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 const char * const bch2_inode_opts[] = {
  24 #define x(name, ...)    #name,
  25         BCH_INODE_OPTS()
  26 #undef  x
  27         NULL,
  28 };
  29
  30 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
  31
  32 static int inode_decode_field(const u8 *in, const u8 *end,
  33                               u64 out[2], unsigned *out_bits)
  34 {
  35         __be64 be[2] = { 0, 0 };
  36         unsigned bytes, shift;
  37         u8 *p;
  38
  39         if (in >= end)
  40                 return -1;
  41
  42         if (!*in)
  43                 return -1;
  44
  45         /*
  46          * position of highest set bit indicates number of bytes:
  47          * shift = number of bits to remove in high byte:
  48          */
  49         shift   = 8 - __fls(*in); /* 1 <= shift <= 8 */
  50         bytes   = byte_table[shift - 1];
  51
  52         if (in + bytes > end)
  53                 return -1;
  54
  55         p = (u8 *) be + 16 - bytes;
  56         memcpy(p, in, bytes);
  57         *p ^= (1 << 8) >> shift;
  58
  59         out[0] = be64_to_cpu(be[0]);
  60         out[1] = be64_to_cpu(be[1]);
  61         *out_bits = out[0] ? 64 + fls64(out[0]) : fls64(out[1]);
  62
  63         return bytes;
  64 }
  65
  66 static inline void bch2_inode_pack_inlined(struct bkey_inode_buf *packed,
  67                                            const struct bch_inode_unpacked *inode)
  68 {
  69         struct bkey_i_inode_v3 *k = &packed->inode;
  70         u8 *out = k->v.fields;
  71         u8 *end = (void *) &packed[1];
  72         u8 *last_nonzero_field = out;
  73         unsigned nr_fields = 0, last_nonzero_fieldnr = 0;
  74         unsigned bytes;
  75         int ret;
  76
  77         bkey_inode_v3_init(&packed->inode.k_i);
  78         packed->inode.k.p.offset        = inode->bi_inum;
  79         packed->inode.v.bi_journal_seq  = cpu_to_le64(inode->bi_journal_seq);
  80         packed->inode.v.bi_hash_seed    = inode->bi_hash_seed;
  81         packed->inode.v.bi_flags        = cpu_to_le64(inode->bi_flags);
  82         packed->inode.v.bi_sectors      = cpu_to_le64(inode->bi_sectors);
  83         packed->inode.v.bi_size         = cpu_to_le64(inode->bi_size);
  84         packed->inode.v.bi_version      = cpu_to_le64(inode->bi_version);
  85         SET_INODEv3_MODE(&packed->inode.v, inode->bi_mode);
  86         SET_INODEv3_FIELDS_START(&packed->inode.v, INODEv3_FIELDS_START_CUR);
  87
  88
  89 #define x(_name, _bits)                                                 \
  90         nr_fields++;                                                    \
  91                                                                         \
  92         if (inode->_name) {                                             \
  93                 ret = bch2_varint_encode_fast(out, inode->_name);       \
  94                 out += ret;                                             \
  95                                                                         \
  96                 if (_bits > 64)                                         \
  97                         *out++ = 0;                                     \
  98                                                                         \
  99                 last_nonzero_field = out;                               \
 100                 last_nonzero_fieldnr = nr_fields;                       \
 101         } else {                                                        \
 102                 *out++ = 0;                                             \
 103                                                                         \
 104                 if (_bits > 64)                                         \
 105                         *out++ = 0;                                     \
 106         }
 107
 108         BCH_INODE_FIELDS_v3()
 109 #undef  x
 110         BUG_ON(out > end);
 111
 112         out = last_nonzero_field;
 113         nr_fields = last_nonzero_fieldnr;
 114
 115         bytes = out - (u8 *) &packed->inode.v;
 116         set_bkey_val_bytes(&packed->inode.k, bytes);
 117         memset_u64s_tail(&packed->inode.v, 0, bytes);
 118
 119         SET_INODEv3_NR_FIELDS(&k->v, nr_fields);
 120
 121         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
 122                 struct bch_inode_unpacked unpacked;
 123
 124                 ret = bch2_inode_unpack(bkey_i_to_s_c(&packed->inode.k_i), &unpacked);
 125                 BUG_ON(ret);
 126                 BUG_ON(unpacked.bi_inum         != inode->bi_inum);
 127                 BUG_ON(unpacked.bi_hash_seed    != inode->bi_hash_seed);
 128                 BUG_ON(unpacked.bi_sectors      != inode->bi_sectors);
 129                 BUG_ON(unpacked.bi_size         != inode->bi_size);
 130                 BUG_ON(unpacked.bi_version      != inode->bi_version);
 131                 BUG_ON(unpacked.bi_mode         != inode->bi_mode);
 132
 133 #define x(_name, _bits) if (unpacked._name != inode->_name)             \
 134                         panic("unpacked %llu should be %llu",           \
 135                               (u64) unpacked._name, (u64) inode->_name);
 136                 BCH_INODE_FIELDS_v3()
 137 #undef  x
 138         }
 139 }
 140
 141 void bch2_inode_pack(struct bkey_inode_buf *packed,
 142                      const struct bch_inode_unpacked *inode)
 143 {
 144         bch2_inode_pack_inlined(packed, inode);
 145 }
 146
 147 static noinline int bch2_inode_unpack_v1(struct bkey_s_c_inode inode,
 148                                 struct bch_inode_unpacked *unpacked)
 149 {
 150         const u8 *in = inode.v->fields;
 151         const u8 *end = bkey_val_end(inode);
 152         u64 field[2];
 153         unsigned fieldnr = 0, field_bits;
 154         int ret;
 155
 156 #define x(_name, _bits)                                 \
 157         if (fieldnr++ == INODE_NR_FIELDS(inode.v)) {                    \
 158                 unsigned offset = offsetof(struct bch_inode_unpacked, _name);\
 159                 memset((void *) unpacked + offset, 0,                   \
 160                        sizeof(*unpacked) - offset);                     \
 161                 return 0;                                               \
 162         }                                                               \
 163                                                                         \
 164         ret = inode_decode_field(in, end, field, &field_bits);          \
 165         if (ret < 0)                                                    \
 166                 return ret;                                             \
 167                                                                         \
 168         if (field_bits > sizeof(unpacked->_name) * 8)                   \
 169                 return -1;                                              \
 170                                                                         \
 171         unpacked->_name = field[1];                                     \
 172         in += ret;
 173
 174         BCH_INODE_FIELDS_v2()
 175 #undef  x
 176
 177         /* XXX: signal if there were more fields than expected? */
 178         return 0;
 179 }
 180
 181 static int bch2_inode_unpack_v2(struct bch_inode_unpacked *unpacked,
 182                                 const u8 *in, const u8 *end,
 183                                 unsigned nr_fields)
 184 {
 185         unsigned fieldnr = 0;
 186         int ret;
 187         u64 v[2];
 188
 189 #define x(_name, _bits)                                                 \
 190         if (fieldnr < nr_fields) {                                      \
 191                 ret = bch2_varint_decode_fast(in, end, &v[0]);          \
 192                 if (ret < 0)                                            \
 193                         return ret;                                     \
 194                 in += ret;                                              \
 195                                                                         \
 196                 if (_bits > 64) {                                       \
 197                         ret = bch2_varint_decode_fast(in, end, &v[1]);  \
 198                         if (ret < 0)                                    \
 199                                 return ret;                             \
 200                         in += ret;                                      \
 201                 } else {                                                \
 202                         v[1] = 0;                                       \
 203                 }                                                       \
 204         } else {                                                        \
 205                 v[0] = v[1] = 0;                                        \
 206         }                                                               \
 207                                                                         \
 208         unpacked->_name = v[0];                                         \
 209         if (v[1] || v[0] != unpacked->_name)                            \
 210                 return -1;                                              \
 211         fieldnr++;
 212
 213         BCH_INODE_FIELDS_v2()
 214 #undef  x
 215
 216         /* XXX: signal if there were more fields than expected? */
 217         return 0;
 218 }
 219
 220 static int bch2_inode_unpack_v3(struct bkey_s_c k,
 221                                 struct bch_inode_unpacked *unpacked)
 222 {
 223         struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
 224         const u8 *in = inode.v->fields;
 225         const u8 *end = bkey_val_end(inode);
 226         unsigned nr_fields = INODEv3_NR_FIELDS(inode.v);
 227         unsigned fieldnr = 0;
 228         int ret;
 229         u64 v[2];
 230
 231         unpacked->bi_inum       = inode.k->p.offset;
 232         unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
 233         unpacked->bi_hash_seed  = inode.v->bi_hash_seed;
 234         unpacked->bi_flags      = le64_to_cpu(inode.v->bi_flags);
 235         unpacked->bi_sectors    = le64_to_cpu(inode.v->bi_sectors);
 236         unpacked->bi_size       = le64_to_cpu(inode.v->bi_size);
 237         unpacked->bi_version    = le64_to_cpu(inode.v->bi_version);
 238         unpacked->bi_mode       = INODEv3_MODE(inode.v);
 239
 240 #define x(_name, _bits)                                                 \
 241         if (fieldnr < nr_fields) {                                      \
 242                 ret = bch2_varint_decode_fast(in, end, &v[0]);          \
 243                 if (ret < 0)                                            \
 244                         return ret;                                     \
 245                 in += ret;                                              \
 246                                                                         \
 247                 if (_bits > 64) {                                       \
 248                         ret = bch2_varint_decode_fast(in, end, &v[1]);  \
 249                         if (ret < 0)                                    \
 250                                 return ret;                             \
 251                         in += ret;                                      \
 252                 } else {                                                \
 253                         v[1] = 0;                                       \
 254                 }                                                       \
 255         } else {                                                        \
 256                 v[0] = v[1] = 0;                                        \
 257         }                                                               \
 258                                                                         \
 259         unpacked->_name = v[0];                                         \
 260         if (v[1] || v[0] != unpacked->_name)                            \
 261                 return -1;                                              \
 262         fieldnr++;
 263
 264         BCH_INODE_FIELDS_v3()
 265 #undef  x
 266
 267         /* XXX: signal if there were more fields than expected? */
 268         return 0;
 269 }
 270
 271 static noinline int bch2_inode_unpack_slowpath(struct bkey_s_c k,
 272                                                struct bch_inode_unpacked *unpacked)
 273 {
 274         memset(unpacked, 0, sizeof(*unpacked));
 275
 276         switch (k.k->type) {
 277         case KEY_TYPE_inode: {
 278                 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
 279
 280                 unpacked->bi_inum       = inode.k->p.offset;
 281                 unpacked->bi_journal_seq= 0;
 282                 unpacked->bi_hash_seed  = inode.v->bi_hash_seed;
 283                 unpacked->bi_flags      = le32_to_cpu(inode.v->bi_flags);
 284                 unpacked->bi_mode       = le16_to_cpu(inode.v->bi_mode);
 285
 286                 if (INODE_NEW_VARINT(inode.v)) {
 287                         return bch2_inode_unpack_v2(unpacked, inode.v->fields,
 288                                                     bkey_val_end(inode),
 289                                                     INODE_NR_FIELDS(inode.v));
 290                 } else {
 291                         return bch2_inode_unpack_v1(inode, unpacked);
 292                 }
 293                 break;
 294         }
 295         case KEY_TYPE_inode_v2: {
 296                 struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
 297
 298                 unpacked->bi_inum       = inode.k->p.offset;
 299                 unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
 300                 unpacked->bi_hash_seed  = inode.v->bi_hash_seed;
 301                 unpacked->bi_flags      = le64_to_cpu(inode.v->bi_flags);
 302                 unpacked->bi_mode       = le16_to_cpu(inode.v->bi_mode);
 303
 304                 return bch2_inode_unpack_v2(unpacked, inode.v->fields,
 305                                             bkey_val_end(inode),
 306                                             INODEv2_NR_FIELDS(inode.v));
 307         }
 308         default:
 309                 BUG();
 310         }
 311 }
 312
 313 int bch2_inode_unpack(struct bkey_s_c k,
 314                       struct bch_inode_unpacked *unpacked)
 315 {
 316         if (likely(k.k->type == KEY_TYPE_inode_v3))
 317                 return bch2_inode_unpack_v3(k, unpacked);
 318         return bch2_inode_unpack_slowpath(k, unpacked);
 319 }
 320
 321 static int bch2_inode_peek_nowarn(struct btree_trans *trans,
 322                     struct btree_iter *iter,
 323                     struct bch_inode_unpacked *inode,
 324                     subvol_inum inum, unsigned flags)
 325 {
 326         struct bkey_s_c k;
 327         u32 snapshot;
 328         int ret;
 329
 330         ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
 331         if (ret)
 332                 return ret;
 333
 334         k = bch2_bkey_get_iter(trans, iter, BTREE_ID_inodes,
 335                                SPOS(0, inum.inum, snapshot),
 336                                flags|BTREE_ITER_CACHED);
 337         ret = bkey_err(k);
 338         if (ret)
 339                 return ret;
 340
 341         ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
 342         if (ret)
 343                 goto err;
 344
 345         ret = bch2_inode_unpack(k, inode);
 346         if (ret)
 347                 goto err;
 348
 349         return 0;
 350 err:
 351         bch2_trans_iter_exit(trans, iter);
 352         return ret;
 353 }
 354
 355 int bch2_inode_peek(struct btree_trans *trans,
 356                     struct btree_iter *iter,
 357                     struct bch_inode_unpacked *inode,
 358                     subvol_inum inum, unsigned flags)
 359 {
 360         int ret = bch2_inode_peek_nowarn(trans, iter, inode, inum, flags);
 361         bch_err_msg(trans->c, ret, "looking up inum %u:%llu:", inum.subvol, inum.inum);
 362         return ret;
 363 }
 364
 365 int bch2_inode_write(struct btree_trans *trans,
 366                      struct btree_iter *iter,
 367                      struct bch_inode_unpacked *inode)
 368 {
 369         struct bkey_inode_buf *inode_p;
 370
 371         inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
 372         if (IS_ERR(inode_p))
 373                 return PTR_ERR(inode_p);
 374
 375         bch2_inode_pack_inlined(inode_p, inode);
 376         inode_p->inode.k.p.snapshot = iter->snapshot;
 377         return bch2_trans_update(trans, iter, &inode_p->inode.k_i, 0);
 378 }
 379
 380 struct bkey_i *bch2_inode_to_v3(struct btree_trans *trans, struct bkey_i *k)
 381 {
 382         struct bch_inode_unpacked u;
 383         struct bkey_inode_buf *inode_p;
 384         int ret;
 385
 386         if (!bkey_is_inode(&k->k))
 387                 return ERR_PTR(-ENOENT);
 388
 389         inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
 390         if (IS_ERR(inode_p))
 391                 return ERR_CAST(inode_p);
 392
 393         ret = bch2_inode_unpack(bkey_i_to_s_c(k), &u);
 394         if (ret)
 395                 return ERR_PTR(ret);
 396
 397         bch2_inode_pack(inode_p, &u);
 398         return &inode_p->inode.k_i;
 399 }
 400
 401 static int __bch2_inode_invalid(struct bkey_s_c k, struct printbuf *err)
 402 {
 403         struct bch_inode_unpacked unpacked;
 404
 405         if (k.k->p.inode) {
 406                 prt_printf(err, "nonzero k.p.inode");
 407                 return -BCH_ERR_invalid_bkey;
 408         }
 409
 410         if (k.k->p.offset < BLOCKDEV_INODE_MAX) {
 411                 prt_printf(err, "fs inode in blockdev range");
 412                 return -BCH_ERR_invalid_bkey;
 413         }
 414
 415         if (bch2_inode_unpack(k, &unpacked)) {
 416                 prt_printf(err, "invalid variable length fields");
 417                 return -BCH_ERR_invalid_bkey;
 418         }
 419
 420         if (unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1) {
 421                 prt_printf(err, "invalid data checksum type (%u >= %u",
 422                         unpacked.bi_data_checksum, BCH_CSUM_OPT_NR + 1);
 423                 return -BCH_ERR_invalid_bkey;
 424         }
 425
 426         if (unpacked.bi_compression &&
 427             !bch2_compression_opt_valid(unpacked.bi_compression - 1)) {
 428                 prt_printf(err, "invalid compression opt %u",
 429                            unpacked.bi_compression - 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         struct bpos end = POS(inum.inum, U64_MAX);
 786         u32 snapshot;
 787         int ret = 0;
 788
 789         /*
 790          * We're never going to be deleting partial extents, no need to use an
 791          * extent iterator:
 792          */
 793         bch2_trans_iter_init(trans, &iter, id, POS(inum.inum, 0),
 794                              BTREE_ITER_INTENT);
 795
 796         while (1) {
 797                 bch2_trans_begin(trans);
 798
 799                 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
 800                 if (ret)
 801                         goto err;
 802
 803                 bch2_btree_iter_set_snapshot(&iter, snapshot);
 804
 805                 k = bch2_btree_iter_peek_upto(&iter, end);
 806                 ret = bkey_err(k);
 807                 if (ret)
 808                         goto err;
 809
 810                 if (!k.k)
 811                         break;
 812
 813                 bkey_init(&delete.k);
 814                 delete.k.p = iter.pos;
 815
 816                 if (iter.flags & BTREE_ITER_IS_EXTENTS)
 817                         bch2_key_resize(&delete.k,
 818                                         bpos_min(end, k.k->p).offset -
 819                                         iter.pos.offset);
 820
 821                 ret = bch2_trans_update(trans, &iter, &delete, 0) ?:
 822                       bch2_trans_commit(trans, NULL, NULL,
 823                                         BTREE_INSERT_NOFAIL);
 824 err:
 825                 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
 826                         break;
 827         }
 828
 829         bch2_trans_iter_exit(trans, &iter);
 830         return ret;
 831 }
 832
 833 int bch2_inode_rm(struct bch_fs *c, subvol_inum inum)
 834 {
 835         struct btree_trans *trans = bch2_trans_get(c);
 836         struct btree_iter iter = { NULL };
 837         struct bkey_i_inode_generation delete;
 838         struct bch_inode_unpacked inode_u;
 839         struct bkey_s_c k;
 840         u32 snapshot;
 841         int ret;
 842
 843         /*
 844          * If this was a directory, there shouldn't be any real dirents left -
 845          * but there could be whiteouts (from hash collisions) that we should
 846          * delete:
 847          *
 848          * XXX: the dirent could ideally would delete whiteouts when they're no
 849          * longer needed
 850          */
 851         ret   = bch2_inode_delete_keys(trans, inum, BTREE_ID_extents) ?:
 852                 bch2_inode_delete_keys(trans, inum, BTREE_ID_xattrs) ?:
 853                 bch2_inode_delete_keys(trans, inum, BTREE_ID_dirents);
 854         if (ret)
 855                 goto err;
 856 retry:
 857         bch2_trans_begin(trans);
 858
 859         ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
 860         if (ret)
 861                 goto err;
 862
 863         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
 864                                SPOS(0, inum.inum, snapshot),
 865                                BTREE_ITER_INTENT|BTREE_ITER_CACHED);
 866         ret = bkey_err(k);
 867         if (ret)
 868                 goto err;
 869
 870         if (!bkey_is_inode(k.k)) {
 871                 bch2_fs_inconsistent(c,
 872                                      "inode %llu:%u not found when deleting",
 873                                      inum.inum, snapshot);
 874                 ret = -EIO;
 875                 goto err;
 876         }
 877
 878         bch2_inode_unpack(k, &inode_u);
 879
 880         bkey_inode_generation_init(&delete.k_i);
 881         delete.k.p = iter.pos;
 882         delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
 883
 884         ret   = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
 885                 bch2_trans_commit(trans, NULL, NULL,
 886                                 BTREE_INSERT_NOFAIL);
 887 err:
 888         bch2_trans_iter_exit(trans, &iter);
 889         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 890                 goto retry;
 891
 892         bch2_trans_put(trans);
 893         return ret;
 894 }
 895
 896 int bch2_inode_find_by_inum_nowarn_trans(struct btree_trans *trans,
 897                                   subvol_inum inum,
 898                                   struct bch_inode_unpacked *inode)
 899 {
 900         struct btree_iter iter;
 901         int ret;
 902
 903         ret = bch2_inode_peek_nowarn(trans, &iter, inode, inum, 0);
 904         if (!ret)
 905                 bch2_trans_iter_exit(trans, &iter);
 906         return ret;
 907 }
 908
 909 int bch2_inode_find_by_inum_trans(struct btree_trans *trans,
 910                                   subvol_inum inum,
 911                                   struct bch_inode_unpacked *inode)
 912 {
 913         struct btree_iter iter;
 914         int ret;
 915
 916         ret = bch2_inode_peek(trans, &iter, inode, inum, 0);
 917         if (!ret)
 918                 bch2_trans_iter_exit(trans, &iter);
 919         return ret;
 920 }
 921
 922 int bch2_inode_find_by_inum(struct bch_fs *c, subvol_inum inum,
 923                             struct bch_inode_unpacked *inode)
 924 {
 925         return bch2_trans_do(c, NULL, NULL, 0,
 926                 bch2_inode_find_by_inum_trans(trans, inum, inode));
 927 }
 928
 929 int bch2_inode_nlink_inc(struct bch_inode_unpacked *bi)
 930 {
 931         if (bi->bi_flags & BCH_INODE_UNLINKED)
 932                 bi->bi_flags &= ~BCH_INODE_UNLINKED;
 933         else {
 934                 if (bi->bi_nlink == U32_MAX)
 935                         return -EINVAL;
 936
 937                 bi->bi_nlink++;
 938         }
 939
 940         return 0;
 941 }
 942
 943 void bch2_inode_nlink_dec(struct btree_trans *trans, struct bch_inode_unpacked *bi)
 944 {
 945         if (bi->bi_nlink && (bi->bi_flags & BCH_INODE_UNLINKED)) {
 946                 bch2_trans_inconsistent(trans, "inode %llu unlinked but link count nonzero",
 947                                         bi->bi_inum);
 948                 return;
 949         }
 950
 951         if (bi->bi_flags & BCH_INODE_UNLINKED) {
 952                 bch2_trans_inconsistent(trans, "inode %llu link count underflow", bi->bi_inum);
 953                 return;
 954         }
 955
 956         if (bi->bi_nlink)
 957                 bi->bi_nlink--;
 958         else
 959                 bi->bi_flags |= BCH_INODE_UNLINKED;
 960 }
 961
 962 struct bch_opts bch2_inode_opts_to_opts(struct bch_inode_unpacked *inode)
 963 {
 964         struct bch_opts ret = { 0 };
 965 #define x(_name, _bits)                                                 \
 966         if (inode->bi_##_name)                                          \
 967                 opt_set(ret, _name, inode->bi_##_name - 1);
 968         BCH_INODE_OPTS()
 969 #undef x
 970         return ret;
 971 }
 972
 973 void bch2_inode_opts_get(struct bch_io_opts *opts, struct bch_fs *c,
 974                          struct bch_inode_unpacked *inode)
 975 {
 976 #define x(_name, _bits)         opts->_name = inode_opt_get(c, inode, _name);
 977         BCH_INODE_OPTS()
 978 #undef x
 979
 980         if (opts->nocow)
 981                 opts->compression = opts->background_compression = opts->data_checksum = opts->erasure_code = 0;
 982 }
 983
 984 int bch2_inum_opts_get(struct btree_trans *trans, subvol_inum inum, struct bch_io_opts *opts)
 985 {
 986         struct bch_inode_unpacked inode;
 987         int ret = lockrestart_do(trans, bch2_inode_find_by_inum_trans(trans, inum, &inode));
 988
 989         if (ret)
 990                 return ret;
 991
 992         bch2_inode_opts_get(opts, trans->c, &inode);
 993         return 0;
 994 }
 995
 996 int bch2_inode_rm_snapshot(struct btree_trans *trans, u64 inum, u32 snapshot)
 997 {
 998         struct bch_fs *c = trans->c;
 999         struct btree_iter iter = { NULL };
1000         struct bkey_i_inode_generation delete;
1001         struct bch_inode_unpacked inode_u;
1002         struct bkey_s_c k;
1003         int ret;
1004
1005         do {
1006                 ret   = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
1007                                                       SPOS(inum, 0, snapshot),
1008                                                       SPOS(inum, U64_MAX, snapshot),
1009                                                       0, NULL) ?:
1010                         bch2_btree_delete_range_trans(trans, BTREE_ID_dirents,
1011                                                       SPOS(inum, 0, snapshot),
1012                                                       SPOS(inum, U64_MAX, snapshot),
1013                                                       0, NULL) ?:
1014                         bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs,
1015                                                       SPOS(inum, 0, snapshot),
1016                                                       SPOS(inum, U64_MAX, snapshot),
1017                                                       0, NULL);
1018         } while (ret == -BCH_ERR_transaction_restart_nested);
1019         if (ret)
1020                 goto err;
1021 retry:
1022         bch2_trans_begin(trans);
1023
1024         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
1025                                SPOS(0, inum, snapshot), BTREE_ITER_INTENT);
1026         ret = bkey_err(k);
1027         if (ret)
1028                 goto err;
1029
1030         if (!bkey_is_inode(k.k)) {
1031                 bch2_fs_inconsistent(c,
1032                                      "inode %llu:%u not found when deleting",
1033                                      inum, snapshot);
1034                 ret = -EIO;
1035                 goto err;
1036         }
1037
1038         bch2_inode_unpack(k, &inode_u);
1039
1040         /* Subvolume root? */
1041         if (inode_u.bi_subvol)
1042                 bch_warn(c, "deleting inode %llu marked as unlinked, but also a subvolume root!?", inode_u.bi_inum);
1043
1044         bkey_inode_generation_init(&delete.k_i);
1045         delete.k.p = iter.pos;
1046         delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
1047
1048         ret   = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
1049                 bch2_trans_commit(trans, NULL, NULL,
1050                                 BTREE_INSERT_NOFAIL);
1051 err:
1052         bch2_trans_iter_exit(trans, &iter);
1053         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1054                 goto retry;
1055
1056         return ret ?: -BCH_ERR_transaction_restart_nested;
1057 }
1058
1059 static int may_delete_deleted_inode(struct btree_trans *trans, struct bpos pos)
1060 {
1061         struct bch_fs *c = trans->c;
1062         struct btree_iter iter;
1063         struct bkey_s_c k;
1064         struct bch_inode_unpacked inode;
1065         int ret;
1066
1067         if (bch2_snapshot_is_internal_node(c, pos.snapshot))
1068                 return 0;
1069
1070         if (!fsck_err_on(c->sb.clean, c,
1071                          "filesystem marked as clean but have deleted inode %llu:%u",
1072                          pos.offset, pos.snapshot))
1073                 return 0;
1074
1075         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes, pos, BTREE_ITER_CACHED);
1076         ret = bkey_err(k);
1077         if (ret)
1078                 return ret;
1079
1080         ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
1081         if (fsck_err_on(!bkey_is_inode(k.k), c,
1082                         "nonexistent inode %llu:%u in deleted_inodes btree",
1083                         pos.offset, pos.snapshot))
1084                 goto delete;
1085
1086         ret = bch2_inode_unpack(k, &inode);
1087         if (ret)
1088                 goto err;
1089
1090         if (fsck_err_on(S_ISDIR(inode.bi_mode), c,
1091                         "directory %llu:%u in deleted_inodes btree",
1092                         pos.offset, pos.snapshot))
1093                 goto delete;
1094
1095         if (fsck_err_on(!(inode.bi_flags & BCH_INODE_UNLINKED), c,
1096                         "non-deleted inode %llu:%u in deleted_inodes btree",
1097                         pos.offset, pos.snapshot))
1098                 goto delete;
1099
1100         return 1;
1101 err:
1102 fsck_err:
1103         return ret;
1104 delete:
1105         return bch2_btree_bit_mod(trans, BTREE_ID_deleted_inodes, pos, false);
1106 }
1107
1108 int bch2_delete_dead_inodes(struct bch_fs *c)
1109 {
1110         struct btree_trans *trans = bch2_trans_get(c);
1111         struct btree_iter iter;
1112         struct bkey_s_c k;
1113         int ret;
1114
1115         ret = bch2_btree_write_buffer_flush_sync(trans);
1116         if (ret)
1117                 goto err;
1118
1119         /*
1120          * Weird transaction restart handling here because on successful delete,
1121          * bch2_inode_rm_snapshot() will return a nested transaction restart,
1122          * but we can't retry because the btree write buffer won't have been
1123          * flushed and we'd spin:
1124          */
1125         for_each_btree_key(trans, iter, BTREE_ID_deleted_inodes, POS_MIN,
1126                            BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
1127                 ret = lockrestart_do(trans, may_delete_deleted_inode(trans, k.k->p));
1128                 if (ret < 0)
1129                         break;
1130
1131                 if (ret) {
1132                         if (!test_bit(BCH_FS_RW, &c->flags)) {
1133                                 bch2_trans_unlock(trans);
1134                                 bch2_fs_lazy_rw(c);
1135                         }
1136
1137                         ret = bch2_inode_rm_snapshot(trans, k.k->p.offset, k.k->p.snapshot);
1138                         if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
1139                                 break;
1140                 }
1141         }
1142         bch2_trans_iter_exit(trans, &iter);
1143 err:
1144         bch2_trans_put(trans);
1145
1146         return ret;
1147 }