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) 2013 by Delphix. All rights reserved.
24 * Copyright 2013 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/fs/zfs.h>
34 #include <sys/zio.h>
35 #include <sys/sunldi.h>
36 #include <sys/efi_partition.h>
37 #include <sys/fm/fs/zfs.h>
38
39 /*
40 * Virtual device vector for disks.
41 */
42
43 extern ldi_ident_t zfs_li;
44
45 static void vdev_disk_close(vdev_t *);
46
47 typedef struct vdev_disk_ldi_cb {
48 list_node_t lcb_next;
49 ldi_callback_id_t lcb_id;
50 } vdev_disk_ldi_cb_t;
51
52 static void
53 vdev_disk_alloc(vdev_t *vd)
54 {
55 vdev_disk_t *dvd;
56
57 dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
58 /*
59 * Create the LDI event callback list.
60 */
61 list_create(&dvd->vd_ldi_cbs, sizeof (vdev_disk_ldi_cb_t),
62 offsetof(vdev_disk_ldi_cb_t, lcb_next));
63 }
64
65 static void
66 vdev_disk_free(vdev_t *vd)
67 {
68 vdev_disk_t *dvd = vd->vdev_tsd;
69 vdev_disk_ldi_cb_t *lcb;
70
71 if (dvd == NULL)
72 return;
73
74 /*
75 * We have already closed the LDI handle. Clean up the LDI event
76 * callbacks and free vd->vdev_tsd.
77 */
78 while ((lcb = list_head(&dvd->vd_ldi_cbs)) != NULL) {
79 list_remove(&dvd->vd_ldi_cbs, lcb);
80 (void) ldi_ev_remove_callbacks(lcb->lcb_id);
81 kmem_free(lcb, sizeof (vdev_disk_ldi_cb_t));
82 }
83 list_destroy(&dvd->vd_ldi_cbs);
84 kmem_free(dvd, sizeof (vdev_disk_t));
85 vd->vdev_tsd = NULL;
86 }
87
88 /* ARGSUSED */
89 static int
90 vdev_disk_off_notify(ldi_handle_t lh, ldi_ev_cookie_t ecookie, void *arg,
91 void *ev_data)
92 {
93 vdev_t *vd = (vdev_t *)arg;
94 vdev_disk_t *dvd = vd->vdev_tsd;
95
96 /*
97 * Ignore events other than offline.
98 */
99 if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_OFFLINE) != 0)
100 return (LDI_EV_SUCCESS);
101
102 /*
103 * All LDI handles must be closed for the state change to succeed, so
104 * call on vdev_disk_close() to do this.
105 *
106 * We inform vdev_disk_close that it is being called from offline
107 * notify context so it will defer cleanup of LDI event callbacks and
108 * freeing of vd->vdev_tsd to the offline finalize or a reopen.
109 */
110 dvd->vd_ldi_offline = B_TRUE;
111 vdev_disk_close(vd);
112
113 /*
114 * Now that the device is closed, request that the spa_async_thread
115 * mark the device as REMOVED and notify FMA of the removal.
116 */
117 zfs_post_remove(vd->vdev_spa, vd);
118 vd->vdev_remove_wanted = B_TRUE;
119 spa_async_request(vd->vdev_spa, SPA_ASYNC_REMOVE);
120
121 return (LDI_EV_SUCCESS);
122 }
123
124 /* ARGSUSED */
125 static void
126 vdev_disk_off_finalize(ldi_handle_t lh, ldi_ev_cookie_t ecookie,
127 int ldi_result, void *arg, void *ev_data)
128 {
129 vdev_t *vd = (vdev_t *)arg;
130
131 /*
132 * Ignore events other than offline.
133 */
134 if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_OFFLINE) != 0)
135 return;
136
137 /*
138 * We have already closed the LDI handle in notify.
139 * Clean up the LDI event callbacks and free vd->vdev_tsd.
140 */
141 vdev_disk_free(vd);
142
143 /*
144 * Request that the vdev be reopened if the offline state change was
145 * unsuccessful.
146 */
147 if (ldi_result != LDI_EV_SUCCESS) {
148 vd->vdev_probe_wanted = B_TRUE;
149 spa_async_request(vd->vdev_spa, SPA_ASYNC_PROBE);
150 }
151 }
152
153 static ldi_ev_callback_t vdev_disk_off_callb = {
154 .cb_vers = LDI_EV_CB_VERS,
155 .cb_notify = vdev_disk_off_notify,
156 .cb_finalize = vdev_disk_off_finalize
157 };
158
159 /* ARGSUSED */
160 static void
161 vdev_disk_dgrd_finalize(ldi_handle_t lh, ldi_ev_cookie_t ecookie,
162 int ldi_result, void *arg, void *ev_data)
163 {
164 vdev_t *vd = (vdev_t *)arg;
165
166 /*
167 * Ignore events other than degrade.
168 */
169 if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_DEGRADE) != 0)
170 return;
171
172 /*
173 * Degrade events always succeed. Mark the vdev as degraded.
174 * This status is purely informative for the user.
175 */
176 (void) vdev_degrade(vd->vdev_spa, vd->vdev_guid, 0);
177 }
178
179 static ldi_ev_callback_t vdev_disk_dgrd_callb = {
180 .cb_vers = LDI_EV_CB_VERS,
181 .cb_notify = NULL,
182 .cb_finalize = vdev_disk_dgrd_finalize
183 };
184
185 static void
186 vdev_disk_hold(vdev_t *vd)
187 {
188 ddi_devid_t devid;
189 char *minor;
190
191 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
192
193 /*
194 * We must have a pathname, and it must be absolute.
195 */
196 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/')
197 return;
198
199 /*
200 * Only prefetch path and devid info if the device has
201 * never been opened.
202 */
203 if (vd->vdev_tsd != NULL)
204 return;
205
206 if (vd->vdev_wholedisk == -1ULL) {
207 size_t len = strlen(vd->vdev_path) + 3;
208 char *buf = kmem_alloc(len, KM_SLEEP);
209
210 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
211
212 (void) ldi_vp_from_name(buf, &vd->vdev_name_vp);
213 kmem_free(buf, len);
214 }
215
216 if (vd->vdev_name_vp == NULL)
217 (void) ldi_vp_from_name(vd->vdev_path, &vd->vdev_name_vp);
218
219 if (vd->vdev_devid != NULL &&
220 ddi_devid_str_decode(vd->vdev_devid, &devid, &minor) == 0) {
221 (void) ldi_vp_from_devid(devid, minor, &vd->vdev_devid_vp);
222 ddi_devid_str_free(minor);
223 ddi_devid_free(devid);
224 }
225 }
226
227 static void
228 vdev_disk_rele(vdev_t *vd)
229 {
230 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
231
232 if (vd->vdev_name_vp) {
233 VN_RELE_ASYNC(vd->vdev_name_vp,
234 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
235 vd->vdev_name_vp = NULL;
236 }
237 if (vd->vdev_devid_vp) {
238 VN_RELE_ASYNC(vd->vdev_devid_vp,
239 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
240 vd->vdev_devid_vp = NULL;
241 }
242 }
243
244 static uint64_t
245 vdev_disk_get_space(vdev_t *vd, uint64_t capacity, uint_t blksz)
246 {
247 ASSERT(vd->vdev_wholedisk);
248
249 vdev_disk_t *dvd = vd->vdev_tsd;
250 dk_efi_t dk_ioc;
251 efi_gpt_t *efi;
252 uint64_t avail_space = 0;
253 int efisize = EFI_LABEL_SIZE * 2;
254
255 dk_ioc.dki_data = kmem_alloc(efisize, KM_SLEEP);
256 dk_ioc.dki_lba = 1;
257 dk_ioc.dki_length = efisize;
258 dk_ioc.dki_data_64 = (uint64_t)(uintptr_t)dk_ioc.dki_data;
259 efi = dk_ioc.dki_data;
260
261 if (ldi_ioctl(dvd->vd_lh, DKIOCGETEFI, (intptr_t)&dk_ioc,
262 FKIOCTL, kcred, NULL) == 0) {
263 uint64_t efi_altern_lba = LE_64(efi->efi_gpt_AlternateLBA);
264
265 zfs_dbgmsg("vdev %s, capacity %llu, altern lba %llu",
266 vd->vdev_path, capacity, efi_altern_lba);
267 if (capacity > efi_altern_lba)
268 avail_space = (capacity - efi_altern_lba) * blksz;
269 }
270 kmem_free(dk_ioc.dki_data, efisize);
271 return (avail_space);
272 }
273
274 /*
275 * We want to be loud in DEBUG kernels when DKIOCGMEDIAINFOEXT fails, or when
276 * even a fallback to DKIOCGMEDIAINFO fails.
277 */
278 #ifdef DEBUG
279 #define VDEV_DEBUG(...) cmn_err(CE_NOTE, __VA_ARGS__)
280 #else
281 #define VDEV_DEBUG(...) /* Nothing... */
282 #endif
283
284 static int
285 vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
286 uint64_t *ashift)
287 {
288 spa_t *spa = vd->vdev_spa;
289 vdev_disk_t *dvd = vd->vdev_tsd;
290 ldi_ev_cookie_t ecookie;
291 vdev_disk_ldi_cb_t *lcb;
292 union {
293 struct dk_minfo_ext ude;
294 struct dk_minfo ud;
295 } dks;
296 struct dk_minfo_ext *dkmext = &dks.ude;
297 struct dk_minfo *dkm = &dks.ud;
298 int error;
299 dev_t dev;
300 int otyp;
301 boolean_t validate_devid = B_FALSE;
302 ddi_devid_t devid;
303 uint64_t capacity = 0, blksz = 0, pbsize;
304
305 /*
306 * We must have a pathname, and it must be absolute.
307 */
308 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
309 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
310 return (SET_ERROR(EINVAL));
311 }
312
313 /*
314 * Reopen the device if it's not currently open. Otherwise,
315 * just update the physical size of the device.
316 */
317 if (dvd != NULL) {
318 if (dvd->vd_ldi_offline && dvd->vd_lh == NULL) {
319 /*
320 * If we are opening a device in its offline notify
321 * context, the LDI handle was just closed. Clean
322 * up the LDI event callbacks and free vd->vdev_tsd.
323 */
324 vdev_disk_free(vd);
325 } else {
326 ASSERT(vd->vdev_reopening);
327 goto skip_open;
328 }
329 }
330
331 /*
332 * Create vd->vdev_tsd.
333 */
334 vdev_disk_alloc(vd);
335 dvd = vd->vdev_tsd;
336
337 /*
338 * When opening a disk device, we want to preserve the user's original
339 * intent. We always want to open the device by the path the user gave
340 * us, even if it is one of multiple paths to the same device. But we
341 * also want to be able to survive disks being removed/recabled.
342 * Therefore the sequence of opening devices is:
343 *
344 * 1. Try opening the device by path. For legacy pools without the
345 * 'whole_disk' property, attempt to fix the path by appending 's0'.
346 *
347 * 2. If the devid of the device matches the stored value, return
348 * success.
349 *
350 * 3. Otherwise, the device may have moved. Try opening the device
351 * by the devid instead.
352 */
353 if (vd->vdev_devid != NULL) {
354 if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
355 &dvd->vd_minor) != 0) {
356 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
357 return (SET_ERROR(EINVAL));
358 }
359 }
360
361 error = EINVAL; /* presume failure */
362
363 if (vd->vdev_path != NULL) {
364
365 if (vd->vdev_wholedisk == -1ULL) {
366 size_t len = strlen(vd->vdev_path) + 3;
367 char *buf = kmem_alloc(len, KM_SLEEP);
368
369 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
370
371 error = ldi_open_by_name(buf, spa_mode(spa), kcred,
372 &dvd->vd_lh, zfs_li);
373 if (error == 0) {
374 spa_strfree(vd->vdev_path);
375 vd->vdev_path = buf;
376 vd->vdev_wholedisk = 1ULL;
377 } else {
378 kmem_free(buf, len);
379 }
380 }
381
382 /*
383 * If we have not yet opened the device, try to open it by the
384 * specified path.
385 */
386 if (error != 0) {
387 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
388 kcred, &dvd->vd_lh, zfs_li);
389 }
390
391 /*
392 * Compare the devid to the stored value.
393 */
394 if (error == 0 && vd->vdev_devid != NULL &&
395 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
396 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
397 error = SET_ERROR(EINVAL);
398 (void) ldi_close(dvd->vd_lh, spa_mode(spa),
399 kcred);
400 dvd->vd_lh = NULL;
401 }
402 ddi_devid_free(devid);
403 }
404
405 /*
406 * If we succeeded in opening the device, but 'vdev_wholedisk'
407 * is not yet set, then this must be a slice.
408 */
409 if (error == 0 && vd->vdev_wholedisk == -1ULL)
410 vd->vdev_wholedisk = 0;
411 }
412
413 /*
414 * If we were unable to open by path, or the devid check fails, open by
415 * devid instead.
416 */
417 if (error != 0 && vd->vdev_devid != NULL) {
418 error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
419 spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
420 }
421
422 /*
423 * If all else fails, then try opening by physical path (if available)
424 * or the logical path (if we failed due to the devid check). While not
425 * as reliable as the devid, this will give us something, and the higher
426 * level vdev validation will prevent us from opening the wrong device.
427 */
428 if (error) {
429 if (vd->vdev_devid != NULL)
430 validate_devid = B_TRUE;
431
432 if (vd->vdev_physpath != NULL &&
433 (dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV)
434 error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
435 kcred, &dvd->vd_lh, zfs_li);
436
437 /*
438 * Note that we don't support the legacy auto-wholedisk support
439 * as above. This hasn't been used in a very long time and we
440 * don't need to propagate its oddities to this edge condition.
441 */
442 if (error && vd->vdev_path != NULL)
443 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
444 kcred, &dvd->vd_lh, zfs_li);
445 }
446
447 if (error) {
448 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
449 return (error);
450 }
451
452 /*
453 * Now that the device has been successfully opened, update the devid
454 * if necessary.
455 */
456 if (validate_devid && spa_writeable(spa) &&
457 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
458 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
459 char *vd_devid;
460
461 vd_devid = ddi_devid_str_encode(devid, dvd->vd_minor);
462 zfs_dbgmsg("vdev %s: update devid from %s, "
463 "to %s", vd->vdev_path, vd->vdev_devid, vd_devid);
464 spa_strfree(vd->vdev_devid);
465 vd->vdev_devid = spa_strdup(vd_devid);
466 ddi_devid_str_free(vd_devid);
467 }
468 ddi_devid_free(devid);
469 }
470
471 /*
472 * Once a device is opened, verify that the physical device path (if
473 * available) is up to date.
474 */
475 if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
476 ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
477 char *physpath, *minorname;
478
479 physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
480 minorname = NULL;
481 if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
482 ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
483 (vd->vdev_physpath == NULL ||
484 strcmp(vd->vdev_physpath, physpath) != 0)) {
485 if (vd->vdev_physpath)
486 spa_strfree(vd->vdev_physpath);
487 (void) strlcat(physpath, ":", MAXPATHLEN);
488 (void) strlcat(physpath, minorname, MAXPATHLEN);
489 vd->vdev_physpath = spa_strdup(physpath);
490 }
491 if (minorname)
492 kmem_free(minorname, strlen(minorname) + 1);
493 kmem_free(physpath, MAXPATHLEN);
494 }
495
496 /*
497 * Register callbacks for the LDI offline event.
498 */
499 if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_OFFLINE, &ecookie) ==
500 LDI_EV_SUCCESS) {
501 lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
502 list_insert_tail(&dvd->vd_ldi_cbs, lcb);
503 (void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
504 &vdev_disk_off_callb, (void *) vd, &lcb->lcb_id);
505 }
506
507 /*
508 * Register callbacks for the LDI degrade event.
509 */
510 if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_DEGRADE, &ecookie) ==
511 LDI_EV_SUCCESS) {
512 lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
513 list_insert_tail(&dvd->vd_ldi_cbs, lcb);
514 (void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
515 &vdev_disk_dgrd_callb, (void *) vd, &lcb->lcb_id);
516 }
517 skip_open:
518 /*
519 * Determine the actual size of the device.
520 */
521 if (ldi_get_size(dvd->vd_lh, psize) != 0) {
522 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
523 return (SET_ERROR(EINVAL));
524 }
525
526 *max_psize = *psize;
527
528 /*
529 * Determine the device's minimum transfer size.
530 * If the ioctl isn't supported, assume DEV_BSIZE.
531 */
532 if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT,
533 (intptr_t)dkmext, FKIOCTL, kcred, NULL)) == 0) {
534 capacity = dkmext->dki_capacity - 1;
535 blksz = dkmext->dki_lbsize;
536 pbsize = dkmext->dki_pbsize;
537 } else if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFO,
538 (intptr_t)dkm, FKIOCTL, kcred, NULL)) == 0) {
539 VDEV_DEBUG(
540 "vdev_disk_open(\"%s\"): fallback to DKIOCGMEDIAINFO\n",
541 vd->vdev_path);
542 capacity = dkm->dki_capacity - 1;
543 blksz = dkm->dki_lbsize;
544 pbsize = blksz;
545 } else {
546 VDEV_DEBUG("vdev_disk_open(\"%s\"): "
547 "both DKIOCGMEDIAINFO{,EXT} calls failed, %d\n",
548 vd->vdev_path, error);
549 pbsize = DEV_BSIZE;
550 }
551
552 *ashift = highbit(MAX(pbsize, SPA_MINBLOCKSIZE)) - 1;
553
554 if (vd->vdev_wholedisk == 1) {
555 int wce = 1;
556
557 if (error == 0) {
558 /*
559 * If we have the capability to expand, we'd have
560 * found out via success from DKIOCGMEDIAINFO{,EXT}.
561 * Adjust max_psize upward accordingly since we know
562 * we own the whole disk now.
563 */
564 *max_psize += vdev_disk_get_space(vd, capacity, blksz);
565 zfs_dbgmsg("capacity change: vdev %s, psize %llu, "
566 "max_psize %llu", vd->vdev_path, *psize,
567 *max_psize);
568 }
569
570 /*
571 * Since we own the whole disk, try to enable disk write
572 * caching. We ignore errors because it's OK if we can't do it.
573 */
574 (void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
575 FKIOCTL, kcred, NULL);
576 }
577
578 /*
579 * Clear the nowritecache bit, so that on a vdev_reopen() we will
580 * try again.
581 */
582 vd->vdev_nowritecache = B_FALSE;
583
584 return (0);
585 }
586
587 static void
588 vdev_disk_close(vdev_t *vd)
589 {
590 vdev_disk_t *dvd = vd->vdev_tsd;
591
592 if (vd->vdev_reopening || dvd == NULL)
593 return;
594
595 if (dvd->vd_minor != NULL) {
596 ddi_devid_str_free(dvd->vd_minor);
597 dvd->vd_minor = NULL;
598 }
599
600 if (dvd->vd_devid != NULL) {
601 ddi_devid_free(dvd->vd_devid);
602 dvd->vd_devid = NULL;
603 }
604
605 if (dvd->vd_lh != NULL) {
606 (void) ldi_close(dvd->vd_lh, spa_mode(vd->vdev_spa), kcred);
607 dvd->vd_lh = NULL;
608 }
609
610 vd->vdev_delayed_close = B_FALSE;
611 /*
612 * If we closed the LDI handle due to an offline notify from LDI,
613 * don't free vd->vdev_tsd or unregister the callbacks here;
614 * the offline finalize callback or a reopen will take care of it.
615 */
616 if (dvd->vd_ldi_offline)
617 return;
618
619 vdev_disk_free(vd);
620 }
621
622 int
623 vdev_disk_physio(vdev_t *vd, caddr_t data,
624 size_t size, uint64_t offset, int flags, boolean_t isdump)
625 {
626 vdev_disk_t *dvd = vd->vdev_tsd;
627
628 /*
629 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
630 * Nothing to be done here but return failure.
631 */
632 if (dvd == NULL || (dvd->vd_ldi_offline && dvd->vd_lh == NULL))
633 return (EIO);
634
635 ASSERT(vd->vdev_ops == &vdev_disk_ops);
636
637 /*
638 * If in the context of an active crash dump, use the ldi_dump(9F)
639 * call instead of ldi_strategy(9F) as usual.
640 */
641 if (isdump) {
642 ASSERT3P(dvd, !=, NULL);
643 return (ldi_dump(dvd->vd_lh, data, lbtodb(offset),
644 lbtodb(size)));
645 }
646
647 return (vdev_disk_ldi_physio(dvd->vd_lh, data, size, offset, flags));
648 }
649
650 int
651 vdev_disk_ldi_physio(ldi_handle_t vd_lh, caddr_t data,
652 size_t size, uint64_t offset, int flags)
653 {
654 buf_t *bp;
655 int error = 0;
656
657 if (vd_lh == NULL)
658 return (SET_ERROR(EINVAL));
659
660 ASSERT(flags & B_READ || flags & B_WRITE);
661
662 bp = getrbuf(KM_SLEEP);
663 bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST;
664 bp->b_bcount = size;
665 bp->b_un.b_addr = (void *)data;
666 bp->b_lblkno = lbtodb(offset);
667 bp->b_bufsize = size;
668
669 error = ldi_strategy(vd_lh, bp);
670 ASSERT(error == 0);
671 if ((error = biowait(bp)) == 0 && bp->b_resid != 0)
672 error = SET_ERROR(EIO);
673 freerbuf(bp);
674
675 return (error);
676 }
677
678 static void
679 vdev_disk_io_intr(buf_t *bp)
680 {
681 vdev_buf_t *vb = (vdev_buf_t *)bp;
682 zio_t *zio = vb->vb_io;
683
684 /*
685 * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO.
686 * Rather than teach the rest of the stack about other error
687 * possibilities (EFAULT, etc), we normalize the error value here.
688 */
689 zio->io_error = (geterror(bp) != 0 ? EIO : 0);
690
691 if (zio->io_error == 0 && bp->b_resid != 0)
692 zio->io_error = SET_ERROR(EIO);
693
694 kmem_free(vb, sizeof (vdev_buf_t));
695
696 zio_interrupt(zio);
697 }
698
699 static void
700 vdev_disk_ioctl_free(zio_t *zio)
701 {
702 kmem_free(zio->io_vsd, sizeof (struct dk_callback));
703 }
704
705 static const zio_vsd_ops_t vdev_disk_vsd_ops = {
706 vdev_disk_ioctl_free,
707 zio_vsd_default_cksum_report
708 };
709
710 static void
711 vdev_disk_ioctl_done(void *zio_arg, int error)
712 {
713 zio_t *zio = zio_arg;
714
715 zio->io_error = error;
716
717 zio_interrupt(zio);
718 }
719
720 static int
721 vdev_disk_io_start(zio_t *zio)
722 {
723 vdev_t *vd = zio->io_vd;
724 vdev_disk_t *dvd = vd->vdev_tsd;
725 vdev_buf_t *vb;
726 struct dk_callback *dkc;
727 buf_t *bp;
728 int error;
729
730 /*
731 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
732 * Nothing to be done here but return failure.
733 */
734 if (dvd == NULL || (dvd->vd_ldi_offline && dvd->vd_lh == NULL)) {
735 zio->io_error = ENXIO;
736 return (ZIO_PIPELINE_CONTINUE);
737 }
738
739 if (zio->io_type == ZIO_TYPE_IOCTL) {
740 /* XXPOLICY */
741 if (!vdev_readable(vd)) {
742 zio->io_error = SET_ERROR(ENXIO);
743 return (ZIO_PIPELINE_CONTINUE);
744 }
745
746 switch (zio->io_cmd) {
747
748 case DKIOCFLUSHWRITECACHE:
749
750 if (zfs_nocacheflush)
751 break;
752
753 if (vd->vdev_nowritecache) {
754 zio->io_error = SET_ERROR(ENOTSUP);
755 break;
756 }
757
758 zio->io_vsd = dkc = kmem_alloc(sizeof (*dkc), KM_SLEEP);
759 zio->io_vsd_ops = &vdev_disk_vsd_ops;
760
761 dkc->dkc_callback = vdev_disk_ioctl_done;
762 dkc->dkc_flag = FLUSH_VOLATILE;
763 dkc->dkc_cookie = zio;
764
765 error = ldi_ioctl(dvd->vd_lh, zio->io_cmd,
766 (uintptr_t)dkc, FKIOCTL, kcred, NULL);
767
768 if (error == 0) {
769 /*
770 * The ioctl will be done asychronously,
771 * and will call vdev_disk_ioctl_done()
772 * upon completion.
773 */
774 return (ZIO_PIPELINE_STOP);
775 }
776
777 if (error == ENOTSUP || error == ENOTTY) {
778 /*
779 * If we get ENOTSUP or ENOTTY, we know that
780 * no future attempts will ever succeed.
781 * In this case we set a persistent bit so
782 * that we don't bother with the ioctl in the
783 * future.
784 */
785 vd->vdev_nowritecache = B_TRUE;
786 }
787 zio->io_error = error;
788
789 break;
790
791 default:
792 zio->io_error = SET_ERROR(ENOTSUP);
793 }
794
795 return (ZIO_PIPELINE_CONTINUE);
796 }
797
798 vb = kmem_alloc(sizeof (vdev_buf_t), KM_SLEEP);
799
800 vb->vb_io = zio;
801 bp = &vb->vb_buf;
802
803 bioinit(bp);
804 bp->b_flags = B_BUSY | B_NOCACHE |
805 (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
806 if (!(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))
807 bp->b_flags |= B_FAILFAST;
808 bp->b_bcount = zio->io_size;
809 bp->b_un.b_addr = zio->io_data;
810 bp->b_lblkno = lbtodb(zio->io_offset);
811 bp->b_bufsize = zio->io_size;
812 bp->b_iodone = (int (*)())vdev_disk_io_intr;
813
814 /* ldi_strategy() will return non-zero only on programming errors */
815 VERIFY(ldi_strategy(dvd->vd_lh, bp) == 0);
816
817 return (ZIO_PIPELINE_STOP);
818 }
819
820 static void
821 vdev_disk_io_done(zio_t *zio)
822 {
823 vdev_t *vd = zio->io_vd;
824
825 /*
826 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
827 * the device has been removed. If this is the case, then we trigger an
828 * asynchronous removal of the device. Otherwise, probe the device and
829 * make sure it's still accessible.
830 */
831 if (zio->io_error == EIO && !vd->vdev_remove_wanted) {
832 vdev_disk_t *dvd = vd->vdev_tsd;
833 int state = DKIO_NONE;
834
835 if (ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state,
836 FKIOCTL, kcred, NULL) == 0 && state != DKIO_INSERTED) {
837 /*
838 * We post the resource as soon as possible, instead of
839 * when the async removal actually happens, because the
840 * DE is using this information to discard previous I/O
841 * errors.
842 */
843 zfs_post_remove(zio->io_spa, vd);
844 vd->vdev_remove_wanted = B_TRUE;
845 spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
846 } else if (!vd->vdev_delayed_close) {
847 vd->vdev_delayed_close = B_TRUE;
848 }
849 }
850 }
851
852 vdev_ops_t vdev_disk_ops = {
853 vdev_disk_open,
854 vdev_disk_close,
855 vdev_default_asize,
856 vdev_disk_io_start,
857 vdev_disk_io_done,
858 NULL,
859 vdev_disk_hold,
860 vdev_disk_rele,
861 VDEV_TYPE_DISK, /* name of this vdev type */
862 B_TRUE /* leaf vdev */
863 };
864
865 /*
866 * Given the root disk device devid or pathname, read the label from
867 * the device, and construct a configuration nvlist.
868 */
869 int
870 vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config)
871 {
872 ldi_handle_t vd_lh;
873 vdev_label_t *label;
874 uint64_t s, size;
875 int l;
876 ddi_devid_t tmpdevid;
877 int error = -1;
878 char *minor_name;
879
880 /*
881 * Read the device label and build the nvlist.
882 */
883 if (devid != NULL && ddi_devid_str_decode(devid, &tmpdevid,
884 &minor_name) == 0) {
885 error = ldi_open_by_devid(tmpdevid, minor_name,
886 FREAD, kcred, &vd_lh, zfs_li);
887 ddi_devid_free(tmpdevid);
888 ddi_devid_str_free(minor_name);
889 }
890
891 if (error && (error = ldi_open_by_name(devpath, FREAD, kcred, &vd_lh,
892 zfs_li)))
893 return (error);
894
895 if (ldi_get_size(vd_lh, &s)) {
896 (void) ldi_close(vd_lh, FREAD, kcred);
897 return (SET_ERROR(EIO));
898 }
899
900 size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t);
901 label = kmem_alloc(sizeof (vdev_label_t), KM_SLEEP);
902
903 *config = NULL;
904 for (l = 0; l < VDEV_LABELS; l++) {
905 uint64_t offset, state, txg = 0;
906
907 /* read vdev label */
908 offset = vdev_label_offset(size, l, 0);
909 if (vdev_disk_ldi_physio(vd_lh, (caddr_t)label,
910 VDEV_SKIP_SIZE + VDEV_PHYS_SIZE, offset, B_READ) != 0)
911 continue;
912
913 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
914 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0) {
915 *config = NULL;
916 continue;
917 }
918
919 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
920 &state) != 0 || state >= POOL_STATE_DESTROYED) {
921 nvlist_free(*config);
922 *config = NULL;
923 continue;
924 }
925
926 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
927 &txg) != 0 || txg == 0) {
928 nvlist_free(*config);
929 *config = NULL;
930 continue;
931 }
932
933 break;
934 }
935
936 kmem_free(label, sizeof (vdev_label_t));
937 (void) ldi_close(vd_lh, FREAD, kcred);
938 if (*config == NULL)
939 error = SET_ERROR(EIDRM);
940
941 return (error);
942 }