xref: /illumos-gate/usr/src/cmd/fm/modules/common/zfs-retire/zfs_retire.c (revision 7a088f03b431bdffa96c3b2175964d4d38420caa)
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 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*
27  * The ZFS retire agent is responsible for managing hot spares across all pools.
28  * When we see a device fault or a device removal, we try to open the associated
29  * pool and look for any hot spares.  We iterate over any available hot spares
30  * and attempt a 'zpool replace' for each one.
31  *
32  * For vdevs diagnosed as faulty, the agent is also responsible for proactively
33  * marking the vdev FAULTY (for I/O errors) or DEGRADED (for checksum errors).
34  */
35 
36 #include <fm/fmd_api.h>
37 #include <sys/fs/zfs.h>
38 #include <sys/fm/protocol.h>
39 #include <sys/fm/fs/zfs.h>
40 #include <libzfs.h>
41 #include <string.h>
42 
43 /*
44  * Find a pool with a matching GUID.
45  */
46 typedef struct find_cbdata {
47 	uint64_t	cb_guid;
48 	zpool_handle_t	*cb_zhp;
49 } find_cbdata_t;
50 
51 static int
52 find_pool(zpool_handle_t *zhp, void *data)
53 {
54 	find_cbdata_t *cbp = data;
55 
56 	if (cbp->cb_guid ==
57 	    zpool_get_prop_int(zhp, ZPOOL_PROP_GUID, NULL)) {
58 		cbp->cb_zhp = zhp;
59 		return (1);
60 	}
61 
62 	zpool_close(zhp);
63 	return (0);
64 }
65 
66 /*
67  * Find a vdev within a tree with a matching GUID.
68  */
69 static nvlist_t *
70 find_vdev(nvlist_t *nv, uint64_t search)
71 {
72 	uint64_t guid;
73 	nvlist_t **child;
74 	uint_t c, children;
75 	nvlist_t *ret;
76 
77 	if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0 &&
78 	    guid == search)
79 		return (nv);
80 
81 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
82 	    &child, &children) != 0)
83 		return (NULL);
84 
85 	for (c = 0; c < children; c++) {
86 		if ((ret = find_vdev(child[c], search)) != NULL)
87 			return (ret);
88 	}
89 
90 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
91 	    &child, &children) != 0)
92 		return (NULL);
93 
94 	for (c = 0; c < children; c++) {
95 		if ((ret = find_vdev(child[c], search)) != NULL)
96 			return (ret);
97 	}
98 
99 	return (NULL);
100 }
101 
102 /*
103  * Given a (pool, vdev) GUID pair, find the matching pool and vdev.
104  */
105 static zpool_handle_t *
106 find_by_guid(libzfs_handle_t *zhdl, uint64_t pool_guid, uint64_t vdev_guid,
107     nvlist_t **vdevp)
108 {
109 	find_cbdata_t cb;
110 	zpool_handle_t *zhp;
111 	nvlist_t *config, *nvroot;
112 
113 	/*
114 	 * Find the corresponding pool and make sure the vdev still exists.
115 	 */
116 	cb.cb_guid = pool_guid;
117 	if (zpool_iter(zhdl, find_pool, &cb) != 1)
118 		return (NULL);
119 
120 	zhp = cb.cb_zhp;
121 	config = zpool_get_config(zhp, NULL);
122 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
123 	    &nvroot) != 0) {
124 		zpool_close(zhp);
125 		return (NULL);
126 	}
127 
128 	if ((*vdevp = find_vdev(nvroot, vdev_guid)) == NULL) {
129 		zpool_close(zhp);
130 		return (NULL);
131 	}
132 
133 	return (zhp);
134 }
135 
136 /*
137  * Given a vdev, attempt to replace it with every known spare until one
138  * succeeds.
139  */
140 static void
141 replace_with_spare(zpool_handle_t *zhp, nvlist_t *vdev)
142 {
143 	nvlist_t *config, *nvroot, *replacement;
144 	nvlist_t **spares;
145 	uint_t s, nspares;
146 	char *dev_name;
147 
148 	config = zpool_get_config(zhp, NULL);
149 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
150 	    &nvroot) != 0)
151 		return;
152 
153 	/*
154 	 * Find out if there are any hot spares available in the pool.
155 	 */
156 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
157 	    &spares, &nspares) != 0)
158 		return;
159 
160 	if (nvlist_alloc(&replacement, NV_UNIQUE_NAME, 0) != 0)
161 		return;
162 
163 	if (nvlist_add_string(replacement, ZPOOL_CONFIG_TYPE,
164 	    VDEV_TYPE_ROOT) != 0) {
165 		nvlist_free(replacement);
166 		return;
167 	}
168 
169 	dev_name = zpool_vdev_name(NULL, zhp, vdev);
170 
171 	/*
172 	 * Try to replace each spare, ending when we successfully
173 	 * replace it.
174 	 */
175 	for (s = 0; s < nspares; s++) {
176 		char *spare_name;
177 
178 		if (nvlist_lookup_string(spares[s], ZPOOL_CONFIG_PATH,
179 		    &spare_name) != 0)
180 			continue;
181 
182 		if (nvlist_add_nvlist_array(replacement,
183 		    ZPOOL_CONFIG_CHILDREN, &spares[s], 1) != 0)
184 			continue;
185 
186 		if (zpool_vdev_attach(zhp, dev_name, spare_name,
187 		    replacement, B_TRUE) == 0)
188 			break;
189 	}
190 
191 	free(dev_name);
192 	nvlist_free(replacement);
193 }
194 
195 /*ARGSUSED*/
196 static void
197 zfs_retire_recv(fmd_hdl_t *hdl, fmd_event_t *ep, nvlist_t *nvl,
198     const char *class)
199 {
200 	uint64_t pool_guid, vdev_guid;
201 	zpool_handle_t *zhp;
202 	nvlist_t *resource, *fault;
203 	nvlist_t **faults;
204 	uint_t f, nfaults;
205 	libzfs_handle_t *zhdl = fmd_hdl_getspecific(hdl);
206 	boolean_t fault_device, degrade_device;
207 	boolean_t is_repair;
208 	char *scheme;
209 	nvlist_t *vdev;
210 	char *uuid;
211 	int repair_done = 0;
212 	boolean_t retire;
213 
214 	/*
215 	 * If this is a resource notifying us of device removal, then simply
216 	 * check for an available spare and continue.
217 	 */
218 	if (strcmp(class, "resource.fs.zfs.removed") == 0) {
219 		if (nvlist_lookup_uint64(nvl, FM_EREPORT_PAYLOAD_ZFS_POOL_GUID,
220 		    &pool_guid) != 0 ||
221 		    nvlist_lookup_uint64(nvl, FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID,
222 		    &vdev_guid) != 0)
223 			return;
224 
225 		if ((zhp = find_by_guid(zhdl, pool_guid, vdev_guid,
226 		    &vdev)) == NULL)
227 			return;
228 
229 		if (fmd_prop_get_int32(hdl, "spare_on_remove"))
230 			replace_with_spare(zhp, vdev);
231 		zpool_close(zhp);
232 		return;
233 	}
234 
235 	if (strcmp(class, FM_LIST_RESOLVED_CLASS) == 0)
236 		return;
237 
238 	if (strcmp(class, FM_LIST_REPAIRED_CLASS) == 0)
239 		is_repair = B_TRUE;
240 	else
241 		is_repair = B_FALSE;
242 
243 	/*
244 	 * We subscribe to zfs faults as well as all repair events.
245 	 */
246 	if (nvlist_lookup_nvlist_array(nvl, FM_SUSPECT_FAULT_LIST,
247 	    &faults, &nfaults) != 0)
248 		return;
249 
250 	for (f = 0; f < nfaults; f++) {
251 		fault = faults[f];
252 
253 		fault_device = B_FALSE;
254 		degrade_device = B_FALSE;
255 
256 		if (nvlist_lookup_boolean_value(fault, FM_SUSPECT_RETIRE,
257 		    &retire) == 0 && retire == 0)
258 			continue;
259 
260 		/*
261 		 * While we subscribe to fault.fs.zfs.*, we only take action
262 		 * for faults targeting a specific vdev (open failure or SERD
263 		 * failure).
264 		 */
265 		if (fmd_nvl_class_match(hdl, fault, "fault.fs.zfs.vdev.io"))
266 			fault_device = B_TRUE;
267 		else if (fmd_nvl_class_match(hdl, fault,
268 		    "fault.fs.zfs.vdev.checksum"))
269 			degrade_device = B_TRUE;
270 		else if (fmd_nvl_class_match(hdl, fault, "fault.fs.zfs.device"))
271 			fault_device = B_FALSE;
272 		else
273 			continue;
274 
275 		if (nvlist_lookup_nvlist(fault, FM_FAULT_RESOURCE,
276 		    &resource) != 0 ||
277 		    nvlist_lookup_string(resource, FM_FMRI_SCHEME,
278 		    &scheme) != 0)
279 			continue;
280 
281 		if (strcmp(scheme, FM_FMRI_SCHEME_ZFS) != 0)
282 			continue;
283 
284 		if (nvlist_lookup_uint64(resource, FM_FMRI_ZFS_POOL,
285 		    &pool_guid) != 0 ||
286 		    nvlist_lookup_uint64(resource, FM_FMRI_ZFS_VDEV,
287 		    &vdev_guid) != 0)
288 			continue;
289 
290 		if ((zhp = find_by_guid(zhdl, pool_guid, vdev_guid,
291 		    &vdev)) == NULL)
292 			continue;
293 
294 		/*
295 		 * If this is a repair event, then mark the vdev as repaired and
296 		 * continue.
297 		 */
298 		if (is_repair) {
299 			repair_done = 1;
300 			(void) zpool_vdev_clear(zhp, vdev_guid);
301 			zpool_close(zhp);
302 			continue;
303 		}
304 
305 		/*
306 		 * Actively fault the device if needed.
307 		 */
308 		if (fault_device)
309 			(void) zpool_vdev_fault(zhp, vdev_guid);
310 		if (degrade_device)
311 			(void) zpool_vdev_degrade(zhp, vdev_guid);
312 
313 		/*
314 		 * Attempt to substitute a hot spare.
315 		 */
316 		replace_with_spare(zhp, vdev);
317 		zpool_close(zhp);
318 	}
319 
320 	if (strcmp(class, FM_LIST_REPAIRED_CLASS) == 0 && repair_done &&
321 	    nvlist_lookup_string(nvl, FM_SUSPECT_UUID, &uuid) == 0)
322 		fmd_case_uuresolved(hdl, uuid);
323 }
324 
325 static const fmd_hdl_ops_t fmd_ops = {
326 	zfs_retire_recv,	/* fmdo_recv */
327 	NULL,			/* fmdo_timeout */
328 	NULL,			/* fmdo_close */
329 	NULL,			/* fmdo_stats */
330 	NULL,			/* fmdo_gc */
331 };
332 
333 static const fmd_prop_t fmd_props[] = {
334 	{ "spare_on_remove", FMD_TYPE_BOOL, "true" },
335 	{ NULL, 0, NULL }
336 };
337 
338 static const fmd_hdl_info_t fmd_info = {
339 	"ZFS Retire Agent", "1.0", &fmd_ops, fmd_props
340 };
341 
342 void
343 _fmd_init(fmd_hdl_t *hdl)
344 {
345 	libzfs_handle_t *zhdl;
346 
347 	if ((zhdl = libzfs_init()) == NULL)
348 		return;
349 
350 	if (fmd_hdl_register(hdl, FMD_API_VERSION, &fmd_info) != 0) {
351 		libzfs_fini(zhdl);
352 		return;
353 	}
354 
355 	fmd_hdl_setspecific(hdl, zhdl);
356 }
357 
358 void
359 _fmd_fini(fmd_hdl_t *hdl)
360 {
361 	libzfs_handle_t *zhdl = fmd_hdl_getspecific(hdl);
362 
363 	if (zhdl != NULL)
364 		libzfs_fini(zhdl);
365 }
366