src/journal/journald-native.c

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