xref: /titanic_41/usr/src/uts/common/fs/zfs/vdev_file.c (revision 51ece83525fa18f5e72627610f480dffc7e492fd)
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 2008 Sun Microsystems, Inc.  All rights reserved.
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_file.h>
31 #include <sys/vdev_impl.h>
32 #include <sys/zio.h>
33 #include <sys/fs/zfs.h>
34 #include <sys/fm/fs/zfs.h>
35 
36 /*
37  * Virtual device vector for files.
38  */
39 
40 static int
41 vdev_file_open_common(vdev_t *vd)
42 {
43 	vdev_file_t *vf;
44 	vnode_t *vp;
45 	int error;
46 
47 	/*
48 	 * We must have a pathname, and it must be absolute.
49 	 */
50 	if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
51 		vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
52 		return (EINVAL);
53 	}
54 
55 	vf = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_file_t), KM_SLEEP);
56 
57 	/*
58 	 * We always open the files from the root of the global zone, even if
59 	 * we're in a local zone.  If the user has gotten to this point, the
60 	 * administrator has already decided that the pool should be available
61 	 * to local zone users, so the underlying devices should be as well.
62 	 */
63 	ASSERT(vd->vdev_path != NULL && vd->vdev_path[0] == '/');
64 	error = vn_openat(vd->vdev_path + 1, UIO_SYSSPACE,
65 	    spa_mode | FOFFMAX, 0, &vp, 0, 0, rootdir, -1);
66 
67 	if (error) {
68 		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
69 		return (error);
70 	}
71 
72 	vf->vf_vnode = vp;
73 
74 #ifdef _KERNEL
75 	/*
76 	 * Make sure it's a regular file.
77 	 */
78 	if (vp->v_type != VREG) {
79 		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
80 		return (ENODEV);
81 	}
82 #endif
83 
84 	return (0);
85 }
86 
87 static int
88 vdev_file_open(vdev_t *vd, uint64_t *psize, uint64_t *ashift)
89 {
90 	vdev_file_t *vf;
91 	vattr_t vattr;
92 	int error;
93 
94 	if ((error = vdev_file_open_common(vd)) != 0)
95 		return (error);
96 
97 	vf = vd->vdev_tsd;
98 
99 	/*
100 	 * Determine the physical size of the file.
101 	 */
102 	vattr.va_mask = AT_SIZE;
103 	error = VOP_GETATTR(vf->vf_vnode, &vattr, 0, kcred, NULL);
104 	if (error) {
105 		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
106 		return (error);
107 	}
108 
109 	*psize = vattr.va_size;
110 	*ashift = SPA_MINBLOCKSHIFT;
111 
112 	return (0);
113 }
114 
115 static void
116 vdev_file_close(vdev_t *vd)
117 {
118 	vdev_file_t *vf = vd->vdev_tsd;
119 
120 	if (vf == NULL)
121 		return;
122 
123 	if (vf->vf_vnode != NULL) {
124 		(void) VOP_PUTPAGE(vf->vf_vnode, 0, 0, B_INVAL, kcred, NULL);
125 		(void) VOP_CLOSE(vf->vf_vnode, spa_mode, 1, 0, kcred, NULL);
126 		VN_RELE(vf->vf_vnode);
127 	}
128 
129 	kmem_free(vf, sizeof (vdev_file_t));
130 	vd->vdev_tsd = NULL;
131 }
132 
133 static int
134 vdev_file_probe_io(vdev_t *vd, caddr_t data, size_t size, uint64_t offset,
135     enum uio_rw rw)
136 {
137 	vdev_file_t *vf = vd ? vd->vdev_tsd : NULL;
138 	ssize_t resid;
139 	int error = 0;
140 
141 	if (vd == NULL || vf == NULL || vf->vf_vnode == NULL)
142 		return (EINVAL);
143 
144 	ASSERT(rw == UIO_READ || rw ==  UIO_WRITE);
145 
146 	error = vn_rdwr(rw, vf->vf_vnode, data, size, offset, UIO_SYSSPACE,
147 	    0, RLIM64_INFINITY, kcred, &resid);
148 
149 	if (error || resid != 0)
150 		return (EIO);
151 
152 	if (zio_injection_enabled)
153 		error = zio_handle_device_injection(vd, EIO);
154 
155 	return (error);
156 }
157 
158 /*
159  * Determine if the underlying device is accessible by reading and writing
160  * to a known location. We must be able to do this during syncing context
161  * and thus we cannot set the vdev state directly.
162  */
163 static int
164 vdev_file_probe(vdev_t *vd)
165 {
166 	vdev_t *nvd;
167 	char *vl_boot;
168 	uint64_t offset;
169 	int l, error = 0, retries = 0;
170 
171 	if (vd == NULL)
172 		return (EINVAL);
173 
174 	/* Hijack the current vdev */
175 	nvd = vd;
176 
177 	/*
178 	 * Pick a random label to rewrite.
179 	 */
180 	l = spa_get_random(VDEV_LABELS);
181 	ASSERT(l < VDEV_LABELS);
182 
183 	offset = vdev_label_offset(vd->vdev_psize, l,
184 	    offsetof(vdev_label_t, vl_boot_header));
185 
186 	vl_boot = kmem_alloc(VDEV_BOOT_HEADER_SIZE, KM_SLEEP);
187 
188 	while ((error = vdev_file_probe_io(nvd, vl_boot, VDEV_BOOT_HEADER_SIZE,
189 	    offset, UIO_READ)) != 0 && retries == 0) {
190 
191 		/*
192 		 * If we failed with the vdev that was passed in then
193 		 * try allocating a new one and try again.
194 		 */
195 		nvd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
196 		if (vd->vdev_path)
197 			nvd->vdev_path = spa_strdup(vd->vdev_path);
198 		nvd->vdev_guid = vd->vdev_guid;
199 		retries++;
200 
201 		if (vdev_file_open_common(nvd) != 0)
202 			break;
203 	}
204 
205 	if ((spa_mode & FWRITE) && !error) {
206 		error = vdev_file_probe_io(nvd, vl_boot, VDEV_BOOT_HEADER_SIZE,
207 		    offset, UIO_WRITE);
208 	}
209 
210 	if (retries) {
211 		vdev_file_close(nvd);
212 		if (nvd->vdev_path)
213 			spa_strfree(nvd->vdev_path);
214 		kmem_free(nvd, sizeof (vdev_t));
215 	}
216 	kmem_free(vl_boot, VDEV_BOOT_HEADER_SIZE);
217 
218 	if (!error)
219 		vd->vdev_is_failing = B_FALSE;
220 
221 	return (error);
222 }
223 
224 static int
225 vdev_file_io_start(zio_t *zio)
226 {
227 	vdev_t *vd = zio->io_vd;
228 	vdev_file_t *vf = vd->vdev_tsd;
229 	ssize_t resid;
230 	int error;
231 
232 	if (zio->io_type == ZIO_TYPE_IOCTL) {
233 		zio_vdev_io_bypass(zio);
234 
235 		/* XXPOLICY */
236 		if (!vdev_readable(vd)) {
237 			zio->io_error = ENXIO;
238 			return (ZIO_PIPELINE_CONTINUE);
239 		}
240 
241 		switch (zio->io_cmd) {
242 		case DKIOCFLUSHWRITECACHE:
243 			zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC | FDSYNC,
244 			    kcred, NULL);
245 			dprintf("fsync(%s) = %d\n", vdev_description(vd),
246 			    zio->io_error);
247 			break;
248 		default:
249 			zio->io_error = ENOTSUP;
250 		}
251 
252 		return (ZIO_PIPELINE_CONTINUE);
253 	}
254 
255 	/*
256 	 * In the kernel, don't bother double-caching, but in userland,
257 	 * we want to test the vdev_cache code.
258 	 */
259 #ifndef _KERNEL
260 	if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio) == 0)
261 		return (ZIO_PIPELINE_STOP);
262 #endif
263 
264 	if ((zio = vdev_queue_io(zio)) == NULL)
265 		return (ZIO_PIPELINE_STOP);
266 
267 	/* XXPOLICY */
268 	if (zio->io_type == ZIO_TYPE_WRITE)
269 		error = vdev_writeable(vd) ? vdev_error_inject(vd, zio) : ENXIO;
270 	else
271 		error = vdev_readable(vd) ? vdev_error_inject(vd, zio) : ENXIO;
272 	error = (vd->vdev_remove_wanted || vd->vdev_is_failing) ? ENXIO : error;
273 	if (error) {
274 		zio->io_error = error;
275 		zio_interrupt(zio);
276 		return (ZIO_PIPELINE_STOP);
277 	}
278 
279 	zio->io_error = vn_rdwr(zio->io_type == ZIO_TYPE_READ ?
280 	    UIO_READ : UIO_WRITE, vf->vf_vnode, zio->io_data,
281 	    zio->io_size, zio->io_offset, UIO_SYSSPACE,
282 	    0, RLIM64_INFINITY, kcred, &resid);
283 
284 	if (resid != 0 && zio->io_error == 0)
285 		zio->io_error = ENOSPC;
286 
287 	zio_interrupt(zio);
288 
289 	return (ZIO_PIPELINE_STOP);
290 }
291 
292 static int
293 vdev_file_io_done(zio_t *zio)
294 {
295 	vdev_t *vd = zio->io_vd;
296 
297 	if (zio_injection_enabled && zio->io_error == 0)
298 		zio->io_error = zio_handle_device_injection(vd, EIO);
299 
300 	/*
301 	 * If an error has been encountered then attempt to probe the device
302 	 * to determine if it's still accessible.
303 	 */
304 	if (zio->io_error == EIO && vdev_probe(vd) != 0) {
305 		if (!vd->vdev_is_failing) {
306 			vd->vdev_is_failing = B_TRUE;
307 			zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
308 			    vd->vdev_spa, vd, zio, 0, 0);
309 		}
310 	}
311 
312 	vdev_queue_io_done(zio);
313 
314 #ifndef _KERNEL
315 	if (zio->io_type == ZIO_TYPE_WRITE)
316 		vdev_cache_write(zio);
317 #endif
318 
319 	return (ZIO_PIPELINE_CONTINUE);
320 }
321 
322 vdev_ops_t vdev_file_ops = {
323 	vdev_file_open,
324 	vdev_file_close,
325 	vdev_file_probe,
326 	vdev_default_asize,
327 	vdev_file_io_start,
328 	vdev_file_io_done,
329 	NULL,
330 	VDEV_TYPE_FILE,		/* name of this vdev type */
331 	B_TRUE			/* leaf vdev */
332 };
333 
334 /*
335  * From userland we access disks just like files.
336  */
337 #ifndef _KERNEL
338 
339 vdev_ops_t vdev_disk_ops = {
340 	vdev_file_open,
341 	vdev_file_close,
342 	vdev_file_probe,
343 	vdev_default_asize,
344 	vdev_file_io_start,
345 	vdev_file_io_done,
346 	NULL,
347 	VDEV_TYPE_DISK,		/* name of this vdev type */
348 	B_TRUE			/* leaf vdev */
349 };
350 
351 #endif
352