nfs_lock.c revision 214048 
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/nfs/nfs_lock.c 214048 2010-10-19 00:20:00Z rmacklem $");
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/priv.h>
47#include <sys/proc.h>
48#include <sys/resourcevar.h>
49#include <sys/socket.h>
50#include <sys/socket.h>
51#include <sys/unistd.h>
52#include <sys/vnode.h>
53
54#include <net/if.h>
55
56#include <nfs/nfsproto.h>
57#include <nfs/nfs_lock.h>
58#include <nfsclient/nfs.h>
59#include <nfsclient/nfsmount.h>
60#include <nfsclient/nfsnode.h>
61#include <nfsclient/nlminfo.h>
62
63extern void (*nlminfo_release_p)(struct proc *p);
64
65vop_advlock_t	*nfs_advlock_p = nfs_dolock;
66vop_reclaim_t	*nfs_reclaim_p = NULL;
67
68MALLOC_DEFINE(M_NFSLOCK, "nfsclient_lock", "NFS lock request");
69MALLOC_DEFINE(M_NLMINFO, "nfsclient_nlminfo", "NFS lock process structure");
70
71static int nfslockdans(struct thread *td, struct lockd_ans *ansp);
72static void nlminfo_release(struct proc *p);
73/*
74 * --------------------------------------------------------------------
75 * A miniature device driver which the userland uses to talk to us.
76 *
77 */
78
79static struct cdev *nfslock_dev;
80static struct mtx nfslock_mtx;
81static int nfslock_isopen;
82static TAILQ_HEAD(,__lock_msg)	nfslock_list;
83
84static int
85nfslock_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
86{
87	int error;
88
89	error = priv_check(td, PRIV_NFS_LOCKD);
90	if (error)
91		return (error);
92
93	mtx_lock(&nfslock_mtx);
94	if (!nfslock_isopen) {
95		error = 0;
96		nfslock_isopen = 1;
97	} else {
98		error = EOPNOTSUPP;
99	}
100	mtx_unlock(&nfslock_mtx);
101
102	return (error);
103}
104
105static int
106nfslock_close(struct cdev *dev, int fflag, int devtype, struct thread *td)
107{
108	struct __lock_msg *lm;
109
110	mtx_lock(&nfslock_mtx);
111	nfslock_isopen = 0;
112	while (!TAILQ_EMPTY(&nfslock_list)) {
113		lm = TAILQ_FIRST(&nfslock_list);
114		/* XXX: answer request */
115		TAILQ_REMOVE(&nfslock_list, lm, lm_link);
116		free(lm, M_NFSLOCK);
117	}
118	mtx_unlock(&nfslock_mtx);
119	return (0);
120}
121
122static int
123nfslock_read(struct cdev *dev, struct uio *uio, int ioflag)
124{
125	int error;
126	struct __lock_msg *lm;
127
128	if (uio->uio_resid != sizeof *lm)
129		return (EOPNOTSUPP);
130	lm = NULL;
131	error = 0;
132	mtx_lock(&nfslock_mtx);
133	while (TAILQ_EMPTY(&nfslock_list)) {
134		error = msleep(&nfslock_list, &nfslock_mtx, PSOCK | PCATCH,
135		    "nfslockd", 0);
136		if (error)
137			break;
138	}
139	if (!error) {
140		lm = TAILQ_FIRST(&nfslock_list);
141		TAILQ_REMOVE(&nfslock_list, lm, lm_link);
142	}
143	mtx_unlock(&nfslock_mtx);
144	if (!error) {
145		error = uiomove(lm, sizeof *lm, uio);
146		free(lm, M_NFSLOCK);
147	}
148	return (error);
149}
150
151static int
152nfslock_write(struct cdev *dev, struct uio *uio, int ioflag)
153{
154	struct lockd_ans la;
155	int error;
156
157	if (uio->uio_resid != sizeof la)
158		return (EOPNOTSUPP);
159	error = uiomove(&la, sizeof la, uio);
160	if (!error)
161		error = nfslockdans(curthread, &la);
162	return (error);
163}
164
165static int
166nfslock_send(struct __lock_msg *lm)
167{
168	struct __lock_msg *lm2;
169	int error;
170
171	error = 0;
172	lm2 = malloc(sizeof *lm2, M_NFSLOCK, M_WAITOK);
173	mtx_lock(&nfslock_mtx);
174	if (nfslock_isopen) {
175		memcpy(lm2, lm, sizeof *lm2);
176		TAILQ_INSERT_TAIL(&nfslock_list, lm2, lm_link);
177		wakeup(&nfslock_list);
178	} else {
179		error = EOPNOTSUPP;
180	}
181	mtx_unlock(&nfslock_mtx);
182	if (error)
183		free(lm2, M_NFSLOCK);
184	return (error);
185}
186
187static struct cdevsw nfslock_cdevsw = {
188	.d_version =	D_VERSION,
189	.d_open =	nfslock_open,
190	.d_close =	nfslock_close,
191	.d_read =	nfslock_read,
192	.d_write =	nfslock_write,
193	.d_name =	"nfslock"
194};
195
196static int
197nfslock_modevent(module_t mod __unused, int type, void *data __unused)
198{
199
200	switch (type) {
201	case MOD_LOAD:
202		if (bootverbose)
203			printf("nfslock: pseudo-device\n");
204		mtx_init(&nfslock_mtx, "nfslock", NULL, MTX_DEF);
205		TAILQ_INIT(&nfslock_list);
206		nlminfo_release_p = nlminfo_release;
207		nfslock_dev = make_dev(&nfslock_cdevsw, 0,
208		    UID_ROOT, GID_KMEM, 0600, _PATH_NFSLCKDEV);
209		return (0);
210	default:
211		return (EOPNOTSUPP);
212	}
213}
214
215DEV_MODULE(nfslock, nfslock_modevent, NULL);
216MODULE_VERSION(nfslock, 1);
217
218
219/*
220 * XXX
221 * We have to let the process know if the call succeeded.  I'm using an extra
222 * field in the p_nlminfo field in the proc structure, as it is already for
223 * lockd stuff.
224 */
225
226/*
227 * nfs_advlock --
228 *      NFS advisory byte-level locks.
229 *
230 * The vnode shall be (shared) locked on the entry, it is
231 * unconditionally unlocked after.
232 */
233int
234nfs_dolock(struct vop_advlock_args *ap)
235{
236	LOCKD_MSG msg;
237	struct thread *td;
238	struct vnode *vp;
239	int error;
240	struct flock *fl;
241	struct proc *p;
242	struct nfsmount *nmp;
243
244	td = curthread;
245	p = td->td_proc;
246
247	vp = ap->a_vp;
248	fl = ap->a_fl;
249	nmp = VFSTONFS(vp->v_mount);
250
251	ASSERT_VOP_LOCKED(vp, "nfs_dolock");
252
253	nmp->nm_getinfo(vp, msg.lm_fh, &msg.lm_fh_len, &msg.lm_addr,
254	    &msg.lm_nfsv3, NULL);
255	VOP_UNLOCK(vp, 0);
256
257	/*
258	 * the NLM protocol doesn't allow the server to return an error
259	 * on ranges, so we do it.
260	 */
261	if (fl->l_whence != SEEK_END) {
262		if ((fl->l_whence != SEEK_CUR && fl->l_whence != SEEK_SET) ||
263		    fl->l_start < 0 ||
264		    (fl->l_len < 0 &&
265		     (fl->l_start == 0 || fl->l_start + fl->l_len < 0)))
266			return (EINVAL);
267		if (fl->l_len > 0 &&
268			 (fl->l_len - 1 > OFF_MAX - fl->l_start))
269			return (EOVERFLOW);
270	}
271
272	/*
273	 * Fill in the information structure.
274	 */
275	msg.lm_version = LOCKD_MSG_VERSION;
276	msg.lm_msg_ident.pid = p->p_pid;
277
278	mtx_lock(&Giant);
279	/*
280	 * if there is no nfsowner table yet, allocate one.
281	 */
282	if (p->p_nlminfo == NULL) {
283		p->p_nlminfo = malloc(sizeof(struct nlminfo),
284		    M_NLMINFO, M_WAITOK | M_ZERO);
285		p->p_nlminfo->pid_start = p->p_stats->p_start;
286		timevaladd(&p->p_nlminfo->pid_start, &boottime);
287	}
288	msg.lm_msg_ident.pid_start = p->p_nlminfo->pid_start;
289	msg.lm_msg_ident.msg_seq = ++(p->p_nlminfo->msg_seq);
290
291	msg.lm_fl = *fl;
292	msg.lm_wait = ap->a_flags & F_WAIT;
293	msg.lm_getlk = ap->a_op == F_GETLK;
294	cru2x(td->td_ucred, &msg.lm_cred);
295
296	for (;;) {
297		error = nfslock_send(&msg);
298		if (error)
299			goto out;
300
301		/* Unlocks succeed immediately.  */
302		if (fl->l_type == F_UNLCK)
303			goto out;
304
305		/*
306		 * Retry after 20 seconds if we haven't gotten a response yet.
307		 * This number was picked out of thin air... but is longer
308		 * then even a reasonably loaded system should take (at least
309		 * on a local network).  XXX Probably should use a back-off
310		 * scheme.
311		 *
312		 * XXX: No PCATCH here since we currently have no useful
313		 * way to signal to the userland rpc.lockd that the request
314		 * has been aborted.  Once the rpc.lockd implementation
315		 * can handle aborts, and we report them properly,
316		 * PCATCH can be put back.  In the mean time, if we did
317		 * permit aborting, the lock attempt would "get lost"
318		 * and the lock would get stuck in the locked state.
319		 */
320		error = tsleep(p->p_nlminfo, PUSER, "lockd", 20*hz);
321		if (error != 0) {
322			if (error == EWOULDBLOCK) {
323				/*
324				 * We timed out, so we rewrite the request
325				 * to the fifo.
326				 */
327				continue;
328			}
329
330			break;
331		}
332
333		if (msg.lm_getlk && p->p_nlminfo->retcode == 0) {
334			if (p->p_nlminfo->set_getlk_pid) {
335				fl->l_sysid = 0; /* XXX */
336				fl->l_pid = p->p_nlminfo->getlk_pid;
337			} else {
338				fl->l_type = F_UNLCK;
339			}
340		}
341		error = p->p_nlminfo->retcode;
342		break;
343	}
344 out:
345	mtx_unlock(&Giant);
346	return (error);
347}
348
349/*
350 * nfslockdans --
351 *      NFS advisory byte-level locks answer from the lock daemon.
352 */
353static int
354nfslockdans(struct thread *td, struct lockd_ans *ansp)
355{
356	struct proc *targetp;
357
358	/* the version should match, or we're out of sync */
359	if (ansp->la_vers != LOCKD_ANS_VERSION)
360		return (EINVAL);
361
362	/* Find the process, set its return errno and wake it up. */
363	if ((targetp = pfind(ansp->la_msg_ident.pid)) == NULL)
364		return (ESRCH);
365
366	/* verify the pid hasn't been reused (if we can), and it isn't waiting
367	 * for an answer from a more recent request.  We return an EPIPE if
368	 * the match fails, because we've already used ESRCH above, and this
369	 * is sort of like writing on a pipe after the reader has closed it.
370	 */
371	if (targetp->p_nlminfo == NULL ||
372	    ((ansp->la_msg_ident.msg_seq != -1) &&
373	      (timevalcmp(&targetp->p_nlminfo->pid_start,
374			&ansp->la_msg_ident.pid_start, !=) ||
375	       targetp->p_nlminfo->msg_seq != ansp->la_msg_ident.msg_seq))) {
376		PROC_UNLOCK(targetp);
377		return (EPIPE);
378	}
379
380	targetp->p_nlminfo->retcode = ansp->la_errno;
381	targetp->p_nlminfo->set_getlk_pid = ansp->la_set_getlk_pid;
382	targetp->p_nlminfo->getlk_pid = ansp->la_getlk_pid;
383
384	wakeup(targetp->p_nlminfo);
385
386	PROC_UNLOCK(targetp);
387	return (0);
388}
389
390/*
391 * Free nlminfo attached to process.
392 */
393void
394nlminfo_release(struct proc *p)
395{
396	free(p->p_nlminfo, M_NLMINFO);
397	p->p_nlminfo = NULL;
398}
399