fs/bcachefs/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 "dirent.h"
  11 #include "error.h"
  12 #include "extents.h"
  13 #include "extent_update.h"
  14 #include "inode.h"
  15 #include "str_hash.h"
  16 #include "snapshot.h"
  17 #include "subvolume.h"
  18 #include "varint.h"
  19
  20 #include <linux/random.h>
  21
  22 #include <asm/unaligned.h>
  23
  24 #define x(name, ...)    #name,
  25 const char * const bch2_inode_opts[] = {
  26         BCH_INODE_OPTS()
  27         NULL,
  28 };
  29
  30 static const char * const bch2_inode_flag_strs[] = {
  31         BCH_INODE_FLAGS()
  32         NULL
  33 };
  34 #undef  x
  35
  36 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
  37
  38 static int inode_decode_field(const u8 *in, const u8 *end,
  39                               u64 out[2], unsigned *out_bits)
  40 {
  41         __be64 be[2] = { 0, 0 };
  42         unsigned bytes, shift;
  43         u8 *p;
  44
  45         if (in >= end)
  46                 return -1;
  47
  48         if (!*in)
  49                 return -1;
  50
  51         /*
  52          * position of highest set bit indicates number of bytes:
  53          * shift = number of bits to remove in high byte:
  54          */
  55         shift   = 8 - __fls(*in); /* 1 <= shift <= 8 */
  56         bytes   = byte_table[shift - 1];
  57
  58         if (in + bytes > end)
  59                 return -1;
  60
  61         p = (u8 *) be + 16 - bytes;
  62         memcpy(p, in, bytes);
  63         *p ^= (1 << 8) >> shift;
  64
  65         out[0] = be64_to_cpu(be[0]);
  66         out[1] = be64_to_cpu(be[1]);
  67         *out_bits = out[0] ? 64 + fls64(out[0]) : fls64(out[1]);
  68
  69         return bytes;
  70 }
  71
  72 static inline void bch2_inode_pack_inlined(struct bkey_inode_buf *packed,
  73                                            const struct bch_inode_unpacked *inode)
  74 {
  75         struct bkey_i_inode_v3 *k = &packed->inode;
  76         u8 *out = k->v.fields;
  77         u8 *end = (void *) &packed[1];
  78         u8 *last_nonzero_field = out;
  79         unsigned nr_fields = 0, last_nonzero_fieldnr = 0;
  80         unsigned bytes;
  81         int ret;
  82
  83         bkey_inode_v3_init(&packed->inode.k_i);
  84         packed->inode.k.p.offset        = inode->bi_inum;
  85         packed->inode.v.bi_journal_seq  = cpu_to_le64(inode->bi_journal_seq);
  86         packed->inode.v.bi_hash_seed    = inode->bi_hash_seed;
  87         packed->inode.v.bi_flags        = cpu_to_le64(inode->bi_flags);
  88         packed->inode.v.bi_sectors      = cpu_to_le64(inode->bi_sectors);
  89         packed->inode.v.bi_size         = cpu_to_le64(inode->bi_size);
  90         packed->inode.v.bi_version      = cpu_to_le64(inode->bi_version);
  91         SET_INODEv3_MODE(&packed->inode.v, inode->bi_mode);
  92         SET_INODEv3_FIELDS_START(&packed->inode.v, INODEv3_FIELDS_START_CUR);
  93
  94
  95 #define x(_name, _bits)                                                 \
  96         nr_fields++;                                                    \
  97                                                                         \
  98         if (inode->_name) {                                             \
  99                 ret = bch2_varint_encode_fast(out, inode->_name);       \
 100                 out += ret;                                             \
 101                                                                         \
 102                 if (_bits > 64)                                         \
 103                         *out++ = 0;                                     \
 104                                                                         \
 105                 last_nonzero_field = out;                               \
 106                 last_nonzero_fieldnr = nr_fields;                       \
 107         } else {                                                        \
 108                 *out++ = 0;                                             \
 109                                                                         \
 110                 if (_bits > 64)                                         \
 111                         *out++ = 0;                                     \
 112         }
 113
 114         BCH_INODE_FIELDS_v3()
 115 #undef  x
 116         BUG_ON(out > end);
 117
 118         out = last_nonzero_field;
 119         nr_fields = last_nonzero_fieldnr;
 120
 121         bytes = out - (u8 *) &packed->inode.v;
 122         set_bkey_val_bytes(&packed->inode.k, bytes);
 123         memset_u64s_tail(&packed->inode.v, 0, bytes);
 124
 125         SET_INODEv3_NR_FIELDS(&k->v, nr_fields);
 126
 127         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
 128                 struct bch_inode_unpacked unpacked;
 129
 130                 ret = bch2_inode_unpack(bkey_i_to_s_c(&packed->inode.k_i), &unpacked);
 131                 BUG_ON(ret);
 132                 BUG_ON(unpacked.bi_inum         != inode->bi_inum);
 133                 BUG_ON(unpacked.bi_hash_seed    != inode->bi_hash_seed);
 134                 BUG_ON(unpacked.bi_sectors      != inode->bi_sectors);
 135                 BUG_ON(unpacked.bi_size         != inode->bi_size);
 136                 BUG_ON(unpacked.bi_version      != inode->bi_version);
 137                 BUG_ON(unpacked.bi_mode         != inode->bi_mode);
 138
 139 #define x(_name, _bits) if (unpacked._name != inode->_name)             \
 140                         panic("unpacked %llu should be %llu",           \
 141                               (u64) unpacked._name, (u64) inode->_name);
 142                 BCH_INODE_FIELDS_v3()
 143 #undef  x
 144         }
 145 }
 146
 147 void bch2_inode_pack(struct bkey_inode_buf *packed,
 148                      const struct bch_inode_unpacked *inode)
 149 {
 150         bch2_inode_pack_inlined(packed, inode);
 151 }
 152
 153 static noinline int bch2_inode_unpack_v1(struct bkey_s_c_inode inode,
 154                                 struct bch_inode_unpacked *unpacked)
 155 {
 156         const u8 *in = inode.v->fields;
 157         const u8 *end = bkey_val_end(inode);
 158         u64 field[2];
 159         unsigned fieldnr = 0, field_bits;
 160         int ret;
 161
 162 #define x(_name, _bits)                                 \
 163         if (fieldnr++ == INODE_NR_FIELDS(inode.v)) {                    \
 164                 unsigned offset = offsetof(struct bch_inode_unpacked, _name);\
 165                 memset((void *) unpacked + offset, 0,                   \
 166                        sizeof(*unpacked) - offset);                     \
 167                 return 0;                                               \
 168         }                                                               \
 169                                                                         \
 170         ret = inode_decode_field(in, end, field, &field_bits);          \
 171         if (ret < 0)                                                    \
 172                 return ret;                                             \
 173                                                                         \
 174         if (field_bits > sizeof(unpacked->_name) * 8)                   \
 175                 return -1;                                              \
 176                                                                         \
 177         unpacked->_name = field[1];                                     \
 178         in += ret;
 179
 180         BCH_INODE_FIELDS_v2()
 181 #undef  x
 182
 183         /* XXX: signal if there were more fields than expected? */
 184         return 0;
 185 }
 186
 187 static int bch2_inode_unpack_v2(struct bch_inode_unpacked *unpacked,
 188                                 const u8 *in, const u8 *end,
 189                                 unsigned nr_fields)
 190 {
 191         unsigned fieldnr = 0;
 192         int ret;
 193         u64 v[2];
 194
 195 #define x(_name, _bits)                                                 \
 196         if (fieldnr < nr_fields) {                                      \
 197                 ret = bch2_varint_decode_fast(in, end, &v[0]);          \
 198                 if (ret < 0)                                            \
 199                         return ret;                                     \
 200                 in += ret;                                              \
 201                                                                         \
 202                 if (_bits > 64) {                                       \
 203                         ret = bch2_varint_decode_fast(in, end, &v[1]);  \
 204                         if (ret < 0)                                    \
 205                                 return ret;                             \
 206                         in += ret;                                      \
 207                 } else {                                                \
 208                         v[1] = 0;                                       \
 209                 }                                                       \
 210         } else {                                                        \
 211                 v[0] = v[1] = 0;                                        \
 212         }                                                               \
 213                                                                         \
 214         unpacked->_name = v[0];                                         \
 215         if (v[1] || v[0] != unpacked->_name)                            \
 216                 return -1;                                              \
 217         fieldnr++;
 218
 219         BCH_INODE_FIELDS_v2()
 220 #undef  x
 221
 222         /* XXX: signal if there were more fields than expected? */
 223         return 0;
 224 }
 225
 226 static int bch2_inode_unpack_v3(struct bkey_s_c k,
 227                                 struct bch_inode_unpacked *unpacked)
 228 {
 229         struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
 230         const u8 *in = inode.v->fields;
 231         const u8 *end = bkey_val_end(inode);
 232         unsigned nr_fields = INODEv3_NR_FIELDS(inode.v);
 233         unsigned fieldnr = 0;
 234         int ret;
 235         u64 v[2];
 236
 237         unpacked->bi_inum       = inode.k->p.offset;
 238         unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
 239         unpacked->bi_hash_seed  = inode.v->bi_hash_seed;
 240         unpacked->bi_flags      = le64_to_cpu(inode.v->bi_flags);
 241         unpacked->bi_sectors    = le64_to_cpu(inode.v->bi_sectors);
 242         unpacked->bi_size       = le64_to_cpu(inode.v->bi_size);
 243         unpacked->bi_version    = le64_to_cpu(inode.v->bi_version);
 244         unpacked->bi_mode       = INODEv3_MODE(inode.v);
 245
 246 #define x(_name, _bits)                                                 \
 247         if (fieldnr < nr_fields) {                                      \
 248                 ret = bch2_varint_decode_fast(in, end, &v[0]);          \
 249                 if (ret < 0)                                            \
 250                         return ret;                                     \
 251                 in += ret;                                              \
 252                                                                         \
 253                 if (_bits > 64) {                                       \
 254                         ret = bch2_varint_decode_fast(in, end, &v[1]);  \
 255                         if (ret < 0)                                    \
 256                                 return ret;                             \
 257                         in += ret;                                      \
 258                 } else {                                                \
 259                         v[1] = 0;                                       \
 260                 }                                                       \
 261         } else {                                                        \
 262                 v[0] = v[1] = 0;                                        \
 263         }                                                               \
 264                                                                         \
 265         unpacked->_name = v[0];                                         \
 266         if (v[1] || v[0] != unpacked->_name)                            \
 267                 return -1;                                              \
 268         fieldnr++;
 269
 270         BCH_INODE_FIELDS_v3()
 271 #undef  x
 272
 273         /* XXX: signal if there were more fields than expected? */
 274         return 0;
 275 }
 276
 277 static noinline int bch2_inode_unpack_slowpath(struct bkey_s_c k,
 278                                                struct bch_inode_unpacked *unpacked)
 279 {
 280         memset(unpacked, 0, sizeof(*unpacked));
 281
 282         switch (k.k->type) {
 283         case KEY_TYPE_inode: {
 284                 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
 285
 286                 unpacked->bi_inum       = inode.k->p.offset;
 287                 unpacked->bi_journal_seq= 0;
 288                 unpacked->bi_hash_seed  = inode.v->bi_hash_seed;
 289                 unpacked->bi_flags      = le32_to_cpu(inode.v->bi_flags);
 290                 unpacked->bi_mode       = le16_to_cpu(inode.v->bi_mode);
 291
 292                 if (INODE_NEW_VARINT(inode.v)) {
 293                         return bch2_inode_unpack_v2(unpacked, inode.v->fields,
 294                                                     bkey_val_end(inode),
 295                                                     INODE_NR_FIELDS(inode.v));
 296                 } else {
 297                         return bch2_inode_unpack_v1(inode, unpacked);
 298                 }
 299                 break;
 300         }
 301         case KEY_TYPE_inode_v2: {
 302                 struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
 303
 304                 unpacked->bi_inum       = inode.k->p.offset;
 305                 unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
 306                 unpacked->bi_hash_seed  = inode.v->bi_hash_seed;
 307                 unpacked->bi_flags      = le64_to_cpu(inode.v->bi_flags);
 308                 unpacked->bi_mode       = le16_to_cpu(inode.v->bi_mode);
 309
 310                 return bch2_inode_unpack_v2(unpacked, inode.v->fields,
 311                                             bkey_val_end(inode),
 312                                             INODEv2_NR_FIELDS(inode.v));
 313         }
 314         default:
 315                 BUG();
 316         }
 317 }
 318
 319 int bch2_inode_unpack(struct bkey_s_c k,
 320                       struct bch_inode_unpacked *unpacked)
 321 {
 322         if (likely(k.k->type == KEY_TYPE_inode_v3))
 323                 return bch2_inode_unpack_v3(k, unpacked);
 324         return bch2_inode_unpack_slowpath(k, unpacked);
 325 }
 326
 327 static int bch2_inode_peek_nowarn(struct btree_trans *trans,
 328                     struct btree_iter *iter,
 329                     struct bch_inode_unpacked *inode,
 330                     subvol_inum inum, unsigned flags)
 331 {
 332         struct bkey_s_c k;
 333         u32 snapshot;
 334         int ret;
 335
 336         ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
 337         if (ret)
 338                 return ret;
 339
 340         k = bch2_bkey_get_iter(trans, iter, BTREE_ID_inodes,
 341                                SPOS(0, inum.inum, snapshot),
 342                                flags|BTREE_ITER_CACHED);
 343         ret = bkey_err(k);
 344         if (ret)
 345                 return ret;
 346
 347         ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
 348         if (ret)
 349                 goto err;
 350
 351         ret = bch2_inode_unpack(k, inode);
 352         if (ret)
 353                 goto err;
 354
 355         return 0;
 356 err:
 357         bch2_trans_iter_exit(trans, iter);
 358         return ret;
 359 }
 360
 361 int bch2_inode_peek(struct btree_trans *trans,
 362                     struct btree_iter *iter,
 363                     struct bch_inode_unpacked *inode,
 364                     subvol_inum inum, unsigned flags)
 365 {
 366         int ret = bch2_inode_peek_nowarn(trans, iter, inode, inum, flags);
 367         bch_err_msg(trans->c, ret, "looking up inum %u:%llu:", inum.subvol, inum.inum);
 368         return ret;
 369 }
 370
 371 int bch2_inode_write_flags(struct btree_trans *trans,
 372                      struct btree_iter *iter,
 373                      struct bch_inode_unpacked *inode,
 374                      enum btree_update_flags flags)
 375 {
 376         struct bkey_inode_buf *inode_p;
 377
 378         inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
 379         if (IS_ERR(inode_p))
 380                 return PTR_ERR(inode_p);
 381
 382         bch2_inode_pack_inlined(inode_p, inode);
 383         inode_p->inode.k.p.snapshot = iter->snapshot;
 384         return bch2_trans_update(trans, iter, &inode_p->inode.k_i, flags);
 385 }
 386
 387 struct bkey_i *bch2_inode_to_v3(struct btree_trans *trans, struct bkey_i *k)
 388 {
 389         struct bch_inode_unpacked u;
 390         struct bkey_inode_buf *inode_p;
 391         int ret;
 392
 393         if (!bkey_is_inode(&k->k))
 394                 return ERR_PTR(-ENOENT);
 395
 396         inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
 397         if (IS_ERR(inode_p))
 398                 return ERR_CAST(inode_p);
 399
 400         ret = bch2_inode_unpack(bkey_i_to_s_c(k), &u);
 401         if (ret)
 402                 return ERR_PTR(ret);
 403
 404         bch2_inode_pack(inode_p, &u);
 405         return &inode_p->inode.k_i;
 406 }
 407
 408 static int __bch2_inode_invalid(struct bch_fs *c, struct bkey_s_c k, struct printbuf *err)
 409 {
 410         struct bch_inode_unpacked unpacked;
 411         int ret = 0;
 412
 413         bkey_fsck_err_on(k.k->p.inode, c, err,
 414                          inode_pos_inode_nonzero,
 415                          "nonzero k.p.inode");
 416
 417         bkey_fsck_err_on(k.k->p.offset < BLOCKDEV_INODE_MAX, c, err,
 418                          inode_pos_blockdev_range,
 419                          "fs inode in blockdev range");
 420
 421         bkey_fsck_err_on(bch2_inode_unpack(k, &unpacked), c, err,
 422                          inode_unpack_error,
 423                          "invalid variable length fields");
 424
 425         bkey_fsck_err_on(unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1, c, err,
 426                          inode_checksum_type_invalid,
 427                          "invalid data checksum type (%u >= %u",
 428                          unpacked.bi_data_checksum, BCH_CSUM_OPT_NR + 1);
 429
 430         bkey_fsck_err_on(unpacked.bi_compression &&
 431                          !bch2_compression_opt_valid(unpacked.bi_compression - 1), c, err,
 432                          inode_compression_type_invalid,
 433                          "invalid compression opt %u", unpacked.bi_compression - 1);
 434
 435         bkey_fsck_err_on((unpacked.bi_flags & BCH_INODE_unlinked) &&
 436                          unpacked.bi_nlink != 0, c, err,
 437                          inode_unlinked_but_nlink_nonzero,
 438                          "flagged as unlinked but bi_nlink != 0");
 439
 440         bkey_fsck_err_on(unpacked.bi_subvol && !S_ISDIR(unpacked.bi_mode), c, err,
 441                          inode_subvol_root_but_not_dir,
 442                          "subvolume root but not a directory");
 443 fsck_err:
 444         return ret;
 445 }
 446
 447 int bch2_inode_invalid(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         int ret = 0;
 453
 454         bkey_fsck_err_on(INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR, c, err,
 455                          inode_str_hash_invalid,
 456                          "invalid str hash type (%llu >= %u)",
 457                          INODE_STR_HASH(inode.v), BCH_STR_HASH_NR);
 458
 459         ret = __bch2_inode_invalid(c, k, err);
 460 fsck_err:
 461         return ret;
 462 }
 463
 464 int bch2_inode_v2_invalid(struct bch_fs *c, struct bkey_s_c k,
 465                           enum bkey_invalid_flags flags,
 466                           struct printbuf *err)
 467 {
 468         struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
 469         int ret = 0;
 470
 471         bkey_fsck_err_on(INODEv2_STR_HASH(inode.v) >= BCH_STR_HASH_NR, c, err,
 472                          inode_str_hash_invalid,
 473                          "invalid str hash type (%llu >= %u)",
 474                          INODEv2_STR_HASH(inode.v), BCH_STR_HASH_NR);
 475
 476         ret = __bch2_inode_invalid(c, k, err);
 477 fsck_err:
 478         return ret;
 479 }
 480
 481 int bch2_inode_v3_invalid(struct bch_fs *c, struct bkey_s_c k,
 482                           enum bkey_invalid_flags flags,
 483                           struct printbuf *err)
 484 {
 485         struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
 486         int ret = 0;
 487
 488         bkey_fsck_err_on(INODEv3_FIELDS_START(inode.v) < INODEv3_FIELDS_START_INITIAL ||
 489                          INODEv3_FIELDS_START(inode.v) > bkey_val_u64s(inode.k), c, err,
 490                          inode_v3_fields_start_bad,
 491                          "invalid fields_start (got %llu, min %u max %zu)",
 492                          INODEv3_FIELDS_START(inode.v),
 493                          INODEv3_FIELDS_START_INITIAL,
 494                          bkey_val_u64s(inode.k));
 495
 496         bkey_fsck_err_on(INODEv3_STR_HASH(inode.v) >= BCH_STR_HASH_NR, c, err,
 497                          inode_str_hash_invalid,
 498                          "invalid str hash type (%llu >= %u)",
 499                          INODEv3_STR_HASH(inode.v), BCH_STR_HASH_NR);
 500
 501         ret = __bch2_inode_invalid(c, k, err);
 502 fsck_err:
 503         return ret;
 504 }
 505
 506 static void __bch2_inode_unpacked_to_text(struct printbuf *out,
 507                                           struct bch_inode_unpacked *inode)
 508 {
 509         printbuf_indent_add(out, 2);
 510         prt_printf(out, "mode=%o", inode->bi_mode);
 511         prt_newline(out);
 512
 513         prt_str(out, "flags=");
 514         prt_bitflags(out, bch2_inode_flag_strs, inode->bi_flags & ((1U << 20) - 1));
 515         prt_printf(out, " (%x)", inode->bi_flags);
 516         prt_newline(out);
 517
 518         prt_printf(out, "journal_seq=%llu", inode->bi_journal_seq);
 519         prt_newline(out);
 520
 521         prt_printf(out, "bi_size=%llu", inode->bi_size);
 522         prt_newline(out);
 523
 524         prt_printf(out, "bi_sectors=%llu", inode->bi_sectors);
 525         prt_newline(out);
 526
 527         prt_newline(out);
 528         prt_printf(out, "bi_version=%llu", inode->bi_version);
 529
 530 #define x(_name, _bits)                                         \
 531         prt_printf(out, #_name "=%llu", (u64) inode->_name);    \
 532         prt_newline(out);
 533         BCH_INODE_FIELDS_v3()
 534 #undef  x
 535         printbuf_indent_sub(out, 2);
 536 }
 537
 538 void bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked *inode)
 539 {
 540         prt_printf(out, "inum: %llu ", inode->bi_inum);
 541         __bch2_inode_unpacked_to_text(out, inode);
 542 }
 543
 544 void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
 545 {
 546         struct bch_inode_unpacked inode;
 547
 548         if (bch2_inode_unpack(k, &inode)) {
 549                 prt_printf(out, "(unpack error)");
 550                 return;
 551         }
 552
 553         __bch2_inode_unpacked_to_text(out, &inode);
 554 }
 555
 556 static inline u64 bkey_inode_flags(struct bkey_s_c k)
 557 {
 558         switch (k.k->type) {
 559         case KEY_TYPE_inode:
 560                 return le32_to_cpu(bkey_s_c_to_inode(k).v->bi_flags);
 561         case KEY_TYPE_inode_v2:
 562                 return le64_to_cpu(bkey_s_c_to_inode_v2(k).v->bi_flags);
 563         case KEY_TYPE_inode_v3:
 564                 return le64_to_cpu(bkey_s_c_to_inode_v3(k).v->bi_flags);
 565         default:
 566                 return 0;
 567         }
 568 }
 569
 570 static inline bool bkey_is_deleted_inode(struct bkey_s_c k)
 571 {
 572         return bkey_inode_flags(k) & BCH_INODE_unlinked;
 573 }
 574
 575 int bch2_trigger_inode(struct btree_trans *trans,
 576                        enum btree_id btree_id, unsigned level,
 577                        struct bkey_s_c old,
 578                        struct bkey_s new,
 579                        unsigned flags)
 580 {
 581         s64 nr = bkey_is_inode(new.k) - bkey_is_inode(old.k);
 582
 583         if (flags & BTREE_TRIGGER_TRANSACTIONAL) {
 584                 if (nr) {
 585                         int ret = bch2_replicas_deltas_realloc(trans, 0);
 586                         if (ret)
 587                                 return ret;
 588
 589                         trans->fs_usage_deltas->nr_inodes += nr;
 590                 }
 591
 592                 bool old_deleted = bkey_is_deleted_inode(old);
 593                 bool new_deleted = bkey_is_deleted_inode(new.s_c);
 594                 if (old_deleted != new_deleted) {
 595                         int ret = bch2_btree_bit_mod(trans, BTREE_ID_deleted_inodes, new.k->p, new_deleted);
 596                         if (ret)
 597                                 return ret;
 598                 }
 599         }
 600
 601         if ((flags & BTREE_TRIGGER_ATOMIC) && (flags & BTREE_TRIGGER_INSERT)) {
 602                 BUG_ON(!trans->journal_res.seq);
 603
 604                 bkey_s_to_inode_v3(new).v->bi_journal_seq = cpu_to_le64(trans->journal_res.seq);
 605         }
 606
 607         if (flags & BTREE_TRIGGER_GC) {
 608                 struct bch_fs *c = trans->c;
 609
 610                 percpu_down_read(&c->mark_lock);
 611                 this_cpu_add(c->usage_gc->b.nr_inodes, nr);
 612                 percpu_up_read(&c->mark_lock);
 613         }
 614
 615         return 0;
 616 }
 617
 618 int bch2_inode_generation_invalid(struct bch_fs *c, struct bkey_s_c k,
 619                                   enum bkey_invalid_flags flags,
 620                                   struct printbuf *err)
 621 {
 622         int ret = 0;
 623
 624         bkey_fsck_err_on(k.k->p.inode, c, err,
 625                          inode_pos_inode_nonzero,
 626                          "nonzero k.p.inode");
 627 fsck_err:
 628         return ret;
 629 }
 630
 631 void bch2_inode_generation_to_text(struct printbuf *out, struct bch_fs *c,
 632                                    struct bkey_s_c k)
 633 {
 634         struct bkey_s_c_inode_generation gen = bkey_s_c_to_inode_generation(k);
 635
 636         prt_printf(out, "generation: %u", le32_to_cpu(gen.v->bi_generation));
 637 }
 638
 639 void bch2_inode_init_early(struct bch_fs *c,
 640                            struct bch_inode_unpacked *inode_u)
 641 {
 642         enum bch_str_hash_type str_hash =
 643                 bch2_str_hash_opt_to_type(c, c->opts.str_hash);
 644
 645         memset(inode_u, 0, sizeof(*inode_u));
 646
 647         /* ick */
 648         inode_u->bi_flags |= str_hash << INODE_STR_HASH_OFFSET;
 649         get_random_bytes(&inode_u->bi_hash_seed,
 650                          sizeof(inode_u->bi_hash_seed));
 651 }
 652
 653 void bch2_inode_init_late(struct bch_inode_unpacked *inode_u, u64 now,
 654                           uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
 655                           struct bch_inode_unpacked *parent)
 656 {
 657         inode_u->bi_mode        = mode;
 658         inode_u->bi_uid         = uid;
 659         inode_u->bi_gid         = gid;
 660         inode_u->bi_dev         = rdev;
 661         inode_u->bi_atime       = now;
 662         inode_u->bi_mtime       = now;
 663         inode_u->bi_ctime       = now;
 664         inode_u->bi_otime       = now;
 665
 666         if (parent && parent->bi_mode & S_ISGID) {
 667                 inode_u->bi_gid = parent->bi_gid;
 668                 if (S_ISDIR(mode))
 669                         inode_u->bi_mode |= S_ISGID;
 670         }
 671
 672         if (parent) {
 673 #define x(_name, ...)   inode_u->bi_##_name = parent->bi_##_name;
 674                 BCH_INODE_OPTS()
 675 #undef x
 676         }
 677 }
 678
 679 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
 680                      uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
 681                      struct bch_inode_unpacked *parent)
 682 {
 683         bch2_inode_init_early(c, inode_u);
 684         bch2_inode_init_late(inode_u, bch2_current_time(c),
 685                              uid, gid, mode, rdev, parent);
 686 }
 687
 688 static inline u32 bkey_generation(struct bkey_s_c k)
 689 {
 690         switch (k.k->type) {
 691         case KEY_TYPE_inode:
 692         case KEY_TYPE_inode_v2:
 693                 BUG();
 694         case KEY_TYPE_inode_generation:
 695                 return le32_to_cpu(bkey_s_c_to_inode_generation(k).v->bi_generation);
 696         default:
 697                 return 0;
 698         }
 699 }
 700
 701 /*
 702  * This just finds an empty slot:
 703  */
 704 int bch2_inode_create(struct btree_trans *trans,
 705                       struct btree_iter *iter,
 706                       struct bch_inode_unpacked *inode_u,
 707                       u32 snapshot, u64 cpu)
 708 {
 709         struct bch_fs *c = trans->c;
 710         struct bkey_s_c k;
 711         u64 min, max, start, pos, *hint;
 712         int ret = 0;
 713         unsigned bits = (c->opts.inodes_32bit ? 31 : 63);
 714
 715         if (c->opts.shard_inode_numbers) {
 716                 bits -= c->inode_shard_bits;
 717
 718                 min = (cpu << bits);
 719                 max = (cpu << bits) | ~(ULLONG_MAX << bits);
 720
 721                 min = max_t(u64, min, BLOCKDEV_INODE_MAX);
 722                 hint = c->unused_inode_hints + cpu;
 723         } else {
 724                 min = BLOCKDEV_INODE_MAX;
 725                 max = ~(ULLONG_MAX << bits);
 726                 hint = c->unused_inode_hints;
 727         }
 728
 729         start = READ_ONCE(*hint);
 730
 731         if (start >= max || start < min)
 732                 start = min;
 733
 734         pos = start;
 735         bch2_trans_iter_init(trans, iter, BTREE_ID_inodes, POS(0, pos),
 736                              BTREE_ITER_ALL_SNAPSHOTS|
 737                              BTREE_ITER_INTENT);
 738 again:
 739         while ((k = bch2_btree_iter_peek(iter)).k &&
 740                !(ret = bkey_err(k)) &&
 741                bkey_lt(k.k->p, POS(0, max))) {
 742                 if (pos < iter->pos.offset)
 743                         goto found_slot;
 744
 745                 /*
 746                  * We don't need to iterate over keys in every snapshot once
 747                  * we've found just one:
 748                  */
 749                 pos = iter->pos.offset + 1;
 750                 bch2_btree_iter_set_pos(iter, POS(0, pos));
 751         }
 752
 753         if (!ret && pos < max)
 754                 goto found_slot;
 755
 756         if (!ret && start == min)
 757                 ret = -BCH_ERR_ENOSPC_inode_create;
 758
 759         if (ret) {
 760                 bch2_trans_iter_exit(trans, iter);
 761                 return ret;
 762         }
 763
 764         /* Retry from start */
 765         pos = start = min;
 766         bch2_btree_iter_set_pos(iter, POS(0, pos));
 767         goto again;
 768 found_slot:
 769         bch2_btree_iter_set_pos(iter, SPOS(0, pos, snapshot));
 770         k = bch2_btree_iter_peek_slot(iter);
 771         ret = bkey_err(k);
 772         if (ret) {
 773                 bch2_trans_iter_exit(trans, iter);
 774                 return ret;
 775         }
 776
 777         *hint                   = k.k->p.offset;
 778         inode_u->bi_inum        = k.k->p.offset;
 779         inode_u->bi_generation  = bkey_generation(k);
 780         return 0;
 781 }
 782
 783 static int bch2_inode_delete_keys(struct btree_trans *trans,
 784                                   subvol_inum inum, enum btree_id id)
 785 {
 786         struct btree_iter iter;
 787         struct bkey_s_c k;
 788         struct bkey_i delete;
 789         struct bpos end = POS(inum.inum, U64_MAX);
 790         u32 snapshot;
 791         int ret = 0;
 792
 793         /*
 794          * We're never going to be deleting partial extents, no need to use an
 795          * extent iterator:
 796          */
 797         bch2_trans_iter_init(trans, &iter, id, POS(inum.inum, 0),
 798                              BTREE_ITER_INTENT);
 799
 800         while (1) {
 801                 bch2_trans_begin(trans);
 802
 803                 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
 804                 if (ret)
 805                         goto err;
 806
 807                 bch2_btree_iter_set_snapshot(&iter, snapshot);
 808
 809                 k = bch2_btree_iter_peek_upto(&iter, end);
 810                 ret = bkey_err(k);
 811                 if (ret)
 812                         goto err;
 813
 814                 if (!k.k)
 815                         break;
 816
 817                 bkey_init(&delete.k);
 818                 delete.k.p = iter.pos;
 819
 820                 if (iter.flags & BTREE_ITER_IS_EXTENTS)
 821                         bch2_key_resize(&delete.k,
 822                                         bpos_min(end, k.k->p).offset -
 823                                         iter.pos.offset);
 824
 825                 ret = bch2_trans_update(trans, &iter, &delete, 0) ?:
 826                       bch2_trans_commit(trans, NULL, NULL,
 827                                         BCH_TRANS_COMMIT_no_enospc);
 828 err:
 829                 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
 830                         break;
 831         }
 832
 833         bch2_trans_iter_exit(trans, &iter);
 834         return ret;
 835 }
 836
 837 int bch2_inode_rm(struct bch_fs *c, subvol_inum inum)
 838 {
 839         struct btree_trans *trans = bch2_trans_get(c);
 840         struct btree_iter iter = { NULL };
 841         struct bkey_i_inode_generation delete;
 842         struct bch_inode_unpacked inode_u;
 843         struct bkey_s_c k;
 844         u32 snapshot;
 845         int ret;
 846
 847         /*
 848          * If this was a directory, there shouldn't be any real dirents left -
 849          * but there could be whiteouts (from hash collisions) that we should
 850          * delete:
 851          *
 852          * XXX: the dirent could ideally would delete whiteouts when they're no
 853          * longer needed
 854          */
 855         ret   = bch2_inode_delete_keys(trans, inum, BTREE_ID_extents) ?:
 856                 bch2_inode_delete_keys(trans, inum, BTREE_ID_xattrs) ?:
 857                 bch2_inode_delete_keys(trans, inum, BTREE_ID_dirents);
 858         if (ret)
 859                 goto err;
 860 retry:
 861         bch2_trans_begin(trans);
 862
 863         ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
 864         if (ret)
 865                 goto err;
 866
 867         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
 868                                SPOS(0, inum.inum, snapshot),
 869                                BTREE_ITER_INTENT|BTREE_ITER_CACHED);
 870         ret = bkey_err(k);
 871         if (ret)
 872                 goto err;
 873
 874         if (!bkey_is_inode(k.k)) {
 875                 bch2_fs_inconsistent(c,
 876                                      "inode %llu:%u not found when deleting",
 877                                      inum.inum, snapshot);
 878                 ret = -EIO;
 879                 goto err;
 880         }
 881
 882         bch2_inode_unpack(k, &inode_u);
 883
 884         bkey_inode_generation_init(&delete.k_i);
 885         delete.k.p = iter.pos;
 886         delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
 887
 888         ret   = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
 889                 bch2_trans_commit(trans, NULL, NULL,
 890                                 BCH_TRANS_COMMIT_no_enospc);
 891 err:
 892         bch2_trans_iter_exit(trans, &iter);
 893         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 894                 goto retry;
 895
 896         bch2_trans_put(trans);
 897         return ret;
 898 }
 899
 900 int bch2_inode_find_by_inum_nowarn_trans(struct btree_trans *trans,
 901                                   subvol_inum inum,
 902                                   struct bch_inode_unpacked *inode)
 903 {
 904         struct btree_iter iter;
 905         int ret;
 906
 907         ret = bch2_inode_peek_nowarn(trans, &iter, inode, inum, 0);
 908         if (!ret)
 909                 bch2_trans_iter_exit(trans, &iter);
 910         return ret;
 911 }
 912
 913 int bch2_inode_find_by_inum_trans(struct btree_trans *trans,
 914                                   subvol_inum inum,
 915                                   struct bch_inode_unpacked *inode)
 916 {
 917         struct btree_iter iter;
 918         int ret;
 919
 920         ret = bch2_inode_peek(trans, &iter, inode, inum, 0);
 921         if (!ret)
 922                 bch2_trans_iter_exit(trans, &iter);
 923         return ret;
 924 }
 925
 926 int bch2_inode_find_by_inum(struct bch_fs *c, subvol_inum inum,
 927                             struct bch_inode_unpacked *inode)
 928 {
 929         return bch2_trans_do(c, NULL, NULL, 0,
 930                 bch2_inode_find_by_inum_trans(trans, inum, inode));
 931 }
 932
 933 int bch2_inode_nlink_inc(struct bch_inode_unpacked *bi)
 934 {
 935         if (bi->bi_flags & BCH_INODE_unlinked)
 936                 bi->bi_flags &= ~BCH_INODE_unlinked;
 937         else {
 938                 if (bi->bi_nlink == U32_MAX)
 939                         return -EINVAL;
 940
 941                 bi->bi_nlink++;
 942         }
 943
 944         return 0;
 945 }
 946
 947 void bch2_inode_nlink_dec(struct btree_trans *trans, struct bch_inode_unpacked *bi)
 948 {
 949         if (bi->bi_nlink && (bi->bi_flags & BCH_INODE_unlinked)) {
 950                 bch2_trans_inconsistent(trans, "inode %llu unlinked but link count nonzero",
 951                                         bi->bi_inum);
 952                 return;
 953         }
 954
 955         if (bi->bi_flags & BCH_INODE_unlinked) {
 956                 bch2_trans_inconsistent(trans, "inode %llu link count underflow", bi->bi_inum);
 957                 return;
 958         }
 959
 960         if (bi->bi_nlink)
 961                 bi->bi_nlink--;
 962         else
 963                 bi->bi_flags |= BCH_INODE_unlinked;
 964 }
 965
 966 struct bch_opts bch2_inode_opts_to_opts(struct bch_inode_unpacked *inode)
 967 {
 968         struct bch_opts ret = { 0 };
 969 #define x(_name, _bits)                                                 \
 970         if (inode->bi_##_name)                                          \
 971                 opt_set(ret, _name, inode->bi_##_name - 1);
 972         BCH_INODE_OPTS()
 973 #undef x
 974         return ret;
 975 }
 976
 977 void bch2_inode_opts_get(struct bch_io_opts *opts, struct bch_fs *c,
 978                          struct bch_inode_unpacked *inode)
 979 {
 980 #define x(_name, _bits)         opts->_name = inode_opt_get(c, inode, _name);
 981         BCH_INODE_OPTS()
 982 #undef x
 983
 984         if (opts->nocow)
 985                 opts->compression = opts->background_compression = opts->data_checksum = opts->erasure_code = 0;
 986 }
 987
 988 int bch2_inum_opts_get(struct btree_trans *trans, subvol_inum inum, struct bch_io_opts *opts)
 989 {
 990         struct bch_inode_unpacked inode;
 991         int ret = lockrestart_do(trans, bch2_inode_find_by_inum_trans(trans, inum, &inode));
 992
 993         if (ret)
 994                 return ret;
 995
 996         bch2_inode_opts_get(opts, trans->c, &inode);
 997         return 0;
 998 }
 999
1000 int bch2_inode_rm_snapshot(struct btree_trans *trans, u64 inum, u32 snapshot)
1001 {
1002         struct bch_fs *c = trans->c;
1003         struct btree_iter iter = { NULL };
1004         struct bkey_i_inode_generation delete;
1005         struct bch_inode_unpacked inode_u;
1006         struct bkey_s_c k;
1007         int ret;
1008
1009         do {
1010                 ret   = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
1011                                                       SPOS(inum, 0, snapshot),
1012                                                       SPOS(inum, U64_MAX, snapshot),
1013                                                       0, NULL) ?:
1014                         bch2_btree_delete_range_trans(trans, BTREE_ID_dirents,
1015                                                       SPOS(inum, 0, snapshot),
1016                                                       SPOS(inum, U64_MAX, snapshot),
1017                                                       0, NULL) ?:
1018                         bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs,
1019                                                       SPOS(inum, 0, snapshot),
1020                                                       SPOS(inum, U64_MAX, snapshot),
1021                                                       0, NULL);
1022         } while (ret == -BCH_ERR_transaction_restart_nested);
1023         if (ret)
1024                 goto err;
1025 retry:
1026         bch2_trans_begin(trans);
1027
1028         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
1029                                SPOS(0, inum, snapshot), BTREE_ITER_INTENT);
1030         ret = bkey_err(k);
1031         if (ret)
1032                 goto err;
1033
1034         if (!bkey_is_inode(k.k)) {
1035                 bch2_fs_inconsistent(c,
1036                                      "inode %llu:%u not found when deleting",
1037                                      inum, snapshot);
1038                 ret = -EIO;
1039                 goto err;
1040         }
1041
1042         bch2_inode_unpack(k, &inode_u);
1043
1044         /* Subvolume root? */
1045         if (inode_u.bi_subvol)
1046                 bch_warn(c, "deleting inode %llu marked as unlinked, but also a subvolume root!?", inode_u.bi_inum);
1047
1048         bkey_inode_generation_init(&delete.k_i);
1049         delete.k.p = iter.pos;
1050         delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
1051
1052         ret   = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
1053                 bch2_trans_commit(trans, NULL, NULL,
1054                                 BCH_TRANS_COMMIT_no_enospc);
1055 err:
1056         bch2_trans_iter_exit(trans, &iter);
1057         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1058                 goto retry;
1059
1060         return ret ?: -BCH_ERR_transaction_restart_nested;
1061 }
1062
1063 static int may_delete_deleted_inode(struct btree_trans *trans,
1064                                     struct btree_iter *iter,
1065                                     struct bpos pos,
1066                                     bool *need_another_pass)
1067 {
1068         struct bch_fs *c = trans->c;
1069         struct btree_iter inode_iter;
1070         struct bkey_s_c k;
1071         struct bch_inode_unpacked inode;
1072         int ret;
1073
1074         k = bch2_bkey_get_iter(trans, &inode_iter, BTREE_ID_inodes, pos, BTREE_ITER_CACHED);
1075         ret = bkey_err(k);
1076         if (ret)
1077                 return ret;
1078
1079         ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
1080         if (fsck_err_on(!bkey_is_inode(k.k), c,
1081                         deleted_inode_missing,
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 out;
1089
1090         if (S_ISDIR(inode.bi_mode)) {
1091                 ret = bch2_empty_dir_snapshot(trans, pos.offset, pos.snapshot);
1092                 if (fsck_err_on(ret == -ENOTEMPTY, c, deleted_inode_is_dir,
1093                                 "non empty directory %llu:%u in deleted_inodes btree",
1094                                 pos.offset, pos.snapshot))
1095                         goto delete;
1096                 if (ret)
1097                         goto out;
1098         }
1099
1100         if (fsck_err_on(!(inode.bi_flags & BCH_INODE_unlinked), c,
1101                         deleted_inode_not_unlinked,
1102                         "non-deleted inode %llu:%u in deleted_inodes btree",
1103                         pos.offset, pos.snapshot))
1104                 goto delete;
1105
1106         if (c->sb.clean &&
1107             !fsck_err(c,
1108                       deleted_inode_but_clean,
1109                       "filesystem marked as clean but have deleted inode %llu:%u",
1110                       pos.offset, pos.snapshot)) {
1111                 ret = 0;
1112                 goto out;
1113         }
1114
1115         if (bch2_snapshot_is_internal_node(c, pos.snapshot)) {
1116                 struct bpos new_min_pos;
1117
1118                 ret = bch2_propagate_key_to_snapshot_leaves(trans, inode_iter.btree_id, k, &new_min_pos);
1119                 if (ret)
1120                         goto out;
1121
1122                 inode.bi_flags &= ~BCH_INODE_unlinked;
1123
1124                 ret = bch2_inode_write_flags(trans, &inode_iter, &inode,
1125                                              BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1126                 bch_err_msg(c, ret, "clearing inode unlinked flag");
1127                 if (ret)
1128                         goto out;
1129
1130                 /*
1131                  * We'll need another write buffer flush to pick up the new
1132                  * unlinked inodes in the snapshot leaves:
1133                  */
1134                 *need_another_pass = true;
1135                 goto out;
1136         }
1137
1138         ret = 1;
1139 out:
1140 fsck_err:
1141         bch2_trans_iter_exit(trans, &inode_iter);
1142         return ret;
1143 delete:
1144         ret = bch2_btree_bit_mod(trans, BTREE_ID_deleted_inodes, pos, false);
1145         goto out;
1146 }
1147
1148 int bch2_delete_dead_inodes(struct bch_fs *c)
1149 {
1150         struct btree_trans *trans = bch2_trans_get(c);
1151         bool need_another_pass;
1152         int ret;
1153 again:
1154         need_another_pass = false;
1155
1156         /*
1157          * Weird transaction restart handling here because on successful delete,
1158          * bch2_inode_rm_snapshot() will return a nested transaction restart,
1159          * but we can't retry because the btree write buffer won't have been
1160          * flushed and we'd spin:
1161          */
1162         ret = for_each_btree_key_commit(trans, iter, BTREE_ID_deleted_inodes, POS_MIN,
1163                                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1164                                         NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
1165                 ret = may_delete_deleted_inode(trans, &iter, k.k->p, &need_another_pass);
1166                 if (ret > 0) {
1167                         bch_verbose(c, "deleting unlinked inode %llu:%u", k.k->p.offset, k.k->p.snapshot);
1168
1169                         ret = bch2_inode_rm_snapshot(trans, k.k->p.offset, k.k->p.snapshot);
1170                         /*
1171                          * We don't want to loop here: a transaction restart
1172                          * error here means we handled a transaction restart and
1173                          * we're actually done, but if we loop we'll retry the
1174                          * same key because the write buffer hasn't been flushed
1175                          * yet
1176                          */
1177                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
1178                                 ret = 0;
1179                                 continue;
1180                         }
1181                 }
1182
1183                 ret;
1184         }));
1185
1186         if (!ret && need_another_pass) {
1187                 ret = bch2_btree_write_buffer_flush_sync(trans);
1188                 if (ret)
1189                         goto err;
1190                 goto again;
1191         }
1192 err:
1193         bch2_trans_put(trans);
1194         return ret;
1195 }