xref: /illumos-gate/usr/src/uts/common/fs/zfs/vdev_file.c (revision 24da5b34f49324ed742a340010ed5bd3d4e06625)
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 	/*
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, spa_mode | FOFFMAX,
65 	    0, &vp, 0, 0, rootdir);
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 	/*
85 	 * Determine the physical size of the file.
86 	 */
87 	vattr.va_mask = AT_SIZE;
88 	error = VOP_GETATTR(vp, &vattr, 0, kcred);
89 	if (error) {
90 		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
91 		return (error);
92 	}
93 
94 	*psize = vattr.va_size;
95 	*ashift = SPA_MINBLOCKSHIFT;
96 
97 	return (0);
98 }
99 
100 static void
101 vdev_file_close(vdev_t *vd)
102 {
103 	vdev_file_t *vf = vd->vdev_tsd;
104 
105 	if (vf == NULL)
106 		return;
107 
108 	if (vf->vf_vnode != NULL) {
109 		(void) VOP_PUTPAGE(vf->vf_vnode, 0, 0, B_INVAL, kcred);
110 		(void) VOP_CLOSE(vf->vf_vnode, spa_mode, 1, 0, kcred);
111 		VN_RELE(vf->vf_vnode);
112 	}
113 
114 	kmem_free(vf, sizeof (vdev_file_t));
115 	vd->vdev_tsd = NULL;
116 }
117 
118 static void
119 vdev_file_io_start(zio_t *zio)
120 {
121 	vdev_t *vd = zio->io_vd;
122 	vdev_file_t *vf = vd->vdev_tsd;
123 	ssize_t resid;
124 	int error;
125 
126 	if (zio->io_type == ZIO_TYPE_IOCTL) {
127 		zio_vdev_io_bypass(zio);
128 
129 		/* XXPOLICY */
130 		if (vdev_is_dead(vd)) {
131 			zio->io_error = ENXIO;
132 			zio_next_stage_async(zio);
133 			return;
134 		}
135 
136 		switch (zio->io_cmd) {
137 		case DKIOCFLUSHWRITECACHE:
138 			zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC | FDSYNC,
139 			    kcred);
140 			dprintf("fsync(%s) = %d\n", vdev_description(vd),
141 			    zio->io_error);
142 			break;
143 		default:
144 			zio->io_error = ENOTSUP;
145 		}
146 
147 		zio_next_stage_async(zio);
148 		return;
149 	}
150 
151 	/*
152 	 * In the kernel, don't bother double-caching, but in userland,
153 	 * we want to test the vdev_cache code.
154 	 */
155 #ifndef _KERNEL
156 	if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio) == 0)
157 		return;
158 #endif
159 
160 	if ((zio = vdev_queue_io(zio)) == NULL)
161 		return;
162 
163 	/* XXPOLICY */
164 	error = vdev_is_dead(vd) ? ENXIO : vdev_error_inject(vd, zio);
165 	if (error) {
166 		zio->io_error = error;
167 		zio_next_stage_async(zio);
168 		return;
169 	}
170 
171 	zio->io_error = vn_rdwr(zio->io_type == ZIO_TYPE_READ ?
172 	    UIO_READ : UIO_WRITE, vf->vf_vnode, zio->io_data,
173 	    zio->io_size, zio->io_offset, UIO_SYSSPACE,
174 	    0, RLIM64_INFINITY, kcred, &resid);
175 
176 	if (resid != 0 && zio->io_error == 0)
177 		zio->io_error = ENOSPC;
178 
179 	zio_next_stage_async(zio);
180 }
181 
182 static void
183 vdev_file_io_done(zio_t *zio)
184 {
185 	vdev_queue_io_done(zio);
186 
187 #ifndef _KERNEL
188 	if (zio->io_type == ZIO_TYPE_WRITE)
189 		vdev_cache_write(zio);
190 #endif
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