author | ek110237 |
Thu, 26 Oct 2006 14:05:48 -0700 | |
changeset 2986 | c782fcf7a319 |
parent 2885 | c0259887ebbc |
child 4451 | 24fbf2d7a5d7 |
permissions | -rw-r--r-- |
789 | 1 |
/* |
2 |
* CDDL HEADER START |
|
3 |
* |
|
4 |
* The contents of this file are subject to the terms of the |
|
1489
fa842259660e
6322205 Enable disk write cache if ZFS owns the disk
webaker
parents:
1171
diff
changeset
|
5 |
* Common Development and Distribution License (the "License"). |
fa842259660e
6322205 Enable disk write cache if ZFS owns the disk
webaker
parents:
1171
diff
changeset
|
6 |
* You may not use this file except in compliance with the License. |
789 | 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 |
/* |
|
1489
fa842259660e
6322205 Enable disk write cache if ZFS owns the disk
webaker
parents:
1171
diff
changeset
|
22 |
* Copyright 2006 Sun Microsystems, Inc. All rights reserved. |
789 | 23 |
* Use is subject to license terms. |
24 |
*/ |
|
25 |
||
26 |
#pragma ident "%Z%%M% %I% %E% SMI" |
|
27 |
||
28 |
#include <sys/zfs_context.h> |
|
29 |
#include <sys/spa.h> |
|
30 |
#include <sys/vdev_disk.h> |
|
31 |
#include <sys/vdev_impl.h> |
|
32 |
#include <sys/fs/zfs.h> |
|
33 |
#include <sys/zio.h> |
|
1171 | 34 |
#include <sys/sunldi.h> |
789 | 35 |
|
36 |
/* |
|
37 |
* Virtual device vector for disks. |
|
38 |
*/ |
|
39 |
||
40 |
extern ldi_ident_t zfs_li; |
|
41 |
||
42 |
typedef struct vdev_disk_buf { |
|
43 |
buf_t vdb_buf; |
|
44 |
zio_t *vdb_io; |
|
45 |
} vdev_disk_buf_t; |
|
46 |
||
47 |
static int |
|
48 |
vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *ashift) |
|
49 |
{ |
|
50 |
vdev_disk_t *dvd; |
|
1732 | 51 |
struct dk_minfo dkm; |
789 | 52 |
int error; |
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 |
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL; |
|
59 |
return (EINVAL); |
|
60 |
} |
|
61 |
||
62 |
dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP); |
|
63 |
||
64 |
/* |
|
65 |
* When opening a disk device, we want to preserve the user's original |
|
66 |
* intent. We always want to open the device by the path the user gave |
|
67 |
* us, even if it is one of multiple paths to the save device. But we |
|
68 |
* also want to be able to survive disks being removed/recabled. |
|
69 |
* Therefore the sequence of opening devices is: |
|
70 |
* |
|
1171 | 71 |
* 1. Try opening the device by path. For legacy pools without the |
72 |
* 'whole_disk' property, attempt to fix the path by appending 's0'. |
|
789 | 73 |
* |
74 |
* 2. If the devid of the device matches the stored value, return |
|
75 |
* success. |
|
76 |
* |
|
77 |
* 3. Otherwise, the device may have moved. Try opening the device |
|
78 |
* by the devid instead. |
|
79 |
* |
|
80 |
*/ |
|
81 |
if (vd->vdev_devid != NULL) { |
|
82 |
if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid, |
|
83 |
&dvd->vd_minor) != 0) { |
|
84 |
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL; |
|
85 |
return (EINVAL); |
|
86 |
} |
|
87 |
} |
|
88 |
||
89 |
error = EINVAL; /* presume failure */ |
|
90 |
||
91 |
if (vd->vdev_path != NULL) { |
|
92 |
ddi_devid_t devid; |
|
93 |
||
1171 | 94 |
if (vd->vdev_wholedisk == -1ULL) { |
95 |
size_t len = strlen(vd->vdev_path) + 3; |
|
96 |
char *buf = kmem_alloc(len, KM_SLEEP); |
|
97 |
ldi_handle_t lh; |
|
98 |
||
99 |
(void) snprintf(buf, len, "%ss0", vd->vdev_path); |
|
789 | 100 |
|
1171 | 101 |
if (ldi_open_by_name(buf, spa_mode, kcred, |
102 |
&lh, zfs_li) == 0) { |
|
103 |
spa_strfree(vd->vdev_path); |
|
104 |
vd->vdev_path = buf; |
|
105 |
vd->vdev_wholedisk = 1ULL; |
|
106 |
(void) ldi_close(lh, spa_mode, kcred); |
|
107 |
} else { |
|
108 |
kmem_free(buf, len); |
|
109 |
} |
|
110 |
} |
|
789 | 111 |
|
1171 | 112 |
error = ldi_open_by_name(vd->vdev_path, spa_mode, kcred, |
113 |
&dvd->vd_lh, zfs_li); |
|
789 | 114 |
|
115 |
/* |
|
116 |
* Compare the devid to the stored value. |
|
117 |
*/ |
|
118 |
if (error == 0 && vd->vdev_devid != NULL && |
|
119 |
ldi_get_devid(dvd->vd_lh, &devid) == 0) { |
|
120 |
if (ddi_devid_compare(devid, dvd->vd_devid) != 0) { |
|
121 |
error = EINVAL; |
|
122 |
(void) ldi_close(dvd->vd_lh, spa_mode, kcred); |
|
123 |
dvd->vd_lh = NULL; |
|
124 |
} |
|
125 |
ddi_devid_free(devid); |
|
126 |
} |
|
1171 | 127 |
|
128 |
/* |
|
129 |
* If we succeeded in opening the device, but 'vdev_wholedisk' |
|
130 |
* is not yet set, then this must be a slice. |
|
131 |
*/ |
|
132 |
if (error == 0 && vd->vdev_wholedisk == -1ULL) |
|
133 |
vd->vdev_wholedisk = 0; |
|
789 | 134 |
} |
135 |
||
136 |
/* |
|
137 |
* If we were unable to open by path, or the devid check fails, open by |
|
138 |
* devid instead. |
|
139 |
*/ |
|
140 |
if (error != 0 && vd->vdev_devid != NULL) |
|
141 |
error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor, |
|
142 |
spa_mode, kcred, &dvd->vd_lh, zfs_li); |
|
143 |
||
144 |
if (error) { |
|
145 |
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED; |
|
146 |
return (error); |
|
147 |
} |
|
148 |
||
149 |
/* |
|
150 |
* Determine the actual size of the device. |
|
151 |
*/ |
|
152 |
if (ldi_get_size(dvd->vd_lh, psize) != 0) { |
|
153 |
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED; |
|
154 |
return (EINVAL); |
|
155 |
} |
|
156 |
||
1732 | 157 |
/* |
158 |
* If we own the whole disk, try to enable disk write caching. |
|
159 |
* We ignore errors because it's OK if we can't do it. |
|
160 |
*/ |
|
1489
fa842259660e
6322205 Enable disk write cache if ZFS owns the disk
webaker
parents:
1171
diff
changeset
|
161 |
if (vd->vdev_wholedisk == 1) { |
1732 | 162 |
int wce = 1; |
163 |
(void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce, |
|
164 |
FKIOCTL, kcred, NULL); |
|
165 |
} |
|
1489
fa842259660e
6322205 Enable disk write cache if ZFS owns the disk
webaker
parents:
1171
diff
changeset
|
166 |
|
1732 | 167 |
/* |
168 |
* Determine the device's minimum transfer size. |
|
169 |
* If the ioctl isn't supported, assume DEV_BSIZE. |
|
170 |
*/ |
|
171 |
if (ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFO, (intptr_t)&dkm, |
|
172 |
FKIOCTL, kcred, NULL) != 0) |
|
173 |
dkm.dki_lbsize = DEV_BSIZE; |
|
1489
fa842259660e
6322205 Enable disk write cache if ZFS owns the disk
webaker
parents:
1171
diff
changeset
|
174 |
|
1732 | 175 |
*ashift = highbit(MAX(dkm.dki_lbsize, SPA_MINBLOCKSIZE)) - 1; |
1489
fa842259660e
6322205 Enable disk write cache if ZFS owns the disk
webaker
parents:
1171
diff
changeset
|
176 |
|
1773
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
177 |
/* |
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
178 |
* Clear the nowritecache bit, so that on a vdev_reopen() we will |
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
179 |
* try again. |
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
180 |
*/ |
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
181 |
vd->vdev_nowritecache = B_FALSE; |
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
182 |
|
789 | 183 |
return (0); |
184 |
} |
|
185 |
||
186 |
static void |
|
187 |
vdev_disk_close(vdev_t *vd) |
|
188 |
{ |
|
189 |
vdev_disk_t *dvd = vd->vdev_tsd; |
|
190 |
||
191 |
if (dvd == NULL) |
|
192 |
return; |
|
193 |
||
194 |
dprintf("removing disk %s, devid %s\n", |
|
195 |
vd->vdev_path ? vd->vdev_path : "<none>", |
|
196 |
vd->vdev_devid ? vd->vdev_devid : "<none>"); |
|
197 |
||
198 |
if (dvd->vd_minor != NULL) |
|
199 |
ddi_devid_str_free(dvd->vd_minor); |
|
200 |
||
201 |
if (dvd->vd_devid != NULL) |
|
202 |
ddi_devid_free(dvd->vd_devid); |
|
203 |
||
204 |
if (dvd->vd_lh != NULL) |
|
205 |
(void) ldi_close(dvd->vd_lh, spa_mode, kcred); |
|
206 |
||
207 |
kmem_free(dvd, sizeof (vdev_disk_t)); |
|
208 |
vd->vdev_tsd = NULL; |
|
209 |
} |
|
210 |
||
211 |
static void |
|
212 |
vdev_disk_io_intr(buf_t *bp) |
|
213 |
{ |
|
214 |
vdev_disk_buf_t *vdb = (vdev_disk_buf_t *)bp; |
|
215 |
zio_t *zio = vdb->vdb_io; |
|
216 |
||
217 |
if ((zio->io_error = geterror(bp)) == 0 && bp->b_resid != 0) |
|
218 |
zio->io_error = EIO; |
|
219 |
||
220 |
kmem_free(vdb, sizeof (vdev_disk_buf_t)); |
|
221 |
||
222 |
zio_next_stage_async(zio); |
|
223 |
} |
|
224 |
||
225 |
static void |
|
226 |
vdev_disk_ioctl_done(void *zio_arg, int error) |
|
227 |
{ |
|
228 |
zio_t *zio = zio_arg; |
|
229 |
||
230 |
zio->io_error = error; |
|
231 |
||
232 |
zio_next_stage_async(zio); |
|
233 |
} |
|
234 |
||
235 |
static void |
|
236 |
vdev_disk_io_start(zio_t *zio) |
|
237 |
{ |
|
238 |
vdev_t *vd = zio->io_vd; |
|
239 |
vdev_disk_t *dvd = vd->vdev_tsd; |
|
240 |
vdev_disk_buf_t *vdb; |
|
241 |
buf_t *bp; |
|
242 |
int flags, error; |
|
243 |
||
244 |
if (zio->io_type == ZIO_TYPE_IOCTL) { |
|
245 |
zio_vdev_io_bypass(zio); |
|
246 |
||
247 |
/* XXPOLICY */ |
|
248 |
if (vdev_is_dead(vd)) { |
|
249 |
zio->io_error = ENXIO; |
|
250 |
zio_next_stage_async(zio); |
|
251 |
return; |
|
252 |
} |
|
253 |
||
254 |
switch (zio->io_cmd) { |
|
255 |
||
256 |
case DKIOCFLUSHWRITECACHE: |
|
257 |
||
2885 | 258 |
if (zfs_nocacheflush) |
259 |
break; |
|
260 |
||
1773
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
261 |
if (vd->vdev_nowritecache) { |
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
262 |
zio->io_error = ENOTSUP; |
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
263 |
break; |
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
264 |
} |
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
265 |
|
789 | 266 |
zio->io_dk_callback.dkc_callback = vdev_disk_ioctl_done; |
267 |
zio->io_dk_callback.dkc_cookie = zio; |
|
268 |
||
269 |
error = ldi_ioctl(dvd->vd_lh, zio->io_cmd, |
|
270 |
(uintptr_t)&zio->io_dk_callback, |
|
271 |
FKIOCTL, kcred, NULL); |
|
272 |
||
273 |
if (error == 0) { |
|
274 |
/* |
|
275 |
* The ioctl will be done asychronously, |
|
276 |
* and will call vdev_disk_ioctl_done() |
|
277 |
* upon completion. |
|
278 |
*/ |
|
279 |
return; |
|
1773
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
280 |
} else if (error == ENOTSUP) { |
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
281 |
/* |
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
282 |
* If we get ENOTSUP, we know that no future |
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
283 |
* attempts will ever succeed. In this case we |
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
284 |
* set a persistent bit so that we don't bother |
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
285 |
* with the ioctl in the future. |
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
286 |
*/ |
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
287 |
vd->vdev_nowritecache = B_TRUE; |
789 | 288 |
} |
289 |
zio->io_error = error; |
|
1773
d6e4f2855c14
6407791 bringover into ZFS results in s. files newer than extracted source
eschrock
parents:
1732
diff
changeset
|
290 |
|
789 | 291 |
break; |
292 |
||
293 |
default: |
|
294 |
zio->io_error = ENOTSUP; |
|
295 |
} |
|
296 |
||
297 |
zio_next_stage_async(zio); |
|
298 |
return; |
|
299 |
} |
|
300 |
||
301 |
if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio) == 0) |
|
302 |
return; |
|
303 |
||
304 |
if ((zio = vdev_queue_io(zio)) == NULL) |
|
305 |
return; |
|
306 |
||
307 |
flags = (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE); |
|
308 |
flags |= B_BUSY | B_NOCACHE; |
|
309 |
if (zio->io_flags & ZIO_FLAG_FAILFAST) |
|
310 |
flags |= B_FAILFAST; |
|
311 |
||
312 |
vdb = kmem_alloc(sizeof (vdev_disk_buf_t), KM_SLEEP); |
|
313 |
||
314 |
vdb->vdb_io = zio; |
|
315 |
bp = &vdb->vdb_buf; |
|
316 |
||
317 |
bioinit(bp); |
|
318 |
bp->b_flags = flags; |
|
319 |
bp->b_bcount = zio->io_size; |
|
320 |
bp->b_un.b_addr = zio->io_data; |
|
321 |
bp->b_lblkno = lbtodb(zio->io_offset); |
|
322 |
bp->b_bufsize = zio->io_size; |
|
323 |
bp->b_iodone = (int (*)())vdev_disk_io_intr; |
|
324 |
||
325 |
/* XXPOLICY */ |
|
326 |
error = vdev_is_dead(vd) ? ENXIO : vdev_error_inject(vd, zio); |
|
327 |
if (error) { |
|
328 |
zio->io_error = error; |
|
329 |
bioerror(bp, error); |
|
330 |
bp->b_resid = bp->b_bcount; |
|
331 |
bp->b_iodone(bp); |
|
332 |
return; |
|
333 |
} |
|
334 |
||
335 |
error = ldi_strategy(dvd->vd_lh, bp); |
|
336 |
/* ldi_strategy() will return non-zero only on programming errors */ |
|
337 |
ASSERT(error == 0); |
|
338 |
} |
|
339 |
||
340 |
static void |
|
341 |
vdev_disk_io_done(zio_t *zio) |
|
342 |
{ |
|
343 |
vdev_queue_io_done(zio); |
|
344 |
||
345 |
if (zio->io_type == ZIO_TYPE_WRITE) |
|
346 |
vdev_cache_write(zio); |
|
347 |
||
1544 | 348 |
if (zio_injection_enabled && zio->io_error == 0) |
349 |
zio->io_error = zio_handle_device_injection(zio->io_vd, EIO); |
|
350 |
||
789 | 351 |
zio_next_stage(zio); |
352 |
} |
|
353 |
||
354 |
vdev_ops_t vdev_disk_ops = { |
|
355 |
vdev_disk_open, |
|
356 |
vdev_disk_close, |
|
357 |
vdev_default_asize, |
|
358 |
vdev_disk_io_start, |
|
359 |
vdev_disk_io_done, |
|
360 |
NULL, |
|
361 |
VDEV_TYPE_DISK, /* name of this vdev type */ |
|
362 |
B_TRUE /* leaf vdev */ |
|
363 |
}; |