181
 182         /* XXX: signal if there were more fields than expected? */
 183         return 0;
 184 }
 185
 186 static int bch2_inode_unpack_v2(struct bch_inode_unpacked *unpacked,
 187                                 const u8 *in, const u8 *end,
 188                                 unsigned nr_fields)
 189 {
 190         unsigned fieldnr = 0;
 191         int ret;
 192         u64 v[2];
 193
 194 #define x(_name, _bits)                                                 \
 195         if (fieldnr < nr_fields) {                                      \
 196                 ret = bch2_varint_decode_fast(in, end, &v[0]);          \
 197                 if (ret < 0)                                            \
 198                         return ret;                                     \
 199                 in += ret;                                              \
 200                                                                         \
 201                 if (_bits > 64) {                                       \
 202                         ret = bch2_varint_decode_fast(in, end, &v[1]);  \
 203                         if (ret < 0)                                    \
 204                                 return ret;                             \
 205                         in += ret;                                      \
 206                 } else {                                                \
 207                         v[1] = 0;                                       \
 208                 }                                                       \
 209         } else {                                                        \
 210                 v[0] = v[1] = 0;                                        \
 211         }                                                               \
 212                                                                         \
 213         unpacked->_name = v[0];                                         \
 214         if (v[1] || v[0] != unpacked->_name)                            \
 215                 return -1;                                              \
 216         fieldnr++;
 217
 218         BCH_INODE_FIELDS_v2()
 219 #undef  x
 220
 221         /* XXX: signal if there were more fields than expected? */
 222         return 0;
 223 }
 224
 225 static int bch2_inode_unpack_v3(struct bkey_s_c k,
 226                                 struct bch_inode_unpacked *unpacked)
 227 {
 228         struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
 229         const u8 *in = inode.v->fields;
 230         const u8 *end = bkey_val_end(inode);
 231         unsigned nr_fields = INODEv3_NR_FIELDS(inode.v);
 232         unsigned fieldnr = 0;
 233         int ret;
 234         u64 v[2];
 235
 236         unpacked->bi_inum       = inode.k->p.offset;
 237         unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
 238         unpacked->bi_hash_seed  = inode.v->bi_hash_seed;
 239         unpacked->bi_flags      = le64_to_cpu(inode.v->bi_flags);
 240         unpacked->bi_sectors    = le64_to_cpu(inode.v->bi_sectors);
 241         unpacked->bi_size       = le64_to_cpu(inode.v->bi_size);
 242         unpacked->bi_version    = le64_to_cpu(inode.v->bi_version);
 243         unpacked->bi_mode       = INODEv3_MODE(inode.v);
 244
 245 #define x(_name, _bits)                                                 \
 246         if (fieldnr < nr_fields) {                                      \
 247                 ret = bch2_varint_decode_fast(in, end, &v[0]);          \
 248                 if (ret < 0)                                            \
 249                         return ret;                                     \
 250                 in += ret;                                              \
 251                                                                         \
 252                 if (_bits > 64) {                                       \
 253                         ret = bch2_varint_decode_fast(in, end, &v[1]);  \
 254                         if (ret < 0)                                    \
 255                                 return ret;                             \
 256                         in += ret;                                      \
 257                 } else {                                                \
 258                         v[1] = 0;                                       \
 259                 }                                                       \
 260         } else {                                                        \
 261                 v[0] = v[1] = 0;                                        \
 262         }                                                               \
 263                                                                         \
 264         unpacked->_name = v[0];                                         \
 265         if (v[1] || v[0] != unpacked->_name)                            \
 266                 return -1;                                              \
 267         fieldnr++;
 268
 269         BCH_INODE_FIELDS_v3()
 270 #undef  x
 271
 272         /* XXX: signal if there were more fields than expected? */
 273         return 0;
 274 }
 275
 276 static noinline int bch2_inode_unpack_slowpath(struct bkey_s_c k,
 277                                                struct bch_inode_unpacked *unpacked)
 278 {
 279         memset(unpacked, 0, sizeof(*unpacked));
 280
 281         switch (k.k->type) {
 282         case KEY_TYPE_inode: {
 283                 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
 284
 285                 unpacked->bi_inum       = inode.k->p.offset;
 286                 unpacked->bi_journal_seq= 0;
 287                 unpacked->bi_hash_seed  = inode.v->bi_hash_seed;
 288                 unpacked->bi_flags      = le32_to_cpu(inode.v->bi_flags);
 289                 unpacked->bi_mode       = le16_to_cpu(inode.v->bi_mode);
 290
 291                 if (INODE_NEW_VARINT(inode.v)) {
 292                         return bch2_inode_unpack_v2(unpacked, inode.v->fields,
 293                                                     bkey_val_end(inode),
 294                                                     INODE_NR_FIELDS(inode.v));
 295                 } else {
 296                         return bch2_inode_unpack_v1(inode, unpacked);
 297                 }
 298                 break;
 299         }
 300         case KEY_TYPE_inode_v2: {
 301                 struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
 302
 303                 unpacked->bi_inum       = inode.k->p.offset;
 304                 unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
 305                 unpacked->bi_hash_seed  = inode.v->bi_hash_seed;
 306                 unpacked->bi_flags      = le64_to_cpu(inode.v->bi_flags);
 307                 unpacked->bi_mode       = le16_to_cpu(inode.v->bi_mode);
 308
 309                 return bch2_inode_unpack_v2(unpacked, inode.v->fields,
 310                                             bkey_val_end(inode),
 311                                             INODEv2_NR_FIELDS(inode.v));
 312         }
 313         default:
 314                 BUG();
 315         }
 316 }
 317
 318 int bch2_inode_unpack(struct bkey_s_c k,
 319                       struct bch_inode_unpacked *unpacked)
 320 {
 321         if (likely(k.k->type == KEY_TYPE_inode_v3))
 322                 return bch2_inode_unpack_v3(k, unpacked);
 323         return bch2_inode_unpack_slowpath(k, unpacked);
 324 }
 325
 326 static int bch2_inode_peek_nowarn(struct btree_trans *trans,
 327                     struct btree_iter *iter,
 328                     struct bch_inode_unpacked *inode,
 329                     subvol_inum inum, unsigned flags)
 330 {
 331         struct bkey_s_c k;
 332         u32 snapshot;
 333         int ret;
 334
 335         ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
 336         if (ret)
 337                 return ret;
 338
 339         k = bch2_bkey_get_iter(trans, iter, BTREE_ID_inodes,
 340                                SPOS(0, inum.inum, snapshot),
 341                                flags|BTREE_ITER_CACHED);
 342         ret = bkey_err(k);
 343         if (ret)
 344                 return ret;
 345
 346         ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
 347         if (ret)
 348                 goto err;
 349
 350         ret = bch2_inode_unpack(k, inode);
 351         if (ret)
 352                 goto err;
 353
 354         return 0;
 355 err:
 356         bch2_trans_iter_exit(trans, iter);
 357         return ret;
 358 }
 359
 360 int bch2_inode_peek(struct btree_trans *trans,
 361                     struct btree_iter *iter,
 362                     struct bch_inode_unpacked *inode,
 363                     subvol_inum inum, unsigned flags)
 364 {
 365         int ret = bch2_inode_peek_nowarn(trans, iter, inode, inum, flags);
 366         bch_err_msg(trans->c, ret, "looking up inum %u:%llu:", inum.subvol, inum.inum);
 367         return ret;
 368 }
 369
 370 int bch2_inode_write(struct btree_trans *trans,
 371                      struct btree_iter *iter,
 372                      struct bch_inode_unpacked *inode)
 373 {
 374         struct bkey_inode_buf *inode_p;
 375
 376         inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
 377         if (IS_ERR(inode_p))
 378                 return PTR_ERR(inode_p);
 379
 380         bch2_inode_pack_inlined(inode_p, inode);
 381         inode_p->inode.k.p.snapshot = iter->snapshot;
 382         return bch2_trans_update(trans, iter, &inode_p->inode.k_i, 0);
 383 }
 384
 385 struct bkey_i *bch2_inode_to_v3(struct btree_trans *trans, struct bkey_i *k)
 386 {
 387         struct bch_inode_unpacked u;
 388         struct bkey_inode_buf *inode_p;
 389         int ret;
 390
 391         if (!bkey_is_inode(&k->k))
 392                 return ERR_PTR(-ENOENT);
 393
 394         inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
 395         if (IS_ERR(inode_p))
 396                 return ERR_CAST(inode_p);
 397
 398         ret = bch2_inode_unpack(bkey_i_to_s_c(k), &u);
 399         if (ret)
 400                 return ERR_PTR(ret);
 401
 402         bch2_inode_pack(inode_p, &u);
 403         return &inode_p->inode.k_i;
 404 }
 405
 406 static int __bch2_inode_invalid(struct bch_fs *c, struct bkey_s_c k, struct printbuf *err)
 407 {
 408         struct bch_inode_unpacked unpacked;
 409         int ret = 0;
 410
 411         bkey_fsck_err_on(k.k->p.inode, c, err,
 412                          inode_pos_inode_nonzero,
 413                          "nonzero k.p.inode");
 414
 415         bkey_fsck_err_on(k.k->p.offset < BLOCKDEV_INODE_MAX, c, err,
 416                          inode_pos_blockdev_range,
 417                          "fs inode in blockdev range");
 418
 419         bkey_fsck_err_on(bch2_inode_unpack(k, &unpacked), c, err,
 420                          inode_unpack_error,
 421                          "invalid variable length fields");
 422
 423         bkey_fsck_err_on(unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1, c, err,
 424                          inode_checksum_type_invalid,
 425                          "invalid data checksum type (%u >= %u",
 426                          unpacked.bi_data_checksum, BCH_CSUM_OPT_NR + 1);
 427
 428         bkey_fsck_err_on(unpacked.bi_compression &&
 429                          !bch2_compression_opt_valid(unpacked.bi_compression - 1), c, err,
 430                          inode_compression_type_invalid,
 431                          "invalid compression opt %u", unpacked.bi_compression - 1);
 432
 433         bkey_fsck_err_on((unpacked.bi_flags & BCH_INODE_unlinked) &&
 434                          unpacked.bi_nlink != 0, c, err,
 435                          inode_unlinked_but_nlink_nonzero,
 436                          "flagged as unlinked but bi_nlink != 0");
 437
 438         bkey_fsck_err_on(unpacked.bi_subvol && !S_ISDIR(unpacked.bi_mode), c, err,
 439                          inode_subvol_root_but_not_dir,
 440                          "subvolume root but not a directory");
 441 fsck_err:
 442         return ret;
 443 }
 444
 445 int bch2_inode_invalid(struct bch_fs *c, struct bkey_s_c k,
 446                        enum bkey_invalid_flags flags,
 447                        struct printbuf *err)
 448 {
 449         struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
 450         int ret = 0;
 451
 452         bkey_fsck_err_on(INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR, c, err,
 453                          inode_str_hash_invalid,
 454                          "invalid str hash type (%llu >= %u)",
 455                          INODE_STR_HASH(inode.v), BCH_STR_HASH_NR);
 456
 457         ret = __bch2_inode_invalid(c, k, err);
 458 fsck_err:
 459         return ret;
 460 }
 461
 462 int bch2_inode_v2_invalid(struct bch_fs *c, struct bkey_s_c k,
 463                           enum bkey_invalid_flags flags,
 464                           struct printbuf *err)
 465 {
 466         struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
 467         int ret = 0;
 468
 469         bkey_fsck_err_on(INODEv2_STR_HASH(inode.v) >= BCH_STR_HASH_NR, c, err,
 470                          inode_str_hash_invalid,
 471                          "invalid str hash type (%llu >= %u)",
 472                          INODEv2_STR_HASH(inode.v), BCH_STR_HASH_NR);
 473
 474         ret = __bch2_inode_invalid(c, k, err);
 475 fsck_err:
 476         return ret;
 477 }
 478
 479 int bch2_inode_v3_invalid(struct bch_fs *c, struct bkey_s_c k,
 480                           enum bkey_invalid_flags flags,
 481                           struct printbuf *err)
 482 {
 483         struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
 484         int ret = 0;
 485
 486         bkey_fsck_err_on(INODEv3_FIELDS_START(inode.v) < INODEv3_FIELDS_START_INITIAL ||
 487                          INODEv3_FIELDS_START(inode.v) > bkey_val_u64s(inode.k), c, err,
 488                          inode_v3_fields_start_bad,
 489                          "invalid fields_start (got %llu, min %u max %zu)",
 490                          INODEv3_FIELDS_START(inode.v),
 491                          INODEv3_FIELDS_START_INITIAL,
 492                          bkey_val_u64s(inode.k));
 493
 494         bkey_fsck_err_on(INODEv3_STR_HASH(inode.v) >= BCH_STR_HASH_NR, c, err,
 495                          inode_str_hash_invalid,
 496                          "invalid str hash type (%llu >= %u)",
 497                          INODEv3_STR_HASH(inode.v), BCH_STR_HASH_NR);
 498
 499         ret = __bch2_inode_invalid(c, k, err);
 500 fsck_err:
 501         return ret;
 502 }
 503
 504 static void __bch2_inode_unpacked_to_text(struct printbuf *out,
 505                                           struct bch_inode_unpacked *inode)
 506 {
 507         prt_printf(out, "mode=%o ", inode->bi_mode);
 508
 509         prt_str(out, "flags=");
 510         prt_bitflags(out, bch2_inode_flag_strs, inode->bi_flags & ((1U << 20) - 1));
 511         prt_printf(out, " (%x)", inode->bi_flags);
 512
 513         prt_printf(out, " journal_seq=%llu bi_size=%llu bi_sectors=%llu bi_version=%llu",
 514                inode->bi_journal_seq,
 515                inode->bi_size,
 516                inode->bi_sectors,
 517                inode->bi_version);
 518
 519 #define x(_name, _bits)                                         \
 520         prt_printf(out, " "#_name "=%llu", (u64) inode->_name);
 521         BCH_INODE_FIELDS_v3()
 522 #undef  x
 523 }
 524
 525 void bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked *inode)
 526 {
 527         prt_printf(out, "inum: %llu ", inode->bi_inum);
 528         __bch2_inode_unpacked_to_text(out, inode);
 529 }
 530
 531 void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
 532 {
 533         struct bch_inode_unpacked inode;
 534
 535         if (bch2_inode_unpack(k, &inode)) {
 536                 prt_printf(out, "(unpack error)");
 537                 return;
 538         }
 539
 540         __bch2_inode_unpacked_to_text(out, &inode);
 541 }
 542
 543 static inline u64 bkey_inode_flags(struct bkey_s_c k)
 544 {
 545         switch (k.k->type) {
 546         case KEY_TYPE_inode:
 547                 return le32_to_cpu(bkey_s_c_to_inode(k).v->bi_flags);
 548         case KEY_TYPE_inode_v2:
 549                 return le64_to_cpu(bkey_s_c_to_inode_v2(k).v->bi_flags);
 550         case KEY_TYPE_inode_v3:
 551                 return le64_to_cpu(bkey_s_c_to_inode_v3(k).v->bi_flags);
 552         default:
 553                 return 0;
 554         }
 555 }
 556
 557 static inline bool bkey_is_deleted_inode(struct bkey_s_c k)
 558 {
 559         return bkey_inode_flags(k) & BCH_INODE_unlinked;
 560 }
 561
 562 int bch2_trans_mark_inode(struct btree_trans *trans,
 563                           enum btree_id btree_id, unsigned level,
 564                           struct bkey_s_c old,
 565                           struct bkey_i *new,
 566                           unsigned flags)
 567 {
 568         int nr = bkey_is_inode(&new->k) - bkey_is_inode(old.k);
 569         bool old_deleted = bkey_is_deleted_inode(old);
 570         bool new_deleted = bkey_is_deleted_inode(bkey_i_to_s_c(new));
 571
 572         if (nr) {
 573                 int ret = bch2_replicas_deltas_realloc(trans, 0);
 574                 struct replicas_delta_list *d = trans->fs_usage_deltas;
 575
 576                 if (ret)
 577                         return ret;
 578
 579                 d->nr_inodes += nr;
 580         }
 581
 582         if (old_deleted != new_deleted) {
 583                 int ret = bch2_btree_bit_mod(trans, BTREE_ID_deleted_inodes, new->k.p, new_deleted);
 584                 if (ret)
 585                         return ret;
 586         }
 587
 588         return 0;
 589 }
 590
 591 int bch2_mark_inode(struct btree_trans *trans,
 592                     enum btree_id btree_id, unsigned level,
 593                     struct bkey_s_c old, struct bkey_s_c new,
 594                     unsigned flags)
 595 {
 596         struct bch_fs *c = trans->c;
 597         struct bch_fs_usage *fs_usage;
 598         u64 journal_seq = trans->journal_res.seq;
 599
 600         if (flags & BTREE_TRIGGER_INSERT) {
 601                 struct bch_inode_v3 *v = (struct bch_inode_v3 *) new.v;
 602
 603                 BUG_ON(!journal_seq);
 604                 BUG_ON(new.k->type != KEY_TYPE_inode_v3);
 605
 606                 v->bi_journal_seq = cpu_to_le64(journal_seq);
 607         }
 608
 609         if (flags & BTREE_TRIGGER_GC) {
 610                 percpu_down_read(&c->mark_lock);
 611                 preempt_disable();
 612
 613                 fs_usage = fs_usage_ptr(c, journal_seq, flags & BTREE_TRIGGER_GC);
 614                 fs_usage->nr_inodes += bkey_is_inode(new.k);
 615                 fs_usage->nr_inodes -= bkey_is_inode(old.k);
 616
 617                 preempt_enable();
 618                 percpu_up_read(&c->mark_lock);
 619         }
 620         return 0;
 621 }
 622
 623 int bch2_inode_generation_invalid(struct bch_fs *c, struct bkey_s_c k,
 624                                   enum bkey_invalid_flags flags,
 625                                   struct printbuf *err)
 626 {
 627         int ret = 0;
 628
 629         bkey_fsck_err_on(k.k->p.inode, c, err,
 630                          inode_pos_inode_nonzero,
 631                          "nonzero k.p.inode");
 632 fsck_err:
 633         return ret;
 634 }
 635
 636 void bch2_inode_generation_to_text(struct printbuf *out, struct bch_fs *c,
 637                                    struct bkey_s_c k)
 638 {
 639         struct bkey_s_c_inode_generation gen = bkey_s_c_to_inode_generation(k);
 640
 641         prt_printf(out, "generation: %u", le32_to_cpu(gen.v->bi_generation));
 642 }
 643
 644 void bch2_inode_init_early(struct bch_fs *c,
 645                            struct bch_inode_unpacked *inode_u)
 646 {
 647         enum bch_str_hash_type str_hash =
 648                 bch2_str_hash_opt_to_type(c, c->opts.str_hash);
 649
 650         memset(inode_u, 0, sizeof(*inode_u));
 651
 652         /* ick */
 653         inode_u->bi_flags |= str_hash << INODE_STR_HASH_OFFSET;
 654         get_random_bytes(&inode_u->bi_hash_seed,
 655                          sizeof(inode_u->bi_hash_seed));
 656 }
 657
 658 void bch2_inode_init_late(struct bch_inode_unpacked *inode_u, u64 now,
 659                           uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
 660                           struct bch_inode_unpacked *parent)
 661 {
 662         inode_u->bi_mode        = mode;
 663         inode_u->bi_uid         = uid;
 664         inode_u->bi_gid         = gid;
 665         inode_u->bi_dev         = rdev;
 666         inode_u->bi_atime       = now;
 667         inode_u->bi_mtime       = now;
 668         inode_u->bi_ctime       = now;
 669         inode_u->bi_otime       = now;
 670
 671         if (parent && parent->bi_mode & S_ISGID) {
 672                 inode_u->bi_gid = parent->bi_gid;
 673                 if (S_ISDIR(mode))
 674                         inode_u->bi_mode |= S_ISGID;
 675         }
 676
 677         if (parent) {
 678 #define x(_name, ...)   inode_u->bi_##_name = parent->bi_##_name;
 679                 BCH_INODE_OPTS()
 680 #undef x
 681         }
 682 }
 683
 684 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
 685                      uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
 686                      struct bch_inode_unpacked *parent)
 687 {
 688         bch2_inode_init_early(c, inode_u);
 689         bch2_inode_init_late(inode_u, bch2_current_time(c),
 690                              uid, gid, mode, rdev, parent);
 691 }
 692
 693 static inline u32 bkey_generation(struct bkey_s_c k)
 694 {
 695         switch (k.k->type) {
 696         case KEY_TYPE_inode:
 697         case KEY_TYPE_inode_v2:
 698                 BUG();
 699         case KEY_TYPE_inode_generation:
 700                 return le32_to_cpu(bkey_s_c_to_inode_generation(k).v->bi_generation);
 701         default:
 702                 return 0;
 703         }
 704 }
 705
 706 /*
 707  * This just finds an empty slot:
 708  */
 709 int bch2_inode_create(struct btree_trans *trans,
 710                       struct btree_iter *iter,
 711                       struct bch_inode_unpacked *inode_u,
 712                       u32 snapshot, u64 cpu)
 713 {
 714         struct bch_fs *c = trans->c;
 715         struct bkey_s_c k;
 716         u64 min, max, start, pos, *hint;
 717         int ret = 0;
 718         unsigned bits = (c->opts.inodes_32bit ? 31 : 63);
 719
 720         if (c->opts.shard_inode_numbers) {
 721                 bits -= c->inode_shard_bits;
 722
 723                 min = (cpu << bits);
 724                 max = (cpu << bits) | ~(ULLONG_MAX << bits);
 725
 726                 min = max_t(u64, min, BLOCKDEV_INODE_MAX);
 727                 hint = c->unused_inode_hints + cpu;
 728         } else {
 729                 min = BLOCKDEV_INODE_MAX;
 730                 max = ~(ULLONG_MAX << bits);
 731                 hint = c->unused_inode_hints;
 732         }
 733
 734         start = READ_ONCE(*hint);
 735
 736         if (start >= max || start < min)
 737                 start = min;
 738
 739         pos = start;
 740         bch2_trans_iter_init(trans, iter, BTREE_ID_inodes, POS(0, pos),
 741                              BTREE_ITER_ALL_SNAPSHOTS|
 742                              BTREE_ITER_INTENT);
 743 again:
 744         while ((k = bch2_btree_iter_peek(iter)).k &&
 745                !(ret = bkey_err(k)) &&
 746                bkey_lt(k.k->p, POS(0, max))) {
 747                 if (pos < iter->pos.offset)
 748                         goto found_slot;
 749
 750                 /*
 751                  * We don't need to iterate over keys in every snapshot once
 752                  * we've found just one:
 753                  */
 754                 pos = iter->pos.offset + 1;
 755                 bch2_btree_iter_set_pos(iter, POS(0, pos));
 756         }
 757
 758         if (!ret && pos < max)
 759                 goto found_slot;
 760
 761         if (!ret && start == min)
 762                 ret = -BCH_ERR_ENOSPC_inode_create;
 763
 764         if (ret) {
 765                 bch2_trans_iter_exit(trans, iter);
 766                 return ret;
 767         }
 768
 769         /* Retry from start */
 770         pos = start = min;
 771         bch2_btree_iter_set_pos(iter, POS(0, pos));
 772         goto again;
 773 found_slot:
 774         bch2_btree_iter_set_pos(iter, SPOS(0, pos, snapshot));
 775         k = bch2_btree_iter_peek_slot(iter);
 776         ret = bkey_err(k);
 777         if (ret) {
 778                 bch2_trans_iter_exit(trans, iter);
 779                 return ret;
 780         }
 781
 782         *hint                   = k.k->p.offset;
 783         inode_u->bi_inum        = k.k->p.offset;
 784         inode_u->bi_generation  = bkey_generation(k);
 785         return 0;
 786 }
 787
 788 static int bch2_inode_delete_keys(struct btree_trans *trans,
 789                                   subvol_inum inum, enum btree_id id)
 790 {
 791         struct btree_iter iter;
 792         struct bkey_s_c k;
 793         struct bkey_i delete;
 794         struct bpos end = POS(inum.inum, U64_MAX);
 795         u32 snapshot;
 796         int ret = 0;
 797
 798         /*
 799          * We're never going to be deleting partial extents, no need to use an
 800          * extent iterator:
 801          */
 802         bch2_trans_iter_init(trans, &iter, id, POS(inum.inum, 0),
 803                              BTREE_ITER_INTENT);
 804
 805         while (1) {
 806                 bch2_trans_begin(trans);
 807
 808                 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
 809                 if (ret)
 810                         goto err;
 811
 812                 bch2_btree_iter_set_snapshot(&iter, snapshot);
 813
 814                 k = bch2_btree_iter_peek_upto(&iter, end);
 815                 ret = bkey_err(k);
 816                 if (ret)
 817                         goto err;
 818
 819                 if (!k.k)
 820                         break;
 821
 822                 bkey_init(&delete.k);
 823                 delete.k.p = iter.pos;
 824
 825                 if (iter.flags & BTREE_ITER_IS_EXTENTS)
 826                         bch2_key_resize(&delete.k,
 827                                         bpos_min(end, k.k->p).offset -
 828                                         iter.pos.offset);
 829
 830                 ret = bch2_trans_update(trans, &iter, &delete, 0) ?:
 831                       bch2_trans_commit(trans, NULL, NULL,
 832                                         BTREE_INSERT_NOFAIL);
 833 err:
 834                 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
 835                         break;
 836         }
 837
 838         bch2_trans_iter_exit(trans, &iter);
 839         return ret;
 840 }
 841
 842 int bch2_inode_rm(struct bch_fs *c, subvol_inum inum)
 843 {
 844         struct btree_trans *trans = bch2_trans_get(c);
 845         struct btree_iter iter = { NULL };
 846         struct bkey_i_inode_generation delete;
 847         struct bch_inode_unpacked inode_u;
 848         struct bkey_s_c k;
 849         u32 snapshot;
 850         int ret;
 851
 852         /*
 853          * If this was a directory, there shouldn't be any real dirents left -
 854          * but there could be whiteouts (from hash collisions) that we should
 855          * delete:
 856          *
 857          * XXX: the dirent could ideally would delete whiteouts when they're no
 858          * longer needed
 859          */
 860         ret   = bch2_inode_delete_keys(trans, inum, BTREE_ID_extents) ?:
 861                 bch2_inode_delete_keys(trans, inum, BTREE_ID_xattrs) ?:
 862                 bch2_inode_delete_keys(trans, inum, BTREE_ID_dirents);
 863         if (ret)
 864                 goto err;
 865 retry:
 866         bch2_trans_begin(trans);
 867
 868         ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
 869         if (ret)
 870                 goto err;
 871
 872         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
 873                                SPOS(0, inum.inum, snapshot),
 874                                BTREE_ITER_INTENT|BTREE_ITER_CACHED);
 875         ret = bkey_err(k);
 876         if (ret)
 877                 goto err;
 878
 879         if (!bkey_is_inode(k.k)) {
 880                 bch2_fs_inconsistent(c,
 881                                      "inode %llu:%u not found when deleting",
 882                                      inum.inum, snapshot);
 883                 ret = -EIO;
 884                 goto err;
 885         }
 886
 887         bch2_inode_unpack(k, &inode_u);
 888
 889         bkey_inode_generation_init(&delete.k_i);
 890         delete.k.p = iter.pos;
 891         delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
 892
 893         ret   = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
 894                 bch2_trans_commit(trans, NULL, NULL,
 895                                 BTREE_INSERT_NOFAIL);
 896 err:
 897         bch2_trans_iter_exit(trans, &iter);
 898         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 899                 goto retry;
 900
 901         bch2_trans_put(trans);
 902         return ret;
 903 }
 904
 905 int bch2_inode_find_by_inum_nowarn_trans(struct btree_trans *trans,
 906                                   subvol_inum inum,
 907                                   struct bch_inode_unpacked *inode)
 908 {
 909         struct btree_iter iter;
 910         int ret;
 911
 912         ret = bch2_inode_peek_nowarn(trans, &iter, inode, inum, 0);
 913         if (!ret)
 914                 bch2_trans_iter_exit(trans, &iter);
 915         return ret;
 916 }
 917
 918 int bch2_inode_find_by_inum_trans(struct btree_trans *trans,
 919                                   subvol_inum inum,
 920                                   struct bch_inode_unpacked *inode)
 921 {
 922         struct btree_iter iter;
 923         int ret;
 924
 925         ret = bch2_inode_peek(trans, &iter, inode, inum, 0);
 926         if (!ret)
 927                 bch2_trans_iter_exit(trans, &iter);
 928         return ret;
 929 }
 930
 931 int bch2_inode_find_by_inum(struct bch_fs *c, subvol_inum inum,
 932                             struct bch_inode_unpacked *inode)
 933 {
 934         return bch2_trans_do(c, NULL, NULL, 0,
 935                 bch2_inode_find_by_inum_trans(trans, inum, inode));
 936 }
 937
 938 int bch2_inode_nlink_inc(struct bch_inode_unpacked *bi)
 939 {
 940         if (bi->bi_flags & BCH_INODE_unlinked)
 941                 bi->bi_flags &= ~BCH_INODE_unlinked;
 942         else {
 943                 if (bi->bi_nlink == U32_MAX)
 944                         return -EINVAL;
 945
 946                 bi->bi_nlink++;
 947         }
 948
 949         return 0;
 950 }
 951
 952 void bch2_inode_nlink_dec(struct btree_trans *trans, struct bch_inode_unpacked *bi)
 953 {
 954         if (bi->bi_nlink && (bi->bi_flags & BCH_INODE_unlinked)) {
 955                 bch2_trans_inconsistent(trans, "inode %llu unlinked but link count nonzero",
 956                                         bi->bi_inum);
 957                 return;
 958         }
 959
 960         if (bi->bi_flags & BCH_INODE_unlinked) {
 961                 bch2_trans_inconsistent(trans, "inode %llu link count underflow", bi->bi_inum);
 962                 return;
 963         }
 964
 965         if (bi->bi_nlink)
 966                 bi->bi_nlink--;
 967         else
 968                 bi->bi_flags |= BCH_INODE_unlinked;
 969 }
 970
 971 struct bch_opts bch2_inode_opts_to_opts(struct bch_inode_unpacked *inode)
 972 {
 973         struct bch_opts ret = { 0 };
 974 #define x(_name, _bits)                                                 \
 975         if (inode->bi_##_name)                                          \
 976                 opt_set(ret, _name, inode->bi_##_name - 1);
 977         BCH_INODE_OPTS()
 978 #undef x
 979         return ret;
 980 }
 981
 982 void bch2_inode_opts_get(struct bch_io_opts *opts, struct bch_fs *c,
 983                          struct bch_inode_unpacked *inode)
 984 {
 985 #define x(_name, _bits)         opts->_name = inode_opt_get(c, inode, _name);
 986         BCH_INODE_OPTS()
 987 #undef x
 988
 989         if (opts->nocow)
 990                 opts->compression = opts->background_compression = opts->data_checksum = opts->erasure_code = 0;
 991 }
 992
 993 int bch2_inum_opts_get(struct btree_trans *trans, subvol_inum inum, struct bch_io_opts *opts)
 994 {
 995         struct bch_inode_unpacked inode;
 996         int ret = lockrestart_do(trans, bch2_inode_find_by_inum_trans(trans, inum, &inode));
 997
 998         if (ret)
 999                 return ret;
1000
1001         bch2_inode_opts_get(opts, trans->c, &inode);
1002         return 0;
1003 }
1004
1005 int bch2_inode_rm_snapshot(struct btree_trans *trans, u64 inum, u32 snapshot)
1006 {
1007         struct bch_fs *c = trans->c;
1008         struct btree_iter iter = { NULL };
1009         struct bkey_i_inode_generation delete;
1010         struct bch_inode_unpacked inode_u;
1011         struct bkey_s_c k;
1012         int ret;
1013
1014         do {
1015                 ret   = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
1016                                                       SPOS(inum, 0, snapshot),
1017                                                       SPOS(inum, U64_MAX, snapshot),
1018                                                       0, NULL) ?:
1019                         bch2_btree_delete_range_trans(trans, BTREE_ID_dirents,
1020                                                       SPOS(inum, 0, snapshot),
1021                                                       SPOS(inum, U64_MAX, snapshot),
1022                                                       0, NULL) ?:
1023                         bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs,
1024                                                       SPOS(inum, 0, snapshot),
1025                                                       SPOS(inum, U64_MAX, snapshot),
1026                                                       0, NULL);
1027         } while (ret == -BCH_ERR_transaction_restart_nested);
1028         if (ret)
1029                 goto err;
1030 retry:
1031         bch2_trans_begin(trans);
1032
1033         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
1034                                SPOS(0, inum, snapshot), BTREE_ITER_INTENT);
1035         ret = bkey_err(k);
1036         if (ret)
1037                 goto err;
1038
1039         if (!bkey_is_inode(k.k)) {
1040                 bch2_fs_inconsistent(c,
1041                                      "inode %llu:%u not found when deleting",
1042                                      inum, snapshot);
1043                 ret = -EIO;
1044                 goto err;
1045         }
1046
1047         bch2_inode_unpack(k, &inode_u);
1048
1049         /* Subvolume root? */
1050         if (inode_u.bi_subvol)
1051                 bch_warn(c, "deleting inode %llu marked as unlinked, but also a subvolume root!?", inode_u.bi_inum);
1052
1053         bkey_inode_generation_init(&delete.k_i);
1054         delete.k.p = iter.pos;
1055         delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
1056
1057         ret   = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
1058                 bch2_trans_commit(trans, NULL, NULL,
1059                                 BTREE_INSERT_NOFAIL);
1060 err:
1061         bch2_trans_iter_exit(trans, &iter);
1062         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1063                 goto retry;
1064
1065         return ret ?: -BCH_ERR_transaction_restart_nested;
1066 }
1067
1068 static int may_delete_deleted_inode(struct btree_trans *trans, struct bpos pos)
1069 {
1070         struct bch_fs *c = trans->c;
1071         struct btree_iter iter;
1072         struct bkey_s_c k;
1073         struct bch_inode_unpacked inode;
1074         int ret;
1075
1076         if (bch2_snapshot_is_internal_node(c, pos.snapshot))
1077                 return 0;
1078
1079         if (!fsck_err_on(c->sb.clean, c,
1080                          deleted_inode_but_clean,
1081                          "filesystem marked as clean but have deleted inode %llu:%u",
1082                          pos.offset, pos.snapshot))
1083                 return 0;
1084
1085         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes, pos, BTREE_ITER_CACHED);
1086         ret = bkey_err(k);
1087         if (ret)
1088                 return ret;
1089
1090         ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
1091         if (fsck_err_on(!bkey_is_inode(k.k), c,
1092                         deleted_inode_missing,
1093                         "nonexistent inode %llu:%u in deleted_inodes btree",
1094                         pos.offset, pos.snapshot))
1095                 goto delete;
1096
1097         ret = bch2_inode_unpack(k, &inode);
1098         if (ret)
1099                 goto err;
1100
1101         if (fsck_err_on(S_ISDIR(inode.bi_mode), c,
1102                         deleted_inode_is_dir,
1103                         "directory %llu:%u in deleted_inodes btree",
1104                         pos.offset, pos.snapshot))
1105                 goto delete;
1106
1107         if (fsck_err_on(!(inode.bi_flags & BCH_INODE_unlinked), c,
1108                         deleted_inode_not_unlinked,
1109                         "non-deleted inode %llu:%u in deleted_inodes btree",
1110                         pos.offset, pos.snapshot))
1111                 goto delete;
1112
1113         return 1;
1114 err:
1115 fsck_err:
1116         return ret;
1117 delete:
1118         return bch2_btree_bit_mod(trans, BTREE_ID_deleted_inodes, pos, false);
1119 }
1120
1121 int bch2_delete_dead_inodes(struct bch_fs *c)
1122 {
1123         struct btree_trans *trans = bch2_trans_get(c);
1124         struct btree_iter iter;
1125         struct bkey_s_c k;
1126         int ret;
1127
1128         ret = bch2_btree_write_buffer_flush_sync(trans);
1129         if (ret)
1130                 goto err;
1131
1132         /*
1133          * Weird transaction restart handling here because on successful delete,
1134          * bch2_inode_rm_snapshot() will return a nested transaction restart,
1135          * but we can't retry because the btree write buffer won't have been
1136          * flushed and we'd spin:
1137          */
1138         for_each_btree_key(trans, iter, BTREE_ID_deleted_inodes, POS_MIN,
1139                            BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
1140                 ret = lockrestart_do(trans, may_delete_deleted_inode(trans, k.k->p));
1141                 if (ret < 0)
1142                         break;
1143
1144                 if (ret) {
1145                         if (!test_bit(BCH_FS_RW, &c->flags)) {
1146                                 bch2_trans_unlock(trans);
1147                                 bch2_fs_lazy_rw(c);
1148                         }
1149
1150                         ret = bch2_inode_rm_snapshot(trans, k.k->p.offset, k.k->p.snapshot);
1151                         if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
1152                                 break;
1153                 }
1154         }
1155         bch2_trans_iter_exit(trans, &iter);
1156 err:
1157         bch2_trans_put(trans);
1158
1159         return ret;
1160 }