xref: /illumos-gate/usr/src/uts/common/fs/zfs/vdev_disk.c (revision a07094369b21309434206d9b3601d162693466fc)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include <sys/zfs_context.h>
30 #include <sys/spa.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 
37 /*
38  * Virtual device vector for disks.
39  */
40 
41 extern ldi_ident_t zfs_li;
42 
43 typedef struct vdev_disk_buf {
44 	buf_t	vdb_buf;
45 	zio_t	*vdb_io;
46 } vdev_disk_buf_t;
47 
48 static int
49 vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *ashift)
50 {
51 	vdev_disk_t *dvd;
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 	 *
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'.
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 
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);
100 
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 		}
111 
112 		error = ldi_open_by_name(vd->vdev_path, spa_mode, kcred,
113 		    &dvd->vd_lh, zfs_li);
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 		}
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;
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 
157 	*ashift = SPA_MINBLOCKSHIFT;
158 
159 	return (0);
160 }
161 
162 static void
163 vdev_disk_close(vdev_t *vd)
164 {
165 	vdev_disk_t *dvd = vd->vdev_tsd;
166 
167 	if (dvd == NULL)
168 		return;
169 
170 	dprintf("removing disk %s, devid %s\n",
171 	    vd->vdev_path ? vd->vdev_path : "<none>",
172 	    vd->vdev_devid ? vd->vdev_devid : "<none>");
173 
174 	if (dvd->vd_minor != NULL)
175 		ddi_devid_str_free(dvd->vd_minor);
176 
177 	if (dvd->vd_devid != NULL)
178 		ddi_devid_free(dvd->vd_devid);
179 
180 	if (dvd->vd_lh != NULL)
181 		(void) ldi_close(dvd->vd_lh, spa_mode, kcred);
182 
183 	kmem_free(dvd, sizeof (vdev_disk_t));
184 	vd->vdev_tsd = NULL;
185 }
186 
187 static void
188 vdev_disk_io_intr(buf_t *bp)
189 {
190 	vdev_disk_buf_t *vdb = (vdev_disk_buf_t *)bp;
191 	zio_t *zio = vdb->vdb_io;
192 
193 	if ((zio->io_error = geterror(bp)) == 0 && bp->b_resid != 0)
194 		zio->io_error = EIO;
195 
196 	kmem_free(vdb, sizeof (vdev_disk_buf_t));
197 
198 	zio_next_stage_async(zio);
199 }
200 
201 static void
202 vdev_disk_ioctl_done(void *zio_arg, int error)
203 {
204 	zio_t *zio = zio_arg;
205 
206 	zio->io_error = error;
207 
208 	zio_next_stage_async(zio);
209 }
210 
211 static void
212 vdev_disk_io_start(zio_t *zio)
213 {
214 	vdev_t *vd = zio->io_vd;
215 	vdev_disk_t *dvd = vd->vdev_tsd;
216 	vdev_disk_buf_t *vdb;
217 	buf_t *bp;
218 	int flags, error;
219 
220 	if (zio->io_type == ZIO_TYPE_IOCTL) {
221 		zio_vdev_io_bypass(zio);
222 
223 		/* XXPOLICY */
224 		if (vdev_is_dead(vd)) {
225 			zio->io_error = ENXIO;
226 			zio_next_stage_async(zio);
227 			return;
228 		}
229 
230 		switch (zio->io_cmd) {
231 
232 		case DKIOCFLUSHWRITECACHE:
233 
234 			zio->io_dk_callback.dkc_callback = vdev_disk_ioctl_done;
235 			zio->io_dk_callback.dkc_cookie = zio;
236 
237 			error = ldi_ioctl(dvd->vd_lh, zio->io_cmd,
238 			    (uintptr_t)&zio->io_dk_callback,
239 			    FKIOCTL, kcred, NULL);
240 
241 			if (error == 0) {
242 				/*
243 				 * The ioctl will be done asychronously,
244 				 * and will call vdev_disk_ioctl_done()
245 				 * upon completion.
246 				 */
247 				return;
248 			}
249 			zio->io_error = error;
250 			break;
251 
252 		default:
253 			zio->io_error = ENOTSUP;
254 		}
255 
256 		zio_next_stage_async(zio);
257 		return;
258 	}
259 
260 	if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio) == 0)
261 		return;
262 
263 	if ((zio = vdev_queue_io(zio)) == NULL)
264 		return;
265 
266 	flags = (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
267 	flags |= B_BUSY | B_NOCACHE;
268 	if (zio->io_flags & ZIO_FLAG_FAILFAST)
269 		flags |= B_FAILFAST;
270 
271 	vdb = kmem_alloc(sizeof (vdev_disk_buf_t), KM_SLEEP);
272 
273 	vdb->vdb_io = zio;
274 	bp = &vdb->vdb_buf;
275 
276 	bioinit(bp);
277 	bp->b_flags = flags;
278 	bp->b_bcount = zio->io_size;
279 	bp->b_un.b_addr = zio->io_data;
280 	bp->b_lblkno = lbtodb(zio->io_offset);
281 	bp->b_bufsize = zio->io_size;
282 	bp->b_iodone = (int (*)())vdev_disk_io_intr;
283 
284 	/* XXPOLICY */
285 	error = vdev_is_dead(vd) ? ENXIO : vdev_error_inject(vd, zio);
286 	if (error) {
287 		zio->io_error = error;
288 		bioerror(bp, error);
289 		bp->b_resid = bp->b_bcount;
290 		bp->b_iodone(bp);
291 		return;
292 	}
293 
294 	error = ldi_strategy(dvd->vd_lh, bp);
295 	/* ldi_strategy() will return non-zero only on programming errors */
296 	ASSERT(error == 0);
297 }
298 
299 static void
300 vdev_disk_io_done(zio_t *zio)
301 {
302 	vdev_queue_io_done(zio);
303 
304 	if (zio->io_type == ZIO_TYPE_WRITE)
305 		vdev_cache_write(zio);
306 
307 	zio_next_stage(zio);
308 }
309 
310 vdev_ops_t vdev_disk_ops = {
311 	vdev_disk_open,
312 	vdev_disk_close,
313 	vdev_default_asize,
314 	vdev_disk_io_start,
315 	vdev_disk_io_done,
316 	NULL,
317 	VDEV_TYPE_DISK,		/* name of this vdev type */
318 	B_TRUE			/* leaf vdev */
319 };
320