xref: /titanic_41/usr/src/cmd/fm/modules/common/disk-transport/disk_transport.c (revision 184cd04c26b064536977dfbb913a1240eaf6f708)
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 2007 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 /*
29  * Disk error transport module
30  *
31  * This transport module is responsible for translating between disk errors
32  * and FMA ereports.  It is a read-only transport module, and checks for the
33  * following failures:
34  *
35  * 	- overtemp
36  * 	- predictive failure
37  * 	- self-test failure
38  *
39  * These failures are detected via the TOPO_METH_DISK_STATUS method, which
40  * leverages libdiskstatus to do the actual analysis.  This transport module is
41  * in charge of the following tasks:
42  *
43  * 	- discovering available devices
44  * 	- periodically checking devices
45  * 	- managing device addition/removal
46  */
47 
48 #include <ctype.h>
49 #include <fm/fmd_api.h>
50 #include <fm/libdiskstatus.h>
51 #include <fm/libtopo.h>
52 #include <fm/topo_hc.h>
53 #include <fm/topo_mod.h>
54 #include <limits.h>
55 #include <string.h>
56 #include <sys/fm/io/scsi.h>
57 #include <sys/fm/protocol.h>
58 
59 static struct dt_stat {
60 	fmd_stat_t dropped;
61 } dt_stats = {
62 	{ "dropped", FMD_TYPE_UINT64, "number of dropped ereports" }
63 };
64 
65 typedef struct disk_monitor {
66 	fmd_hdl_t	*dm_hdl;
67 	fmd_xprt_t	*dm_xprt;
68 	id_t		dm_timer;
69 	hrtime_t	dm_interval;
70 	char		*dm_sim_search;
71 	char		*dm_sim_file;
72 	boolean_t	dm_timer_istopo;
73 } disk_monitor_t;
74 
75 static void
dt_post_ereport(fmd_hdl_t * hdl,fmd_xprt_t * xprt,const char * protocol,const char * faultname,uint64_t ena,nvlist_t * detector,nvlist_t * payload)76 dt_post_ereport(fmd_hdl_t *hdl, fmd_xprt_t *xprt, const char *protocol,
77     const char *faultname, uint64_t ena, nvlist_t *detector, nvlist_t *payload)
78 {
79 	nvlist_t *nvl;
80 	int e = 0;
81 	char fullclass[PATH_MAX];
82 
83 	(void) snprintf(fullclass, sizeof (fullclass), "%s.io.%s.disk.%s",
84 	    FM_EREPORT_CLASS, protocol, faultname);
85 
86 	if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) == 0) {
87 		e |= nvlist_add_string(nvl, FM_CLASS, fullclass);
88 		e |= nvlist_add_uint8(nvl, FM_VERSION, FM_EREPORT_VERSION);
89 		e |= nvlist_add_uint64(nvl, FM_EREPORT_ENA, ena);
90 		e |= nvlist_add_nvlist(nvl, FM_EREPORT_DETECTOR, detector);
91 		e |= nvlist_merge(nvl, payload, 0);
92 
93 		if (e == 0) {
94 			fmd_xprt_post(hdl, xprt, nvl, 0);
95 		} else {
96 			nvlist_free(nvl);
97 			dt_stats.dropped.fmds_value.ui64++;
98 		}
99 	} else {
100 		dt_stats.dropped.fmds_value.ui64++;
101 	}
102 }
103 
104 /*
105  * Check a single topo node for failure.  This simply invokes the disk status
106  * method, and generates any ereports as necessary.
107  */
108 static int
dt_analyze_disk(topo_hdl_t * thp,tnode_t * node,void * arg)109 dt_analyze_disk(topo_hdl_t *thp, tnode_t *node, void *arg)
110 {
111 	nvlist_t *result;
112 	nvlist_t *fmri, *faults;
113 	char *protocol;
114 	int err;
115 	disk_monitor_t *dmp = arg;
116 	uint64_t ena;
117 	nvpair_t *elem;
118 	boolean_t fault;
119 	nvlist_t *details;
120 	char *fmristr;
121 	nvlist_t *in = NULL;
122 
123 	if (topo_node_resource(node, &fmri, &err) != 0) {
124 		fmd_hdl_error(dmp->dm_hdl, "failed to get fmri: %s\n",
125 		    topo_strerror(err));
126 		return (TOPO_WALK_ERR);
127 	}
128 
129 	if (topo_hdl_nvalloc(thp, &in, NV_UNIQUE_NAME) != 0) {
130 		nvlist_free(fmri);
131 		return (TOPO_WALK_ERR);
132 	}
133 
134 	if (dmp->dm_sim_search) {
135 		fmristr = NULL;
136 		if (topo_fmri_nvl2str(thp, fmri, &fmristr, &err) == 0 &&
137 		    strstr(fmristr, dmp->dm_sim_search) != 0)
138 			(void) nvlist_add_string(in, "path", dmp->dm_sim_file);
139 		topo_hdl_strfree(thp, fmristr);
140 	}
141 
142 	/*
143 	 * Try to invoke the method.  If this fails (most likely because the
144 	 * method is not supported), then ignore this node.
145 	 */
146 	if (topo_method_invoke(node, TOPO_METH_DISK_STATUS,
147 	    TOPO_METH_DISK_STATUS_VERSION, in, &result, &err) != 0) {
148 		nvlist_free(fmri);
149 		nvlist_free(in);
150 		return (TOPO_WALK_NEXT);
151 	}
152 
153 	nvlist_free(in);
154 
155 	ena = fmd_event_ena_create(dmp->dm_hdl);
156 
157 	/*
158 	 * Add any faults.
159 	 */
160 	if (nvlist_lookup_nvlist(result, "faults", &faults) == 0 &&
161 	    nvlist_lookup_string(result, "protocol", &protocol) == 0) {
162 		elem = NULL;
163 		while ((elem = nvlist_next_nvpair(faults, elem)) != NULL) {
164 			if (nvpair_type(elem) != DATA_TYPE_BOOLEAN_VALUE)
165 				continue;
166 
167 			(void) nvpair_value_boolean_value(elem, &fault);
168 			if (!fault ||
169 			    nvlist_lookup_nvlist(result, nvpair_name(elem),
170 			    &details) != 0)
171 				continue;
172 
173 			dt_post_ereport(dmp->dm_hdl, dmp->dm_xprt, protocol,
174 			    nvpair_name(elem), ena, fmri, details);
175 		}
176 	}
177 
178 	nvlist_free(result);
179 	nvlist_free(fmri);
180 
181 	return (TOPO_WALK_NEXT);
182 }
183 
184 /*
185  * Periodic timeout.  Iterates over all hc:// topo nodes, calling
186  * dt_analyze_disk() for each one.
187  */
188 /*ARGSUSED*/
189 static void
dt_timeout(fmd_hdl_t * hdl,id_t id,void * data)190 dt_timeout(fmd_hdl_t *hdl, id_t id, void *data)
191 {
192 	topo_hdl_t *thp;
193 	topo_walk_t *twp;
194 	int err;
195 	disk_monitor_t *dmp = fmd_hdl_getspecific(hdl);
196 
197 	dmp->dm_hdl = hdl;
198 
199 	thp = fmd_hdl_topo_hold(hdl, TOPO_VERSION);
200 	if ((twp = topo_walk_init(thp, FM_FMRI_SCHEME_HC, dt_analyze_disk,
201 	    dmp, &err)) == NULL) {
202 		fmd_hdl_topo_rele(hdl, thp);
203 		fmd_hdl_error(hdl, "failed to get topology: %s\n",
204 		    topo_strerror(err));
205 		return;
206 	}
207 
208 	if (topo_walk_step(twp, TOPO_WALK_CHILD) == TOPO_WALK_ERR) {
209 		topo_walk_fini(twp);
210 		fmd_hdl_topo_rele(hdl, thp);
211 		fmd_hdl_error(hdl, "failed to walk topology\n");
212 		return;
213 	}
214 
215 	topo_walk_fini(twp);
216 	fmd_hdl_topo_rele(hdl, thp);
217 
218 	dmp->dm_timer = fmd_timer_install(hdl, NULL, NULL, dmp->dm_interval);
219 	dmp->dm_timer_istopo = B_FALSE;
220 }
221 
222 /*
223  * Called when the topology may have changed.  We want to examine all disks in
224  * case a new one has been inserted, but we don't want to overwhelm the system
225  * in the event of a flurry of topology changes, as most likely only a small
226  * number of disks are changing.  To avoid this, we set the timer for a small
227  * but non-trivial interval (by default 1 minute), and ignore intervening
228  * changes during this period.  This still gives us a reasonable response time
229  * to newly inserted devices without overwhelming the system if lots of hotplug
230  * activity is going on.
231  */
232 /*ARGSUSED*/
233 static void
dt_topo_change(fmd_hdl_t * hdl,topo_hdl_t * thp)234 dt_topo_change(fmd_hdl_t *hdl, topo_hdl_t *thp)
235 {
236 	disk_monitor_t *dmp = fmd_hdl_getspecific(hdl);
237 
238 	if (dmp->dm_timer_istopo)
239 		return;
240 
241 	fmd_timer_remove(hdl, dmp->dm_timer);
242 	dmp->dm_timer = fmd_timer_install(hdl, NULL, NULL,
243 	    fmd_prop_get_int64(hdl, "min-interval"));
244 	dmp->dm_timer_istopo = B_TRUE;
245 }
246 
247 static const fmd_prop_t fmd_props[] = {
248 	{ "interval", FMD_TYPE_TIME, "1h" },
249 	{ "min-interval", FMD_TYPE_TIME, "1min" },
250 	{ "simulate", FMD_TYPE_STRING, "" },
251 	{ NULL, 0, NULL }
252 };
253 
254 static const fmd_hdl_ops_t fmd_ops = {
255 	NULL,			/* fmdo_recv */
256 	dt_timeout,		/* fmdo_timeout */
257 	NULL, 			/* fmdo_close */
258 	NULL,			/* fmdo_stats */
259 	NULL,			/* fmdo_gc */
260 	NULL,			/* fmdo_send */
261 	dt_topo_change,		/* fmdo_topo_change */
262 };
263 
264 static const fmd_hdl_info_t fmd_info = {
265 	"Disk Transport Agent", "1.0", &fmd_ops, fmd_props
266 };
267 
268 void
_fmd_init(fmd_hdl_t * hdl)269 _fmd_init(fmd_hdl_t *hdl)
270 {
271 	disk_monitor_t *dmp;
272 	char *simulate;
273 
274 	if (fmd_hdl_register(hdl, FMD_API_VERSION, &fmd_info) != 0)
275 		return;
276 
277 	(void) fmd_stat_create(hdl, FMD_STAT_NOALLOC,
278 	    sizeof (dt_stats) / sizeof (fmd_stat_t),
279 	    (fmd_stat_t *)&dt_stats);
280 
281 	dmp = fmd_hdl_zalloc(hdl, sizeof (disk_monitor_t), FMD_SLEEP);
282 	fmd_hdl_setspecific(hdl, dmp);
283 
284 	dmp->dm_xprt = fmd_xprt_open(hdl, FMD_XPRT_RDONLY, NULL, NULL);
285 	dmp->dm_interval = fmd_prop_get_int64(hdl, "interval");
286 
287 	/*
288 	 * Determine if we have the simulate property set.  This property allows
289 	 * the developer to substitute a faulty device based off all or part of
290 	 * an FMRI string.  For example, one could do:
291 	 *
292 	 * 	setprop simulate "bay=4/disk=4	/path/to/sim.so"
293 	 *
294 	 * When the transport module encounters an FMRI containing the given
295 	 * string, then it will open the simulator file instead of the
296 	 * corresponding device.  This can be any file, but is intended to be a
297 	 * libdiskstatus simulator shared object, capable of faking up SCSI
298 	 * responses.
299 	 *
300 	 * The property consists of two strings, an FMRI fragment and an
301 	 * absolute path, separated by whitespace.
302 	 */
303 	simulate = fmd_prop_get_string(hdl, "simulate");
304 	if (simulate[0] != '\0') {
305 		const char *sep;
306 		size_t len;
307 
308 		for (sep = simulate; *sep != '\0'; sep++) {
309 			if (isspace(*sep))
310 				break;
311 		}
312 
313 		if (*sep != '\0') {
314 			len = sep - simulate;
315 
316 			dmp->dm_sim_search = fmd_hdl_alloc(hdl,
317 			    len + 1, FMD_SLEEP);
318 			(void) memcpy(dmp->dm_sim_search, simulate, len);
319 			dmp->dm_sim_search[len] = '\0';
320 		}
321 
322 		for (; *sep != '\0'; sep++) {
323 			if (!isspace(*sep))
324 				break;
325 		}
326 
327 		if (*sep != '\0') {
328 			dmp->dm_sim_file = fmd_hdl_strdup(hdl, sep, FMD_SLEEP);
329 		} else if (dmp->dm_sim_search) {
330 			fmd_hdl_strfree(hdl, dmp->dm_sim_search);
331 			dmp->dm_sim_search = NULL;
332 		}
333 	}
334 	fmd_prop_free_string(hdl, simulate);
335 
336 	/*
337 	 * Call our initial timer routine.  This will do an initial check of all
338 	 * the disks, and then start the periodic timeout.
339 	 */
340 	dmp->dm_timer = fmd_timer_install(hdl, NULL, NULL, 0);
341 }
342 
343 void
_fmd_fini(fmd_hdl_t * hdl)344 _fmd_fini(fmd_hdl_t *hdl)
345 {
346 	disk_monitor_t *dmp;
347 
348 	dmp = fmd_hdl_getspecific(hdl);
349 	if (dmp) {
350 		fmd_xprt_close(hdl, dmp->dm_xprt);
351 		fmd_hdl_strfree(hdl, dmp->dm_sim_search);
352 		fmd_hdl_strfree(hdl, dmp->dm_sim_file);
353 		fmd_hdl_free(hdl, dmp, sizeof (*dmp));
354 	}
355 }
356