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 /*
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
24 * Use is subject to license terms.
25 */
26
27 #include <dlfcn.h>
28 #include <errno.h>
29 #include <libintl.h>
30 #include <link.h>
31 #include <pthread.h>
32 #include <strings.h>
33 #include <unistd.h>
34
35 #include <libzfs.h>
36
37 #include <fm/libtopo.h>
38 #include <sys/fm/protocol.h>
39 #include <sys/systeminfo.h>
40
41 #include "libzfs_impl.h"
42
43 /*
44 * This file is responsible for determining the relationship between I/O
45 * devices paths and physical locations. In the world of MPxIO and external
46 * enclosures, the device path is not synonymous with the physical location.
47 * If you remove a drive and insert it into a different slot, it will end up
48 * with the same path under MPxIO. If you recable storage enclosures, the
49 * device paths may change. All of this makes it difficult to implement the
50 * 'autoreplace' property, which is supposed to automatically manage disk
51 * replacement based on physical slot.
52 *
53 * In order to work around these limitations, we have a per-vdev FRU property
54 * that is the libtopo path (minus disk-specific authority information) to the
55 * physical location of the device on the system. This is an optional
56 * property, and is only needed when using the 'autoreplace' property or when
57 * generating FMA faults against vdevs.
58 */
59
60 /*
61 * Because the FMA packages depend on ZFS, we have to dlopen() libtopo in case
62 * it is not present. We only need this once per library instance, so it is
63 * not part of the libzfs handle.
64 */
65 static void *_topo_dlhandle;
66 static topo_hdl_t *(*_topo_open)(int, const char *, int *);
67 static void (*_topo_close)(topo_hdl_t *);
68 static char *(*_topo_snap_hold)(topo_hdl_t *, const char *, int *);
69 static void (*_topo_snap_release)(topo_hdl_t *);
70 static topo_walk_t *(*_topo_walk_init)(topo_hdl_t *, const char *,
71 topo_walk_cb_t, void *, int *);
72 static int (*_topo_walk_step)(topo_walk_t *, int);
73 static void (*_topo_walk_fini)(topo_walk_t *);
74 static void (*_topo_hdl_strfree)(topo_hdl_t *, char *);
75 static char *(*_topo_node_name)(tnode_t *);
76 static int (*_topo_prop_get_string)(tnode_t *, const char *, const char *,
77 char **, int *);
78 static int (*_topo_node_fru)(tnode_t *, nvlist_t **, nvlist_t *, int *);
79 static int (*_topo_fmri_nvl2str)(topo_hdl_t *, nvlist_t *, char **, int *);
80 static int (*_topo_fmri_strcmp_noauth)(topo_hdl_t *, const char *,
81 const char *);
82
83 #define ZFS_FRU_HASH_SIZE 257
84
85 static size_t
86 fru_strhash(const char *key)
87 {
88 ulong_t g, h = 0;
89 const char *p;
90
91 for (p = key; *p != '\0'; p++) {
92 h = (h << 4) + *p;
93
94 if ((g = (h & 0xf0000000)) != 0) {
95 h ^= (g >> 24);
96 h ^= g;
97 }
98 }
99
100 return (h % ZFS_FRU_HASH_SIZE);
101 }
102
103 static int
104 libzfs_fru_gather(topo_hdl_t *thp, tnode_t *tn, void *arg)
105 {
106 libzfs_handle_t *hdl = arg;
107 nvlist_t *fru;
108 char *devpath, *frustr;
109 int err;
110 libzfs_fru_t *frup;
111 size_t idx;
112
113 /*
114 * If this is the chassis node, and we don't yet have the system
115 * chassis ID, then fill in this value now.
116 */
117 if (hdl->libzfs_chassis_id[0] == '\0' &&
118 strcmp(_topo_node_name(tn), "chassis") == 0) {
119 if (_topo_prop_get_string(tn, FM_FMRI_AUTHORITY,
120 FM_FMRI_AUTH_CHASSIS, &devpath, &err) == 0)
121 (void) strlcpy(hdl->libzfs_chassis_id, devpath,
122 sizeof (hdl->libzfs_chassis_id));
123 }
124
125 /*
126 * Skip non-disk nodes.
127 */
128 if (strcmp(_topo_node_name(tn), "disk") != 0)
129 return (TOPO_WALK_NEXT);
130
131 /*
132 * Get the devfs path and FRU.
133 */
134 if (_topo_prop_get_string(tn, "io", "devfs-path", &devpath, &err) != 0)
135 return (TOPO_WALK_NEXT);
136
137 if (libzfs_fru_lookup(hdl, devpath) != NULL) {
138 _topo_hdl_strfree(thp, devpath);
139 return (TOPO_WALK_NEXT);
140 }
141
142 if (_topo_node_fru(tn, &fru, NULL, &err) != 0) {
143 _topo_hdl_strfree(thp, devpath);
144 return (TOPO_WALK_NEXT);
145 }
146
147 /*
148 * Convert the FRU into a string.
149 */
150 if (_topo_fmri_nvl2str(thp, fru, &frustr, &err) != 0) {
151 nvlist_free(fru);
152 _topo_hdl_strfree(thp, devpath);
153 return (TOPO_WALK_NEXT);
154 }
155
156 nvlist_free(fru);
157
158 /*
159 * Finally, we have a FRU string and device path. Add it to the hash.
160 */
161 if ((frup = calloc(sizeof (libzfs_fru_t), 1)) == NULL) {
162 _topo_hdl_strfree(thp, devpath);
163 _topo_hdl_strfree(thp, frustr);
164 return (TOPO_WALK_NEXT);
165 }
166
167 if ((frup->zf_device = strdup(devpath)) == NULL ||
168 (frup->zf_fru = strdup(frustr)) == NULL) {
169 free(frup->zf_device);
170 free(frup);
171 _topo_hdl_strfree(thp, devpath);
172 _topo_hdl_strfree(thp, frustr);
173 return (TOPO_WALK_NEXT);
174 }
175
176 _topo_hdl_strfree(thp, devpath);
177 _topo_hdl_strfree(thp, frustr);
178
179 idx = fru_strhash(frup->zf_device);
180 frup->zf_chain = hdl->libzfs_fru_hash[idx];
181 hdl->libzfs_fru_hash[idx] = frup;
182 frup->zf_next = hdl->libzfs_fru_list;
183 hdl->libzfs_fru_list = frup;
184
185 return (TOPO_WALK_NEXT);
186 }
187
188 /*
189 * Called during initialization to setup the dynamic libtopo connection.
190 */
191 #pragma init(libzfs_init_fru)
192 static void
193 libzfs_init_fru(void)
194 {
195 char path[MAXPATHLEN];
196 char isa[257];
197
198 #if defined(_LP64)
199 if (sysinfo(SI_ARCHITECTURE_64, isa, sizeof (isa)) < 0)
200 isa[0] = '\0';
201 #else
202 isa[0] = '\0';
203 #endif
204 (void) snprintf(path, sizeof (path),
205 "/usr/lib/fm/%s/libtopo.so", isa);
206
207 if ((_topo_dlhandle = dlopen(path, RTLD_LAZY)) == NULL)
208 return;
209
210 _topo_open = (topo_hdl_t *(*)())
211 dlsym(_topo_dlhandle, "topo_open");
212 _topo_close = (void (*)())
213 dlsym(_topo_dlhandle, "topo_close");
214 _topo_snap_hold = (char *(*)())
215 dlsym(_topo_dlhandle, "topo_snap_hold");
216 _topo_snap_release = (void (*)())
217 dlsym(_topo_dlhandle, "topo_snap_release");
218 _topo_walk_init = (topo_walk_t *(*)())
219 dlsym(_topo_dlhandle, "topo_walk_init");
220 _topo_walk_step = (int (*)())
221 dlsym(_topo_dlhandle, "topo_walk_step");
222 _topo_walk_fini = (void (*)())
223 dlsym(_topo_dlhandle, "topo_walk_fini");
224 _topo_hdl_strfree = (void (*)())
225 dlsym(_topo_dlhandle, "topo_hdl_strfree");
226 _topo_node_name = (char *(*)())
227 dlsym(_topo_dlhandle, "topo_node_name");
228 _topo_prop_get_string = (int (*)())
229 dlsym(_topo_dlhandle, "topo_prop_get_string");
230 _topo_node_fru = (int (*)())
231 dlsym(_topo_dlhandle, "topo_node_fru");
232 _topo_fmri_nvl2str = (int (*)())
233 dlsym(_topo_dlhandle, "topo_fmri_nvl2str");
234 _topo_fmri_strcmp_noauth = (int (*)())
235 dlsym(_topo_dlhandle, "topo_fmri_strcmp_noauth");
236
237 if (_topo_open == NULL || _topo_close == NULL ||
238 _topo_snap_hold == NULL || _topo_snap_release == NULL ||
239 _topo_walk_init == NULL || _topo_walk_step == NULL ||
240 _topo_walk_fini == NULL || _topo_hdl_strfree == NULL ||
241 _topo_node_name == NULL || _topo_prop_get_string == NULL ||
242 _topo_node_fru == NULL || _topo_fmri_nvl2str == NULL ||
243 _topo_fmri_strcmp_noauth == NULL) {
244 (void) dlclose(_topo_dlhandle);
245 _topo_dlhandle = NULL;
246 }
247 }
248
249 /*
250 * Refresh the mappings from device path -> FMRI. We do this by walking the
251 * hc topology looking for disk nodes, and recording the io/devfs-path and FRU.
252 * Note that we strip out the disk-specific authority information (serial,
253 * part, revision, etc) so that we are left with only the identifying
254 * characteristics of the slot (hc path and chassis-id).
255 */
256 void
257 libzfs_fru_refresh(libzfs_handle_t *hdl)
258 {
259 int err;
260 char *uuid;
261 topo_hdl_t *thp;
262 topo_walk_t *twp;
263
264 if (_topo_dlhandle == NULL)
265 return;
266
267 /*
268 * Clear the FRU hash and initialize our basic structures.
269 */
270 libzfs_fru_clear(hdl, B_FALSE);
271
272 if ((hdl->libzfs_topo_hdl = _topo_open(TOPO_VERSION,
273 NULL, &err)) == NULL)
274 return;
275
276 thp = hdl->libzfs_topo_hdl;
277
278 if ((uuid = _topo_snap_hold(thp, NULL, &err)) == NULL)
279 return;
280
281 _topo_hdl_strfree(thp, uuid);
282
283 if (hdl->libzfs_fru_hash == NULL &&
284 (hdl->libzfs_fru_hash =
285 calloc(ZFS_FRU_HASH_SIZE, sizeof (void *))) == NULL)
286 return;
287
288 /*
289 * We now have a topo snapshot, so iterate over the hc topology looking
290 * for disks to add to the hash.
291 */
292 twp = _topo_walk_init(thp, FM_FMRI_SCHEME_HC,
293 libzfs_fru_gather, hdl, &err);
294 if (twp != NULL) {
295 (void) _topo_walk_step(twp, TOPO_WALK_CHILD);
296 _topo_walk_fini(twp);
297 }
298 }
299
300 /*
301 * Given a devfs path, return the FRU for the device, if known. This will
302 * automatically call libzfs_fru_refresh() if it hasn't already been called by
303 * the consumer. The string returned is valid until the next call to
304 * libzfs_fru_refresh().
305 */
306 const char *
307 libzfs_fru_lookup(libzfs_handle_t *hdl, const char *devpath)
308 {
309 size_t idx = fru_strhash(devpath);
310 libzfs_fru_t *frup;
311
312 if (hdl->libzfs_fru_hash == NULL)
313 libzfs_fru_refresh(hdl);
314
315 if (hdl->libzfs_fru_hash == NULL)
316 return (NULL);
317
318 for (frup = hdl->libzfs_fru_hash[idx]; frup != NULL;
319 frup = frup->zf_chain) {
320 if (strcmp(devpath, frup->zf_device) == 0)
321 return (frup->zf_fru);
322 }
323
324 return (NULL);
325 }
326
327 /*
328 * Given a fru path, return the device path. This will automatically call
329 * libzfs_fru_refresh() if it hasn't already been called by the consumer. The
330 * string returned is valid until the next call to libzfs_fru_refresh().
331 */
332 const char *
333 libzfs_fru_devpath(libzfs_handle_t *hdl, const char *fru)
334 {
335 libzfs_fru_t *frup;
336 size_t idx;
337
338 if (hdl->libzfs_fru_hash == NULL)
339 libzfs_fru_refresh(hdl);
340
341 if (hdl->libzfs_fru_hash == NULL)
342 return (NULL);
343
344 for (idx = 0; idx < ZFS_FRU_HASH_SIZE; idx++) {
345 for (frup = hdl->libzfs_fru_hash[idx]; frup != NULL;
346 frup = frup->zf_next) {
347 if (_topo_fmri_strcmp_noauth(hdl->libzfs_topo_hdl,
348 fru, frup->zf_fru))
349 return (frup->zf_device);
350 }
351 }
352
353 return (NULL);
354 }
355
356 /*
357 * Change the stored FRU for the given vdev.
358 */
359 int
360 zpool_fru_set(zpool_handle_t *zhp, uint64_t vdev_guid, const char *fru)
361 {
362 zfs_cmd_t zc = { 0 };
363
364 (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
365 (void) strncpy(zc.zc_value, fru, sizeof (zc.zc_value));
366 zc.zc_guid = vdev_guid;
367
368 if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_VDEV_SETFRU, &zc) != 0)
369 return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
370 dgettext(TEXT_DOMAIN, "cannot set FRU")));
371
372 return (0);
373 }
374
375 /*
376 * Compare to two FRUs, ignoring any authority information.
377 */
378 boolean_t
379 libzfs_fru_compare(libzfs_handle_t *hdl, const char *a, const char *b)
380 {
381 if (hdl->libzfs_fru_hash == NULL)
382 libzfs_fru_refresh(hdl);
383
384 if (hdl->libzfs_fru_hash == NULL)
385 return (strcmp(a, b) == 0);
386
387 return (_topo_fmri_strcmp_noauth(hdl->libzfs_topo_hdl, a, b));
388 }
389
390 /*
391 * This special function checks to see whether the FRU indicates it's supposed
392 * to be in the system chassis, but the chassis-id doesn't match. This can
393 * happen in a clustered case, where both head nodes have the same logical
394 * disk, but opening the device on the other head node is meaningless.
395 */
396 boolean_t
397 libzfs_fru_notself(libzfs_handle_t *hdl, const char *fru)
398 {
399 const char *chassisid;
400 size_t len;
401
402 if (hdl->libzfs_fru_hash == NULL)
403 libzfs_fru_refresh(hdl);
404
405 if (hdl->libzfs_chassis_id[0] == '\0')
406 return (B_FALSE);
407
408 if (strstr(fru, "/chassis=0/") == NULL)
409 return (B_FALSE);
410
411 if ((chassisid = strstr(fru, ":chassis-id=")) == NULL)
412 return (B_FALSE);
413
414 chassisid += 12;
415 len = strlen(hdl->libzfs_chassis_id);
416 if (strncmp(chassisid, hdl->libzfs_chassis_id, len) == 0 &&
417 (chassisid[len] == '/' || chassisid[len] == ':'))
418 return (B_FALSE);
419
420 return (B_TRUE);
421 }
422
423 /*
424 * Clear memory associated with the FRU hash.
425 */
426 void
427 libzfs_fru_clear(libzfs_handle_t *hdl, boolean_t final)
428 {
429 libzfs_fru_t *frup;
430
431 while ((frup = hdl->libzfs_fru_list) != NULL) {
432 hdl->libzfs_fru_list = frup->zf_next;
433 free(frup->zf_device);
434 free(frup->zf_fru);
435 free(frup);
436 }
437
438 hdl->libzfs_fru_list = NULL;
439
440 if (hdl->libzfs_topo_hdl != NULL) {
441 _topo_snap_release(hdl->libzfs_topo_hdl);
442 _topo_close(hdl->libzfs_topo_hdl);
443 hdl->libzfs_topo_hdl = NULL;
444 }
445
446 if (final) {
447 free(hdl->libzfs_fru_hash);
448 } else if (hdl->libzfs_fru_hash != NULL) {
449 bzero(hdl->libzfs_fru_hash,
450 ZFS_FRU_HASH_SIZE * sizeof (void *));
451 }
452 }