syslog.c revision 627351e35aacd29d28d29ff70845f23e5b0da5c8
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
23 * Use is subject to license terms.
24 */
25
26#pragma ident	"%Z%%M%	%I%	%E% SMI"
27
28#include <sys/fm/protocol.h>
29#include <sys/strlog.h>
30#include <fm/fmd_api.h>
31#include <fm/fmd_msg.h>
32
33#include <stropts.h>
34#include <syslog.h>
35#include <locale.h>
36#include <strings.h>
37#include <stdlib.h>
38#include <unistd.h>
39#include <limits.h>
40#include <alloca.h>
41#include <errno.h>
42#include <fcntl.h>
43#include <time.h>
44
45/*
46 * SYSLOG_DOMAIN and SYSLOG_TEMPLATE define the dgettext() parameters the agent
47 * can use to retrieve the localized format string for diagnosis messages.
48 * The format string retrieved from SYSLOG_DOMAIN is the default format
49 * string, but when processing each suspect list, dgettext() is also called
50 * for the domain that matches the diagcode dictname and if SYSLOG_TEMPLATE
51 * is defined, it overrides the default for that suspect list only.
52 *
53 * Similarly, SYSLOG_URL is also checked to see if syslog_url
54 * should be overridden for each suspect list.
55 *
56 * The net effect of all this is that for a given diagcode DICT-1234-56:
57 *
58 *	- If DICT.mo defines syslog-msgs-message-template, it is used
59 *	  as the format string for the diagnosis message.
60 *
61 *	- Otherwise, syslog-msgs-message-template from FMD.mo is used.
62 *
63 *	- If DICT.mo defines syslog-url, it is used when filling
64 *	  in the %s in the "description" message.
65 *
66 *	- Otherwise, if syslog-msgs.conf defines a "url" property, that
67 *	  value is used.
68 *
69 *	- Otherwise, the default "http://sun.com/msg/" is used (via the
70 *	  fmd_props[] table defined in this file).
71 */
72static const char SYSLOG_DOMAIN[] = "FMD";
73static const char SYSLOG_TEMPLATE[] = "syslog-msgs-message-template";
74static const char SYSLOG_URL[] = "syslog-url";
75
76static struct stats {
77	fmd_stat_t bad_vers;
78	fmd_stat_t bad_fmri;
79	fmd_stat_t bad_code;
80	fmd_stat_t bad_time;
81	fmd_stat_t log_err;
82	fmd_stat_t msg_err;
83	fmd_stat_t no_msg;
84} syslog_stats = {
85	{ "bad_vers", FMD_TYPE_UINT64, "event version is missing or invalid" },
86	{ "bad_fmri", FMD_TYPE_UINT64, "event fmri is missing or invalid" },
87	{ "bad_code", FMD_TYPE_UINT64, "event code has no dictionary name" },
88	{ "bad_time", FMD_TYPE_UINT64, "event time is not properly encoded" },
89	{ "log_err", FMD_TYPE_UINT64, "failed to log message to log(7D)" },
90	{ "msg_err", FMD_TYPE_UINT64, "failed to log message to sysmsg(7D)" },
91	{ "no_msg", FMD_TYPE_UINT64, "message logging suppressed" }
92};
93
94static const struct facility {
95	const char *fac_name;
96	int fac_value;
97} syslog_facs[] = {
98	{ "LOG_DAEMON", LOG_DAEMON },
99	{ "LOG_LOCAL0", LOG_LOCAL0 },
100	{ "LOG_LOCAL1", LOG_LOCAL1 },
101	{ "LOG_LOCAL2", LOG_LOCAL2 },
102	{ "LOG_LOCAL3", LOG_LOCAL3 },
103	{ "LOG_LOCAL4", LOG_LOCAL4 },
104	{ "LOG_LOCAL5", LOG_LOCAL5 },
105	{ "LOG_LOCAL6", LOG_LOCAL6 },
106	{ "LOG_LOCAL7", LOG_LOCAL7 },
107	{ NULL, 0 }
108};
109
110static char *syslog_locdir;	/* l10n messages directory (if alternate) */
111static char *syslog_url;	/* current value of "url" property */
112static int syslog_msgall;	/* set to message all faults */
113static log_ctl_t syslog_ctl;	/* log(7D) meta-data for each msg */
114static int syslog_logfd = -1;	/* log(7D) file descriptor */
115static int syslog_msgfd = -1;	/* sysmsg(7D) file descriptor */
116static int syslog_file;		/* log to syslog_logfd */
117static int syslog_cons;		/* log to syslog_msgfd */
118
119/*
120 * Ideally we would just use syslog(3C) for outputting our messages, but our
121 * messaging standard defines a nice multi-line format and syslogd(1M) is very
122 * inflexible and stupid when it comes to multi-line messages.  It pulls data
123 * out of log(7D) and splits it up by \n, printing each line to the console
124 * with its usual prefix of date and sender; it uses the same behavior for the
125 * messages file as well.  Further, syslog(3C) provides no CE_CONT equivalent
126 * for userland callers (which at least works around repeated file prefixing).
127 * So with a multi-line message format, your file and console end up like this:
128 *
129 * Dec 02 18:08:40 hostname this is my nicely formatted
130 * Dec 02 18:08:40 hostname message designed for 80 cols
131 * ...
132 *
133 * To resolve these issues, we use our own syslog_emit() wrapper to emit
134 * messages and some knowledge of how the Solaris log drivers work.  We first
135 * construct an enlarged format string containing the appropriate msgid(1).
136 * We then format the caller's message using the provided format and buffer.
137 * We send this message to log(7D) using putmsg() with SL_CONSOLE | SL_LOGONLY
138 * set in the log_ctl_t.  The log driver allows us to set SL_LOGONLY when we
139 * construct messages ourself, indicating that syslogd should only emit the
140 * message to /var/adm/messages and any remote hosts, and skip the console.
141 * Then we emit the message a second time, without the special prefix, to the
142 * sysmsg(7D) device, which handles console redirection and also permits us
143 * to output any characters we like to the console, including \n and \r.
144 */
145/*PRINTFLIKE4*/
146static void
147syslog_emit(fmd_hdl_t *hdl, char *buf, size_t len, const char *msgformat, ...)
148{
149	struct strbuf ctl, dat;
150	uint32_t msgid;
151
152	char *format;
153	size_t formatlen;
154	va_list ap;
155
156	formatlen = strlen(msgformat) + 64; /* +64 for prefix and \0 */
157	format = alloca(formatlen);
158
159	STRLOG_MAKE_MSGID(msgformat, msgid);
160	(void) snprintf(format, formatlen,
161	    "fmd: [ID %u FACILITY_AND_PRIORITY] %s", msgid, msgformat);
162
163	va_start(ap, msgformat);
164	(void) vsnprintf(buf, len, format, ap);
165	va_end(ap);
166
167	ctl.buf = (void *)&syslog_ctl;
168	ctl.len = sizeof (syslog_ctl);
169
170	dat.buf = buf;
171	dat.len = strlen(buf) + 1;
172
173	if (syslog_file && putmsg(syslog_logfd, &ctl, &dat, 0) != 0) {
174		fmd_hdl_debug(hdl, "putmsg failed: %s\n", strerror(errno));
175		syslog_stats.log_err.fmds_value.ui64++;
176	}
177
178	dat.buf = strchr(buf, ']');
179	dat.len -= (size_t)(dat.buf - buf);
180
181	dat.buf[0] = '\r'; /* overwrite ']' with carriage return */
182	dat.buf[1] = '\n'; /* overwrite ' ' with newline */
183
184	if (syslog_cons && write(syslog_msgfd, dat.buf, dat.len) != dat.len) {
185		fmd_hdl_debug(hdl, "write failed: %s\n", strerror(errno));
186		syslog_stats.msg_err.fmds_value.ui64++;
187	}
188}
189
190/*ARGSUSED*/
191static void
192syslog_recv(fmd_hdl_t *hdl, fmd_event_t *ep, nvlist_t *nvl, const char *class)
193{
194	char *uuid, *code, *dict, *url, *urlcode, *template, *p;
195	char *src_name, *src_vers, *platform, *chassis, *server;
196	char *typ, *sev, *fmt, *trfmt, *rsp, *imp, *act, *locdir;
197	char msg[1024], desc[1024], date[64];
198	boolean_t domsg;
199
200	nvlist_t *fmri, *auth;
201	uint8_t version;
202	struct tm tm, *tmp;
203	int64_t *tv;
204	time_t sec;
205	uint_t tn = 0;
206	char *olang = NULL;
207	int locale_c = 0;
208	size_t len;
209
210	if (nvlist_lookup_uint8(nvl, FM_VERSION, &version) != 0 ||
211	    version > FM_SUSPECT_VERSION) {
212		fmd_hdl_debug(hdl, "invalid event version: %u\n", version);
213		syslog_stats.bad_vers.fmds_value.ui64++;
214		return; /* invalid event version */
215	}
216
217	if (!syslog_msgall && nvlist_lookup_boolean_value(nvl,
218	    FM_SUSPECT_MESSAGE, &domsg) == 0 && !domsg) {
219		fmd_hdl_debug(hdl, "%s requested no message\n", class);
220		syslog_stats.no_msg.fmds_value.ui64++;
221		return; /* event is not to be messaged */
222	}
223
224	/*
225	 * Extract the DE element, which is an FMRI for the diagnosis engine
226	 * that made this event, and validate its meta-data before continuing.
227	 */
228	if (nvlist_lookup_nvlist(nvl, FM_SUSPECT_DE, &fmri) != 0 ||
229	    nvlist_lookup_string(fmri, FM_FMRI_SCHEME, &p) != 0 ||
230	    strcmp(p, FM_FMRI_SCHEME_FMD) != 0 ||
231	    nvlist_lookup_uint8(fmri, FM_VERSION, &version) != 0 ||
232	    version > FM_FMD_SCHEME_VERSION ||
233	    nvlist_lookup_nvlist(fmri, FM_FMRI_AUTHORITY, &auth) != 0 ||
234	    nvlist_lookup_uint8(auth, FM_VERSION, &version) != 0 ||
235	    version > FM_FMRI_AUTH_VERSION) {
236		syslog_stats.bad_fmri.fmds_value.ui64++;
237		return; /* invalid de fmri */
238	}
239
240	/*
241	 * Extract the relevant identifying elements of the FMRI and authority.
242	 * Note: for now, we ignore FM_FMRI_AUTH_DOMAIN (only for SPs).
243	 */
244	(void) nvlist_lookup_string(fmri, FM_FMRI_FMD_NAME, &src_name);
245	(void) nvlist_lookup_string(fmri, FM_FMRI_FMD_VERSION, &src_vers);
246	(void) nvlist_lookup_string(auth, FM_FMRI_AUTH_PRODUCT, &platform);
247	(void) nvlist_lookup_string(auth, FM_FMRI_AUTH_SERVER, &server);
248
249	if (nvlist_lookup_string(auth, FM_FMRI_AUTH_CHASSIS, &chassis) != 0)
250		chassis = "-"; /* chassis serial number may not be present */
251
252	/*
253	 * Extract the uuid and diagcode dictionary from the event code.  The
254	 * dictionary name is the text preceding the first "-" in the code.
255	 */
256	(void) nvlist_lookup_string(nvl, FM_SUSPECT_UUID, &uuid);
257	(void) nvlist_lookup_string(nvl, FM_SUSPECT_DIAG_CODE, &code);
258
259	if ((p = strchr(code, '-')) == NULL || p == code) {
260		fmd_hdl_debug(hdl, "invalid diagnosis code: %s\n", code);
261		syslog_stats.bad_code.fmds_value.ui64++;
262		return; /* invalid diagnosis code */
263	}
264
265	dict = alloca((size_t)(p - code) + 1);
266	(void) strncpy(dict, code, (size_t)(p - code));
267	dict[(size_t)(p - code)] = '\0';
268
269	/*
270	 * Alloca a hunk of memory and use it to create the msgid strings
271	 * <code>.type, <code>.severity, <code>.description, and so forth.
272	 * These form the msgids we will use to look up the localized text.
273	 * Since we've allocated things to be of the right size, we know
274	 * than snprintf() can't overflow: INT_MAX is used shut lint up and
275	 * avoid code to needlessly recompute the remaining buffer space.
276	 */
277	typ = alloca(6 * (strlen(code) + 16));
278	sev = typ + snprintf(typ, INT_MAX, "%s.type", code) + 1;
279	fmt = sev + snprintf(sev, INT_MAX, "%s.severity", code) + 1;
280	rsp = fmt + snprintf(fmt, INT_MAX, "%s.description", code) + 1;
281	imp = rsp + snprintf(rsp, INT_MAX, "%s.response", code) + 1;
282	act = imp + snprintf(imp, INT_MAX, "%s.impact", code) + 1;
283	(void) snprintf(act, INT_MAX, "%s.action", code);
284
285	fmd_msg_lock();
286
287	if (syslog_locdir != NULL)
288		locdir =  bindtextdomain(dict, syslog_locdir);
289
290	if ((trfmt = dgettext(dict, fmt)) == fmt) {
291		/*
292		 * We didn't find a translation in the dictionary for the
293		 * current language.  The string we passed to gettext is merely
294		 * an index - it isn't sufficient, on its own, to be used as the
295		 * message.  Fall back to C and try again.
296		 */
297		olang = setlocale(LC_MESSAGES, NULL);
298		if (olang) {
299			p = alloca(strlen(olang) + 1);
300			olang = strcpy(p, olang);
301		}
302		locale_c = 1;
303		(void) setlocale(LC_MESSAGES, "C");
304		trfmt = dgettext(dict, fmt);
305	}
306
307	if ((url = dgettext(dict, SYSLOG_URL)) == SYSLOG_URL)
308		url = syslog_url;
309
310	/*
311	 * If the URL ends with a slash, that indicates the code should be
312	 * appended to it.  After formatting the URL, reformat the DESC
313	 * text using the URL as an snprintf argument.
314	 */
315	len = strlen(url);
316	if (url[len - 1] == '/') {
317		urlcode = alloca(len + strlen(code) + 1);
318		(void) snprintf(urlcode, INT_MAX, "%s%s", url, code);
319	} else {
320		urlcode = url;
321	}
322	/* LINTED - variable format specifier to snprintf() */
323	(void) snprintf(desc, sizeof (desc), trfmt, urlcode);
324
325	/*
326	 * Extract the diagnosis time and format it using the locale's default.
327	 * strftime() will use GMT or local time based on our "gmt" setting.
328	 */
329	if (nvlist_lookup_int64_array(nvl, FM_SUSPECT_DIAG_TIME,
330	    &tv, &tn) == 0 && tn == 2 && (sec = (time_t)tv[0]) != (time_t)-1 &&
331	    (tmp = localtime_r(&sec, &tm)) != NULL)
332		(void) strftime(date, sizeof (date), "%C", tmp);
333	else {
334		syslog_stats.bad_time.fmds_value.ui64++;
335		(void) strcpy(date, "-");
336	}
337
338	/*
339	 * Create and log the final string by filling in the template with the
340	 * strings we've created and the strings from the message dictionary.
341	 * If a template is provided for this dictionary, use it, otherwise
342	 * fall back to the default template.
343	 */
344	if ((template = dgettext(dict, SYSLOG_TEMPLATE)) == SYSLOG_TEMPLATE)
345		template = dgettext(SYSLOG_DOMAIN, SYSLOG_TEMPLATE);
346
347	syslog_ctl.pri &= LOG_FACMASK;
348	if (strcmp(class, FM_LIST_REPAIRED_CLASS) == 0)
349		syslog_ctl.pri |= LOG_NOTICE;
350	else
351		syslog_ctl.pri |= LOG_ERR;
352	syslog_emit(hdl, msg, sizeof (msg),
353	    template, code, dgettext(dict, typ),
354	    dgettext(dict, sev), date, platform, chassis, server, src_name,
355	    src_vers, uuid, desc, dgettext(dict, rsp), dgettext(dict, imp),
356	    dgettext(dict, act));
357
358	/*
359	 * Switch back to our original language if we had to fall back to C.
360	 */
361	if (olang != NULL)
362		(void) setlocale(LC_MESSAGES, olang);
363
364	if (syslog_locdir != NULL)
365		(void) bindtextdomain(dict, locdir);
366
367	fmd_msg_unlock();
368
369	if (locale_c) {
370		fmd_hdl_debug(hdl,
371		    trfmt == fmt ?
372		    "dgettext(%s, %s) in %s and C failed\n" :
373		    "dgettext(%s, %s) in %s failed; C used\n",
374		    dict, fmt, olang ? olang : "<null>");
375	}
376}
377
378static const fmd_prop_t fmd_props[] = {
379	{ "console", FMD_TYPE_BOOL, "true" },
380	{ "facility", FMD_TYPE_STRING, "LOG_DAEMON" },
381	{ "gmt", FMD_TYPE_BOOL, "false" },
382	{ "syslogd", FMD_TYPE_BOOL, "true" },
383	{ "url", FMD_TYPE_STRING, "http://sun.com/msg/" },
384	{ "message_all", FMD_TYPE_BOOL, "false" },
385	{ NULL, 0, NULL }
386};
387
388static const fmd_hdl_ops_t fmd_ops = {
389	syslog_recv,	/* fmdo_recv */
390	NULL,		/* fmdo_timeout */
391	NULL,		/* fmdo_close */
392	NULL,		/* fmdo_stats */
393	NULL,		/* fmdo_gc */
394};
395
396static const fmd_hdl_info_t fmd_info = {
397	"Syslog Messaging Agent", "1.0", &fmd_ops, fmd_props
398};
399
400void
401_fmd_init(fmd_hdl_t *hdl)
402{
403	const struct facility *fp;
404	char *facname, *tz, *rootdir, *locdir, *locale, *p;
405
406	if (fmd_hdl_register(hdl, FMD_API_VERSION, &fmd_info) != 0)
407		return; /* invalid data in configuration file */
408
409	(void) fmd_stat_create(hdl, FMD_STAT_NOALLOC, sizeof (syslog_stats) /
410	    sizeof (fmd_stat_t), (fmd_stat_t *)&syslog_stats);
411
412	if ((syslog_logfd = open("/dev/conslog", O_WRONLY | O_NOCTTY)) == -1)
413		fmd_hdl_abort(hdl, "syslog-msgs failed to open /dev/conslog");
414
415	if ((syslog_msgfd = open("/dev/sysmsg", O_WRONLY | O_NOCTTY)) == -1)
416		fmd_hdl_abort(hdl, "syslog-msgs failed to open /dev/sysmsg");
417
418	/*
419	 * All FMA event dictionaries use msgfmt(1) message objects to produce
420	 * messages, even for the C locale.  We therefore want to use dgettext
421	 * for all message lookups, but its defined behavior in the C locale is
422	 * to return the input string.  Since our input strings are event codes
423	 * and not format strings, this doesn't help us.  We resolve this nit
424	 * by setting NLSPATH to a non-existent file: the presence of NLSPATH
425	 * is defined to force dgettext(3C) to do a full lookup even for C.
426	 */
427	if (getenv("NLSPATH") == NULL && putenv(fmd_hdl_strdup(hdl,
428	    "NLSPATH=/usr/lib/fm/fmd/fmd.cat", FMD_SLEEP)) != 0)
429		fmd_hdl_abort(hdl, "syslog-msgs failed to set NLSPATH");
430
431	fmd_msg_lock();
432	(void) setlocale(LC_MESSAGES, "");
433	locale = setlocale(LC_MESSAGES, NULL);
434	if (locale) {
435		p = alloca(strlen(locale) + 1);
436		locale = strcpy(p, locale);
437	} else {
438		locale = "<null>";
439	}
440	fmd_msg_unlock();
441	fmd_hdl_debug(hdl, "locale=%s\n", locale);
442
443	/*
444	 * If the "gmt" property is set to true, force our EVENT-TIME to be
445	 * reported in GMT time; otherwise we use localtime.  tzset() affects
446	 * the results of subsequent calls to strftime(3C) above.
447	 */
448	if (fmd_prop_get_int32(hdl, "gmt") == FMD_B_TRUE &&
449	    ((tz = getenv("TZ")) == NULL || strcmp(tz, "GMT") != 0)) {
450		(void) putenv(fmd_hdl_strdup(hdl, "TZ=GMT", FMD_SLEEP));
451		tzset(); /* reload env */
452	}
453
454	/*
455	 * Look up the value of the "facility" property and use it to determine
456	 * what syslog LOG_* facility value we use to fill in our log_ctl_t.
457	 * The details of our logging method are described above syslog_emit().
458	 */
459	facname = fmd_prop_get_string(hdl, "facility");
460
461	for (fp = syslog_facs; fp->fac_name != NULL; fp++) {
462		if (strcmp(fp->fac_name, facname) == 0)
463			break;
464	}
465
466	if (fp->fac_name == NULL)
467		fmd_hdl_abort(hdl, "invalid 'facility' setting: %s\n", facname);
468
469	fmd_prop_free_string(hdl, facname);
470	syslog_ctl.pri = fp->fac_value;
471	syslog_ctl.flags = SL_CONSOLE | SL_LOGONLY;
472
473	/*
474	 * Cache any properties we use every time we receive an event and
475	 * subscribe to list.suspect events regardless of the .conf file.
476	 */
477	syslog_file = fmd_prop_get_int32(hdl, "syslogd");
478	syslog_cons = fmd_prop_get_int32(hdl, "console");
479	syslog_url = fmd_prop_get_string(hdl, "url");
480	syslog_msgall = fmd_prop_get_int32(hdl, "message_all");
481
482	/*
483	 * If fmd's rootdir property is set to a non-default root, then we are
484	 * going to need to rebind the text domains we use for dgettext() as
485	 * we go.  Look up the default l10n messages directory and make
486	 * syslog_locdir be this path with fmd.rootdir prepended to it.
487	 */
488	rootdir = fmd_prop_get_string(hdl, "fmd.rootdir");
489
490	if (*rootdir != '\0' && strcmp(rootdir, "/") != 0) {
491		fmd_msg_lock();
492		locdir = bindtextdomain(SYSLOG_DOMAIN, NULL);
493		fmd_msg_unlock();
494		if (locdir != NULL) {
495			size_t len = strlen(rootdir) + strlen(locdir) + 1;
496			syslog_locdir = fmd_hdl_alloc(hdl, len, FMD_SLEEP);
497			(void) snprintf(syslog_locdir, len, "%s%s", rootdir,
498			    locdir);
499			fmd_hdl_debug(hdl,
500			    "binding textdomain to %s for syslog\n",
501			    syslog_locdir);
502		}
503	}
504
505	fmd_prop_free_string(hdl, rootdir);
506	fmd_hdl_subscribe(hdl, FM_LIST_SUSPECT_CLASS);
507	fmd_hdl_subscribe(hdl, FM_LIST_REPAIRED_CLASS);
508}
509
510void
511_fmd_fini(fmd_hdl_t *hdl)
512{
513	fmd_hdl_strfree(hdl, syslog_locdir);
514	fmd_prop_free_string(hdl, syslog_url);
515
516	(void) close(syslog_logfd);
517	(void) close(syslog_msgfd);
518}
519