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

   1
   2 #include "bcachefs.h"
   3 #include "bkey_methods.h"
   4 #include "btree_update.h"
   5 #include "error.h"
   6 #include "extents.h"
   7 #include "inode.h"
   8 #include "io.h"
   9 #include "keylist.h"
  10
  11 #include <linux/random.h>
  12
  13 #include <asm/unaligned.h>
  14
  15 const char * const bch2_inode_opts[] = {
  16 #define x(name, ...)    #name,
  17         BCH_INODE_OPTS()
  18 #undef  x
  19         NULL,
  20 };
  21
  22 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
  23 static const u8 bits_table[8] = {
  24         1  * 8 - 1,
  25         2  * 8 - 2,
  26         3  * 8 - 3,
  27         4  * 8 - 4,
  28         6  * 8 - 5,
  29         8  * 8 - 6,
  30         10 * 8 - 7,
  31         13 * 8 - 8,
  32 };
  33
  34 static int inode_encode_field(u8 *out, u8 *end, u64 hi, u64 lo)
  35 {
  36         __be64 in[2] = { cpu_to_be64(hi), cpu_to_be64(lo), };
  37         unsigned shift, bytes, bits = likely(!hi)
  38                 ? fls64(lo)
  39                 : fls64(hi) + 64;
  40
  41         for (shift = 1; shift <= 8; shift++)
  42                 if (bits < bits_table[shift - 1])
  43                         goto got_shift;
  44
  45         BUG();
  46 got_shift:
  47         bytes = byte_table[shift - 1];
  48
  49         BUG_ON(out + bytes > end);
  50
  51         memcpy(out, (u8 *) in + 16 - bytes, bytes);
  52         *out |= (1 << 8) >> shift;
  53
  54         return bytes;
  55 }
  56
  57 static int inode_decode_field(const u8 *in, const u8 *end,
  58                               u64 out[2], unsigned *out_bits)
  59 {
  60         __be64 be[2] = { 0, 0 };
  61         unsigned bytes, shift;
  62         u8 *p;
  63
  64         if (in >= end)
  65                 return -1;
  66
  67         if (!*in)
  68                 return -1;
  69
  70         /*
  71          * position of highest set bit indicates number of bytes:
  72          * shift = number of bits to remove in high byte:
  73          */
  74         shift   = 8 - __fls(*in); /* 1 <= shift <= 8 */
  75         bytes   = byte_table[shift - 1];
  76
  77         if (in + bytes > end)
  78                 return -1;
  79
  80         p = (u8 *) be + 16 - bytes;
  81         memcpy(p, in, bytes);
  82         *p ^= (1 << 8) >> shift;
  83
  84         out[0] = be64_to_cpu(be[0]);
  85         out[1] = be64_to_cpu(be[1]);
  86         *out_bits = out[0] ? 64 + fls64(out[0]) : fls64(out[1]);
  87
  88         return bytes;
  89 }
  90
  91 void bch2_inode_pack(struct bkey_inode_buf *packed,
  92                      const struct bch_inode_unpacked *inode)
  93 {
  94         u8 *out = packed->inode.v.fields;
  95         u8 *end = (void *) &packed[1];
  96         u8 *last_nonzero_field = out;
  97         unsigned nr_fields = 0, last_nonzero_fieldnr = 0;
  98
  99         bkey_inode_init(&packed->inode.k_i);
 100         packed->inode.k.p.inode         = inode->bi_inum;
 101         packed->inode.v.bi_hash_seed    = inode->bi_hash_seed;
 102         packed->inode.v.bi_flags        = cpu_to_le32(inode->bi_flags);
 103         packed->inode.v.bi_mode         = cpu_to_le16(inode->bi_mode);
 104
 105 #define x(_name, _bits)                                 \
 106         out += inode_encode_field(out, end, 0, inode->_name);           \
 107         nr_fields++;                                                    \
 108                                                                         \
 109         if (inode->_name) {                                             \
 110                 last_nonzero_field = out;                               \
 111                 last_nonzero_fieldnr = nr_fields;                       \
 112         }
 113
 114         BCH_INODE_FIELDS()
 115 #undef  x
 116
 117         out = last_nonzero_field;
 118         nr_fields = last_nonzero_fieldnr;
 119
 120         set_bkey_val_bytes(&packed->inode.k, out - (u8 *) &packed->inode.v);
 121         memset(out, 0,
 122                (u8 *) &packed->inode.v +
 123                bkey_val_bytes(&packed->inode.k) - out);
 124
 125         SET_INODE_NR_FIELDS(&packed->inode.v, nr_fields);
 126
 127         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
 128                 struct bch_inode_unpacked unpacked;
 129
 130                 int ret = bch2_inode_unpack(inode_i_to_s_c(&packed->inode),
 131                                            &unpacked);
 132                 BUG_ON(ret);
 133                 BUG_ON(unpacked.bi_inum         != inode->bi_inum);
 134                 BUG_ON(unpacked.bi_hash_seed    != inode->bi_hash_seed);
 135                 BUG_ON(unpacked.bi_mode         != inode->bi_mode);
 136
 137 #define x(_name, _bits) BUG_ON(unpacked._name != inode->_name);
 138                 BCH_INODE_FIELDS()
 139 #undef  x
 140         }
 141 }
 142
 143 int bch2_inode_unpack(struct bkey_s_c_inode inode,
 144                       struct bch_inode_unpacked *unpacked)
 145 {
 146         const u8 *in = inode.v->fields;
 147         const u8 *end = (void *) inode.v + bkey_val_bytes(inode.k);
 148         u64 field[2];
 149         unsigned fieldnr = 0, field_bits;
 150         int ret;
 151
 152         unpacked->bi_inum       = inode.k->p.inode;
 153         unpacked->bi_hash_seed  = inode.v->bi_hash_seed;
 154         unpacked->bi_flags      = le32_to_cpu(inode.v->bi_flags);
 155         unpacked->bi_mode       = le16_to_cpu(inode.v->bi_mode);
 156
 157 #define x(_name, _bits)                                 \
 158         if (fieldnr++ == INODE_NR_FIELDS(inode.v)) {                    \
 159                 memset(&unpacked->_name, 0,                             \
 160                        sizeof(*unpacked) -                              \
 161                        offsetof(struct bch_inode_unpacked, _name));     \
 162                 return 0;                                               \
 163         }                                                               \
 164                                                                         \
 165         ret = inode_decode_field(in, end, field, &field_bits);          \
 166         if (ret < 0)                                                    \
 167                 return ret;                                             \
 168                                                                         \
 169         if (field_bits > sizeof(unpacked->_name) * 8)                   \
 170                 return -1;                                              \
 171                                                                         \
 172         unpacked->_name = field[1];                                     \
 173         in += ret;
 174
 175         BCH_INODE_FIELDS()
 176 #undef  x
 177
 178         /* XXX: signal if there were more fields than expected? */
 179
 180         return 0;
 181 }
 182
 183 const char *bch2_inode_invalid(const struct bch_fs *c, struct bkey_s_c k)
 184 {
 185                 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
 186                 struct bch_inode_unpacked unpacked;
 187
 188         if (k.k->p.offset)
 189                 return "nonzero offset";
 190
 191         if (bkey_val_bytes(k.k) < sizeof(struct bch_inode))
 192                 return "incorrect value size";
 193
 194         if (k.k->p.inode < BLOCKDEV_INODE_MAX)
 195                 return "fs inode in blockdev range";
 196
 197         if (INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR)
 198                 return "invalid str hash type";
 199
 200         if (bch2_inode_unpack(inode, &unpacked))
 201                 return "invalid variable length fields";
 202
 203         if (unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1)
 204                 return "invalid data checksum type";
 205
 206         if (unpacked.bi_compression >= BCH_COMPRESSION_OPT_NR + 1)
 207                 return "invalid data checksum type";
 208
 209         if ((unpacked.bi_flags & BCH_INODE_UNLINKED) &&
 210             unpacked.bi_nlink != 0)
 211                 return "flagged as unlinked but bi_nlink != 0";
 212
 213         return NULL;
 214 }
 215
 216 void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c,
 217                        struct bkey_s_c k)
 218 {
 219         struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
 220         struct bch_inode_unpacked unpacked;
 221
 222         if (bch2_inode_unpack(inode, &unpacked)) {
 223                 pr_buf(out, "(unpack error)");
 224                 return;
 225         }
 226
 227 #define x(_name, _bits)                                         \
 228         pr_buf(out, #_name ": %llu ", (u64) unpacked._name);
 229         BCH_INODE_FIELDS()
 230 #undef  x
 231 }
 232
 233 const char *bch2_inode_generation_invalid(const struct bch_fs *c,
 234                                           struct bkey_s_c k)
 235 {
 236         if (k.k->p.offset)
 237                 return "nonzero offset";
 238
 239         if (bkey_val_bytes(k.k) != sizeof(struct bch_inode_generation))
 240                 return "incorrect value size";
 241
 242         return NULL;
 243 }
 244
 245 void bch2_inode_generation_to_text(struct printbuf *out, struct bch_fs *c,
 246                                    struct bkey_s_c k)
 247 {
 248 }
 249
 250 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
 251                      uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
 252                      struct bch_inode_unpacked *parent)
 253 {
 254         s64 now = bch2_current_time(c);
 255
 256         memset(inode_u, 0, sizeof(*inode_u));
 257
 258         /* ick */
 259         inode_u->bi_flags |= c->opts.str_hash << INODE_STR_HASH_OFFSET;
 260         get_random_bytes(&inode_u->bi_hash_seed,
 261                          sizeof(inode_u->bi_hash_seed));
 262
 263         inode_u->bi_mode        = mode;
 264         inode_u->bi_uid         = uid;
 265         inode_u->bi_gid         = gid;
 266         inode_u->bi_dev         = rdev;
 267         inode_u->bi_atime       = now;
 268         inode_u->bi_mtime       = now;
 269         inode_u->bi_ctime       = now;
 270         inode_u->bi_otime       = now;
 271
 272         if (parent) {
 273 #define x(_name, ...)   inode_u->bi_##_name = parent->bi_##_name;
 274                 BCH_INODE_OPTS()
 275 #undef x
 276         }
 277 }
 278
 279 static inline u32 bkey_generation(struct bkey_s_c k)
 280 {
 281         switch (k.k->type) {
 282         case KEY_TYPE_inode:
 283                 BUG();
 284         case KEY_TYPE_inode_generation:
 285                 return le32_to_cpu(bkey_s_c_to_inode_generation(k).v->bi_generation);
 286         default:
 287                 return 0;
 288         }
 289 }
 290
 291 int __bch2_inode_create(struct btree_trans *trans,
 292                         struct bch_inode_unpacked *inode_u,
 293                         u64 min, u64 max, u64 *hint)
 294 {
 295         struct bch_fs *c = trans->c;
 296         struct bkey_inode_buf *inode_p;
 297         struct btree_iter *iter;
 298         u64 start;
 299         int ret;
 300
 301         if (!max)
 302                 max = ULLONG_MAX;
 303
 304         if (c->opts.inodes_32bit)
 305                 max = min_t(u64, max, U32_MAX);
 306
 307         start = READ_ONCE(*hint);
 308
 309         if (start >= max || start < min)
 310                 start = min;
 311
 312         inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
 313         if (IS_ERR(inode_p))
 314                 return PTR_ERR(inode_p);
 315
 316         iter = bch2_trans_get_iter(trans,
 317                         BTREE_ID_INODES, POS(start, 0),
 318                         BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 319         if (IS_ERR(iter))
 320                 return PTR_ERR(iter);
 321 again:
 322         while (1) {
 323                 struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
 324
 325                 ret = bkey_err(k);
 326                 if (ret)
 327                         return ret;
 328
 329                 switch (k.k->type) {
 330                 case KEY_TYPE_inode:
 331                         /* slot used */
 332                         if (iter->pos.inode >= max)
 333                                 goto out;
 334
 335                         bch2_btree_iter_next_slot(iter);
 336                         break;
 337
 338                 default:
 339                         *hint                   = k.k->p.inode;
 340                         inode_u->bi_inum        = k.k->p.inode;
 341                         inode_u->bi_generation  = bkey_generation(k);
 342
 343                         bch2_inode_pack(inode_p, inode_u);
 344                         bch2_trans_update(trans,
 345                                 BTREE_INSERT_ENTRY(iter, &inode_p->inode.k_i));
 346                         return 0;
 347                 }
 348         }
 349 out:
 350         if (start != min) {
 351                 /* Retry from start */
 352                 start = min;
 353                 bch2_btree_iter_set_pos(iter, POS(start, 0));
 354                 goto again;
 355         }
 356
 357         return -ENOSPC;
 358 }
 359
 360 int bch2_inode_create(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
 361                       u64 min, u64 max, u64 *hint)
 362 {
 363         return bch2_trans_do(c, NULL, BTREE_INSERT_ATOMIC,
 364                         __bch2_inode_create(&trans, inode_u, min, max, hint));
 365 }
 366
 367 int bch2_inode_rm(struct bch_fs *c, u64 inode_nr)
 368 {
 369         struct btree_trans trans;
 370         struct btree_iter *iter;
 371         struct bkey_i_inode_generation delete;
 372         struct bpos start = POS(inode_nr, 0);
 373         struct bpos end = POS(inode_nr + 1, 0);
 374         int ret;
 375
 376         /*
 377          * If this was a directory, there shouldn't be any real dirents left -
 378          * but there could be whiteouts (from hash collisions) that we should
 379          * delete:
 380          *
 381          * XXX: the dirent could ideally would delete whiteouts when they're no
 382          * longer needed
 383          */
 384         ret   = bch2_btree_delete_range(c, BTREE_ID_EXTENTS,
 385                                         start, end, NULL) ?:
 386                 bch2_btree_delete_range(c, BTREE_ID_XATTRS,
 387                                         start, end, NULL) ?:
 388                 bch2_btree_delete_range(c, BTREE_ID_DIRENTS,
 389                                         start, end, NULL);
 390         if (ret)
 391                 return ret;
 392
 393         bch2_trans_init(&trans, c);
 394
 395         iter = bch2_trans_get_iter(&trans, BTREE_ID_INODES, POS(inode_nr, 0),
 396                                    BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 397         do {
 398                 struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
 399                 u32 bi_generation = 0;
 400
 401                 ret = bkey_err(k);
 402                 if (ret)
 403                         break;
 404
 405                 bch2_fs_inconsistent_on(k.k->type != KEY_TYPE_inode, c,
 406                                         "inode %llu not found when deleting",
 407                                         inode_nr);
 408
 409                 switch (k.k->type) {
 410                 case KEY_TYPE_inode: {
 411                         struct bch_inode_unpacked inode_u;
 412
 413                         if (!bch2_inode_unpack(bkey_s_c_to_inode(k), &inode_u))
 414                                 bi_generation = inode_u.bi_generation + 1;
 415                         break;
 416                 }
 417                 case KEY_TYPE_inode_generation: {
 418                         struct bkey_s_c_inode_generation g =
 419                                 bkey_s_c_to_inode_generation(k);
 420                         bi_generation = le32_to_cpu(g.v->bi_generation);
 421                         break;
 422                 }
 423                 }
 424
 425                 if (!bi_generation) {
 426                         bkey_init(&delete.k);
 427                         delete.k.p.inode = inode_nr;
 428                 } else {
 429                         bkey_inode_generation_init(&delete.k_i);
 430                         delete.k.p.inode = inode_nr;
 431                         delete.v.bi_generation = cpu_to_le32(bi_generation);
 432                 }
 433
 434                 bch2_trans_update(&trans,
 435                                   BTREE_INSERT_ENTRY(iter, &delete.k_i));
 436
 437                 ret = bch2_trans_commit(&trans, NULL, NULL,
 438                                         BTREE_INSERT_ATOMIC|
 439                                         BTREE_INSERT_NOFAIL);
 440         } while (ret == -EINTR);
 441
 442         bch2_trans_exit(&trans);
 443         return ret;
 444 }
 445
 446 int bch2_inode_find_by_inum_trans(struct btree_trans *trans, u64 inode_nr,
 447                                   struct bch_inode_unpacked *inode)
 448 {
 449         struct btree_iter *iter;
 450         struct bkey_s_c k;
 451         int ret = -ENOENT;
 452
 453         iter = bch2_trans_get_iter(trans, BTREE_ID_INODES,
 454                         POS(inode_nr, 0), BTREE_ITER_SLOTS);
 455         if (IS_ERR(iter))
 456                 return PTR_ERR(iter);
 457
 458         k = bch2_btree_iter_peek_slot(iter);
 459         if (k.k->type == KEY_TYPE_inode)
 460                 ret = bch2_inode_unpack(bkey_s_c_to_inode(k), inode);
 461
 462         bch2_trans_iter_put(trans, iter);
 463
 464         return ret;
 465 }
 466
 467 int bch2_inode_find_by_inum(struct bch_fs *c, u64 inode_nr,
 468                             struct bch_inode_unpacked *inode)
 469 {
 470         return bch2_trans_do(c, NULL, 0,
 471                 bch2_inode_find_by_inum_trans(&trans, inode_nr, inode));
 472 }
 473
 474 #ifdef CONFIG_BCACHEFS_DEBUG
 475 void bch2_inode_pack_test(void)
 476 {
 477         struct bch_inode_unpacked *u, test_inodes[] = {
 478                 {
 479                         .bi_atime       = U64_MAX,
 480                         .bi_ctime       = U64_MAX,
 481                         .bi_mtime       = U64_MAX,
 482                         .bi_otime       = U64_MAX,
 483                         .bi_size        = U64_MAX,
 484                         .bi_sectors     = U64_MAX,
 485                         .bi_uid         = U32_MAX,
 486                         .bi_gid         = U32_MAX,
 487                         .bi_nlink       = U32_MAX,
 488                         .bi_generation  = U32_MAX,
 489                         .bi_dev         = U32_MAX,
 490                 },
 491         };
 492
 493         for (u = test_inodes;
 494              u < test_inodes + ARRAY_SIZE(test_inodes);
 495              u++) {
 496                 struct bkey_inode_buf p;
 497
 498                 bch2_inode_pack(&p, u);
 499         }
 500 }
 501 #endif