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;
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)
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) {
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;
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);
|
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) 2013, Joyent, Inc. All rights reserved.
25 * Copyright 2012 Nexenta Systems, Inc. All rights reserved.
26 */
27
28 #include <sys/zfs_context.h>
29 #include <sys/zfs_zone.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/fs/zfs.h>
35 #include <sys/zio.h>
36 #include <sys/sunldi.h>
37 #include <sys/efi_partition.h>
38 #include <sys/fm/fs/zfs.h>
39
40 /*
41 * Virtual device vector for disks.
42 */
43
44 extern ldi_ident_t zfs_li;
45
46 static void vdev_disk_close(vdev_t *);
47
48 typedef struct vdev_disk_buf {
49 buf_t vdb_buf;
50 zio_t *vdb_io;
51 } vdev_disk_buf_t;
52
53 typedef struct vdev_disk_ldi_cb {
54 list_node_t lcb_next;
55 ldi_callback_id_t lcb_id;
56 } vdev_disk_ldi_cb_t;
57
58 static void vdev_disk_alloc(vdev_t *vd)
59 {
60 vdev_disk_t *dvd;
61
62 dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
63 /*
64 * Create the LDI event callback list.
65 */
66 list_create(&dvd->vd_ldi_cbs, sizeof (vdev_disk_ldi_cb_t),
67 offsetof(vdev_disk_ldi_cb_t, lcb_next));
68 }
69
70 static void vdev_disk_free(vdev_t *vd)
71 {
72 vdev_disk_t *dvd = vd->vdev_tsd;
73 vdev_disk_ldi_cb_t *lcb;
74
75 /*
76 * We have already closed the LDI handle. Clean up the LDI event
77 * callbacks and free vd->vdev_tsd.
78 */
79 while ((lcb = list_head(&dvd->vd_ldi_cbs)) != NULL) {
80 list_remove(&dvd->vd_ldi_cbs, lcb);
81 (void) ldi_ev_remove_callbacks(lcb->lcb_id);
82 kmem_free(lcb, sizeof (vdev_disk_ldi_cb_t));
83 }
84 list_destroy(&dvd->vd_ldi_cbs);
85 kmem_free(dvd, sizeof (vdev_disk_t));
86 vd->vdev_tsd = NULL;
87 }
88
89 /* ARGSUSED */
90 static int
91 vdev_disk_off_notify(ldi_handle_t lh, ldi_ev_cookie_t ecookie, void *arg,
92 void *ev_data)
93 {
94 vdev_t *vd = (vdev_t *)arg;
95 vdev_disk_t *dvd = vd->vdev_tsd;
96
97 /*
98 * Ignore events other than offline.
99 */
100 if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_OFFLINE) != 0)
101 return (LDI_EV_SUCCESS);
102
103 /*
104 * All LDI handles must be closed for the state change to succeed, so
105 * call on vdev_disk_close() to do this.
106 *
107 * We inform vdev_disk_close that it is being called from offline
108 * notify context so it will defer cleanup of LDI event callbacks and
109 * freeing of vd->vdev_tsd to the offline finalize or a reopen.
110 */
111 dvd->vd_ldi_offline = B_TRUE;
112 vdev_disk_close(vd);
113
114 /*
115 * Now that the device is closed, request that the spa_async_thread
116 * mark the device as REMOVED and notify FMA of the removal.
117 */
118 zfs_post_remove(vd->vdev_spa, vd);
119 vd->vdev_remove_wanted = B_TRUE;
120 spa_async_request(vd->vdev_spa, SPA_ASYNC_REMOVE);
121
122 return (LDI_EV_SUCCESS);
123 }
124
125 /* ARGSUSED */
126 static void
127 vdev_disk_off_finalize(ldi_handle_t lh, ldi_ev_cookie_t ecookie,
128 int ldi_result, void *arg, void *ev_data)
129 {
130 vdev_t *vd = (vdev_t *)arg;
131 vdev_disk_t *dvd = vd->vdev_tsd;
132 vdev_disk_ldi_cb_t *lcb;
133
134 /*
135 * Ignore events other than offline.
136 */
137 if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_OFFLINE) != 0)
138 return;
139
140 /*
141 * We have already closed the LDI handle in notify.
142 * Clean up the LDI event callbacks and free vd->vdev_tsd.
143 */
144 vdev_disk_free(vd);
145
146 /*
147 * Request that the vdev be reopened if the offline state change was
148 * unsuccessful.
149 */
150 if (ldi_result != LDI_EV_SUCCESS) {
151 vd->vdev_probe_wanted = B_TRUE;
152 spa_async_request(vd->vdev_spa, SPA_ASYNC_PROBE);
153 }
154 }
155
156 static ldi_ev_callback_t vdev_disk_off_callb = {
157 .cb_vers = LDI_EV_CB_VERS,
158 .cb_notify = vdev_disk_off_notify,
159 .cb_finalize = vdev_disk_off_finalize
160 };
161
162 /* ARGSUSED */
163 static void
164 vdev_disk_dgrd_finalize(ldi_handle_t lh, ldi_ev_cookie_t ecookie,
165 int ldi_result, void *arg, void *ev_data)
166 {
167 vdev_t *vd = (vdev_t *)arg;
168
169 /*
170 * Ignore events other than degrade.
171 */
172 if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_DEGRADE) != 0)
173 return;
174
175 /*
176 * Degrade events always succeed. Mark the vdev as degraded.
177 * This status is purely informative for the user.
178 */
179 (void) vdev_degrade(vd->vdev_spa, vd->vdev_guid, 0);
180 }
181
182 static ldi_ev_callback_t vdev_disk_dgrd_callb = {
183 .cb_vers = LDI_EV_CB_VERS,
184 .cb_notify = NULL,
185 .cb_finalize = vdev_disk_dgrd_finalize
186 };
187
188 static void
189 vdev_disk_hold(vdev_t *vd)
190 {
191 ddi_devid_t devid;
192 char *minor;
193
194 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
195
196 /*
197 * We must have a pathname, and it must be absolute.
198 */
199 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/')
200 return;
201
202 /*
203 * Only prefetch path and devid info if the device has
204 * never been opened.
205 */
206 if (vd->vdev_tsd != NULL)
207 return;
262 efi = dk_ioc.dki_data;
263
264 if (ldi_ioctl(dvd->vd_lh, DKIOCGETEFI, (intptr_t)&dk_ioc,
265 FKIOCTL, kcred, NULL) == 0) {
266 uint64_t efi_altern_lba = LE_64(efi->efi_gpt_AlternateLBA);
267
268 zfs_dbgmsg("vdev %s, capacity %llu, altern lba %llu",
269 vd->vdev_path, capacity, efi_altern_lba);
270 if (capacity > efi_altern_lba)
271 avail_space = (capacity - efi_altern_lba) * blksz;
272 }
273 kmem_free(dk_ioc.dki_data, efisize);
274 return (avail_space);
275 }
276
277 static int
278 vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
279 uint64_t *ashift)
280 {
281 spa_t *spa = vd->vdev_spa;
282 vdev_disk_t *dvd = vd->vdev_tsd;
283 struct dk_minfo_ext dkmext;
284 ldi_ev_cookie_t ecookie;
285 vdev_disk_ldi_cb_t *lcb;
286 int error;
287 dev_t dev;
288 int otyp;
289
290 /*
291 * We must have a pathname, and it must be absolute.
292 */
293 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
294 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
295 return (EINVAL);
296 }
297
298 /*
299 * Reopen the device if it's not currently open. Otherwise,
300 * just update the physical size of the device.
301 */
302 if (dvd != NULL) {
303 if (dvd->vd_ldi_offline && dvd->vd_lh == NULL) {
304 /*
305 * If we are opening a device in its offline notify
306 * context, the LDI handle was just closed. Clean
307 * up the LDI event callbacks and free vd->vdev_tsd.
308 */
309 vdev_disk_free(vd);
310 } else {
311 ASSERT(vd->vdev_reopening);
312 goto skip_open;
313 }
314 }
315
316 /*
317 * Create vd->vdev_tsd.
318 */
319 vdev_disk_alloc(vd);
320 dvd = vd->vdev_tsd;
321
322 /*
323 * When opening a disk device, we want to preserve the user's original
324 * intent. We always want to open the device by the path the user gave
325 * us, even if it is one of multiple paths to the same device. But we
326 * also want to be able to survive disks being removed/recabled.
327 * Therefore the sequence of opening devices is:
328 *
329 * 1. Try opening the device by path. For legacy pools without the
330 * 'whole_disk' property, attempt to fix the path by appending 's0'.
331 *
332 * 2. If the devid of the device matches the stored value, return
333 * success.
334 *
335 * 3. Otherwise, the device may have moved. Try opening the device
336 * by the devid instead.
337 */
338 if (vd->vdev_devid != NULL) {
339 if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
340 &dvd->vd_minor) != 0) {
341 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
342 return (EINVAL);
343 }
344 }
345
346 error = EINVAL; /* presume failure */
347
348 if (vd->vdev_path != NULL) {
349 ddi_devid_t devid;
350
351 if (vd->vdev_wholedisk == -1ULL) {
352 size_t len = strlen(vd->vdev_path) + 3;
353 char *buf = kmem_alloc(len, KM_SLEEP);
354
355 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
356
357 error = ldi_open_by_name(buf, spa_mode(spa), kcred,
358 &dvd->vd_lh, zfs_li);
359 if (error == 0) {
360 spa_strfree(vd->vdev_path);
361 vd->vdev_path = buf;
362 vd->vdev_wholedisk = 1ULL;
363 } else {
364 kmem_free(buf, len);
365 }
366 }
367
368 /*
369 * If we have not yet opened the device, try to open it by the
370 * specified path.
371 */
372 if (error != 0) {
373 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
374 kcred, &dvd->vd_lh, zfs_li);
375 }
376
377 /*
378 * Compare the devid to the stored value.
379 */
380 if (error == 0 && vd->vdev_devid != NULL &&
381 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
382 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
383 error = EINVAL;
384 (void) ldi_close(dvd->vd_lh, spa_mode(spa),
385 kcred);
386 dvd->vd_lh = NULL;
387 }
388 ddi_devid_free(devid);
389 }
390
391 /*
392 * If we succeeded in opening the device, but 'vdev_wholedisk'
393 * is not yet set, then this must be a slice.
394 */
395 if (error == 0 && vd->vdev_wholedisk == -1ULL)
439 ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
440 char *physpath, *minorname;
441
442 physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
443 minorname = NULL;
444 if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
445 ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
446 (vd->vdev_physpath == NULL ||
447 strcmp(vd->vdev_physpath, physpath) != 0)) {
448 if (vd->vdev_physpath)
449 spa_strfree(vd->vdev_physpath);
450 (void) strlcat(physpath, ":", MAXPATHLEN);
451 (void) strlcat(physpath, minorname, MAXPATHLEN);
452 vd->vdev_physpath = spa_strdup(physpath);
453 }
454 if (minorname)
455 kmem_free(minorname, strlen(minorname) + 1);
456 kmem_free(physpath, MAXPATHLEN);
457 }
458
459 /*
460 * Register callbacks for the LDI offline event.
461 */
462 if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_OFFLINE, &ecookie) ==
463 LDI_EV_SUCCESS) {
464 lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
465 list_insert_tail(&dvd->vd_ldi_cbs, lcb);
466 (void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
467 &vdev_disk_off_callb, (void *) vd, &lcb->lcb_id);
468 }
469
470 /*
471 * Register callbacks for the LDI degrade event.
472 */
473 if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_DEGRADE, &ecookie) ==
474 LDI_EV_SUCCESS) {
475 lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
476 list_insert_tail(&dvd->vd_ldi_cbs, lcb);
477 (void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
478 &vdev_disk_dgrd_callb, (void *) vd, &lcb->lcb_id);
479 }
480 skip_open:
481 /*
482 * Determine the actual size of the device.
483 */
484 if (ldi_get_size(dvd->vd_lh, psize) != 0) {
485 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
486 return (EINVAL);
487 }
488
489 /*
490 * Determine the device's minimum transfer size.
491 * If the ioctl isn't supported, assume DEV_BSIZE.
492 */
493 if (ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT, (intptr_t)&dkmext,
494 FKIOCTL, kcred, NULL) != 0)
495 dkmext.dki_pbsize = DEV_BSIZE;
496
497 *ashift = highbit(MAX(dkmext.dki_pbsize, SPA_MINBLOCKSIZE)) - 1;
498
499 if (vd->vdev_wholedisk == 1) {
511 *max_psize = *psize + vdev_disk_get_space(vd, capacity, blksz);
512 zfs_dbgmsg("capacity change: vdev %s, psize %llu, "
513 "max_psize %llu", vd->vdev_path, *psize, *max_psize);
514 } else {
515 *max_psize = *psize;
516 }
517
518 /*
519 * Clear the nowritecache bit, so that on a vdev_reopen() we will
520 * try again.
521 */
522 vd->vdev_nowritecache = B_FALSE;
523
524 return (0);
525 }
526
527 static void
528 vdev_disk_close(vdev_t *vd)
529 {
530 vdev_disk_t *dvd = vd->vdev_tsd;
531 vdev_disk_ldi_cb_t *lcb;
532
533 if (vd->vdev_reopening || dvd == NULL)
534 return;
535
536 if (dvd->vd_minor != NULL) {
537 ddi_devid_str_free(dvd->vd_minor);
538 dvd->vd_minor = NULL;
539 }
540
541 if (dvd->vd_devid != NULL) {
542 ddi_devid_free(dvd->vd_devid);
543 dvd->vd_devid = NULL;
544 }
545
546 if (dvd->vd_lh != NULL) {
547 (void) ldi_close(dvd->vd_lh, spa_mode(vd->vdev_spa), kcred);
548 dvd->vd_lh = NULL;
549 }
550
551 vd->vdev_delayed_close = B_FALSE;
552 /*
553 * If we closed the LDI handle due to an offline notify from LDI,
554 * don't free vd->vdev_tsd or unregister the callbacks here;
555 * the offline finalize callback or a reopen will take care of it.
556 */
557 if (dvd->vd_ldi_offline)
558 return;
559
560 vdev_disk_free(vd);
561 }
562
563 int
564 vdev_disk_physio(vdev_t *vd, caddr_t data,
565 size_t size, uint64_t offset, int flags)
566 {
567 vdev_disk_t *dvd = vd->vdev_tsd;
568
569 /*
570 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
571 * Nothing to be done here but return failure.
572 */
573 if (dvd == NULL || (dvd->vd_ldi_offline && dvd->vd_lh == NULL))
574 return (EIO);
575
576 ASSERT(vd->vdev_ops == &vdev_disk_ops);
577 return (vdev_disk_ldi_physio(dvd->vd_lh, data, size, offset, flags));
578 }
579
580 int
581 vdev_disk_ldi_physio(ldi_handle_t vd_lh, caddr_t data,
582 size_t size, uint64_t offset, int flags)
583 {
584 buf_t *bp;
585 int error = 0;
586
587 if (vd_lh == NULL)
588 return (EINVAL);
589
590 ASSERT(flags & B_READ || flags & B_WRITE);
591
592 bp = getrbuf(KM_SLEEP);
593 bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST;
640 static void
641 vdev_disk_ioctl_done(void *zio_arg, int error)
642 {
643 zio_t *zio = zio_arg;
644
645 zio->io_error = error;
646
647 zio_interrupt(zio);
648 }
649
650 static int
651 vdev_disk_io_start(zio_t *zio)
652 {
653 vdev_t *vd = zio->io_vd;
654 vdev_disk_t *dvd = vd->vdev_tsd;
655 vdev_buf_t *vb;
656 struct dk_callback *dkc;
657 buf_t *bp;
658 int error;
659
660 /*
661 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
662 * Nothing to be done here but return failure.
663 */
664 if (dvd == NULL || (dvd->vd_ldi_offline && dvd->vd_lh == NULL)) {
665 zio->io_error = ENXIO;
666 return (ZIO_PIPELINE_CONTINUE);
667 }
668
669 if (zio->io_type == ZIO_TYPE_IOCTL) {
670 /* XXPOLICY */
671 if (!vdev_readable(vd)) {
672 zio->io_error = ENXIO;
673 return (ZIO_PIPELINE_CONTINUE);
674 }
675
676 switch (zio->io_cmd) {
677
678 case DKIOCFLUSHWRITECACHE:
679
680 if (zfs_nocacheflush)
681 break;
682
683 if (vd->vdev_nowritecache) {
684 zio->io_error = ENOTSUP;
685 break;
686 }
687
688 zio->io_vsd = dkc = kmem_alloc(sizeof (*dkc), KM_SLEEP);
|