| 1 | /* |
| 2 | * SPDX-License-Identifier: MIT |
| 3 | * |
| 4 | * Copyright 2016-2017 Philippe Proulx <pproulx@efficios.com> |
| 5 | * Copyright 2016 Jérémie Galarneau <jeremie.galarneau@efficios.com> |
| 6 | * Copyright 2010-2011 EfficiOS Inc. and Linux Foundation |
| 7 | */ |
| 8 | |
| 9 | #include <glib.h> |
| 10 | #include <stdint.h> |
| 11 | #include <stdio.h> |
| 12 | |
| 13 | #include "compat/endian.h" /* IWYU pragma: keep */ |
| 14 | #include "compat/mman.h" /* IWYU: pragma keep */ |
| 15 | #include "cpp-common/bt2s/make-unique.hpp" |
| 16 | #include "cpp-common/vendor/fmt/format.h" |
| 17 | |
| 18 | #include "../common/src/msg-iter/msg-iter.hpp" |
| 19 | #include "data-stream-file.hpp" |
| 20 | #include "file.hpp" |
| 21 | #include "fs.hpp" |
| 22 | #include "lttng-index.hpp" |
| 23 | |
| 24 | static inline size_t remaining_mmap_bytes(struct ctf_fs_ds_file *ds_file) |
| 25 | { |
| 26 | BT_ASSERT_DBG(ds_file->mmap_len >= ds_file->request_offset_in_mapping); |
| 27 | return ds_file->mmap_len - ds_file->request_offset_in_mapping; |
| 28 | } |
| 29 | |
| 30 | /* |
| 31 | * Return true if `offset_in_file` is in the current mapping. |
| 32 | */ |
| 33 | |
| 34 | static bool offset_ist_mapped(struct ctf_fs_ds_file *ds_file, off_t offset_in_file) |
| 35 | { |
| 36 | return offset_in_file >= ds_file->mmap_offset_in_file && |
| 37 | offset_in_file < (ds_file->mmap_offset_in_file + ds_file->mmap_len); |
| 38 | } |
| 39 | |
| 40 | static enum ctf_msg_iter_medium_status ds_file_munmap(struct ctf_fs_ds_file *ds_file) |
| 41 | { |
| 42 | enum ctf_msg_iter_medium_status status; |
| 43 | |
| 44 | BT_ASSERT(ds_file); |
| 45 | |
| 46 | if (!ds_file->mmap_addr) { |
| 47 | status = CTF_MSG_ITER_MEDIUM_STATUS_OK; |
| 48 | goto end; |
| 49 | } |
| 50 | |
| 51 | if (bt_munmap(ds_file->mmap_addr, ds_file->mmap_len)) { |
| 52 | BT_CPPLOGE_ERRNO_SPEC(ds_file->logger, "Cannot memory-unmap file", |
| 53 | ": address={}, size={}, file_path=\"{}\", file={}", |
| 54 | fmt::ptr(ds_file->mmap_addr), ds_file->mmap_len, |
| 55 | ds_file->file ? ds_file->file->path : "NULL", |
| 56 | ds_file->file ? fmt::ptr(ds_file->file->fp) : NULL); |
| 57 | status = CTF_MSG_ITER_MEDIUM_STATUS_ERROR; |
| 58 | goto end; |
| 59 | } |
| 60 | |
| 61 | ds_file->mmap_addr = NULL; |
| 62 | |
| 63 | status = CTF_MSG_ITER_MEDIUM_STATUS_OK; |
| 64 | end: |
| 65 | return status; |
| 66 | } |
| 67 | |
| 68 | /* |
| 69 | * mmap a region of `ds_file` such that `requested_offset_in_file` is in the |
| 70 | * mapping. If the currently mmap-ed region already contains |
| 71 | * `requested_offset_in_file`, the mapping is kept. |
| 72 | * |
| 73 | * Set `ds_file->requested_offset_in_mapping` based on `request_offset_in_file`, |
| 74 | * such that the next call to `request_bytes` will return bytes starting at that |
| 75 | * position. |
| 76 | * |
| 77 | * `requested_offset_in_file` must be a valid offset in the file. |
| 78 | */ |
| 79 | static enum ctf_msg_iter_medium_status ds_file_mmap(struct ctf_fs_ds_file *ds_file, |
| 80 | off_t requested_offset_in_file) |
| 81 | { |
| 82 | enum ctf_msg_iter_medium_status status; |
| 83 | |
| 84 | /* Ensure the requested offset is in the file range. */ |
| 85 | BT_ASSERT(requested_offset_in_file >= 0); |
| 86 | BT_ASSERT(requested_offset_in_file < ds_file->file->size); |
| 87 | |
| 88 | /* |
| 89 | * If the mapping already contains the requested offset, just adjust |
| 90 | * requested_offset_in_mapping. |
| 91 | */ |
| 92 | if (offset_ist_mapped(ds_file, requested_offset_in_file)) { |
| 93 | ds_file->request_offset_in_mapping = |
| 94 | requested_offset_in_file - ds_file->mmap_offset_in_file; |
| 95 | status = CTF_MSG_ITER_MEDIUM_STATUS_OK; |
| 96 | goto end; |
| 97 | } |
| 98 | |
| 99 | /* Unmap old region */ |
| 100 | status = ds_file_munmap(ds_file); |
| 101 | if (status != CTF_MSG_ITER_MEDIUM_STATUS_OK) { |
| 102 | goto end; |
| 103 | } |
| 104 | |
| 105 | /* |
| 106 | * Compute a mapping that has the required alignment properties and |
| 107 | * contains `requested_offset_in_file`. |
| 108 | */ |
| 109 | ds_file->request_offset_in_mapping = |
| 110 | requested_offset_in_file % |
| 111 | bt_mmap_get_offset_align_size(static_cast<int>(ds_file->logger.level())); |
| 112 | ds_file->mmap_offset_in_file = requested_offset_in_file - ds_file->request_offset_in_mapping; |
| 113 | ds_file->mmap_len = |
| 114 | MIN(ds_file->file->size - ds_file->mmap_offset_in_file, ds_file->mmap_max_len); |
| 115 | |
| 116 | BT_ASSERT(ds_file->mmap_len > 0); |
| 117 | |
| 118 | ds_file->mmap_addr = |
| 119 | bt_mmap(ds_file->mmap_len, PROT_READ, MAP_PRIVATE, fileno(ds_file->file->fp.get()), |
| 120 | ds_file->mmap_offset_in_file, static_cast<int>(ds_file->logger.level())); |
| 121 | if (ds_file->mmap_addr == MAP_FAILED) { |
| 122 | BT_CPPLOGE_SPEC(ds_file->logger, |
| 123 | "Cannot memory-map address (size {}) of file \"{}\" ({}) at offset {}: {}", |
| 124 | ds_file->mmap_len, ds_file->file->path, fmt::ptr(ds_file->file->fp), |
| 125 | (intmax_t) ds_file->mmap_offset_in_file, strerror(errno)); |
| 126 | status = CTF_MSG_ITER_MEDIUM_STATUS_ERROR; |
| 127 | goto end; |
| 128 | } |
| 129 | |
| 130 | status = CTF_MSG_ITER_MEDIUM_STATUS_OK; |
| 131 | |
| 132 | end: |
| 133 | return status; |
| 134 | } |
| 135 | |
| 136 | /* |
| 137 | * Change the mapping of the file to read the region that follows the current |
| 138 | * mapping. |
| 139 | * |
| 140 | * If the file hasn't been mapped yet, then everything (mmap_offset_in_file, |
| 141 | * mmap_len, request_offset_in_mapping) should have the value 0, which will |
| 142 | * result in the beginning of the file getting mapped. |
| 143 | * |
| 144 | * return _EOF if the current mapping is the end of the file. |
| 145 | */ |
| 146 | |
| 147 | static enum ctf_msg_iter_medium_status ds_file_mmap_next(struct ctf_fs_ds_file *ds_file) |
| 148 | { |
| 149 | enum ctf_msg_iter_medium_status status; |
| 150 | |
| 151 | /* |
| 152 | * If we're called, it's because more bytes are requested but we have |
| 153 | * given all the bytes of the current mapping. |
| 154 | */ |
| 155 | BT_ASSERT(ds_file->request_offset_in_mapping == ds_file->mmap_len); |
| 156 | |
| 157 | /* |
| 158 | * If the current mapping coincides with the end of the file, there is |
| 159 | * no next mapping. |
| 160 | */ |
| 161 | if (ds_file->mmap_offset_in_file + ds_file->mmap_len == ds_file->file->size) { |
| 162 | status = CTF_MSG_ITER_MEDIUM_STATUS_EOF; |
| 163 | goto end; |
| 164 | } |
| 165 | |
| 166 | status = ds_file_mmap(ds_file, ds_file->mmap_offset_in_file + ds_file->mmap_len); |
| 167 | |
| 168 | end: |
| 169 | return status; |
| 170 | } |
| 171 | |
| 172 | static enum ctf_msg_iter_medium_status medop_request_bytes(size_t request_sz, uint8_t **buffer_addr, |
| 173 | size_t *buffer_sz, void *data) |
| 174 | { |
| 175 | enum ctf_msg_iter_medium_status status = CTF_MSG_ITER_MEDIUM_STATUS_OK; |
| 176 | struct ctf_fs_ds_file *ds_file = (struct ctf_fs_ds_file *) data; |
| 177 | |
| 178 | BT_ASSERT(request_sz > 0); |
| 179 | |
| 180 | /* |
| 181 | * Check if we have at least one memory-mapped byte left. If we don't, |
| 182 | * mmap the next file. |
| 183 | */ |
| 184 | if (remaining_mmap_bytes(ds_file) == 0) { |
| 185 | /* Are we at the end of the file? */ |
| 186 | if (ds_file->mmap_offset_in_file >= ds_file->file->size) { |
| 187 | BT_CPPLOGD_SPEC(ds_file->logger, "Reached end of file \"{}\" ({})", ds_file->file->path, |
| 188 | fmt::ptr(ds_file->file->fp)); |
| 189 | status = CTF_MSG_ITER_MEDIUM_STATUS_EOF; |
| 190 | goto end; |
| 191 | } |
| 192 | |
| 193 | status = ds_file_mmap_next(ds_file); |
| 194 | switch (status) { |
| 195 | case CTF_MSG_ITER_MEDIUM_STATUS_OK: |
| 196 | break; |
| 197 | case CTF_MSG_ITER_MEDIUM_STATUS_EOF: |
| 198 | goto end; |
| 199 | default: |
| 200 | BT_CPPLOGE_SPEC(ds_file->logger, "Cannot memory-map next region of file \"{}\" ({})", |
| 201 | ds_file->file->path, fmt::ptr(ds_file->file->fp)); |
| 202 | goto error; |
| 203 | } |
| 204 | } |
| 205 | |
| 206 | BT_ASSERT(remaining_mmap_bytes(ds_file) > 0); |
| 207 | *buffer_sz = MIN(remaining_mmap_bytes(ds_file), request_sz); |
| 208 | |
| 209 | BT_ASSERT(ds_file->mmap_addr); |
| 210 | *buffer_addr = ((uint8_t *) ds_file->mmap_addr) + ds_file->request_offset_in_mapping; |
| 211 | |
| 212 | ds_file->request_offset_in_mapping += *buffer_sz; |
| 213 | goto end; |
| 214 | |
| 215 | error: |
| 216 | status = CTF_MSG_ITER_MEDIUM_STATUS_ERROR; |
| 217 | |
| 218 | end: |
| 219 | return status; |
| 220 | } |
| 221 | |
| 222 | static bt_stream *medop_borrow_stream(bt_stream_class *stream_class, int64_t, void *data) |
| 223 | { |
| 224 | struct ctf_fs_ds_file *ds_file = (struct ctf_fs_ds_file *) data; |
| 225 | bt_stream_class *ds_file_stream_class; |
| 226 | bt_stream *stream = NULL; |
| 227 | |
| 228 | ds_file_stream_class = ds_file->stream->cls().libObjPtr(); |
| 229 | |
| 230 | if (stream_class != ds_file_stream_class) { |
| 231 | /* |
| 232 | * Not supported: two packets described by two different |
| 233 | * stream classes within the same data stream file. |
| 234 | */ |
| 235 | goto end; |
| 236 | } |
| 237 | |
| 238 | stream = ds_file->stream->libObjPtr(); |
| 239 | |
| 240 | end: |
| 241 | return stream; |
| 242 | } |
| 243 | |
| 244 | static enum ctf_msg_iter_medium_status medop_seek(off_t offset, void *data) |
| 245 | { |
| 246 | struct ctf_fs_ds_file *ds_file = (struct ctf_fs_ds_file *) data; |
| 247 | |
| 248 | BT_ASSERT(offset >= 0); |
| 249 | BT_ASSERT(offset < ds_file->file->size); |
| 250 | |
| 251 | return ds_file_mmap(ds_file, offset); |
| 252 | } |
| 253 | |
| 254 | struct ctf_msg_iter_medium_ops ctf_fs_ds_file_medops = { |
| 255 | medop_request_bytes, |
| 256 | medop_seek, |
| 257 | nullptr, |
| 258 | medop_borrow_stream, |
| 259 | }; |
| 260 | |
| 261 | struct ctf_fs_ds_group_medops_data |
| 262 | { |
| 263 | explicit ctf_fs_ds_group_medops_data(const bt2c::Logger& parentLogger) : |
| 264 | logger {parentLogger, "PLUGIN/SRC.CTF.FS/DS-GROUP-MEDOPS"} |
| 265 | { |
| 266 | } |
| 267 | |
| 268 | bt2c::Logger logger; |
| 269 | |
| 270 | /* Weak, set once at creation time. */ |
| 271 | struct ctf_fs_ds_file_group *ds_file_group = nullptr; |
| 272 | |
| 273 | /* |
| 274 | * Index (as in element rank) of the index entry of ds_file_groups' |
| 275 | * index we will read next (so, the one after the one we are reading |
| 276 | * right now). |
| 277 | */ |
| 278 | guint next_index_entry_index = 0; |
| 279 | |
| 280 | /* |
| 281 | * File we are currently reading. Changes whenever we switch to |
| 282 | * reading another data file. |
| 283 | * |
| 284 | * Owned by this. |
| 285 | */ |
| 286 | struct ctf_fs_ds_file *file = nullptr; |
| 287 | |
| 288 | /* Weak, for context / logging / appending causes. */ |
| 289 | bt_self_message_iterator *self_msg_iter = nullptr; |
| 290 | }; |
| 291 | |
| 292 | static enum ctf_msg_iter_medium_status medop_group_request_bytes(size_t request_sz, |
| 293 | uint8_t **buffer_addr, |
| 294 | size_t *buffer_sz, void *void_data) |
| 295 | { |
| 296 | struct ctf_fs_ds_group_medops_data *data = (struct ctf_fs_ds_group_medops_data *) void_data; |
| 297 | |
| 298 | /* Return bytes from the current file. */ |
| 299 | return medop_request_bytes(request_sz, buffer_addr, buffer_sz, data->file); |
| 300 | } |
| 301 | |
| 302 | static bt_stream *medop_group_borrow_stream(bt_stream_class *stream_class, int64_t stream_id, |
| 303 | void *void_data) |
| 304 | { |
| 305 | struct ctf_fs_ds_group_medops_data *data = (struct ctf_fs_ds_group_medops_data *) void_data; |
| 306 | |
| 307 | return medop_borrow_stream(stream_class, stream_id, data->file); |
| 308 | } |
| 309 | |
| 310 | /* |
| 311 | * Set `data->file` to prepare it to read the packet described |
| 312 | * by `index_entry`. |
| 313 | */ |
| 314 | |
| 315 | static enum ctf_msg_iter_medium_status |
| 316 | ctf_fs_ds_group_medops_set_file(struct ctf_fs_ds_group_medops_data *data, |
| 317 | struct ctf_fs_ds_index_entry *index_entry) |
| 318 | { |
| 319 | enum ctf_msg_iter_medium_status status; |
| 320 | |
| 321 | BT_ASSERT(data); |
| 322 | BT_ASSERT(index_entry); |
| 323 | |
| 324 | /* Check if that file is already the one mapped. */ |
| 325 | if (!data->file || data->file->file->path != index_entry->path) { |
| 326 | /* Destroy the previously used file. */ |
| 327 | ctf_fs_ds_file_destroy(data->file); |
| 328 | |
| 329 | /* Create the new file. */ |
| 330 | data->file = |
| 331 | ctf_fs_ds_file_create(data->ds_file_group->ctf_fs_trace, data->ds_file_group->stream, |
| 332 | index_entry->path, data->logger); |
| 333 | if (!data->file) { |
| 334 | BT_CPPLOGE_APPEND_CAUSE_SPEC(data->logger, "failed to create ctf_fs_ds_file."); |
| 335 | status = CTF_MSG_ITER_MEDIUM_STATUS_ERROR; |
| 336 | goto end; |
| 337 | } |
| 338 | } |
| 339 | |
| 340 | /* |
| 341 | * Ensure the right portion of the file will be returned on the next |
| 342 | * request_bytes call. |
| 343 | */ |
| 344 | status = ds_file_mmap(data->file, index_entry->offset.bytes()); |
| 345 | if (status != CTF_MSG_ITER_MEDIUM_STATUS_OK) { |
| 346 | goto end; |
| 347 | } |
| 348 | |
| 349 | status = CTF_MSG_ITER_MEDIUM_STATUS_OK; |
| 350 | |
| 351 | end: |
| 352 | return status; |
| 353 | } |
| 354 | |
| 355 | static enum ctf_msg_iter_medium_status medop_group_switch_packet(void *void_data) |
| 356 | { |
| 357 | struct ctf_fs_ds_group_medops_data *data = (struct ctf_fs_ds_group_medops_data *) void_data; |
| 358 | struct ctf_fs_ds_index_entry *index_entry; |
| 359 | enum ctf_msg_iter_medium_status status; |
| 360 | |
| 361 | /* If we have gone through all index entries, we are done. */ |
| 362 | if (data->next_index_entry_index >= data->ds_file_group->index->entries.size()) { |
| 363 | status = CTF_MSG_ITER_MEDIUM_STATUS_EOF; |
| 364 | goto end; |
| 365 | } |
| 366 | |
| 367 | /* |
| 368 | * Otherwise, look up the next index entry / packet and prepare it |
| 369 | * for reading. |
| 370 | */ |
| 371 | index_entry = data->ds_file_group->index->entries[data->next_index_entry_index].get(); |
| 372 | |
| 373 | status = ctf_fs_ds_group_medops_set_file(data, index_entry); |
| 374 | if (status != CTF_MSG_ITER_MEDIUM_STATUS_OK) { |
| 375 | goto end; |
| 376 | } |
| 377 | |
| 378 | data->next_index_entry_index++; |
| 379 | |
| 380 | status = CTF_MSG_ITER_MEDIUM_STATUS_OK; |
| 381 | end: |
| 382 | return status; |
| 383 | } |
| 384 | |
| 385 | void ctf_fs_ds_group_medops_data_destroy(struct ctf_fs_ds_group_medops_data *data) |
| 386 | { |
| 387 | if (!data) { |
| 388 | goto end; |
| 389 | } |
| 390 | |
| 391 | ctf_fs_ds_file_destroy(data->file); |
| 392 | |
| 393 | delete data; |
| 394 | |
| 395 | end: |
| 396 | return; |
| 397 | } |
| 398 | |
| 399 | enum ctf_msg_iter_medium_status ctf_fs_ds_group_medops_data_create( |
| 400 | struct ctf_fs_ds_file_group *ds_file_group, bt_self_message_iterator *self_msg_iter, |
| 401 | const bt2c::Logger& parentLogger, struct ctf_fs_ds_group_medops_data **out) |
| 402 | { |
| 403 | BT_ASSERT(self_msg_iter); |
| 404 | BT_ASSERT(ds_file_group); |
| 405 | BT_ASSERT(ds_file_group->index); |
| 406 | BT_ASSERT(!ds_file_group->index->entries.empty()); |
| 407 | |
| 408 | ctf_fs_ds_group_medops_data *data = new ctf_fs_ds_group_medops_data {parentLogger}; |
| 409 | data->ds_file_group = ds_file_group; |
| 410 | data->self_msg_iter = self_msg_iter; |
| 411 | |
| 412 | /* |
| 413 | * No need to prepare the first file. ctf_msg_iter will call |
| 414 | * switch_packet before reading the first packet, it will be |
| 415 | * done then. |
| 416 | */ |
| 417 | |
| 418 | *out = data; |
| 419 | return CTF_MSG_ITER_MEDIUM_STATUS_OK; |
| 420 | } |
| 421 | |
| 422 | void ctf_fs_ds_group_medops_data_reset(struct ctf_fs_ds_group_medops_data *data) |
| 423 | { |
| 424 | data->next_index_entry_index = 0; |
| 425 | } |
| 426 | |
| 427 | struct ctf_msg_iter_medium_ops ctf_fs_ds_group_medops = { |
| 428 | .request_bytes = medop_group_request_bytes, |
| 429 | |
| 430 | /* |
| 431 | * We don't support seeking using this medops. It would probably be |
| 432 | * possible, but it's not needed at the moment. |
| 433 | */ |
| 434 | .seek = NULL, |
| 435 | |
| 436 | .switch_packet = medop_group_switch_packet, |
| 437 | .borrow_stream = medop_group_borrow_stream, |
| 438 | }; |
| 439 | |
| 440 | static ctf_fs_ds_index_entry::UP ctf_fs_ds_index_entry_create(const bt2c::DataLen offset, |
| 441 | const bt2c::DataLen packetSize) |
| 442 | { |
| 443 | ctf_fs_ds_index_entry::UP entry = bt2s::make_unique<ctf_fs_ds_index_entry>(offset, packetSize); |
| 444 | |
| 445 | entry->packet_seq_num = UINT64_MAX; |
| 446 | |
| 447 | return entry; |
| 448 | } |
| 449 | |
| 450 | static int convert_cycles_to_ns(struct ctf_clock_class *clock_class, uint64_t cycles, int64_t *ns) |
| 451 | { |
| 452 | return bt_util_clock_cycles_to_ns_from_origin(cycles, clock_class->frequency, |
| 453 | clock_class->offset_seconds, |
| 454 | clock_class->offset_cycles, ns); |
| 455 | } |
| 456 | |
| 457 | static ctf_fs_ds_index::UP build_index_from_idx_file(struct ctf_fs_ds_file *ds_file, |
| 458 | struct ctf_fs_ds_file_info *file_info, |
| 459 | struct ctf_msg_iter *msg_iter) |
| 460 | { |
| 461 | int ret; |
| 462 | gchar *directory = NULL; |
| 463 | gchar *basename = NULL; |
| 464 | GString *index_basename = NULL; |
| 465 | gchar *index_file_path = NULL; |
| 466 | GMappedFile *mapped_file = NULL; |
| 467 | gsize filesize; |
| 468 | const char *mmap_begin = NULL, *file_pos = NULL; |
| 469 | const struct ctf_packet_index_file_hdr *header = NULL; |
| 470 | ctf_fs_ds_index::UP index; |
| 471 | ctf_fs_ds_index_entry::UP index_entry; |
| 472 | ctf_fs_ds_index_entry *prev_index_entry = NULL; |
| 473 | auto totalPacketsSize = bt2c::DataLen::fromBytes(0); |
| 474 | size_t file_index_entry_size; |
| 475 | size_t file_entry_count; |
| 476 | size_t i; |
| 477 | struct ctf_stream_class *sc; |
| 478 | struct ctf_msg_iter_packet_properties props; |
| 479 | uint32_t version_major, version_minor; |
| 480 | |
| 481 | BT_CPPLOGI_SPEC(ds_file->logger, "Building index from .idx file of stream file {}", |
| 482 | ds_file->file->path); |
| 483 | ret = ctf_msg_iter_get_packet_properties(msg_iter, &props); |
| 484 | if (ret) { |
| 485 | BT_CPPLOGI_STR_SPEC(ds_file->logger, |
| 486 | "Cannot read first packet's header and context fields."); |
| 487 | goto error; |
| 488 | } |
| 489 | |
| 490 | sc = ctf_trace_class_borrow_stream_class_by_id(ds_file->metadata->tc, props.stream_class_id); |
| 491 | BT_ASSERT(sc); |
| 492 | if (!sc->default_clock_class) { |
| 493 | BT_CPPLOGI_STR_SPEC(ds_file->logger, "Cannot find stream class's default clock class."); |
| 494 | goto error; |
| 495 | } |
| 496 | |
| 497 | /* Look for index file in relative path index/name.idx. */ |
| 498 | basename = g_path_get_basename(ds_file->file->path.c_str()); |
| 499 | if (!basename) { |
| 500 | BT_CPPLOGE_SPEC(ds_file->logger, "Cannot get the basename of datastream file {}", |
| 501 | ds_file->file->path); |
| 502 | goto error; |
| 503 | } |
| 504 | |
| 505 | directory = g_path_get_dirname(ds_file->file->path.c_str()); |
| 506 | if (!directory) { |
| 507 | BT_CPPLOGE_SPEC(ds_file->logger, "Cannot get dirname of datastream file {}", |
| 508 | ds_file->file->path); |
| 509 | goto error; |
| 510 | } |
| 511 | |
| 512 | index_basename = g_string_new(basename); |
| 513 | if (!index_basename) { |
| 514 | BT_CPPLOGE_STR_SPEC(ds_file->logger, "Cannot allocate index file basename string"); |
| 515 | goto error; |
| 516 | } |
| 517 | |
| 518 | g_string_append(index_basename, ".idx"); |
| 519 | index_file_path = g_build_filename(directory, "index", index_basename->str, NULL); |
| 520 | mapped_file = g_mapped_file_new(index_file_path, FALSE, NULL); |
| 521 | if (!mapped_file) { |
| 522 | BT_CPPLOGD_SPEC(ds_file->logger, "Cannot create new mapped file {}", index_file_path); |
| 523 | goto error; |
| 524 | } |
| 525 | |
| 526 | /* |
| 527 | * The g_mapped_file API limits us to 4GB files on 32-bit. |
| 528 | * Traces with such large indexes have never been seen in the wild, |
| 529 | * but this would need to be adjusted to support them. |
| 530 | */ |
| 531 | filesize = g_mapped_file_get_length(mapped_file); |
| 532 | if (filesize < sizeof(*header)) { |
| 533 | BT_CPPLOGW_SPEC(ds_file->logger, |
| 534 | "Invalid LTTng trace index file: " |
| 535 | "file size ({} bytes) < header size ({} bytes)", |
| 536 | filesize, sizeof(*header)); |
| 537 | goto error; |
| 538 | } |
| 539 | |
| 540 | mmap_begin = g_mapped_file_get_contents(mapped_file); |
| 541 | header = (struct ctf_packet_index_file_hdr *) mmap_begin; |
| 542 | |
| 543 | file_pos = g_mapped_file_get_contents(mapped_file) + sizeof(*header); |
| 544 | if (be32toh(header->magic) != CTF_INDEX_MAGIC) { |
| 545 | BT_CPPLOGW_STR_SPEC(ds_file->logger, |
| 546 | "Invalid LTTng trace index: \"magic\" field validation failed"); |
| 547 | goto error; |
| 548 | } |
| 549 | |
| 550 | version_major = be32toh(header->index_major); |
| 551 | version_minor = be32toh(header->index_minor); |
| 552 | if (version_major != 1) { |
| 553 | BT_CPPLOGW_SPEC(ds_file->logger, "Unknown LTTng trace index version: major={}, minor={}", |
| 554 | version_major, version_minor); |
| 555 | goto error; |
| 556 | } |
| 557 | |
| 558 | file_index_entry_size = be32toh(header->packet_index_len); |
| 559 | if (file_index_entry_size < CTF_INDEX_1_0_SIZE) { |
| 560 | BT_CPPLOGW_SPEC( |
| 561 | ds_file->logger, |
| 562 | "Invalid `packet_index_len` in LTTng trace index file (`packet_index_len` < CTF index 1.0 index entry size): " |
| 563 | "packet_index_len={}, CTF_INDEX_1_0_SIZE={}", |
| 564 | file_index_entry_size, CTF_INDEX_1_0_SIZE); |
| 565 | goto error; |
| 566 | } |
| 567 | |
| 568 | file_entry_count = (filesize - sizeof(*header)) / file_index_entry_size; |
| 569 | if ((filesize - sizeof(*header)) % file_index_entry_size) { |
| 570 | BT_CPPLOGW_SPEC(ds_file->logger, |
| 571 | "Invalid LTTng trace index: the index's size after the header " |
| 572 | "({} bytes) is not a multiple of the index entry size " |
| 573 | "({} bytes)", |
| 574 | (filesize - sizeof(*header)), sizeof(*header)); |
| 575 | goto error; |
| 576 | } |
| 577 | |
| 578 | index = ctf_fs_ds_index_create(); |
| 579 | if (!index) { |
| 580 | goto error; |
| 581 | } |
| 582 | |
| 583 | for (i = 0; i < file_entry_count; i++) { |
| 584 | struct ctf_packet_index *file_index = (struct ctf_packet_index *) file_pos; |
| 585 | const auto packetSize = bt2c::DataLen::fromBits(be64toh(file_index->packet_size)); |
| 586 | |
| 587 | if (packetSize.hasExtraBits()) { |
| 588 | BT_CPPLOGW_SPEC(ds_file->logger, |
| 589 | "Invalid packet size encountered in LTTng trace index file"); |
| 590 | goto error; |
| 591 | } |
| 592 | |
| 593 | const auto offset = bt2c::DataLen::fromBytes(be64toh(file_index->offset)); |
| 594 | |
| 595 | if (i != 0 && offset < prev_index_entry->offset) { |
| 596 | BT_CPPLOGW_SPEC( |
| 597 | ds_file->logger, |
| 598 | "Invalid, non-monotonic, packet offset encountered in LTTng trace index file: " |
| 599 | "previous offset={} bytes, current offset={} bytes", |
| 600 | prev_index_entry->offset.bytes(), offset.bytes()); |
| 601 | goto error; |
| 602 | } |
| 603 | |
| 604 | index_entry = ctf_fs_ds_index_entry_create(offset, packetSize); |
| 605 | if (!index_entry) { |
| 606 | BT_CPPLOGE_APPEND_CAUSE_SPEC(ds_file->logger, |
| 607 | "Failed to create a ctf_fs_ds_index_entry."); |
| 608 | goto error; |
| 609 | } |
| 610 | |
| 611 | /* Set path to stream file. */ |
| 612 | index_entry->path = file_info->path.c_str(); |
| 613 | |
| 614 | index_entry->timestamp_begin = be64toh(file_index->timestamp_begin); |
| 615 | index_entry->timestamp_end = be64toh(file_index->timestamp_end); |
| 616 | if (index_entry->timestamp_end < index_entry->timestamp_begin) { |
| 617 | BT_CPPLOGW_SPEC( |
| 618 | ds_file->logger, |
| 619 | "Invalid packet time bounds encountered in LTTng trace index file (begin > end): " |
| 620 | "timestamp_begin={}, timestamp_end={}", |
| 621 | index_entry->timestamp_begin, index_entry->timestamp_end); |
| 622 | goto error; |
| 623 | } |
| 624 | |
| 625 | /* Convert the packet's bound to nanoseconds since Epoch. */ |
| 626 | ret = convert_cycles_to_ns(sc->default_clock_class, index_entry->timestamp_begin, |
| 627 | &index_entry->timestamp_begin_ns); |
| 628 | if (ret) { |
| 629 | BT_CPPLOGI_STR_SPEC( |
| 630 | ds_file->logger, |
| 631 | "Failed to convert raw timestamp to nanoseconds since Epoch during index parsing"); |
| 632 | goto error; |
| 633 | } |
| 634 | ret = convert_cycles_to_ns(sc->default_clock_class, index_entry->timestamp_end, |
| 635 | &index_entry->timestamp_end_ns); |
| 636 | if (ret) { |
| 637 | BT_CPPLOGI_STR_SPEC( |
| 638 | ds_file->logger, |
| 639 | "Failed to convert raw timestamp to nanoseconds since Epoch during LTTng trace index parsing"); |
| 640 | goto error; |
| 641 | } |
| 642 | |
| 643 | if (version_minor >= 1) { |
| 644 | index_entry->packet_seq_num = be64toh(file_index->packet_seq_num); |
| 645 | } |
| 646 | |
| 647 | totalPacketsSize += packetSize; |
| 648 | file_pos += file_index_entry_size; |
| 649 | |
| 650 | prev_index_entry = index_entry.get(); |
| 651 | |
| 652 | index->entries.emplace_back(std::move(index_entry)); |
| 653 | } |
| 654 | |
| 655 | /* Validate that the index addresses the complete stream. */ |
| 656 | if (ds_file->file->size != totalPacketsSize.bytes()) { |
| 657 | BT_CPPLOGW_SPEC(ds_file->logger, |
| 658 | "Invalid LTTng trace index file; indexed size != stream file size: " |
| 659 | "file-size={} bytes, total-packets-size={} bytes", |
| 660 | ds_file->file->size, totalPacketsSize.bytes()); |
| 661 | goto error; |
| 662 | } |
| 663 | end: |
| 664 | g_free(directory); |
| 665 | g_free(basename); |
| 666 | g_free(index_file_path); |
| 667 | if (index_basename) { |
| 668 | g_string_free(index_basename, TRUE); |
| 669 | } |
| 670 | if (mapped_file) { |
| 671 | g_mapped_file_unref(mapped_file); |
| 672 | } |
| 673 | return index; |
| 674 | error: |
| 675 | index.reset(); |
| 676 | goto end; |
| 677 | } |
| 678 | |
| 679 | static int init_index_entry(struct ctf_fs_ds_index_entry *entry, struct ctf_fs_ds_file *ds_file, |
| 680 | struct ctf_msg_iter_packet_properties *props) |
| 681 | { |
| 682 | int ret = 0; |
| 683 | struct ctf_stream_class *sc; |
| 684 | |
| 685 | sc = ctf_trace_class_borrow_stream_class_by_id(ds_file->metadata->tc, props->stream_class_id); |
| 686 | BT_ASSERT(sc); |
| 687 | |
| 688 | if (props->snapshots.beginning_clock != UINT64_C(-1)) { |
| 689 | entry->timestamp_begin = props->snapshots.beginning_clock; |
| 690 | |
| 691 | /* Convert the packet's bound to nanoseconds since Epoch. */ |
| 692 | ret = convert_cycles_to_ns(sc->default_clock_class, props->snapshots.beginning_clock, |
| 693 | &entry->timestamp_begin_ns); |
| 694 | if (ret) { |
| 695 | BT_CPPLOGI_STR_SPEC(ds_file->logger, |
| 696 | "Failed to convert raw timestamp to nanoseconds since Epoch."); |
| 697 | goto end; |
| 698 | } |
| 699 | } else { |
| 700 | entry->timestamp_begin = UINT64_C(-1); |
| 701 | entry->timestamp_begin_ns = UINT64_C(-1); |
| 702 | } |
| 703 | |
| 704 | if (props->snapshots.end_clock != UINT64_C(-1)) { |
| 705 | entry->timestamp_end = props->snapshots.end_clock; |
| 706 | |
| 707 | /* Convert the packet's bound to nanoseconds since Epoch. */ |
| 708 | ret = convert_cycles_to_ns(sc->default_clock_class, props->snapshots.end_clock, |
| 709 | &entry->timestamp_end_ns); |
| 710 | if (ret) { |
| 711 | BT_CPPLOGI_STR_SPEC(ds_file->logger, |
| 712 | "Failed to convert raw timestamp to nanoseconds since Epoch."); |
| 713 | goto end; |
| 714 | } |
| 715 | } else { |
| 716 | entry->timestamp_end = UINT64_C(-1); |
| 717 | entry->timestamp_end_ns = UINT64_C(-1); |
| 718 | } |
| 719 | |
| 720 | end: |
| 721 | return ret; |
| 722 | } |
| 723 | |
| 724 | static ctf_fs_ds_index::UP build_index_from_stream_file(struct ctf_fs_ds_file *ds_file, |
| 725 | struct ctf_fs_ds_file_info *file_info, |
| 726 | struct ctf_msg_iter *msg_iter) |
| 727 | { |
| 728 | int ret; |
| 729 | ctf_fs_ds_index::UP index; |
| 730 | enum ctf_msg_iter_status iter_status = CTF_MSG_ITER_STATUS_OK; |
| 731 | auto currentPacketOffset = bt2c::DataLen::fromBytes(0); |
| 732 | |
| 733 | BT_CPPLOGI_SPEC(ds_file->logger, "Indexing stream file {}", ds_file->file->path); |
| 734 | |
| 735 | index = ctf_fs_ds_index_create(); |
| 736 | if (!index) { |
| 737 | goto error; |
| 738 | } |
| 739 | |
| 740 | while (true) { |
| 741 | struct ctf_msg_iter_packet_properties props; |
| 742 | |
| 743 | if (currentPacketOffset.bytes() > ds_file->file->size) { |
| 744 | BT_CPPLOGE_STR_SPEC(ds_file->logger, |
| 745 | "Unexpected current packet's offset (larger than file)."); |
| 746 | goto error; |
| 747 | } else if (currentPacketOffset.bytes() == ds_file->file->size) { |
| 748 | /* No more data */ |
| 749 | break; |
| 750 | } |
| 751 | |
| 752 | iter_status = ctf_msg_iter_seek(msg_iter, currentPacketOffset.bytes()); |
| 753 | if (iter_status != CTF_MSG_ITER_STATUS_OK) { |
| 754 | goto error; |
| 755 | } |
| 756 | |
| 757 | iter_status = ctf_msg_iter_get_packet_properties(msg_iter, &props); |
| 758 | if (iter_status != CTF_MSG_ITER_STATUS_OK) { |
| 759 | goto error; |
| 760 | } |
| 761 | |
| 762 | /* |
| 763 | * Get the current packet size from the packet header, if set. Else, |
| 764 | * assume there is a single packet in the file, so take the file size |
| 765 | * as the packet size. |
| 766 | */ |
| 767 | const auto currentPacketSize = props.exp_packet_total_size >= 0 ? |
| 768 | bt2c::DataLen::fromBits(props.exp_packet_total_size) : |
| 769 | bt2c::DataLen::fromBytes(ds_file->file->size); |
| 770 | |
| 771 | if ((currentPacketOffset + currentPacketSize).bytes() > ds_file->file->size) { |
| 772 | BT_CPPLOGW_SPEC(ds_file->logger, |
| 773 | "Invalid packet size reported in file: stream=\"{}\", " |
| 774 | "packet-offset-bytes={}, packet-size-bytes={}, " |
| 775 | "file-size-bytes={}", |
| 776 | ds_file->file->path, currentPacketOffset.bytes(), |
| 777 | currentPacketSize.bytes(), ds_file->file->size); |
| 778 | goto error; |
| 779 | } |
| 780 | |
| 781 | auto index_entry = ctf_fs_ds_index_entry_create(currentPacketOffset, currentPacketSize); |
| 782 | if (!index_entry) { |
| 783 | BT_CPPLOGE_APPEND_CAUSE_SPEC(ds_file->logger, |
| 784 | "Failed to create a ctf_fs_ds_index_entry."); |
| 785 | goto error; |
| 786 | } |
| 787 | |
| 788 | /* Set path to stream file. */ |
| 789 | index_entry->path = file_info->path.c_str(); |
| 790 | |
| 791 | ret = init_index_entry(index_entry.get(), ds_file, &props); |
| 792 | if (ret) { |
| 793 | goto error; |
| 794 | } |
| 795 | |
| 796 | index->entries.emplace_back(std::move(index_entry)); |
| 797 | |
| 798 | currentPacketOffset += currentPacketSize; |
| 799 | BT_CPPLOGD_SPEC(ds_file->logger, |
| 800 | "Seeking to next packet: current-packet-offset-bytes={}, " |
| 801 | "next-packet-offset-bytes={}", |
| 802 | (currentPacketOffset - currentPacketSize).bytes(), |
| 803 | currentPacketOffset.bytes()); |
| 804 | } |
| 805 | |
| 806 | end: |
| 807 | return index; |
| 808 | |
| 809 | error: |
| 810 | index.reset(); |
| 811 | goto end; |
| 812 | } |
| 813 | |
| 814 | struct ctf_fs_ds_file *ctf_fs_ds_file_create(struct ctf_fs_trace *ctf_fs_trace, |
| 815 | bt2::Stream::Shared stream, const char *path, |
| 816 | const bt2c::Logger& parentLogger) |
| 817 | { |
| 818 | int ret; |
| 819 | size_t offset_align; |
| 820 | ctf_fs_ds_file *ds_file = new ctf_fs_ds_file {parentLogger}; |
| 821 | |
| 822 | if (!ds_file) { |
| 823 | goto error; |
| 824 | } |
| 825 | |
| 826 | ds_file->file = ctf_fs_file_create(parentLogger); |
| 827 | if (!ds_file->file) { |
| 828 | goto error; |
| 829 | } |
| 830 | |
| 831 | ds_file->stream = std::move(stream); |
| 832 | ds_file->metadata = ctf_fs_trace->metadata; |
| 833 | ds_file->file->path = path; |
| 834 | ret = ctf_fs_file_open(ds_file->file.get(), "rb"); |
| 835 | if (ret) { |
| 836 | goto error; |
| 837 | } |
| 838 | |
| 839 | offset_align = bt_mmap_get_offset_align_size(static_cast<int>(ds_file->logger.level())); |
| 840 | ds_file->mmap_max_len = offset_align * 2048; |
| 841 | |
| 842 | goto end; |
| 843 | |
| 844 | error: |
| 845 | /* Do not touch "borrowed" file. */ |
| 846 | ctf_fs_ds_file_destroy(ds_file); |
| 847 | ds_file = NULL; |
| 848 | |
| 849 | end: |
| 850 | return ds_file; |
| 851 | } |
| 852 | |
| 853 | ctf_fs_ds_index::UP ctf_fs_ds_file_build_index(struct ctf_fs_ds_file *ds_file, |
| 854 | struct ctf_fs_ds_file_info *file_info, |
| 855 | struct ctf_msg_iter *msg_iter) |
| 856 | { |
| 857 | auto index = build_index_from_idx_file(ds_file, file_info, msg_iter); |
| 858 | if (index) { |
| 859 | goto end; |
| 860 | } |
| 861 | |
| 862 | BT_CPPLOGI_SPEC(ds_file->logger, "Failed to build index from .index file; " |
| 863 | "falling back to stream indexing."); |
| 864 | index = build_index_from_stream_file(ds_file, file_info, msg_iter); |
| 865 | end: |
| 866 | return index; |
| 867 | } |
| 868 | |
| 869 | ctf_fs_ds_index::UP ctf_fs_ds_index_create() |
| 870 | { |
| 871 | return bt2s::make_unique<ctf_fs_ds_index>(); |
| 872 | } |
| 873 | |
| 874 | void ctf_fs_ds_file_destroy(struct ctf_fs_ds_file *ds_file) |
| 875 | { |
| 876 | if (!ds_file) { |
| 877 | return; |
| 878 | } |
| 879 | |
| 880 | (void) ds_file_munmap(ds_file); |
| 881 | |
| 882 | delete ds_file; |
| 883 | } |
| 884 | |
| 885 | ctf_fs_ds_file_info::UP ctf_fs_ds_file_info_create(const char *path, int64_t begin_ns) |
| 886 | { |
| 887 | ctf_fs_ds_file_info::UP ds_file_info = bt2s::make_unique<ctf_fs_ds_file_info>(); |
| 888 | |
| 889 | ds_file_info->path = path; |
| 890 | ds_file_info->begin_ns = begin_ns; |
| 891 | return ds_file_info; |
| 892 | } |
| 893 | |
| 894 | ctf_fs_ds_file_group::UP ctf_fs_ds_file_group_create(struct ctf_fs_trace *ctf_fs_trace, |
| 895 | struct ctf_stream_class *sc, |
| 896 | uint64_t stream_instance_id, |
| 897 | ctf_fs_ds_index::UP index) |
| 898 | { |
| 899 | ctf_fs_ds_file_group::UP ds_file_group {new ctf_fs_ds_file_group}; |
| 900 | |
| 901 | ds_file_group->index = std::move(index); |
| 902 | |
| 903 | ds_file_group->stream_id = stream_instance_id; |
| 904 | BT_ASSERT(sc); |
| 905 | ds_file_group->sc = sc; |
| 906 | ds_file_group->ctf_fs_trace = ctf_fs_trace; |
| 907 | |
| 908 | return ds_file_group; |
| 909 | } |