src/import/pull-job.c

   1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
   2
   3 #include <fcntl.h>
   4 #include <sys/stat.h>
   5 #include <sys/xattr.h>
   6
   7 #include "alloc-util.h"
   8 #include "fd-util.h"
   9 #include "format-util.h"
  10 #include "gcrypt-util.h"
  11 #include "hexdecoct.h"
  12 #include "import-util.h"
  13 #include "io-util.h"
  14 #include "machine-pool.h"
  15 #include "parse-util.h"
  16 #include "pull-common.h"
  17 #include "pull-job.h"
  18 #include "string-util.h"
  19 #include "strv.h"
  20 #include "xattr-util.h"
  21
  22 PullJob* pull_job_unref(PullJob *j) {
  23         if (!j)
  24                 return NULL;
  25
  26         curl_glue_remove_and_free(j->glue, j->curl);
  27         curl_slist_free_all(j->request_header);
  28
  29         safe_close(j->disk_fd);
  30
  31         import_compress_free(&j->compress);
  32
  33         if (j->checksum_context)
  34                 gcry_md_close(j->checksum_context);
  35
  36         free(j->url);
  37         free(j->etag);
  38         strv_free(j->old_etags);
  39         free(j->payload);
  40         free(j->checksum);
  41
  42         return mfree(j);
  43 }
  44
  45 static void pull_job_finish(PullJob *j, int ret) {
  46         assert(j);
  47
  48         if (IN_SET(j->state, PULL_JOB_DONE, PULL_JOB_FAILED))
  49                 return;
  50
  51         if (ret == 0) {
  52                 j->state = PULL_JOB_DONE;
  53                 j->progress_percent = 100;
  54                 log_info("Download of %s complete.", j->url);
  55         } else {
  56                 j->state = PULL_JOB_FAILED;
  57                 j->error = ret;
  58         }
  59
  60         if (j->on_finished)
  61                 j->on_finished(j);
  62 }
  63
  64 static int pull_job_restart(PullJob *j, const char *new_url) {
  65         int r;
  66
  67         assert(j);
  68         assert(new_url);
  69
  70         r = free_and_strdup(&j->url, new_url);
  71         if (r < 0)
  72                 return r;
  73
  74         j->state = PULL_JOB_INIT;
  75         j->error = 0;
  76         j->payload = mfree(j->payload);
  77         j->payload_size = 0;
  78         j->written_compressed = 0;
  79         j->written_uncompressed = 0;
  80         j->content_length = UINT64_MAX;
  81         j->etag = mfree(j->etag);
  82         j->etag_exists = false;
  83         j->mtime = 0;
  84         j->checksum = mfree(j->checksum);
  85
  86         curl_glue_remove_and_free(j->glue, j->curl);
  87         j->curl = NULL;
  88
  89         curl_slist_free_all(j->request_header);
  90         j->request_header = NULL;
  91
  92         import_compress_free(&j->compress);
  93
  94         if (j->checksum_context) {
  95                 gcry_md_close(j->checksum_context);
  96                 j->checksum_context = NULL;
  97         }
  98
  99         r = pull_job_begin(j);
 100         if (r < 0)
 101                 return r;
 102
 103         return 0;
 104 }
 105
 106 void pull_job_curl_on_finished(CurlGlue *g, CURL *curl, CURLcode result) {
 107         PullJob *j = NULL;
 108         CURLcode code;
 109         long status;
 110         int r;
 111
 112         if (curl_easy_getinfo(curl, CURLINFO_PRIVATE, (char **)&j) != CURLE_OK)
 113                 return;
 114
 115         if (!j || IN_SET(j->state, PULL_JOB_DONE, PULL_JOB_FAILED))
 116                 return;
 117
 118         if (result != CURLE_OK) {
 119                 log_error("Transfer failed: %s", curl_easy_strerror(result));
 120                 r = -EIO;
 121                 goto finish;
 122         }
 123
 124         code = curl_easy_getinfo(curl, CURLINFO_RESPONSE_CODE, &status);
 125         if (code != CURLE_OK) {
 126                 log_error("Failed to retrieve response code: %s", curl_easy_strerror(code));
 127                 r = -EIO;
 128                 goto finish;
 129         } else if (status == 304) {
 130                 log_info("Image already downloaded. Skipping download.");
 131                 j->etag_exists = true;
 132                 r = 0;
 133                 goto finish;
 134         } else if (status >= 300) {
 135
 136                 if (status == 404 && j->on_not_found) {
 137                         _cleanup_free_ char *new_url = NULL;
 138
 139                         /* This resource wasn't found, but the implementor wants to maybe let us know a new URL, query for it. */
 140                         r = j->on_not_found(j, &new_url);
 141                         if (r < 0)
 142                                 goto finish;
 143
 144                         if (r > 0) { /* A new url to use */
 145                                 assert(new_url);
 146
 147                                 r = pull_job_restart(j, new_url);
 148                                 if (r < 0)
 149                                         goto finish;
 150
 151                                 code = curl_easy_getinfo(j->curl, CURLINFO_RESPONSE_CODE, &status);
 152                                 if (code != CURLE_OK) {
 153                                         log_error("Failed to retrieve response code: %s", curl_easy_strerror(code));
 154                                         r = -EIO;
 155                                         goto finish;
 156                                 }
 157
 158                                 if (status == 0)
 159                                         return;
 160                         }
 161                 }
 162
 163                 log_error("HTTP request to %s failed with code %li.", j->url, status);
 164                 r = -EIO;
 165                 goto finish;
 166         } else if (status < 200) {
 167                 log_error("HTTP request to %s finished with unexpected code %li.", j->url, status);
 168                 r = -EIO;
 169                 goto finish;
 170         }
 171
 172         if (j->state != PULL_JOB_RUNNING) {
 173                 log_error("Premature connection termination.");
 174                 r = -EIO;
 175                 goto finish;
 176         }
 177
 178         if (j->content_length != UINT64_MAX &&
 179             j->content_length != j->written_compressed) {
 180                 log_error("Download truncated.");
 181                 r = -EIO;
 182                 goto finish;
 183         }
 184
 185         if (j->checksum_context) {
 186                 uint8_t *k;
 187
 188                 k = gcry_md_read(j->checksum_context, GCRY_MD_SHA256);
 189                 if (!k) {
 190                         log_error("Failed to get checksum.");
 191                         r = -EIO;
 192                         goto finish;
 193                 }
 194
 195                 j->checksum = hexmem(k, gcry_md_get_algo_dlen(GCRY_MD_SHA256));
 196                 if (!j->checksum) {
 197                         r = log_oom();
 198                         goto finish;
 199                 }
 200
 201                 log_debug("SHA256 of %s is %s.", j->url, j->checksum);
 202         }
 203
 204         if (j->disk_fd >= 0 && j->allow_sparse) {
 205                 /* Make sure the file size is right, in case the file was
 206                  * sparse and we just seeked for the last part */
 207
 208                 if (ftruncate(j->disk_fd, j->written_uncompressed) < 0) {
 209                         r = log_error_errno(errno, "Failed to truncate file: %m");
 210                         goto finish;
 211                 }
 212
 213                 if (j->etag)
 214                         (void) fsetxattr(j->disk_fd, "user.source_etag", j->etag, strlen(j->etag), 0);
 215                 if (j->url)
 216                         (void) fsetxattr(j->disk_fd, "user.source_url", j->url, strlen(j->url), 0);
 217
 218                 if (j->mtime != 0) {
 219                         struct timespec ut[2];
 220
 221                         timespec_store(&ut[0], j->mtime);
 222                         ut[1] = ut[0];
 223                         (void) futimens(j->disk_fd, ut);
 224
 225                         (void) fd_setcrtime(j->disk_fd, j->mtime);
 226                 }
 227         }
 228
 229         r = 0;
 230
 231 finish:
 232         pull_job_finish(j, r);
 233 }
 234
 235 static int pull_job_write_uncompressed(const void *p, size_t sz, void *userdata) {
 236         PullJob *j = userdata;
 237         ssize_t n;
 238
 239         assert(j);
 240         assert(p);
 241
 242         if (sz <= 0)
 243                 return 0;
 244
 245         if (j->written_uncompressed + sz < j->written_uncompressed)
 246                 return log_error_errno(SYNTHETIC_ERRNO(EOVERFLOW),
 247                                        "File too large, overflow");
 248
 249         if (j->written_uncompressed + sz > j->uncompressed_max)
 250                 return log_error_errno(SYNTHETIC_ERRNO(EFBIG),
 251                                        "File overly large, refusing");
 252
 253         if (j->disk_fd >= 0) {
 254
 255                 if (j->allow_sparse)
 256                         n = sparse_write(j->disk_fd, p, sz, 64);
 257                 else {
 258                         n = write(j->disk_fd, p, sz);
 259                         if (n < 0)
 260                                 n = -errno;
 261                 }
 262                 if (n < 0)
 263                         return log_error_errno((int) n, "Failed to write file: %m");
 264                 if ((size_t) n < sz)
 265                         return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short write");
 266         } else {
 267
 268                 if (!GREEDY_REALLOC(j->payload, j->payload_size + sz))
 269                         return log_oom();
 270
 271                 memcpy(j->payload + j->payload_size, p, sz);
 272                 j->payload_size += sz;
 273         }
 274
 275         j->written_uncompressed += sz;
 276
 277         return 0;
 278 }
 279
 280 static int pull_job_write_compressed(PullJob *j, void *p, size_t sz) {
 281         int r;
 282
 283         assert(j);
 284         assert(p);
 285
 286         if (sz <= 0)
 287                 return 0;
 288
 289         if (j->written_compressed + sz < j->written_compressed)
 290                 return log_error_errno(SYNTHETIC_ERRNO(EOVERFLOW), "File too large, overflow");
 291
 292         if (j->written_compressed + sz > j->compressed_max)
 293                 return log_error_errno(SYNTHETIC_ERRNO(EFBIG), "File overly large, refusing.");
 294
 295         if (j->content_length != UINT64_MAX &&
 296             j->written_compressed + sz > j->content_length)
 297                 return log_error_errno(SYNTHETIC_ERRNO(EFBIG),
 298                                        "Content length incorrect.");
 299
 300         if (j->checksum_context)
 301                 gcry_md_write(j->checksum_context, p, sz);
 302
 303         r = import_uncompress(&j->compress, p, sz, pull_job_write_uncompressed, j);
 304         if (r < 0)
 305                 return r;
 306
 307         j->written_compressed += sz;
 308
 309         return 0;
 310 }
 311
 312 static int pull_job_open_disk(PullJob *j) {
 313         int r;
 314
 315         assert(j);
 316
 317         if (j->on_open_disk) {
 318                 r = j->on_open_disk(j);
 319                 if (r < 0)
 320                         return r;
 321         }
 322
 323         if (j->disk_fd >= 0) {
 324                 /* Check if we can do sparse files */
 325
 326                 if (lseek(j->disk_fd, SEEK_SET, 0) == 0)
 327                         j->allow_sparse = true;
 328                 else {
 329                         if (errno != ESPIPE)
 330                                 return log_error_errno(errno, "Failed to seek on file descriptor: %m");
 331
 332                         j->allow_sparse = false;
 333                 }
 334         }
 335
 336         if (j->calc_checksum) {
 337                 initialize_libgcrypt(false);
 338
 339                 if (gcry_md_open(&j->checksum_context, GCRY_MD_SHA256, 0) != 0)
 340                         return log_error_errno(SYNTHETIC_ERRNO(EIO),
 341                                                "Failed to initialize hash context.");
 342         }
 343
 344         return 0;
 345 }
 346
 347 static int pull_job_detect_compression(PullJob *j) {
 348         _cleanup_free_ uint8_t *stub = NULL;
 349         size_t stub_size;
 350
 351         int r;
 352
 353         assert(j);
 354
 355         r = import_uncompress_detect(&j->compress, j->payload, j->payload_size);
 356         if (r < 0)
 357                 return log_error_errno(r, "Failed to initialize compressor: %m");
 358         if (r == 0)
 359                 return 0;
 360
 361         log_debug("Stream is compressed: %s", import_compress_type_to_string(j->compress.type));
 362
 363         r = pull_job_open_disk(j);
 364         if (r < 0)
 365                 return r;
 366
 367         /* Now, take the payload we read so far, and decompress it */
 368         stub = j->payload;
 369         stub_size = j->payload_size;
 370
 371         j->payload = NULL;
 372         j->payload_size = 0;
 373
 374         j->state = PULL_JOB_RUNNING;
 375
 376         r = pull_job_write_compressed(j, stub, stub_size);
 377         if (r < 0)
 378                 return r;
 379
 380         return 0;
 381 }
 382
 383 static size_t pull_job_write_callback(void *contents, size_t size, size_t nmemb, void *userdata) {
 384         PullJob *j = userdata;
 385         size_t sz = size * nmemb;
 386         int r;
 387
 388         assert(contents);
 389         assert(j);
 390
 391         switch (j->state) {
 392
 393         case PULL_JOB_ANALYZING:
 394                 /* Let's first check what it actually is */
 395
 396                 if (!GREEDY_REALLOC(j->payload, j->payload_size + sz)) {
 397                         r = log_oom();
 398                         goto fail;
 399                 }
 400
 401                 memcpy(j->payload + j->payload_size, contents, sz);
 402                 j->payload_size += sz;
 403
 404                 r = pull_job_detect_compression(j);
 405                 if (r < 0)
 406                         goto fail;
 407
 408                 break;
 409
 410         case PULL_JOB_RUNNING:
 411
 412                 r = pull_job_write_compressed(j, contents, sz);
 413                 if (r < 0)
 414                         goto fail;
 415
 416                 break;
 417
 418         case PULL_JOB_DONE:
 419         case PULL_JOB_FAILED:
 420                 r = -ESTALE;
 421                 goto fail;
 422
 423         default:
 424                 assert_not_reached("Impossible state.");
 425         }
 426
 427         return sz;
 428
 429 fail:
 430         pull_job_finish(j, r);
 431         return 0;
 432 }
 433
 434 static int http_status_ok(CURLcode status) {
 435         /* Consider all HTTP status code in the 2xx range as OK */
 436         return status >= 200 && status <= 299;
 437 }
 438
 439 static int http_status_etag_exists(CURLcode status) {
 440         /* This one is special, it's triggered by our etag mgmt logic */
 441         return status == 304;
 442 }
 443
 444 static size_t pull_job_header_callback(void *contents, size_t size, size_t nmemb, void *userdata) {
 445         _cleanup_free_ char *length = NULL, *last_modified = NULL, *etag = NULL;
 446         size_t sz = size * nmemb;
 447         PullJob *j = userdata;
 448         CURLcode code;
 449         long status;
 450         int r;
 451
 452         assert(contents);
 453         assert(j);
 454
 455         if (IN_SET(j->state, PULL_JOB_DONE, PULL_JOB_FAILED)) {
 456                 r = -ESTALE;
 457                 goto fail;
 458         }
 459
 460         assert(j->state == PULL_JOB_ANALYZING);
 461
 462         code = curl_easy_getinfo(j->curl, CURLINFO_RESPONSE_CODE, &status);
 463         if (code != CURLE_OK) {
 464                 log_error("Failed to retrieve response code: %s", curl_easy_strerror(code));
 465                 r = -EIO;
 466                 goto fail;
 467         }
 468
 469         if (http_status_ok(status) || http_status_etag_exists(status)) {
 470                 /* Check Etag on OK and etag exists responses. */
 471
 472                 r = curl_header_strdup(contents, sz, "ETag:", &etag);
 473                 if (r < 0) {
 474                         log_oom();
 475                         goto fail;
 476                 }
 477                 if (r > 0) {
 478                         free_and_replace(j->etag, etag);
 479
 480                         if (strv_contains(j->old_etags, j->etag)) {
 481                                 log_info("Image already downloaded. Skipping download. (%s)", j->etag);
 482                                 j->etag_exists = true;
 483                                 pull_job_finish(j, 0);
 484                                 return sz;
 485                         }
 486
 487                         return sz;
 488                 }
 489         }
 490
 491         if (!http_status_ok(status)) /* Let's ignore the rest here, these requests are probably redirects and
 492                                       * stuff where the headers aren't interesting to us */
 493                 return sz;
 494
 495         r = curl_header_strdup(contents, sz, "Content-Length:", &length);
 496         if (r < 0) {
 497                 log_oom();
 498                 goto fail;
 499         }
 500         if (r > 0) {
 501                 (void) safe_atou64(length, &j->content_length);
 502
 503                 if (j->content_length != UINT64_MAX) {
 504                         if (j->content_length > j->compressed_max) {
 505                                 r = log_error_errno(SYNTHETIC_ERRNO(EFBIG), "Content too large.");
 506                                 goto fail;
 507                         }
 508
 509                         log_info("Downloading %s for %s.", FORMAT_BYTES(j->content_length), j->url);
 510                 }
 511
 512                 return sz;
 513         }
 514
 515         r = curl_header_strdup(contents, sz, "Last-Modified:", &last_modified);
 516         if (r < 0) {
 517                 log_oom();
 518                 goto fail;
 519         }
 520         if (r > 0) {
 521                 (void) curl_parse_http_time(last_modified, &j->mtime);
 522                 return sz;
 523         }
 524
 525         if (j->on_header) {
 526                 r = j->on_header(j, contents, sz);
 527                 if (r < 0)
 528                         goto fail;
 529         }
 530
 531         return sz;
 532
 533 fail:
 534         pull_job_finish(j, r);
 535         return 0;
 536 }
 537
 538 static int pull_job_progress_callback(void *userdata, curl_off_t dltotal, curl_off_t dlnow, curl_off_t ultotal, curl_off_t ulnow) {
 539         PullJob *j = userdata;
 540         unsigned percent;
 541         usec_t n;
 542
 543         assert(j);
 544
 545         if (dltotal <= 0)
 546                 return 0;
 547
 548         percent = ((100 * dlnow) / dltotal);
 549         n = now(CLOCK_MONOTONIC);
 550
 551         if (n > j->last_status_usec + USEC_PER_SEC &&
 552             percent != j->progress_percent &&
 553             dlnow < dltotal) {
 554
 555                 if (n - j->start_usec > USEC_PER_SEC && dlnow > 0) {
 556                         usec_t left, done;
 557
 558                         done = n - j->start_usec;
 559                         left = (usec_t) (((double) done * (double) dltotal) / dlnow) - done;
 560
 561                         log_info("Got %u%% of %s. %s left at %s/s.",
 562                                  percent,
 563                                  j->url,
 564                                  FORMAT_TIMESPAN(left, USEC_PER_SEC),
 565                                  FORMAT_BYTES((uint64_t) ((double) dlnow / ((double) done / (double) USEC_PER_SEC))));
 566                 } else
 567                         log_info("Got %u%% of %s.", percent, j->url);
 568
 569                 j->progress_percent = percent;
 570                 j->last_status_usec = n;
 571
 572                 if (j->on_progress)
 573                         j->on_progress(j);
 574         }
 575
 576         return 0;
 577 }
 578
 579 int pull_job_new(PullJob **ret, const char *url, CurlGlue *glue, void *userdata) {
 580         _cleanup_(pull_job_unrefp) PullJob *j = NULL;
 581         _cleanup_free_ char *u = NULL;
 582
 583         assert(url);
 584         assert(glue);
 585         assert(ret);
 586
 587         u = strdup(url);
 588         if (!u)
 589                 return -ENOMEM;
 590
 591         j = new(PullJob, 1);
 592         if (!j)
 593                 return -ENOMEM;
 594
 595         *j = (PullJob) {
 596                 .state = PULL_JOB_INIT,
 597                 .disk_fd = -1,
 598                 .userdata = userdata,
 599                 .glue = glue,
 600                 .content_length = UINT64_MAX,
 601                 .start_usec = now(CLOCK_MONOTONIC),
 602                 .compressed_max = 64LLU * 1024LLU * 1024LLU * 1024LLU, /* 64GB safety limit */
 603                 .uncompressed_max = 64LLU * 1024LLU * 1024LLU * 1024LLU, /* 64GB safety limit */
 604                 .url = TAKE_PTR(u),
 605         };
 606
 607         *ret = TAKE_PTR(j);
 608
 609         return 0;
 610 }
 611
 612 int pull_job_begin(PullJob *j) {
 613         int r;
 614
 615         assert(j);
 616
 617         if (j->state != PULL_JOB_INIT)
 618                 return -EBUSY;
 619
 620         r = curl_glue_make(&j->curl, j->url, j);
 621         if (r < 0)
 622                 return r;
 623
 624         if (!strv_isempty(j->old_etags)) {
 625                 _cleanup_free_ char *cc = NULL, *hdr = NULL;
 626
 627                 cc = strv_join(j->old_etags, ", ");
 628                 if (!cc)
 629                         return -ENOMEM;
 630
 631                 hdr = strjoin("If-None-Match: ", cc);
 632                 if (!hdr)
 633                         return -ENOMEM;
 634
 635                 if (!j->request_header) {
 636                         j->request_header = curl_slist_new(hdr, NULL);
 637                         if (!j->request_header)
 638                                 return -ENOMEM;
 639                 } else {
 640                         struct curl_slist *l;
 641
 642                         l = curl_slist_append(j->request_header, hdr);
 643                         if (!l)
 644                                 return -ENOMEM;
 645
 646                         j->request_header = l;
 647                 }
 648         }
 649
 650         if (j->request_header) {
 651                 if (curl_easy_setopt(j->curl, CURLOPT_HTTPHEADER, j->request_header) != CURLE_OK)
 652                         return -EIO;
 653         }
 654
 655         if (curl_easy_setopt(j->curl, CURLOPT_WRITEFUNCTION, pull_job_write_callback) != CURLE_OK)
 656                 return -EIO;
 657
 658         if (curl_easy_setopt(j->curl, CURLOPT_WRITEDATA, j) != CURLE_OK)
 659                 return -EIO;
 660
 661         if (curl_easy_setopt(j->curl, CURLOPT_HEADERFUNCTION, pull_job_header_callback) != CURLE_OK)
 662                 return -EIO;
 663
 664         if (curl_easy_setopt(j->curl, CURLOPT_HEADERDATA, j) != CURLE_OK)
 665                 return -EIO;
 666
 667         if (curl_easy_setopt(j->curl, CURLOPT_XFERINFOFUNCTION, pull_job_progress_callback) != CURLE_OK)
 668                 return -EIO;
 669
 670         if (curl_easy_setopt(j->curl, CURLOPT_XFERINFODATA, j) != CURLE_OK)
 671                 return -EIO;
 672
 673         if (curl_easy_setopt(j->curl, CURLOPT_NOPROGRESS, 0) != CURLE_OK)
 674                 return -EIO;
 675
 676         r = curl_glue_add(j->glue, j->curl);
 677         if (r < 0)
 678                 return r;
 679
 680         j->state = PULL_JOB_ANALYZING;
 681
 682         return 0;
 683 }