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 "extents.h"
   6 #include "inode.h"
   7 #include "io.h"
   8 #include "keylist.h"
   9
  10 #include <linux/random.h>
  11
  12 #include <asm/unaligned.h>
  13
  14 #define FIELD_BYTES()                                           \
  15
  16 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
  17 static const u8 bits_table[8] = {
  18         1  * 8 - 1,
  19         2  * 8 - 2,
  20         3  * 8 - 3,
  21         4  * 8 - 4,
  22         6  * 8 - 5,
  23         8  * 8 - 6,
  24         10 * 8 - 7,
  25         13 * 8 - 8,
  26 };
  27
  28 static int inode_encode_field(u8 *out, u8 *end, u64 hi, u64 lo)
  29 {
  30         __be64 in[2] = { cpu_to_be64(hi), cpu_to_be64(lo), };
  31         unsigned shift, bytes, bits = likely(!hi)
  32                 ? fls64(lo)
  33                 : fls64(hi) + 64;
  34
  35         for (shift = 1; shift <= 8; shift++)
  36                 if (bits < bits_table[shift - 1])
  37                         goto got_shift;
  38
  39         BUG();
  40 got_shift:
  41         bytes = byte_table[shift - 1];
  42
  43         BUG_ON(out + bytes > end);
  44
  45         memcpy(out, (u8 *) in + 16 - bytes, bytes);
  46         *out |= (1 << 8) >> shift;
  47
  48         return bytes;
  49 }
  50
  51 static int inode_decode_field(const u8 *in, const u8 *end,
  52                               u64 out[2], unsigned *out_bits)
  53 {
  54         __be64 be[2] = { 0, 0 };
  55         unsigned bytes, shift;
  56         u8 *p;
  57
  58         if (in >= end)
  59                 return -1;
  60
  61         if (!*in)
  62                 return -1;
  63
  64         /*
  65          * position of highest set bit indicates number of bytes:
  66          * shift = number of bits to remove in high byte:
  67          */
  68         shift   = 8 - __fls(*in); /* 1 <= shift <= 8 */
  69         bytes   = byte_table[shift - 1];
  70
  71         if (in + bytes > end)
  72                 return -1;
  73
  74         p = (u8 *) be + 16 - bytes;
  75         memcpy(p, in, bytes);
  76         *p ^= (1 << 8) >> shift;
  77
  78         out[0] = be64_to_cpu(be[0]);
  79         out[1] = be64_to_cpu(be[1]);
  80         *out_bits = out[0] ? 64 + fls64(out[0]) : fls64(out[1]);
  81
  82         return bytes;
  83 }
  84
  85 void bch2_inode_pack(struct bkey_inode_buf *packed,
  86                      const struct bch_inode_unpacked *inode)
  87 {
  88         u8 *out = packed->inode.v.fields;
  89         u8 *end = (void *) &packed[1];
  90         u8 *last_nonzero_field = out;
  91         unsigned nr_fields = 0, last_nonzero_fieldnr = 0;
  92
  93         bkey_inode_init(&packed->inode.k_i);
  94         packed->inode.k.p.inode         = inode->inum;
  95         packed->inode.v.i_hash_seed     = inode->i_hash_seed;
  96         packed->inode.v.i_flags         = cpu_to_le32(inode->i_flags);
  97         packed->inode.v.i_mode          = cpu_to_le16(inode->i_mode);
  98
  99 #define BCH_INODE_FIELD(_name, _bits)                                   \
 100         out += inode_encode_field(out, end, 0, inode->_name);           \
 101         nr_fields++;                                                    \
 102                                                                         \
 103         if (inode->_name) {                                             \
 104                 last_nonzero_field = out;                               \
 105                 last_nonzero_fieldnr = nr_fields;                       \
 106         }
 107
 108         BCH_INODE_FIELDS()
 109 #undef  BCH_INODE_FIELD
 110
 111         out = last_nonzero_field;
 112         nr_fields = last_nonzero_fieldnr;
 113
 114         set_bkey_val_bytes(&packed->inode.k, out - (u8 *) &packed->inode.v);
 115         memset(out, 0,
 116                (u8 *) &packed->inode.v +
 117                bkey_val_bytes(&packed->inode.k) - out);
 118
 119         SET_INODE_NR_FIELDS(&packed->inode.v, nr_fields);
 120
 121         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
 122                 struct bch_inode_unpacked unpacked;
 123
 124                 int ret = bch2_inode_unpack(inode_i_to_s_c(&packed->inode),
 125                                            &unpacked);
 126                 BUG_ON(ret);
 127                 BUG_ON(unpacked.inum            != inode->inum);
 128                 BUG_ON(unpacked.i_hash_seed     != inode->i_hash_seed);
 129                 BUG_ON(unpacked.i_mode          != inode->i_mode);
 130
 131 #define BCH_INODE_FIELD(_name, _bits)   BUG_ON(unpacked._name != inode->_name);
 132                 BCH_INODE_FIELDS()
 133 #undef  BCH_INODE_FIELD
 134         }
 135 }
 136
 137 int bch2_inode_unpack(struct bkey_s_c_inode inode,
 138                       struct bch_inode_unpacked *unpacked)
 139 {
 140         const u8 *in = inode.v->fields;
 141         const u8 *end = (void *) inode.v + bkey_val_bytes(inode.k);
 142         u64 field[2];
 143         unsigned fieldnr = 0, field_bits;
 144         int ret;
 145
 146         unpacked->inum          = inode.k->p.inode;
 147         unpacked->i_hash_seed   = inode.v->i_hash_seed;
 148         unpacked->i_flags       = le32_to_cpu(inode.v->i_flags);
 149         unpacked->i_mode        = le16_to_cpu(inode.v->i_mode);
 150
 151 #define BCH_INODE_FIELD(_name, _bits)                                   \
 152         if (fieldnr++ == INODE_NR_FIELDS(inode.v)) {                    \
 153                 memset(&unpacked->_name, 0,                             \
 154                        sizeof(*unpacked) -                              \
 155                        offsetof(struct bch_inode_unpacked, _name));     \
 156                 return 0;                                               \
 157         }                                                               \
 158                                                                         \
 159         ret = inode_decode_field(in, end, field, &field_bits);          \
 160         if (ret < 0)                                                    \
 161                 return ret;                                             \
 162                                                                         \
 163         if (field_bits > sizeof(unpacked->_name) * 8)                   \
 164                 return -1;                                              \
 165                                                                         \
 166         unpacked->_name = field[1];                                     \
 167         in += ret;
 168
 169         BCH_INODE_FIELDS()
 170 #undef  BCH_INODE_FIELD
 171
 172         /* XXX: signal if there were more fields than expected? */
 173
 174         return 0;
 175 }
 176
 177 static const char *bch2_inode_invalid(const struct bch_fs *c,
 178                                       struct bkey_s_c k)
 179 {
 180         if (k.k->p.offset)
 181                 return "nonzero offset";
 182
 183         switch (k.k->type) {
 184         case BCH_INODE_FS: {
 185                 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
 186                 struct bch_inode_unpacked unpacked;
 187
 188                 if (bkey_val_bytes(k.k) < sizeof(struct bch_inode))
 189                         return "incorrect value size";
 190
 191                 if (k.k->p.inode < BLOCKDEV_INODE_MAX)
 192                         return "fs inode in blockdev range";
 193
 194                 if (INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR)
 195                         return "invalid str hash type";
 196
 197                 if (bch2_inode_unpack(inode, &unpacked))
 198                         return "invalid variable length fields";
 199
 200                 return NULL;
 201         }
 202         case BCH_INODE_BLOCKDEV:
 203                 if (bkey_val_bytes(k.k) != sizeof(struct bch_inode_blockdev))
 204                         return "incorrect value size";
 205
 206                 if (k.k->p.inode >= BLOCKDEV_INODE_MAX)
 207                         return "blockdev inode in fs range";
 208
 209                 return NULL;
 210         default:
 211                 return "invalid type";
 212         }
 213 }
 214
 215 static void bch2_inode_to_text(struct bch_fs *c, char *buf,
 216                                size_t size, struct bkey_s_c k)
 217 {
 218         struct bkey_s_c_inode inode;
 219         struct bch_inode_unpacked unpacked;
 220
 221         switch (k.k->type) {
 222         case BCH_INODE_FS:
 223                 inode = bkey_s_c_to_inode(k);
 224                 if (bch2_inode_unpack(inode, &unpacked)) {
 225                         scnprintf(buf, size, "(unpack error)");
 226                         break;
 227                 }
 228
 229                 scnprintf(buf, size, "i_size %llu", unpacked.i_size);
 230                 break;
 231         }
 232 }
 233
 234 const struct bkey_ops bch2_bkey_inode_ops = {
 235         .key_invalid    = bch2_inode_invalid,
 236         .val_to_text    = bch2_inode_to_text,
 237 };
 238
 239 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
 240                      uid_t uid, gid_t gid, umode_t mode, dev_t rdev)
 241 {
 242         s64 now = timespec_to_bch2_time(c, CURRENT_TIME);
 243
 244         memset(inode_u, 0, sizeof(*inode_u));
 245
 246         /* ick */
 247         inode_u->i_flags |= c->sb.str_hash_type << INODE_STR_HASH_OFFSET;
 248         get_random_bytes(&inode_u->i_hash_seed, sizeof(inode_u->i_hash_seed));
 249
 250         inode_u->i_mode         = mode;
 251         inode_u->i_uid          = uid;
 252         inode_u->i_gid          = gid;
 253         inode_u->i_dev          = rdev;
 254         inode_u->i_atime        = now;
 255         inode_u->i_mtime        = now;
 256         inode_u->i_ctime        = now;
 257         inode_u->i_otime        = now;
 258 }
 259
 260 int bch2_inode_create(struct bch_fs *c, struct bkey_i *inode,
 261                       u64 min, u64 max, u64 *hint)
 262 {
 263         struct btree_iter iter;
 264         bool searched_from_start = false;
 265         int ret;
 266
 267         if (!max)
 268                 max = ULLONG_MAX;
 269
 270         if (c->opts.inodes_32bit)
 271                 max = min_t(u64, max, U32_MAX);
 272
 273         if (*hint >= max || *hint < min)
 274                 *hint = min;
 275
 276         if (*hint == min)
 277                 searched_from_start = true;
 278 again:
 279         bch2_btree_iter_init(&iter, c, BTREE_ID_INODES, POS(*hint, 0),
 280                              BTREE_ITER_INTENT);
 281
 282         while (1) {
 283                 struct bkey_s_c k = bch2_btree_iter_peek_with_holes(&iter);
 284
 285                 ret = btree_iter_err(k);
 286                 if (ret) {
 287                         bch2_btree_iter_unlock(&iter);
 288                         return ret;
 289                 }
 290
 291                 if (k.k->type < BCH_INODE_FS) {
 292                         inode->k.p = k.k->p;
 293
 294                         pr_debug("inserting inode %llu (size %u)",
 295                                  inode->k.p.inode, inode->k.u64s);
 296
 297                         ret = bch2_btree_insert_at(c, NULL, NULL, NULL,
 298                                         BTREE_INSERT_ATOMIC,
 299                                         BTREE_INSERT_ENTRY(&iter, inode));
 300
 301                         if (ret == -EINTR)
 302                                 continue;
 303
 304                         bch2_btree_iter_unlock(&iter);
 305                         if (!ret)
 306                                 *hint = k.k->p.inode + 1;
 307
 308                         return ret;
 309                 } else {
 310                         if (iter.pos.inode == max)
 311                                 break;
 312                         /* slot used */
 313                         bch2_btree_iter_advance_pos(&iter);
 314                 }
 315         }
 316         bch2_btree_iter_unlock(&iter);
 317
 318         if (!searched_from_start) {
 319                 /* Retry from start */
 320                 *hint = min;
 321                 searched_from_start = true;
 322                 goto again;
 323         }
 324
 325         return -ENOSPC;
 326 }
 327
 328 int bch2_inode_truncate(struct bch_fs *c, u64 inode_nr, u64 new_size,
 329                         struct extent_insert_hook *hook, u64 *journal_seq)
 330 {
 331         return bch2_discard(c, POS(inode_nr, new_size), POS(inode_nr + 1, 0),
 332                            ZERO_VERSION, NULL, hook, journal_seq);
 333 }
 334
 335 int bch2_inode_rm(struct bch_fs *c, u64 inode_nr)
 336 {
 337         struct bkey_i delete;
 338         int ret;
 339
 340         ret = bch2_inode_truncate(c, inode_nr, 0, NULL, NULL);
 341         if (ret < 0)
 342                 return ret;
 343
 344         ret = bch2_btree_delete_range(c, BTREE_ID_XATTRS,
 345                                      POS(inode_nr, 0),
 346                                      POS(inode_nr + 1, 0),
 347                                      ZERO_VERSION, NULL, NULL, NULL);
 348         if (ret < 0)
 349                 return ret;
 350
 351         /*
 352          * If this was a directory, there shouldn't be any real dirents left -
 353          * but there could be whiteouts (from hash collisions) that we should
 354          * delete:
 355          *
 356          * XXX: the dirent could ideally would delete whitouts when they're no
 357          * longer needed
 358          */
 359         ret = bch2_btree_delete_range(c, BTREE_ID_DIRENTS,
 360                                      POS(inode_nr, 0),
 361                                      POS(inode_nr + 1, 0),
 362                                      ZERO_VERSION, NULL, NULL, NULL);
 363         if (ret < 0)
 364                 return ret;
 365
 366         bkey_init(&delete.k);
 367         delete.k.p.inode = inode_nr;
 368
 369         return bch2_btree_insert(c, BTREE_ID_INODES, &delete, NULL,
 370                                 NULL, NULL, BTREE_INSERT_NOFAIL);
 371 }
 372
 373 int bch2_inode_find_by_inum(struct bch_fs *c, u64 inode_nr,
 374                             struct bch_inode_unpacked *inode)
 375 {
 376         struct btree_iter iter;
 377         struct bkey_s_c k;
 378         int ret = -ENOENT;
 379
 380         for_each_btree_key(&iter, c, BTREE_ID_INODES,
 381                            POS(inode_nr, 0),
 382                            BTREE_ITER_WITH_HOLES, k) {
 383                 switch (k.k->type) {
 384                 case BCH_INODE_FS:
 385                         ret = bch2_inode_unpack(bkey_s_c_to_inode(k), inode);
 386                         break;
 387                 default:
 388                         /* hole, not found */
 389                         break;
 390                 }
 391
 392                 break;
 393
 394         }
 395
 396         return bch2_btree_iter_unlock(&iter) ?: ret;
 397 }
 398
 399 int bch2_cached_dev_inode_find_by_uuid(struct bch_fs *c, uuid_le *uuid,
 400                                        struct bkey_i_inode_blockdev *ret)
 401 {
 402         struct btree_iter iter;
 403         struct bkey_s_c k;
 404
 405         for_each_btree_key(&iter, c, BTREE_ID_INODES, POS(0, 0), 0, k) {
 406                 if (k.k->p.inode >= BLOCKDEV_INODE_MAX)
 407                         break;
 408
 409                 if (k.k->type == BCH_INODE_BLOCKDEV) {
 410                         struct bkey_s_c_inode_blockdev inode =
 411                                 bkey_s_c_to_inode_blockdev(k);
 412
 413                         pr_debug("found inode %llu: %pU (u64s %u)",
 414                                  inode.k->p.inode, inode.v->i_uuid.b,
 415                                  inode.k->u64s);
 416
 417                         if (CACHED_DEV(inode.v) &&
 418                             !memcmp(uuid, &inode.v->i_uuid, 16)) {
 419                                 bkey_reassemble(&ret->k_i, k);
 420                                 bch2_btree_iter_unlock(&iter);
 421                                 return 0;
 422                         }
 423                 }
 424
 425                 bch2_btree_iter_cond_resched(&iter);
 426         }
 427         bch2_btree_iter_unlock(&iter);
 428         return -ENOENT;
 429 }
 430
 431 #ifdef CONFIG_BCACHEFS_DEBUG
 432 void bch2_inode_pack_test(void)
 433 {
 434         struct bch_inode_unpacked *u, test_inodes[] = {
 435                 {
 436                         .i_atime        = U64_MAX,
 437                         .i_ctime        = U64_MAX,
 438                         .i_mtime        = U64_MAX,
 439                         .i_otime        = U64_MAX,
 440                         .i_size         = U64_MAX,
 441                         .i_sectors      = U64_MAX,
 442                         .i_uid          = U32_MAX,
 443                         .i_gid          = U32_MAX,
 444                         .i_nlink        = U32_MAX,
 445                         .i_generation   = U32_MAX,
 446                         .i_dev          = U32_MAX,
 447                 },
 448         };
 449
 450         for (u = test_inodes;
 451              u < test_inodes + ARRAY_SIZE(test_inodes);
 452              u++) {
 453                 struct bkey_inode_buf p;
 454
 455                 bch2_inode_pack(&p, u);
 456         }
 457 }
 458 #endif