]>
Commit | Line | Data |
---|---|---|
ef63833d LP |
1 | /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/ |
2 | ||
3 | /*** | |
4 | This file is part of systemd. | |
5 | ||
6 | Copyright 2011 Lennart Poettering | |
7 | ||
8 | systemd is free software; you can redistribute it and/or modify it | |
9 | under the terms of the GNU Lesser General Public License as published by | |
10 | the Free Software Foundation; either version 2.1 of the License, or | |
11 | (at your option) any later version. | |
12 | ||
13 | systemd is distributed in the hope that it will be useful, but | |
14 | WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
16 | Lesser General Public License for more details. | |
17 | ||
18 | You should have received a copy of the GNU Lesser General Public License | |
19 | along with systemd; If not, see <http://www.gnu.org/licenses/>. | |
20 | ***/ | |
21 | ||
22 | #include <unistd.h> | |
23 | #include <sys/epoll.h> | |
24 | #include <fcntl.h> | |
25 | #include <sys/mman.h> | |
26 | ||
27 | #include <systemd/sd-messages.h> | |
28 | #include <libudev.h> | |
29 | ||
30 | #include "journald.h" | |
31 | #include "journald-kmsg.h" | |
32 | ||
33 | void server_forward_kmsg( | |
34 | Server *s, | |
35 | int priority, | |
36 | const char *identifier, | |
37 | const char *message, | |
38 | struct ucred *ucred) { | |
39 | ||
40 | struct iovec iovec[5]; | |
41 | char header_priority[6], header_pid[16]; | |
42 | int n = 0; | |
43 | char *ident_buf = NULL; | |
44 | ||
45 | assert(s); | |
46 | assert(priority >= 0); | |
47 | assert(priority <= 999); | |
48 | assert(message); | |
49 | ||
50 | if (_unlikely_(LOG_PRI(priority) > s->max_level_kmsg)) | |
51 | return; | |
52 | ||
53 | if (_unlikely_(s->dev_kmsg_fd < 0)) | |
54 | return; | |
55 | ||
56 | /* Never allow messages with kernel facility to be written to | |
57 | * kmsg, regardless where the data comes from. */ | |
58 | priority = syslog_fixup_facility(priority); | |
59 | ||
60 | /* First: priority field */ | |
61 | snprintf(header_priority, sizeof(header_priority), "<%i>", priority); | |
62 | char_array_0(header_priority); | |
63 | IOVEC_SET_STRING(iovec[n++], header_priority); | |
64 | ||
65 | /* Second: identifier and PID */ | |
66 | if (ucred) { | |
67 | if (!identifier) { | |
68 | get_process_comm(ucred->pid, &ident_buf); | |
69 | identifier = ident_buf; | |
70 | } | |
71 | ||
72 | snprintf(header_pid, sizeof(header_pid), "[%lu]: ", (unsigned long) ucred->pid); | |
73 | char_array_0(header_pid); | |
74 | ||
75 | if (identifier) | |
76 | IOVEC_SET_STRING(iovec[n++], identifier); | |
77 | ||
78 | IOVEC_SET_STRING(iovec[n++], header_pid); | |
79 | } else if (identifier) { | |
80 | IOVEC_SET_STRING(iovec[n++], identifier); | |
81 | IOVEC_SET_STRING(iovec[n++], ": "); | |
82 | } | |
83 | ||
84 | /* Fourth: message */ | |
85 | IOVEC_SET_STRING(iovec[n++], message); | |
86 | IOVEC_SET_STRING(iovec[n++], "\n"); | |
87 | ||
88 | if (writev(s->dev_kmsg_fd, iovec, n) < 0) | |
89 | log_debug("Failed to write to /dev/kmsg for logging: %s", strerror(errno)); | |
90 | ||
91 | free(ident_buf); | |
92 | } | |
93 | ||
94 | static bool is_us(const char *pid) { | |
95 | pid_t t; | |
96 | ||
97 | assert(pid); | |
98 | ||
99 | if (parse_pid(pid, &t) < 0) | |
100 | return false; | |
101 | ||
102 | return t == getpid(); | |
103 | } | |
104 | ||
105 | static void dev_kmsg_record(Server *s, char *p, size_t l) { | |
106 | struct iovec iovec[N_IOVEC_META_FIELDS + 7 + N_IOVEC_KERNEL_FIELDS + 2 + N_IOVEC_UDEV_FIELDS]; | |
107 | char *message = NULL, *syslog_priority = NULL, *syslog_pid = NULL, *syslog_facility = NULL, *syslog_identifier = NULL, *source_time = NULL; | |
108 | int priority, r; | |
109 | unsigned n = 0, z = 0, j; | |
110 | usec_t usec; | |
111 | char *identifier = NULL, *pid = NULL, *e, *f, *k; | |
112 | uint64_t serial; | |
113 | size_t pl; | |
114 | char *kernel_device = NULL; | |
115 | ||
116 | assert(s); | |
117 | assert(p); | |
118 | ||
119 | if (l <= 0) | |
120 | return; | |
121 | ||
122 | e = memchr(p, ',', l); | |
123 | if (!e) | |
124 | return; | |
125 | *e = 0; | |
126 | ||
127 | r = safe_atoi(p, &priority); | |
128 | if (r < 0 || priority < 0 || priority > 999) | |
129 | return; | |
130 | ||
131 | if (s->forward_to_kmsg && (priority & LOG_FACMASK) != LOG_KERN) | |
132 | return; | |
133 | ||
134 | l -= (e - p) + 1; | |
135 | p = e + 1; | |
136 | e = memchr(p, ',', l); | |
137 | if (!e) | |
138 | return; | |
139 | *e = 0; | |
140 | ||
141 | r = safe_atou64(p, &serial); | |
142 | if (r < 0) | |
143 | return; | |
144 | ||
145 | if (s->kernel_seqnum) { | |
146 | /* We already read this one? */ | |
147 | if (serial < *s->kernel_seqnum) | |
148 | return; | |
149 | ||
150 | /* Did we lose any? */ | |
151 | if (serial > *s->kernel_seqnum) | |
152 | server_driver_message(s, SD_MESSAGE_JOURNAL_MISSED, "Missed %llu kernel messages", (unsigned long long) serial - *s->kernel_seqnum - 1); | |
153 | ||
154 | /* Make sure we never read this one again. Note that | |
155 | * we always store the next message serial we expect | |
156 | * here, simply because this makes handling the first | |
157 | * message with serial 0 easy. */ | |
158 | *s->kernel_seqnum = serial + 1; | |
159 | } | |
160 | ||
161 | l -= (e - p) + 1; | |
162 | p = e + 1; | |
163 | f = memchr(p, ';', l); | |
164 | if (!f) | |
165 | return; | |
166 | /* Kernel 3.6 has the flags field, kernel 3.5 lacks that */ | |
167 | e = memchr(p, ',', l); | |
168 | if (!e || f < e) | |
169 | e = f; | |
170 | *e = 0; | |
171 | ||
172 | r = parse_usec(p, &usec); | |
173 | if (r < 0) | |
174 | return; | |
175 | ||
176 | l -= (f - p) + 1; | |
177 | p = f + 1; | |
178 | e = memchr(p, '\n', l); | |
179 | if (!e) | |
180 | return; | |
181 | *e = 0; | |
182 | ||
183 | pl = e - p; | |
184 | l -= (e - p) + 1; | |
185 | k = e + 1; | |
186 | ||
187 | for (j = 0; l > 0 && j < N_IOVEC_KERNEL_FIELDS; j++) { | |
188 | char *m; | |
189 | /* Meta data fields attached */ | |
190 | ||
191 | if (*k != ' ') | |
192 | break; | |
193 | ||
194 | k ++, l --; | |
195 | ||
196 | e = memchr(k, '\n', l); | |
197 | if (!e) | |
198 | return; | |
199 | ||
200 | *e = 0; | |
201 | ||
202 | m = cunescape_length_with_prefix(k, e - k, "_KERNEL_"); | |
203 | if (!m) | |
204 | break; | |
205 | ||
206 | if (startswith(m, "_KERNEL_DEVICE=")) | |
207 | kernel_device = m + 15; | |
208 | ||
209 | IOVEC_SET_STRING(iovec[n++], m); | |
210 | z++; | |
211 | ||
212 | l -= (e - k) + 1; | |
213 | k = e + 1; | |
214 | } | |
215 | ||
216 | if (kernel_device) { | |
217 | struct udev_device *ud; | |
218 | ||
219 | ud = udev_device_new_from_device_id(s->udev, kernel_device); | |
220 | if (ud) { | |
221 | const char *g; | |
222 | struct udev_list_entry *ll; | |
223 | char *b; | |
224 | ||
225 | g = udev_device_get_devnode(ud); | |
226 | if (g) { | |
227 | b = strappend("_UDEV_DEVNODE=", g); | |
228 | if (b) { | |
229 | IOVEC_SET_STRING(iovec[n++], b); | |
230 | z++; | |
231 | } | |
232 | } | |
233 | ||
234 | g = udev_device_get_sysname(ud); | |
235 | if (g) { | |
236 | b = strappend("_UDEV_SYSNAME=", g); | |
237 | if (b) { | |
238 | IOVEC_SET_STRING(iovec[n++], b); | |
239 | z++; | |
240 | } | |
241 | } | |
242 | ||
243 | j = 0; | |
244 | ll = udev_device_get_devlinks_list_entry(ud); | |
245 | udev_list_entry_foreach(ll, ll) { | |
246 | ||
247 | if (j > N_IOVEC_UDEV_FIELDS) | |
248 | break; | |
249 | ||
250 | g = udev_list_entry_get_name(ll); | |
251 | b = strappend("_UDEV_DEVLINK=", g); | |
252 | if (g) { | |
253 | IOVEC_SET_STRING(iovec[n++], b); | |
254 | z++; | |
255 | } | |
256 | ||
257 | j++; | |
258 | } | |
259 | ||
260 | udev_device_unref(ud); | |
261 | } | |
262 | } | |
263 | ||
264 | if (asprintf(&source_time, "_SOURCE_MONOTONIC_TIMESTAMP=%llu", | |
265 | (unsigned long long) usec) >= 0) | |
266 | IOVEC_SET_STRING(iovec[n++], source_time); | |
267 | ||
268 | IOVEC_SET_STRING(iovec[n++], "_TRANSPORT=kernel"); | |
269 | ||
270 | if (asprintf(&syslog_priority, "PRIORITY=%i", priority & LOG_PRIMASK) >= 0) | |
271 | IOVEC_SET_STRING(iovec[n++], syslog_priority); | |
272 | ||
273 | if ((priority & LOG_FACMASK) == LOG_KERN) | |
274 | IOVEC_SET_STRING(iovec[n++], "SYSLOG_IDENTIFIER=kernel"); | |
275 | else { | |
276 | syslog_read_identifier((const char**) &p, &identifier, &pid); | |
277 | ||
278 | /* Avoid any messages we generated ourselves via | |
279 | * log_info() and friends. */ | |
280 | if (pid && is_us(pid)) | |
281 | goto finish; | |
282 | ||
283 | if (identifier) { | |
284 | syslog_identifier = strappend("SYSLOG_IDENTIFIER=", identifier); | |
285 | if (syslog_identifier) | |
286 | IOVEC_SET_STRING(iovec[n++], syslog_identifier); | |
287 | } | |
288 | ||
289 | if (pid) { | |
290 | syslog_pid = strappend("SYSLOG_PID=", pid); | |
291 | if (syslog_pid) | |
292 | IOVEC_SET_STRING(iovec[n++], syslog_pid); | |
293 | } | |
294 | ||
295 | if (asprintf(&syslog_facility, "SYSLOG_FACILITY=%i", LOG_FAC(priority)) >= 0) | |
296 | IOVEC_SET_STRING(iovec[n++], syslog_facility); | |
297 | } | |
298 | ||
299 | message = cunescape_length_with_prefix(p, pl, "MESSAGE="); | |
300 | if (message) | |
301 | IOVEC_SET_STRING(iovec[n++], message); | |
302 | ||
303 | server_dispatch_message(s, iovec, n, ELEMENTSOF(iovec), NULL, NULL, NULL, 0, NULL, priority); | |
304 | ||
305 | finish: | |
306 | for (j = 0; j < z; j++) | |
307 | free(iovec[j].iov_base); | |
308 | ||
309 | free(message); | |
310 | free(syslog_priority); | |
311 | free(syslog_identifier); | |
312 | free(syslog_pid); | |
313 | free(syslog_facility); | |
314 | free(source_time); | |
315 | free(identifier); | |
316 | free(pid); | |
317 | } | |
318 | ||
319 | int server_read_dev_kmsg(Server *s) { | |
320 | char buffer[8192+1]; /* the kernel-side limit per record is 8K currently */ | |
321 | ssize_t l; | |
322 | ||
323 | assert(s); | |
324 | assert(s->dev_kmsg_fd >= 0); | |
325 | ||
326 | l = read(s->dev_kmsg_fd, buffer, sizeof(buffer) - 1); | |
327 | if (l == 0) | |
328 | return 0; | |
329 | if (l < 0) { | |
330 | /* Old kernels who don't allow reading from /dev/kmsg | |
331 | * return EINVAL when we try. So handle this cleanly, | |
332 | * but don' try to ever read from it again. */ | |
333 | if (errno == EINVAL) { | |
334 | epoll_ctl(s->epoll_fd, EPOLL_CTL_DEL, s->dev_kmsg_fd, NULL); | |
335 | return 0; | |
336 | } | |
337 | ||
338 | if (errno == EAGAIN || errno == EINTR || errno == EPIPE) | |
339 | return 0; | |
340 | ||
341 | log_error("Failed to read from kernel: %m"); | |
342 | return -errno; | |
343 | } | |
344 | ||
345 | dev_kmsg_record(s, buffer, l); | |
346 | return 1; | |
347 | } | |
348 | ||
349 | int server_flush_dev_kmsg(Server *s) { | |
350 | int r; | |
351 | ||
352 | assert(s); | |
353 | ||
354 | if (s->dev_kmsg_fd < 0) | |
355 | return 0; | |
356 | ||
357 | if (!s->dev_kmsg_readable) | |
358 | return 0; | |
359 | ||
360 | log_info("Flushing /dev/kmsg..."); | |
361 | ||
362 | for (;;) { | |
363 | r = server_read_dev_kmsg(s); | |
364 | if (r < 0) | |
365 | return r; | |
366 | ||
367 | if (r == 0) | |
368 | break; | |
369 | } | |
370 | ||
371 | return 0; | |
372 | } | |
373 | ||
374 | int server_open_dev_kmsg(Server *s) { | |
375 | struct epoll_event ev; | |
376 | ||
377 | assert(s); | |
378 | ||
379 | s->dev_kmsg_fd = open("/dev/kmsg", O_RDWR|O_CLOEXEC|O_NONBLOCK|O_NOCTTY); | |
380 | if (s->dev_kmsg_fd < 0) { | |
381 | log_warning("Failed to open /dev/kmsg, ignoring: %m"); | |
382 | return 0; | |
383 | } | |
384 | ||
385 | zero(ev); | |
386 | ev.events = EPOLLIN; | |
387 | ev.data.fd = s->dev_kmsg_fd; | |
388 | if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->dev_kmsg_fd, &ev) < 0) { | |
389 | ||
390 | /* This will fail with EPERM on older kernels where | |
391 | * /dev/kmsg is not readable. */ | |
392 | if (errno == EPERM) | |
393 | return 0; | |
394 | ||
395 | log_error("Failed to add /dev/kmsg fd to epoll object: %m"); | |
396 | return -errno; | |
397 | } | |
398 | ||
399 | s->dev_kmsg_readable = true; | |
400 | ||
401 | return 0; | |
402 | } | |
403 | ||
404 | int server_open_kernel_seqnum(Server *s) { | |
405 | int fd; | |
406 | uint64_t *p; | |
407 | ||
408 | assert(s); | |
409 | ||
410 | /* We store the seqnum we last read in an mmaped file. That | |
411 | * way we can just use it like a variable, but it is | |
412 | * persistant and automatically flushed at reboot. */ | |
413 | ||
414 | fd = open("/run/systemd/journal/kernel-seqnum", O_RDWR|O_CREAT|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW, 0644); | |
415 | if (fd < 0) { | |
416 | log_error("Failed to open /run/systemd/journal/kernel-seqnum, ignoring: %m"); | |
417 | return 0; | |
418 | } | |
419 | ||
420 | if (posix_fallocate(fd, 0, sizeof(uint64_t)) < 0) { | |
421 | log_error("Failed to allocate sequential number file, ignoring: %m"); | |
422 | close_nointr_nofail(fd); | |
423 | return 0; | |
424 | } | |
425 | ||
426 | p = mmap(NULL, sizeof(uint64_t), PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0); | |
427 | if (p == MAP_FAILED) { | |
428 | log_error("Failed to map sequential number file, ignoring: %m"); | |
429 | close_nointr_nofail(fd); | |
430 | return 0; | |
431 | } | |
432 | ||
433 | close_nointr_nofail(fd); | |
434 | s->kernel_seqnum = p; | |
435 | ||
436 | return 0; | |
437 | } |