1/*	$NetBSD: nfs_lock.c,v 1.3 2020/01/02 15:42:26 thorpej Exp $	*/
2/*-
3 * Copyright (c) 1997 Berkeley Software Design, Inc. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 * 3. Berkeley Software Design Inc's name may not be used to endorse or
14 *    promote products derived from this software without specific prior
15 *    written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY BERKELEY SOFTWARE DESIGN INC ``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 BERKELEY SOFTWARE DESIGN INC 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 *      from BSDI nfs_lock.c,v 2.4 1998/12/14 23:49:56 jch Exp
30 */
31
32#include <sys/cdefs.h>
33/* __FBSDID("FreeBSD: head/sys/nfs/nfs_lock.c 303382 2016-07-27 11:08:59Z kib "); */
34__RCSID("$NetBSD: nfs_lock.c,v 1.3 2020/01/02 15:42:26 thorpej Exp $");
35
36#include <sys/param.h>
37#include <sys/systm.h>
38#include <sys/conf.h>
39#include <sys/fcntl.h>
40#include <sys/kernel.h>		/* for hz */
41#include <sys/limits.h>
42#include <sys/lock.h>
43#include <sys/malloc.h>
44#include <sys/lockf.h>		/* for hz */ /* Must come after sys/malloc.h */
45#include <sys/mbuf.h>
46#include <sys/mount.h>
47#include <sys/namei.h>
48#include <sys/priv.h>
49#include <sys/proc.h>
50#include <sys/resourcevar.h>
51#include <sys/socket.h>
52#include <sys/socket.h>
53#include <sys/unistd.h>
54#include <sys/vnode.h>
55
56#include <net/if.h>
57
58#include <fs/nfs/common/nfsproto.h>
59#include <fs/nfs/common/nfs_lock.h>
60#include <fs/nfs/client/nfs.h>
61#include <fs/nfs/client/nfsmount.h>
62#include <fs/nfs/client/nfsnode.h>
63#include <fs/nfs/client/nlminfo.h>
64
65extern void (*nlminfo_release_p)(struct proc *p);
66
67vop_advlock_t	*nfs_advlock_p = nfs_dolock;
68vop_reclaim_t	*nfs_reclaim_p = NULL;
69
70static MALLOC_DEFINE(M_NFSLOCK, "nfsclient_lock", "NFS lock request");
71static MALLOC_DEFINE(M_NLMINFO, "nfsclient_nlminfo",
72    "NFS lock process structure");
73
74static int nfslockdans(struct thread *td, struct lockd_ans *ansp);
75static void nlminfo_release(struct proc *p);
76/*
77 * --------------------------------------------------------------------
78 * A miniature device driver which the userland uses to talk to us.
79 *
80 */
81
82static struct cdev *nfslock_dev;
83static struct mtx nfslock_mtx;
84static int nfslock_isopen;
85static TAILQ_HEAD(,__lock_msg)	nfslock_list;
86
87static int
88nfslock_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
89{
90	int error;
91
92	error = priv_check(td, PRIV_NFS_LOCKD);
93	if (error)
94		return (error);
95
96	mtx_lock(&nfslock_mtx);
97	if (!nfslock_isopen) {
98		error = 0;
99		nfslock_isopen = 1;
100	} else {
101		error = EOPNOTSUPP;
102	}
103	mtx_unlock(&nfslock_mtx);
104
105	return (error);
106}
107
108static int
109nfslock_close(struct cdev *dev, int fflag, int devtype, struct thread *td)
110{
111	struct __lock_msg *lm;
112
113	mtx_lock(&nfslock_mtx);
114	nfslock_isopen = 0;
115	while (!TAILQ_EMPTY(&nfslock_list)) {
116		lm = TAILQ_FIRST(&nfslock_list);
117		/* XXX: answer request */
118		TAILQ_REMOVE(&nfslock_list, lm, lm_link);
119		free(lm, M_NFSLOCK);
120	}
121	mtx_unlock(&nfslock_mtx);
122	return (0);
123}
124
125static int
126nfslock_read(struct cdev *dev, struct uio *uio, int ioflag)
127{
128	int error;
129	struct __lock_msg *lm;
130
131	if (uio->uio_resid != sizeof *lm)
132		return (EOPNOTSUPP);
133	lm = NULL;
134	error = 0;
135	mtx_lock(&nfslock_mtx);
136	while (TAILQ_EMPTY(&nfslock_list)) {
137		error = msleep(&nfslock_list, &nfslock_mtx, PSOCK | PCATCH,
138		    "nfslockd", 0);
139		if (error)
140			break;
141	}
142	if (!error) {
143		lm = TAILQ_FIRST(&nfslock_list);
144		TAILQ_REMOVE(&nfslock_list, lm, lm_link);
145	}
146	mtx_unlock(&nfslock_mtx);
147	if (!error) {
148		error = uiomove(lm, sizeof *lm, uio);
149		free(lm, M_NFSLOCK);
150	}
151	return (error);
152}
153
154static int
155nfslock_write(struct cdev *dev, struct uio *uio, int ioflag)
156{
157	struct lockd_ans la;
158	int error;
159
160	if (uio->uio_resid != sizeof la)
161		return (EOPNOTSUPP);
162	error = uiomove(&la, sizeof la, uio);
163	if (!error)
164		error = nfslockdans(curthread, &la);
165	return (error);
166}
167
168static int
169nfslock_send(struct __lock_msg *lm)
170{
171	struct __lock_msg *lm2;
172	int error;
173
174	error = 0;
175	lm2 = malloc(sizeof *lm2, M_NFSLOCK, M_WAITOK);
176	mtx_lock(&nfslock_mtx);
177	if (nfslock_isopen) {
178		memcpy(lm2, lm, sizeof *lm2);
179		TAILQ_INSERT_TAIL(&nfslock_list, lm2, lm_link);
180		wakeup(&nfslock_list);
181	} else {
182		error = EOPNOTSUPP;
183	}
184	mtx_unlock(&nfslock_mtx);
185	if (error)
186		free(lm2, M_NFSLOCK);
187	return (error);
188}
189
190static struct cdevsw nfslock_cdevsw = {
191	.d_version =	D_VERSION,
192	.d_open =	nfslock_open,
193	.d_close =	nfslock_close,
194	.d_read =	nfslock_read,
195	.d_write =	nfslock_write,
196	.d_name =	"nfslock"
197};
198
199static int
200nfslock_modevent(module_t mod __unused, int type, void *data __unused)
201{
202
203	switch (type) {
204	case MOD_LOAD:
205		if (bootverbose)
206			printf("nfslock: pseudo-device\n");
207		mtx_init(&nfslock_mtx, "nfslock", NULL, MTX_DEF);
208		TAILQ_INIT(&nfslock_list);
209		nlminfo_release_p = nlminfo_release;
210		nfslock_dev = make_dev(&nfslock_cdevsw, 0,
211		    UID_ROOT, GID_KMEM, 0600, _PATH_NFSLCKDEV);
212		return (0);
213	default:
214		return (EOPNOTSUPP);
215	}
216}
217
218DEV_MODULE(nfslock, nfslock_modevent, NULL);
219MODULE_VERSION(nfslock, 1);
220
221
222/*
223 * XXX
224 * We have to let the process know if the call succeeded.  I'm using an extra
225 * field in the p_nlminfo field in the proc structure, as it is already for
226 * lockd stuff.
227 */
228
229/*
230 * nfs_advlock --
231 *      NFS advisory byte-level locks.
232 *
233 * The vnode shall be (shared) locked on the entry, it is
234 * unconditionally unlocked after.
235 */
236int
237nfs_dolock(struct vop_advlock_args *ap)
238{
239	LOCKD_MSG msg;
240	struct thread *td;
241	struct vnode *vp;
242	int error;
243	struct flock *fl;
244	struct proc *p;
245	struct nfsmount *nmp;
246	struct timeval btv;
247
248	td = curthread;
249	p = td->td_proc;
250
251	vp = ap->a_vp;
252	fl = ap->a_fl;
253	nmp = VFSTONFS(vp->v_mount);
254
255	ASSERT_VOP_LOCKED(vp, "nfs_dolock");
256
257	nmp->nm_getinfo(vp, msg.lm_fh, &msg.lm_fh_len, &msg.lm_addr,
258	    &msg.lm_nfsv3, NULL, NULL);
259	VOP_UNLOCK(vp, 0);
260
261	/*
262	 * the NLM protocol doesn't allow the server to return an error
263	 * on ranges, so we do it.
264	 */
265	if (fl->l_whence != SEEK_END) {
266		if ((fl->l_whence != SEEK_CUR && fl->l_whence != SEEK_SET) ||
267		    fl->l_start < 0 ||
268		    (fl->l_len < 0 &&
269		     (fl->l_start == 0 || fl->l_start + fl->l_len < 0)))
270			return (EINVAL);
271		if (fl->l_len > 0 &&
272			 (fl->l_len - 1 > OFF_MAX - fl->l_start))
273			return (EOVERFLOW);
274	}
275
276	/*
277	 * Fill in the information structure.
278	 */
279	msg.lm_version = LOCKD_MSG_VERSION;
280	msg.lm_msg_ident.pid = p->p_pid;
281
282	mtx_lock(&Giant);
283	/*
284	 * if there is no nfsowner table yet, allocate one.
285	 */
286	if (p->p_nlminfo == NULL) {
287		p->p_nlminfo = malloc(sizeof(struct nlminfo),
288		    M_NLMINFO, M_WAITOK | M_ZERO);
289		p->p_nlminfo->pid_start = p->p_stats->p_start;
290		getmicroboottime(&btv);
291		timevaladd(&p->p_nlminfo->pid_start, &btv);
292	}
293	msg.lm_msg_ident.pid_start = p->p_nlminfo->pid_start;
294	msg.lm_msg_ident.msg_seq = ++(p->p_nlminfo->msg_seq);
295
296	msg.lm_fl = *fl;
297	msg.lm_wait = ap->a_flags & F_WAIT;
298	msg.lm_getlk = ap->a_op == F_GETLK;
299	cru2x(td->td_ucred, &msg.lm_cred);
300
301	for (;;) {
302		error = nfslock_send(&msg);
303		if (error)
304			goto out;
305
306		/* Unlocks succeed immediately.  */
307		if (fl->l_type == F_UNLCK)
308			goto out;
309
310		/*
311		 * Retry after 20 seconds if we haven't gotten a response yet.
312		 * This number was picked out of thin air... but is longer
313		 * then even a reasonably loaded system should take (at least
314		 * on a local network).  XXX Probably should use a back-off
315		 * scheme.
316		 *
317		 * XXX: No PCATCH here since we currently have no useful
318		 * way to signal to the userland rpc.lockd that the request
319		 * has been aborted.  Once the rpc.lockd implementation
320		 * can handle aborts, and we report them properly,
321		 * PCATCH can be put back.  In the mean time, if we did
322		 * permit aborting, the lock attempt would "get lost"
323		 * and the lock would get stuck in the locked state.
324		 */
325		error = tsleep(p->p_nlminfo, PUSER, "lockd", 20*hz);
326		if (error != 0) {
327			if (error == EWOULDBLOCK) {
328				/*
329				 * We timed out, so we rewrite the request
330				 * to the fifo.
331				 */
332				continue;
333			}
334
335			break;
336		}
337
338		if (msg.lm_getlk && p->p_nlminfo->retcode == 0) {
339			if (p->p_nlminfo->set_getlk_pid) {
340				fl->l_sysid = 0; /* XXX */
341				fl->l_pid = p->p_nlminfo->getlk_pid;
342			} else {
343				fl->l_type = F_UNLCK;
344			}
345		}
346		error = p->p_nlminfo->retcode;
347		break;
348	}
349 out:
350	mtx_unlock(&Giant);
351	return (error);
352}
353
354/*
355 * nfslockdans --
356 *      NFS advisory byte-level locks answer from the lock daemon.
357 */
358static int
359nfslockdans(struct thread *td, struct lockd_ans *ansp)
360{
361	struct proc *targetp;
362
363	/* the version should match, or we're out of sync */
364	if (ansp->la_vers != LOCKD_ANS_VERSION)
365		return (EINVAL);
366
367	/* Find the process, set its return errno and wake it up. */
368	if ((targetp = pfind(ansp->la_msg_ident.pid)) == NULL)
369		return (ESRCH);
370
371	/* verify the pid hasn't been reused (if we can), and it isn't waiting
372	 * for an answer from a more recent request.  We return an EPIPE if
373	 * the match fails, because we've already used ESRCH above, and this
374	 * is sort of like writing on a pipe after the reader has closed it.
375	 */
376	if (targetp->p_nlminfo == NULL ||
377	    ((ansp->la_msg_ident.msg_seq != -1) &&
378	      (timevalcmp(&targetp->p_nlminfo->pid_start,
379			&ansp->la_msg_ident.pid_start, !=) ||
380	       targetp->p_nlminfo->msg_seq != ansp->la_msg_ident.msg_seq))) {
381		PROC_UNLOCK(targetp);
382		return (EPIPE);
383	}
384
385	targetp->p_nlminfo->retcode = ansp->la_errno;
386	targetp->p_nlminfo->set_getlk_pid = ansp->la_set_getlk_pid;
387	targetp->p_nlminfo->getlk_pid = ansp->la_getlk_pid;
388
389	wakeup(targetp->p_nlminfo);
390
391	PROC_UNLOCK(targetp);
392	return (0);
393}
394
395/*
396 * Free nlminfo attached to process.
397 */
398void
399nlminfo_release(struct proc *p)
400{
401	free(p->p_nlminfo, M_NLMINFO);
402	p->p_nlminfo = NULL;
403}
404