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, 2015 by Delphix. All rights reserved.
24 * Copyright 2016 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2013 Joyent, Inc. All rights reserved.
26 * Copyright (c) 2017 James S Blachly, MD <james.blachly@gmail.com>
27 */
28
29 #include <sys/zfs_context.h>
30 #include <sys/spa_impl.h>
31 #include <sys/refcount.h>
32 #include <sys/vdev_disk.h>
33 #include <sys/vdev_impl.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 * Virtual device vector for disks.
43 */
44
45 extern ldi_ident_t zfs_li;
46
47 static void vdev_disk_close(vdev_t *);
48
49 typedef struct vdev_disk_ldi_cb {
50 list_node_t lcb_next;
51 ldi_callback_id_t lcb_id;
52 } vdev_disk_ldi_cb_t;
53
54 static void
55 vdev_disk_alloc(vdev_t *vd)
56 {
57 vdev_disk_t *dvd;
58
59 dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
60 /*
61 * Create the LDI event callback list.
62 */
63 list_create(&dvd->vd_ldi_cbs, sizeof (vdev_disk_ldi_cb_t),
64 offsetof(vdev_disk_ldi_cb_t, lcb_next));
65 }
66
67 static void
68 vdev_disk_free(vdev_t *vd)
69 {
70 vdev_disk_t *dvd = vd->vdev_tsd;
71 vdev_disk_ldi_cb_t *lcb;
72
73 if (dvd == NULL)
74 return;
75
76 /*
77 * We have already closed the LDI handle. Clean up the LDI event
78 * callbacks and free vd->vdev_tsd.
79 */
80 while ((lcb = list_head(&dvd->vd_ldi_cbs)) != NULL) {
81 list_remove(&dvd->vd_ldi_cbs, lcb);
82 (void) ldi_ev_remove_callbacks(lcb->lcb_id);
83 kmem_free(lcb, sizeof (vdev_disk_ldi_cb_t));
84 }
85 list_destroy(&dvd->vd_ldi_cbs);
86 kmem_free(dvd, sizeof (vdev_disk_t));
87 vd->vdev_tsd = NULL;
88 }
89
90 /* ARGSUSED */
91 static int
92 vdev_disk_off_notify(ldi_handle_t lh, ldi_ev_cookie_t ecookie, void *arg,
93 void *ev_data)
94 {
95 vdev_t *vd = (vdev_t *)arg;
96 vdev_disk_t *dvd = vd->vdev_tsd;
97
98 /*
99 * Ignore events other than offline.
100 */
101 if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_OFFLINE) != 0)
102 return (LDI_EV_SUCCESS);
103
104 /*
105 * All LDI handles must be closed for the state change to succeed, so
106 * call on vdev_disk_close() to do this.
107 *
108 * We inform vdev_disk_close that it is being called from offline
109 * notify context so it will defer cleanup of LDI event callbacks and
110 * freeing of vd->vdev_tsd to the offline finalize or a reopen.
111 */
112 dvd->vd_ldi_offline = B_TRUE;
113 vdev_disk_close(vd);
114
115 /*
116 * Now that the device is closed, request that the spa_async_thread
117 * mark the device as REMOVED and notify FMA of the removal.
118 */
119 zfs_post_remove(vd->vdev_spa, vd);
120 vd->vdev_remove_wanted = B_TRUE;
121 spa_async_request(vd->vdev_spa, SPA_ASYNC_REMOVE);
122
123 return (LDI_EV_SUCCESS);
124 }
125
126 /* ARGSUSED */
127 static void
128 vdev_disk_off_finalize(ldi_handle_t lh, ldi_ev_cookie_t ecookie,
129 int ldi_result, void *arg, void *ev_data)
130 {
131 vdev_t *vd = (vdev_t *)arg;
132
133 /*
134 * Ignore events other than offline.
135 */
136 if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_OFFLINE) != 0)
137 return;
138
139 /*
140 * We have already closed the LDI handle in notify.
141 * Clean up the LDI event callbacks and free vd->vdev_tsd.
142 */
143 vdev_disk_free(vd);
144
145 /*
146 * Request that the vdev be reopened if the offline state change was
147 * unsuccessful.
148 */
149 if (ldi_result != LDI_EV_SUCCESS) {
150 vd->vdev_probe_wanted = B_TRUE;
151 spa_async_request(vd->vdev_spa, SPA_ASYNC_PROBE);
152 }
153 }
154
155 static ldi_ev_callback_t vdev_disk_off_callb = {
156 .cb_vers = LDI_EV_CB_VERS,
157 .cb_notify = vdev_disk_off_notify,
158 .cb_finalize = vdev_disk_off_finalize
159 };
160
161 /* ARGSUSED */
162 static void
163 vdev_disk_dgrd_finalize(ldi_handle_t lh, ldi_ev_cookie_t ecookie,
164 int ldi_result, void *arg, void *ev_data)
165 {
166 vdev_t *vd = (vdev_t *)arg;
167
168 /*
169 * Ignore events other than degrade.
170 */
171 if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_DEGRADE) != 0)
172 return;
173
174 /*
175 * Degrade events always succeed. Mark the vdev as degraded.
176 * This status is purely informative for the user.
177 */
178 (void) vdev_degrade(vd->vdev_spa, vd->vdev_guid, 0);
179 }
180
181 static ldi_ev_callback_t vdev_disk_dgrd_callb = {
182 .cb_vers = LDI_EV_CB_VERS,
183 .cb_notify = NULL,
184 .cb_finalize = vdev_disk_dgrd_finalize
185 };
186
187 static void
188 vdev_disk_hold(vdev_t *vd)
189 {
190 ddi_devid_t devid;
191 char *minor;
192
193 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
194
195 /*
196 * We must have a pathname, and it must be absolute.
197 */
198 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/')
199 return;
200
201 /*
202 * Only prefetch path and devid info if the device has
203 * never been opened.
204 */
205 if (vd->vdev_tsd != NULL)
206 return;
207
208 if (vd->vdev_wholedisk == -1ULL) {
209 size_t len = strlen(vd->vdev_path) + 3;
210 char *buf = kmem_alloc(len, KM_SLEEP);
211
212 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
213
214 (void) ldi_vp_from_name(buf, &vd->vdev_name_vp);
215 kmem_free(buf, len);
216 }
217
218 if (vd->vdev_name_vp == NULL)
219 (void) ldi_vp_from_name(vd->vdev_path, &vd->vdev_name_vp);
220
221 if (vd->vdev_devid != NULL &&
222 ddi_devid_str_decode(vd->vdev_devid, &devid, &minor) == 0) {
223 (void) ldi_vp_from_devid(devid, minor, &vd->vdev_devid_vp);
224 ddi_devid_str_free(minor);
225 ddi_devid_free(devid);
226 }
227 }
228
229 static void
230 vdev_disk_rele(vdev_t *vd)
231 {
232 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
233
234 if (vd->vdev_name_vp) {
235 VN_RELE_ASYNC(vd->vdev_name_vp,
236 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
237 vd->vdev_name_vp = NULL;
238 }
239 if (vd->vdev_devid_vp) {
240 VN_RELE_ASYNC(vd->vdev_devid_vp,
241 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
242 vd->vdev_devid_vp = NULL;
243 }
244 }
245
246 /*
247 * We want to be loud in DEBUG kernels when DKIOCGMEDIAINFOEXT fails, or when
248 * even a fallback to DKIOCGMEDIAINFO fails.
249 */
250 #ifdef DEBUG
251 #define VDEV_DEBUG(...) cmn_err(CE_NOTE, __VA_ARGS__)
252 #else
253 #define VDEV_DEBUG(...) /* Nothing... */
254 #endif
255
256 static int
257 vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
258 uint64_t *ashift)
259 {
260 spa_t *spa = vd->vdev_spa;
261 vdev_disk_t *dvd = vd->vdev_tsd;
262 ldi_ev_cookie_t ecookie;
263 vdev_disk_ldi_cb_t *lcb;
264 union {
265 struct dk_minfo_ext ude;
266 struct dk_minfo ud;
267 } dks;
268 struct dk_minfo_ext *dkmext = &dks.ude;
269 struct dk_minfo *dkm = &dks.ud;
270 int error;
271 dev_t dev;
272 int otyp;
273 boolean_t validate_devid = B_FALSE;
274 ddi_devid_t devid;
275 uint64_t capacity = 0, blksz = 0, pbsize;
276 int device_rotational;
277
278 /*
279 * We must have a pathname, and it must be absolute.
280 */
281 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
282 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
283 return (SET_ERROR(EINVAL));
284 }
285
286 /*
287 * Reopen the device if it's not currently open. Otherwise,
288 * just update the physical size of the device.
289 */
290 if (dvd != NULL) {
291 if (dvd->vd_ldi_offline && dvd->vd_lh == NULL) {
292 /*
293 * If we are opening a device in its offline notify
294 * context, the LDI handle was just closed. Clean
295 * up the LDI event callbacks and free vd->vdev_tsd.
296 */
297 vdev_disk_free(vd);
298 } else {
299 ASSERT(vd->vdev_reopening);
300 goto skip_open;
301 }
302 }
303
304 /*
305 * Create vd->vdev_tsd.
306 */
307 vdev_disk_alloc(vd);
308 dvd = vd->vdev_tsd;
309
310 /*
311 * When opening a disk device, we want to preserve the user's original
312 * intent. We always want to open the device by the path the user gave
313 * us, even if it is one of multiple paths to the same device. But we
314 * also want to be able to survive disks being removed/recabled.
315 * Therefore the sequence of opening devices is:
316 *
317 * 1. Try opening the device by path. For legacy pools without the
318 * 'whole_disk' property, attempt to fix the path by appending 's0'.
319 *
320 * 2. If the devid of the device matches the stored value, return
321 * success.
322 *
323 * 3. Otherwise, the device may have moved. Try opening the device
324 * by the devid instead.
325 */
326 if (vd->vdev_devid != NULL) {
327 if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
328 &dvd->vd_minor) != 0) {
329 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
330 return (SET_ERROR(EINVAL));
331 }
332 }
333
334 error = EINVAL; /* presume failure */
335
336 if (vd->vdev_path != NULL) {
337
338 if (vd->vdev_wholedisk == -1ULL) {
339 size_t len = strlen(vd->vdev_path) + 3;
340 char *buf = kmem_alloc(len, KM_SLEEP);
341
342 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
343
344 error = ldi_open_by_name(buf, spa_mode(spa), kcred,
345 &dvd->vd_lh, zfs_li);
346 if (error == 0) {
347 spa_strfree(vd->vdev_path);
348 vd->vdev_path = buf;
349 vd->vdev_wholedisk = 1ULL;
350 } else {
351 kmem_free(buf, len);
352 }
353 }
354
355 /*
356 * If we have not yet opened the device, try to open it by the
357 * specified path.
358 */
359 if (error != 0) {
360 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
361 kcred, &dvd->vd_lh, zfs_li);
362 }
363
364 /*
365 * Compare the devid to the stored value.
366 */
367 if (error == 0 && vd->vdev_devid != NULL &&
368 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
369 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
370 error = SET_ERROR(EINVAL);
371 (void) ldi_close(dvd->vd_lh, spa_mode(spa),
372 kcred);
373 dvd->vd_lh = NULL;
374 }
375 ddi_devid_free(devid);
376 }
377
378 /*
379 * If we succeeded in opening the device, but 'vdev_wholedisk'
380 * is not yet set, then this must be a slice.
381 */
382 if (error == 0 && vd->vdev_wholedisk == -1ULL)
383 vd->vdev_wholedisk = 0;
384 }
385
386 /*
387 * If we were unable to open by path, or the devid check fails, open by
388 * devid instead.
389 */
390 if (error != 0 && vd->vdev_devid != NULL) {
391 error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
392 spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
393 }
394
395 /*
396 * If all else fails, then try opening by physical path (if available)
397 * or the logical path (if we failed due to the devid check). While not
398 * as reliable as the devid, this will give us something, and the higher
399 * level vdev validation will prevent us from opening the wrong device.
400 */
401 if (error) {
402 if (vd->vdev_devid != NULL)
403 validate_devid = B_TRUE;
404
405 if (vd->vdev_physpath != NULL &&
406 (dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV)
407 error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
408 kcred, &dvd->vd_lh, zfs_li);
409
410 /*
411 * Note that we don't support the legacy auto-wholedisk support
412 * as above. This hasn't been used in a very long time and we
413 * don't need to propagate its oddities to this edge condition.
414 */
415 if (error && vd->vdev_path != NULL)
416 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
417 kcred, &dvd->vd_lh, zfs_li);
418 }
419
420 if (error) {
421 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
422 return (error);
423 }
424
425 /*
426 * Now that the device has been successfully opened, update the devid
427 * if necessary.
428 */
429 if (validate_devid && spa_writeable(spa) &&
430 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
431 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
432 char *vd_devid;
433
434 vd_devid = ddi_devid_str_encode(devid, dvd->vd_minor);
435 zfs_dbgmsg("vdev %s: update devid from %s, "
436 "to %s", vd->vdev_path, vd->vdev_devid, vd_devid);
437 spa_strfree(vd->vdev_devid);
438 vd->vdev_devid = spa_strdup(vd_devid);
439 ddi_devid_str_free(vd_devid);
440 }
441 ddi_devid_free(devid);
442 }
443
444 /*
445 * Once a device is opened, verify that the physical device path (if
446 * available) is up to date.
447 */
448 if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
449 ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
450 char *physpath, *minorname;
451
452 physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
453 minorname = NULL;
454 if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
455 ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
456 (vd->vdev_physpath == NULL ||
457 strcmp(vd->vdev_physpath, physpath) != 0)) {
458 if (vd->vdev_physpath)
459 spa_strfree(vd->vdev_physpath);
460 (void) strlcat(physpath, ":", MAXPATHLEN);
461 (void) strlcat(physpath, minorname, MAXPATHLEN);
462 vd->vdev_physpath = spa_strdup(physpath);
463 }
464 if (minorname)
465 kmem_free(minorname, strlen(minorname) + 1);
466 kmem_free(physpath, MAXPATHLEN);
467 }
468
469 /*
470 * Register callbacks for the LDI offline event.
471 */
472 if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_OFFLINE, &ecookie) ==
473 LDI_EV_SUCCESS) {
474 lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
475 list_insert_tail(&dvd->vd_ldi_cbs, lcb);
476 (void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
477 &vdev_disk_off_callb, (void *) vd, &lcb->lcb_id);
478 }
479
480 /*
481 * Register callbacks for the LDI degrade event.
482 */
483 if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_DEGRADE, &ecookie) ==
484 LDI_EV_SUCCESS) {
485 lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
486 list_insert_tail(&dvd->vd_ldi_cbs, lcb);
487 (void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
488 &vdev_disk_dgrd_callb, (void *) vd, &lcb->lcb_id);
489 }
490 skip_open:
491 /*
492 * Determine the actual size of the device.
493 */
494 if (ldi_get_size(dvd->vd_lh, psize) != 0) {
495 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
496 return (SET_ERROR(EINVAL));
497 }
498
499 *max_psize = *psize;
500
501 /*
502 * Determine the device's minimum transfer size.
503 * If the ioctl isn't supported, assume DEV_BSIZE.
504 */
505 if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT,
506 (intptr_t)dkmext, FKIOCTL, kcred, NULL)) == 0) {
507 capacity = dkmext->dki_capacity - 1;
508 blksz = dkmext->dki_lbsize;
509 pbsize = dkmext->dki_pbsize;
510 } else if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFO,
511 (intptr_t)dkm, FKIOCTL, kcred, NULL)) == 0) {
512 VDEV_DEBUG(
513 "vdev_disk_open(\"%s\"): fallback to DKIOCGMEDIAINFO\n",
514 vd->vdev_path);
515 capacity = dkm->dki_capacity - 1;
516 blksz = dkm->dki_lbsize;
517 pbsize = blksz;
518 } else {
519 VDEV_DEBUG("vdev_disk_open(\"%s\"): "
520 "both DKIOCGMEDIAINFO{,EXT} calls failed, %d\n",
521 vd->vdev_path, error);
522 pbsize = DEV_BSIZE;
523 }
524
525 *ashift = highbit64(MAX(pbsize, SPA_MINBLOCKSIZE)) - 1;
526
527 if (vd->vdev_wholedisk == 1) {
528 int wce = 1;
529
530 if (error == 0) {
531 /*
532 * If we have the capability to expand, we'd have
533 * found out via success from DKIOCGMEDIAINFO{,EXT}.
534 * Adjust max_psize upward accordingly since we know
535 * we own the whole disk now.
536 */
537 *max_psize = capacity * blksz;
538 }
539
540 /*
541 * Since we own the whole disk, try to enable disk write
542 * caching. We ignore errors because it's OK if we can't do it.
543 */
544 (void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
545 FKIOCTL, kcred, NULL);
546 }
547
548 /*
549 * Inform the ZIO pipeline if we are non-rotational
550 */
551 device_rotational = ldi_prop_get_int(dvd->vd_lh, LDI_DEV_T_ANY,
552 "device-rotational", 0);
553 vd->vdev_nonrot = (device_rotational ? B_FALSE : B_TRUE);
554
555 cmn_err(CE_NOTE, "![vdev_disk_open] %s :: vd->vdev_nonrot == %d\n",
556 vd->vdev_path, (int) vd->vdev_nonrot);
557
558 /*
559 * Clear the nowritecache bit, so that on a vdev_reopen() we will
560 * try again.
561 */
562 vd->vdev_nowritecache = B_FALSE;
563
564 return (0);
565 }
566
567 static void
568 vdev_disk_close(vdev_t *vd)
569 {
570 vdev_disk_t *dvd = vd->vdev_tsd;
571
572 if (vd->vdev_reopening || dvd == NULL)
573 return;
574
575 if (dvd->vd_minor != NULL) {
576 ddi_devid_str_free(dvd->vd_minor);
577 dvd->vd_minor = NULL;
578 }
579
580 if (dvd->vd_devid != NULL) {
581 ddi_devid_free(dvd->vd_devid);
582 dvd->vd_devid = NULL;
583 }
584
585 if (dvd->vd_lh != NULL) {
586 (void) ldi_close(dvd->vd_lh, spa_mode(vd->vdev_spa), kcred);
587 dvd->vd_lh = NULL;
588 }
589
590 vd->vdev_delayed_close = B_FALSE;
591 /*
592 * If we closed the LDI handle due to an offline notify from LDI,
593 * don't free vd->vdev_tsd or unregister the callbacks here;
594 * the offline finalize callback or a reopen will take care of it.
595 */
596 if (dvd->vd_ldi_offline)
597 return;
598
599 vdev_disk_free(vd);
600 }
601
602 int
603 vdev_disk_physio(vdev_t *vd, caddr_t data,
604 size_t size, uint64_t offset, int flags, boolean_t isdump)
605 {
606 vdev_disk_t *dvd = vd->vdev_tsd;
607
608 /*
609 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
610 * Nothing to be done here but return failure.
611 */
612 if (dvd == NULL || (dvd->vd_ldi_offline && dvd->vd_lh == NULL))
613 return (EIO);
614
615 ASSERT(vd->vdev_ops == &vdev_disk_ops);
616
617 /*
618 * If in the context of an active crash dump, use the ldi_dump(9F)
619 * call instead of ldi_strategy(9F) as usual.
620 */
621 if (isdump) {
622 ASSERT3P(dvd, !=, NULL);
623 return (ldi_dump(dvd->vd_lh, data, lbtodb(offset),
624 lbtodb(size)));
625 }
626
627 return (vdev_disk_ldi_physio(dvd->vd_lh, data, size, offset, flags));
628 }
629
630 int
631 vdev_disk_ldi_physio(ldi_handle_t vd_lh, caddr_t data,
632 size_t size, uint64_t offset, int flags)
633 {
634 buf_t *bp;
635 int error = 0;
636
637 if (vd_lh == NULL)
638 return (SET_ERROR(EINVAL));
639
640 ASSERT(flags & B_READ || flags & B_WRITE);
641
642 bp = getrbuf(KM_SLEEP);
643 bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST;
644 bp->b_bcount = size;
645 bp->b_un.b_addr = (void *)data;
646 bp->b_lblkno = lbtodb(offset);
647 bp->b_bufsize = size;
648
649 error = ldi_strategy(vd_lh, bp);
650 ASSERT(error == 0);
651 if ((error = biowait(bp)) == 0 && bp->b_resid != 0)
652 error = SET_ERROR(EIO);
653 freerbuf(bp);
654
655 return (error);
656 }
657
658 static void
659 vdev_disk_io_intr(buf_t *bp)
660 {
661 vdev_buf_t *vb = (vdev_buf_t *)bp;
662 zio_t *zio = vb->vb_io;
663
664 /*
665 * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO.
666 * Rather than teach the rest of the stack about other error
667 * possibilities (EFAULT, etc), we normalize the error value here.
668 */
669 zio->io_error = (geterror(bp) != 0 ? EIO : 0);
670
671 if (zio->io_error == 0 && bp->b_resid != 0)
672 zio->io_error = SET_ERROR(EIO);
673
674 if (zio->io_type == ZIO_TYPE_READ) {
675 abd_return_buf_copy(zio->io_abd, bp->b_un.b_addr, zio->io_size);
676 } else {
677 abd_return_buf(zio->io_abd, bp->b_un.b_addr, zio->io_size);
678 }
679
680 kmem_free(vb, sizeof (vdev_buf_t));
681
682 zio_delay_interrupt(zio);
683 }
684
685 static void
686 vdev_disk_ioctl_free(zio_t *zio)
687 {
688 kmem_free(zio->io_vsd, sizeof (struct dk_callback));
689 }
690
691 static const zio_vsd_ops_t vdev_disk_vsd_ops = {
692 vdev_disk_ioctl_free,
693 zio_vsd_default_cksum_report
694 };
695
696 static void
697 vdev_disk_ioctl_done(void *zio_arg, int error)
698 {
699 zio_t *zio = zio_arg;
700
701 zio->io_error = error;
702
703 zio_interrupt(zio);
704 }
705
706 static void
707 vdev_disk_io_start(zio_t *zio)
708 {
709 vdev_t *vd = zio->io_vd;
710 vdev_disk_t *dvd = vd->vdev_tsd;
711 vdev_buf_t *vb;
712 struct dk_callback *dkc;
713 buf_t *bp;
714 int error;
715
716 /*
717 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
718 * Nothing to be done here but return failure.
719 */
720 if (dvd == NULL || (dvd->vd_ldi_offline && dvd->vd_lh == NULL)) {
721 zio->io_error = ENXIO;
722 zio_interrupt(zio);
723 return;
724 }
725
726 if (zio->io_type == ZIO_TYPE_IOCTL) {
727 /* XXPOLICY */
728 if (!vdev_readable(vd)) {
729 zio->io_error = SET_ERROR(ENXIO);
730 zio_interrupt(zio);
731 return;
732 }
733
734 switch (zio->io_cmd) {
735
736 case DKIOCFLUSHWRITECACHE:
737
738 if (zfs_nocacheflush)
739 break;
740
741 if (vd->vdev_nowritecache) {
742 zio->io_error = SET_ERROR(ENOTSUP);
743 break;
744 }
745
746 zio->io_vsd = dkc = kmem_alloc(sizeof (*dkc), KM_SLEEP);
747 zio->io_vsd_ops = &vdev_disk_vsd_ops;
748
749 dkc->dkc_callback = vdev_disk_ioctl_done;
750 dkc->dkc_flag = FLUSH_VOLATILE;
751 dkc->dkc_cookie = zio;
752
753 error = ldi_ioctl(dvd->vd_lh, zio->io_cmd,
754 (uintptr_t)dkc, FKIOCTL, kcred, NULL);
755
756 if (error == 0) {
757 /*
758 * The ioctl will be done asychronously,
759 * and will call vdev_disk_ioctl_done()
760 * upon completion.
761 */
762 return;
763 }
764
765 zio->io_error = error;
766
767 break;
768
769 default:
770 zio->io_error = SET_ERROR(ENOTSUP);
771 }
772
773 zio_execute(zio);
774 return;
775 }
776
777 ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE);
778 zio->io_target_timestamp = zio_handle_io_delay(zio);
779
780 vb = kmem_alloc(sizeof (vdev_buf_t), KM_SLEEP);
781
782 vb->vb_io = zio;
783 bp = &vb->vb_buf;
784
785 bioinit(bp);
786 bp->b_flags = B_BUSY | B_NOCACHE |
787 (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
788 if (!(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))
789 bp->b_flags |= B_FAILFAST;
790 bp->b_bcount = zio->io_size;
791
792 if (zio->io_type == ZIO_TYPE_READ) {
793 bp->b_un.b_addr =
794 abd_borrow_buf(zio->io_abd, zio->io_size);
795 } else {
796 bp->b_un.b_addr =
797 abd_borrow_buf_copy(zio->io_abd, zio->io_size);
798 }
799
800 bp->b_lblkno = lbtodb(zio->io_offset);
801 bp->b_bufsize = zio->io_size;
802 bp->b_iodone = (int (*)())vdev_disk_io_intr;
803
804 /* ldi_strategy() will return non-zero only on programming errors */
805 VERIFY(ldi_strategy(dvd->vd_lh, bp) == 0);
806 }
807
808 static void
809 vdev_disk_io_done(zio_t *zio)
810 {
811 vdev_t *vd = zio->io_vd;
812
813 /*
814 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
815 * the device has been removed. If this is the case, then we trigger an
816 * asynchronous removal of the device. Otherwise, probe the device and
817 * make sure it's still accessible.
818 */
819 if (zio->io_error == EIO && !vd->vdev_remove_wanted) {
820 vdev_disk_t *dvd = vd->vdev_tsd;
821 int state = DKIO_NONE;
822
823 if (ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state,
824 FKIOCTL, kcred, NULL) == 0 && state != DKIO_INSERTED) {
825 /*
826 * We post the resource as soon as possible, instead of
827 * when the async removal actually happens, because the
828 * DE is using this information to discard previous I/O
829 * errors.
830 */
831 zfs_post_remove(zio->io_spa, vd);
832 vd->vdev_remove_wanted = B_TRUE;
833 spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
834 } else if (!vd->vdev_delayed_close) {
835 vd->vdev_delayed_close = B_TRUE;
836 }
837 }
838 }
839
840 vdev_ops_t vdev_disk_ops = {
841 vdev_disk_open,
842 vdev_disk_close,
843 vdev_default_asize,
844 vdev_disk_io_start,
845 vdev_disk_io_done,
846 NULL,
847 vdev_disk_hold,
848 vdev_disk_rele,
849 VDEV_TYPE_DISK, /* name of this vdev type */
850 B_TRUE /* leaf vdev */
851 };
852
853 /*
854 * Given the root disk device devid or pathname, read the label from
855 * the device, and construct a configuration nvlist.
856 */
857 int
858 vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config)
859 {
860 ldi_handle_t vd_lh;
861 vdev_label_t *label;
862 uint64_t s, size;
863 int l;
864 ddi_devid_t tmpdevid;
865 int error = -1;
866 char *minor_name;
867
868 /*
869 * Read the device label and build the nvlist.
870 */
871 if (devid != NULL && ddi_devid_str_decode(devid, &tmpdevid,
872 &minor_name) == 0) {
873 error = ldi_open_by_devid(tmpdevid, minor_name,
874 FREAD, kcred, &vd_lh, zfs_li);
875 ddi_devid_free(tmpdevid);
876 ddi_devid_str_free(minor_name);
877 }
878
879 if (error && (error = ldi_open_by_name(devpath, FREAD, kcred, &vd_lh,
880 zfs_li)))
881 return (error);
882
883 if (ldi_get_size(vd_lh, &s)) {
884 (void) ldi_close(vd_lh, FREAD, kcred);
885 return (SET_ERROR(EIO));
886 }
887
888 size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t);
889 label = kmem_alloc(sizeof (vdev_label_t), KM_SLEEP);
890
891 *config = NULL;
892 for (l = 0; l < VDEV_LABELS; l++) {
893 uint64_t offset, state, txg = 0;
894
895 /* read vdev label */
896 offset = vdev_label_offset(size, l, 0);
897 if (vdev_disk_ldi_physio(vd_lh, (caddr_t)label,
898 VDEV_SKIP_SIZE + VDEV_PHYS_SIZE, offset, B_READ) != 0)
899 continue;
900
901 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
902 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0) {
903 *config = NULL;
904 continue;
905 }
906
907 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
908 &state) != 0 || state >= POOL_STATE_DESTROYED) {
909 nvlist_free(*config);
910 *config = NULL;
911 continue;
912 }
913
914 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
915 &txg) != 0 || txg == 0) {
916 nvlist_free(*config);
917 *config = NULL;
918 continue;
919 }
920
921 break;
922 }
923
924 kmem_free(label, sizeof (vdev_label_t));
925 (void) ldi_close(vd_lh, FREAD, kcred);
926 if (*config == NULL)
927 error = SET_ERROR(EIDRM);
928
929 return (error);
930 }