xref: /freebsd/sys/contrib/openzfs/lib/libzfs/libzfs_import.c (revision ca53e5aedfebcc1b4091b68e01b2d5cae923f85e)
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 2015 Nexenta Systems, Inc. All rights reserved.
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
25  * Copyright 2015 RackTop Systems.
26  * Copyright (c) 2016, Intel Corporation.
27  */
28 
29 #include <errno.h>
30 #include <libintl.h>
31 #include <libgen.h>
32 #include <stddef.h>
33 #include <stdlib.h>
34 #include <string.h>
35 #include <sys/stat.h>
36 #include <unistd.h>
37 #include <sys/vdev_impl.h>
38 #include <libzfs.h>
39 #include <libzfs_impl.h>
40 #include <libzutil.h>
41 #include <sys/arc_impl.h>
42 
43 /*
44  * Returns true if the named pool matches the given GUID.
45  */
46 static int
47 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
48     boolean_t *isactive)
49 {
50 	zpool_handle_t *zhp;
51 	uint64_t theguid;
52 
53 	if (zpool_open_silent(hdl, name, &zhp) != 0)
54 		return (-1);
55 
56 	if (zhp == NULL) {
57 		*isactive = B_FALSE;
58 		return (0);
59 	}
60 
61 	verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
62 	    &theguid) == 0);
63 
64 	zpool_close(zhp);
65 
66 	*isactive = (theguid == guid);
67 	return (0);
68 }
69 
70 static nvlist_t *
71 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
72 {
73 	nvlist_t *nvl;
74 	zfs_cmd_t zc = {"\0"};
75 	int err, dstbuf_size;
76 
77 	if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
78 		return (NULL);
79 
80 	dstbuf_size = MAX(CONFIG_BUF_MINSIZE, zc.zc_nvlist_conf_size * 4);
81 
82 	if (zcmd_alloc_dst_nvlist(hdl, &zc, dstbuf_size) != 0) {
83 		zcmd_free_nvlists(&zc);
84 		return (NULL);
85 	}
86 
87 	while ((err = zfs_ioctl(hdl, ZFS_IOC_POOL_TRYIMPORT,
88 	    &zc)) != 0 && errno == ENOMEM) {
89 		if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
90 			zcmd_free_nvlists(&zc);
91 			return (NULL);
92 		}
93 	}
94 
95 	if (err) {
96 		zcmd_free_nvlists(&zc);
97 		return (NULL);
98 	}
99 
100 	if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
101 		zcmd_free_nvlists(&zc);
102 		return (NULL);
103 	}
104 
105 	zcmd_free_nvlists(&zc);
106 	return (nvl);
107 }
108 
109 static nvlist_t *
110 refresh_config_libzfs(void *handle, nvlist_t *tryconfig)
111 {
112 	return (refresh_config((libzfs_handle_t *)handle, tryconfig));
113 }
114 
115 
116 static int
117 pool_active_libzfs(void *handle, const char *name, uint64_t guid,
118     boolean_t *isactive)
119 {
120 	return (pool_active((libzfs_handle_t *)handle, name, guid, isactive));
121 }
122 
123 const pool_config_ops_t libzfs_config_ops = {
124 	.pco_refresh_config = refresh_config_libzfs,
125 	.pco_pool_active = pool_active_libzfs,
126 };
127 
128 /*
129  * Return the offset of the given label.
130  */
131 static uint64_t
132 label_offset(uint64_t size, int l)
133 {
134 	ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
135 	return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
136 	    0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
137 }
138 
139 /*
140  * Given a file descriptor, clear (zero) the label information.  This function
141  * is used in the appliance stack as part of the ZFS sysevent module and
142  * to implement the "zpool labelclear" command.
143  */
144 int
145 zpool_clear_label(int fd)
146 {
147 	struct stat64 statbuf;
148 	int l;
149 	vdev_label_t *label;
150 	l2arc_dev_hdr_phys_t *l2dhdr;
151 	uint64_t size;
152 	int labels_cleared = 0, header_cleared = 0;
153 	boolean_t clear_l2arc_header = B_FALSE;
154 
155 	if (fstat64_blk(fd, &statbuf) == -1)
156 		return (0);
157 
158 	size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
159 
160 	if ((label = calloc(1, sizeof (vdev_label_t))) == NULL)
161 		return (-1);
162 
163 	if ((l2dhdr = calloc(1, sizeof (l2arc_dev_hdr_phys_t))) == NULL) {
164 		free(label);
165 		return (-1);
166 	}
167 
168 	for (l = 0; l < VDEV_LABELS; l++) {
169 		uint64_t state, guid, l2cache;
170 		nvlist_t *config;
171 
172 		if (pread64(fd, label, sizeof (vdev_label_t),
173 		    label_offset(size, l)) != sizeof (vdev_label_t)) {
174 			continue;
175 		}
176 
177 		if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
178 		    sizeof (label->vl_vdev_phys.vp_nvlist), &config, 0) != 0) {
179 			continue;
180 		}
181 
182 		/* Skip labels which do not have a valid guid. */
183 		if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
184 		    &guid) != 0 || guid == 0) {
185 			nvlist_free(config);
186 			continue;
187 		}
188 
189 		/* Skip labels which are not in a known valid state. */
190 		if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
191 		    &state) != 0 || state > POOL_STATE_L2CACHE) {
192 			nvlist_free(config);
193 			continue;
194 		}
195 
196 		/* If the device is a cache device clear the header. */
197 		if (!clear_l2arc_header) {
198 			if (nvlist_lookup_uint64(config,
199 			    ZPOOL_CONFIG_POOL_STATE, &l2cache) == 0 &&
200 			    l2cache == POOL_STATE_L2CACHE) {
201 				clear_l2arc_header = B_TRUE;
202 			}
203 		}
204 
205 		nvlist_free(config);
206 
207 		/*
208 		 * A valid label was found, overwrite this label's nvlist
209 		 * and uberblocks with zeros on disk.  This is done to prevent
210 		 * system utilities, like blkid, from incorrectly detecting a
211 		 * partial label.  The leading pad space is left untouched.
212 		 */
213 		memset(label, 0, sizeof (vdev_label_t));
214 		size_t label_size = sizeof (vdev_label_t) - (2 * VDEV_PAD_SIZE);
215 
216 		if (pwrite64(fd, label, label_size, label_offset(size, l) +
217 		    (2 * VDEV_PAD_SIZE)) == label_size) {
218 			labels_cleared++;
219 		}
220 	}
221 
222 	/* Clear the L2ARC header. */
223 	if (clear_l2arc_header) {
224 		memset(l2dhdr, 0, sizeof (l2arc_dev_hdr_phys_t));
225 		if (pwrite64(fd, l2dhdr, sizeof (l2arc_dev_hdr_phys_t),
226 		    VDEV_LABEL_START_SIZE) == sizeof (l2arc_dev_hdr_phys_t)) {
227 			header_cleared++;
228 		}
229 	}
230 
231 	free(label);
232 	free(l2dhdr);
233 
234 	if (labels_cleared == 0)
235 		return (-1);
236 
237 	return (0);
238 }
239 
240 static boolean_t
241 find_guid(nvlist_t *nv, uint64_t guid)
242 {
243 	uint64_t tmp;
244 	nvlist_t **child;
245 	uint_t c, children;
246 
247 	verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
248 	if (tmp == guid)
249 		return (B_TRUE);
250 
251 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
252 	    &child, &children) == 0) {
253 		for (c = 0; c < children; c++)
254 			if (find_guid(child[c], guid))
255 				return (B_TRUE);
256 	}
257 
258 	return (B_FALSE);
259 }
260 
261 typedef struct aux_cbdata {
262 	const char	*cb_type;
263 	uint64_t	cb_guid;
264 	zpool_handle_t	*cb_zhp;
265 } aux_cbdata_t;
266 
267 static int
268 find_aux(zpool_handle_t *zhp, void *data)
269 {
270 	aux_cbdata_t *cbp = data;
271 	nvlist_t **list;
272 	uint_t i, count;
273 	uint64_t guid;
274 	nvlist_t *nvroot;
275 
276 	verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
277 	    &nvroot) == 0);
278 
279 	if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
280 	    &list, &count) == 0) {
281 		for (i = 0; i < count; i++) {
282 			verify(nvlist_lookup_uint64(list[i],
283 			    ZPOOL_CONFIG_GUID, &guid) == 0);
284 			if (guid == cbp->cb_guid) {
285 				cbp->cb_zhp = zhp;
286 				return (1);
287 			}
288 		}
289 	}
290 
291 	zpool_close(zhp);
292 	return (0);
293 }
294 
295 /*
296  * Determines if the pool is in use.  If so, it returns true and the state of
297  * the pool as well as the name of the pool.  Name string is allocated and
298  * must be freed by the caller.
299  */
300 int
301 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
302     boolean_t *inuse)
303 {
304 	nvlist_t *config;
305 	char *name;
306 	boolean_t ret;
307 	uint64_t guid, vdev_guid;
308 	zpool_handle_t *zhp;
309 	nvlist_t *pool_config;
310 	uint64_t stateval, isspare;
311 	aux_cbdata_t cb = { 0 };
312 	boolean_t isactive;
313 
314 	*inuse = B_FALSE;
315 
316 	if (zpool_read_label(fd, &config, NULL) != 0) {
317 		(void) no_memory(hdl);
318 		return (-1);
319 	}
320 
321 	if (config == NULL)
322 		return (0);
323 
324 	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
325 	    &stateval) == 0);
326 	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
327 	    &vdev_guid) == 0);
328 
329 	if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
330 		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
331 		    &name) == 0);
332 		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
333 		    &guid) == 0);
334 	}
335 
336 	switch (stateval) {
337 	case POOL_STATE_EXPORTED:
338 		/*
339 		 * A pool with an exported state may in fact be imported
340 		 * read-only, so check the in-core state to see if it's
341 		 * active and imported read-only.  If it is, set
342 		 * its state to active.
343 		 */
344 		if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
345 		    (zhp = zpool_open_canfail(hdl, name)) != NULL) {
346 			if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
347 				stateval = POOL_STATE_ACTIVE;
348 
349 			/*
350 			 * All we needed the zpool handle for is the
351 			 * readonly prop check.
352 			 */
353 			zpool_close(zhp);
354 		}
355 
356 		ret = B_TRUE;
357 		break;
358 
359 	case POOL_STATE_ACTIVE:
360 		/*
361 		 * For an active pool, we have to determine if it's really part
362 		 * of a currently active pool (in which case the pool will exist
363 		 * and the guid will be the same), or whether it's part of an
364 		 * active pool that was disconnected without being explicitly
365 		 * exported.
366 		 */
367 		if (pool_active(hdl, name, guid, &isactive) != 0) {
368 			nvlist_free(config);
369 			return (-1);
370 		}
371 
372 		if (isactive) {
373 			/*
374 			 * Because the device may have been removed while
375 			 * offlined, we only report it as active if the vdev is
376 			 * still present in the config.  Otherwise, pretend like
377 			 * it's not in use.
378 			 */
379 			if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
380 			    (pool_config = zpool_get_config(zhp, NULL))
381 			    != NULL) {
382 				nvlist_t *nvroot;
383 
384 				verify(nvlist_lookup_nvlist(pool_config,
385 				    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
386 				ret = find_guid(nvroot, vdev_guid);
387 			} else {
388 				ret = B_FALSE;
389 			}
390 
391 			/*
392 			 * If this is an active spare within another pool, we
393 			 * treat it like an unused hot spare.  This allows the
394 			 * user to create a pool with a hot spare that currently
395 			 * in use within another pool.  Since we return B_TRUE,
396 			 * libdiskmgt will continue to prevent generic consumers
397 			 * from using the device.
398 			 */
399 			if (ret && nvlist_lookup_uint64(config,
400 			    ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
401 				stateval = POOL_STATE_SPARE;
402 
403 			if (zhp != NULL)
404 				zpool_close(zhp);
405 		} else {
406 			stateval = POOL_STATE_POTENTIALLY_ACTIVE;
407 			ret = B_TRUE;
408 		}
409 		break;
410 
411 	case POOL_STATE_SPARE:
412 		/*
413 		 * For a hot spare, it can be either definitively in use, or
414 		 * potentially active.  To determine if it's in use, we iterate
415 		 * over all pools in the system and search for one with a spare
416 		 * with a matching guid.
417 		 *
418 		 * Due to the shared nature of spares, we don't actually report
419 		 * the potentially active case as in use.  This means the user
420 		 * can freely create pools on the hot spares of exported pools,
421 		 * but to do otherwise makes the resulting code complicated, and
422 		 * we end up having to deal with this case anyway.
423 		 */
424 		cb.cb_zhp = NULL;
425 		cb.cb_guid = vdev_guid;
426 		cb.cb_type = ZPOOL_CONFIG_SPARES;
427 		if (zpool_iter(hdl, find_aux, &cb) == 1) {
428 			name = (char *)zpool_get_name(cb.cb_zhp);
429 			ret = B_TRUE;
430 		} else {
431 			ret = B_FALSE;
432 		}
433 		break;
434 
435 	case POOL_STATE_L2CACHE:
436 
437 		/*
438 		 * Check if any pool is currently using this l2cache device.
439 		 */
440 		cb.cb_zhp = NULL;
441 		cb.cb_guid = vdev_guid;
442 		cb.cb_type = ZPOOL_CONFIG_L2CACHE;
443 		if (zpool_iter(hdl, find_aux, &cb) == 1) {
444 			name = (char *)zpool_get_name(cb.cb_zhp);
445 			ret = B_TRUE;
446 		} else {
447 			ret = B_FALSE;
448 		}
449 		break;
450 
451 	default:
452 		ret = B_FALSE;
453 	}
454 
455 
456 	if (ret) {
457 		if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
458 			if (cb.cb_zhp)
459 				zpool_close(cb.cb_zhp);
460 			nvlist_free(config);
461 			return (-1);
462 		}
463 		*state = (pool_state_t)stateval;
464 	}
465 
466 	if (cb.cb_zhp)
467 		zpool_close(cb.cb_zhp);
468 
469 	nvlist_free(config);
470 	*inuse = ret;
471 	return (0);
472 }
473