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 * Copyright 2019 Joyent, Inc.
25 */
26
27#include <sys/zfs_context.h>
28#include <sys/spa.h>
29#include <sys/spa_impl.h>
30#include <sys/vdev_file.h>
31#include <sys/vdev_impl.h>
32#include <sys/vdev_trim.h>
33#include <sys/zio.h>
34#include <sys/fs/zfs.h>
35#include <sys/fm/fs/zfs.h>
36#include <sys/abd.h>
37#include <sys/fcntl.h>
38#include <sys/vnode.h>
39
40/*
41 * Virtual device vector for files.
42 */
43
44static void
45vdev_file_hold(vdev_t *vd)
46{
47	ASSERT(vd->vdev_path != NULL);
48}
49
50static void
51vdev_file_rele(vdev_t *vd)
52{
53	ASSERT(vd->vdev_path != NULL);
54}
55
56static int
57vdev_file_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
58    uint64_t *ashift)
59{
60	vdev_file_t *vf;
61	vnode_t *vp;
62	vattr_t vattr;
63	int error;
64
65	/*
66	 * Rotational optimizations only make sense on block devices.
67	 */
68	vd->vdev_nonrot = B_TRUE;
69
70	/*
71	 * Allow TRIM on file based vdevs.  This may not always be supported,
72	 * since it depends on your kernel version and underlying filesystem
73	 * type but it is always safe to attempt.
74	 */
75	vd->vdev_has_trim = B_TRUE;
76
77	/*
78	 * Disable secure TRIM on file based vdevs.  There is no way to
79	 * request this behavior from the underlying filesystem.
80	 */
81	vd->vdev_has_securetrim = B_FALSE;
82
83	/*
84	 * We must have a pathname, and it must be absolute.
85	 */
86	if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
87		vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
88		return (SET_ERROR(EINVAL));
89	}
90
91	/*
92	 * Reopen the device if it's not currently open.  Otherwise,
93	 * just update the physical size of the device.
94	 */
95	if (vd->vdev_tsd != NULL) {
96		ASSERT(vd->vdev_reopening);
97		vf = vd->vdev_tsd;
98		goto skip_open;
99	}
100
101	vf = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_file_t), KM_SLEEP);
102
103	/*
104	 * We always open the files from the root of the global zone, even if
105	 * we're in a local zone.  If the user has gotten to this point, the
106	 * administrator has already decided that the pool should be available
107	 * to local zone users, so the underlying devices should be as well.
108	 */
109	ASSERT(vd->vdev_path != NULL && vd->vdev_path[0] == '/');
110	error = vn_openat(vd->vdev_path + 1, UIO_SYSSPACE,
111	    spa_mode(vd->vdev_spa) | FOFFMAX, 0, &vp, 0, 0, rootdir, -1);
112
113	if (error) {
114		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
115		return (error);
116	}
117
118	vf->vf_vnode = vp;
119
120#ifdef _KERNEL
121	/*
122	 * Make sure it's a regular file.
123	 */
124	if (vp->v_type != VREG) {
125		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
126		return (SET_ERROR(ENODEV));
127	}
128#endif
129
130skip_open:
131	/*
132	 * Determine the physical size of the file.
133	 */
134	vattr.va_mask = AT_SIZE;
135	error = VOP_GETATTR(vf->vf_vnode, &vattr, 0, kcred, NULL);
136	if (error) {
137		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
138		return (error);
139	}
140
141	*max_psize = *psize = vattr.va_size;
142	*ashift = SPA_MINBLOCKSHIFT;
143
144	return (0);
145}
146
147static void
148vdev_file_close(vdev_t *vd)
149{
150	vdev_file_t *vf = vd->vdev_tsd;
151
152	if (vd->vdev_reopening || vf == NULL)
153		return;
154
155	if (vf->vf_vnode != NULL) {
156		(void) VOP_PUTPAGE(vf->vf_vnode, 0, 0, B_INVAL, kcred, NULL);
157		(void) VOP_CLOSE(vf->vf_vnode, spa_mode(vd->vdev_spa), 1, 0,
158		    kcred, NULL);
159		VN_RELE(vf->vf_vnode);
160	}
161
162	vd->vdev_delayed_close = B_FALSE;
163	kmem_free(vf, sizeof (vdev_file_t));
164	vd->vdev_tsd = NULL;
165}
166
167/*
168 * Implements the interrupt side for file vdev types. This routine will be
169 * called when the I/O completes allowing us to transfer the I/O to the
170 * interrupt taskqs. For consistency, the code structure mimics disk vdev
171 * types.
172 */
173static int
174vdev_file_io_intr(buf_t *bp)
175{
176	vdev_buf_t *vb = (vdev_buf_t *)bp;
177	zio_t *zio = vb->vb_io;
178
179	zio->io_error = (geterror(bp) != 0 ? EIO : 0);
180	if (zio->io_error == 0 && bp->b_resid != 0)
181		zio->io_error = SET_ERROR(ENOSPC);
182
183	if (zio->io_type == ZIO_TYPE_READ) {
184		abd_return_buf_copy(zio->io_abd, bp->b_un.b_addr, zio->io_size);
185	} else {
186		abd_return_buf(zio->io_abd, bp->b_un.b_addr, zio->io_size);
187	}
188
189	kmem_free(vb, sizeof (vdev_buf_t));
190	zio_delay_interrupt(zio);
191	return (0);
192}
193
194static void
195vdev_file_io_strategy(void *arg)
196{
197	buf_t *bp = arg;
198	vnode_t *vp = bp->b_private;
199	ssize_t resid;
200	int error;
201
202	error = vn_rdwr((bp->b_flags & B_READ) ? UIO_READ : UIO_WRITE,
203	    vp, bp->b_un.b_addr, bp->b_bcount, ldbtob(bp->b_lblkno),
204	    UIO_SYSSPACE, 0, RLIM64_INFINITY, kcred, &resid);
205
206	if (error == 0) {
207		bp->b_resid = resid;
208		biodone(bp);
209	} else {
210		bioerror(bp, error);
211		biodone(bp);
212	}
213}
214
215static void
216vdev_file_io_start(zio_t *zio)
217{
218	vdev_t *vd = zio->io_vd;
219	vdev_file_t *vf = vd->vdev_tsd;
220	vdev_buf_t *vb;
221	buf_t *bp;
222
223	if (zio->io_type == ZIO_TYPE_IOCTL) {
224		/* XXPOLICY */
225		if (!vdev_readable(vd)) {
226			zio->io_error = SET_ERROR(ENXIO);
227			zio_interrupt(zio);
228			return;
229		}
230
231		switch (zio->io_cmd) {
232		case DKIOCFLUSHWRITECACHE:
233			zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC | FDSYNC,
234			    kcred, NULL);
235			break;
236		default:
237			zio->io_error = SET_ERROR(ENOTSUP);
238		}
239
240		zio_execute(zio);
241		return;
242	} else if (zio->io_type == ZIO_TYPE_TRIM) {
243		struct flock64 flck;
244
245		ASSERT3U(zio->io_size, !=, 0);
246		bzero(&flck, sizeof (flck));
247		flck.l_type = F_FREESP;
248		flck.l_start = zio->io_offset;
249		flck.l_len = zio->io_size;
250		flck.l_whence = 0;
251
252		zio->io_error = VOP_SPACE(vf->vf_vnode, F_FREESP, &flck,
253		    0, 0, kcred, NULL);
254
255		zio_execute(zio);
256		return;
257	}
258
259	ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE);
260	zio->io_target_timestamp = zio_handle_io_delay(zio);
261
262	vb = kmem_alloc(sizeof (vdev_buf_t), KM_SLEEP);
263
264	vb->vb_io = zio;
265	bp = &vb->vb_buf;
266
267	bioinit(bp);
268	bp->b_flags = (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
269	bp->b_bcount = zio->io_size;
270
271	if (zio->io_type == ZIO_TYPE_READ) {
272		bp->b_un.b_addr =
273		    abd_borrow_buf(zio->io_abd, zio->io_size);
274	} else {
275		bp->b_un.b_addr =
276		    abd_borrow_buf_copy(zio->io_abd, zio->io_size);
277	}
278
279	bp->b_lblkno = lbtodb(zio->io_offset);
280	bp->b_bufsize = zio->io_size;
281	bp->b_private = vf->vf_vnode;
282	bp->b_iodone = vdev_file_io_intr;
283
284	VERIFY3U(taskq_dispatch(system_taskq, vdev_file_io_strategy, bp,
285	    TQ_SLEEP), !=, TASKQID_INVALID);
286}
287
288/* ARGSUSED */
289static void
290vdev_file_io_done(zio_t *zio)
291{
292}
293
294vdev_ops_t vdev_file_ops = {
295	.vdev_op_open = vdev_file_open,
296	.vdev_op_close = vdev_file_close,
297	.vdev_op_asize = vdev_default_asize,
298	.vdev_op_io_start = vdev_file_io_start,
299	.vdev_op_io_done = vdev_file_io_done,
300	.vdev_op_state_change = NULL,
301	.vdev_op_need_resilver = NULL,
302	.vdev_op_hold = vdev_file_hold,
303	.vdev_op_rele = vdev_file_rele,
304	.vdev_op_remap = NULL,
305	.vdev_op_xlate = vdev_default_xlate,
306	.vdev_op_type = VDEV_TYPE_FILE,		/* name of this vdev type */
307	.vdev_op_leaf = B_TRUE			/* leaf vdev */
308};
309
310/*
311 * From userland we access disks just like files.
312 */
313#ifndef _KERNEL
314
315vdev_ops_t vdev_disk_ops = {
316	.vdev_op_open = vdev_file_open,
317	.vdev_op_close = vdev_file_close,
318	.vdev_op_asize = vdev_default_asize,
319	.vdev_op_io_start = vdev_file_io_start,
320	.vdev_op_io_done = vdev_file_io_done,
321	.vdev_op_state_change = NULL,
322	.vdev_op_need_resilver = NULL,
323	.vdev_op_hold = vdev_file_hold,
324	.vdev_op_rele = vdev_file_rele,
325	.vdev_op_remap = NULL,
326	.vdev_op_xlate = vdev_default_xlate,
327	.vdev_op_type = VDEV_TYPE_DISK,		/* name of this vdev type */
328	.vdev_op_leaf = B_TRUE			/* leaf vdev */
329};
330
331#endif
332