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