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 "strv.h"
  29 #include "unaligned.h"
  30
  31 static bool allow_object_pid(const struct ucred *ucred) {
  32         return ucred && ucred->uid == 0;
  33 }
  34
  35 static void server_process_entry_meta(
  36                 const char *p, size_t l,
  37                 const struct ucred *ucred,
  38                 int *priority,
  39                 char **identifier,
  40                 char **message,
  41                 pid_t *object_pid) {
  42
  43         /* We need to determine the priority of this entry for the rate limiting logic */
  44
  45         if (l == 10 &&
  46             startswith(p, "PRIORITY=") &&
  47             p[9] >= '0' && p[9] <= '9')
  48                 *priority = (*priority & LOG_FACMASK) | (p[9] - '0');
  49
  50         else if (l == 17 &&
  51                  startswith(p, "SYSLOG_FACILITY=") &&
  52                  p[16] >= '0' && p[16] <= '9')
  53                 *priority = (*priority & LOG_PRIMASK) | ((p[16] - '0') << 3);
  54
  55         else if (l == 18 &&
  56                  startswith(p, "SYSLOG_FACILITY=") &&
  57                  p[16] >= '0' && p[16] <= '9' &&
  58                  p[17] >= '0' && p[17] <= '9')
  59                 *priority = (*priority & LOG_PRIMASK) | (((p[16] - '0')*10 + (p[17] - '0')) << 3);
  60
  61         else if (l >= 19 &&
  62                  startswith(p, "SYSLOG_IDENTIFIER=")) {
  63                 char *t;
  64
  65                 t = strndup(p + 18, l - 18);
  66                 if (t) {
  67                         free(*identifier);
  68                         *identifier = t;
  69                 }
  70
  71         } else if (l >= 8 &&
  72                    startswith(p, "MESSAGE=")) {
  73                 char *t;
  74
  75                 t = strndup(p + 8, l - 8);
  76                 if (t) {
  77                         free(*message);
  78                         *message = t;
  79                 }
  80
  81         } else if (l > STRLEN("OBJECT_PID=") &&
  82                    l < STRLEN("OBJECT_PID=")  + DECIMAL_STR_MAX(pid_t) &&
  83                    startswith(p, "OBJECT_PID=") &&
  84                    allow_object_pid(ucred)) {
  85                 char buf[DECIMAL_STR_MAX(pid_t)];
  86                 memcpy(buf, p + STRLEN("OBJECT_PID="),
  87                        l - STRLEN("OBJECT_PID="));
  88                 buf[l-STRLEN("OBJECT_PID=")] = '\0';
  89
  90                 (void) parse_pid(buf, object_pid);
  91         }
  92 }
  93
  94 static int server_process_entry(
  95                 Server *s,
  96                 const void *buffer, size_t *remaining,
  97                 ClientContext *context,
  98                 const struct ucred *ucred,
  99                 const struct timeval *tv,
 100                 const char *label, size_t label_len) {
 101
 102         /* Process a single entry from a native message. Returns 0 if nothing special happened and the message
 103          * processing should continue, and a negative or positive value otherwise.
 104          *
 105          * Note that *remaining is altered on both success and failure. */
 106
 107         size_t n = 0, j, tn = (size_t) -1, m = 0, entry_size = 0;
 108         char *identifier = NULL, *message = NULL;
 109         struct iovec *iovec = NULL;
 110         int priority = LOG_INFO;
 111         pid_t object_pid = 0;
 112         const char *p;
 113         int r = 0;
 114
 115         p = buffer;
 116
 117         while (*remaining > 0) {
 118                 const char *e, *q;
 119
 120                 e = memchr(p, '\n', *remaining);
 121
 122                 if (!e) {
 123                         /* Trailing noise, let's ignore it, and flush what we collected */
 124                         log_debug("Received message with trailing noise, ignoring.");
 125                         r = 1; /* finish processing of the message */
 126                         break;
 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
 137                          * comments too. */
 138                         *remaining -= (e - p) + 1;
 139                         p = e + 1;
 140                         continue;
 141                 }
 142
 143                 /* A property follows */
 144
 145                 /* n existing properties, 1 new, +1 for _TRANSPORT */
 146                 if (!GREEDY_REALLOC(iovec, m,
 147                                     n + 2 +
 148                                     N_IOVEC_META_FIELDS + N_IOVEC_OBJECT_FIELDS +
 149                                     client_context_extra_fields_n_iovec(context))) {
 150                         r = log_oom();
 151                         break;
 152                 }
 153
 154                 q = memchr(p, '=', e - p);
 155                 if (q) {
 156                         if (journal_field_valid(p, q - p, false)) {
 157                                 size_t l;
 158
 159                                 l = e - p;
 160
 161                                 /* If the field name starts with an underscore, skip the variable, since that indicates
 162                                  * a trusted field */
 163                                 iovec[n++] = IOVEC_MAKE((char*) p, l);
 164                                 entry_size += l;
 165
 166                                 server_process_entry_meta(p, l, ucred,
 167                                                           &priority,
 168                                                           &identifier,
 169                                                           &message,
 170                                                           &object_pid);
 171                         }
 172
 173                         *remaining -= (e - p) + 1;
 174                         p = e + 1;
 175                         continue;
 176                 } else {
 177                         uint64_t l;
 178                         char *k;
 179
 180                         if (*remaining < e - p + 1 + sizeof(uint64_t) + 1) {
 181                                 log_debug("Failed to parse message, ignoring.");
 182                                 break;
 183                         }
 184
 185                         l = unaligned_read_le64(e + 1);
 186
 187                         if (l > DATA_SIZE_MAX) {
 188                                 log_debug("Received binary data block of %"PRIu64" bytes is too large, ignoring.", l);
 189                                 break;
 190                         }
 191
 192                         if ((uint64_t) *remaining < e - p + 1 + sizeof(uint64_t) + l + 1 ||
 193                             e[1+sizeof(uint64_t)+l] != '\n') {
 194                                 log_debug("Failed to parse message, ignoring.");
 195                                 break;
 196                         }
 197
 198                         k = malloc((e - p) + 1 + l);
 199                         if (!k) {
 200                                 log_oom();
 201                                 break;
 202                         }
 203
 204                         memcpy(k, p, e - p);
 205                         k[e - p] = '=';
 206                         memcpy(k + (e - p) + 1, e + 1 + sizeof(uint64_t), l);
 207
 208                         if (journal_field_valid(p, e - p, false)) {
 209                                 iovec[n] = IOVEC_MAKE(k, (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_SET(k, "/dev/shm/", "/tmp/", "/var/tmp/");
 341                 if (!e) {
 342                         log_error("Received file outside of allowed directories. Refusing.");
 343                         return;
 344                 }
 345
 346                 if (!filename_is_valid(e)) {
 347                         log_error("Received file in subdirectory of allowed directories. Refusing.");
 348                         return;
 349                 }
 350         }
 351
 352         if (fstat(fd, &st) < 0) {
 353                 log_error_errno(errno, "Failed to stat passed file, ignoring: %m");
 354                 return;
 355         }
 356
 357         if (!S_ISREG(st.st_mode)) {
 358                 log_error("File passed is not regular. Ignoring.");
 359                 return;
 360         }
 361
 362         if (st.st_size <= 0)
 363                 return;
 364
 365         if (st.st_size > ENTRY_SIZE_MAX) {
 366                 log_error("File passed too large. Ignoring.");
 367                 return;
 368         }
 369
 370         if (sealed) {
 371                 void *p;
 372                 size_t ps;
 373
 374                 /* The file is sealed, we can just map it and use it. */
 375
 376                 ps = PAGE_ALIGN(st.st_size);
 377                 p = mmap(NULL, ps, PROT_READ, MAP_PRIVATE, fd, 0);
 378                 if (p == MAP_FAILED) {
 379                         log_error_errno(errno, "Failed to map memfd, ignoring: %m");
 380                         return;
 381                 }
 382
 383                 server_process_native_message(s, p, st.st_size, ucred, tv, label, label_len);
 384                 assert_se(munmap(p, ps) >= 0);
 385         } else {
 386                 _cleanup_free_ void *p = NULL;
 387                 struct statvfs vfs;
 388                 ssize_t n;
 389
 390                 if (fstatvfs(fd, &vfs) < 0) {
 391                         log_error_errno(errno, "Failed to stat file system of passed file, ignoring: %m");
 392                         return;
 393                 }
 394
 395                 /* Refuse operating on file systems that have
 396                  * mandatory locking enabled, see:
 397                  *
 398                  * https://github.com/systemd/systemd/issues/1822
 399                  */
 400                 if (vfs.f_flag & ST_MANDLOCK) {
 401                         log_error("Received file descriptor from file system with mandatory locking enabled, refusing.");
 402                         return;
 403                 }
 404
 405                 /* Make the fd non-blocking. On regular files this has
 406                  * the effect of bypassing mandatory locking. Of
 407                  * course, this should normally not be necessary given
 408                  * the check above, but let's better be safe than
 409                  * sorry, after all NFS is pretty confusing regarding
 410                  * file system flags, and we better don't trust it,
 411                  * and so is SMB. */
 412                 r = fd_nonblock(fd, true);
 413                 if (r < 0) {
 414                         log_error_errno(r, "Failed to make fd non-blocking, ignoring: %m");
 415                         return;
 416                 }
 417
 418                 /* The file is not sealed, we can't map the file here, since
 419                  * clients might then truncate it and trigger a SIGBUS for
 420                  * us. So let's stupidly read it */
 421
 422                 p = malloc(st.st_size);
 423                 if (!p) {
 424                         log_oom();
 425                         return;
 426                 }
 427
 428                 n = pread(fd, p, st.st_size, 0);
 429                 if (n < 0)
 430                         log_error_errno(errno, "Failed to read file, ignoring: %m");
 431                 else if (n > 0)
 432                         server_process_native_message(s, p, n, ucred, tv, label, label_len);
 433         }
 434 }
 435
 436 int server_open_native_socket(Server *s) {
 437
 438         static const union sockaddr_union sa = {
 439                 .un.sun_family = AF_UNIX,
 440                 .un.sun_path = "/run/systemd/journal/socket",
 441         };
 442         int r;
 443
 444         assert(s);
 445
 446         if (s->native_fd < 0) {
 447                 s->native_fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
 448                 if (s->native_fd < 0)
 449                         return log_error_errno(errno, "socket() failed: %m");
 450
 451                 (void) sockaddr_un_unlink(&sa.un);
 452
 453                 r = bind(s->native_fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
 454                 if (r < 0)
 455                         return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
 456
 457                 (void) chmod(sa.un.sun_path, 0666);
 458         } else
 459                 (void) fd_nonblock(s->native_fd, true);
 460
 461         r = setsockopt_int(s->native_fd, SOL_SOCKET, SO_PASSCRED, true);
 462         if (r < 0)
 463                 return log_error_errno(r, "SO_PASSCRED failed: %m");
 464
 465 #if HAVE_SELINUX
 466         if (mac_selinux_use()) {
 467                 r = setsockopt_int(s->native_fd, SOL_SOCKET, SO_PASSSEC, true);
 468                 if (r < 0)
 469                         log_warning_errno(r, "SO_PASSSEC failed: %m");
 470         }
 471 #endif
 472
 473         r = setsockopt_int(s->native_fd, SOL_SOCKET, SO_TIMESTAMP, true);
 474         if (r < 0)
 475                 return log_error_errno(r, "SO_TIMESTAMP failed: %m");
 476
 477         r = sd_event_add_io(s->event, &s->native_event_source, s->native_fd, EPOLLIN, server_process_datagram, s);
 478         if (r < 0)
 479                 return log_error_errno(r, "Failed to add native server fd to event loop: %m");
 480
 481         r = sd_event_source_set_priority(s->native_event_source, SD_EVENT_PRIORITY_NORMAL+5);
 482         if (r < 0)
 483                 return log_error_errno(r, "Failed to adjust native event source priority: %m");
 484
 485         return 0;
 486 }