xref: /freebsd/sys/ufs/ffs/ffs_suspend.c (revision 2da0fcde21e0b90384f61fd50b87ea7dbd820233)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2012 The FreeBSD Foundation
5  * All rights reserved.
6  *
7  * This software was developed by Edward Tomasz Napierala under sponsorship
8  * from the FreeBSD Foundation.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  * $FreeBSD$
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/buf.h>
40 #include <sys/ioccom.h>
41 #include <sys/mount.h>
42 #include <sys/vnode.h>
43 #include <sys/conf.h>
44 #include <sys/jail.h>
45 #include <sys/sx.h>
46 
47 #include <security/mac/mac_framework.h>
48 
49 #include <ufs/ufs/extattr.h>
50 #include <ufs/ufs/quota.h>
51 #include <ufs/ufs/ufsmount.h>
52 #include <ufs/ufs/inode.h>
53 
54 #include <ufs/ffs/fs.h>
55 #include <ufs/ffs/ffs_extern.h>
56 
57 static d_open_t ffs_susp_open;
58 static d_write_t ffs_susp_rdwr;
59 static d_ioctl_t ffs_susp_ioctl;
60 
61 static struct cdevsw ffs_susp_cdevsw = {
62 	.d_version =	D_VERSION,
63 	.d_open =	ffs_susp_open,
64 	.d_read =	ffs_susp_rdwr,
65 	.d_write =	ffs_susp_rdwr,
66 	.d_ioctl =	ffs_susp_ioctl,
67 	.d_name =	"ffs_susp",
68 };
69 
70 static struct cdev *ffs_susp_dev;
71 static struct sx ffs_susp_lock;
72 
73 static int
74 ffs_susp_suspended(struct mount *mp)
75 {
76 	struct ufsmount *ump;
77 
78 	sx_assert(&ffs_susp_lock, SA_LOCKED);
79 
80 	ump = VFSTOUFS(mp);
81 	if ((ump->um_flags & UM_WRITESUSPENDED) != 0)
82 		return (1);
83 	return (0);
84 }
85 
86 static int
87 ffs_susp_open(struct cdev *dev __unused, int flags __unused,
88     int fmt __unused, struct thread *td __unused)
89 {
90 
91 	return (0);
92 }
93 
94 static int
95 ffs_susp_rdwr(struct cdev *dev, struct uio *uio, int ioflag)
96 {
97 	int error, i;
98 	struct vnode *devvp;
99 	struct mount *mp;
100 	struct ufsmount *ump;
101 	struct buf *bp;
102 	void *base;
103 	size_t len;
104 	ssize_t cnt;
105 	struct fs *fs;
106 
107 	sx_slock(&ffs_susp_lock);
108 
109 	error = devfs_get_cdevpriv((void **)&mp);
110 	if (error != 0) {
111 		sx_sunlock(&ffs_susp_lock);
112 		return (ENXIO);
113 	}
114 
115 	ump = VFSTOUFS(mp);
116 	devvp = ump->um_devvp;
117 	fs = ump->um_fs;
118 
119 	if (ffs_susp_suspended(mp) == 0) {
120 		sx_sunlock(&ffs_susp_lock);
121 		return (ENXIO);
122 	}
123 
124 	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
125 	    ("neither UIO_READ or UIO_WRITE"));
126 	KASSERT(uio->uio_segflg == UIO_USERSPACE,
127 	    ("uio->uio_segflg != UIO_USERSPACE"));
128 
129 	cnt = uio->uio_resid;
130 
131 	for (i = 0; i < uio->uio_iovcnt; i++) {
132 		while (uio->uio_iov[i].iov_len) {
133 			base = uio->uio_iov[i].iov_base;
134 			len = uio->uio_iov[i].iov_len;
135 			if (len > fs->fs_bsize)
136 				len = fs->fs_bsize;
137 			if (fragoff(fs, uio->uio_offset) != 0 ||
138 			    fragoff(fs, len) != 0) {
139 				error = EINVAL;
140 				goto out;
141 			}
142 			error = bread(devvp, btodb(uio->uio_offset), len,
143 			    NOCRED, &bp);
144 			if (error != 0)
145 				goto out;
146 			if (uio->uio_rw == UIO_WRITE) {
147 				error = copyin(base, bp->b_data, len);
148 				if (error != 0) {
149 					bp->b_flags |= B_INVAL | B_NOCACHE;
150 					brelse(bp);
151 					goto out;
152 				}
153 				error = bwrite(bp);
154 				if (error != 0)
155 					goto out;
156 			} else {
157 				error = copyout(bp->b_data, base, len);
158 				brelse(bp);
159 				if (error != 0)
160 					goto out;
161 			}
162 			uio->uio_iov[i].iov_base =
163 			    (char *)uio->uio_iov[i].iov_base + len;
164 			uio->uio_iov[i].iov_len -= len;
165 			uio->uio_resid -= len;
166 			uio->uio_offset += len;
167 		}
168 	}
169 
170 out:
171 	sx_sunlock(&ffs_susp_lock);
172 
173 	if (uio->uio_resid < cnt)
174 		return (0);
175 
176 	return (error);
177 }
178 
179 static int
180 ffs_susp_suspend(struct mount *mp)
181 {
182 	struct ufsmount *ump;
183 	int error;
184 
185 	sx_assert(&ffs_susp_lock, SA_XLOCKED);
186 
187 	if (!ffs_own_mount(mp))
188 		return (EINVAL);
189 	if (ffs_susp_suspended(mp))
190 		return (EBUSY);
191 
192 	ump = VFSTOUFS(mp);
193 
194 	/*
195 	 * Make sure the calling thread is permitted to access the mounted
196 	 * device.  The permissions can change after we unlock the vnode;
197 	 * it's harmless.
198 	 */
199 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
200 	error = VOP_ACCESS(ump->um_devvp, VREAD | VWRITE,
201 	    curthread->td_ucred, curthread);
202 	VOP_UNLOCK(ump->um_devvp, 0);
203 	if (error != 0)
204 		return (error);
205 #ifdef MAC
206 	if (mac_mount_check_stat(curthread->td_ucred, mp) != 0)
207 		return (EPERM);
208 #endif
209 
210 	if ((error = vfs_write_suspend(mp, VS_SKIP_UNMOUNT)) != 0)
211 		return (error);
212 
213 	ump->um_flags |= UM_WRITESUSPENDED;
214 
215 	return (0);
216 }
217 
218 static void
219 ffs_susp_dtor(void *data)
220 {
221 	struct fs *fs;
222 	struct ufsmount *ump;
223 	struct mount *mp;
224 	int error;
225 
226 	sx_xlock(&ffs_susp_lock);
227 
228 	mp = (struct mount *)data;
229 	ump = VFSTOUFS(mp);
230 	fs = ump->um_fs;
231 
232 	if (ffs_susp_suspended(mp) == 0) {
233 		sx_xunlock(&ffs_susp_lock);
234 		return;
235 	}
236 
237 	KASSERT((mp->mnt_kern_flag & MNTK_SUSPEND) != 0,
238 	    ("MNTK_SUSPEND not set"));
239 
240 	error = ffs_reload(mp, curthread, FFSR_FORCE | FFSR_UNSUSPEND);
241 	if (error != 0)
242 		panic("failed to unsuspend writes on %s", fs->fs_fsmnt);
243 
244 	/*
245 	 * XXX: The status is kept per-process; the vfs_write_resume() routine
246 	 * 	asserts that the resuming thread is the same one that called
247 	 * 	vfs_write_suspend().  The cdevpriv data, however, is attached
248 	 * 	to the file descriptor, e.g. is inherited during fork.  Thus,
249 	 * 	it's possible that the resuming process will be different from
250 	 * 	the one that started the suspension.
251 	 *
252 	 * 	Work around by fooling the check in vfs_write_resume().
253 	 */
254 	mp->mnt_susp_owner = curthread;
255 
256 	vfs_write_resume(mp, 0);
257 	vfs_unbusy(mp);
258 	ump->um_flags &= ~UM_WRITESUSPENDED;
259 
260 	sx_xunlock(&ffs_susp_lock);
261 }
262 
263 static int
264 ffs_susp_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
265     struct thread *td)
266 {
267 	struct mount *mp;
268 	fsid_t *fsidp;
269 	int error;
270 
271 	/*
272 	 * No suspend inside the jail.  Allowing it would require making
273 	 * sure that e.g. the devfs ruleset for that jail permits access
274 	 * to the devvp.
275 	 */
276 	if (jailed(td->td_ucred))
277 		return (EPERM);
278 
279 	sx_xlock(&ffs_susp_lock);
280 
281 	switch (cmd) {
282 	case UFSSUSPEND:
283 		fsidp = (fsid_t *)addr;
284 		mp = vfs_getvfs(fsidp);
285 		if (mp == NULL) {
286 			error = ENOENT;
287 			break;
288 		}
289 		error = vfs_busy(mp, 0);
290 		vfs_rel(mp);
291 		if (error != 0)
292 			break;
293 		error = ffs_susp_suspend(mp);
294 		if (error != 0) {
295 			vfs_unbusy(mp);
296 			break;
297 		}
298 		error = devfs_set_cdevpriv(mp, ffs_susp_dtor);
299 		KASSERT(error == 0, ("devfs_set_cdevpriv failed"));
300 		break;
301 	case UFSRESUME:
302 		error = devfs_get_cdevpriv((void **)&mp);
303 		if (error != 0)
304 			break;
305 		/*
306 		 * This calls ffs_susp_dtor, which in turn unsuspends the fs.
307 		 * The dtor expects to be called without lock held, because
308 		 * sometimes it's called from here, and sometimes due to the
309 		 * file being closed or process exiting.
310 		 */
311 		sx_xunlock(&ffs_susp_lock);
312 		devfs_clear_cdevpriv();
313 		return (0);
314 	default:
315 		error = ENXIO;
316 		break;
317 	}
318 
319 	sx_xunlock(&ffs_susp_lock);
320 
321 	return (error);
322 }
323 
324 void
325 ffs_susp_initialize(void)
326 {
327 
328 	sx_init(&ffs_susp_lock, "ffs_susp");
329 	ffs_susp_dev = make_dev(&ffs_susp_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600,
330 	    "ufssuspend");
331 }
332 
333 void
334 ffs_susp_uninitialize(void)
335 {
336 
337 	destroy_dev(ffs_susp_dev);
338 	sx_destroy(&ffs_susp_lock);
339 }
340