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 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
24 */
25
26#include <sys/zfs_context.h>
27#include <sys/spa.h>
28#include <sys/vdev_file.h>
29#include <sys/vdev_impl.h>
30#include <sys/zio.h>
31#include <sys/fs/zfs.h>
32#include <sys/fm/fs/zfs.h>
33#include <sys/abd.h>
34
35/*
36 * Virtual device vector for files.
37 */
38
39static taskq_t *vdev_file_taskq;
40
41void
42vdev_file_init(void)
43{
44	vdev_file_taskq = taskq_create("z_vdev_file", MAX(max_ncpus, 16),
45	    minclsyspri, max_ncpus, INT_MAX, 0);
46}
47
48void
49vdev_file_fini(void)
50{
51	taskq_destroy(vdev_file_taskq);
52}
53
54static void
55vdev_file_hold(vdev_t *vd)
56{
57	ASSERT(vd->vdev_path != NULL);
58}
59
60static void
61vdev_file_rele(vdev_t *vd)
62{
63	ASSERT(vd->vdev_path != NULL);
64}
65
66static int
67vdev_file_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
68    uint64_t *logical_ashift, uint64_t *physical_ashift)
69{
70	vdev_file_t *vf;
71	vnode_t *vp;
72	vattr_t vattr;
73	int error;
74
75	/* Rotational optimizations only make sense on block devices */
76	vd->vdev_nonrot = B_TRUE;
77
78	/*
79	 * We must have a pathname, and it must be absolute.
80	 */
81	if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
82		vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
83		return (SET_ERROR(EINVAL));
84	}
85
86	/*
87	 * Reopen the device if it's not currently open.  Otherwise,
88	 * just update the physical size of the device.
89	 */
90	if (vd->vdev_tsd != NULL) {
91		ASSERT(vd->vdev_reopening);
92		vf = vd->vdev_tsd;
93		vp = vf->vf_vnode;
94		goto skip_open;
95	}
96
97	vf = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_file_t), KM_SLEEP);
98
99	/*
100	 * We always open the files from the root of the global zone, even if
101	 * we're in a local zone.  If the user has gotten to this point, the
102	 * administrator has already decided that the pool should be available
103	 * to local zone users, so the underlying devices should be as well.
104	 */
105	ASSERT(vd->vdev_path != NULL && vd->vdev_path[0] == '/');
106	error = vn_openat(vd->vdev_path + 1, UIO_SYSSPACE,
107	    spa_mode(vd->vdev_spa) | FOFFMAX, 0, &vp, 0, 0, rootdir, -1);
108
109	if (error) {
110		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
111		kmem_free(vd->vdev_tsd, sizeof (vdev_file_t));
112		vd->vdev_tsd = NULL;
113		return (error);
114	}
115
116	vf->vf_vnode = vp;
117
118#ifdef _KERNEL
119	/*
120	 * Make sure it's a regular file.
121	 */
122	if (vp->v_type != VREG) {
123#ifdef __FreeBSD__
124		(void) VOP_CLOSE(vp, spa_mode(vd->vdev_spa), 1, 0, kcred, NULL);
125#endif
126		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
127#ifdef __FreeBSD__
128		kmem_free(vd->vdev_tsd, sizeof (vdev_file_t));
129		vd->vdev_tsd = NULL;
130#endif
131		return (SET_ERROR(ENODEV));
132	}
133#endif	/* _KERNEL */
134
135skip_open:
136	/*
137	 * Determine the physical size of the file.
138	 */
139	vattr.va_mask = AT_SIZE;
140	vn_lock(vp, LK_SHARED | LK_RETRY);
141	error = VOP_GETATTR(vp, &vattr, kcred);
142	VOP_UNLOCK(vp, 0);
143	if (error) {
144		(void) VOP_CLOSE(vp, spa_mode(vd->vdev_spa), 1, 0, kcred, NULL);
145		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
146		kmem_free(vd->vdev_tsd, sizeof (vdev_file_t));
147		vd->vdev_tsd = NULL;
148		return (error);
149	}
150
151	vd->vdev_notrim = B_TRUE;
152
153	*max_psize = *psize = vattr.va_size;
154	*logical_ashift = SPA_MINBLOCKSHIFT;
155	*physical_ashift = SPA_MINBLOCKSHIFT;
156
157	return (0);
158}
159
160static void
161vdev_file_close(vdev_t *vd)
162{
163	vdev_file_t *vf = vd->vdev_tsd;
164
165	if (vd->vdev_reopening || vf == NULL)
166		return;
167
168	if (vf->vf_vnode != NULL) {
169		(void) VOP_CLOSE(vf->vf_vnode, spa_mode(vd->vdev_spa), 1, 0,
170		    kcred, NULL);
171	}
172
173	vd->vdev_delayed_close = B_FALSE;
174	kmem_free(vf, sizeof (vdev_file_t));
175	vd->vdev_tsd = NULL;
176}
177
178/*
179 * Implements the interrupt side for file vdev types. This routine will be
180 * called when the I/O completes allowing us to transfer the I/O to the
181 * interrupt taskqs. For consistency, the code structure mimics disk vdev
182 * types.
183 */
184static void
185vdev_file_io_intr(zio_t *zio)
186{
187	zio_delay_interrupt(zio);
188}
189
190static void
191vdev_file_io_strategy(void *arg)
192{
193	zio_t *zio = arg;
194	vdev_t *vd = zio->io_vd;
195	vdev_file_t *vf;
196	vnode_t *vp;
197	void *addr;
198	ssize_t resid;
199
200	vf = vd->vdev_tsd;
201	vp = vf->vf_vnode;
202
203	ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE);
204	if (zio->io_type == ZIO_TYPE_READ) {
205		addr = abd_borrow_buf(zio->io_abd, zio->io_size);
206	} else {
207		addr = abd_borrow_buf_copy(zio->io_abd, zio->io_size);
208	}
209
210	zio->io_error = vn_rdwr(zio->io_type == ZIO_TYPE_READ ?
211	    UIO_READ : UIO_WRITE, vp, addr, zio->io_size,
212	    zio->io_offset, UIO_SYSSPACE, 0, RLIM64_INFINITY, kcred, &resid);
213
214	if (zio->io_type == ZIO_TYPE_READ) {
215		abd_return_buf_copy(zio->io_abd, addr, zio->io_size);
216	} else {
217		abd_return_buf(zio->io_abd, addr, zio->io_size);
218	}
219
220	if (resid != 0 && zio->io_error == 0)
221		zio->io_error = ENOSPC;
222
223	vdev_file_io_intr(zio);
224}
225
226static void
227vdev_file_io_start(zio_t *zio)
228{
229	vdev_t *vd = zio->io_vd;
230	vdev_file_t *vf = vd->vdev_tsd;
231
232	if (zio->io_type == ZIO_TYPE_IOCTL) {
233		/* XXPOLICY */
234		if (!vdev_readable(vd)) {
235			zio->io_error = SET_ERROR(ENXIO);
236			zio_interrupt(zio);
237			return;
238		}
239
240		switch (zio->io_cmd) {
241		case DKIOCFLUSHWRITECACHE:
242			zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC | FDSYNC,
243			    kcred, NULL);
244			break;
245		default:
246			zio->io_error = SET_ERROR(ENOTSUP);
247		}
248
249		zio_execute(zio);
250		return;
251	}
252
253	ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE);
254	zio->io_target_timestamp = zio_handle_io_delay(zio);
255
256	VERIFY3U(taskq_dispatch(vdev_file_taskq, vdev_file_io_strategy, zio,
257	    TQ_SLEEP), !=, 0);
258}
259
260/* ARGSUSED */
261static void
262vdev_file_io_done(zio_t *zio)
263{
264}
265
266vdev_ops_t vdev_file_ops = {
267	vdev_file_open,
268	vdev_file_close,
269	vdev_default_asize,
270	vdev_file_io_start,
271	vdev_file_io_done,
272	NULL,
273	NULL,
274	vdev_file_hold,
275	vdev_file_rele,
276	NULL,
277	vdev_default_xlate,
278	VDEV_TYPE_FILE,		/* name of this vdev type */
279	B_TRUE			/* leaf vdev */
280};
281
282/*
283 * From userland we access disks just like files.
284 */
285#ifndef _KERNEL
286
287vdev_ops_t vdev_disk_ops = {
288	vdev_file_open,
289	vdev_file_close,
290	vdev_default_asize,
291	vdev_file_io_start,
292	vdev_file_io_done,
293	NULL,
294	NULL,
295	vdev_file_hold,
296	vdev_file_rele,
297	NULL,
298	vdev_default_xlate,
299	VDEV_TYPE_DISK,		/* name of this vdev type */
300	B_TRUE			/* leaf vdev */
301};
302
303#endif
304