src/journal/journald-native.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2 /***
   3   This file is part of systemd.
   4
   5   Copyright 2011 Lennart Poettering
   6 ***/
   7
   8 #include <stddef.h>
   9 #include <sys/epoll.h>
  10 #include <sys/mman.h>
  11 #include <sys/statvfs.h>
  12 #include <unistd.h>
  13
  14 #include "alloc-util.h"
  15 #include "fd-util.h"
  16 #include "fs-util.h"
  17 #include "io-util.h"
  18 #include "journal-importer.h"
  19 #include "journal-util.h"
  20 #include "journald-console.h"
  21 #include "journald-kmsg.h"
  22 #include "journald-native.h"
  23 #include "journald-server.h"
  24 #include "journald-syslog.h"
  25 #include "journald-wall.h"
  26 #include "memfd-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 "string-util.h"
  33 #include "unaligned.h"
  34
  35 static bool allow_object_pid(const struct ucred *ucred) {
  36         return ucred && ucred->uid == 0;
  37 }
  38
  39 static void server_process_entry_meta(
  40                 const char *p, size_t l,
  41                 const struct ucred *ucred,
  42                 int *priority,
  43                 char **identifier,
  44                 char **message,
  45                 pid_t *object_pid) {
  46
  47         /* We need to determine the priority of this entry for the rate limiting logic */
  48
  49         if (l == 10 &&
  50             startswith(p, "PRIORITY=") &&
  51             p[9] >= '0' && p[9] <= '9')
  52                 *priority = (*priority & LOG_FACMASK) | (p[9] - '0');
  53
  54         else if (l == 17 &&
  55                  startswith(p, "SYSLOG_FACILITY=") &&
  56                  p[16] >= '0' && p[16] <= '9')
  57                 *priority = (*priority & LOG_PRIMASK) | ((p[16] - '0') << 3);
  58
  59         else if (l == 18 &&
  60                  startswith(p, "SYSLOG_FACILITY=") &&
  61                  p[16] >= '0' && p[16] <= '9' &&
  62                  p[17] >= '0' && p[17] <= '9')
  63                 *priority = (*priority & LOG_PRIMASK) | (((p[16] - '0')*10 + (p[17] - '0')) << 3);
  64
  65         else if (l >= 19 &&
  66                  startswith(p, "SYSLOG_IDENTIFIER=")) {
  67                 char *t;
  68
  69                 t = strndup(p + 18, l - 18);
  70                 if (t) {
  71                         free(*identifier);
  72                         *identifier = t;
  73                 }
  74
  75         } else if (l >= 8 &&
  76                    startswith(p, "MESSAGE=")) {
  77                 char *t;
  78
  79                 t = strndup(p + 8, l - 8);
  80                 if (t) {
  81                         free(*message);
  82                         *message = t;
  83                 }
  84
  85         } else if (l > STRLEN("OBJECT_PID=") &&
  86                    l < STRLEN("OBJECT_PID=")  + DECIMAL_STR_MAX(pid_t) &&
  87                    startswith(p, "OBJECT_PID=") &&
  88                    allow_object_pid(ucred)) {
  89                 char buf[DECIMAL_STR_MAX(pid_t)];
  90                 memcpy(buf, p + STRLEN("OBJECT_PID="),
  91                        l - STRLEN("OBJECT_PID="));
  92                 buf[l-STRLEN("OBJECT_PID=")] = '\0';
  93
  94                 (void) parse_pid(buf, object_pid);
  95         }
  96 }
  97
  98 static int server_process_entry(
  99                 Server *s,
 100                 const void *buffer, size_t *remaining,
 101                 ClientContext *context,
 102                 const struct ucred *ucred,
 103                 const struct timeval *tv,
 104                 const char *label, size_t label_len) {
 105
 106         /* Process a single entry from a native message. Returns 0 if nothing special happened and the message
 107          * processing should continue, and a negative or positive value otherwise.
 108          *
 109          * Note that *remaining is altered on both success and failure. */
 110
 111         size_t n = 0, j, tn = (size_t) -1, m = 0, entry_size = 0;
 112         char *identifier = NULL, *message = NULL;
 113         struct iovec *iovec = NULL;
 114         int priority = LOG_INFO;
 115         pid_t object_pid = 0;
 116         const char *p;
 117         int r = 0;
 118
 119         p = buffer;
 120
 121         while (*remaining > 0) {
 122                 const char *e, *q;
 123
 124                 e = memchr(p, '\n', *remaining);
 125
 126                 if (!e) {
 127                         /* Trailing noise, let's ignore it, and flush what we collected */
 128                         log_debug("Received message with trailing noise, ignoring.");
 129                         r = 1; /* finish processing of the message */
 130                         break;
 131                 }
 132
 133                 if (e == p) {
 134                         /* Entry separator */
 135                         *remaining -= 1;
 136                         break;
 137                 }
 138
 139                 if (IN_SET(*p, '.', '#')) {
 140                         /* Ignore control commands for now, and
 141                          * comments too. */
 142                         *remaining -= (e - p) + 1;
 143                         p = e + 1;
 144                         continue;
 145                 }
 146
 147                 /* A property follows */
 148
 149                 /* n existing properties, 1 new, +1 for _TRANSPORT */
 150                 if (!GREEDY_REALLOC(iovec, m,
 151                                     n + 2 +
 152                                     N_IOVEC_META_FIELDS + N_IOVEC_OBJECT_FIELDS +
 153                                     client_context_extra_fields_n_iovec(context))) {
 154                         r = log_oom();
 155                         break;
 156                 }
 157
 158                 q = memchr(p, '=', e - p);
 159                 if (q) {
 160                         if (journal_field_valid(p, q - p, false)) {
 161                                 size_t l;
 162
 163                                 l = e - p;
 164
 165                                 /* If the field name starts with an underscore, skip the variable, since that indicates
 166                                  * a trusted field */
 167                                 iovec[n++] = IOVEC_MAKE((char*) p, l);
 168                                 entry_size += l;
 169
 170                                 server_process_entry_meta(p, l, ucred,
 171                                                           &priority,
 172                                                           &identifier,
 173                                                           &message,
 174                                                           &object_pid);
 175                         }
 176
 177                         *remaining -= (e - p) + 1;
 178                         p = e + 1;
 179                         continue;
 180                 } else {
 181                         uint64_t l;
 182                         char *k;
 183
 184                         if (*remaining < e - p + 1 + sizeof(uint64_t) + 1) {
 185                                 log_debug("Failed to parse message, ignoring.");
 186                                 break;
 187                         }
 188
 189                         l = unaligned_read_le64(e + 1);
 190
 191                         if (l > DATA_SIZE_MAX) {
 192                                 log_debug("Received binary data block of %"PRIu64" bytes is too large, ignoring.", l);
 193                                 break;
 194                         }
 195
 196                         if ((uint64_t) *remaining < e - p + 1 + sizeof(uint64_t) + l + 1 ||
 197                             e[1+sizeof(uint64_t)+l] != '\n') {
 198                                 log_debug("Failed to parse message, ignoring.");
 199                                 break;
 200                         }
 201
 202                         k = malloc((e - p) + 1 + l);
 203                         if (!k) {
 204                                 log_oom();
 205                                 break;
 206                         }
 207
 208                         memcpy(k, p, e - p);
 209                         k[e - p] = '=';
 210                         memcpy(k + (e - p) + 1, e + 1 + sizeof(uint64_t), l);
 211
 212                         if (journal_field_valid(p, e - p, false)) {
 213                                 iovec[n].iov_base = k;
 214                                 iovec[n].iov_len = (e - p) + 1 + l;
 215                                 entry_size += iovec[n].iov_len;
 216                                 n++;
 217
 218                                 server_process_entry_meta(k, (e - p) + 1 + l, ucred,
 219                                                           &priority,
 220                                                           &identifier,
 221                                                           &message,
 222                                                           &object_pid);
 223                         } else
 224                                 free(k);
 225
 226                         *remaining -= (e - p) + 1 + sizeof(uint64_t) + l + 1;
 227                         p = e + 1 + sizeof(uint64_t) + l + 1;
 228                 }
 229         }
 230
 231         if (n <= 0) {
 232                 r = 1;
 233                 goto finish;
 234         }
 235
 236         if (!client_context_test_priority(context, priority)) {
 237                 r = 0;
 238                 goto finish;
 239         }
 240
 241         tn = n++;
 242         iovec[tn] = IOVEC_MAKE_STRING("_TRANSPORT=journal");
 243         entry_size += STRLEN("_TRANSPORT=journal");
 244
 245         if (entry_size + n + 1 > ENTRY_SIZE_MAX) { /* data + separators + trailer */
 246                 log_debug("Entry is too big with %zu properties and %zu bytes, ignoring.", n, entry_size);
 247                 goto finish;
 248         }
 249
 250         if (message) {
 251                 if (s->forward_to_syslog)
 252                         server_forward_syslog(s, syslog_fixup_facility(priority), identifier, message, ucred, tv);
 253
 254                 if (s->forward_to_kmsg)
 255                         server_forward_kmsg(s, priority, identifier, message, ucred);
 256
 257                 if (s->forward_to_console)
 258                         server_forward_console(s, priority, identifier, message, ucred);
 259
 260                 if (s->forward_to_wall)
 261                         server_forward_wall(s, priority, identifier, message, ucred);
 262         }
 263
 264         server_dispatch_message(s, iovec, n, m, context, tv, priority, object_pid);
 265
 266 finish:
 267         for (j = 0; j < n; j++)  {
 268                 if (j == tn)
 269                         continue;
 270
 271                 if (iovec[j].iov_base < buffer ||
 272                     (const char*) iovec[j].iov_base >= p + *remaining)
 273                         free(iovec[j].iov_base);
 274         }
 275
 276         free(iovec);
 277         free(identifier);
 278         free(message);
 279
 280         return r;
 281 }
 282
 283 void server_process_native_message(
 284                 Server *s,
 285                 const void *buffer, size_t buffer_size,
 286                 const struct ucred *ucred,
 287                 const struct timeval *tv,
 288                 const char *label, size_t label_len) {
 289
 290         size_t remaining = buffer_size;
 291         ClientContext *context = NULL;
 292         int r;
 293
 294         assert(s);
 295         assert(buffer || buffer_size == 0);
 296
 297         if (ucred && pid_is_valid(ucred->pid)) {
 298                 r = client_context_get(s, ucred->pid, ucred, label, label_len, NULL, &context);
 299                 if (r < 0)
 300                         log_warning_errno(r, "Failed to retrieve credentials for PID " PID_FMT ", ignoring: %m", ucred->pid);
 301         }
 302
 303         do {
 304                 r = server_process_entry(s,
 305                                          (const uint8_t*) buffer + (buffer_size - remaining), &remaining,
 306                                          context, ucred, tv, label, label_len);
 307         } while (r == 0);
 308 }
 309
 310 void server_process_native_file(
 311                 Server *s,
 312                 int fd,
 313                 const struct ucred *ucred,
 314                 const struct timeval *tv,
 315                 const char *label, size_t label_len) {
 316
 317         struct stat st;
 318         bool sealed;
 319         int r;
 320
 321         /* Data is in the passed fd, since it didn't fit in a
 322          * datagram. */
 323
 324         assert(s);
 325         assert(fd >= 0);
 326
 327         /* If it's a memfd, check if it is sealed. If so, we can just
 328          * use map it and use it, and do not need to copy the data
 329          * out. */
 330         sealed = memfd_get_sealed(fd) > 0;
 331
 332         if (!sealed && (!ucred || ucred->uid != 0)) {
 333                 _cleanup_free_ char *k = NULL;
 334                 const char *e;
 335
 336                 /* If this is not a sealed memfd, and the peer is unknown or
 337                  * unprivileged, then verify the path. */
 338
 339                 r = fd_get_path(fd, &k);
 340                 if (r < 0) {
 341                         log_error_errno(r, "readlink(/proc/self/fd/%i) failed: %m", fd);
 342                         return;
 343                 }
 344
 345                 e = path_startswith(k, "/dev/shm/");
 346                 if (!e)
 347                         e = path_startswith(k, "/tmp/");
 348                 if (!e)
 349                         e = path_startswith(k, "/var/tmp/");
 350                 if (!e) {
 351                         log_error("Received file outside of allowed directories. Refusing.");
 352                         return;
 353                 }
 354
 355                 if (!filename_is_valid(e)) {
 356                         log_error("Received file in subdirectory of allowed directories. Refusing.");
 357                         return;
 358                 }
 359         }
 360
 361         if (fstat(fd, &st) < 0) {
 362                 log_error_errno(errno, "Failed to stat passed file, ignoring: %m");
 363                 return;
 364         }
 365
 366         if (!S_ISREG(st.st_mode)) {
 367                 log_error("File passed is not regular. Ignoring.");
 368                 return;
 369         }
 370
 371         if (st.st_size <= 0)
 372                 return;
 373
 374         if (st.st_size > ENTRY_SIZE_MAX) {
 375                 log_error("File passed too large. Ignoring.");
 376                 return;
 377         }
 378
 379         if (sealed) {
 380                 void *p;
 381                 size_t ps;
 382
 383                 /* The file is sealed, we can just map it and use it. */
 384
 385                 ps = PAGE_ALIGN(st.st_size);
 386                 p = mmap(NULL, ps, PROT_READ, MAP_PRIVATE, fd, 0);
 387                 if (p == MAP_FAILED) {
 388                         log_error_errno(errno, "Failed to map memfd, ignoring: %m");
 389                         return;
 390                 }
 391
 392                 server_process_native_message(s, p, st.st_size, ucred, tv, label, label_len);
 393                 assert_se(munmap(p, ps) >= 0);
 394         } else {
 395                 _cleanup_free_ void *p = NULL;
 396                 struct statvfs vfs;
 397                 ssize_t n;
 398
 399                 if (fstatvfs(fd, &vfs) < 0) {
 400                         log_error_errno(errno, "Failed to stat file system of passed file, ignoring: %m");
 401                         return;
 402                 }
 403
 404                 /* Refuse operating on file systems that have
 405                  * mandatory locking enabled, see:
 406                  *
 407                  * https://github.com/systemd/systemd/issues/1822
 408                  */
 409                 if (vfs.f_flag & ST_MANDLOCK) {
 410                         log_error("Received file descriptor from file system with mandatory locking enabled, refusing.");
 411                         return;
 412                 }
 413
 414                 /* Make the fd non-blocking. On regular files this has
 415                  * the effect of bypassing mandatory locking. Of
 416                  * course, this should normally not be necessary given
 417                  * the check above, but let's better be safe than
 418                  * sorry, after all NFS is pretty confusing regarding
 419                  * file system flags, and we better don't trust it,
 420                  * and so is SMB. */
 421                 r = fd_nonblock(fd, true);
 422                 if (r < 0) {
 423                         log_error_errno(r, "Failed to make fd non-blocking, ignoring: %m");
 424                         return;
 425                 }
 426
 427                 /* The file is not sealed, we can't map the file here, since
 428                  * clients might then truncate it and trigger a SIGBUS for
 429                  * us. So let's stupidly read it */
 430
 431                 p = malloc(st.st_size);
 432                 if (!p) {
 433                         log_oom();
 434                         return;
 435                 }
 436
 437                 n = pread(fd, p, st.st_size, 0);
 438                 if (n < 0)
 439                         log_error_errno(errno, "Failed to read file, ignoring: %m");
 440                 else if (n > 0)
 441                         server_process_native_message(s, p, n, ucred, tv, label, label_len);
 442         }
 443 }
 444
 445 int server_open_native_socket(Server*s) {
 446
 447         static const union sockaddr_union sa = {
 448                 .un.sun_family = AF_UNIX,
 449                 .un.sun_path = "/run/systemd/journal/socket",
 450         };
 451         static const int one = 1;
 452         int r;
 453
 454         assert(s);
 455
 456         if (s->native_fd < 0) {
 457                 s->native_fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
 458                 if (s->native_fd < 0)
 459                         return log_error_errno(errno, "socket() failed: %m");
 460
 461                 (void) unlink(sa.un.sun_path);
 462
 463                 r = bind(s->native_fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
 464                 if (r < 0)
 465                         return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
 466
 467                 (void) chmod(sa.un.sun_path, 0666);
 468         } else
 469                 fd_nonblock(s->native_fd, 1);
 470
 471         r = setsockopt(s->native_fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
 472         if (r < 0)
 473                 return log_error_errno(errno, "SO_PASSCRED failed: %m");
 474
 475 #if HAVE_SELINUX
 476         if (mac_selinux_use()) {
 477                 r = setsockopt(s->native_fd, SOL_SOCKET, SO_PASSSEC, &one, sizeof(one));
 478                 if (r < 0)
 479                         log_warning_errno(errno, "SO_PASSSEC failed: %m");
 480         }
 481 #endif
 482
 483         r = setsockopt(s->native_fd, SOL_SOCKET, SO_TIMESTAMP, &one, sizeof(one));
 484         if (r < 0)
 485                 return log_error_errno(errno, "SO_TIMESTAMP failed: %m");
 486
 487         r = sd_event_add_io(s->event, &s->native_event_source, s->native_fd, EPOLLIN, server_process_datagram, s);
 488         if (r < 0)
 489                 return log_error_errno(r, "Failed to add native server fd to event loop: %m");
 490
 491         r = sd_event_source_set_priority(s->native_event_source, SD_EVENT_PRIORITY_NORMAL+5);
 492         if (r < 0)
 493                 return log_error_errno(r, "Failed to adjust native event source priority: %m");
 494
 495         return 0;
 496 }