xref: /illumos-gate/usr/src/uts/common/fs/zfs/vdev_file.c (revision 2caf0dcd2abc26b477e317999994020212790d38)
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 2006 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 
35 /*
36  * Virtual device vector for files.
37  */
38 
39 static int
40 vdev_file_open(vdev_t *vd, uint64_t *psize, uint64_t *ashift)
41 {
42 	vdev_file_t *vf;
43 	vnode_t *vp;
44 	vattr_t vattr;
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 #ifdef _KERNEL
58 	/*
59 	 * When using a file vdev in kernel context, the underlying filesystem
60 	 * will already be caching the data.  Don't cache it again here.
61 	 */
62 	vd->vdev_cache.vc_size = 0;
63 #endif
64 
65 	/*
66 	 * We always open the files from the root of the global zone, even if
67 	 * we're in a local zone.  If the user has gotten to this point, the
68 	 * administrator has already decided that the pool should be available
69 	 * to local zone users, so the underlying devices should be as well.
70 	 */
71 	ASSERT(vd->vdev_path != NULL && vd->vdev_path[0] == '/');
72 	error = vn_openat(vd->vdev_path + 1, UIO_SYSSPACE, spa_mode | FOFFMAX,
73 	    0, &vp, 0, 0, rootdir);
74 
75 	if (error) {
76 		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
77 		return (error);
78 	}
79 
80 	vf->vf_vnode = vp;
81 
82 #ifdef _KERNEL
83 	/*
84 	 * Make sure it's a regular file.
85 	 */
86 	if (vp->v_type != VREG) {
87 		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
88 		return (ENODEV);
89 	}
90 #endif
91 
92 	/*
93 	 * Determine the physical size of the file.
94 	 */
95 	vattr.va_mask = AT_SIZE;
96 	error = VOP_GETATTR(vp, &vattr, 0, kcred);
97 	if (error) {
98 		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
99 		return (error);
100 	}
101 
102 	*psize = vattr.va_size;
103 	*ashift = SPA_MINBLOCKSHIFT;
104 
105 	return (0);
106 }
107 
108 static void
109 vdev_file_close(vdev_t *vd)
110 {
111 	vdev_file_t *vf = vd->vdev_tsd;
112 
113 	if (vf == NULL)
114 		return;
115 
116 	if (vf->vf_vnode != NULL) {
117 		(void) VOP_PUTPAGE(vf->vf_vnode, 0, 0, B_INVAL, kcred);
118 		(void) VOP_CLOSE(vf->vf_vnode, spa_mode, 1, 0, kcred);
119 		VN_RELE(vf->vf_vnode);
120 	}
121 
122 	kmem_free(vf, sizeof (vdev_file_t));
123 	vd->vdev_tsd = NULL;
124 }
125 
126 static void
127 vdev_file_io_start(zio_t *zio)
128 {
129 	vdev_t *vd = zio->io_vd;
130 	vdev_file_t *vf = vd->vdev_tsd;
131 	ssize_t resid;
132 	int error;
133 
134 	if (zio->io_type == ZIO_TYPE_IOCTL) {
135 		zio_vdev_io_bypass(zio);
136 
137 		/* XXPOLICY */
138 		if (vdev_is_dead(vd)) {
139 			zio->io_error = ENXIO;
140 			zio_next_stage_async(zio);
141 			return;
142 		}
143 
144 		switch (zio->io_cmd) {
145 		case DKIOCFLUSHWRITECACHE:
146 			zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC | FDSYNC,
147 			    kcred);
148 			dprintf("fsync(%s) = %d\n", vdev_description(vd),
149 			    zio->io_error);
150 			break;
151 		default:
152 			zio->io_error = ENOTSUP;
153 		}
154 
155 		zio_next_stage_async(zio);
156 		return;
157 	}
158 
159 	if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio) == 0)
160 		return;
161 
162 	if ((zio = vdev_queue_io(zio)) == NULL)
163 		return;
164 
165 	/* XXPOLICY */
166 	error = vdev_is_dead(vd) ? ENXIO : vdev_error_inject(vd, zio);
167 	if (error) {
168 		zio->io_error = error;
169 		zio_next_stage_async(zio);
170 		return;
171 	}
172 
173 	zio->io_error = vn_rdwr(zio->io_type == ZIO_TYPE_READ ?
174 	    UIO_READ : UIO_WRITE, vf->vf_vnode, zio->io_data,
175 	    zio->io_size, zio->io_offset, UIO_SYSSPACE,
176 	    0, RLIM64_INFINITY, kcred, &resid);
177 
178 	if (resid != 0 && zio->io_error == 0)
179 		zio->io_error = ENOSPC;
180 
181 	zio_next_stage_async(zio);
182 }
183 
184 static void
185 vdev_file_io_done(zio_t *zio)
186 {
187 	vdev_queue_io_done(zio);
188 
189 	if (zio->io_type == ZIO_TYPE_WRITE)
190 		vdev_cache_write(zio);
191 
192 	if (zio_injection_enabled && zio->io_error == 0)
193 		zio->io_error = zio_handle_device_injection(zio->io_vd, EIO);
194 
195 	zio_next_stage(zio);
196 }
197 
198 vdev_ops_t vdev_file_ops = {
199 	vdev_file_open,
200 	vdev_file_close,
201 	vdev_default_asize,
202 	vdev_file_io_start,
203 	vdev_file_io_done,
204 	NULL,
205 	VDEV_TYPE_FILE,		/* name of this vdev type */
206 	B_TRUE			/* leaf vdev */
207 };
208 
209 /*
210  * From userland we access disks just like files.
211  */
212 #ifndef _KERNEL
213 
214 vdev_ops_t vdev_disk_ops = {
215 	vdev_file_open,
216 	vdev_file_close,
217 	vdev_default_asize,
218 	vdev_file_io_start,
219 	vdev_file_io_done,
220 	NULL,
221 	VDEV_TYPE_DISK,		/* name of this vdev type */
222 	B_TRUE			/* leaf vdev */
223 };
224 
225 #endif
226