src/journal/journald-native.c

   1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
   2
   3 #include <sys/statvfs.h>
   4 #include <unistd.h>
   5
   6 #include "sd-event.h"
   7
   8 #include "alloc-util.h"
   9 #include "errno-util.h"
  10 #include "fd-util.h"
  11 #include "format-util.h"
  12 #include "iovec-util.h"
  13 #include "journal-importer.h"
  14 #include "journal-internal.h"
  15 #include "journald-client.h"
  16 #include "journald-console.h"
  17 #include "journald-context.h"
  18 #include "journald-kmsg.h"
  19 #include "journald-manager.h"
  20 #include "journald-native.h"
  21 #include "journald-syslog.h"
  22 #include "journald-wall.h"
  23 #include "log.h"
  24 #include "log-ratelimit.h"
  25 #include "memfd-util.h"
  26 #include "memory-util.h"
  27 #include "parse-util.h"
  28 #include "path-util.h"
  29 #include "process-util.h"
  30 #include "selinux-util.h"
  31 #include "socket-util.h"
  32 #include "stat-util.h"
  33 #include "string-util.h"
  34 #include "unaligned.h"
  35
  36 static bool allow_object_pid(const struct ucred *ucred) {
  37         return ucred && ucred->uid == 0;
  38 }
  39
  40 static void manager_process_entry_meta(
  41                 const char *p, size_t l,
  42                 const struct ucred *ucred,
  43                 int *priority,
  44                 char **identifier,
  45                 char **message,
  46                 pid_t *object_pid) {
  47
  48         /* We need to determine the priority of this entry for the rate limiting logic */
  49
  50         if (l == 10 &&
  51             startswith(p, "PRIORITY=") &&
  52             p[9] >= '0' && p[9] <= '9')
  53                 *priority = (*priority & LOG_FACMASK) | (p[9] - '0');
  54
  55         else if (l == 17 &&
  56                  startswith(p, "SYSLOG_FACILITY=") &&
  57                  p[16] >= '0' && p[16] <= '9')
  58                 *priority = LOG_PRI(*priority) | ((p[16] - '0') << 3);
  59
  60         else if (l == 18 &&
  61                  startswith(p, "SYSLOG_FACILITY=") &&
  62                  p[16] >= '0' && p[16] <= '9' &&
  63                  p[17] >= '0' && p[17] <= '9')
  64                 *priority = LOG_PRI(*priority) | (((p[16] - '0')*10 + (p[17] - '0')) << 3);
  65
  66         else if (l >= 19 &&
  67                  startswith(p, "SYSLOG_IDENTIFIER=")) {
  68                 char *t;
  69
  70                 t = memdup_suffix0(p + 18, l - 18);
  71                 if (t)
  72                         free_and_replace(*identifier, t);
  73
  74         } else if (l >= 8 &&
  75                    startswith(p, "MESSAGE=")) {
  76                 char *t;
  77
  78                 t = memdup_suffix0(p + 8, l - 8);
  79                 if (t)
  80                         free_and_replace(*message, t);
  81
  82         } else if (l > STRLEN("OBJECT_PID=") &&
  83                    l < STRLEN("OBJECT_PID=")  + DECIMAL_STR_MAX(pid_t) &&
  84                    startswith(p, "OBJECT_PID=") &&
  85                    allow_object_pid(ucred)) {
  86                 char buf[DECIMAL_STR_MAX(pid_t)];
  87                 memcpy(buf, p + STRLEN("OBJECT_PID="),
  88                        l - STRLEN("OBJECT_PID="));
  89                 buf[l-STRLEN("OBJECT_PID=")] = '\0';
  90
  91                 (void) parse_pid(buf, object_pid);
  92         }
  93 }
  94
  95 static int manager_process_entry(
  96                 Manager *m,
  97                 const void *buffer, size_t *remaining,
  98                 ClientContext *context,
  99                 const struct ucred *ucred,
 100                 const struct timeval *tv,
 101                 const char *label, size_t label_len) {
 102
 103         /* Process a single entry from a native message. Returns 0 if nothing special happened and the message
 104          * processing should continue, and a negative or positive value otherwise.
 105          *
 106          * Note that *remaining is altered on both success and failure. */
 107
 108         size_t n = 0, j, tn = SIZE_MAX, entry_size = 0;
 109         char *identifier = NULL, *message = NULL;
 110         struct iovec *iovec = NULL;
 111         int priority = LOG_INFO;
 112         pid_t object_pid = 0;
 113         const char *p;
 114         int r = 1;
 115
 116         p = buffer;
 117
 118         while (*remaining > 0) {
 119                 const char *e, *q;
 120
 121                 e = memchr(p, '\n', *remaining);
 122
 123                 if (!e) {
 124                         /* Trailing noise, let's ignore it, and flush what we collected */
 125                         log_debug("Received message with trailing noise, ignoring.");
 126                         break; /* finish processing of the message */
 127                 }
 128
 129                 if (e == p) {
 130                         /* Entry separator */
 131                         *remaining -= 1;
 132                         break;
 133                 }
 134
 135                 if (IN_SET(*p, '.', '#')) {
 136                         /* Ignore control commands for now, and comments too. */
 137                         *remaining -= (e - p) + 1;
 138                         p = e + 1;
 139                         continue;
 140                 }
 141
 142                 /* A property follows */
 143                 if (n > ENTRY_FIELD_COUNT_MAX) {
 144                         log_debug("Received an entry that has more than " STRINGIFY(ENTRY_FIELD_COUNT_MAX) " fields, ignoring entry.");
 145                         goto finish;
 146                 }
 147
 148                 /* n existing properties, 1 new, +1 for _TRANSPORT */
 149                 if (!GREEDY_REALLOC(iovec,
 150                                     n + 2 +
 151                                     N_IOVEC_META_FIELDS + N_IOVEC_OBJECT_FIELDS +
 152                                     client_context_extra_fields_n_iovec(context))) {
 153                         r = log_oom();
 154                         goto finish;
 155                 }
 156
 157                 q = memchr(p, '=', e - p);
 158                 if (q) {
 159                         if (journal_field_valid(p, q - p, false)) {
 160                                 size_t l;
 161
 162                                 l = e - p;
 163                                 if (l > DATA_SIZE_MAX) {
 164                                         log_debug("Received text block of %zu bytes is too large, ignoring entry.", l);
 165                                         goto finish;
 166                                 }
 167
 168                                 if (entry_size + l + n + 1 > ENTRY_SIZE_MAX) { /* data + separators + trailer */
 169                                         log_debug("Entry is too big (%zu bytes after processing %zu entries), ignoring entry.",
 170                                                   entry_size + l, n + 1);
 171                                         goto finish;
 172                                 }
 173
 174                                 /* If the field name starts with an underscore, skip the variable, since that indicates
 175                                  * a trusted field */
 176                                 iovec[n++] = IOVEC_MAKE((char*) p, l);
 177                                 entry_size += l;
 178
 179                                 manager_process_entry_meta(p, l, ucred,
 180                                                           &priority,
 181                                                           &identifier,
 182                                                           &message,
 183                                                           &object_pid);
 184                         }
 185
 186                         *remaining -= (e - p) + 1;
 187                         p = e + 1;
 188                         continue;
 189                 } else {
 190                         uint64_t l, total;
 191                         char *k;
 192
 193                         if (*remaining < e - p + 1 + sizeof(uint64_t) + 1) {
 194                                 log_debug("Failed to parse message, ignoring.");
 195                                 break;
 196                         }
 197
 198                         l = unaligned_read_le64(e + 1);
 199                         if (l > DATA_SIZE_MAX) {
 200                                 log_debug("Received binary data block of %"PRIu64" bytes is too large, ignoring entry.", l);
 201                                 goto finish;
 202                         }
 203
 204                         total = (e - p) + 1 + l;
 205                         if (entry_size + total + n + 1 > ENTRY_SIZE_MAX) { /* data + separators + trailer */
 206                                 log_debug("Entry is too big (%"PRIu64"bytes after processing %zu fields), ignoring.",
 207                                           entry_size + total, n + 1);
 208                                 goto finish;
 209                         }
 210
 211                         if ((uint64_t) *remaining < e - p + 1 + sizeof(uint64_t) + l + 1 ||
 212                             e[1+sizeof(uint64_t)+l] != '\n') {
 213                                 log_debug("Failed to parse message, ignoring.");
 214                                 break;
 215                         }
 216
 217                         k = malloc(total);
 218                         if (!k) {
 219                                 log_oom();
 220                                 break;
 221                         }
 222
 223                         memcpy(k, p, e - p);
 224                         k[e - p] = '=';
 225                         memcpy(k + (e - p) + 1, e + 1 + sizeof(uint64_t), l);
 226
 227                         if (journal_field_valid(p, e - p, false)) {
 228                                 iovec[n] = IOVEC_MAKE(k, (e - p) + 1 + l);
 229                                 entry_size += iovec[n].iov_len;
 230                                 n++;
 231
 232                                 manager_process_entry_meta(k, (e - p) + 1 + l, ucred,
 233                                                           &priority,
 234                                                           &identifier,
 235                                                           &message,
 236                                                           &object_pid);
 237                         } else
 238                                 free(k);
 239
 240                         *remaining -= (e - p) + 1 + sizeof(uint64_t) + l + 1;
 241                         p = e + 1 + sizeof(uint64_t) + l + 1;
 242                 }
 243         }
 244
 245         if (n <= 0)
 246                 goto finish;
 247
 248         tn = n++;
 249         iovec[tn] = IOVEC_MAKE_STRING("_TRANSPORT=journal");
 250         entry_size += STRLEN("_TRANSPORT=journal");
 251
 252         if (entry_size + n + 1 > ENTRY_SIZE_MAX) { /* data + separators + trailer */
 253                 log_debug("Entry is too big with %zu properties and %zu bytes, ignoring.", n, entry_size);
 254                 goto finish;
 255         }
 256
 257         r = 0; /* Success, we read the message. */
 258
 259         if (!client_context_test_priority(context, priority))
 260                 goto finish;
 261
 262         if (message) {
 263                 /* Ensure message is not NULL, otherwise strlen(message) would crash. This check needs to
 264                  * be here until manager_process_entry() is able to process messages containing \0 characters,
 265                  * as we would have access to the actual size of message. */
 266                 r = client_context_check_keep_log(context, message, strlen(message));
 267                 if (r <= 0)
 268                         goto finish;
 269
 270                 if (m->forward_to_syslog)
 271                         manager_forward_syslog(m, syslog_fixup_facility(priority), identifier, message, ucred, tv);
 272
 273                 if (m->forward_to_kmsg)
 274                         manager_forward_kmsg(m, priority, identifier, message, ucred);
 275
 276                 if (m->forward_to_console)
 277                         manager_forward_console(m, priority, identifier, message, ucred);
 278
 279                 if (m->forward_to_wall)
 280                         manager_forward_wall(m, priority, identifier, message, ucred);
 281         }
 282
 283         manager_dispatch_message(m, iovec, n, MALLOC_ELEMENTSOF(iovec), context, tv, priority, object_pid);
 284
 285 finish:
 286         for (j = 0; j < n; j++)  {
 287                 if (j == tn)
 288                         continue;
 289
 290                 if (iovec[j].iov_base < buffer ||
 291                     (const char*) iovec[j].iov_base >= p + *remaining)
 292                         free(iovec[j].iov_base);
 293         }
 294
 295         free(iovec);
 296         free(identifier);
 297         free(message);
 298
 299         return r;
 300 }
 301
 302 void manager_process_native_message(
 303                 Manager *m,
 304                 const char *buffer, size_t buffer_size,
 305                 const struct ucred *ucred,
 306                 const struct timeval *tv,
 307                 const char *label, size_t label_len) {
 308
 309         size_t remaining = buffer_size;
 310         ClientContext *context = NULL;
 311         int r;
 312
 313         assert(m);
 314         assert(buffer || buffer_size == 0);
 315
 316         if (ucred && pid_is_valid(ucred->pid)) {
 317                 r = client_context_get(m, ucred->pid, ucred, label, label_len, NULL, &context);
 318                 if (r < 0)
 319                         log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT,
 320                                                     "Failed to retrieve credentials for PID " PID_FMT ", ignoring: %m",
 321                                                     ucred->pid);
 322         }
 323
 324         do {
 325                 r = manager_process_entry(m,
 326                                          (const uint8_t*) buffer + (buffer_size - remaining), &remaining,
 327                                          context, ucred, tv, label, label_len);
 328         } while (r == 0);
 329 }
 330
 331 int manager_process_native_file(
 332                 Manager *m,
 333                 int fd,
 334                 const struct ucred *ucred,
 335                 const struct timeval *tv,
 336                 const char *label, size_t label_len) {
 337
 338         struct stat st;
 339         bool sealed;
 340         int r;
 341
 342         /* Data is in the passed fd, probably it didn't fit in a datagram. */
 343
 344         assert(m);
 345         assert(fd >= 0);
 346
 347         if (fstat(fd, &st) < 0)
 348                 return log_ratelimit_error_errno(errno, JOURNAL_LOG_RATELIMIT,
 349                                                  "Failed to stat passed file: %m");
 350
 351         r = stat_verify_regular(&st);
 352         if (r < 0)
 353                 return log_ratelimit_error_errno(r, JOURNAL_LOG_RATELIMIT,
 354                                                  "File passed is not regular, ignoring message: %m");
 355
 356         if (st.st_size <= 0)
 357                 return 0;
 358
 359         r = fd_verify_safe_flags(fd);
 360         if (r == -EREMOTEIO)
 361                 return log_ratelimit_error_errno(r, JOURNAL_LOG_RATELIMIT,
 362                                                  "Unexpected flags of passed memory fd, ignoring message.");
 363         if (r < 0)
 364                 return log_ratelimit_error_errno(r, JOURNAL_LOG_RATELIMIT,
 365                                                  "Failed to get flags of passed file: %m");
 366
 367         /* If it's a memfd, check if it is sealed. If so, we can just mmap it and use it, and do not need to
 368          * copy the data out. */
 369         sealed = memfd_get_sealed(fd) > 0;
 370
 371         if (!sealed && (!ucred || ucred->uid != 0)) {
 372                 _cleanup_free_ char *k = NULL;
 373                 const char *e;
 374
 375                 /* If this is not a sealed memfd, and the peer is unknown or unprivileged, then verify the
 376                  * path. */
 377
 378                 r = fd_get_path(fd, &k);
 379                 if (r < 0)
 380                         return log_ratelimit_error_errno(r, JOURNAL_LOG_RATELIMIT,
 381                                                          "Failed to get path of passed fd: %m");
 382
 383                 e = PATH_STARTSWITH_SET(k, "/dev/shm/", "/tmp/", "/var/tmp/");
 384                 if (!e)
 385                         return log_ratelimit_error_errno(SYNTHETIC_ERRNO(EPERM), JOURNAL_LOG_RATELIMIT,
 386                                                          "Received file outside of allowed directories, refusing.");
 387
 388                 if (!filename_is_valid(e))
 389                         return log_ratelimit_error_errno(SYNTHETIC_ERRNO(EPERM), JOURNAL_LOG_RATELIMIT,
 390                                                          "Received file in subdirectory of allowed directories, refusing.");
 391         }
 392
 393         /* When !sealed, set a lower memory limit. We have to read the file, effectively doubling memory
 394          * use. */
 395         if (st.st_size > ENTRY_SIZE_MAX / (sealed ? 1 : 2))
 396                 return log_ratelimit_error_errno(SYNTHETIC_ERRNO(EFBIG), JOURNAL_LOG_RATELIMIT,
 397                                                  "File passed too large (%"PRIu64" bytes), refusing.",
 398                                                  (uint64_t) st.st_size);
 399
 400         if (sealed) {
 401                 void *p;
 402                 size_t ps;
 403
 404                 /* The file is sealed, we can just map it and use it. */
 405
 406                 ps = PAGE_ALIGN(st.st_size);
 407                 assert(ps < SIZE_MAX);
 408                 p = mmap(NULL, ps, PROT_READ, MAP_PRIVATE, fd, 0);
 409                 if (p == MAP_FAILED)
 410                         return log_ratelimit_error_errno(errno, JOURNAL_LOG_RATELIMIT,
 411                                                          "Failed to map memfd: %m");
 412
 413                 manager_process_native_message(m, p, st.st_size, ucred, tv, label, label_len);
 414                 assert_se(munmap(p, ps) >= 0);
 415
 416                 return 0;
 417         }
 418
 419         _cleanup_free_ void *p = NULL;
 420         struct statvfs vfs;
 421         ssize_t n;
 422
 423         if (fstatvfs(fd, &vfs) < 0)
 424                 return log_ratelimit_error_errno(errno, JOURNAL_LOG_RATELIMIT,
 425                                                  "Failed to stat file system of passed file: %m");
 426
 427         /* Refuse operating on file systems that have mandatory locking enabled.
 428          * See also: https://github.com/systemd/systemd/issues/1822 */
 429         if (FLAGS_SET(vfs.f_flag, ST_MANDLOCK))
 430                 return log_ratelimit_error_errno(SYNTHETIC_ERRNO(EPERM), JOURNAL_LOG_RATELIMIT,
 431                                                  "Received file descriptor from file system with mandatory locking enabled, not processing it.");
 432
 433         /* Make the fd non-blocking. On regular files this has the effect of bypassing mandatory
 434          * locking. Of course, this should normally not be necessary given the check above, but let's
 435          * better be safe than sorry, after all NFS is pretty confusing regarding file system flags,
 436          * and we better don't trust it, and so is SMB. */
 437         r = fd_nonblock(fd, true);
 438         if (r < 0)
 439                 return log_ratelimit_error_errno(r, JOURNAL_LOG_RATELIMIT,
 440                                                  "Failed to make fd non-blocking: %m");
 441
 442         /* The file is not sealed, we can't map the file here, since clients might then truncate it
 443          * and trigger a SIGBUS for us. So let's stupidly read it. */
 444
 445         p = malloc(st.st_size);
 446         if (!p)
 447                 return log_oom();
 448
 449         n = pread(fd, p, st.st_size, 0);
 450         if (n < 0)
 451                 return log_ratelimit_error_errno(errno, JOURNAL_LOG_RATELIMIT,
 452                                                  "Failed to read file: %m");
 453         if (n > 0)
 454                 manager_process_native_message(m, p, n, ucred, tv, label, label_len);
 455
 456         return 0;
 457 }
 458
 459 int manager_open_native_socket(Manager *m, const char *native_socket) {
 460         int r;
 461
 462         assert(m);
 463         assert(native_socket);
 464
 465         if (m->native_fd < 0) {
 466                 union sockaddr_union sa;
 467                 size_t sa_len;
 468
 469                 r = sockaddr_un_set_path(&sa.un, native_socket);
 470                 if (r < 0)
 471                         return log_error_errno(r, "Unable to use namespace path %s for AF_UNIX socket: %m", native_socket);
 472                 sa_len = r;
 473
 474                 m->native_fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
 475                 if (m->native_fd < 0)
 476                         return log_error_errno(errno, "socket() failed: %m");
 477
 478                 (void) sockaddr_un_unlink(&sa.un);
 479
 480                 r = bind(m->native_fd, &sa.sa, sa_len);
 481                 if (r < 0)
 482                         return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
 483
 484                 (void) chmod(sa.un.sun_path, 0666);
 485         } else
 486                 (void) fd_nonblock(m->native_fd, true);
 487
 488         r = setsockopt_int(m->native_fd, SOL_SOCKET, SO_PASSCRED, true);
 489         if (r < 0)
 490                 return log_error_errno(r, "SO_PASSCRED failed: %m");
 491
 492         if (mac_selinux_use()) {
 493                 r = setsockopt_int(m->native_fd, SOL_SOCKET, SO_PASSSEC, true);
 494                 if (r < 0)
 495                         log_full_errno(ERRNO_IS_NEG_NOT_SUPPORTED(r) ? LOG_DEBUG : LOG_WARNING, r, "SO_PASSSEC failed, ignoring: %m");
 496         }
 497
 498         r = setsockopt_int(m->native_fd, SOL_SOCKET, SO_TIMESTAMP, true);
 499         if (r < 0)
 500                 return log_error_errno(r, "SO_TIMESTAMP failed: %m");
 501
 502         r = sd_event_add_io(m->event, &m->native_event_source, m->native_fd, EPOLLIN, manager_process_datagram, m);
 503         if (r < 0)
 504                 return log_error_errno(r, "Failed to add native manager fd to event loop: %m");
 505
 506         r = sd_event_source_set_priority(m->native_event_source, SD_EVENT_PRIORITY_NORMAL+5);
 507         if (r < 0)
 508                 return log_error_errno(r, "Failed to adjust native event source priority: %m");
 509
 510         return 0;
 511 }