| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | #include "rrdengine.h" |
| 3 | |
| 4 | // the default value is set in ND_PROFILE, not here |
| 5 | time_t dbengine_journal_v2_unmount_time = 120; |
| 6 | |
| 7 | /* Careful to always call this before creating a new journal file */ |
| 8 | int journalfile_v1_extent_write(struct rrdengine_instance *ctx, struct rrdengine_datafile *datafile, WAL *wal) |
| 9 | { |
| 10 | uv_fs_t request; |
| 11 | struct rrdengine_journalfile *journalfile = datafile->journalfile; |
| 12 | uv_buf_t iov; |
| 13 | |
| 14 | if (wal->size < wal->buf_size) { |
| 15 | /* simulate an empty transaction to skip the rest of the block */ |
| 16 | *(uint8_t *) (wal->buf + wal->size) = STORE_PADDING; |
| 17 | } |
| 18 | |
| 19 | uint64_t journalfile_position; |
| 20 | spinlock_lock(&journalfile->unsafe.spinlock); |
| 21 | journalfile_position = journalfile->unsafe.pos; |
| 22 | journalfile->unsafe.pos += wal->buf_size; |
| 23 | spinlock_unlock(&journalfile->unsafe.spinlock); |
| 24 | |
| 25 | iov = uv_buf_init(wal->buf, wal->buf_size); |
| 26 | |
| 27 | int retries = 10; |
| 28 | int ret = -1; |
| 29 | while (ret < 0 && --retries) { |
| 30 | ret = uv_fs_write(NULL, &request, journalfile->file, &iov, 1, (int64_t)journalfile_position, NULL); |
| 31 | uv_fs_req_cleanup(&request); |
| 32 | if (ret < 0) { |
| 33 | if (ret == -ENOSPC || ret == -EBADF || ret == -EACCES || ret == -EROFS || ret == -EINVAL) |
| 34 | break; |
| 35 | sleep_usec(300 * USEC_PER_MS); |
| 36 | } |
| 37 | } |
| 38 | |
| 39 | if (unlikely(ret < 0)) { |
| 40 | ctx_io_error(ctx); |
| 41 | goto done; |
| 42 | } |
| 43 | |
| 44 | ctx_current_disk_space_increase(ctx, wal->buf_size); |
| 45 | ctx_io_write_op_bytes(ctx, wal->buf_size); |
| 46 | |
| 47 | done: |
| 48 | wal_release(wal); |
| 49 | worker_is_idle(); |
| 50 | return ret; |
| 51 | } |
| 52 | |
| 53 | void journalfile_v2_generate_path(struct rrdengine_datafile *datafile, char *str, size_t maxlen) |
| 54 | { |
| 55 | (void) snprintfz(str, maxlen, "%s/" WALFILE_PREFIX RRDENG_FILE_NUMBER_PRINT_TMPL WALFILE_EXTENSION_V2, |
| 56 | datafile_ctx(datafile)->config.dbfiles_path, datafile->tier, datafile->fileno); |
| 57 | } |
| 58 | |
| 59 | void journalfile_v1_generate_path(struct rrdengine_datafile *datafile, char *str, size_t maxlen) |
| 60 | { |
| 61 | (void) snprintfz(str, maxlen - 1, "%s/" WALFILE_PREFIX RRDENG_FILE_NUMBER_PRINT_TMPL WALFILE_EXTENSION, |
| 62 | datafile_ctx(datafile)->config.dbfiles_path, datafile->tier, datafile->fileno); |
| 63 | } |
| 64 | |
| 65 | // ---------------------------------------------------------------------------- |
| 66 | |
| 67 | ALWAYS_INLINE struct rrdengine_datafile *njfv2idx_find_and_acquire_j2_header(NJFV2IDX_FIND_STATE *s) { |
| 68 | if (unlikely(!s)) return NULL; |
| 69 | |
| 70 | struct rrdengine_datafile *datafile = NULL; |
| 71 | |
| 72 | rw_spinlock_read_lock(&s->ctx->njfv2idx.spinlock); |
| 73 | |
| 74 | Pvoid_t *PValue = NULL; |
| 75 | |
| 76 | if(unlikely(!s->init)) { |
| 77 | s->init = true; |
| 78 | s->last = s->wanted_start_time_s; |
| 79 | |
| 80 | PValue = JudyLPrev(s->ctx->njfv2idx.JudyL, &s->last, PJE0); |
| 81 | if (unlikely(PValue == PJERR)) |
| 82 | fatal("DBENGINE: NJFV2IDX corrupted judy array"); |
| 83 | |
| 84 | if(!PValue) { |
| 85 | s->last = 0; |
| 86 | PValue = JudyLFirst(s->ctx->njfv2idx.JudyL, &s->last, PJE0); |
| 87 | if (unlikely(PValue == PJERR)) |
| 88 | fatal("DBENGINE: NJFV2IDX corrupted judy array"); |
| 89 | |
| 90 | if(!PValue) |
| 91 | s->last = s->wanted_start_time_s; |
| 92 | } |
| 93 | } |
| 94 | |
| 95 | while(1) { |
| 96 | if (likely(!PValue)) { |
| 97 | PValue = JudyLNext(s->ctx->njfv2idx.JudyL, &s->last, PJE0); |
| 98 | if (unlikely(PValue == PJERR)) |
| 99 | fatal("DBENGINE: NJFV2IDX corrupted judy array"); |
| 100 | |
| 101 | if(!PValue) { |
| 102 | // cannot find anything after that point |
| 103 | datafile = NULL; |
| 104 | break; |
| 105 | } |
| 106 | } |
| 107 | |
| 108 | datafile = *PValue; |
| 109 | |
| 110 | if (!datafile || !datafile_acquire(datafile, DATAFILE_ACQUIRE_PAGE_DETAILS)) { |
| 111 | datafile = NULL; |
| 112 | PValue = NULL; |
| 113 | continue; |
| 114 | } |
| 115 | |
| 116 | struct rrdengine_journalfile *journalfile = datafile->journalfile; |
| 117 | |
| 118 | if (!journalfile) { |
| 119 | datafile_release(datafile, DATAFILE_ACQUIRE_PAGE_DETAILS); |
| 120 | datafile = NULL; |
| 121 | PValue = NULL; |
| 122 | continue; |
| 123 | } |
| 124 | |
| 125 | TIME_RANGE_COMPARE rc = is_page_in_time_range(journalfile->v2.first_time_s, |
| 126 | journalfile->v2.last_time_s, |
| 127 | s->wanted_start_time_s, |
| 128 | s->wanted_end_time_s); |
| 129 | |
| 130 | if(rc == PAGE_IS_IN_RANGE) { |
| 131 | s->j2_header_acquired = journalfile_v2_data_acquire(journalfile, NULL, |
| 132 | s->wanted_start_time_s, |
| 133 | s->wanted_end_time_s); |
| 134 | if(s->j2_header_acquired) { |
| 135 | // this is good to return |
| 136 | break; |
| 137 | } |
| 138 | |
| 139 | datafile_release(datafile, DATAFILE_ACQUIRE_PAGE_DETAILS); |
| 140 | datafile = NULL; |
| 141 | PValue = NULL; |
| 142 | continue; |
| 143 | } |
| 144 | else if(rc == PAGE_IS_IN_THE_PAST) { |
| 145 | // continue to get the next |
| 146 | datafile_release(datafile, DATAFILE_ACQUIRE_PAGE_DETAILS); |
| 147 | datafile = NULL; |
| 148 | PValue = NULL; |
| 149 | continue; |
| 150 | } |
| 151 | else /* PAGE_IS_IN_THE_FUTURE */ { |
| 152 | // we finished - no more datafiles |
| 153 | datafile_release(datafile, DATAFILE_ACQUIRE_PAGE_DETAILS); |
| 154 | datafile = NULL; |
| 155 | PValue = NULL; |
| 156 | break; |
| 157 | } |
| 158 | } |
| 159 | |
| 160 | if(!datafile) |
| 161 | s->j2_header_acquired = NULL; |
| 162 | |
| 163 | rw_spinlock_read_unlock(&s->ctx->njfv2idx.spinlock); |
| 164 | |
| 165 | return datafile; |
| 166 | } |
| 167 | |
| 168 | static void njfv2idx_add(struct rrdengine_datafile *datafile) { |
| 169 | if(unlikely(!datafile)) |
| 170 | fatal("DBENGINE: NJFV2IDX trying to index a journal file with no datafile"); |
| 171 | |
| 172 | struct rrdengine_instance *ctx = datafile_ctx(datafile); |
| 173 | |
| 174 | internal_fatal(datafile->journalfile->v2.last_time_s <= 0, "DBENGINE: NJFV2IDX trying to index a journal file with invalid first_time_s"); |
| 175 | |
| 176 | rw_spinlock_write_lock(&ctx->njfv2idx.spinlock); |
| 177 | datafile->journalfile->njfv2idx.indexed_as = datafile->journalfile->v2.last_time_s; |
| 178 | |
| 179 | do { |
| 180 | internal_fatal(datafile->journalfile->njfv2idx.indexed_as <= 0, "DBENGINE: NJFV2IDX journalfile is already indexed"); |
| 181 | |
| 182 | Pvoid_t *PValue = JudyLIns(&ctx->njfv2idx.JudyL, datafile->journalfile->njfv2idx.indexed_as, PJE0); |
| 183 | if (!PValue || PValue == PJERR) |
| 184 | fatal("DBENGINE: NJFV2IDX corrupted judy array"); |
| 185 | |
| 186 | if (unlikely(*PValue)) { |
| 187 | // already there |
| 188 | datafile->journalfile->njfv2idx.indexed_as++; |
| 189 | } |
| 190 | else { |
| 191 | *PValue = datafile; |
| 192 | break; |
| 193 | } |
| 194 | } while(1); |
| 195 | |
| 196 | rw_spinlock_write_unlock(&ctx->njfv2idx.spinlock); |
| 197 | } |
| 198 | |
| 199 | static void njfv2idx_remove(struct rrdengine_datafile *datafile) { |
| 200 | internal_fatal(!datafile->journalfile->njfv2idx.indexed_as, "DBENGINE: NJFV2IDX journalfile to remove is not indexed"); |
| 201 | |
| 202 | struct rrdengine_instance *ctx = datafile_ctx(datafile); |
| 203 | rw_spinlock_write_lock(&ctx->njfv2idx.spinlock); |
| 204 | |
| 205 | int rc = JudyLDel(&ctx->njfv2idx.JudyL, datafile->journalfile->njfv2idx.indexed_as, PJE0); |
| 206 | (void)rc; |
| 207 | internal_fatal(!rc, "DBENGINE: NJFV2IDX cannot remove entry"); |
| 208 | |
| 209 | datafile->journalfile->njfv2idx.indexed_as = 0; |
| 210 | |
| 211 | rw_spinlock_write_unlock(&ctx->njfv2idx.spinlock); |
| 212 | } |
| 213 | |
| 214 | // ---------------------------------------------------------------------------- |
| 215 | |
| 216 | static struct journal_v2_header *journalfile_v2_mounted_data_get(struct rrdengine_journalfile *journalfile, size_t *data_size) { |
| 217 | struct journal_v2_header *j2_header = NULL; |
| 218 | |
| 219 | spinlock_lock(&journalfile->data_spinlock); |
| 220 | |
| 221 | if(!journalfile->mmap.data) { |
| 222 | journalfile->mmap.data = nd_mmap(NULL, journalfile->mmap.size, PROT_READ, MAP_SHARED, journalfile->mmap.fd, 0); |
| 223 | if (journalfile->mmap.data == MAP_FAILED) { |
| 224 | internal_fatal(true, "DBENGINE: failed to re-mmap() journal file v2"); |
| 225 | close(journalfile->mmap.fd); |
| 226 | journalfile->mmap.fd = -1; |
| 227 | journalfile->mmap.data = NULL; |
| 228 | journalfile->mmap.size = 0; |
| 229 | |
| 230 | journalfile->v2.flags &= ~(JOURNALFILE_FLAG_IS_AVAILABLE | JOURNALFILE_FLAG_IS_MOUNTED); |
| 231 | |
| 232 | ctx_fs_error(datafile_ctx(journalfile->datafile)); |
| 233 | } |
| 234 | else { |
| 235 | __atomic_add_fetch(&rrdeng_cache_efficiency_stats.journal_v2_mapped, 1, __ATOMIC_RELAXED); |
| 236 | |
| 237 | madvise_dontfork(journalfile->mmap.data, journalfile->mmap.size); |
| 238 | madvise_dontdump(journalfile->mmap.data, journalfile->mmap.size); |
| 239 | // madvise_dontneed(journalfile->mmap.data, journalfile->mmap.size); |
| 240 | |
| 241 | journalfile->v2.flags |= JOURNALFILE_FLAG_IS_AVAILABLE | JOURNALFILE_FLAG_IS_MOUNTED; |
| 242 | JOURNALFILE_FLAGS flags = journalfile->v2.flags; |
| 243 | |
| 244 | if(flags & JOURNALFILE_FLAG_MOUNTED_FOR_RETENTION) { |
| 245 | // sequential access pattern during MRG population |
| 246 | madvise_sequential(journalfile->mmap.data, journalfile->v2.size_of_directory); |
| 247 | madvise_willneed(journalfile->mmap.data, journalfile->v2.size_of_directory); |
| 248 | } |
| 249 | else { |
| 250 | madvise_random(journalfile->mmap.data, journalfile->mmap.size); |
| 251 | } |
| 252 | } |
| 253 | } |
| 254 | |
| 255 | if(journalfile->mmap.data) { |
| 256 | j2_header = journalfile->mmap.data; |
| 257 | |
| 258 | if (data_size) |
| 259 | *data_size = journalfile->mmap.size; |
| 260 | } |
| 261 | |
| 262 | spinlock_unlock(&journalfile->data_spinlock); |
| 263 | |
| 264 | return j2_header; |
| 265 | } |
| 266 | |
| 267 | static bool journalfile_v2_mounted_data_unmount(struct rrdengine_journalfile *journalfile, bool have_locks, bool wait) { |
| 268 | bool unmounted = false; |
| 269 | |
| 270 | if(!have_locks) { |
| 271 | if(!wait) { |
| 272 | if (!spinlock_trylock(&journalfile->data_spinlock)) |
| 273 | return false; |
| 274 | } |
| 275 | else |
| 276 | spinlock_lock(&journalfile->data_spinlock); |
| 277 | } |
| 278 | |
| 279 | if(!journalfile->v2.refcount) { |
| 280 | if(journalfile->mmap.data) { |
| 281 | if (nd_munmap(journalfile->mmap.data, journalfile->mmap.size)) { |
| 282 | char path[RRDENG_PATH_MAX]; |
| 283 | journalfile_v2_generate_path(journalfile->datafile, path, sizeof(path)); |
| 284 | netdata_log_error("DBENGINE: failed to unmap index file \"%s\"", path); |
| 285 | internal_fatal(true, "DBENGINE: failed to unmap file \"%s\"", path); |
| 286 | ctx_fs_error(datafile_ctx(journalfile->datafile)); |
| 287 | } |
| 288 | else { |
| 289 | __atomic_add_fetch(&rrdeng_cache_efficiency_stats.journal_v2_unmapped, 1, __ATOMIC_RELAXED); |
| 290 | journalfile->mmap.data = NULL; |
| 291 | journalfile->v2.flags &= ~JOURNALFILE_FLAG_IS_MOUNTED; |
| 292 | } |
| 293 | } |
| 294 | |
| 295 | unmounted = true; |
| 296 | } |
| 297 | |
| 298 | if(!have_locks) { |
| 299 | spinlock_unlock(&journalfile->data_spinlock); |
| 300 | } |
| 301 | |
| 302 | return unmounted; |
| 303 | } |
| 304 | |
| 305 | void journalfile_v2_data_unmount_cleanup(time_t now_s) { |
| 306 | // DO NOT WAIT ON ANY LOCK!!! |
| 307 | |
| 308 | for(size_t tier = 0; tier < (size_t)nd_profile.storage_tiers;tier++) { |
| 309 | struct rrdengine_instance *ctx = multidb_ctx[tier]; |
| 310 | if(!ctx) continue; |
| 311 | |
| 312 | struct rrdengine_datafile *datafile; |
| 313 | if(netdata_rwlock_tryrdlock(&ctx->datafiles.rwlock) != 0) |
| 314 | continue; |
| 315 | |
| 316 | bool first_then_next = true; |
| 317 | Pvoid_t *Pvalue = NULL; |
| 318 | Word_t Index = 0; |
| 319 | |
| 320 | while((Pvalue = JudyLFirstThenNext(ctx->datafiles.JudyL, &Index, &first_then_next))) { |
| 321 | |
| 322 | datafile = *Pvalue; |
| 323 | if (!datafile) |
| 324 | continue; |
| 325 | |
| 326 | struct rrdengine_journalfile *journalfile = datafile->journalfile; |
| 327 | |
| 328 | if(!spinlock_trylock(&journalfile->data_spinlock)) |
| 329 | continue; |
| 330 | |
| 331 | bool unmount = false; |
| 332 | if (!journalfile->v2.refcount && (journalfile->v2.flags & JOURNALFILE_FLAG_IS_MOUNTED)) { |
| 333 | // this journal has no references and it is mounted |
| 334 | |
| 335 | if (!journalfile->v2.not_needed_since_s) |
| 336 | journalfile->v2.not_needed_since_s = now_s; |
| 337 | |
| 338 | else if ( |
| 339 | dbengine_journal_v2_unmount_time && now_s - journalfile->v2.not_needed_since_s >= dbengine_journal_v2_unmount_time) |
| 340 | // enough time has passed since we last needed this journal |
| 341 | unmount = true; |
| 342 | } |
| 343 | spinlock_unlock(&journalfile->data_spinlock); |
| 344 | |
| 345 | if (unmount) |
| 346 | journalfile_v2_mounted_data_unmount(journalfile, false, false); |
| 347 | } |
| 348 | netdata_rwlock_rdunlock(&ctx->datafiles.rwlock); |
| 349 | } |
| 350 | } |
| 351 | |
| 352 | ALWAYS_INLINE struct journal_v2_header *journalfile_v2_data_acquire_with_hint(struct rrdengine_journalfile *journalfile, |
| 353 | size_t *data_size, time_t wanted_first_time_s, time_t wanted_last_time_s, JOURNALFILE_V2_ACCESS_HINT hint) |
| 354 | { |
| 355 | spinlock_lock(&journalfile->data_spinlock); |
| 356 | |
| 357 | bool has_data = (journalfile->v2.flags & JOURNALFILE_FLAG_IS_AVAILABLE); |
| 358 | bool is_mounted = (journalfile->v2.flags & JOURNALFILE_FLAG_IS_MOUNTED); |
| 359 | bool do_we_need_it = false; |
| 360 | |
| 361 | if(has_data) { |
| 362 | if (!wanted_first_time_s || !wanted_last_time_s || |
| 363 | is_page_in_time_range(journalfile->v2.first_time_s, journalfile->v2.last_time_s, |
| 364 | wanted_first_time_s, wanted_last_time_s) == PAGE_IS_IN_RANGE) { |
| 365 | bool was_sequential = (journalfile->v2.flags & JOURNALFILE_FLAG_MOUNTED_FOR_RETENTION); |
| 366 | |
| 367 | journalfile->v2.refcount++; |
| 368 | |
| 369 | do_we_need_it = true; |
| 370 | |
| 371 | if(hint == JOURNALFILE_V2_ACCESS_AUTO) |
| 372 | hint = (!wanted_first_time_s && !wanted_last_time_s) ? |
| 373 | JOURNALFILE_V2_ACCESS_SEQUENTIAL_DIRECTORY : |
| 374 | JOURNALFILE_V2_ACCESS_RANDOM; |
| 375 | |
| 376 | bool want_sequential = (hint == JOURNALFILE_V2_ACCESS_SEQUENTIAL_DIRECTORY); |
| 377 | if (want_sequential) |
| 378 | journalfile->v2.flags |= JOURNALFILE_FLAG_MOUNTED_FOR_RETENTION; |
| 379 | else |
| 380 | journalfile->v2.flags &= ~JOURNALFILE_FLAG_MOUNTED_FOR_RETENTION; |
| 381 | |
| 382 | if (is_mounted && journalfile->mmap.data) { |
| 383 | if (!was_sequential && want_sequential) { |
| 384 | madvise_sequential(journalfile->mmap.data, journalfile->v2.size_of_directory); |
| 385 | madvise_willneed(journalfile->mmap.data, journalfile->v2.size_of_directory); |
| 386 | } |
| 387 | else if (was_sequential && !want_sequential) |
| 388 | madvise_random(journalfile->mmap.data, journalfile->mmap.size); |
| 389 | } |
| 390 | |
| 391 | } |
| 392 | } |
| 393 | spinlock_unlock(&journalfile->data_spinlock); |
| 394 | |
| 395 | if(do_we_need_it) |
| 396 | return journalfile_v2_mounted_data_get(journalfile, data_size); |
| 397 | |
| 398 | return NULL; |
| 399 | } |
| 400 | |
| 401 | ALWAYS_INLINE struct journal_v2_header *journalfile_v2_data_acquire(struct rrdengine_journalfile *journalfile, size_t *data_size, time_t wanted_first_time_s, time_t wanted_last_time_s) { |
| 402 | return journalfile_v2_data_acquire_with_hint(journalfile, data_size, wanted_first_time_s, wanted_last_time_s, JOURNALFILE_V2_ACCESS_AUTO); |
| 403 | } |
| 404 | |
| 405 | ALWAYS_INLINE void journalfile_v2_data_release(struct rrdengine_journalfile *journalfile) { |
| 406 | spinlock_lock(&journalfile->data_spinlock); |
| 407 | |
| 408 | internal_fatal(!journalfile->mmap.data, "trying to release a journalfile without data"); |
| 409 | internal_fatal(journalfile->v2.refcount < 1, "trying to release a non-acquired journalfile"); |
| 410 | |
| 411 | bool unmount = false; |
| 412 | |
| 413 | journalfile->v2.refcount--; |
| 414 | |
| 415 | if(journalfile->v2.refcount == 0) { |
| 416 | journalfile->v2.not_needed_since_s = 0; |
| 417 | |
| 418 | if(journalfile->v2.flags & JOURNALFILE_FLAG_MOUNTED_FOR_RETENTION) |
| 419 | unmount = true; |
| 420 | } |
| 421 | spinlock_unlock(&journalfile->data_spinlock); |
| 422 | |
| 423 | if(unmount) |
| 424 | journalfile_v2_mounted_data_unmount(journalfile, false, true); |
| 425 | } |
| 426 | |
| 427 | bool journalfile_v2_data_available(struct rrdengine_journalfile *journalfile) { |
| 428 | |
| 429 | spinlock_lock(&journalfile->data_spinlock); |
| 430 | bool has_data = (journalfile->v2.flags & JOURNALFILE_FLAG_IS_AVAILABLE); |
| 431 | spinlock_unlock(&journalfile->data_spinlock); |
| 432 | |
| 433 | return has_data; |
| 434 | } |
| 435 | |
| 436 | size_t journalfile_v2_data_size_get(struct rrdengine_journalfile *journalfile) { |
| 437 | |
| 438 | spinlock_lock(&journalfile->data_spinlock); |
| 439 | size_t data_size = journalfile->mmap.size; |
| 440 | spinlock_unlock(&journalfile->data_spinlock); |
| 441 | |
| 442 | return data_size; |
| 443 | } |
| 444 | |
| 445 | void journalfile_v2_data_set(struct rrdengine_journalfile *journalfile, int fd, void *journal_data, uint32_t journal_data_size) { |
| 446 | if(unlikely(!journalfile)) |
| 447 | fatal("DBENGINE: JOURNALFILE: trying to set journal data without a journalfile"); |
| 448 | |
| 449 | if(unlikely(!journalfile->datafile)) |
| 450 | fatal("DBENGINE: JOURNALFILE: trying to set journal data without a datafile"); |
| 451 | |
| 452 | spinlock_lock(&journalfile->data_spinlock); |
| 453 | |
| 454 | internal_fatal(journalfile->mmap.fd != -1, "DBENGINE JOURNALFILE: trying to re-set journal fd"); |
| 455 | internal_fatal(journalfile->mmap.data, "DBENGINE JOURNALFILE: trying to re-set journal_data"); |
| 456 | internal_fatal(journalfile->v2.refcount, "DBENGINE JOURNALFILE: trying to re-set journal_data of referenced journalfile"); |
| 457 | |
| 458 | journalfile->mmap.fd = fd; |
| 459 | journalfile->mmap.data = journal_data; |
| 460 | journalfile->mmap.size = journal_data_size; |
| 461 | journalfile->v2.not_needed_since_s = now_monotonic_sec(); |
| 462 | journalfile->v2.flags |= JOURNALFILE_FLAG_IS_AVAILABLE | JOURNALFILE_FLAG_IS_MOUNTED; |
| 463 | |
| 464 | struct journal_v2_header *j2_header = journalfile->mmap.data; |
| 465 | journalfile->v2.first_time_s = (time_t)(j2_header->start_time_ut / USEC_PER_SEC); |
| 466 | journalfile->v2.last_time_s = (time_t)(j2_header->end_time_ut / USEC_PER_SEC); |
| 467 | journalfile->v2.size_of_directory = j2_header->metric_offset + j2_header->metric_count * sizeof(struct journal_metric_list); |
| 468 | |
| 469 | journalfile_v2_mounted_data_unmount(journalfile, true, true); |
| 470 | |
| 471 | spinlock_unlock(&journalfile->data_spinlock); |
| 472 | |
| 473 | njfv2idx_add(journalfile->datafile); |
| 474 | } |
| 475 | |
| 476 | static void journalfile_v2_data_unmap_permanently(struct rrdengine_journalfile *journalfile) { |
| 477 | njfv2idx_remove(journalfile->datafile); |
| 478 | |
| 479 | bool has_references = false; |
| 480 | char path_v2[RRDENG_PATH_MAX]; |
| 481 | |
| 482 | journalfile_v2_generate_path(journalfile->datafile, path_v2, sizeof(path_v2)); |
| 483 | |
| 484 | do { |
| 485 | if (has_references) |
| 486 | sleep_usec(10 * USEC_PER_MS); |
| 487 | |
| 488 | spinlock_lock(&journalfile->data_spinlock); |
| 489 | |
| 490 | if(journalfile_v2_mounted_data_unmount(journalfile, true, true)) { |
| 491 | if(journalfile->mmap.fd != -1) |
| 492 | close(journalfile->mmap.fd); |
| 493 | |
| 494 | journalfile->mmap.fd = -1; |
| 495 | journalfile->mmap.data = NULL; |
| 496 | journalfile->mmap.size = 0; |
| 497 | journalfile->v2.first_time_s = 0; |
| 498 | journalfile->v2.last_time_s = 0; |
| 499 | journalfile->v2.flags = 0; |
| 500 | has_references = false; |
| 501 | } |
| 502 | else { |
| 503 | has_references = true; |
| 504 | nd_log_limit_static_global_var(journalfile_erl, 10, 0); |
| 505 | nd_log_limit(&journalfile_erl, NDLS_DAEMON, NDLP_WARNING, "DBENGINE: journalfile \"%s\" is not available for unmap", path_v2); |
| 506 | } |
| 507 | |
| 508 | spinlock_unlock(&journalfile->data_spinlock); |
| 509 | |
| 510 | } while(has_references); |
| 511 | } |
| 512 | |
| 513 | struct rrdengine_journalfile *journalfile_alloc_and_init(struct rrdengine_datafile *datafile) |
| 514 | { |
| 515 | struct rrdengine_journalfile *journalfile = callocz(1, sizeof(struct rrdengine_journalfile)); |
| 516 | journalfile->datafile = datafile; |
| 517 | spinlock_init(&journalfile->data_spinlock); |
| 518 | spinlock_init(&journalfile->unsafe.spinlock); |
| 519 | journalfile->mmap.fd = -1; |
| 520 | datafile->journalfile = journalfile; |
| 521 | return journalfile; |
| 522 | } |
| 523 | |
| 524 | static int close_uv_file(struct rrdengine_datafile *datafile, uv_file file) |
| 525 | { |
| 526 | int ret; |
| 527 | char path[RRDENG_PATH_MAX]; |
| 528 | |
| 529 | struct rrdengine_instance *ctx = datafile_ctx(datafile); |
| 530 | journalfile_v1_generate_path(datafile, path, sizeof(path)); |
| 531 | |
| 532 | CLOSE_FILE(ctx, path, file, ret); |
| 533 | return ret; |
| 534 | } |
| 535 | |
| 536 | int journalfile_close(struct rrdengine_journalfile *journalfile, struct rrdengine_datafile *datafile) |
| 537 | { |
| 538 | if(journalfile_v2_data_available(journalfile)) { |
| 539 | journalfile_v2_data_unmap_permanently(journalfile); |
| 540 | return 0; |
| 541 | } |
| 542 | |
| 543 | return close_uv_file(datafile, journalfile->file); |
| 544 | } |
| 545 | |
| 546 | int journalfile_unlink(struct rrdengine_journalfile *journalfile) |
| 547 | { |
| 548 | struct rrdengine_datafile *datafile = journalfile->datafile; |
| 549 | struct rrdengine_instance *ctx = datafile_ctx(datafile); |
| 550 | int ret; |
| 551 | |
| 552 | char path[RRDENG_PATH_MAX]; |
| 553 | journalfile_v1_generate_path(datafile, path, sizeof(path)); |
| 554 | |
| 555 | UNLINK_FILE(ctx, path, ret); |
| 556 | if (ret == 0) |
| 557 | __atomic_add_fetch(&ctx->stats.journalfile_deletions, 1, __ATOMIC_RELAXED); |
| 558 | |
| 559 | return ret; |
| 560 | } |
| 561 | |
| 562 | uint8_t journalfile_destroy_unsafe(struct rrdengine_journalfile *journalfile, struct rrdengine_datafile *datafile) |
| 563 | { |
| 564 | struct rrdengine_instance *ctx = datafile_ctx(datafile); |
| 565 | int ret; |
| 566 | uv_fs_t req_v2 = { 0 }; |
| 567 | uv_fs_t req_v1 = { 0 }; |
| 568 | char path[RRDENG_PATH_MAX]; |
| 569 | char path_v2[RRDENG_PATH_MAX]; |
| 570 | |
| 571 | journalfile_v1_generate_path(datafile, path, sizeof(path)); |
| 572 | journalfile_v2_generate_path(datafile, path_v2, sizeof(path)); |
| 573 | |
| 574 | if (journalfile->file) |
| 575 | (void)close_uv_file(datafile, journalfile->file); |
| 576 | |
| 577 | // Wait for all references to be released and unmap before deleting files. |
| 578 | // This prevents SIGBUS when threads are still accessing the mmap'd data. |
| 579 | if(journalfile_v2_data_available(journalfile)) |
| 580 | journalfile_v2_data_unmap_permanently(journalfile); |
| 581 | |
| 582 | // Now safe to delete the files - no threads are accessing them |
| 583 | uint8_t deleted = 0; |
| 584 | |
| 585 | ret = uv_fs_unlink(NULL, &req_v2, path_v2, NULL); |
| 586 | if (ret == 0) |
| 587 | deleted |= JOURNALFILE_DELETED_V2; |
| 588 | else if (ret != UV_ENOENT) { |
| 589 | netdata_log_error("DBENGINE: uv_fs_unlink(\"%s\"): %s", path_v2, uv_strerror(ret)); |
| 590 | ctx_fs_error(ctx); |
| 591 | } |
| 592 | uv_fs_req_cleanup(&req_v2); |
| 593 | |
| 594 | ret = uv_fs_unlink(NULL, &req_v1, path, NULL); |
| 595 | if (ret == 0) |
| 596 | deleted |= JOURNALFILE_DELETED_V1; |
| 597 | else if (ret != UV_ENOENT) { |
| 598 | netdata_log_error("DBENGINE: uv_fs_unlink(\"%s\"): %s", path, uv_strerror(ret)); |
| 599 | ctx_fs_error(ctx); |
| 600 | } |
| 601 | uv_fs_req_cleanup(&req_v1); |
| 602 | |
| 603 | __atomic_add_fetch(&ctx->stats.journalfile_deletions, __builtin_popcount(deleted), __ATOMIC_RELAXED); |
| 604 | |
| 605 | return deleted; |
| 606 | } |
| 607 | |
| 608 | int journalfile_create(struct rrdengine_journalfile *journalfile, struct rrdengine_datafile *datafile) |
| 609 | { |
| 610 | struct rrdengine_instance *ctx = datafile_ctx(datafile); |
| 611 | uv_fs_t req; |
| 612 | uv_file file; |
| 613 | int ret, fd; |
| 614 | struct rrdeng_jf_sb *superblock = NULL; |
| 615 | uv_buf_t iov; |
| 616 | char path[RRDENG_PATH_MAX]; |
| 617 | |
| 618 | journalfile_v1_generate_path(datafile, path, sizeof(path)); |
| 619 | fd = open_file_for_io(path, O_CREAT | O_RDWR | O_TRUNC, &file, dbengine_use_direct_io); |
| 620 | if (fd < 0) { |
| 621 | ctx_fs_error(ctx); |
| 622 | return fd; |
| 623 | } |
| 624 | journalfile->file = file; |
| 625 | |
| 626 | (void)posix_memalignz((void *)&superblock, RRDFILE_ALIGNMENT, sizeof(*superblock)); |
| 627 | memset(superblock, 0, sizeof(*superblock)); |
| 628 | (void) strncpy(superblock->magic_number, RRDENG_JF_MAGIC, RRDENG_MAGIC_SZ); |
| 629 | (void) strncpy(superblock->version, RRDENG_JF_VER, RRDENG_VER_SZ); |
| 630 | |
| 631 | iov = uv_buf_init((void *)superblock, sizeof(*superblock)); |
| 632 | |
| 633 | int retries = 10; |
| 634 | ret = -1; |
| 635 | while (ret < 0 && --retries) { |
| 636 | ret = uv_fs_write(NULL, &req, file, &iov, 1, 0, NULL); |
| 637 | uv_fs_req_cleanup(&req); |
| 638 | if (ret < 0) { |
| 639 | if (ret == -ENOSPC || ret == -EBADF || ret == -EACCES || ret == -EROFS || ret == -EINVAL) |
| 640 | break; |
| 641 | sleep_usec(300 * USEC_PER_MS); |
| 642 | } |
| 643 | } |
| 644 | |
| 645 | posix_memalign_freez(superblock); |
| 646 | |
| 647 | if (ret < 0) { |
| 648 | journalfile_destroy_unsafe(journalfile, datafile); |
| 649 | ctx_io_error(ctx); |
| 650 | nd_log_limit_static_global_var(dbengine_erl, 10, 0); |
| 651 | nd_log_limit(&dbengine_erl, NDLS_DAEMON, NDLP_ERR, "DBENGINE: Failed to create journlfile \"%s\"", path); |
| 652 | return ret; |
| 653 | } |
| 654 | |
| 655 | __atomic_add_fetch(&ctx->stats.journalfile_creations, 1, __ATOMIC_RELAXED); |
| 656 | journalfile->unsafe.pos = sizeof(*superblock); |
| 657 | ctx_io_write_op_bytes(ctx, sizeof(*superblock)); |
| 658 | |
| 659 | return 0; |
| 660 | } |
| 661 | |
| 662 | static int journalfile_check_superblock(uv_file file) |
| 663 | { |
| 664 | int ret; |
| 665 | struct rrdeng_jf_sb *superblock = NULL; |
| 666 | uv_buf_t iov; |
| 667 | uv_fs_t req; |
| 668 | |
| 669 | (void)posix_memalignz((void *)&superblock, RRDFILE_ALIGNMENT, sizeof(*superblock)); |
| 670 | iov = uv_buf_init((void *)superblock, sizeof(*superblock)); |
| 671 | |
| 672 | ret = uv_fs_read(NULL, &req, file, &iov, 1, 0, NULL); |
| 673 | if (ret < 0) { |
| 674 | netdata_log_error("DBENGINE: uv_fs_read: %s", uv_strerror(ret)); |
| 675 | uv_fs_req_cleanup(&req); |
| 676 | goto error; |
| 677 | } |
| 678 | fatal_assert(req.result >= 0); |
| 679 | uv_fs_req_cleanup(&req); |
| 680 | |
| 681 | |
| 682 | char jf_magic[RRDENG_MAGIC_SZ] = RRDENG_JF_MAGIC; |
| 683 | char jf_ver[RRDENG_VER_SZ] = RRDENG_JF_VER; |
| 684 | if (strncmp(superblock->magic_number, jf_magic, RRDENG_MAGIC_SZ) != 0 || |
| 685 | strncmp(superblock->version, jf_ver, RRDENG_VER_SZ) != 0) { |
| 686 | nd_log(NDLS_DAEMON, NDLP_ERR, "DBENGINE: File has invalid superblock."); |
| 687 | ret = UV_EINVAL; |
| 688 | } else { |
| 689 | ret = 0; |
| 690 | } |
| 691 | error: |
| 692 | posix_memalign_freez(superblock); |
| 693 | return ret; |
| 694 | } |
| 695 | |
| 696 | static void journalfile_restore_extent_metadata(struct rrdengine_instance *ctx, struct rrdengine_journalfile *journalfile, void *buf, unsigned max_size) |
| 697 | { |
| 698 | static bitmap64_t page_error_map = BITMAP64_INITIALIZER; |
| 699 | unsigned i, count, payload_length, descr_size; |
| 700 | struct rrdeng_jf_store_data *jf_metric_data; |
| 701 | |
| 702 | jf_metric_data = buf; |
| 703 | count = jf_metric_data->number_of_pages; |
| 704 | descr_size = sizeof(*jf_metric_data->descr) * count; |
| 705 | payload_length = sizeof(*jf_metric_data) + descr_size; |
| 706 | if (payload_length > max_size || !rrdeng_valid_extent_disk_size(jf_metric_data->extent_size)) { |
| 707 | netdata_log_error("DBENGINE: corrupted transaction payload."); |
| 708 | return; |
| 709 | } |
| 710 | |
| 711 | time_t now_s = max_acceptable_collected_time(); |
| 712 | time_t extent_first_time_s = journalfile->v2.first_time_s ? journalfile->v2.first_time_s : LONG_MAX; |
| 713 | time_t extent_last_time_s = journalfile->v2.last_time_s ? journalfile->v2.last_time_s : 0; |
| 714 | for (i = 0; i < count ; ++i) { |
| 715 | nd_uuid_t *temp_id; |
| 716 | uint8_t page_type = jf_metric_data->descr[i].type; |
| 717 | |
| 718 | if (page_type > RRDENG_PAGE_TYPE_MAX) { |
| 719 | if (!bitmap64_get(&page_error_map, page_type)) { |
| 720 | netdata_log_error("DBENGINE: unknown page type %d encountered.", page_type); |
| 721 | bitmap64_set(&page_error_map, page_type); |
| 722 | } |
| 723 | continue; |
| 724 | } |
| 725 | |
| 726 | temp_id = (nd_uuid_t *)jf_metric_data->descr[i].uuid; |
| 727 | METRIC *metric = mrg_metric_get_and_acquire_by_uuid(main_mrg, temp_id, (Word_t)ctx); |
| 728 | |
| 729 | struct rrdeng_extent_page_descr *descr = &jf_metric_data->descr[i]; |
| 730 | VALIDATED_PAGE_DESCRIPTOR vd = validate_extent_page_descr( |
| 731 | descr, now_s, |
| 732 | (metric) ? mrg_metric_get_update_every_s(main_mrg, metric) : 0, |
| 733 | false); |
| 734 | |
| 735 | if(!vd.is_valid) { |
| 736 | if(metric) |
| 737 | mrg_metric_release(main_mrg, metric); |
| 738 | |
| 739 | continue; |
| 740 | } |
| 741 | |
| 742 | bool update_metric_time = true; |
| 743 | if (!metric) { |
| 744 | MRG_ENTRY entry = { |
| 745 | .uuid = temp_id, |
| 746 | .section = (Word_t)ctx, |
| 747 | .first_time_s = vd.start_time_s, |
| 748 | .last_time_s = vd.end_time_s, |
| 749 | .latest_update_every_s = vd.update_every_s, |
| 750 | }; |
| 751 | |
| 752 | bool added; |
| 753 | metric = mrg_metric_add_and_acquire(main_mrg, entry, &added); |
| 754 | if(added) |
| 755 | update_metric_time = false; |
| 756 | |
| 757 | if (vd.update_every_s) { |
| 758 | uint64_t samples = (vd.end_time_s - vd.start_time_s) / vd.update_every_s; |
| 759 | __atomic_add_fetch(&ctx->atomic.samples, samples, __ATOMIC_RELAXED); |
| 760 | } |
| 761 | } |
| 762 | Word_t metric_id = mrg_metric_id(main_mrg, metric); |
| 763 | |
| 764 | if (update_metric_time) |
| 765 | mrg_metric_expand_retention(main_mrg, metric, vd.start_time_s, vd.end_time_s, vd.update_every_s); |
| 766 | |
| 767 | pgc_open_add_hot_page( |
| 768 | (Word_t)ctx, |
| 769 | metric_id, |
| 770 | vd.start_time_s, |
| 771 | vd.end_time_s, |
| 772 | vd.update_every_s, |
| 773 | journalfile->datafile, |
| 774 | jf_metric_data->extent_offset, |
| 775 | jf_metric_data->extent_size); |
| 776 | |
| 777 | extent_first_time_s = MIN(extent_first_time_s, vd.start_time_s); |
| 778 | extent_last_time_s = MAX(extent_last_time_s, vd.end_time_s); |
| 779 | |
| 780 | mrg_metric_release(main_mrg, metric); |
| 781 | } |
| 782 | |
| 783 | journalfile->v2.first_time_s = extent_first_time_s; |
| 784 | journalfile->v2.last_time_s = extent_last_time_s; |
| 785 | |
| 786 | time_t old = __atomic_load_n(&ctx->atomic.first_time_s, __ATOMIC_RELAXED);; |
| 787 | do { |
| 788 | if(old <= extent_first_time_s) |
| 789 | break; |
| 790 | } while(!__atomic_compare_exchange_n(&ctx->atomic.first_time_s, &old, extent_first_time_s, false, __ATOMIC_RELAXED, __ATOMIC_RELAXED)); |
| 791 | } |
| 792 | |
| 793 | /* |
| 794 | * Replays transaction by interpreting up to max_size bytes from buf. |
| 795 | * Sets id to the current transaction id or to 0 if unknown. |
| 796 | * Returns size of transaction record or 0 for unknown size. |
| 797 | */ |
| 798 | static unsigned journalfile_replay_transaction(struct rrdengine_instance *ctx, struct rrdengine_journalfile *journalfile, |
| 799 | void *buf, uint64_t *id, unsigned max_size) |
| 800 | { |
| 801 | unsigned payload_length, size_bytes; |
| 802 | int ret; |
| 803 | /* persistent structures */ |
| 804 | struct rrdeng_jf_transaction_header *jf_header; |
| 805 | struct rrdeng_jf_transaction_trailer *jf_trailer; |
| 806 | uLong crc; |
| 807 | |
| 808 | *id = 0; |
| 809 | jf_header = buf; |
| 810 | if (STORE_PADDING == jf_header->type) { |
| 811 | netdata_log_debug(D_RRDENGINE, "Skipping padding."); |
| 812 | return 0; |
| 813 | } |
| 814 | if (sizeof(*jf_header) > max_size) { |
| 815 | netdata_log_error("DBENGINE: corrupted transaction record, skipping."); |
| 816 | return 0; |
| 817 | } |
| 818 | *id = jf_header->id; |
| 819 | payload_length = jf_header->payload_length; |
| 820 | size_bytes = sizeof(*jf_header) + payload_length + sizeof(*jf_trailer); |
| 821 | if (size_bytes > max_size) { |
| 822 | netdata_log_error("DBENGINE: corrupted transaction record, skipping."); |
| 823 | return 0; |
| 824 | } |
| 825 | jf_trailer = buf + sizeof(*jf_header) + payload_length; |
| 826 | crc = crc32(0L, Z_NULL, 0); |
| 827 | crc = crc32(crc, buf, sizeof(*jf_header) + payload_length); |
| 828 | ret = crc32cmp(jf_trailer->checksum, crc); |
| 829 | netdata_log_debug(D_RRDENGINE, "Transaction %"PRIu64" was read from disk. CRC32 check: %s", *id, ret ? "FAILED" : "SUCCEEDED"); |
| 830 | if (unlikely(ret)) { |
| 831 | netdata_log_error("DBENGINE: transaction %"PRIu64" was read from disk. CRC32 check: FAILED", *id); |
| 832 | return size_bytes; |
| 833 | } |
| 834 | switch (jf_header->type) { |
| 835 | case STORE_DATA: |
| 836 | netdata_log_debug(D_RRDENGINE, "Replaying transaction %"PRIu64"", jf_header->id); |
| 837 | journalfile_restore_extent_metadata(ctx, journalfile, buf + sizeof(*jf_header), payload_length); |
| 838 | break; |
| 839 | default: |
| 840 | netdata_log_error("DBENGINE: unknown transaction type, skipping record."); |
| 841 | break; |
| 842 | } |
| 843 | |
| 844 | return size_bytes; |
| 845 | } |
| 846 | |
| 847 | |
| 848 | #define READAHEAD_BYTES (RRDENG_BLOCK_SIZE * 256) |
| 849 | /* |
| 850 | * Iterates journal file transactions and populates the page cache. |
| 851 | * Page cache must already be initialized. |
| 852 | * Returns the maximum transaction id it discovered. |
| 853 | */ |
| 854 | static uint64_t journalfile_iterate_transactions(struct rrdengine_instance *ctx, struct rrdengine_journalfile *journalfile) |
| 855 | { |
| 856 | uv_file file; |
| 857 | uint64_t file_size; |
| 858 | int ret; |
| 859 | uint64_t pos, pos_i, max_id, id; |
| 860 | unsigned size_bytes; |
| 861 | void *buf = NULL; |
| 862 | uv_buf_t iov; |
| 863 | uv_fs_t req; |
| 864 | |
| 865 | file = journalfile->file; |
| 866 | file_size = journalfile->unsafe.pos; |
| 867 | |
| 868 | max_id = 1; |
| 869 | (void)posix_memalignz((void *)&buf, RRDFILE_ALIGNMENT, READAHEAD_BYTES); |
| 870 | |
| 871 | for (pos = sizeof(struct rrdeng_jf_sb); pos < file_size; pos += READAHEAD_BYTES) { |
| 872 | size_bytes = MIN(READAHEAD_BYTES, file_size - pos); |
| 873 | iov = uv_buf_init(buf, size_bytes); |
| 874 | ret = uv_fs_read(NULL, &req, file, &iov, 1, pos, NULL); |
| 875 | if (ret < 0) { |
| 876 | netdata_log_error("DBENGINE: uv_fs_read: pos=%" PRIu64 ", %s", pos, uv_strerror(ret)); |
| 877 | uv_fs_req_cleanup(&req); |
| 878 | goto skip_file; |
| 879 | } |
| 880 | fatal_assert(req.result >= 0); |
| 881 | uv_fs_req_cleanup(&req); |
| 882 | ctx_io_read_op_bytes(ctx, size_bytes); |
| 883 | |
| 884 | for (pos_i = 0; pos_i < size_bytes;) { |
| 885 | unsigned max_size; |
| 886 | |
| 887 | max_size = size_bytes - pos_i; |
| 888 | ret = journalfile_replay_transaction(ctx, journalfile, buf + pos_i, &id, max_size); |
| 889 | if (!ret) |
| 890 | /* unknown transaction size, move on to the next block */ |
| 891 | pos_i = ALIGN_BYTES_FLOOR(pos_i + RRDENG_BLOCK_SIZE); |
| 892 | else |
| 893 | pos_i += ret; |
| 894 | max_id = MAX(max_id, id); |
| 895 | } |
| 896 | } |
| 897 | skip_file: |
| 898 | posix_memalign_freez(buf); |
| 899 | return max_id; |
| 900 | } |
| 901 | |
| 902 | // Checks that the extent list checksum is valid |
| 903 | static int journalfile_check_v2_extent_list (void *data_start, size_t file_size) |
| 904 | { |
| 905 | UNUSED(file_size); |
| 906 | uLong crc; |
| 907 | |
| 908 | struct journal_v2_header *j2_header = (void *) data_start; |
| 909 | struct journal_v2_block_trailer *journal_v2_trailer; |
| 910 | |
| 911 | journal_v2_trailer = (struct journal_v2_block_trailer *) ((uint8_t *) data_start + j2_header->extent_trailer_offset); |
| 912 | crc = crc32(0L, Z_NULL, 0); |
| 913 | crc = crc32(crc, (uint8_t *) data_start + j2_header->extent_offset, j2_header->extent_count * sizeof(struct journal_extent_list)); |
| 914 | if (unlikely(crc32cmp(journal_v2_trailer->checksum, crc))) { |
| 915 | netdata_log_error("DBENGINE: extent list CRC32 check: FAILED"); |
| 916 | return 1; |
| 917 | } |
| 918 | |
| 919 | return 0; |
| 920 | } |
| 921 | |
| 922 | // Checks that the metric list (UUIDs) checksum is valid |
| 923 | static int journalfile_check_v2_metric_list(void *data_start, size_t file_size) |
| 924 | { |
| 925 | UNUSED(file_size); |
| 926 | uLong crc; |
| 927 | |
| 928 | struct journal_v2_header *j2_header = (void *) data_start; |
| 929 | struct journal_v2_block_trailer *journal_v2_trailer; |
| 930 | |
| 931 | journal_v2_trailer = (struct journal_v2_block_trailer *) ((uint8_t *) data_start + j2_header->metric_trailer_offset); |
| 932 | crc = crc32(0L, Z_NULL, 0); |
| 933 | crc = crc32(crc, (uint8_t *) data_start + j2_header->metric_offset, j2_header->metric_count * sizeof(struct journal_metric_list)); |
| 934 | if (unlikely(crc32cmp(journal_v2_trailer->checksum, crc))) { |
| 935 | netdata_log_error("DBENGINE: metric list CRC32 check: FAILED"); |
| 936 | return 1; |
| 937 | } |
| 938 | return 0; |
| 939 | } |
| 940 | |
| 941 | // |
| 942 | // Return |
| 943 | // 0 Ok |
| 944 | // 1 Invalid |
| 945 | // 2 Force rebuild |
| 946 | // 3 skip |
| 947 | |
| 948 | static int journalfile_v2_validate(void *data_start, size_t journal_v2_file_size, size_t journal_v1_file_size) |
| 949 | { |
| 950 | int rc; |
| 951 | uLong crc; |
| 952 | |
| 953 | struct journal_v2_header *j2_header = (void *) data_start; |
| 954 | struct journal_v2_block_trailer *journal_v2_trailer; |
| 955 | |
| 956 | if (j2_header->magic == JOURVAL_V2_REBUILD_MAGIC) |
| 957 | return 2; |
| 958 | |
| 959 | if (j2_header->magic == JOURVAL_V2_SKIP_MAGIC) |
| 960 | return 3; |
| 961 | |
| 962 | // Magic failure |
| 963 | if (j2_header->magic != JOURVAL_V2_MAGIC) |
| 964 | return 1; |
| 965 | |
| 966 | if (j2_header->journal_v2_file_size != journal_v2_file_size) |
| 967 | return 1; |
| 968 | |
| 969 | if (journal_v1_file_size && j2_header->journal_v1_file_size != journal_v1_file_size) |
| 970 | return 1; |
| 971 | |
| 972 | journal_v2_trailer = (struct journal_v2_block_trailer *) ((uint8_t *) data_start + journal_v2_file_size - sizeof(*journal_v2_trailer)); |
| 973 | |
| 974 | crc = crc32(0L, Z_NULL, 0); |
| 975 | crc = crc32(crc, (void *) j2_header, sizeof(*j2_header)); |
| 976 | |
| 977 | rc = crc32cmp(journal_v2_trailer->checksum, crc); |
| 978 | if (unlikely(rc)) { |
| 979 | netdata_log_error("DBENGINE: file CRC32 check: FAILED"); |
| 980 | return 1; |
| 981 | } |
| 982 | |
| 983 | rc = journalfile_check_v2_extent_list(data_start, journal_v2_file_size); |
| 984 | if (rc) return 1; |
| 985 | |
| 986 | if (!db_engine_journal_check) |
| 987 | return 0; |
| 988 | |
| 989 | rc = journalfile_check_v2_metric_list(data_start, journal_v2_file_size); |
| 990 | if (rc) return 1; |
| 991 | |
| 992 | // Verify complete UUID chain |
| 993 | |
| 994 | struct journal_metric_list *metric = (void *) (data_start + j2_header->metric_offset); |
| 995 | |
| 996 | unsigned verified = 0; |
| 997 | unsigned entries; |
| 998 | unsigned total_pages = 0; |
| 999 | |
| 1000 | netdata_log_info("DBENGINE: checking %u metrics that exist in the journal", j2_header->metric_count); |
| 1001 | for (entries = 0; entries < j2_header->metric_count; entries++) { |
| 1002 | |
| 1003 | char uuid_str[UUID_STR_LEN]; |
| 1004 | uuid_unparse_lower(metric->uuid, uuid_str); |
| 1005 | struct journal_page_header *metric_list_header = (void *) (data_start + metric->page_offset); |
| 1006 | struct journal_page_header local_metric_list_header = *metric_list_header; |
| 1007 | |
| 1008 | local_metric_list_header.crc = JOURVAL_V2_MAGIC; |
| 1009 | |
| 1010 | crc = crc32(0L, Z_NULL, 0); |
| 1011 | crc = crc32(crc, (void *) &local_metric_list_header, sizeof(local_metric_list_header)); |
| 1012 | rc = crc32cmp(metric_list_header->checksum, crc); |
| 1013 | |
| 1014 | if (!rc) { |
| 1015 | struct journal_v2_block_trailer *journal_trailer = |
| 1016 | (void *) data_start + metric->page_offset + sizeof(struct journal_page_header) + (metric_list_header->entries * sizeof(struct journal_page_list)); |
| 1017 | |
| 1018 | crc = crc32(0L, Z_NULL, 0); |
| 1019 | crc = crc32(crc, (uint8_t *) metric_list_header + sizeof(struct journal_page_header), metric_list_header->entries * sizeof(struct journal_page_list)); |
| 1020 | rc = crc32cmp(journal_trailer->checksum, crc); |
| 1021 | internal_error(rc, "DBENGINE: index %u : %s entries %u at offset %u verified, DATA CRC computed %lu, stored %u", entries, uuid_str, metric->entries, metric->page_offset, |
| 1022 | crc, metric_list_header->crc); |
| 1023 | if (!rc) { |
| 1024 | total_pages += metric_list_header->entries; |
| 1025 | verified++; |
| 1026 | } |
| 1027 | } |
| 1028 | |
| 1029 | metric++; |
| 1030 | if ((uint32_t)((uint8_t *) metric - (uint8_t *) data_start) > (uint32_t) journal_v2_file_size) { |
| 1031 | netdata_log_info("DBENGINE: verification failed EOF reached -- total entries %u, verified %u", entries, verified); |
| 1032 | return 1; |
| 1033 | } |
| 1034 | } |
| 1035 | |
| 1036 | if (entries != verified) { |
| 1037 | netdata_log_info("DBENGINE: verification failed -- total entries %u, verified %u", entries, verified); |
| 1038 | return 1; |
| 1039 | } |
| 1040 | netdata_log_info("DBENGINE: verification succeeded -- total entries %u, verified %u (%u total pages)", entries, verified, total_pages); |
| 1041 | |
| 1042 | return 0; |
| 1043 | } |
| 1044 | |
| 1045 | void journalfile_v2_populate_retention_to_mrg(struct rrdengine_instance *ctx, struct rrdengine_journalfile *journalfile) { |
| 1046 | usec_t started_ut = now_monotonic_usec(); |
| 1047 | |
| 1048 | size_t data_size = 0; |
| 1049 | struct journal_v2_header *j2_header = journalfile_v2_data_acquire_with_hint( |
| 1050 | journalfile, &data_size, 0, 0, JOURNALFILE_V2_ACCESS_SEQUENTIAL_DIRECTORY); |
| 1051 | if(!j2_header) |
| 1052 | return; |
| 1053 | |
| 1054 | uint8_t *data_start = (uint8_t *)j2_header; |
| 1055 | |
| 1056 | char path_v2[RRDENG_PATH_MAX]; |
| 1057 | journalfile_v2_generate_path(journalfile->datafile, path_v2, sizeof(path_v2)); |
| 1058 | time_t global_first_time_s = 0; |
| 1059 | bool failed = false; |
| 1060 | uint32_t entries = 0; |
| 1061 | // Calculate number of samples here and update once the file is loaded |
| 1062 | uint64_t journal_samples = 0; |
| 1063 | |
| 1064 | // Protect the whole walk -- both the optional CRC check and the mrg update |
| 1065 | // read into the mmap'd v2 journal. If a backing page cannot be paged in |
| 1066 | // (file truncated, sparse hole, transient I/O error) the kernel raises |
| 1067 | // SIGBUS; without a protected region active the process aborts. Same |
| 1068 | // pattern as journalfile_v2_validate() in journalfile_v2_load(). |
| 1069 | PROTECTED_ACCESS_SETUP(journalfile->mmap.data, journalfile->mmap.size, path_v2, "mrg-load"); |
| 1070 | if(no_signal_received) { |
| 1071 | // Validate header-controlled offsets against the actual mapping size |
| 1072 | // BEFORE reading through them. PROTECTED_ACCESS_SETUP only catches |
| 1073 | // faults whose address falls within [data_start, data_start+mmap.size); |
| 1074 | // an over-read driven by a corrupted offset past the registered range |
| 1075 | // would not be recovered and the process would still abort. Out-of- |
| 1076 | // range offsets are treated as a rebuild trigger (same outcome as a |
| 1077 | // CRC failure). |
| 1078 | size_t mmap_size = journalfile->mmap.size; |
| 1079 | // Order matters: short-circuit on metric_offset > mmap_size first so the |
| 1080 | // subtraction below cannot underflow, then compare metric_count against |
| 1081 | // the available slot capacity via division rather than multiplying out |
| 1082 | // metric_count * sizeof(...), which would wrap size_t on 32-bit builds |
| 1083 | // and silently pass a malformed header. Only after those two checks is |
| 1084 | // metric_count * sizeof(...) known to fit in size_t, which lets the |
| 1085 | // trailer-ordering check below compute the expected trailer offset |
| 1086 | // safely. The on-disk layout (see journalfile_migrate_to_v2_callback) |
| 1087 | // places the metric-list trailer immediately after the metric list, so |
| 1088 | // any deviation indicates a corrupted header that would otherwise let |
| 1089 | // the CRC compare against bytes inside the metric list itself. |
| 1090 | if ((size_t)j2_header->metric_offset > mmap_size || |
| 1091 | (size_t)j2_header->metric_count > (mmap_size - (size_t)j2_header->metric_offset) / sizeof(struct journal_metric_list) || |
| 1092 | (size_t)j2_header->metric_trailer_offset > mmap_size || |
| 1093 | mmap_size - (size_t)j2_header->metric_trailer_offset < sizeof(struct journal_v2_block_trailer) || |
| 1094 | (size_t)j2_header->metric_trailer_offset != (size_t)j2_header->metric_offset + (size_t)j2_header->metric_count * sizeof(struct journal_metric_list)) { |
| 1095 | // header offsets out of range -- needs rebuild |
| 1096 | nd_log_daemon(NDLP_ERR, |
| 1097 | "DBENGINE: journal v2 \"%s\" has out-of-range header offsets " |
| 1098 | "(metric_offset=%u, metric_count=%u, metric_trailer_offset=%u, mmap_size=%zu); " |
| 1099 | "marking unavailable for rebuild", |
| 1100 | path_v2, |
| 1101 | j2_header->metric_offset, |
| 1102 | j2_header->metric_count, |
| 1103 | j2_header->metric_trailer_offset, |
| 1104 | mmap_size); |
| 1105 | failed = true; |
| 1106 | } |
| 1107 | else if (journalfile->v2.flags & JOURNALFILE_FLAG_METRIC_CRC_CHECK) { |
| 1108 | journalfile->v2.flags &= ~JOURNALFILE_FLAG_METRIC_CRC_CHECK; |
| 1109 | // Pass the verified mmap_size, not the header-controlled |
| 1110 | // j2_header->journal_v2_file_size; the helper currently ignores the |
| 1111 | // size argument (UNUSED) but the value at the call site should |
| 1112 | // still reflect the trusted bound for clarity and future-proofing. |
| 1113 | if (journalfile_check_v2_metric_list(data_start, mmap_size)) { |
| 1114 | // needs rebuild |
| 1115 | failed = true; |
| 1116 | } |
| 1117 | } |
| 1118 | |
| 1119 | if (!failed) { |
| 1120 | entries = j2_header->metric_count; |
| 1121 | struct journal_metric_list *metric = (struct journal_metric_list *) (data_start + j2_header->metric_offset); |
| 1122 | time_t header_start_time_s = (time_t) (j2_header->start_time_ut / USEC_PER_SEC); |
| 1123 | global_first_time_s = header_start_time_s; |
| 1124 | time_t now_s = max_acceptable_collected_time(); |
| 1125 | for (size_t i=0; i < entries; i++) { |
| 1126 | // Copy uuid out of the mmap onto the stack BEFORE calling mrg. |
| 1127 | // If a backing page is unreadable, uuid_copy SIGBUSes here and |
| 1128 | // the protected region recovers cleanly; the mrg call then |
| 1129 | // never executes. If we passed &metric->uuid into mrg, a |
| 1130 | // SIGBUS could fire INSIDE mrg while it holds internal locks, |
| 1131 | // and siglongjmp would skip mrg's unlock paths. |
| 1132 | nd_uuid_t local_uuid; |
| 1133 | uuid_copy(local_uuid, metric->uuid); |
| 1134 | time_t start_time_s = header_start_time_s + metric->delta_start_s; |
| 1135 | time_t end_time_s = header_start_time_s + metric->delta_end_s; |
| 1136 | uint32_t update_every_s = metric->update_every_s; |
| 1137 | |
| 1138 | mrg_update_metric_retention_and_granularity_by_uuid( |
| 1139 | main_mrg, |
| 1140 | (Word_t)ctx, |
| 1141 | &local_uuid, |
| 1142 | start_time_s, |
| 1143 | end_time_s, |
| 1144 | update_every_s, |
| 1145 | now_s, |
| 1146 | &journal_samples); |
| 1147 | |
| 1148 | metric++; |
| 1149 | } |
| 1150 | } |
| 1151 | } |
| 1152 | else { |
| 1153 | // SIGBUS/SIGSEGV inside the mmap walk. The PROTECTED_ACCESS_SETUP |
| 1154 | // macro already rate-limits the error log. |
| 1155 | failed = true; |
| 1156 | } |
| 1157 | |
| 1158 | if (unlikely(failed)) { |
| 1159 | // Clear IS_AVAILABLE under the data spinlock BEFORE releasing our |
| 1160 | // refcount, so no concurrent journalfile_v2_data_acquire_with_hint() |
| 1161 | // can succeed and walk the (potentially corrupted) mmap between here |
| 1162 | // and the permanent unmap. The bounds, CRC, and signal-recovery paths |
| 1163 | // all converge through this single transition; without it, teardown |
| 1164 | // (journalfile_close / journalfile_destroy_unsafe) would gate on |
| 1165 | // IS_AVAILABLE and skip journalfile_v2_data_unmap_permanently(), |
| 1166 | // leaving a dangling njfv2idx entry, an unclosed fd, and an unmapped |
| 1167 | // region. |
| 1168 | spinlock_lock(&journalfile->data_spinlock); |
| 1169 | journalfile->v2.flags &= ~JOURNALFILE_FLAG_IS_AVAILABLE; |
| 1170 | spinlock_unlock(&journalfile->data_spinlock); |
| 1171 | } |
| 1172 | |
| 1173 | journalfile_v2_data_release(journalfile); |
| 1174 | |
| 1175 | if (unlikely(failed)) { |
| 1176 | journalfile_v2_data_unmap_permanently(journalfile); |
| 1177 | return; |
| 1178 | } |
| 1179 | |
| 1180 | __atomic_add_fetch(&ctx->atomic.samples, journal_samples, __ATOMIC_RELAXED); |
| 1181 | |
| 1182 | usec_t ended_ut = now_monotonic_usec(); |
| 1183 | |
| 1184 | nd_log_daemon(NDLP_DEBUG, "DBENGINE: journal v2 of tier %d, datafile %u populated, size: %0.2f MiB, metrics: %0.2f k, %0.2f ms" |
| 1185 | , ctx->config.tier, journalfile->datafile->fileno |
| 1186 | , (double)data_size / 1024 / 1024 |
| 1187 | , (double)entries / 1000 |
| 1188 | , ((double)(ended_ut - started_ut) / USEC_PER_MS) |
| 1189 | ); |
| 1190 | |
| 1191 | time_t old = __atomic_load_n(&ctx->atomic.first_time_s, __ATOMIC_RELAXED);; |
| 1192 | do { |
| 1193 | if(old <= global_first_time_s) |
| 1194 | break; |
| 1195 | } while(!__atomic_compare_exchange_n(&ctx->atomic.first_time_s, &old, global_first_time_s, false, __ATOMIC_RELAXED, __ATOMIC_RELAXED)); |
| 1196 | } |
| 1197 | |
| 1198 | int journalfile_v2_load(struct rrdengine_instance *ctx, struct rrdengine_journalfile *journalfile, struct rrdengine_datafile *datafile) |
| 1199 | { |
| 1200 | int ret, fd; |
| 1201 | char path_v1[RRDENG_PATH_MAX]; |
| 1202 | char path_v2[RRDENG_PATH_MAX]; |
| 1203 | struct stat statbuf; |
| 1204 | size_t journal_v1_file_size = 0; |
| 1205 | size_t journal_v2_file_size; |
| 1206 | |
| 1207 | journalfile_v1_generate_path(datafile, path_v1, sizeof(path_v1)); |
| 1208 | ret = stat(path_v1, &statbuf); |
| 1209 | if (!ret) |
| 1210 | journal_v1_file_size = (uint32_t)statbuf.st_size; |
| 1211 | |
| 1212 | journalfile_v2_generate_path(datafile, path_v2, sizeof(path_v2)); |
| 1213 | fd = open(path_v2, O_RDONLY | O_CLOEXEC); |
| 1214 | if (fd < 0) { |
| 1215 | if (errno == ENOENT) |
| 1216 | return 1; |
| 1217 | ctx_fs_error(ctx); |
| 1218 | netdata_log_error("DBENGINE: failed to open \"%s\"", path_v2); |
| 1219 | return 1; |
| 1220 | } |
| 1221 | |
| 1222 | ret = fstat(fd, &statbuf); |
| 1223 | if (ret) { |
| 1224 | netdata_log_error("DBENGINE: failed to get file information for \"%s\"", path_v2); |
| 1225 | close(fd); |
| 1226 | return 1; |
| 1227 | } |
| 1228 | |
| 1229 | journal_v2_file_size = (size_t)statbuf.st_size; |
| 1230 | |
| 1231 | if (journal_v2_file_size < sizeof(struct journal_v2_header)) { |
| 1232 | error_report("Invalid file \"%s\". Not the expected size", path_v2); |
| 1233 | close(fd); |
| 1234 | return 1; |
| 1235 | } |
| 1236 | |
| 1237 | usec_t mmap_start_ut = now_monotonic_usec(); |
| 1238 | uint8_t *data_start = nd_mmap(NULL, journal_v2_file_size, PROT_READ, MAP_SHARED, fd, 0); |
| 1239 | if (data_start == MAP_FAILED) { |
| 1240 | close(fd); |
| 1241 | return 1; |
| 1242 | } |
| 1243 | |
| 1244 | // validation reads the file sequentially — hint the kernel to prefetch |
| 1245 | madvise_sequential(data_start, journal_v2_file_size); |
| 1246 | madvise_willneed(data_start, journal_v2_file_size); |
| 1247 | |
| 1248 | nd_log_daemon(NDLP_DEBUG, "DBENGINE: checking integrity of \"%s\"", path_v2); |
| 1249 | |
| 1250 | usec_t validation_start_ut = now_monotonic_usec(); |
| 1251 | |
| 1252 | int rc = 0; |
| 1253 | PROTECTED_ACCESS_SETUP(data_start, journal_v2_file_size, path_v2, "validate"); |
| 1254 | if(no_signal_received) { |
| 1255 | rc = journalfile_v2_validate(data_start, journal_v2_file_size, journal_v1_file_size); |
| 1256 | } |
| 1257 | else { |
| 1258 | rc = 2; |
| 1259 | } |
| 1260 | |
| 1261 | if (unlikely(rc)) { |
| 1262 | if (rc == 2) |
| 1263 | error_report("File \"%s\" needs to be rebuilt", path_v2); |
| 1264 | else if (rc == 3) |
| 1265 | error_report("File \"%s\" will be skipped", path_v2); |
| 1266 | else |
| 1267 | error_report("File \"%s\" is invalid and it will be rebuilt", path_v2); |
| 1268 | |
| 1269 | if (unlikely(nd_munmap(data_start, journal_v2_file_size))) |
| 1270 | netdata_log_error("DBENGINE: failed to unmap \"%s\"", path_v2); |
| 1271 | |
| 1272 | close(fd); |
| 1273 | return rc; |
| 1274 | } |
| 1275 | |
| 1276 | struct journal_v2_header *j2_header = (void *) data_start; |
| 1277 | uint32_t entries = j2_header->metric_count; |
| 1278 | |
| 1279 | if (unlikely(!entries)) { |
| 1280 | if (unlikely(nd_munmap(data_start, journal_v2_file_size))) |
| 1281 | netdata_log_error("DBENGINE: failed to unmap \"%s\"", path_v2); |
| 1282 | |
| 1283 | close(fd); |
| 1284 | return 1; |
| 1285 | } |
| 1286 | |
| 1287 | usec_t finished_ut = now_monotonic_usec(); |
| 1288 | |
| 1289 | nd_log_daemon(NDLP_DEBUG, "DBENGINE: journal v2 \"%s\" loaded, size: %0.2f MiB, metrics: %0.2f k, " |
| 1290 | "mmap: %0.2f ms, validate: %0.2f ms" |
| 1291 | , path_v2 |
| 1292 | , (double)journal_v2_file_size / 1024 / 1024 |
| 1293 | , (double)entries / 1000 |
| 1294 | , ((double)(validation_start_ut - mmap_start_ut) / USEC_PER_MS) |
| 1295 | , ((double)(finished_ut - validation_start_ut) / USEC_PER_MS) |
| 1296 | ); |
| 1297 | |
| 1298 | // Initialize the journal file to be able to access the data |
| 1299 | |
| 1300 | if (!db_engine_journal_check) |
| 1301 | journalfile->v2.flags |= JOURNALFILE_FLAG_METRIC_CRC_CHECK; |
| 1302 | |
| 1303 | journalfile_v2_data_set(journalfile, fd, data_start, journal_v2_file_size); |
| 1304 | |
| 1305 | ctx_current_disk_space_increase(ctx, journal_v2_file_size); |
| 1306 | |
| 1307 | // File is OK load it |
| 1308 | return 0; |
| 1309 | } |
| 1310 | |
| 1311 | struct journal_metric_list_to_sort { |
| 1312 | struct jv2_metrics_info *metric_info; |
| 1313 | }; |
| 1314 | |
| 1315 | static int journalfile_metric_compare (const void *item1, const void *item2) |
| 1316 | { |
| 1317 | const struct jv2_metrics_info *metric1 = ((struct journal_metric_list_to_sort *) item1)->metric_info; |
| 1318 | const struct jv2_metrics_info *metric2 = ((struct journal_metric_list_to_sort *) item2)->metric_info; |
| 1319 | |
| 1320 | return memcmp(metric1->uuid, metric2->uuid, sizeof(nd_uuid_t)); |
| 1321 | } |
| 1322 | |
| 1323 | |
| 1324 | // Write list of extents for the journalfile |
| 1325 | void *journalfile_v2_write_extent_list(Pvoid_t JudyL_extents_pos, void *data) |
| 1326 | { |
| 1327 | Pvoid_t *PValue; |
| 1328 | struct journal_extent_list *j2_extent_base = (void *) data; |
| 1329 | struct jv2_extents_info *ext_info; |
| 1330 | |
| 1331 | bool first = true; |
| 1332 | Word_t pos = 0; |
| 1333 | size_t count = 0; |
| 1334 | while ((PValue = JudyLFirstThenNext(JudyL_extents_pos, &pos, &first))) { |
| 1335 | ext_info = *PValue; |
| 1336 | size_t index = ext_info->index; |
| 1337 | j2_extent_base[index].file_index = 0; |
| 1338 | j2_extent_base[index].datafile_offset = BLOCK_TO_OFFSET(ext_info->block); |
| 1339 | j2_extent_base[index].datafile_size = ext_info->bytes; |
| 1340 | j2_extent_base[index].pages = ext_info->number_of_pages; |
| 1341 | count++; |
| 1342 | } |
| 1343 | return j2_extent_base + count; |
| 1344 | } |
| 1345 | |
| 1346 | static int journalfile_verify_space(struct journal_v2_header *j2_header, void *data, uint32_t bytes) |
| 1347 | { |
| 1348 | if ((unsigned long)(((uint8_t *) data - (uint8_t *) j2_header->data) + bytes) > (j2_header->journal_v2_file_size - sizeof(struct journal_v2_block_trailer))) |
| 1349 | return 1; |
| 1350 | |
| 1351 | return 0; |
| 1352 | } |
| 1353 | |
| 1354 | void *journalfile_v2_write_metric_page(struct journal_v2_header *j2_header, void *data, struct jv2_metrics_info *metric_info, uint32_t pages_offset) |
| 1355 | { |
| 1356 | struct journal_metric_list *metric = (void *) data; |
| 1357 | |
| 1358 | if (journalfile_verify_space(j2_header, data, sizeof(*metric))) |
| 1359 | return NULL; |
| 1360 | |
| 1361 | uuid_copy(metric->uuid, *metric_info->uuid); |
| 1362 | metric->entries = metric_info->number_of_pages; |
| 1363 | metric->page_offset = pages_offset; |
| 1364 | metric->delta_start_s = (uint32_t)(metric_info->first_time_s - (time_t)(j2_header->start_time_ut / USEC_PER_SEC)); |
| 1365 | metric->delta_end_s = (uint32_t)(metric_info->last_time_s - (time_t)(j2_header->start_time_ut / USEC_PER_SEC)); |
| 1366 | metric->update_every_s = 0; |
| 1367 | |
| 1368 | return ++metric; |
| 1369 | } |
| 1370 | |
| 1371 | void *journalfile_v2_write_data_page_header(struct journal_v2_header *j2_header __maybe_unused, void *data, struct jv2_metrics_info *metric_info, uint32_t uuid_offset) |
| 1372 | { |
| 1373 | struct journal_page_header *data_page_header = (void *) data; |
| 1374 | uLong crc; |
| 1375 | |
| 1376 | uuid_copy(data_page_header->uuid, *metric_info->uuid); |
| 1377 | data_page_header->entries = metric_info->number_of_pages; |
| 1378 | data_page_header->uuid_offset = uuid_offset; // data header OFFSET poings to METRIC in the directory |
| 1379 | data_page_header->crc = JOURVAL_V2_MAGIC; |
| 1380 | crc = crc32(0L, Z_NULL, 0); |
| 1381 | crc = crc32(crc, (void *) data_page_header, sizeof(*data_page_header)); |
| 1382 | crc32set(data_page_header->checksum, crc); |
| 1383 | return ++data_page_header; |
| 1384 | } |
| 1385 | |
| 1386 | void *journalfile_v2_write_data_page_trailer(struct journal_v2_header *j2_header __maybe_unused, void *data, void *page_header) |
| 1387 | { |
| 1388 | struct journal_page_header *data_page_header = (void *) page_header; |
| 1389 | struct journal_v2_block_trailer *journal_trailer = (void *) data; |
| 1390 | uLong crc; |
| 1391 | |
| 1392 | crc = crc32(0L, Z_NULL, 0); |
| 1393 | crc = crc32(crc, (uint8_t *) page_header + sizeof(struct journal_page_header), data_page_header->entries * sizeof(struct journal_page_list)); |
| 1394 | crc32set(journal_trailer->checksum, crc); |
| 1395 | return ++journal_trailer; |
| 1396 | } |
| 1397 | |
| 1398 | void *journalfile_v2_write_data_page(struct journal_v2_header *j2_header, void *data, struct jv2_page_info *page_info) |
| 1399 | { |
| 1400 | struct journal_page_list *data_page = data; |
| 1401 | |
| 1402 | if (journalfile_verify_space(j2_header, data, sizeof(*data_page))) |
| 1403 | return NULL; |
| 1404 | |
| 1405 | data_page->delta_start_s = (uint32_t) (page_info->start_time_s - (time_t) (j2_header->start_time_ut) / USEC_PER_SEC); |
| 1406 | data_page->delta_end_s = (uint32_t) (page_info->end_time_s - (time_t) (j2_header->start_time_ut) / USEC_PER_SEC); |
| 1407 | data_page->extent_index = page_info->extent_index; |
| 1408 | |
| 1409 | data_page->update_every_s = page_info->update_every_s; |
| 1410 | data_page->page_length = 0; |
| 1411 | data_page->type = 0; |
| 1412 | |
| 1413 | return ++data_page; |
| 1414 | } |
| 1415 | |
| 1416 | // Must be recorded in metric_info->entries |
| 1417 | static void *journalfile_v2_write_descriptors(struct journal_v2_header *j2_header, void *data, struct jv2_metrics_info *metric_info, |
| 1418 | struct journal_metric_list *current_metric) |
| 1419 | { |
| 1420 | Pvoid_t *PValue; |
| 1421 | |
| 1422 | struct journal_page_list *data_page = (void *)data; |
| 1423 | // We need to write all descriptors with index metric_info->min_index_time_s, metric_info->max_index_time_s |
| 1424 | // that belong to this journal file |
| 1425 | Pvoid_t JudyL_array = metric_info->JudyL_pages_by_start_time; |
| 1426 | |
| 1427 | Word_t index_time = 0; |
| 1428 | bool first = true; |
| 1429 | struct jv2_page_info *page_info; |
| 1430 | uint32_t update_every_s = 0; |
| 1431 | while ((PValue = JudyLFirstThenNext(JudyL_array, &index_time, &first))) { |
| 1432 | page_info = *PValue; |
| 1433 | // Write one descriptor and return the next data page location |
| 1434 | data_page = journalfile_v2_write_data_page(j2_header, (void *) data_page, page_info); |
| 1435 | update_every_s = page_info->update_every_s; |
| 1436 | if (NULL == data_page) |
| 1437 | break; |
| 1438 | } |
| 1439 | current_metric->update_every_s = update_every_s; |
| 1440 | return data_page; |
| 1441 | } |
| 1442 | |
| 1443 | // Migrate the journalfile pointed by datafile |
| 1444 | // activate : make the new file active immediately |
| 1445 | // journafile data will be set and descriptors (if deleted) will be repopulated as needed |
| 1446 | // startup : if the migration is done during agent startup |
| 1447 | // this will allow us to optimize certain things |
| 1448 | |
| 1449 | bool journalfile_migrate_to_v2_callback(Word_t section, unsigned datafile_fileno __maybe_unused, uint8_t type __maybe_unused, |
| 1450 | Pvoid_t JudyL_metrics, Pvoid_t JudyL_extents_pos, |
| 1451 | size_t number_of_extents, size_t number_of_metrics, size_t number_of_pages, void *user_data) |
| 1452 | { |
| 1453 | // Nothing to migrate if no metrics |
| 1454 | if (number_of_metrics == 0) |
| 1455 | return true; |
| 1456 | |
| 1457 | char path[RRDENG_PATH_MAX]; |
| 1458 | Pvoid_t *PValue; |
| 1459 | struct rrdengine_instance *ctx = (struct rrdengine_instance *) section; |
| 1460 | struct rrdengine_journalfile *journalfile = (struct rrdengine_journalfile *) user_data; |
| 1461 | struct rrdengine_datafile *datafile = journalfile->datafile; |
| 1462 | time_t min_time_s = LONG_MAX; |
| 1463 | time_t max_time_s = 0; |
| 1464 | struct jv2_metrics_info *metric_info; |
| 1465 | |
| 1466 | journalfile_v2_generate_path(datafile, path, sizeof(path)); |
| 1467 | |
| 1468 | netdata_log_info("DBENGINE: tier %d: indexing " WALFILE_PREFIX RRDENG_FILE_NUMBER_PRINT_TMPL WALFILE_EXTENSION_V2 ": extents %zu, metrics %zu, pages %zu", |
| 1469 | ctx->config.tier, datafile->tier, datafile->fileno, |
| 1470 | number_of_extents, |
| 1471 | number_of_metrics, |
| 1472 | number_of_pages); |
| 1473 | |
| 1474 | #ifdef NETDATA_INTERNAL_CHECKS |
| 1475 | usec_t start_loading = now_monotonic_usec(); |
| 1476 | #endif |
| 1477 | |
| 1478 | size_t total_file_size = 0; |
| 1479 | total_file_size += (sizeof(struct journal_v2_header) + JOURNAL_V2_HEADER_PADDING_SZ); |
| 1480 | |
| 1481 | // Extents will start here |
| 1482 | uint32_t extent_offset = total_file_size; |
| 1483 | total_file_size += (number_of_extents * sizeof(struct journal_extent_list)); |
| 1484 | |
| 1485 | uint32_t extent_offset_trailer = total_file_size; |
| 1486 | total_file_size += sizeof(struct journal_v2_block_trailer); |
| 1487 | |
| 1488 | // UUID list will start here |
| 1489 | uint32_t metrics_offset = total_file_size; |
| 1490 | total_file_size += (number_of_metrics * sizeof(struct journal_metric_list)); |
| 1491 | |
| 1492 | // UUID list trailer |
| 1493 | uint32_t metric_offset_trailer = total_file_size; |
| 1494 | total_file_size += sizeof(struct journal_v2_block_trailer); |
| 1495 | |
| 1496 | // descr @ time will start here |
| 1497 | uint32_t pages_offset = total_file_size; |
| 1498 | total_file_size += (number_of_pages * (sizeof(struct journal_page_list) + sizeof(struct journal_page_header) + sizeof(struct journal_v2_block_trailer))); |
| 1499 | |
| 1500 | // File trailer |
| 1501 | uint32_t trailer_offset = total_file_size; |
| 1502 | total_file_size += sizeof(struct journal_v2_block_trailer); |
| 1503 | |
| 1504 | int fd_v2 = -1; |
| 1505 | uint8_t *data_start = nd_mmap_advanced(path, total_file_size, MAP_SHARED, 0, false, true, &fd_v2); |
| 1506 | if(!data_start) { |
| 1507 | if(fd_v2 != -1) |
| 1508 | close(fd_v2); |
| 1509 | nd_log_daemon(NDLP_WARNING, "DBENGINE: Failed to allocate %"PRIu64" bytes of memory for journal file \"%s\". Will retry later", total_file_size, path); |
| 1510 | return false; |
| 1511 | } |
| 1512 | |
| 1513 | struct journal_metric_list_to_sort *uuid_list = NULL; |
| 1514 | |
| 1515 | PROTECTED_ACCESS_SETUP(data_start, total_file_size, path, "migrate"); |
| 1516 | if(no_signal_received) { |
| 1517 | fatal_assert(extent_offset <= total_file_size); |
| 1518 | memset(data_start, 0, extent_offset); |
| 1519 | |
| 1520 | // Write header |
| 1521 | struct journal_v2_header j2_header; |
| 1522 | memset(&j2_header, 0, sizeof(j2_header)); |
| 1523 | |
| 1524 | j2_header.magic = JOURVAL_V2_MAGIC; |
| 1525 | j2_header.start_time_ut = 0; |
| 1526 | j2_header.end_time_ut = 0; |
| 1527 | j2_header.extent_count = number_of_extents; |
| 1528 | j2_header.extent_offset = extent_offset; |
| 1529 | j2_header.metric_count = number_of_metrics; |
| 1530 | j2_header.metric_offset = metrics_offset; |
| 1531 | j2_header.page_count = number_of_pages; |
| 1532 | j2_header.page_offset = pages_offset; |
| 1533 | j2_header.extent_trailer_offset = extent_offset_trailer; |
| 1534 | j2_header.metric_trailer_offset = metric_offset_trailer; |
| 1535 | j2_header.journal_v2_file_size = total_file_size; |
| 1536 | j2_header.journal_v1_file_size = (uint32_t)journalfile_current_size(journalfile); |
| 1537 | j2_header.data = data_start; // Used during migration |
| 1538 | |
| 1539 | struct journal_v2_block_trailer *journal_v2_trailer; |
| 1540 | |
| 1541 | uint8_t *data = journalfile_v2_write_extent_list(JudyL_extents_pos, data_start + extent_offset); |
| 1542 | internal_error( |
| 1543 | true, "DBENGINE: write extent list so far %llu", (now_monotonic_usec() - start_loading) / USEC_PER_MS); |
| 1544 | |
| 1545 | fatal_assert(data == data_start + extent_offset_trailer); |
| 1546 | |
| 1547 | // Calculate CRC for extents |
| 1548 | journal_v2_trailer = (struct journal_v2_block_trailer *)(data_start + extent_offset_trailer); |
| 1549 | uLong crc; |
| 1550 | crc = crc32(0L, Z_NULL, 0); |
| 1551 | crc = crc32(crc, (uint8_t *)data_start + extent_offset, number_of_extents * sizeof(struct journal_extent_list)); |
| 1552 | crc32set(journal_v2_trailer->checksum, crc); |
| 1553 | |
| 1554 | internal_error( |
| 1555 | true, "DBENGINE: CALCULATE CRC FOR EXTENT %llu", (now_monotonic_usec() - start_loading) / USEC_PER_MS); |
| 1556 | // Skip the trailer, point to the metrics off |
| 1557 | data += sizeof(struct journal_v2_block_trailer); |
| 1558 | |
| 1559 | // Sanity check -- we must be at the metrics_offset |
| 1560 | fatal_assert(data == data_start + metrics_offset); |
| 1561 | |
| 1562 | // Allocate array to sort UUIDs and keep them sorted in the journal because we want to do binary search when we do lookups |
| 1563 | uuid_list = mallocz(number_of_metrics * sizeof(struct journal_metric_list_to_sort)); |
| 1564 | |
| 1565 | Word_t Index = 0; |
| 1566 | size_t count = 0; |
| 1567 | bool first_then_next = true; |
| 1568 | while ((PValue = JudyLFirstThenNext(JudyL_metrics, &Index, &first_then_next))) { |
| 1569 | metric_info = *PValue; |
| 1570 | |
| 1571 | fatal_assert(metric_info != NULL); |
| 1572 | fatal_assert(count < number_of_metrics); |
| 1573 | uuid_list[count++].metric_info = metric_info; |
| 1574 | min_time_s = MIN(min_time_s, metric_info->first_time_s); |
| 1575 | max_time_s = MAX(max_time_s, metric_info->last_time_s); |
| 1576 | } |
| 1577 | |
| 1578 | fatal_assert(count == number_of_metrics); |
| 1579 | |
| 1580 | // Check if not properly set in the loop above to prevent overflow |
| 1581 | if (min_time_s == LONG_MAX) |
| 1582 | min_time_s = 0; |
| 1583 | |
| 1584 | // Store in the header |
| 1585 | j2_header.start_time_ut = min_time_s * USEC_PER_SEC; |
| 1586 | j2_header.end_time_ut = max_time_s * USEC_PER_SEC; |
| 1587 | |
| 1588 | qsort(&uuid_list[0], number_of_metrics, sizeof(struct journal_metric_list_to_sort), journalfile_metric_compare); |
| 1589 | internal_error( |
| 1590 | true, "DBENGINE: traverse and qsort UUID %llu", (now_monotonic_usec() - start_loading) / USEC_PER_MS); |
| 1591 | |
| 1592 | for (Index = 0; Index < number_of_metrics; Index++) { |
| 1593 | metric_info = uuid_list[Index].metric_info; |
| 1594 | |
| 1595 | // Calculate current UUID offset from start of file. We will store this in the data page header |
| 1596 | uint32_t uuid_offset = data - data_start; |
| 1597 | |
| 1598 | struct journal_metric_list *current_metric = (void *)data; |
| 1599 | // Write the UUID we are processing |
| 1600 | data = (void *)journalfile_v2_write_metric_page(&j2_header, data, metric_info, pages_offset); |
| 1601 | if (unlikely(!data)) |
| 1602 | break; |
| 1603 | |
| 1604 | // Next we will write |
| 1605 | // Header |
| 1606 | // Detailed entries (descr @ time) |
| 1607 | // Trailer (checksum) |
| 1608 | |
| 1609 | // Keep the page_list_header, to be used for migration when where agent is running |
| 1610 | metric_info->page_list_header = pages_offset; |
| 1611 | // Write page header |
| 1612 | void *metric_page = |
| 1613 | journalfile_v2_write_data_page_header(&j2_header, data_start + pages_offset, metric_info, uuid_offset); |
| 1614 | |
| 1615 | // Start writing descr @ time |
| 1616 | void *page_trailer = journalfile_v2_write_descriptors(&j2_header, metric_page, metric_info, current_metric); |
| 1617 | if (unlikely(!page_trailer)) |
| 1618 | break; |
| 1619 | |
| 1620 | // Trailer (checksum) |
| 1621 | uint8_t *next_page_address = |
| 1622 | journalfile_v2_write_data_page_trailer(&j2_header, page_trailer, data_start + pages_offset); |
| 1623 | |
| 1624 | // Calculate start of the pages start for next descriptor |
| 1625 | pages_offset += |
| 1626 | (metric_info->number_of_pages * (sizeof(struct journal_page_list)) + |
| 1627 | sizeof(struct journal_page_header) + sizeof(struct journal_v2_block_trailer)); |
| 1628 | // Verify we are at the right location |
| 1629 | if (pages_offset != (uint32_t)(next_page_address - data_start)) { |
| 1630 | // make sure checks fail so that we abort |
| 1631 | data = data_start; |
| 1632 | break; |
| 1633 | } |
| 1634 | } |
| 1635 | |
| 1636 | if (data == data_start + metric_offset_trailer) { |
| 1637 | internal_error( |
| 1638 | true, "DBENGINE: WRITE METRICS AND PAGES %llu", (now_monotonic_usec() - start_loading) / USEC_PER_MS); |
| 1639 | |
| 1640 | // Calculate CRC for metrics |
| 1641 | journal_v2_trailer = (struct journal_v2_block_trailer *)(data_start + metric_offset_trailer); |
| 1642 | crc = crc32(0L, Z_NULL, 0); |
| 1643 | crc = crc32( |
| 1644 | crc, (uint8_t *)data_start + metrics_offset, number_of_metrics * sizeof(struct journal_metric_list)); |
| 1645 | crc32set(journal_v2_trailer->checksum, crc); |
| 1646 | internal_error( |
| 1647 | true, "DBENGINE: CALCULATE CRC FOR UUIDs %llu", (now_monotonic_usec() - start_loading) / USEC_PER_MS); |
| 1648 | |
| 1649 | // Prepare to write checksum for the file |
| 1650 | j2_header.data = NULL; |
| 1651 | journal_v2_trailer = (struct journal_v2_block_trailer *)(data_start + trailer_offset); |
| 1652 | crc = crc32(0L, Z_NULL, 0); |
| 1653 | crc = crc32(crc, (void *)&j2_header, sizeof(j2_header)); |
| 1654 | crc32set(journal_v2_trailer->checksum, crc); |
| 1655 | |
| 1656 | // Write header to the file |
| 1657 | memcpy(data_start, &j2_header, sizeof(j2_header)); |
| 1658 | |
| 1659 | internal_error( |
| 1660 | true, "DBENGINE: FILE COMPLETED --------> %llu", (now_monotonic_usec() - start_loading) / USEC_PER_MS); |
| 1661 | |
| 1662 | char size_for_humans[128]; |
| 1663 | size_snprintf(size_for_humans, sizeof(size_for_humans), total_file_size, "B", false); |
| 1664 | netdata_log_info("DBENGINE: tier %d: migrated " WALFILE_PREFIX RRDENG_FILE_NUMBER_PRINT_TMPL WALFILE_EXTENSION_V2 ", %s", |
| 1665 | ctx->config.tier, datafile->tier, datafile->fileno, size_for_humans); |
| 1666 | |
| 1667 | // msync(data_start, total_file_size, MS_SYNC); |
| 1668 | journalfile_v2_data_set(journalfile, fd_v2, data_start, total_file_size); |
| 1669 | |
| 1670 | internal_error( |
| 1671 | true, "DBENGINE: ACTIVATING NEW INDEX JNL %llu", (now_monotonic_usec() - start_loading) / USEC_PER_MS); |
| 1672 | ctx_current_disk_space_increase(ctx, total_file_size); |
| 1673 | freez(uuid_list); |
| 1674 | return true; |
| 1675 | } |
| 1676 | } |
| 1677 | else { |
| 1678 | nd_log(NDLS_DAEMON, NDLP_ERR, "DBENGINE: failed to write journal file \"%s\" (SIGBUS)", path); |
| 1679 | } |
| 1680 | |
| 1681 | freez(uuid_list); |
| 1682 | |
| 1683 | netdata_log_info("DBENGINE: failed to build index \"%s\", file will be skipped", path); |
| 1684 | |
| 1685 | nd_munmap(data_start, total_file_size); |
| 1686 | if(fd_v2 != -1) |
| 1687 | close(fd_v2); |
| 1688 | unlink(path); |
| 1689 | return false; |
| 1690 | } |
| 1691 | |
| 1692 | int journalfile_load(struct rrdengine_instance *ctx, struct rrdengine_journalfile *journalfile, |
| 1693 | struct rrdengine_datafile *datafile) |
| 1694 | { |
| 1695 | uv_file file; |
| 1696 | int ret, fd, error; |
| 1697 | uint64_t file_size, max_id; |
| 1698 | char path[RRDENG_PATH_MAX]; |
| 1699 | bool loaded_v2 = false; |
| 1700 | |
| 1701 | // Do not try to load jv2 of the latest file |
| 1702 | if (datafile->fileno != ctx_last_fileno_get(ctx)) |
| 1703 | loaded_v2 = journalfile_v2_load(ctx, journalfile, datafile) == 0; |
| 1704 | |
| 1705 | journalfile_v1_generate_path(datafile, path, sizeof(path)); |
| 1706 | |
| 1707 | fd = open_file_for_io(path, O_RDWR, &file, dbengine_use_direct_io); |
| 1708 | if (fd < 0) { |
| 1709 | ctx_fs_error(ctx); |
| 1710 | |
| 1711 | if(loaded_v2) |
| 1712 | return 0; |
| 1713 | |
| 1714 | return fd; |
| 1715 | } |
| 1716 | |
| 1717 | ret = check_file_properties(file, &file_size, sizeof(struct rrdeng_df_sb)); |
| 1718 | if (ret) { |
| 1719 | error = ret; |
| 1720 | goto cleanup; |
| 1721 | } |
| 1722 | |
| 1723 | if(loaded_v2) { |
| 1724 | journalfile->unsafe.pos = file_size; |
| 1725 | error = 0; |
| 1726 | goto cleanup; |
| 1727 | } |
| 1728 | |
| 1729 | file_size = ALIGN_BYTES_FLOOR(file_size); |
| 1730 | journalfile->unsafe.pos = file_size; |
| 1731 | journalfile->file = file; |
| 1732 | |
| 1733 | ret = journalfile_check_superblock(file); |
| 1734 | if (ret) { |
| 1735 | netdata_log_info("DBENGINE: invalid journal file \"%s\" ; superblock check failed.", path); |
| 1736 | error = ret; |
| 1737 | goto cleanup; |
| 1738 | } |
| 1739 | ctx_io_read_op_bytes(ctx, sizeof(struct rrdeng_jf_sb)); |
| 1740 | |
| 1741 | nd_log_daemon(NDLP_DEBUG, "DBENGINE: loading journal file \"%s\"", path); |
| 1742 | |
| 1743 | max_id = journalfile_iterate_transactions(ctx, journalfile); |
| 1744 | |
| 1745 | __atomic_store_n(&ctx->atomic.transaction_id, MAX(__atomic_load_n(&ctx->atomic.transaction_id, __ATOMIC_RELAXED), max_id + 1), __ATOMIC_RELAXED); |
| 1746 | |
| 1747 | nd_log_daemon(NDLP_DEBUG, "DBENGINE: journal file \"%s\" loaded (size:%" PRIu64 ").", path, file_size); |
| 1748 | |
| 1749 | bool is_last_file = (ctx_last_fileno_get(ctx) == journalfile->datafile->fileno); |
| 1750 | bool has_old_data = false; |
| 1751 | if (ctx->config.tier == 0 && journalfile->v2.last_time_s > 0) |
| 1752 | has_old_data = (now_realtime_sec() - journalfile->v2.last_time_s) > 86400; |
| 1753 | |
| 1754 | if (is_last_file && journalfile->datafile->pos <= rrdeng_target_data_file_size(ctx) / 3 && !has_old_data) { |
| 1755 | ctx->loading.create_new_datafile_pair = false; |
| 1756 | return 0; |
| 1757 | } |
| 1758 | |
| 1759 | pgc_open_cache_to_journal_v2( |
| 1760 | open_cache, |
| 1761 | (Word_t)ctx, |
| 1762 | (int)datafile->fileno, |
| 1763 | ctx->config.page_type, |
| 1764 | journalfile_migrate_to_v2_callback, |
| 1765 | (void *)datafile->journalfile, |
| 1766 | true); |
| 1767 | |
| 1768 | if (is_last_file) |
| 1769 | ctx->loading.create_new_datafile_pair = true; |
| 1770 | |
| 1771 | return 0; |
| 1772 | |
| 1773 | cleanup: |
| 1774 | CLOSE_FILE(ctx, path, file, ret); |
| 1775 | return error; |
| 1776 | } |