nfs_lock.c revision 154316 
1/*-
2 * Copyright (c) 1997 Berkeley Software Design, Inc. All rights reserved.
3 *
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
6 * are met:
7 * 1. Redistributions of source code must retain the above copyright
8 *    notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 *    notice, this list of conditions and the following disclaimer in the
11 *    documentation and/or other materials provided with the distribution.
12 * 3. Berkeley Software Design Inc's name may not be used to endorse or
13 *    promote products derived from this software without specific prior
14 *    written permission.
15 *
16 * THIS SOFTWARE IS PROVIDED BY BERKELEY SOFTWARE DESIGN INC ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED.  IN NO EVENT SHALL BERKELEY SOFTWARE DESIGN INC BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 *
28 *      from BSDI nfs_lock.c,v 2.4 1998/12/14 23:49:56 jch Exp
29 */
30
31#include <sys/cdefs.h>
32__FBSDID("$FreeBSD: head/sys/nfsclient/nfs_lock.c 154316 2006-01-13 23:16:29Z rwatson $");
33
34#include <sys/param.h>
35#include <sys/systm.h>
36#include <sys/conf.h>
37#include <sys/fcntl.h>
38#include <sys/kernel.h>		/* for hz */
39#include <sys/limits.h>
40#include <sys/lock.h>
41#include <sys/malloc.h>
42#include <sys/lockf.h>		/* for hz */ /* Must come after sys/malloc.h */
43#include <sys/mbuf.h>
44#include <sys/mount.h>
45#include <sys/namei.h>
46#include <sys/proc.h>
47#include <sys/resourcevar.h>
48#include <sys/socket.h>
49#include <sys/socket.h>
50#include <sys/unistd.h>
51#include <sys/vnode.h>
52
53#include <net/if.h>
54
55#include <rpc/rpcclnt.h>
56
57#include <nfs/rpcv2.h>
58#include <nfs/nfsproto.h>
59#include <nfsclient/nfs.h>
60#include <nfsclient/nfsmount.h>
61#include <nfsclient/nfsnode.h>
62#include <nfsclient/nfs_lock.h>
63#include <nfsclient/nlminfo.h>
64
65extern void (*nlminfo_release_p)(struct proc *p);
66
67MALLOC_DEFINE(M_NFSLOCK, "nfsclient_lock", "NFS lock request");
68MALLOC_DEFINE(M_NLMINFO, "nfsclient_nlminfo", "NFS lock process structure");
69
70static int nfslockdans(struct thread *td, struct lockd_ans *ansp);
71static void nlminfo_release(struct proc *p);
72/*
73 * --------------------------------------------------------------------
74 * A miniature device driver which the userland uses to talk to us.
75 *
76 */
77
78static struct cdev *nfslock_dev;
79static struct mtx nfslock_mtx;
80static int nfslock_isopen;
81static TAILQ_HEAD(,__lock_msg)	nfslock_list;
82
83static int
84nfslock_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
85{
86	int error;
87
88	mtx_lock(&nfslock_mtx);
89	if (!nfslock_isopen) {
90		error = 0;
91		nfslock_isopen = 1;
92	} else {
93		error = EOPNOTSUPP;
94	}
95	mtx_unlock(&nfslock_mtx);
96
97	return (error);
98}
99
100static int
101nfslock_close(struct cdev *dev, int fflag, int devtype, struct thread *td)
102{
103	struct __lock_msg *lm;
104
105	mtx_lock(&nfslock_mtx);
106	nfslock_isopen = 0;
107	while (!TAILQ_EMPTY(&nfslock_list)) {
108		lm = TAILQ_FIRST(&nfslock_list);
109		/* XXX: answer request */
110		TAILQ_REMOVE(&nfslock_list, lm, lm_link);
111		free(lm, M_NFSLOCK);
112	}
113	mtx_unlock(&nfslock_mtx);
114	return (0);
115}
116
117static int
118nfslock_read(struct cdev *dev, struct uio *uio, int ioflag)
119{
120	int error;
121	struct __lock_msg *lm;
122
123	if (uio->uio_resid != sizeof *lm)
124		return (EOPNOTSUPP);
125	lm = NULL;
126	error = 0;
127	mtx_lock(&nfslock_mtx);
128	while (TAILQ_EMPTY(&nfslock_list)) {
129		error = msleep(&nfslock_list, &nfslock_mtx, PSOCK | PCATCH,
130		    "nfslockd", 0);
131		if (error)
132			break;
133	}
134	if (!error) {
135		lm = TAILQ_FIRST(&nfslock_list);
136		TAILQ_REMOVE(&nfslock_list, lm, lm_link);
137	}
138	mtx_unlock(&nfslock_mtx);
139	if (!error) {
140		error = uiomove(lm, sizeof *lm, uio);
141		free(lm, M_NFSLOCK);
142	}
143	return (error);
144}
145
146static int
147nfslock_write(struct cdev *dev, struct uio *uio, int ioflag)
148{
149	struct lockd_ans la;
150	int error;
151
152	if (uio->uio_resid != sizeof la)
153		return (EOPNOTSUPP);
154	error = uiomove(&la, sizeof la, uio);
155	if (!error)
156		error = nfslockdans(curthread, &la);
157	return (error);
158}
159
160static int
161nfslock_send(struct __lock_msg *lm)
162{
163	struct __lock_msg *lm2;
164	int error;
165
166	error = 0;
167	lm2 = malloc(sizeof *lm2, M_NFSLOCK, M_WAITOK);
168	mtx_lock(&nfslock_mtx);
169	if (nfslock_isopen) {
170		memcpy(lm2, lm, sizeof *lm2);
171		TAILQ_INSERT_TAIL(&nfslock_list, lm2, lm_link);
172		wakeup(&nfslock_list);
173	} else {
174		error = EOPNOTSUPP;
175	}
176	mtx_unlock(&nfslock_mtx);
177	if (error)
178		free(lm2, M_NFSLOCK);
179	return (error);
180}
181
182static struct cdevsw nfslock_cdevsw = {
183	.d_version =	D_VERSION,
184	.d_open =	nfslock_open,
185	.d_close =	nfslock_close,
186	.d_read =	nfslock_read,
187	.d_write =	nfslock_write,
188	.d_name =	"nfslock"
189};
190
191static int
192nfslock_modevent(module_t mod __unused, int type, void *data __unused)
193{
194
195	switch (type) {
196	case MOD_LOAD:
197		if (bootverbose)
198			printf("nfslock: pseudo-device\n");
199		mtx_init(&nfslock_mtx, "nfslock", NULL, MTX_DEF);
200		TAILQ_INIT(&nfslock_list);
201		nlminfo_release_p = nlminfo_release;
202		nfslock_dev = make_dev(&nfslock_cdevsw, 0,
203		    UID_ROOT, GID_KMEM, 0600, _PATH_NFSLCKDEV);
204		return (0);
205	default:
206		return (EOPNOTSUPP);
207	}
208}
209
210DEV_MODULE(nfslock, nfslock_modevent, NULL);
211MODULE_VERSION(nfslock, 1);
212
213
214/*
215 * XXX
216 * We have to let the process know if the call succeeded.  I'm using an extra
217 * field in the p_nlminfo field in the proc structure, as it is already for
218 * lockd stuff.
219 */
220
221/*
222 * nfs_advlock --
223 *      NFS advisory byte-level locks.
224 */
225int
226nfs_dolock(struct vop_advlock_args *ap)
227{
228	LOCKD_MSG msg;
229	struct thread *td;
230	struct vnode *vp;
231	int error;
232	struct flock *fl;
233	struct proc *p;
234
235	td = curthread;
236	p = td->td_proc;
237
238	vp = ap->a_vp;
239	fl = ap->a_fl;
240
241	/*
242	 * the NLM protocol doesn't allow the server to return an error
243	 * on ranges, so we do it.
244	 */
245	if (fl->l_whence != SEEK_END) {
246		if ((fl->l_whence != SEEK_CUR && fl->l_whence != SEEK_SET) ||
247		    fl->l_start < 0 ||
248		    (fl->l_len < 0 &&
249		     (fl->l_start == 0 || fl->l_start + fl->l_len < 0)))
250			return (EINVAL);
251		if (fl->l_len > 0 &&
252			 (fl->l_len - 1 > OFF_MAX - fl->l_start))
253			return (EOVERFLOW);
254	}
255
256	/*
257	 * Fill in the information structure.
258	 */
259	msg.lm_version = LOCKD_MSG_VERSION;
260	msg.lm_msg_ident.pid = p->p_pid;
261	/*
262	 * if there is no nfsowner table yet, allocate one.
263	 */
264	if (p->p_nlminfo == NULL) {
265		MALLOC(p->p_nlminfo, struct nlminfo *,
266			sizeof(struct nlminfo), M_NLMINFO, M_WAITOK | M_ZERO);
267		p->p_nlminfo->pid_start = p->p_stats->p_start;
268		timevaladd(&p->p_nlminfo->pid_start, &boottime);
269	}
270	msg.lm_msg_ident.pid_start = p->p_nlminfo->pid_start;
271	msg.lm_msg_ident.msg_seq = ++(p->p_nlminfo->msg_seq);
272
273	msg.lm_fl = *fl;
274	msg.lm_wait = ap->a_flags & F_WAIT;
275	msg.lm_getlk = ap->a_op == F_GETLK;
276	bcopy(VFSTONFS(vp->v_mount)->nm_nam, &msg.lm_addr,
277		min(sizeof msg.lm_addr, VFSTONFS(vp->v_mount)->nm_nam->sa_len));
278	msg.lm_fh_len = NFS_ISV3(vp) ? VTONFS(vp)->n_fhsize : NFSX_V2FH;
279	bcopy(VTONFS(vp)->n_fhp, msg.lm_fh, msg.lm_fh_len);
280	msg.lm_nfsv3 = NFS_ISV3(vp);
281	cru2x(td->td_ucred, &msg.lm_cred);
282
283	for (;;) {
284		error = nfslock_send(&msg);
285		if (error)
286			return (error);
287
288		/* Unlocks succeed immediately.  */
289		if (fl->l_type == F_UNLCK)
290			return (error);
291
292		/*
293		 * retry after 20 seconds if we haven't gotten a responce yet.
294		 * This number was picked out of thin air... but is longer
295		 * then even a reasonably loaded system should take (at least
296		 * on a local network).  XXX Probably should use a back-off
297		 * scheme.
298		 *
299		 * XXX: No PCATCH here since we currently have no useful
300		 * way to signal to the userland rpc.lockd that the request
301		 * has been aborted.  Once the rpc.lockd implementation
302		 * can handle aborts, and we report them properly,
303		 * PCATCH can be put back.  In the mean time, if we did
304		 * permit aborting, the lock attempt would "get lost"
305		 * and the lock would get stuck in the locked state.
306		 */
307		error = tsleep(p->p_nlminfo, PUSER, "lockd", 20*hz);
308		if (error != 0) {
309			if (error == EWOULDBLOCK) {
310				/*
311				 * We timed out, so we rewrite the request
312				 * to the fifo.
313				 */
314				continue;
315			}
316
317			break;
318		}
319
320		if (msg.lm_getlk && p->p_nlminfo->retcode == 0) {
321			if (p->p_nlminfo->set_getlk_pid) {
322				fl->l_pid = p->p_nlminfo->getlk_pid;
323			} else {
324				fl->l_type = F_UNLCK;
325			}
326		}
327		error = p->p_nlminfo->retcode;
328		break;
329	}
330
331	return (error);
332}
333
334/*
335 * nfslockdans --
336 *      NFS advisory byte-level locks answer from the lock daemon.
337 */
338static int
339nfslockdans(struct thread *td, struct lockd_ans *ansp)
340{
341	struct proc *targetp;
342	int error;
343
344	/* Let root, or someone who once was root (lockd generally
345	 * switches to the daemon uid once it is done setting up) make
346	 * this call.
347	 *
348	 * XXX This authorization check is probably not right.
349	 */
350	if ((error = suser(td)) != 0 &&
351	    td->td_ucred->cr_svuid != 0)
352		return (error);
353
354	/* the version should match, or we're out of sync */
355	if (ansp->la_vers != LOCKD_ANS_VERSION)
356		return (EINVAL);
357
358	/* Find the process, set its return errno and wake it up. */
359	if ((targetp = pfind(ansp->la_msg_ident.pid)) == NULL)
360		return (ESRCH);
361
362	/* verify the pid hasn't been reused (if we can), and it isn't waiting
363	 * for an answer from a more recent request.  We return an EPIPE if
364	 * the match fails, because we've already used ESRCH above, and this
365	 * is sort of like writing on a pipe after the reader has closed it.
366	 */
367	if (targetp->p_nlminfo == NULL ||
368	    ((ansp->la_msg_ident.msg_seq != -1) &&
369	      (timevalcmp(&targetp->p_nlminfo->pid_start,
370			&ansp->la_msg_ident.pid_start, !=) ||
371	       targetp->p_nlminfo->msg_seq != ansp->la_msg_ident.msg_seq))) {
372		PROC_UNLOCK(targetp);
373		return (EPIPE);
374	}
375
376	targetp->p_nlminfo->retcode = ansp->la_errno;
377	targetp->p_nlminfo->set_getlk_pid = ansp->la_set_getlk_pid;
378	targetp->p_nlminfo->getlk_pid = ansp->la_getlk_pid;
379
380	wakeup(targetp->p_nlminfo);
381
382	PROC_UNLOCK(targetp);
383	return (0);
384}
385
386/*
387 * Free nlminfo attached to process.
388 */
389void
390nlminfo_release(struct proc *p)
391{
392	free(p->p_nlminfo, M_NLMINFO);
393	p->p_nlminfo = NULL;
394}
395