xref: /illumos-gate/usr/src/uts/common/fs/zfs/vdev_file.c (revision ccd81fdda071e031209c777983199d191c35b0a2)
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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  */
24 
25 #include <sys/zfs_context.h>
26 #include <sys/spa.h>
27 #include <sys/vdev_file.h>
28 #include <sys/vdev_impl.h>
29 #include <sys/zio.h>
30 #include <sys/fs/zfs.h>
31 #include <sys/fm/fs/zfs.h>
32 
33 /*
34  * Virtual device vector for files.
35  */
36 
37 static void
38 vdev_file_hold(vdev_t *vd)
39 {
40 	ASSERT(vd->vdev_path != NULL);
41 }
42 
43 static void
44 vdev_file_rele(vdev_t *vd)
45 {
46 	ASSERT(vd->vdev_path != NULL);
47 }
48 
49 static int
50 vdev_file_open(vdev_t *vd, uint64_t *psize, uint64_t *ashift)
51 {
52 	vdev_file_t *vf;
53 	vnode_t *vp;
54 	vattr_t vattr;
55 	int error;
56 
57 	/*
58 	 * We must have a pathname, and it must be absolute.
59 	 */
60 	if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
61 		vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
62 		return (EINVAL);
63 	}
64 
65 	/*
66 	 * Reopen the device if it's not currently open.  Otherwise,
67 	 * just update the physical size of the device.
68 	 */
69 	if (vd->vdev_tsd != NULL) {
70 		ASSERT(vd->vdev_reopening);
71 		vf = vd->vdev_tsd;
72 		goto skip_open;
73 	}
74 
75 	vf = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_file_t), KM_SLEEP);
76 
77 	/*
78 	 * We always open the files from the root of the global zone, even if
79 	 * we're in a local zone.  If the user has gotten to this point, the
80 	 * administrator has already decided that the pool should be available
81 	 * to local zone users, so the underlying devices should be as well.
82 	 */
83 	ASSERT(vd->vdev_path != NULL && vd->vdev_path[0] == '/');
84 	error = vn_openat(vd->vdev_path + 1, UIO_SYSSPACE,
85 	    spa_mode(vd->vdev_spa) | FOFFMAX, 0, &vp, 0, 0, rootdir, -1);
86 
87 	if (error) {
88 		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
89 		return (error);
90 	}
91 
92 	vf->vf_vnode = vp;
93 
94 #ifdef _KERNEL
95 	/*
96 	 * Make sure it's a regular file.
97 	 */
98 	if (vp->v_type != VREG) {
99 		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
100 		return (ENODEV);
101 	}
102 #endif
103 
104 skip_open:
105 	/*
106 	 * Determine the physical size of the file.
107 	 */
108 	vattr.va_mask = AT_SIZE;
109 	error = VOP_GETATTR(vf->vf_vnode, &vattr, 0, kcred, NULL);
110 	if (error) {
111 		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
112 		return (error);
113 	}
114 
115 	*psize = vattr.va_size;
116 	*ashift = SPA_MINBLOCKSHIFT;
117 
118 	return (0);
119 }
120 
121 static void
122 vdev_file_close(vdev_t *vd)
123 {
124 	vdev_file_t *vf = vd->vdev_tsd;
125 
126 	if (vd->vdev_reopening || vf == NULL)
127 		return;
128 
129 	if (vf->vf_vnode != NULL) {
130 		(void) VOP_PUTPAGE(vf->vf_vnode, 0, 0, B_INVAL, kcred, NULL);
131 		(void) VOP_CLOSE(vf->vf_vnode, spa_mode(vd->vdev_spa), 1, 0,
132 		    kcred, NULL);
133 		VN_RELE(vf->vf_vnode);
134 	}
135 
136 	vd->vdev_delayed_close = B_FALSE;
137 	kmem_free(vf, sizeof (vdev_file_t));
138 	vd->vdev_tsd = NULL;
139 }
140 
141 static int
142 vdev_file_io_start(zio_t *zio)
143 {
144 	vdev_t *vd = zio->io_vd;
145 	vdev_file_t *vf = vd->vdev_tsd;
146 	ssize_t resid;
147 
148 	if (zio->io_type == ZIO_TYPE_IOCTL) {
149 		/* XXPOLICY */
150 		if (!vdev_readable(vd)) {
151 			zio->io_error = ENXIO;
152 			return (ZIO_PIPELINE_CONTINUE);
153 		}
154 
155 		switch (zio->io_cmd) {
156 		case DKIOCFLUSHWRITECACHE:
157 			zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC | FDSYNC,
158 			    kcred, NULL);
159 			break;
160 		default:
161 			zio->io_error = ENOTSUP;
162 		}
163 
164 		return (ZIO_PIPELINE_CONTINUE);
165 	}
166 
167 	zio->io_error = vn_rdwr(zio->io_type == ZIO_TYPE_READ ?
168 	    UIO_READ : UIO_WRITE, vf->vf_vnode, zio->io_data,
169 	    zio->io_size, zio->io_offset, UIO_SYSSPACE,
170 	    0, RLIM64_INFINITY, kcred, &resid);
171 
172 	if (resid != 0 && zio->io_error == 0)
173 		zio->io_error = ENOSPC;
174 
175 	zio_interrupt(zio);
176 
177 	return (ZIO_PIPELINE_STOP);
178 }
179 
180 /* ARGSUSED */
181 static void
182 vdev_file_io_done(zio_t *zio)
183 {
184 }
185 
186 vdev_ops_t vdev_file_ops = {
187 	vdev_file_open,
188 	vdev_file_close,
189 	vdev_default_asize,
190 	vdev_file_io_start,
191 	vdev_file_io_done,
192 	NULL,
193 	vdev_file_hold,
194 	vdev_file_rele,
195 	VDEV_TYPE_FILE,		/* name of this vdev type */
196 	B_TRUE			/* leaf vdev */
197 };
198 
199 /*
200  * From userland we access disks just like files.
201  */
202 #ifndef _KERNEL
203 
204 vdev_ops_t vdev_disk_ops = {
205 	vdev_file_open,
206 	vdev_file_close,
207 	vdev_default_asize,
208 	vdev_file_io_start,
209 	vdev_file_io_done,
210 	NULL,
211 	vdev_file_hold,
212 	vdev_file_rele,
213 	VDEV_TYPE_DISK,		/* name of this vdev type */
214 	B_TRUE			/* leaf vdev */
215 };
216 
217 #endif
218