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 by Delphix. All rights reserved.
24 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
25 */
26
27 #include <sys/zfs_context.h>
28 #include <sys/zfs_zone.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
46 vdev_disk_hold(vdev_t *vd)
47 {
48 ddi_devid_t devid;
49 char *minor;
50
51 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
52
53 /*
54 * We must have a pathname, and it must be absolute.
55 */
56 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/')
57 return;
58
59 /*
60 * Only prefetch path and devid info if the device has
61 * never been opened.
62 */
63 if (vd->vdev_tsd != NULL)
64 return;
65
66 if (vd->vdev_wholedisk == -1ULL) {
67 size_t len = strlen(vd->vdev_path) + 3;
68 char *buf = kmem_alloc(len, KM_SLEEP);
69
70 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
71
72 (void) ldi_vp_from_name(buf, &vd->vdev_name_vp);
73 kmem_free(buf, len);
74 }
75
76 if (vd->vdev_name_vp == NULL)
77 (void) ldi_vp_from_name(vd->vdev_path, &vd->vdev_name_vp);
78
79 if (vd->vdev_devid != NULL &&
80 ddi_devid_str_decode(vd->vdev_devid, &devid, &minor) == 0) {
81 (void) ldi_vp_from_devid(devid, minor, &vd->vdev_devid_vp);
82 ddi_devid_str_free(minor);
83 ddi_devid_free(devid);
84 }
85 }
86
87 static void
88 vdev_disk_rele(vdev_t *vd)
89 {
90 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
91
92 if (vd->vdev_name_vp) {
93 VN_RELE_ASYNC(vd->vdev_name_vp,
94 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
95 vd->vdev_name_vp = NULL;
96 }
97 if (vd->vdev_devid_vp) {
98 VN_RELE_ASYNC(vd->vdev_devid_vp,
99 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
100 vd->vdev_devid_vp = NULL;
101 }
102 }
103
104 static uint64_t
105 vdev_disk_get_space(vdev_t *vd, uint64_t capacity, uint_t blksz)
106 {
107 ASSERT(vd->vdev_wholedisk);
108
109 vdev_disk_t *dvd = vd->vdev_tsd;
110 dk_efi_t dk_ioc;
111 efi_gpt_t *efi;
112 uint64_t avail_space = 0;
113 int efisize = EFI_LABEL_SIZE * 2;
114
115 dk_ioc.dki_data = kmem_alloc(efisize, KM_SLEEP);
116 dk_ioc.dki_lba = 1;
117 dk_ioc.dki_length = efisize;
118 dk_ioc.dki_data_64 = (uint64_t)(uintptr_t)dk_ioc.dki_data;
119 efi = dk_ioc.dki_data;
120
121 if (ldi_ioctl(dvd->vd_lh, DKIOCGETEFI, (intptr_t)&dk_ioc,
122 FKIOCTL, kcred, NULL) == 0) {
123 uint64_t efi_altern_lba = LE_64(efi->efi_gpt_AlternateLBA);
124
125 zfs_dbgmsg("vdev %s, capacity %llu, altern lba %llu",
126 vd->vdev_path, capacity, efi_altern_lba);
127 if (capacity > efi_altern_lba)
128 avail_space = (capacity - efi_altern_lba) * blksz;
129 }
130 kmem_free(dk_ioc.dki_data, efisize);
131 return (avail_space);
132 }
133
134 static int
135 vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
136 uint64_t *ashift)
137 {
138 spa_t *spa = vd->vdev_spa;
139 vdev_disk_t *dvd;
140 struct dk_minfo_ext dkmext;
141 int error;
142 dev_t dev;
143 int otyp;
144
145 /*
146 * We must have a pathname, and it must be absolute.
147 */
148 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
149 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
150 return (EINVAL);
151 }
152
153 /*
154 * Reopen the device if it's not currently open. Otherwise,
155 * just update the physical size of the device.
156 */
157 if (vd->vdev_tsd != NULL) {
158 ASSERT(vd->vdev_reopening);
159 dvd = vd->vdev_tsd;
160 goto skip_open;
161 }
162
163 dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
164
165 /*
166 * When opening a disk device, we want to preserve the user's original
167 * intent. We always want to open the device by the path the user gave
168 * us, even if it is one of multiple paths to the same device. But we
169 * also want to be able to survive disks being removed/recabled.
170 * Therefore the sequence of opening devices is:
171 *
172 * 1. Try opening the device by path. For legacy pools without the
173 * 'whole_disk' property, attempt to fix the path by appending 's0'.
174 *
175 * 2. If the devid of the device matches the stored value, return
176 * success.
177 *
178 * 3. Otherwise, the device may have moved. Try opening the device
179 * by the devid instead.
180 */
181 if (vd->vdev_devid != NULL) {
182 if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
183 &dvd->vd_minor) != 0) {
184 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
185 return (EINVAL);
186 }
187 }
188
189 error = EINVAL; /* presume failure */
190
191 if (vd->vdev_path != NULL) {
192 ddi_devid_t devid;
193
194 if (vd->vdev_wholedisk == -1ULL) {
195 size_t len = strlen(vd->vdev_path) + 3;
196 char *buf = kmem_alloc(len, KM_SLEEP);
197 ldi_handle_t lh;
198
199 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
200
201 if (ldi_open_by_name(buf, spa_mode(spa), kcred,
202 &lh, zfs_li) == 0) {
203 spa_strfree(vd->vdev_path);
204 vd->vdev_path = buf;
205 vd->vdev_wholedisk = 1ULL;
206 (void) ldi_close(lh, spa_mode(spa), kcred);
207 } else {
208 kmem_free(buf, len);
209 }
210 }
211
212 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa), kcred,
213 &dvd->vd_lh, zfs_li);
214
215 /*
216 * Compare the devid to the stored value.
217 */
218 if (error == 0 && vd->vdev_devid != NULL &&
219 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
220 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
221 error = EINVAL;
222 (void) ldi_close(dvd->vd_lh, spa_mode(spa),
223 kcred);
224 dvd->vd_lh = NULL;
225 }
226 ddi_devid_free(devid);
227 }
228
229 /*
230 * If we succeeded in opening the device, but 'vdev_wholedisk'
231 * is not yet set, then this must be a slice.
232 */
233 if (error == 0 && vd->vdev_wholedisk == -1ULL)
234 vd->vdev_wholedisk = 0;
235 }
236
237 /*
238 * If we were unable to open by path, or the devid check fails, open by
239 * devid instead.
240 */
241 if (error != 0 && vd->vdev_devid != NULL)
242 error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
243 spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
244
245 /*
246 * If all else fails, then try opening by physical path (if available)
247 * or the logical path (if we failed due to the devid check). While not
248 * as reliable as the devid, this will give us something, and the higher
249 * level vdev validation will prevent us from opening the wrong device.
250 */
251 if (error) {
252 if (vd->vdev_physpath != NULL &&
253 (dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV)
254 error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
255 kcred, &dvd->vd_lh, zfs_li);
256
257 /*
258 * Note that we don't support the legacy auto-wholedisk support
259 * as above. This hasn't been used in a very long time and we
260 * don't need to propagate its oddities to this edge condition.
261 */
262 if (error && vd->vdev_path != NULL)
263 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
264 kcred, &dvd->vd_lh, zfs_li);
265 }
266
267 if (error) {
268 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
269 return (error);
270 }
271
272 /*
273 * Once a device is opened, verify that the physical device path (if
274 * available) is up to date.
275 */
276 if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
277 ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
278 char *physpath, *minorname;
279
280 physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
281 minorname = NULL;
282 if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
283 ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
284 (vd->vdev_physpath == NULL ||
285 strcmp(vd->vdev_physpath, physpath) != 0)) {
286 if (vd->vdev_physpath)
287 spa_strfree(vd->vdev_physpath);
288 (void) strlcat(physpath, ":", MAXPATHLEN);
289 (void) strlcat(physpath, minorname, MAXPATHLEN);
290 vd->vdev_physpath = spa_strdup(physpath);
291 }
292 if (minorname)
293 kmem_free(minorname, strlen(minorname) + 1);
294 kmem_free(physpath, MAXPATHLEN);
295 }
296
297 skip_open:
298 /*
299 * Determine the actual size of the device.
300 */
301 if (ldi_get_size(dvd->vd_lh, psize) != 0) {
302 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
303 return (EINVAL);
304 }
305
306 /*
307 * Determine the device's minimum transfer size.
308 * If the ioctl isn't supported, assume DEV_BSIZE.
309 */
310 if (ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT, (intptr_t)&dkmext,
311 FKIOCTL, kcred, NULL) != 0)
312 dkmext.dki_pbsize = DEV_BSIZE;
313
314 *ashift = highbit(MAX(dkmext.dki_pbsize, SPA_MINBLOCKSIZE)) - 1;
315
316 if (vd->vdev_wholedisk == 1) {
317 uint64_t capacity = dkmext.dki_capacity - 1;
318 uint64_t blksz = dkmext.dki_lbsize;
319 int wce = 1;
320
321 /*
322 * If we own the whole disk, try to enable disk write caching.
323 * We ignore errors because it's OK if we can't do it.
324 */
325 (void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
326 FKIOCTL, kcred, NULL);
327
328 *max_psize = *psize + vdev_disk_get_space(vd, capacity, blksz);
329 zfs_dbgmsg("capacity change: vdev %s, psize %llu, "
330 "max_psize %llu", vd->vdev_path, *psize, *max_psize);
331 } else {
332 *max_psize = *psize;
333 }
334
335 /*
336 * Clear the nowritecache bit, so that on a vdev_reopen() we will
337 * try again.
338 */
339 vd->vdev_nowritecache = B_FALSE;
340
341 return (0);
342 }
343
344 static void
345 vdev_disk_close(vdev_t *vd)
346 {
347 vdev_disk_t *dvd = vd->vdev_tsd;
348
349 if (vd->vdev_reopening || dvd == NULL)
350 return;
351
352 if (dvd->vd_minor != NULL)
353 ddi_devid_str_free(dvd->vd_minor);
354
355 if (dvd->vd_devid != NULL)
356 ddi_devid_free(dvd->vd_devid);
357
358 if (dvd->vd_lh != NULL)
359 (void) ldi_close(dvd->vd_lh, spa_mode(vd->vdev_spa), kcred);
360
361 vd->vdev_delayed_close = B_FALSE;
362 kmem_free(dvd, sizeof (vdev_disk_t));
363 vd->vdev_tsd = NULL;
364 }
365
366 int
367 vdev_disk_physio(vdev_t *vd, caddr_t data,
368 size_t size, uint64_t offset, int flags)
369 {
370 vdev_disk_t *dvd = vd->vdev_tsd;
371
372 /*
373 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
374 * Nothing to be done here but return failure.
375 */
376 if (dvd == NULL)
377 return (EIO);
378
379 ASSERT(vd->vdev_ops == &vdev_disk_ops);
380 return (vdev_disk_ldi_physio(dvd->vd_lh, data, size, offset, flags));
381 }
382
383 int
384 vdev_disk_ldi_physio(ldi_handle_t vd_lh, caddr_t data,
385 size_t size, uint64_t offset, int flags)
386 {
387 buf_t *bp;
388 int error = 0;
389
390 if (vd_lh == NULL)
391 return (EINVAL);
392
393 ASSERT(flags & B_READ || flags & B_WRITE);
394
395 bp = getrbuf(KM_SLEEP);
396 bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST;
397 bp->b_bcount = size;
398 bp->b_un.b_addr = (void *)data;
399 bp->b_lblkno = lbtodb(offset);
400 bp->b_bufsize = size;
401
402 error = ldi_strategy(vd_lh, bp);
403 ASSERT(error == 0);
404 if ((error = biowait(bp)) == 0 && bp->b_resid != 0)
405 error = EIO;
406 freerbuf(bp);
407
408 return (error);
409 }
410
411 static void
412 vdev_disk_io_intr(buf_t *bp)
413 {
414 vdev_buf_t *vb = (vdev_buf_t *)bp;
415 zio_t *zio = vb->vb_io;
416
417 /*
418 * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO.
419 * Rather than teach the rest of the stack about other error
420 * possibilities (EFAULT, etc), we normalize the error value here.
421 */
422 zio->io_error = (geterror(bp) != 0 ? EIO : 0);
423
424 if (zio->io_error == 0 && bp->b_resid != 0)
425 zio->io_error = EIO;
426
427 kmem_free(vb, sizeof (vdev_buf_t));
428
429 zio_interrupt(zio);
430 }
431
432 static void
433 vdev_disk_ioctl_free(zio_t *zio)
434 {
435 kmem_free(zio->io_vsd, sizeof (struct dk_callback));
436 }
437
438 static const zio_vsd_ops_t vdev_disk_vsd_ops = {
439 vdev_disk_ioctl_free,
440 zio_vsd_default_cksum_report
441 };
442
443 static void
444 vdev_disk_ioctl_done(void *zio_arg, int error)
445 {
446 zio_t *zio = zio_arg;
447
448 zio->io_error = error;
449
450 zio_interrupt(zio);
451 }
452
453 static int
454 vdev_disk_io_start(zio_t *zio)
455 {
456 vdev_t *vd = zio->io_vd;
457 vdev_disk_t *dvd = vd->vdev_tsd;
458 vdev_buf_t *vb;
459 struct dk_callback *dkc;
460 buf_t *bp;
461 int error;
462
463 if (zio->io_type == ZIO_TYPE_IOCTL) {
464 /* XXPOLICY */
465 if (!vdev_readable(vd)) {
466 zio->io_error = ENXIO;
467 return (ZIO_PIPELINE_CONTINUE);
468 }
469
470 switch (zio->io_cmd) {
471
472 case DKIOCFLUSHWRITECACHE:
473
474 if (zfs_nocacheflush)
475 break;
476
477 if (vd->vdev_nowritecache) {
478 zio->io_error = ENOTSUP;
479 break;
480 }
481
482 zio->io_vsd = dkc = kmem_alloc(sizeof (*dkc), KM_SLEEP);
483 zio->io_vsd_ops = &vdev_disk_vsd_ops;
484
485 dkc->dkc_callback = vdev_disk_ioctl_done;
486 dkc->dkc_flag = FLUSH_VOLATILE;
487 dkc->dkc_cookie = zio;
488
489 error = ldi_ioctl(dvd->vd_lh, zio->io_cmd,
490 (uintptr_t)dkc, FKIOCTL, kcred, NULL);
491
492 if (error == 0) {
493 /*
494 * The ioctl will be done asychronously,
495 * and will call vdev_disk_ioctl_done()
496 * upon completion.
497 */
498 return (ZIO_PIPELINE_STOP);
499 }
500
501 if (error == ENOTSUP || error == ENOTTY) {
502 /*
503 * If we get ENOTSUP or ENOTTY, we know that
504 * no future attempts will ever succeed.
505 * In this case we set a persistent bit so
506 * that we don't bother with the ioctl in the
507 * future.
508 */
509 vd->vdev_nowritecache = B_TRUE;
510 }
511 zio->io_error = error;
512
513 break;
514
515 default:
516 zio->io_error = ENOTSUP;
517 }
518
519 return (ZIO_PIPELINE_CONTINUE);
520 }
521
522 vb = kmem_alloc(sizeof (vdev_buf_t), KM_SLEEP);
523
524 vb->vb_io = zio;
525 bp = &vb->vb_buf;
526
527 bioinit(bp);
528 bp->b_flags = B_BUSY | B_NOCACHE |
529 (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
530 if (!(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))
531 bp->b_flags |= B_FAILFAST;
532 bp->b_bcount = zio->io_size;
533 bp->b_un.b_addr = zio->io_data;
534 bp->b_lblkno = lbtodb(zio->io_offset);
535 bp->b_bufsize = zio->io_size;
536 bp->b_iodone = (int (*)())vdev_disk_io_intr;
537
538 zfs_zone_zio_start(zio);
539
540 /* ldi_strategy() will return non-zero only on programming errors */
541 VERIFY(ldi_strategy(dvd->vd_lh, bp) == 0);
542
543 return (ZIO_PIPELINE_STOP);
544 }
545
546 static void
547 vdev_disk_io_done(zio_t *zio)
548 {
549 vdev_t *vd = zio->io_vd;
550
551 zfs_zone_zio_done(zio);
552
553 /*
554 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
555 * the device has been removed. If this is the case, then we trigger an
556 * asynchronous removal of the device. Otherwise, probe the device and
557 * make sure it's still accessible.
558 */
559 if (zio->io_error == EIO && !vd->vdev_remove_wanted) {
560 vdev_disk_t *dvd = vd->vdev_tsd;
561 int state = DKIO_NONE;
562
563 if (ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state,
564 FKIOCTL, kcred, NULL) == 0 && state != DKIO_INSERTED) {
565 /*
566 * We post the resource as soon as possible, instead of
567 * when the async removal actually happens, because the
568 * DE is using this information to discard previous I/O
569 * errors.
570 */
571 zfs_post_remove(zio->io_spa, vd);
572 vd->vdev_remove_wanted = B_TRUE;
573 spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
574 } else if (!vd->vdev_delayed_close) {
575 vd->vdev_delayed_close = B_TRUE;
576 }
577 }
578 }
579
580 vdev_ops_t vdev_disk_ops = {
581 vdev_disk_open,
582 vdev_disk_close,
583 vdev_default_asize,
584 vdev_disk_io_start,
585 vdev_disk_io_done,
586 NULL,
587 vdev_disk_hold,
588 vdev_disk_rele,
589 VDEV_TYPE_DISK, /* name of this vdev type */
590 B_TRUE /* leaf vdev */
591 };
592
593 /*
594 * Given the root disk device devid or pathname, read the label from
595 * the device, and construct a configuration nvlist.
596 */
597 int
598 vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config)
599 {
600 ldi_handle_t vd_lh;
601 vdev_label_t *label;
602 uint64_t s, size;
603 int l;
604 ddi_devid_t tmpdevid;
605 int error = -1;
606 char *minor_name;
607
608 /*
609 * Read the device label and build the nvlist.
610 */
611 if (devid != NULL && ddi_devid_str_decode(devid, &tmpdevid,
612 &minor_name) == 0) {
613 error = ldi_open_by_devid(tmpdevid, minor_name,
614 FREAD, kcred, &vd_lh, zfs_li);
615 ddi_devid_free(tmpdevid);
616 ddi_devid_str_free(minor_name);
617 }
618
619 if (error && (error = ldi_open_by_name(devpath, FREAD, kcred, &vd_lh,
620 zfs_li)))
621 return (error);
622
623 if (ldi_get_size(vd_lh, &s)) {
624 (void) ldi_close(vd_lh, FREAD, kcred);
625 return (EIO);
626 }
627
628 size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t);
629 label = kmem_alloc(sizeof (vdev_label_t), KM_SLEEP);
630
631 *config = NULL;
632 for (l = 0; l < VDEV_LABELS; l++) {
633 uint64_t offset, state, txg = 0;
634
635 /* read vdev label */
636 offset = vdev_label_offset(size, l, 0);
637 if (vdev_disk_ldi_physio(vd_lh, (caddr_t)label,
638 VDEV_SKIP_SIZE + VDEV_PHYS_SIZE, offset, B_READ) != 0)
639 continue;
640
641 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
642 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0) {
643 *config = NULL;
644 continue;
645 }
646
647 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
648 &state) != 0 || state >= POOL_STATE_DESTROYED) {
649 nvlist_free(*config);
650 *config = NULL;
651 continue;
652 }
653
654 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
655 &txg) != 0 || txg == 0) {
656 nvlist_free(*config);
657 *config = NULL;
658 continue;
659 }
660
661 break;
662 }
663
664 kmem_free(label, sizeof (vdev_label_t));
665 (void) ldi_close(vd_lh, FREAD, kcred);
666 if (*config == NULL)
667 error = EIDRM;
668
669 return (error);
670 }