xref: /freebsd/sys/ufs/ffs/ffs_suspend.c (revision e14ddd1f16e7e5788392c50de21ea7c927e0690c)
1 /*-
2  * Copyright (c) 2012 The FreeBSD Foundation
3  * All rights reserved.
4  *
5  * This software was developed by Edward Tomasz Napierala under sponsorship
6  * from the FreeBSD Foundation.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  * $FreeBSD$
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/ioccom.h>
38 #include <sys/mount.h>
39 #include <sys/vnode.h>
40 #include <sys/conf.h>
41 #include <sys/jail.h>
42 #include <sys/sx.h>
43 
44 #include <security/mac/mac_framework.h>
45 
46 #include <ufs/ufs/extattr.h>
47 #include <ufs/ufs/quota.h>
48 #include <ufs/ufs/ufsmount.h>
49 #include <ufs/ufs/inode.h>
50 
51 #include <ufs/ffs/fs.h>
52 #include <ufs/ffs/ffs_extern.h>
53 
54 static d_open_t ffs_susp_open;
55 static d_write_t ffs_susp_rdwr;
56 static d_ioctl_t ffs_susp_ioctl;
57 
58 static struct cdevsw ffs_susp_cdevsw = {
59 	.d_version =	D_VERSION,
60 	.d_open =	ffs_susp_open,
61 	.d_read =	ffs_susp_rdwr,
62 	.d_write =	ffs_susp_rdwr,
63 	.d_ioctl =	ffs_susp_ioctl,
64 	.d_name =	"ffs_susp",
65 };
66 
67 static struct cdev *ffs_susp_dev;
68 static struct sx ffs_susp_lock;
69 
70 static int
71 ffs_susp_suspended(struct mount *mp)
72 {
73 	struct ufsmount *ump;
74 
75 	sx_assert(&ffs_susp_lock, SA_LOCKED);
76 
77 	ump = VFSTOUFS(mp);
78 	if (ump->um_writesuspended)
79 		return (1);
80 	return (0);
81 }
82 
83 static int
84 ffs_susp_open(struct cdev *dev __unused, int flags __unused,
85     int fmt __unused, struct thread *td __unused)
86 {
87 
88 	return (0);
89 }
90 
91 static int
92 ffs_susp_rdwr(struct cdev *dev, struct uio *uio, int ioflag)
93 {
94 	int error, i;
95 	struct vnode *devvp;
96 	struct mount *mp;
97 	struct ufsmount *ump;
98 	struct buf *bp;
99 	void *base;
100 	size_t len;
101 	ssize_t cnt;
102 	struct fs *fs;
103 
104 	sx_slock(&ffs_susp_lock);
105 
106 	error = devfs_get_cdevpriv((void **)&mp);
107 	if (error != 0) {
108 		sx_sunlock(&ffs_susp_lock);
109 		return (ENXIO);
110 	}
111 
112 	ump = VFSTOUFS(mp);
113 	devvp = ump->um_devvp;
114 	fs = ump->um_fs;
115 
116 	if (ffs_susp_suspended(mp) == 0) {
117 		sx_sunlock(&ffs_susp_lock);
118 		return (ENXIO);
119 	}
120 
121 	KASSERT(uio->uio_rw == UIO_READ || uio->uio_rw == UIO_WRITE,
122 	    ("neither UIO_READ or UIO_WRITE"));
123 	KASSERT(uio->uio_segflg == UIO_USERSPACE,
124 	    ("uio->uio_segflg != UIO_USERSPACE"));
125 
126 	cnt = uio->uio_resid;
127 
128 	for (i = 0; i < uio->uio_iovcnt; i++) {
129 		while (uio->uio_iov[i].iov_len) {
130 			base = uio->uio_iov[i].iov_base;
131 			len = uio->uio_iov[i].iov_len;
132 			if (len > fs->fs_bsize)
133 				len = fs->fs_bsize;
134 			if (fragoff(fs, uio->uio_offset) != 0 ||
135 			    fragoff(fs, len) != 0) {
136 				error = EINVAL;
137 				goto out;
138 			}
139 			error = bread(devvp, btodb(uio->uio_offset), len,
140 			    NOCRED, &bp);
141 			if (error != 0)
142 				goto out;
143 			if (uio->uio_rw == UIO_WRITE) {
144 				error = copyin(base, bp->b_data, len);
145 				if (error != 0) {
146 					bp->b_flags |= B_INVAL | B_NOCACHE;
147 					brelse(bp);
148 					goto out;
149 				}
150 				error = bwrite(bp);
151 				if (error != 0)
152 					goto out;
153 			} else {
154 				error = copyout(bp->b_data, base, len);
155 				brelse(bp);
156 				if (error != 0)
157 					goto out;
158 			}
159 			uio->uio_iov[i].iov_base =
160 			    (char *)uio->uio_iov[i].iov_base + len;
161 			uio->uio_iov[i].iov_len -= len;
162 			uio->uio_resid -= len;
163 			uio->uio_offset += len;
164 		}
165 	}
166 
167 out:
168 	sx_sunlock(&ffs_susp_lock);
169 
170 	if (uio->uio_resid < cnt)
171 		return (0);
172 
173 	return (error);
174 }
175 
176 static int
177 ffs_susp_suspend(struct mount *mp)
178 {
179 	struct fs *fs;
180 	struct ufsmount *ump;
181 	int error;
182 
183 	sx_assert(&ffs_susp_lock, SA_XLOCKED);
184 
185 	if (!ffs_own_mount(mp))
186 		return (EINVAL);
187 	if (ffs_susp_suspended(mp))
188 		return (EBUSY);
189 
190 	ump = VFSTOUFS(mp);
191 	fs = ump->um_fs;
192 
193 	/*
194 	 * Make sure the calling thread is permitted to access the mounted
195 	 * device.  The permissions can change after we unlock the vnode;
196 	 * it's harmless.
197 	 */
198 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
199 	error = VOP_ACCESS(ump->um_devvp, VREAD | VWRITE,
200 	    curthread->td_ucred, curthread);
201 	VOP_UNLOCK(ump->um_devvp, 0);
202 	if (error != 0)
203 		return (error);
204 #ifdef MAC
205 	if (mac_mount_check_stat(curthread->td_ucred, mp) != 0)
206 		return (EPERM);
207 #endif
208 
209 	if ((error = vfs_write_suspend(mp)) != 0)
210 		return (error);
211 
212 	ump->um_writesuspended = 1;
213 
214 	return (0);
215 }
216 
217 static void
218 ffs_susp_dtor(void *data)
219 {
220 	struct fs *fs;
221 	struct ufsmount *ump;
222 	struct mount *mp;
223 	int error;
224 
225 	sx_xlock(&ffs_susp_lock);
226 
227 	mp = (struct mount *)data;
228 	ump = VFSTOUFS(mp);
229 	fs = ump->um_fs;
230 
231 	if (ffs_susp_suspended(mp) == 0) {
232 		sx_xunlock(&ffs_susp_lock);
233 		return;
234 	}
235 
236 	KASSERT((mp->mnt_kern_flag & MNTK_SUSPEND) != 0,
237 	    ("MNTK_SUSPEND not set"));
238 
239 	error = ffs_reload(mp, curthread, 1);
240 	if (error != 0)
241 		panic("failed to unsuspend writes on %s", fs->fs_fsmnt);
242 
243 	/*
244 	 * XXX: The status is kept per-process; the vfs_write_resume() routine
245 	 * 	asserts that the resuming thread is the same one that called
246 	 * 	vfs_write_suspend().  The cdevpriv data, however, is attached
247 	 * 	to the file descriptor, e.g. is inherited during fork.  Thus,
248 	 * 	it's possible that the resuming process will be different from
249 	 * 	the one that started the suspension.
250 	 *
251 	 * 	Work around by fooling the check in vfs_write_resume().
252 	 */
253 	mp->mnt_susp_owner = curthread;
254 
255 	vfs_write_resume(mp, 0);
256 	vfs_unbusy(mp);
257 	ump->um_writesuspended = 0;
258 
259 	sx_xunlock(&ffs_susp_lock);
260 }
261 
262 static int
263 ffs_susp_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
264     struct thread *td)
265 {
266 	struct mount *mp;
267 	fsid_t *fsidp;
268 	int error;
269 
270 	/*
271 	 * No suspend inside the jail.  Allowing it would require making
272 	 * sure that e.g. the devfs ruleset for that jail permits access
273 	 * to the devvp.
274 	 */
275 	if (jailed(td->td_ucred))
276 		return (EPERM);
277 
278 	sx_xlock(&ffs_susp_lock);
279 
280 	switch (cmd) {
281 	case UFSSUSPEND:
282 		fsidp = (fsid_t *)addr;
283 		mp = vfs_getvfs(fsidp);
284 		if (mp == NULL) {
285 			error = ENOENT;
286 			break;
287 		}
288 		error = vfs_busy(mp, 0);
289 		vfs_rel(mp);
290 		if (error != 0)
291 			break;
292 		error = ffs_susp_suspend(mp);
293 		if (error != 0) {
294 			vfs_unbusy(mp);
295 			break;
296 		}
297 		error = devfs_set_cdevpriv(mp, ffs_susp_dtor);
298 		KASSERT(error == 0, ("devfs_set_cdevpriv failed"));
299 		break;
300 	case UFSRESUME:
301 		error = devfs_get_cdevpriv((void **)&mp);
302 		if (error != 0)
303 			break;
304 		/*
305 		 * This calls ffs_susp_dtor, which in turn unsuspends the fs.
306 		 * The dtor expects to be called without lock held, because
307 		 * sometimes it's called from here, and sometimes due to the
308 		 * file being closed or process exiting.
309 		 */
310 		sx_xunlock(&ffs_susp_lock);
311 		devfs_clear_cdevpriv();
312 		return (0);
313 	default:
314 		error = ENXIO;
315 		break;
316 	}
317 
318 	sx_xunlock(&ffs_susp_lock);
319 
320 	return (error);
321 }
322 
323 void
324 ffs_susp_initialize(void)
325 {
326 
327 	sx_init(&ffs_susp_lock, "ffs_susp");
328 	ffs_susp_dev = make_dev(&ffs_susp_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600,
329 	    "ufssuspend");
330 }
331 
332 void
333 ffs_susp_uninitialize(void)
334 {
335 
336 	destroy_dev(ffs_susp_dev);
337 	sx_destroy(&ffs_susp_lock);
338 }
339