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 */
24
25#include <sys/zfs_context.h>
26#include <sys/spa_impl.h>
27#include <sys/refcount.h>
28#include <sys/vdev_disk.h>
29#include <sys/vdev_impl.h>
30#include <sys/fs/zfs.h>
31#include <sys/zio.h>
32#include <sys/sunldi.h>
33#include <sys/fm/fs/zfs.h>
34
35/*
36 * Virtual device vector for disks.
37 */
38
39extern ldi_ident_t zfs_li;
40
41typedef struct vdev_disk_buf {
42 buf_t vdb_buf;
43 zio_t *vdb_io;
44} vdev_disk_buf_t;
45
46static void
47vdev_disk_hold(vdev_t *vd)
48{
49 ddi_devid_t devid;
50 char *minor;
51
52 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
53
54 /*
55 * We must have a pathname, and it must be absolute.
56 */
57 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/')
58 return;
59
60 /*
61 * Only prefetch path and devid info if the device has
62 * never been opened.
63 */
64 if (vd->vdev_tsd != NULL)
65 return;
66
67 if (vd->vdev_wholedisk == -1ULL) {
68 size_t len = strlen(vd->vdev_path) + 3;
69 char *buf = kmem_alloc(len, KM_SLEEP);
70
71 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
72
73 (void) ldi_vp_from_name(buf, &vd->vdev_name_vp);
74 kmem_free(buf, len);
75 }
76
77 if (vd->vdev_name_vp == NULL)
78 (void) ldi_vp_from_name(vd->vdev_path, &vd->vdev_name_vp);
79
80 if (vd->vdev_devid != NULL &&
81 ddi_devid_str_decode(vd->vdev_devid, &devid, &minor) == 0) {
82 (void) ldi_vp_from_devid(devid, minor, &vd->vdev_devid_vp);
83 ddi_devid_str_free(minor);
84 ddi_devid_free(devid);
85 }
86}
87
88static void
89vdev_disk_rele(vdev_t *vd)
90{
91 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
92
93 if (vd->vdev_name_vp) {
94 VN_RELE_ASYNC(vd->vdev_name_vp,
95 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
96 vd->vdev_name_vp = NULL;
97 }
98 if (vd->vdev_devid_vp) {
99 VN_RELE_ASYNC(vd->vdev_devid_vp,
100 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
101 vd->vdev_devid_vp = NULL;
102 }
103}
104
105static int
106vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *ashift)
107{
108 spa_t *spa = vd->vdev_spa;
109 vdev_disk_t *dvd;
110 struct dk_minfo_ext dkmext;
111 int error;
112 dev_t dev;
113 int otyp;
114
115 /*
116 * We must have a pathname, and it must be absolute.
117 */
118 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
119 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
120 return (EINVAL);
121 }
122
123 /*
124 * Reopen the device if it's not currently open. Otherwise,
125 * just update the physical size of the device.
126 */
127 if (vd->vdev_tsd != NULL) {
128 ASSERT(vd->vdev_reopening);
129 dvd = vd->vdev_tsd;
130 goto skip_open;
131 }
132
133 dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
134
135 /*
136 * When opening a disk device, we want to preserve the user's original
137 * intent. We always want to open the device by the path the user gave
138 * us, even if it is one of multiple paths to the save device. But we
139 * also want to be able to survive disks being removed/recabled.
140 * Therefore the sequence of opening devices is:
141 *
142 * 1. Try opening the device by path. For legacy pools without the
143 * 'whole_disk' property, attempt to fix the path by appending 's0'.
144 *
145 * 2. If the devid of the device matches the stored value, return
146 * success.
147 *
148 * 3. Otherwise, the device may have moved. Try opening the device
149 * by the devid instead.
150 */
151 if (vd->vdev_devid != NULL) {
152 if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
153 &dvd->vd_minor) != 0) {
154 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
155 return (EINVAL);
156 }
157 }
158
159 error = EINVAL; /* presume failure */
160
161 if (vd->vdev_path != NULL) {
162 ddi_devid_t devid;
163
164 if (vd->vdev_wholedisk == -1ULL) {
165 size_t len = strlen(vd->vdev_path) + 3;
166 char *buf = kmem_alloc(len, KM_SLEEP);
167 ldi_handle_t lh;
168
169 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
170
171 if (ldi_open_by_name(buf, spa_mode(spa), kcred,
172 &lh, zfs_li) == 0) {
173 spa_strfree(vd->vdev_path);
174 vd->vdev_path = buf;
175 vd->vdev_wholedisk = 1ULL;
176 (void) ldi_close(lh, spa_mode(spa), kcred);
177 } else {
178 kmem_free(buf, len);
179 }
180 }
181
182 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa), kcred,
183 &dvd->vd_lh, zfs_li);
184
185 /*
186 * Compare the devid to the stored value.
187 */
188 if (error == 0 && vd->vdev_devid != NULL &&
189 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
190 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
191 error = EINVAL;
192 (void) ldi_close(dvd->vd_lh, spa_mode(spa),
193 kcred);
194 dvd->vd_lh = NULL;
195 }
196 ddi_devid_free(devid);
197 }
198
199 /*
200 * If we succeeded in opening the device, but 'vdev_wholedisk'
201 * is not yet set, then this must be a slice.
202 */
203 if (error == 0 && vd->vdev_wholedisk == -1ULL)
204 vd->vdev_wholedisk = 0;
205 }
206
207 /*
208 * If we were unable to open by path, or the devid check fails, open by
209 * devid instead.
210 */
211 if (error != 0 && vd->vdev_devid != NULL)
212 error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
213 spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
214
215 /*
216 * If all else fails, then try opening by physical path (if available)
217 * or the logical path (if we failed due to the devid check). While not
218 * as reliable as the devid, this will give us something, and the higher
219 * level vdev validation will prevent us from opening the wrong device.
220 */
221 if (error) {
222 if (vd->vdev_physpath != NULL &&
223 (dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV)
224 error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
225 kcred, &dvd->vd_lh, zfs_li);
226
227 /*
228 * Note that we don't support the legacy auto-wholedisk support
229 * as above. This hasn't been used in a very long time and we
230 * don't need to propagate its oddities to this edge condition.
231 */
232 if (error && vd->vdev_path != NULL)
233 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
234 kcred, &dvd->vd_lh, zfs_li);
235 }
236
237 if (error) {
238 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
239 return (error);
240 }
241
242 /*
243 * Once a device is opened, verify that the physical device path (if
244 * available) is up to date.
245 */
246 if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
247 ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
248 char *physpath, *minorname;
249
250 physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
251 minorname = NULL;
252 if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
253 ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
254 (vd->vdev_physpath == NULL ||
255 strcmp(vd->vdev_physpath, physpath) != 0)) {
256 if (vd->vdev_physpath)
257 spa_strfree(vd->vdev_physpath);
258 (void) strlcat(physpath, ":", MAXPATHLEN);
259 (void) strlcat(physpath, minorname, MAXPATHLEN);
260 vd->vdev_physpath = spa_strdup(physpath);
261 }
262 if (minorname)
263 kmem_free(minorname, strlen(minorname) + 1);
264 kmem_free(physpath, MAXPATHLEN);
265 }
266
267skip_open:
268 /*
269 * Determine the actual size of the device.
270 */
271 if (ldi_get_size(dvd->vd_lh, psize) != 0) {
272 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
273 return (EINVAL);
274 }
275
276 /*
277 * If we own the whole disk, try to enable disk write caching.
278 * We ignore errors because it's OK if we can't do it.
279 */
280 if (vd->vdev_wholedisk == 1) {
281 int wce = 1;
282 (void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
283 FKIOCTL, kcred, NULL);
284 }
285
286 /*
287 * Determine the device's minimum transfer size.
288 * If the ioctl isn't supported, assume DEV_BSIZE.
289 */
290 if (ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT, (intptr_t)&dkmext,
291 FKIOCTL, kcred, NULL) != 0)
292 dkmext.dki_pbsize = DEV_BSIZE;
293
294 *ashift = highbit(MAX(dkmext.dki_pbsize, SPA_MINBLOCKSIZE)) - 1;
295
296 /*
297 * Clear the nowritecache bit, so that on a vdev_reopen() we will
298 * try again.
299 */
300 vd->vdev_nowritecache = B_FALSE;
301
302 return (0);
303}
304
305static void
306vdev_disk_close(vdev_t *vd)
307{
308 vdev_disk_t *dvd = vd->vdev_tsd;
309
310 if (vd->vdev_reopening || dvd == NULL)
311 return;
312
313 if (dvd->vd_minor != NULL)
314 ddi_devid_str_free(dvd->vd_minor);
315
316 if (dvd->vd_devid != NULL)
317 ddi_devid_free(dvd->vd_devid);
318
319 if (dvd->vd_lh != NULL)
320 (void) ldi_close(dvd->vd_lh, spa_mode(vd->vdev_spa), kcred);
321
322 vd->vdev_delayed_close = B_FALSE;
323 kmem_free(dvd, sizeof (vdev_disk_t));
324 vd->vdev_tsd = NULL;
325}
326
327int
328vdev_disk_physio(ldi_handle_t vd_lh, caddr_t data, size_t size,
329 uint64_t offset, int flags)
330{
331 buf_t *bp;
332 int error = 0;
333
334 if (vd_lh == NULL)
335 return (EINVAL);
336
337 ASSERT(flags & B_READ || flags & B_WRITE);
338
339 bp = getrbuf(KM_SLEEP);
340 bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST;
341 bp->b_bcount = size;
342 bp->b_un.b_addr = (void *)data;
343 bp->b_lblkno = lbtodb(offset);
344 bp->b_bufsize = size;
345
346 error = ldi_strategy(vd_lh, bp);
347 ASSERT(error == 0);
348 if ((error = biowait(bp)) == 0 && bp->b_resid != 0)
349 error = EIO;
350 freerbuf(bp);
351
352 return (error);
353}
354
355static void
356vdev_disk_io_intr(buf_t *bp)
357{
358 vdev_disk_buf_t *vdb = (vdev_disk_buf_t *)bp;
359 zio_t *zio = vdb->vdb_io;
360
361 /*
362 * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO.
363 * Rather than teach the rest of the stack about other error
364 * possibilities (EFAULT, etc), we normalize the error value here.
365 */
366 zio->io_error = (geterror(bp) != 0 ? EIO : 0);
367
368 if (zio->io_error == 0 && bp->b_resid != 0)
369 zio->io_error = EIO;
370
371 kmem_free(vdb, sizeof (vdev_disk_buf_t));
372
373 zio_interrupt(zio);
374}
375
376static void
377vdev_disk_ioctl_free(zio_t *zio)
378{
379 kmem_free(zio->io_vsd, sizeof (struct dk_callback));
380}
381
382static const zio_vsd_ops_t vdev_disk_vsd_ops = {
383 vdev_disk_ioctl_free,
384 zio_vsd_default_cksum_report
385};
386
387static void
388vdev_disk_ioctl_done(void *zio_arg, int error)
389{
390 zio_t *zio = zio_arg;
391
392 zio->io_error = error;
393
394 zio_interrupt(zio);
395}
396
397static int
398vdev_disk_io_start(zio_t *zio)
399{
400 vdev_t *vd = zio->io_vd;
401 vdev_disk_t *dvd = vd->vdev_tsd;
402 vdev_disk_buf_t *vdb;
403 struct dk_callback *dkc;
404 buf_t *bp;
405 int error;
406
407 if (zio->io_type == ZIO_TYPE_IOCTL) {
408 /* XXPOLICY */
409 if (!vdev_readable(vd)) {
410 zio->io_error = ENXIO;
411 return (ZIO_PIPELINE_CONTINUE);
412 }
413
414 switch (zio->io_cmd) {
415
416 case DKIOCFLUSHWRITECACHE:
417
418 if (zfs_nocacheflush)
419 break;
420
421 if (vd->vdev_nowritecache) {
422 zio->io_error = ENOTSUP;
423 break;
424 }
425
426 zio->io_vsd = dkc = kmem_alloc(sizeof (*dkc), KM_SLEEP);
427 zio->io_vsd_ops = &vdev_disk_vsd_ops;
428
429 dkc->dkc_callback = vdev_disk_ioctl_done;
430 dkc->dkc_flag = FLUSH_VOLATILE;
431 dkc->dkc_cookie = zio;
432
433 error = ldi_ioctl(dvd->vd_lh, zio->io_cmd,
434 (uintptr_t)dkc, FKIOCTL, kcred, NULL);
435
436 if (error == 0) {
437 /*
438 * The ioctl will be done asychronously,
439 * and will call vdev_disk_ioctl_done()
440 * upon completion.
441 */
442 return (ZIO_PIPELINE_STOP);
443 }
444
445 if (error == ENOTSUP || error == ENOTTY) {
446 /*
447 * If we get ENOTSUP or ENOTTY, we know that
448 * no future attempts will ever succeed.
449 * In this case we set a persistent bit so
450 * that we don't bother with the ioctl in the
451 * future.
452 */
453 vd->vdev_nowritecache = B_TRUE;
454 }
455 zio->io_error = error;
456
457 break;
458
459 default:
460 zio->io_error = ENOTSUP;
461 }
462
463 return (ZIO_PIPELINE_CONTINUE);
464 }
465
466 vdb = kmem_alloc(sizeof (vdev_disk_buf_t), KM_SLEEP);
467
468 vdb->vdb_io = zio;
469 bp = &vdb->vdb_buf;
470
471 bioinit(bp);
472 bp->b_flags = B_BUSY | B_NOCACHE |
473 (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
474 if (!(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))
475 bp->b_flags |= B_FAILFAST;
476 bp->b_bcount = zio->io_size;
477 bp->b_un.b_addr = zio->io_data;
478 bp->b_lblkno = lbtodb(zio->io_offset);
479 bp->b_bufsize = zio->io_size;
480 bp->b_iodone = (int (*)())vdev_disk_io_intr;
481
482 /* ldi_strategy() will return non-zero only on programming errors */
483 VERIFY(ldi_strategy(dvd->vd_lh, bp) == 0);
484
485 return (ZIO_PIPELINE_STOP);
486}
487
488static void
489vdev_disk_io_done(zio_t *zio)
490{
491 vdev_t *vd = zio->io_vd;
492
493 /*
494 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
495 * the device has been removed. If this is the case, then we trigger an
496 * asynchronous removal of the device. Otherwise, probe the device and
497 * make sure it's still accessible.
498 */
499 if (zio->io_error == EIO && !vd->vdev_remove_wanted) {
500 vdev_disk_t *dvd = vd->vdev_tsd;
501 int state = DKIO_NONE;
502
503 if (ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state,
504 FKIOCTL, kcred, NULL) == 0 && state != DKIO_INSERTED) {
505 /*
506 * We post the resource as soon as possible, instead of
507 * when the async removal actually happens, because the
508 * DE is using this information to discard previous I/O
509 * errors.
510 */
511 zfs_post_remove(zio->io_spa, vd);
512 vd->vdev_remove_wanted = B_TRUE;
513 spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
514 } else if (!vd->vdev_delayed_close) {
515 vd->vdev_delayed_close = B_TRUE;
516 }
517 }
518}
519
520vdev_ops_t vdev_disk_ops = {
521 vdev_disk_open,
522 vdev_disk_close,
523 vdev_default_asize,
524 vdev_disk_io_start,
525 vdev_disk_io_done,
526 NULL,
527 vdev_disk_hold,
528 vdev_disk_rele,
529 VDEV_TYPE_DISK, /* name of this vdev type */
530 B_TRUE /* leaf vdev */
531};
532
533/*
534 * Given the root disk device devid or pathname, read the label from
535 * the device, and construct a configuration nvlist.
536 */
537int
538vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config)
539{
540 ldi_handle_t vd_lh;
541 vdev_label_t *label;
542 uint64_t s, size;
543 int l;
544 ddi_devid_t tmpdevid;
545 int error = -1;
546 char *minor_name;
547
548 /*
549 * Read the device label and build the nvlist.
550 */
551 if (devid != NULL && ddi_devid_str_decode(devid, &tmpdevid,
552 &minor_name) == 0) {
553 error = ldi_open_by_devid(tmpdevid, minor_name,
554 FREAD, kcred, &vd_lh, zfs_li);
555 ddi_devid_free(tmpdevid);
556 ddi_devid_str_free(minor_name);
557 }
558
559 if (error && (error = ldi_open_by_name(devpath, FREAD, kcred, &vd_lh,
560 zfs_li)))
561 return (error);
562
563 if (ldi_get_size(vd_lh, &s)) {
564 (void) ldi_close(vd_lh, FREAD, kcred);
565 return (EIO);
566 }
567
568 size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t);
569 label = kmem_alloc(sizeof (vdev_label_t), KM_SLEEP);
570
571 *config = NULL;
572 for (l = 0; l < VDEV_LABELS; l++) {
573 uint64_t offset, state, txg = 0;
574
575 /* read vdev label */
576 offset = vdev_label_offset(size, l, 0);
577 if (vdev_disk_physio(vd_lh, (caddr_t)label,
578 VDEV_SKIP_SIZE + VDEV_PHYS_SIZE, offset, B_READ) != 0)
579 continue;
580
581 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
582 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0) {
583 *config = NULL;
584 continue;
585 }
586
587 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
588 &state) != 0 || state >= POOL_STATE_DESTROYED) {
589 nvlist_free(*config);
590 *config = NULL;
591 continue;
592 }
593
594 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
595 &txg) != 0 || txg == 0) {
596 nvlist_free(*config);
597 *config = NULL;
598 continue;
599 }
600
601 break;
602 }
603
604 kmem_free(label, sizeof (vdev_label_t));
605 (void) ldi_close(vd_lh, FREAD, kcred);
606 if (*config == NULL)
607 error = EIDRM;
608
609 return (error);
610}