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 2019 Joyent, Inc.
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/vdev_trim.h>
34#include <sys/abd.h>
35#include <sys/fs/zfs.h>
36#include <sys/zio.h>
37#include <sys/sunldi.h>
38#include <sys/efi_partition.h>
39#include <sys/fm/fs/zfs.h>
40
41/*
42 * Tunable to enable TRIM, which is temporarily disabled by default.
43 */
44uint_t zfs_no_trim = 1;
45
46/*
47 * Tunable parameter for debugging or performance analysis. Setting this
48 * will cause pool corruption on power loss if a volatile out-of-order
49 * write cache is enabled.
50 */
51boolean_t zfs_nocacheflush = B_FALSE;
52
53/*
54 * Virtual device vector for disks.
55 */
56
57extern ldi_ident_t zfs_li;
58
59static void vdev_disk_close(vdev_t *);
60
61typedef struct vdev_disk_ldi_cb {
62	list_node_t		lcb_next;
63	ldi_callback_id_t	lcb_id;
64} vdev_disk_ldi_cb_t;
65
66/*
67 * Bypass the devid when opening a disk vdev.
68 * There have been issues where the devids of several devices were shuffled,
69 * causing pool open failures. Note, that this flag is intended to be used
70 * for pool recovery only.
71 *
72 * Note that if a pool is imported with the devids bypassed, all its vdevs will
73 * cease storing devid information permanently. In practice, the devid is rarely
74 * useful as vdev paths do not tend to change unless the hardware is
75 * reconfigured. That said, if the paths do change and a pool fails to open
76 * automatically at boot, a simple zpool import should re-scan the paths and fix
77 * the issue.
78 */
79boolean_t vdev_disk_bypass_devid = B_FALSE;
80
81static void
82vdev_disk_alloc(vdev_t *vd)
83{
84	vdev_disk_t *dvd;
85
86	dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
87	/*
88	 * Create the LDI event callback list.
89	 */
90	list_create(&dvd->vd_ldi_cbs, sizeof (vdev_disk_ldi_cb_t),
91	    offsetof(vdev_disk_ldi_cb_t, lcb_next));
92}
93
94static void
95vdev_disk_free(vdev_t *vd)
96{
97	vdev_disk_t *dvd = vd->vdev_tsd;
98	vdev_disk_ldi_cb_t *lcb;
99
100	if (dvd == NULL)
101		return;
102
103	/*
104	 * We have already closed the LDI handle. Clean up the LDI event
105	 * callbacks and free vd->vdev_tsd.
106	 */
107	while ((lcb = list_head(&dvd->vd_ldi_cbs)) != NULL) {
108		list_remove(&dvd->vd_ldi_cbs, lcb);
109		(void) ldi_ev_remove_callbacks(lcb->lcb_id);
110		kmem_free(lcb, sizeof (vdev_disk_ldi_cb_t));
111	}
112	list_destroy(&dvd->vd_ldi_cbs);
113	kmem_free(dvd, sizeof (vdev_disk_t));
114	vd->vdev_tsd = NULL;
115}
116
117/* ARGSUSED */
118static int
119vdev_disk_off_notify(ldi_handle_t lh, ldi_ev_cookie_t ecookie, void *arg,
120    void *ev_data)
121{
122	vdev_t *vd = (vdev_t *)arg;
123	vdev_disk_t *dvd = vd->vdev_tsd;
124
125	/*
126	 * Ignore events other than offline.
127	 */
128	if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_OFFLINE) != 0)
129		return (LDI_EV_SUCCESS);
130
131	/*
132	 * All LDI handles must be closed for the state change to succeed, so
133	 * call on vdev_disk_close() to do this.
134	 *
135	 * We inform vdev_disk_close that it is being called from offline
136	 * notify context so it will defer cleanup of LDI event callbacks and
137	 * freeing of vd->vdev_tsd to the offline finalize or a reopen.
138	 */
139	dvd->vd_ldi_offline = B_TRUE;
140	vdev_disk_close(vd);
141
142	/*
143	 * Now that the device is closed, request that the spa_async_thread
144	 * mark the device as REMOVED and notify FMA of the removal.
145	 */
146	zfs_post_remove(vd->vdev_spa, vd);
147	vd->vdev_remove_wanted = B_TRUE;
148	spa_async_request(vd->vdev_spa, SPA_ASYNC_REMOVE);
149
150	return (LDI_EV_SUCCESS);
151}
152
153/* ARGSUSED */
154static void
155vdev_disk_off_finalize(ldi_handle_t lh, ldi_ev_cookie_t ecookie,
156    int ldi_result, void *arg, void *ev_data)
157{
158	vdev_t *vd = (vdev_t *)arg;
159
160	/*
161	 * Ignore events other than offline.
162	 */
163	if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_OFFLINE) != 0)
164		return;
165
166	/*
167	 * We have already closed the LDI handle in notify.
168	 * Clean up the LDI event callbacks and free vd->vdev_tsd.
169	 */
170	vdev_disk_free(vd);
171
172	/*
173	 * Request that the vdev be reopened if the offline state change was
174	 * unsuccessful.
175	 */
176	if (ldi_result != LDI_EV_SUCCESS) {
177		vd->vdev_probe_wanted = B_TRUE;
178		spa_async_request(vd->vdev_spa, SPA_ASYNC_PROBE);
179	}
180}
181
182static ldi_ev_callback_t vdev_disk_off_callb = {
183	.cb_vers = LDI_EV_CB_VERS,
184	.cb_notify = vdev_disk_off_notify,
185	.cb_finalize = vdev_disk_off_finalize
186};
187
188/* ARGSUSED */
189static void
190vdev_disk_dgrd_finalize(ldi_handle_t lh, ldi_ev_cookie_t ecookie,
191    int ldi_result, void *arg, void *ev_data)
192{
193	vdev_t *vd = (vdev_t *)arg;
194
195	/*
196	 * Ignore events other than degrade.
197	 */
198	if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_DEGRADE) != 0)
199		return;
200
201	/*
202	 * Degrade events always succeed. Mark the vdev as degraded.
203	 * This status is purely informative for the user.
204	 */
205	(void) vdev_degrade(vd->vdev_spa, vd->vdev_guid, 0);
206}
207
208static ldi_ev_callback_t vdev_disk_dgrd_callb = {
209	.cb_vers = LDI_EV_CB_VERS,
210	.cb_notify = NULL,
211	.cb_finalize = vdev_disk_dgrd_finalize
212};
213
214static void
215vdev_disk_hold(vdev_t *vd)
216{
217	ddi_devid_t devid;
218	char *minor;
219
220	ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
221
222	/*
223	 * We must have a pathname, and it must be absolute.
224	 */
225	if (vd->vdev_path == NULL || vd->vdev_path[0] != '/')
226		return;
227
228	/*
229	 * Only prefetch path and devid info if the device has
230	 * never been opened.
231	 */
232	if (vd->vdev_tsd != NULL)
233		return;
234
235	if (vd->vdev_wholedisk == -1ULL) {
236		size_t len = strlen(vd->vdev_path) + 3;
237		char *buf = kmem_alloc(len, KM_SLEEP);
238
239		(void) snprintf(buf, len, "%ss0", vd->vdev_path);
240
241		(void) ldi_vp_from_name(buf, &vd->vdev_name_vp);
242		kmem_free(buf, len);
243	}
244
245	if (vd->vdev_name_vp == NULL)
246		(void) ldi_vp_from_name(vd->vdev_path, &vd->vdev_name_vp);
247
248	if (vd->vdev_devid != NULL &&
249	    ddi_devid_str_decode(vd->vdev_devid, &devid, &minor) == 0) {
250		(void) ldi_vp_from_devid(devid, minor, &vd->vdev_devid_vp);
251		ddi_devid_str_free(minor);
252		ddi_devid_free(devid);
253	}
254}
255
256static void
257vdev_disk_rele(vdev_t *vd)
258{
259	ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
260
261	if (vd->vdev_name_vp) {
262		VN_RELE_ASYNC(vd->vdev_name_vp,
263		    dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
264		vd->vdev_name_vp = NULL;
265	}
266	if (vd->vdev_devid_vp) {
267		VN_RELE_ASYNC(vd->vdev_devid_vp,
268		    dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
269		vd->vdev_devid_vp = NULL;
270	}
271}
272
273/*
274 * We want to be loud in DEBUG kernels when DKIOCGMEDIAINFOEXT fails, or when
275 * even a fallback to DKIOCGMEDIAINFO fails.
276 */
277#ifdef DEBUG
278#define	VDEV_DEBUG(...)	cmn_err(CE_NOTE, __VA_ARGS__)
279#else
280#define	VDEV_DEBUG(...)	/* Nothing... */
281#endif
282
283static int
284vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
285    uint64_t *ashift)
286{
287	spa_t *spa = vd->vdev_spa;
288	vdev_disk_t *dvd = vd->vdev_tsd;
289	ldi_ev_cookie_t ecookie;
290	vdev_disk_ldi_cb_t *lcb;
291	union {
292		struct dk_minfo_ext ude;
293		struct dk_minfo ud;
294	} dks;
295	struct dk_minfo_ext *dkmext = &dks.ude;
296	struct dk_minfo *dkm = &dks.ud;
297	int error, can_free;
298	dev_t dev;
299	int otyp;
300	boolean_t validate_devid = B_FALSE;
301	uint64_t capacity = 0, blksz = 0, pbsize;
302
303	/*
304	 * We must have a pathname, and it must be absolute.
305	 */
306	if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
307		vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
308		return (SET_ERROR(EINVAL));
309	}
310
311	/*
312	 * Reopen the device if it's not currently open. Otherwise,
313	 * just update the physical size of the device.
314	 */
315	if (dvd != NULL) {
316		if (dvd->vd_ldi_offline && dvd->vd_lh == NULL) {
317			/*
318			 * If we are opening a device in its offline notify
319			 * context, the LDI handle was just closed. Clean
320			 * up the LDI event callbacks and free vd->vdev_tsd.
321			 */
322			vdev_disk_free(vd);
323		} else {
324			ASSERT(vd->vdev_reopening);
325			goto skip_open;
326		}
327	}
328
329	/*
330	 * Create vd->vdev_tsd.
331	 */
332	vdev_disk_alloc(vd);
333	dvd = vd->vdev_tsd;
334
335	/*
336	 * Allow bypassing the devid.
337	 */
338	if (vd->vdev_devid != NULL && vdev_disk_bypass_devid) {
339		vdev_dbgmsg(vd, "vdev_disk_open, devid %s bypassed",
340		    vd->vdev_devid);
341		spa_strfree(vd->vdev_devid);
342		vd->vdev_devid = NULL;
343	}
344
345	/*
346	 * When opening a disk device, we want to preserve the user's original
347	 * intent.  We always want to open the device by the path the user gave
348	 * us, even if it is one of multiple paths to the same device.  But we
349	 * also want to be able to survive disks being removed/recabled.
350	 * Therefore the sequence of opening devices is:
351	 *
352	 * 1. Try opening the device by path.  For legacy pools without the
353	 *    'whole_disk' property, attempt to fix the path by appending 's0'.
354	 *
355	 * 2. If the devid of the device matches the stored value, return
356	 *    success.
357	 *
358	 * 3. Otherwise, the device may have moved.  Try opening the device
359	 *    by the devid instead.
360	 */
361	if (vd->vdev_devid != NULL) {
362		if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
363		    &dvd->vd_minor) != 0) {
364			vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
365			vdev_dbgmsg(vd, "vdev_disk_open: invalid "
366			    "vdev_devid '%s'", vd->vdev_devid);
367			return (SET_ERROR(EINVAL));
368		}
369	}
370
371	error = EINVAL;		/* presume failure */
372
373	if (vd->vdev_path != NULL) {
374
375		if (vd->vdev_wholedisk == -1ULL) {
376			size_t len = strlen(vd->vdev_path) + 3;
377			char *buf = kmem_alloc(len, KM_SLEEP);
378
379			(void) snprintf(buf, len, "%ss0", vd->vdev_path);
380
381			error = ldi_open_by_name(buf, spa_mode(spa), kcred,
382			    &dvd->vd_lh, zfs_li);
383			if (error == 0) {
384				spa_strfree(vd->vdev_path);
385				vd->vdev_path = buf;
386				vd->vdev_wholedisk = 1ULL;
387			} else {
388				kmem_free(buf, len);
389			}
390		}
391
392		/*
393		 * If we have not yet opened the device, try to open it by the
394		 * specified path.
395		 */
396		if (error != 0) {
397			error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
398			    kcred, &dvd->vd_lh, zfs_li);
399		}
400
401		/*
402		 * Compare the devid to the stored value.
403		 */
404		if (error == 0 && vd->vdev_devid != NULL) {
405			ddi_devid_t devid = NULL;
406
407			if (ldi_get_devid(dvd->vd_lh, &devid) != 0) {
408				/*
409				 * We expected a devid on this device but it no
410				 * longer appears to have one.  The validation
411				 * step may need to remove it from the
412				 * configuration.
413				 */
414				validate_devid = B_TRUE;
415
416			} else if (ddi_devid_compare(devid, dvd->vd_devid) !=
417			    0) {
418				/*
419				 * A mismatch here is unexpected, log it.
420				 */
421				char *devid_str = ddi_devid_str_encode(devid,
422				    dvd->vd_minor);
423				vdev_dbgmsg(vd, "vdev_disk_open: devid "
424				    "mismatch: %s != %s", vd->vdev_devid,
425				    devid_str);
426				cmn_err(CE_NOTE, "vdev_disk_open %s: devid "
427				    "mismatch: %s != %s", vd->vdev_path,
428				    vd->vdev_devid, devid_str);
429				ddi_devid_str_free(devid_str);
430
431				error = SET_ERROR(EINVAL);
432				(void) ldi_close(dvd->vd_lh, spa_mode(spa),
433				    kcred);
434				dvd->vd_lh = NULL;
435			}
436
437			if (devid != NULL) {
438				ddi_devid_free(devid);
439			}
440		}
441
442		/*
443		 * If we succeeded in opening the device, but 'vdev_wholedisk'
444		 * is not yet set, then this must be a slice.
445		 */
446		if (error == 0 && vd->vdev_wholedisk == -1ULL)
447			vd->vdev_wholedisk = 0;
448	}
449
450	/*
451	 * If we were unable to open by path, or the devid check fails, open by
452	 * devid instead.
453	 */
454	if (error != 0 && vd->vdev_devid != NULL) {
455		error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
456		    spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
457		if (error != 0) {
458			vdev_dbgmsg(vd, "Failed to open by devid (%s)",
459			    vd->vdev_devid);
460		}
461	}
462
463	/*
464	 * If all else fails, then try opening by physical path (if available)
465	 * or the logical path (if we failed due to the devid check).  While not
466	 * as reliable as the devid, this will give us something, and the higher
467	 * level vdev validation will prevent us from opening the wrong device.
468	 */
469	if (error != 0) {
470		validate_devid = B_TRUE;
471
472		if (vd->vdev_physpath != NULL &&
473		    (dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV) {
474			error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
475			    kcred, &dvd->vd_lh, zfs_li);
476		}
477
478		/*
479		 * Note that we don't support the legacy auto-wholedisk support
480		 * as above.  This hasn't been used in a very long time and we
481		 * don't need to propagate its oddities to this edge condition.
482		 */
483		if (error != 0 && vd->vdev_path != NULL) {
484			error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
485			    kcred, &dvd->vd_lh, zfs_li);
486		}
487	}
488
489	if (error != 0) {
490		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
491		vdev_dbgmsg(vd, "vdev_disk_open: failed to open [error=%d]",
492		    error);
493		return (error);
494	}
495
496	/*
497	 * Now that the device has been successfully opened, update the devid
498	 * if necessary.
499	 */
500	if (validate_devid) {
501		ddi_devid_t devid = NULL;
502		char *minorname = NULL;
503		char *vd_devid = NULL;
504		boolean_t remove = B_FALSE, update = B_FALSE;
505
506		/*
507		 * Get the current devid and minor name for the device we
508		 * opened.
509		 */
510		if (ldi_get_devid(dvd->vd_lh, &devid) != 0 ||
511		    ldi_get_minor_name(dvd->vd_lh, &minorname) != 0) {
512			/*
513			 * If we are unable to get the devid or the minor name
514			 * for the device, we need to remove them from the
515			 * configuration to prevent potential inconsistencies.
516			 */
517			if (dvd->vd_minor != NULL || dvd->vd_devid != NULL ||
518			    vd->vdev_devid != NULL) {
519				/*
520				 * We only need to remove the devid if one
521				 * exists.
522				 */
523				remove = B_TRUE;
524			}
525
526		} else if (dvd->vd_devid == NULL || dvd->vd_minor == NULL) {
527			/*
528			 * There was previously no devid at all so we need to
529			 * add one.
530			 */
531			update = B_TRUE;
532
533		} else if (ddi_devid_compare(devid, dvd->vd_devid) != 0 ||
534		    strcmp(minorname, dvd->vd_minor) != 0) {
535			/*
536			 * The devid or minor name on file does not match the
537			 * one from the opened device.
538			 */
539			update = B_TRUE;
540		}
541
542		if (update) {
543			/*
544			 * Render the new devid and minor name as a string for
545			 * logging and to store in the vdev configuration.
546			 */
547			vd_devid = ddi_devid_str_encode(devid, minorname);
548		}
549
550		if (update || remove) {
551			vdev_dbgmsg(vd, "vdev_disk_open: update devid from "
552			    "'%s' to '%s'",
553			    vd->vdev_devid != NULL ? vd->vdev_devid : "<none>",
554			    vd_devid != NULL ? vd_devid : "<none>");
555			cmn_err(CE_NOTE, "vdev_disk_open %s: update devid "
556			    "from '%s' to '%s'",
557			    vd->vdev_path != NULL ? vd->vdev_path : "?",
558			    vd->vdev_devid != NULL ? vd->vdev_devid : "<none>",
559			    vd_devid != NULL ? vd_devid : "<none>");
560
561			/*
562			 * Remove and free any existing values.
563			 */
564			if (dvd->vd_minor != NULL) {
565				ddi_devid_str_free(dvd->vd_minor);
566				dvd->vd_minor = NULL;
567			}
568			if (dvd->vd_devid != NULL) {
569				ddi_devid_free(dvd->vd_devid);
570				dvd->vd_devid = NULL;
571			}
572			if (vd->vdev_devid != NULL) {
573				spa_strfree(vd->vdev_devid);
574				vd->vdev_devid = NULL;
575			}
576		}
577
578		if (update) {
579			/*
580			 * Install the new values.
581			 */
582			vd->vdev_devid = vd_devid;
583			dvd->vd_minor = minorname;
584			dvd->vd_devid = devid;
585
586		} else {
587			if (devid != NULL) {
588				ddi_devid_free(devid);
589			}
590			if (minorname != NULL) {
591				kmem_free(minorname, strlen(minorname) + 1);
592			}
593		}
594	}
595
596	/*
597	 * Once a device is opened, verify that the physical device path (if
598	 * available) is up to date.
599	 */
600	if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
601	    ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
602		char *physpath, *minorname;
603
604		physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
605		minorname = NULL;
606		if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
607		    ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
608		    (vd->vdev_physpath == NULL ||
609		    strcmp(vd->vdev_physpath, physpath) != 0)) {
610			if (vd->vdev_physpath)
611				spa_strfree(vd->vdev_physpath);
612			(void) strlcat(physpath, ":", MAXPATHLEN);
613			(void) strlcat(physpath, minorname, MAXPATHLEN);
614			vd->vdev_physpath = spa_strdup(physpath);
615		}
616		if (minorname)
617			kmem_free(minorname, strlen(minorname) + 1);
618		kmem_free(physpath, MAXPATHLEN);
619	}
620
621	/*
622	 * Register callbacks for the LDI offline event.
623	 */
624	if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_OFFLINE, &ecookie) ==
625	    LDI_EV_SUCCESS) {
626		lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
627		list_insert_tail(&dvd->vd_ldi_cbs, lcb);
628		(void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
629		    &vdev_disk_off_callb, (void *) vd, &lcb->lcb_id);
630	}
631
632	/*
633	 * Register callbacks for the LDI degrade event.
634	 */
635	if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_DEGRADE, &ecookie) ==
636	    LDI_EV_SUCCESS) {
637		lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
638		list_insert_tail(&dvd->vd_ldi_cbs, lcb);
639		(void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
640		    &vdev_disk_dgrd_callb, (void *) vd, &lcb->lcb_id);
641	}
642
643skip_open:
644	/*
645	 * Determine the actual size of the device.
646	 */
647	if (ldi_get_size(dvd->vd_lh, psize) != 0) {
648		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
649		vdev_dbgmsg(vd, "vdev_disk_open: failed to get size");
650		return (SET_ERROR(EINVAL));
651	}
652
653	*max_psize = *psize;
654
655	/*
656	 * Determine the device's minimum transfer size.
657	 * If the ioctl isn't supported, assume DEV_BSIZE.
658	 */
659	if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT,
660	    (intptr_t)dkmext, FKIOCTL, kcred, NULL)) == 0) {
661		capacity = dkmext->dki_capacity - 1;
662		blksz = dkmext->dki_lbsize;
663		pbsize = dkmext->dki_pbsize;
664	} else if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFO,
665	    (intptr_t)dkm, FKIOCTL, kcred, NULL)) == 0) {
666		VDEV_DEBUG(
667		    "vdev_disk_open(\"%s\"): fallback to DKIOCGMEDIAINFO\n",
668		    vd->vdev_path);
669		capacity = dkm->dki_capacity - 1;
670		blksz = dkm->dki_lbsize;
671		pbsize = blksz;
672	} else {
673		VDEV_DEBUG("vdev_disk_open(\"%s\"): "
674		    "both DKIOCGMEDIAINFO{,EXT} calls failed, %d\n",
675		    vd->vdev_path, error);
676		pbsize = DEV_BSIZE;
677	}
678
679	*ashift = highbit64(MAX(pbsize, SPA_MINBLOCKSIZE)) - 1;
680
681	if (vd->vdev_wholedisk == 1) {
682		int wce = 1;
683
684		if (error == 0) {
685			/*
686			 * If we have the capability to expand, we'd have
687			 * found out via success from DKIOCGMEDIAINFO{,EXT}.
688			 * Adjust max_psize upward accordingly since we know
689			 * we own the whole disk now.
690			 */
691			*max_psize = capacity * blksz;
692		}
693
694		/*
695		 * Since we own the whole disk, try to enable disk write
696		 * caching.  We ignore errors because it's OK if we can't do it.
697		 */
698		(void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
699		    FKIOCTL, kcred, NULL);
700	}
701
702	/*
703	 * Clear the nowritecache bit, so that on a vdev_reopen() we will
704	 * try again.
705	 */
706	vd->vdev_nowritecache = B_FALSE;
707
708	if (ldi_ioctl(dvd->vd_lh, DKIOC_CANFREE, (intptr_t)&can_free, FKIOCTL,
709	    kcred, NULL) == 0 && can_free == 1) {
710		vd->vdev_has_trim = B_TRUE;
711	} else {
712		vd->vdev_has_trim = B_FALSE;
713	}
714
715	if (zfs_no_trim == 1)
716		vd->vdev_has_trim = B_FALSE;
717
718	/* Currently only supported for ZoL. */
719	vd->vdev_has_securetrim = B_FALSE;
720
721	/* Inform the ZIO pipeline that we are non-rotational */
722	vd->vdev_nonrot = B_FALSE;
723	if (ldi_prop_exists(dvd->vd_lh, DDI_PROP_DONTPASS | DDI_PROP_NOTPROM,
724	    "device-solid-state")) {
725		if (ldi_prop_get_int(dvd->vd_lh,
726		    LDI_DEV_T_ANY | DDI_PROP_DONTPASS | DDI_PROP_NOTPROM,
727		    "device-solid-state", B_FALSE) != 0)
728			vd->vdev_nonrot = B_TRUE;
729	}
730
731	return (0);
732}
733
734static void
735vdev_disk_close(vdev_t *vd)
736{
737	vdev_disk_t *dvd = vd->vdev_tsd;
738
739	if (vd->vdev_reopening || dvd == NULL)
740		return;
741
742	if (dvd->vd_minor != NULL) {
743		ddi_devid_str_free(dvd->vd_minor);
744		dvd->vd_minor = NULL;
745	}
746
747	if (dvd->vd_devid != NULL) {
748		ddi_devid_free(dvd->vd_devid);
749		dvd->vd_devid = NULL;
750	}
751
752	if (dvd->vd_lh != NULL) {
753		(void) ldi_close(dvd->vd_lh, spa_mode(vd->vdev_spa), kcred);
754		dvd->vd_lh = NULL;
755	}
756
757	vd->vdev_delayed_close = B_FALSE;
758	/*
759	 * If we closed the LDI handle due to an offline notify from LDI,
760	 * don't free vd->vdev_tsd or unregister the callbacks here;
761	 * the offline finalize callback or a reopen will take care of it.
762	 */
763	if (dvd->vd_ldi_offline)
764		return;
765
766	vdev_disk_free(vd);
767}
768
769int
770vdev_disk_physio(vdev_t *vd, caddr_t data,
771    size_t size, uint64_t offset, int flags, boolean_t isdump)
772{
773	vdev_disk_t *dvd = vd->vdev_tsd;
774
775	/*
776	 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
777	 * Nothing to be done here but return failure.
778	 */
779	if (dvd == NULL || (dvd->vd_ldi_offline && dvd->vd_lh == NULL))
780		return (EIO);
781
782	ASSERT(vd->vdev_ops == &vdev_disk_ops);
783
784	/*
785	 * If in the context of an active crash dump, use the ldi_dump(9F)
786	 * call instead of ldi_strategy(9F) as usual.
787	 */
788	if (isdump) {
789		ASSERT3P(dvd, !=, NULL);
790		return (ldi_dump(dvd->vd_lh, data, lbtodb(offset),
791		    lbtodb(size)));
792	}
793
794	return (vdev_disk_ldi_physio(dvd->vd_lh, data, size, offset, flags));
795}
796
797int
798vdev_disk_ldi_physio(ldi_handle_t vd_lh, caddr_t data,
799    size_t size, uint64_t offset, int flags)
800{
801	buf_t *bp;
802	int error = 0;
803
804	if (vd_lh == NULL)
805		return (SET_ERROR(EINVAL));
806
807	ASSERT(flags & B_READ || flags & B_WRITE);
808
809	bp = getrbuf(KM_SLEEP);
810	bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST;
811	bp->b_bcount = size;
812	bp->b_un.b_addr = (void *)data;
813	bp->b_lblkno = lbtodb(offset);
814	bp->b_bufsize = size;
815
816	error = ldi_strategy(vd_lh, bp);
817	ASSERT(error == 0);
818	if ((error = biowait(bp)) == 0 && bp->b_resid != 0)
819		error = SET_ERROR(EIO);
820	freerbuf(bp);
821
822	return (error);
823}
824
825static int
826vdev_disk_io_intr(buf_t *bp)
827{
828	vdev_buf_t *vb = (vdev_buf_t *)bp;
829	zio_t *zio = vb->vb_io;
830
831	/*
832	 * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO.
833	 * Rather than teach the rest of the stack about other error
834	 * possibilities (EFAULT, etc), we normalize the error value here.
835	 */
836	zio->io_error = (geterror(bp) != 0 ? EIO : 0);
837
838	if (zio->io_error == 0 && bp->b_resid != 0)
839		zio->io_error = SET_ERROR(EIO);
840
841	if (zio->io_type == ZIO_TYPE_READ) {
842		abd_return_buf_copy(zio->io_abd, bp->b_un.b_addr, zio->io_size);
843	} else {
844		abd_return_buf(zio->io_abd, bp->b_un.b_addr, zio->io_size);
845	}
846
847	kmem_free(vb, sizeof (vdev_buf_t));
848
849	zio_delay_interrupt(zio);
850	return (0);
851}
852
853static void
854vdev_disk_ioctl_free(zio_t *zio)
855{
856	kmem_free(zio->io_vsd, sizeof (struct dk_callback));
857}
858
859static const zio_vsd_ops_t vdev_disk_vsd_ops = {
860	vdev_disk_ioctl_free,
861	zio_vsd_default_cksum_report
862};
863
864static void
865vdev_disk_ioctl_done(void *zio_arg, int error)
866{
867	zio_t *zio = zio_arg;
868
869	zio->io_error = error;
870
871	zio_interrupt(zio);
872}
873
874static void
875vdev_disk_io_start(zio_t *zio)
876{
877	vdev_t *vd = zio->io_vd;
878	vdev_disk_t *dvd = vd->vdev_tsd;
879	unsigned long trim_flags = 0;
880	vdev_buf_t *vb;
881	struct dk_callback *dkc;
882	buf_t *bp;
883	int error;
884
885	/*
886	 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
887	 * Nothing to be done here but return failure.
888	 */
889	if (dvd == NULL || (dvd->vd_ldi_offline && dvd->vd_lh == NULL)) {
890		zio->io_error = ENXIO;
891		zio_interrupt(zio);
892		return;
893	}
894
895	switch (zio->io_type) {
896	case ZIO_TYPE_IOCTL:
897		/* XXPOLICY */
898		if (!vdev_readable(vd)) {
899			zio->io_error = SET_ERROR(ENXIO);
900			zio_interrupt(zio);
901			return;
902		}
903
904		switch (zio->io_cmd) {
905
906		case DKIOCFLUSHWRITECACHE:
907
908			if (zfs_nocacheflush)
909				break;
910
911			if (vd->vdev_nowritecache) {
912				zio->io_error = SET_ERROR(ENOTSUP);
913				break;
914			}
915
916			zio->io_vsd = dkc = kmem_alloc(sizeof (*dkc), KM_SLEEP);
917			zio->io_vsd_ops = &vdev_disk_vsd_ops;
918
919			dkc->dkc_callback = vdev_disk_ioctl_done;
920			dkc->dkc_flag = FLUSH_VOLATILE;
921			dkc->dkc_cookie = zio;
922
923			error = ldi_ioctl(dvd->vd_lh, zio->io_cmd,
924			    (uintptr_t)dkc, FKIOCTL, kcred, NULL);
925
926			if (error == 0) {
927				/*
928				 * The ioctl will be done asychronously,
929				 * and will call vdev_disk_ioctl_done()
930				 * upon completion.
931				 */
932				return;
933			}
934
935			zio->io_error = error;
936
937			break;
938
939		default:
940			zio->io_error = SET_ERROR(ENOTSUP);
941		}
942
943		zio_execute(zio);
944		return;
945
946	case ZIO_TYPE_TRIM:
947		if (zfs_no_trim == 1 || !vd->vdev_has_trim) {
948			zio->io_error = SET_ERROR(ENOTSUP);
949			zio_execute(zio);
950			return;
951		}
952		/* Currently only supported on ZoL. */
953		ASSERT0(zio->io_trim_flags & ZIO_TRIM_SECURE);
954
955		/* dkioc_free_list_t is already declared to hold one entry */
956		dkioc_free_list_t dfl;
957		dfl.dfl_flags = 0;
958		dfl.dfl_num_exts = 1;
959		dfl.dfl_offset = 0;
960		dfl.dfl_exts[0].dfle_start = zio->io_offset;
961		dfl.dfl_exts[0].dfle_length = zio->io_size;
962
963		zio->io_error = ldi_ioctl(dvd->vd_lh, DKIOCFREE,
964		    (uintptr_t)&dfl, FKIOCTL, kcred, NULL);
965
966		if (zio->io_error == ENOTSUP || zio->io_error == ENOTTY) {
967			/*
968			 * The device must have changed and now TRIM is
969			 * no longer supported.
970			 */
971			vd->vdev_has_trim = B_FALSE;
972		}
973
974		zio_interrupt(zio);
975		return;
976	}
977
978	ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE);
979	zio->io_target_timestamp = zio_handle_io_delay(zio);
980
981	vb = kmem_alloc(sizeof (vdev_buf_t), KM_SLEEP);
982
983	vb->vb_io = zio;
984	bp = &vb->vb_buf;
985
986	bioinit(bp);
987	bp->b_flags = B_BUSY | B_NOCACHE |
988	    (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
989	if (!(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))
990		bp->b_flags |= B_FAILFAST;
991	bp->b_bcount = zio->io_size;
992
993	if (zio->io_type == ZIO_TYPE_READ) {
994		bp->b_un.b_addr =
995		    abd_borrow_buf(zio->io_abd, zio->io_size);
996	} else {
997		bp->b_un.b_addr =
998		    abd_borrow_buf_copy(zio->io_abd, zio->io_size);
999	}
1000
1001	bp->b_lblkno = lbtodb(zio->io_offset);
1002	bp->b_bufsize = zio->io_size;
1003	bp->b_iodone = vdev_disk_io_intr;
1004
1005	/*
1006	 * In general we would expect ldi_strategy() to return non-zero only
1007	 * because of programming errors, but we've also seen this fail shortly
1008	 * after a disk dies.
1009	 */
1010	if (ldi_strategy(dvd->vd_lh, bp) != 0) {
1011		zio->io_error = ENXIO;
1012		zio_interrupt(zio);
1013	}
1014}
1015
1016static void
1017vdev_disk_io_done(zio_t *zio)
1018{
1019	vdev_t *vd = zio->io_vd;
1020
1021	/*
1022	 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
1023	 * the device has been removed.  If this is the case, then we trigger an
1024	 * asynchronous removal of the device. Otherwise, probe the device and
1025	 * make sure it's still accessible.
1026	 */
1027	if (zio->io_error == EIO && !vd->vdev_remove_wanted) {
1028		vdev_disk_t *dvd = vd->vdev_tsd;
1029		int state = DKIO_NONE;
1030
1031		if (ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state,
1032		    FKIOCTL, kcred, NULL) == 0 && state != DKIO_INSERTED) {
1033			/*
1034			 * We post the resource as soon as possible, instead of
1035			 * when the async removal actually happens, because the
1036			 * DE is using this information to discard previous I/O
1037			 * errors.
1038			 */
1039			zfs_post_remove(zio->io_spa, vd);
1040			vd->vdev_remove_wanted = B_TRUE;
1041			spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
1042		} else if (!vd->vdev_delayed_close) {
1043			vd->vdev_delayed_close = B_TRUE;
1044		}
1045	}
1046}
1047
1048vdev_ops_t vdev_disk_ops = {
1049	.vdev_op_open = vdev_disk_open,
1050	.vdev_op_close = vdev_disk_close,
1051	.vdev_op_asize = vdev_default_asize,
1052	.vdev_op_io_start = vdev_disk_io_start,
1053	.vdev_op_io_done = vdev_disk_io_done,
1054	.vdev_op_state_change = NULL,
1055	.vdev_op_need_resilver = NULL,
1056	.vdev_op_hold = vdev_disk_hold,
1057	.vdev_op_rele = vdev_disk_rele,
1058	.vdev_op_remap = NULL,
1059	.vdev_op_xlate = vdev_default_xlate,
1060	.vdev_op_type = VDEV_TYPE_DISK,		/* name of this vdev type */
1061	.vdev_op_leaf = B_TRUE			/* leaf vdev */
1062};
1063
1064/*
1065 * Given the root disk device devid or pathname, read the label from
1066 * the device, and construct a configuration nvlist.
1067 */
1068int
1069vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config)
1070{
1071	ldi_handle_t vd_lh;
1072	vdev_label_t *label;
1073	uint64_t s, size;
1074	int l;
1075	ddi_devid_t tmpdevid;
1076	int error = -1;
1077	char *minor_name;
1078
1079	/*
1080	 * Read the device label and build the nvlist.
1081	 */
1082	if (devid != NULL && ddi_devid_str_decode(devid, &tmpdevid,
1083	    &minor_name) == 0) {
1084		error = ldi_open_by_devid(tmpdevid, minor_name,
1085		    FREAD, kcred, &vd_lh, zfs_li);
1086		ddi_devid_free(tmpdevid);
1087		ddi_devid_str_free(minor_name);
1088	}
1089
1090	if (error && (error = ldi_open_by_name(devpath, FREAD, kcred, &vd_lh,
1091	    zfs_li)))
1092		return (error);
1093
1094	if (ldi_get_size(vd_lh, &s)) {
1095		(void) ldi_close(vd_lh, FREAD, kcred);
1096		return (SET_ERROR(EIO));
1097	}
1098
1099	size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t);
1100	label = kmem_alloc(sizeof (vdev_label_t), KM_SLEEP);
1101
1102	*config = NULL;
1103	for (l = 0; l < VDEV_LABELS; l++) {
1104		uint64_t offset, state, txg = 0;
1105
1106		/* read vdev label */
1107		offset = vdev_label_offset(size, l, 0);
1108		if (vdev_disk_ldi_physio(vd_lh, (caddr_t)label,
1109		    VDEV_SKIP_SIZE + VDEV_PHYS_SIZE, offset, B_READ) != 0)
1110			continue;
1111
1112		if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
1113		    sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0) {
1114			*config = NULL;
1115			continue;
1116		}
1117
1118		if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
1119		    &state) != 0 || state >= POOL_STATE_DESTROYED) {
1120			nvlist_free(*config);
1121			*config = NULL;
1122			continue;
1123		}
1124
1125		if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
1126		    &txg) != 0 || txg == 0) {
1127			nvlist_free(*config);
1128			*config = NULL;
1129			continue;
1130		}
1131
1132		break;
1133	}
1134
1135	kmem_free(label, sizeof (vdev_label_t));
1136	(void) ldi_close(vd_lh, FREAD, kcred);
1137	if (*config == NULL)
1138		error = SET_ERROR(EIDRM);
1139
1140	return (error);
1141}
1142