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