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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
24  * Copyright 2016 Nexenta Systems, Inc.  All rights reserved.
25  * Copyright (c) 2013 Joyent, Inc.  All rights reserved.
26  */
27 
28 #include <sys/zfs_context.h>
29 #include <sys/spa_impl.h>
30 #include <sys/refcount.h>
31 #include <sys/vdev_disk.h>
32 #include <sys/vdev_impl.h>
33 #include <sys/abd.h>
34 #include <sys/fs/zfs.h>
35 #include <sys/zio.h>
36 #include <sys/sunldi.h>
37 #include <sys/efi_partition.h>
38 #include <sys/fm/fs/zfs.h>
39 
40 /*
41  * Virtual device vector for disks.
42  */
43 
44 extern ldi_ident_t zfs_li;
45 
46 static void vdev_disk_close(vdev_t *);
47 
48 typedef struct vdev_disk_ldi_cb {
49 	list_node_t		lcb_next;
50 	ldi_callback_id_t	lcb_id;
51 } vdev_disk_ldi_cb_t;
52 
53 /*
54  * Bypass the devid when opening a disk vdev.
55  * There have been issues where the devids of several devices were shuffled,
56  * causing pool open failures. Note, that this flag is intended to be used
57  * for pool recovery only.
58  *
59  * Note that if a pool is imported with the devids bypassed, all its vdevs will
60  * cease storing devid information permanently. In practice, the devid is rarely
61  * useful as vdev paths do not tend to change unless the hardware is
62  * reconfigured. That said, if the paths do change and a pool fails to open
63  * automatically at boot, a simple zpool import should re-scan the paths and fix
64  * the issue.
65  */
66 boolean_t vdev_disk_bypass_devid = B_FALSE;
67 
68 static void
vdev_disk_alloc(vdev_t * vd)69 vdev_disk_alloc(vdev_t *vd)
70 {
71 	vdev_disk_t *dvd;
72 
73 	dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
74 	/*
75 	 * Create the LDI event callback list.
76 	 */
77 	list_create(&dvd->vd_ldi_cbs, sizeof (vdev_disk_ldi_cb_t),
78 	    offsetof(vdev_disk_ldi_cb_t, lcb_next));
79 }
80 
81 static void
vdev_disk_free(vdev_t * vd)82 vdev_disk_free(vdev_t *vd)
83 {
84 	vdev_disk_t *dvd = vd->vdev_tsd;
85 	vdev_disk_ldi_cb_t *lcb;
86 
87 	if (dvd == NULL)
88 		return;
89 
90 	/*
91 	 * We have already closed the LDI handle. Clean up the LDI event
92 	 * callbacks and free vd->vdev_tsd.
93 	 */
94 	while ((lcb = list_head(&dvd->vd_ldi_cbs)) != NULL) {
95 		list_remove(&dvd->vd_ldi_cbs, lcb);
96 		(void) ldi_ev_remove_callbacks(lcb->lcb_id);
97 		kmem_free(lcb, sizeof (vdev_disk_ldi_cb_t));
98 	}
99 	list_destroy(&dvd->vd_ldi_cbs);
100 	kmem_free(dvd, sizeof (vdev_disk_t));
101 	vd->vdev_tsd = NULL;
102 }
103 
104 /* ARGSUSED */
105 static int
vdev_disk_off_notify(ldi_handle_t lh,ldi_ev_cookie_t ecookie,void * arg,void * ev_data)106 vdev_disk_off_notify(ldi_handle_t lh, ldi_ev_cookie_t ecookie, void *arg,
107     void *ev_data)
108 {
109 	vdev_t *vd = (vdev_t *)arg;
110 	vdev_disk_t *dvd = vd->vdev_tsd;
111 
112 	/*
113 	 * Ignore events other than offline.
114 	 */
115 	if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_OFFLINE) != 0)
116 		return (LDI_EV_SUCCESS);
117 
118 	/*
119 	 * All LDI handles must be closed for the state change to succeed, so
120 	 * call on vdev_disk_close() to do this.
121 	 *
122 	 * We inform vdev_disk_close that it is being called from offline
123 	 * notify context so it will defer cleanup of LDI event callbacks and
124 	 * freeing of vd->vdev_tsd to the offline finalize or a reopen.
125 	 */
126 	dvd->vd_ldi_offline = B_TRUE;
127 	vdev_disk_close(vd);
128 
129 	/*
130 	 * Now that the device is closed, request that the spa_async_thread
131 	 * mark the device as REMOVED and notify FMA of the removal.
132 	 */
133 	zfs_post_remove(vd->vdev_spa, vd);
134 	vd->vdev_remove_wanted = B_TRUE;
135 	spa_async_request(vd->vdev_spa, SPA_ASYNC_REMOVE);
136 
137 	return (LDI_EV_SUCCESS);
138 }
139 
140 /* ARGSUSED */
141 static void
vdev_disk_off_finalize(ldi_handle_t lh,ldi_ev_cookie_t ecookie,int ldi_result,void * arg,void * ev_data)142 vdev_disk_off_finalize(ldi_handle_t lh, ldi_ev_cookie_t ecookie,
143     int ldi_result, void *arg, void *ev_data)
144 {
145 	vdev_t *vd = (vdev_t *)arg;
146 
147 	/*
148 	 * Ignore events other than offline.
149 	 */
150 	if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_OFFLINE) != 0)
151 		return;
152 
153 	/*
154 	 * We have already closed the LDI handle in notify.
155 	 * Clean up the LDI event callbacks and free vd->vdev_tsd.
156 	 */
157 	vdev_disk_free(vd);
158 
159 	/*
160 	 * Request that the vdev be reopened if the offline state change was
161 	 * unsuccessful.
162 	 */
163 	if (ldi_result != LDI_EV_SUCCESS) {
164 		vd->vdev_probe_wanted = B_TRUE;
165 		spa_async_request(vd->vdev_spa, SPA_ASYNC_PROBE);
166 	}
167 }
168 
169 static ldi_ev_callback_t vdev_disk_off_callb = {
170 	.cb_vers = LDI_EV_CB_VERS,
171 	.cb_notify = vdev_disk_off_notify,
172 	.cb_finalize = vdev_disk_off_finalize
173 };
174 
175 /* ARGSUSED */
176 static void
vdev_disk_dgrd_finalize(ldi_handle_t lh,ldi_ev_cookie_t ecookie,int ldi_result,void * arg,void * ev_data)177 vdev_disk_dgrd_finalize(ldi_handle_t lh, ldi_ev_cookie_t ecookie,
178     int ldi_result, void *arg, void *ev_data)
179 {
180 	vdev_t *vd = (vdev_t *)arg;
181 
182 	/*
183 	 * Ignore events other than degrade.
184 	 */
185 	if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_DEGRADE) != 0)
186 		return;
187 
188 	/*
189 	 * Degrade events always succeed. Mark the vdev as degraded.
190 	 * This status is purely informative for the user.
191 	 */
192 	(void) vdev_degrade(vd->vdev_spa, vd->vdev_guid, 0);
193 }
194 
195 static ldi_ev_callback_t vdev_disk_dgrd_callb = {
196 	.cb_vers = LDI_EV_CB_VERS,
197 	.cb_notify = NULL,
198 	.cb_finalize = vdev_disk_dgrd_finalize
199 };
200 
201 static void
vdev_disk_hold(vdev_t * vd)202 vdev_disk_hold(vdev_t *vd)
203 {
204 	ddi_devid_t devid;
205 	char *minor;
206 
207 	ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
208 
209 	/*
210 	 * We must have a pathname, and it must be absolute.
211 	 */
212 	if (vd->vdev_path == NULL || vd->vdev_path[0] != '/')
213 		return;
214 
215 	/*
216 	 * Only prefetch path and devid info if the device has
217 	 * never been opened.
218 	 */
219 	if (vd->vdev_tsd != NULL)
220 		return;
221 
222 	if (vd->vdev_wholedisk == -1ULL) {
223 		size_t len = strlen(vd->vdev_path) + 3;
224 		char *buf = kmem_alloc(len, KM_SLEEP);
225 
226 		(void) snprintf(buf, len, "%ss0", vd->vdev_path);
227 
228 		(void) ldi_vp_from_name(buf, &vd->vdev_name_vp);
229 		kmem_free(buf, len);
230 	}
231 
232 	if (vd->vdev_name_vp == NULL)
233 		(void) ldi_vp_from_name(vd->vdev_path, &vd->vdev_name_vp);
234 
235 	if (vd->vdev_devid != NULL &&
236 	    ddi_devid_str_decode(vd->vdev_devid, &devid, &minor) == 0) {
237 		(void) ldi_vp_from_devid(devid, minor, &vd->vdev_devid_vp);
238 		ddi_devid_str_free(minor);
239 		ddi_devid_free(devid);
240 	}
241 }
242 
243 static void
vdev_disk_rele(vdev_t * vd)244 vdev_disk_rele(vdev_t *vd)
245 {
246 	ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
247 
248 	if (vd->vdev_name_vp) {
249 		VN_RELE_ASYNC(vd->vdev_name_vp,
250 		    dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
251 		vd->vdev_name_vp = NULL;
252 	}
253 	if (vd->vdev_devid_vp) {
254 		VN_RELE_ASYNC(vd->vdev_devid_vp,
255 		    dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
256 		vd->vdev_devid_vp = NULL;
257 	}
258 }
259 
260 /*
261  * We want to be loud in DEBUG kernels when DKIOCGMEDIAINFOEXT fails, or when
262  * even a fallback to DKIOCGMEDIAINFO fails.
263  */
264 #ifdef DEBUG
265 #define	VDEV_DEBUG(...)	cmn_err(CE_NOTE, __VA_ARGS__)
266 #else
267 #define	VDEV_DEBUG(...)	/* Nothing... */
268 #endif
269 
270 static int
vdev_disk_open(vdev_t * vd,uint64_t * psize,uint64_t * max_psize,uint64_t * ashift)271 vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
272     uint64_t *ashift)
273 {
274 	spa_t *spa = vd->vdev_spa;
275 	vdev_disk_t *dvd = vd->vdev_tsd;
276 	ldi_ev_cookie_t ecookie;
277 	vdev_disk_ldi_cb_t *lcb;
278 	union {
279 		struct dk_minfo_ext ude;
280 		struct dk_minfo ud;
281 	} dks;
282 	struct dk_minfo_ext *dkmext = &dks.ude;
283 	struct dk_minfo *dkm = &dks.ud;
284 	int error;
285 	dev_t dev;
286 	int otyp;
287 	boolean_t validate_devid = B_FALSE;
288 	ddi_devid_t devid;
289 	uint64_t capacity = 0, blksz = 0, pbsize;
290 
291 	/*
292 	 * We must have a pathname, and it must be absolute.
293 	 */
294 	if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
295 		vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
296 		return (SET_ERROR(EINVAL));
297 	}
298 
299 	/*
300 	 * Reopen the device if it's not currently open. Otherwise,
301 	 * just update the physical size of the device.
302 	 */
303 	if (dvd != NULL) {
304 		if (dvd->vd_ldi_offline && dvd->vd_lh == NULL) {
305 			/*
306 			 * If we are opening a device in its offline notify
307 			 * context, the LDI handle was just closed. Clean
308 			 * up the LDI event callbacks and free vd->vdev_tsd.
309 			 */
310 			vdev_disk_free(vd);
311 		} else {
312 			ASSERT(vd->vdev_reopening);
313 			goto skip_open;
314 		}
315 	}
316 
317 	/*
318 	 * Create vd->vdev_tsd.
319 	 */
320 	vdev_disk_alloc(vd);
321 	dvd = vd->vdev_tsd;
322 
323 	/*
324 	 * Allow bypassing the devid.
325 	 */
326 	if (vd->vdev_devid != NULL && vdev_disk_bypass_devid) {
327 		vdev_dbgmsg(vd, "vdev_disk_open, devid %s bypassed",
328 		    vd->vdev_devid);
329 		spa_strfree(vd->vdev_devid);
330 		vd->vdev_devid = NULL;
331 	}
332 
333 	/*
334 	 * When opening a disk device, we want to preserve the user's original
335 	 * intent.  We always want to open the device by the path the user gave
336 	 * us, even if it is one of multiple paths to the save device.  But we
337 	 * also want to be able to survive disks being removed/recabled.
338 	 * Therefore the sequence of opening devices is:
339 	 *
340 	 * 1. Try opening the device by path.  For legacy pools without the
341 	 *    'whole_disk' property, attempt to fix the path by appending 's0'.
342 	 *
343 	 * 2. If the devid of the device matches the stored value, return
344 	 *    success.
345 	 *
346 	 * 3. Otherwise, the device may have moved.  Try opening the device
347 	 *    by the devid instead.
348 	 */
349 	if (vd->vdev_devid != NULL) {
350 		if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
351 		    &dvd->vd_minor) != 0) {
352 			vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
353 			vdev_dbgmsg(vd, "vdev_disk_open: invalid "
354 			    "vdev_devid '%s'", vd->vdev_devid);
355 			return (SET_ERROR(EINVAL));
356 		}
357 	}
358 
359 	error = EINVAL;		/* presume failure */
360 
361 	if (vd->vdev_path != NULL) {
362 
363 		if (vd->vdev_wholedisk == -1ULL) {
364 			size_t len = strlen(vd->vdev_path) + 3;
365 			char *buf = kmem_alloc(len, KM_SLEEP);
366 
367 			(void) snprintf(buf, len, "%ss0", vd->vdev_path);
368 
369 			error = ldi_open_by_name(buf, spa_mode(spa), kcred,
370 			    &dvd->vd_lh, zfs_li);
371 			if (error == 0) {
372 				spa_strfree(vd->vdev_path);
373 				vd->vdev_path = buf;
374 				vd->vdev_wholedisk = 1ULL;
375 			} else {
376 				kmem_free(buf, len);
377 			}
378 		}
379 
380 		/*
381 		 * If we have not yet opened the device, try to open it by the
382 		 * specified path.
383 		 */
384 		if (error != 0) {
385 			error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
386 			    kcred, &dvd->vd_lh, zfs_li);
387 		}
388 
389 		/*
390 		 * Compare the devid to the stored value.
391 		 */
392 		if (error == 0 && vd->vdev_devid != NULL &&
393 		    ldi_get_devid(dvd->vd_lh, &devid) == 0) {
394 			if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
395 				/*
396 				 * A mismatch here is unexpected, log it.
397 				 */
398 				char *devid_str = ddi_devid_str_encode(devid,
399 				    dvd->vd_minor);
400 				vdev_dbgmsg(vd, "vdev_disk_open: devid "
401 				    "mismatch: %s != %s", vd->vdev_devid,
402 				    devid_str);
403 				cmn_err(CE_NOTE, "vdev_disk_open %s: devid "
404 				    "mismatch: %s != %s", vd->vdev_path,
405 				    vd->vdev_devid, devid_str);
406 				ddi_devid_str_free(devid_str);
407 
408 				error = SET_ERROR(EINVAL);
409 				(void) ldi_close(dvd->vd_lh, spa_mode(spa),
410 				    kcred);
411 				dvd->vd_lh = NULL;
412 			}
413 			ddi_devid_free(devid);
414 		}
415 
416 		/*
417 		 * If we succeeded in opening the device, but 'vdev_wholedisk'
418 		 * is not yet set, then this must be a slice.
419 		 */
420 		if (error == 0 && vd->vdev_wholedisk == -1ULL)
421 			vd->vdev_wholedisk = 0;
422 	}
423 
424 	/*
425 	 * If we were unable to open by path, or the devid check fails, open by
426 	 * devid instead.
427 	 */
428 	if (error != 0 && vd->vdev_devid != NULL) {
429 		error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
430 		    spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
431 		if (error != 0) {
432 			vdev_dbgmsg(vd, "Failed to open by devid (%s)",
433 			    vd->vdev_devid);
434 		}
435 	}
436 
437 	/*
438 	 * If all else fails, then try opening by physical path (if available)
439 	 * or the logical path (if we failed due to the devid check).  While not
440 	 * as reliable as the devid, this will give us something, and the higher
441 	 * level vdev validation will prevent us from opening the wrong device.
442 	 */
443 	if (error) {
444 		if (vd->vdev_devid != NULL)
445 			validate_devid = B_TRUE;
446 
447 		if (vd->vdev_physpath != NULL &&
448 		    (dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV)
449 			error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
450 			    kcred, &dvd->vd_lh, zfs_li);
451 
452 		/*
453 		 * Note that we don't support the legacy auto-wholedisk support
454 		 * as above.  This hasn't been used in a very long time and we
455 		 * don't need to propagate its oddities to this edge condition.
456 		 */
457 		if (error && vd->vdev_path != NULL)
458 			error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
459 			    kcred, &dvd->vd_lh, zfs_li);
460 	}
461 
462 	if (error) {
463 		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
464 		vdev_dbgmsg(vd, "vdev_disk_open: failed to open [error=%d]",
465 		    error);
466 		return (error);
467 	}
468 
469 	/*
470 	 * Now that the device has been successfully opened, update the devid
471 	 * if necessary.
472 	 */
473 	if (validate_devid && spa_writeable(spa) &&
474 	    ldi_get_devid(dvd->vd_lh, &devid) == 0) {
475 		if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
476 			char *vd_devid;
477 
478 			vd_devid = ddi_devid_str_encode(devid, dvd->vd_minor);
479 			vdev_dbgmsg(vd, "vdev_disk_open: update devid from "
480 			    "'%s' to '%s'", vd->vdev_devid, vd_devid);
481 			cmn_err(CE_NOTE, "vdev_disk_open %s: update devid "
482 			    "from '%s' to '%s'", vd->vdev_path != NULL ?
483 			    vd->vdev_path : "?", vd->vdev_devid, vd_devid);
484 			spa_strfree(vd->vdev_devid);
485 			vd->vdev_devid = spa_strdup(vd_devid);
486 			ddi_devid_str_free(vd_devid);
487 		}
488 		ddi_devid_free(devid);
489 	}
490 
491 	/*
492 	 * Once a device is opened, verify that the physical device path (if
493 	 * available) is up to date.
494 	 */
495 	if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
496 	    ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
497 		char *physpath, *minorname;
498 
499 		physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
500 		minorname = NULL;
501 		if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
502 		    ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
503 		    (vd->vdev_physpath == NULL ||
504 		    strcmp(vd->vdev_physpath, physpath) != 0)) {
505 			if (vd->vdev_physpath)
506 				spa_strfree(vd->vdev_physpath);
507 			(void) strlcat(physpath, ":", MAXPATHLEN);
508 			(void) strlcat(physpath, minorname, MAXPATHLEN);
509 			vd->vdev_physpath = spa_strdup(physpath);
510 		}
511 		if (minorname)
512 			kmem_free(minorname, strlen(minorname) + 1);
513 		kmem_free(physpath, MAXPATHLEN);
514 	}
515 
516 	/*
517 	 * Register callbacks for the LDI offline event.
518 	 */
519 	if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_OFFLINE, &ecookie) ==
520 	    LDI_EV_SUCCESS) {
521 		lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
522 		list_insert_tail(&dvd->vd_ldi_cbs, lcb);
523 		(void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
524 		    &vdev_disk_off_callb, (void *) vd, &lcb->lcb_id);
525 	}
526 
527 	/*
528 	 * Register callbacks for the LDI degrade event.
529 	 */
530 	if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_DEGRADE, &ecookie) ==
531 	    LDI_EV_SUCCESS) {
532 		lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
533 		list_insert_tail(&dvd->vd_ldi_cbs, lcb);
534 		(void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
535 		    &vdev_disk_dgrd_callb, (void *) vd, &lcb->lcb_id);
536 	}
537 skip_open:
538 	/*
539 	 * Determine the actual size of the device.
540 	 */
541 	if (ldi_get_size(dvd->vd_lh, psize) != 0) {
542 		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
543 		vdev_dbgmsg(vd, "vdev_disk_open: failed to get size");
544 		return (SET_ERROR(EINVAL));
545 	}
546 
547 	*max_psize = *psize;
548 
549 	/*
550 	 * Determine the device's minimum transfer size.
551 	 * If the ioctl isn't supported, assume DEV_BSIZE.
552 	 */
553 	if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT,
554 	    (intptr_t)dkmext, FKIOCTL, kcred, NULL)) == 0) {
555 		capacity = dkmext->dki_capacity - 1;
556 		blksz = dkmext->dki_lbsize;
557 		pbsize = dkmext->dki_pbsize;
558 	} else if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFO,
559 	    (intptr_t)dkm, FKIOCTL, kcred, NULL)) == 0) {
560 		VDEV_DEBUG(
561 		    "vdev_disk_open(\"%s\"): fallback to DKIOCGMEDIAINFO\n",
562 		    vd->vdev_path);
563 		capacity = dkm->dki_capacity - 1;
564 		blksz = dkm->dki_lbsize;
565 		pbsize = blksz;
566 	} else {
567 		VDEV_DEBUG("vdev_disk_open(\"%s\"): "
568 		    "both DKIOCGMEDIAINFO{,EXT} calls failed, %d\n",
569 		    vd->vdev_path, error);
570 		pbsize = DEV_BSIZE;
571 	}
572 
573 	*ashift = highbit64(MAX(pbsize, SPA_MINBLOCKSIZE)) - 1;
574 
575 	if (vd->vdev_wholedisk == 1) {
576 		int wce = 1;
577 
578 		if (error == 0) {
579 			/*
580 			 * If we have the capability to expand, we'd have
581 			 * found out via success from DKIOCGMEDIAINFO{,EXT}.
582 			 * Adjust max_psize upward accordingly since we know
583 			 * we own the whole disk now.
584 			 */
585 			*max_psize = capacity * blksz;
586 		}
587 
588 		/*
589 		 * Since we own the whole disk, try to enable disk write
590 		 * caching.  We ignore errors because it's OK if we can't do it.
591 		 */
592 		(void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
593 		    FKIOCTL, kcred, NULL);
594 	}
595 
596 	/*
597 	 * Clear the nowritecache bit, so that on a vdev_reopen() we will
598 	 * try again.
599 	 */
600 	vd->vdev_nowritecache = B_FALSE;
601 
602 	return (0);
603 }
604 
605 static void
vdev_disk_close(vdev_t * vd)606 vdev_disk_close(vdev_t *vd)
607 {
608 	vdev_disk_t *dvd = vd->vdev_tsd;
609 
610 	if (vd->vdev_reopening || dvd == NULL)
611 		return;
612 
613 	if (dvd->vd_minor != NULL) {
614 		ddi_devid_str_free(dvd->vd_minor);
615 		dvd->vd_minor = NULL;
616 	}
617 
618 	if (dvd->vd_devid != NULL) {
619 		ddi_devid_free(dvd->vd_devid);
620 		dvd->vd_devid = NULL;
621 	}
622 
623 	if (dvd->vd_lh != NULL) {
624 		(void) ldi_close(dvd->vd_lh, spa_mode(vd->vdev_spa), kcred);
625 		dvd->vd_lh = NULL;
626 	}
627 
628 	vd->vdev_delayed_close = B_FALSE;
629 	/*
630 	 * If we closed the LDI handle due to an offline notify from LDI,
631 	 * don't free vd->vdev_tsd or unregister the callbacks here;
632 	 * the offline finalize callback or a reopen will take care of it.
633 	 */
634 	if (dvd->vd_ldi_offline)
635 		return;
636 
637 	vdev_disk_free(vd);
638 }
639 
640 int
vdev_disk_physio(vdev_t * vd,caddr_t data,size_t size,uint64_t offset,int flags,boolean_t isdump)641 vdev_disk_physio(vdev_t *vd, caddr_t data,
642     size_t size, uint64_t offset, int flags, boolean_t isdump)
643 {
644 	vdev_disk_t *dvd = vd->vdev_tsd;
645 
646 	/*
647 	 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
648 	 * Nothing to be done here but return failure.
649 	 */
650 	if (dvd == NULL || (dvd->vd_ldi_offline && dvd->vd_lh == NULL))
651 		return (EIO);
652 
653 	ASSERT(vd->vdev_ops == &vdev_disk_ops);
654 
655 	/*
656 	 * If in the context of an active crash dump, use the ldi_dump(9F)
657 	 * call instead of ldi_strategy(9F) as usual.
658 	 */
659 	if (isdump) {
660 		ASSERT3P(dvd, !=, NULL);
661 		return (ldi_dump(dvd->vd_lh, data, lbtodb(offset),
662 		    lbtodb(size)));
663 	}
664 
665 	return (vdev_disk_ldi_physio(dvd->vd_lh, data, size, offset, flags));
666 }
667 
668 int
vdev_disk_ldi_physio(ldi_handle_t vd_lh,caddr_t data,size_t size,uint64_t offset,int flags)669 vdev_disk_ldi_physio(ldi_handle_t vd_lh, caddr_t data,
670     size_t size, uint64_t offset, int flags)
671 {
672 	buf_t *bp;
673 	int error = 0;
674 
675 	if (vd_lh == NULL)
676 		return (SET_ERROR(EINVAL));
677 
678 	ASSERT(flags & B_READ || flags & B_WRITE);
679 
680 	bp = getrbuf(KM_SLEEP);
681 	bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST;
682 	bp->b_bcount = size;
683 	bp->b_un.b_addr = (void *)data;
684 	bp->b_lblkno = lbtodb(offset);
685 	bp->b_bufsize = size;
686 
687 	error = ldi_strategy(vd_lh, bp);
688 	ASSERT(error == 0);
689 	if ((error = biowait(bp)) == 0 && bp->b_resid != 0)
690 		error = SET_ERROR(EIO);
691 	freerbuf(bp);
692 
693 	return (error);
694 }
695 
696 static void
vdev_disk_io_intr(buf_t * bp)697 vdev_disk_io_intr(buf_t *bp)
698 {
699 	vdev_buf_t *vb = (vdev_buf_t *)bp;
700 	zio_t *zio = vb->vb_io;
701 
702 	/*
703 	 * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO.
704 	 * Rather than teach the rest of the stack about other error
705 	 * possibilities (EFAULT, etc), we normalize the error value here.
706 	 */
707 	zio->io_error = (geterror(bp) != 0 ? SET_ERROR(EIO) : 0);
708 
709 	if (zio->io_error == 0 && bp->b_resid != 0)
710 		zio->io_error = SET_ERROR(EIO);
711 
712 	if (zio->io_type == ZIO_TYPE_READ) {
713 		abd_return_buf_copy(zio->io_abd, bp->b_un.b_addr, zio->io_size);
714 	} else {
715 		abd_return_buf(zio->io_abd, bp->b_un.b_addr, zio->io_size);
716 	}
717 
718 	kmem_free(vb, sizeof (vdev_buf_t));
719 
720 	zio_delay_interrupt(zio);
721 }
722 
723 static void
vdev_disk_ioctl_free(zio_t * zio)724 vdev_disk_ioctl_free(zio_t *zio)
725 {
726 	kmem_free(zio->io_vsd, sizeof (struct dk_callback));
727 }
728 
729 static const zio_vsd_ops_t vdev_disk_vsd_ops = {
730 	vdev_disk_ioctl_free,
731 	zio_vsd_default_cksum_report
732 };
733 
734 static void
vdev_disk_ioctl_done(void * zio_arg,int error)735 vdev_disk_ioctl_done(void *zio_arg, int error)
736 {
737 	zio_t *zio = zio_arg;
738 
739 	zio->io_error = error;
740 
741 	zio_interrupt(zio);
742 }
743 
744 static void
vdev_disk_io_start(zio_t * zio)745 vdev_disk_io_start(zio_t *zio)
746 {
747 	vdev_t *vd = zio->io_vd;
748 	vdev_disk_t *dvd = vd->vdev_tsd;
749 	vdev_buf_t *vb;
750 	struct dk_callback *dkc;
751 	buf_t *bp;
752 	int error;
753 
754 	/*
755 	 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
756 	 * Nothing to be done here but return failure.
757 	 */
758 	if (dvd == NULL || (dvd->vd_ldi_offline && dvd->vd_lh == NULL)) {
759 		zio->io_error = SET_ERROR(ENXIO);
760 		zio_interrupt(zio);
761 		return;
762 	}
763 
764 	if (zio->io_type == ZIO_TYPE_IOCTL) {
765 		/* XXPOLICY */
766 		if (!vdev_readable(vd)) {
767 			zio->io_error = SET_ERROR(ENXIO);
768 			zio_interrupt(zio);
769 			return;
770 		}
771 
772 		switch (zio->io_cmd) {
773 
774 		case DKIOCFLUSHWRITECACHE:
775 
776 			if (zfs_nocacheflush)
777 				break;
778 
779 			if (vd->vdev_nowritecache) {
780 				zio->io_error = SET_ERROR(ENOTSUP);
781 				break;
782 			}
783 
784 			zio->io_vsd = dkc = kmem_alloc(sizeof (*dkc), KM_SLEEP);
785 			zio->io_vsd_ops = &vdev_disk_vsd_ops;
786 
787 			dkc->dkc_callback = vdev_disk_ioctl_done;
788 			dkc->dkc_flag = FLUSH_VOLATILE;
789 			dkc->dkc_cookie = zio;
790 
791 			error = ldi_ioctl(dvd->vd_lh, zio->io_cmd,
792 			    (uintptr_t)dkc, FKIOCTL, kcred, NULL);
793 
794 			if (error == 0) {
795 				/*
796 				 * The ioctl will be done asychronously,
797 				 * and will call vdev_disk_ioctl_done()
798 				 * upon completion.
799 				 */
800 				return;
801 			}
802 
803 			zio->io_error = error;
804 
805 			break;
806 
807 		default:
808 			zio->io_error = SET_ERROR(ENOTSUP);
809 		}
810 
811 		zio_execute(zio);
812 		return;
813 	}
814 
815 	ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE);
816 	zio->io_target_timestamp = zio_handle_io_delay(zio);
817 
818 	vb = kmem_alloc(sizeof (vdev_buf_t), KM_SLEEP);
819 
820 	vb->vb_io = zio;
821 	bp = &vb->vb_buf;
822 
823 	bioinit(bp);
824 	bp->b_flags = B_BUSY | B_NOCACHE |
825 	    (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
826 	if (!(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))
827 		bp->b_flags |= B_FAILFAST;
828 	bp->b_bcount = zio->io_size;
829 
830 	if (zio->io_type == ZIO_TYPE_READ) {
831 		bp->b_un.b_addr =
832 		    abd_borrow_buf(zio->io_abd, zio->io_size);
833 	} else {
834 		bp->b_un.b_addr =
835 		    abd_borrow_buf_copy(zio->io_abd, zio->io_size);
836 	}
837 
838 	bp->b_lblkno = lbtodb(zio->io_offset);
839 	bp->b_bufsize = zio->io_size;
840 	bp->b_iodone = (int (*)())vdev_disk_io_intr;
841 
842 	/* ldi_strategy() will return non-zero only on programming errors */
843 	VERIFY(ldi_strategy(dvd->vd_lh, bp) == 0);
844 }
845 
846 static void
vdev_disk_io_done(zio_t * zio)847 vdev_disk_io_done(zio_t *zio)
848 {
849 	vdev_t *vd = zio->io_vd;
850 
851 	/*
852 	 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
853 	 * the device has been removed.  If this is the case, then we trigger an
854 	 * asynchronous removal of the device. Otherwise, probe the device and
855 	 * make sure it's still accessible.
856 	 */
857 	if (zio->io_error == EIO && !vd->vdev_remove_wanted) {
858 		vdev_disk_t *dvd = vd->vdev_tsd;
859 		int state = DKIO_NONE;
860 
861 		if (ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state,
862 		    FKIOCTL, kcred, NULL) == 0 && state != DKIO_INSERTED) {
863 			/*
864 			 * We post the resource as soon as possible, instead of
865 			 * when the async removal actually happens, because the
866 			 * DE is using this information to discard previous I/O
867 			 * errors.
868 			 */
869 			zfs_post_remove(zio->io_spa, vd);
870 			vd->vdev_remove_wanted = B_TRUE;
871 			spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
872 		} else if (!vd->vdev_delayed_close) {
873 			vd->vdev_delayed_close = B_TRUE;
874 		}
875 	}
876 }
877 
878 vdev_ops_t vdev_disk_ops = {
879 	vdev_disk_open,
880 	vdev_disk_close,
881 	vdev_default_asize,
882 	vdev_disk_io_start,
883 	vdev_disk_io_done,
884 	NULL,
885 	NULL,
886 	vdev_disk_hold,
887 	vdev_disk_rele,
888 	NULL,
889 	vdev_default_xlate,
890 	VDEV_TYPE_DISK,		/* name of this vdev type */
891 	B_TRUE			/* leaf vdev */
892 };
893 
894 /*
895  * Given the root disk device devid or pathname, read the label from
896  * the device, and construct a configuration nvlist.
897  */
898 int
vdev_disk_read_rootlabel(char * devpath,char * devid,nvlist_t ** config)899 vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config)
900 {
901 	ldi_handle_t vd_lh;
902 	vdev_label_t *label;
903 	uint64_t s, size;
904 	int l;
905 	ddi_devid_t tmpdevid;
906 	int error = -1;
907 	char *minor_name;
908 
909 	/*
910 	 * Read the device label and build the nvlist.
911 	 */
912 	if (devid != NULL && ddi_devid_str_decode(devid, &tmpdevid,
913 	    &minor_name) == 0) {
914 		error = ldi_open_by_devid(tmpdevid, minor_name,
915 		    FREAD, kcred, &vd_lh, zfs_li);
916 		ddi_devid_free(tmpdevid);
917 		ddi_devid_str_free(minor_name);
918 	}
919 
920 	if (error && (error = ldi_open_by_name(devpath, FREAD, kcred, &vd_lh,
921 	    zfs_li)))
922 		return (error);
923 
924 	if (ldi_get_size(vd_lh, &s)) {
925 		(void) ldi_close(vd_lh, FREAD, kcred);
926 		return (SET_ERROR(EIO));
927 	}
928 
929 	size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t);
930 	label = kmem_alloc(sizeof (vdev_label_t), KM_SLEEP);
931 
932 	*config = NULL;
933 	for (l = 0; l < VDEV_LABELS; l++) {
934 		uint64_t offset, state, txg = 0;
935 
936 		/* read vdev label */
937 		offset = vdev_label_offset(size, l, 0);
938 		if (vdev_disk_ldi_physio(vd_lh, (caddr_t)label,
939 		    VDEV_SKIP_SIZE + VDEV_PHYS_SIZE, offset, B_READ) != 0)
940 			continue;
941 
942 		if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
943 		    sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0) {
944 			*config = NULL;
945 			continue;
946 		}
947 
948 		if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
949 		    &state) != 0 || state >= POOL_STATE_DESTROYED) {
950 			nvlist_free(*config);
951 			*config = NULL;
952 			continue;
953 		}
954 
955 		if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
956 		    &txg) != 0 || txg == 0) {
957 			nvlist_free(*config);
958 			*config = NULL;
959 			continue;
960 		}
961 
962 		break;
963 	}
964 
965 	kmem_free(label, sizeof (vdev_label_t));
966 	(void) ldi_close(vd_lh, FREAD, kcred);
967 	if (*config == NULL)
968 		error = SET_ERROR(EIDRM);
969 
970 	return (error);
971 }
972