nfs_clvfsops.c revision 317404
1/*-
2 * Copyright (c) 1989, 1993, 1995
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
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 * 4. Neither the name of the University nor the names of its contributors
17 *    may be used to endorse or promote products derived from this software
18 *    without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 *	from nfs_vfsops.c	8.12 (Berkeley) 5/20/95
33 */
34
35#include <sys/cdefs.h>
36__FBSDID("$FreeBSD: stable/10/sys/fs/nfsclient/nfs_clvfsops.c 317404 2017-04-25 11:36:25Z rmacklem $");
37
38
39#include "opt_bootp.h"
40#include "opt_nfsroot.h"
41
42#include <sys/param.h>
43#include <sys/systm.h>
44#include <sys/kernel.h>
45#include <sys/bio.h>
46#include <sys/buf.h>
47#include <sys/clock.h>
48#include <sys/jail.h>
49#include <sys/limits.h>
50#include <sys/lock.h>
51#include <sys/malloc.h>
52#include <sys/mbuf.h>
53#include <sys/module.h>
54#include <sys/mount.h>
55#include <sys/proc.h>
56#include <sys/socket.h>
57#include <sys/socketvar.h>
58#include <sys/sockio.h>
59#include <sys/sysctl.h>
60#include <sys/vnode.h>
61#include <sys/signalvar.h>
62
63#include <vm/vm.h>
64#include <vm/vm_extern.h>
65#include <vm/uma.h>
66
67#include <net/if.h>
68#include <net/route.h>
69#include <netinet/in.h>
70
71#include <fs/nfs/nfsport.h>
72#include <fs/nfsclient/nfsnode.h>
73#include <fs/nfsclient/nfsmount.h>
74#include <fs/nfsclient/nfs.h>
75#include <nfs/nfsdiskless.h>
76
77FEATURE(nfscl, "NFSv4 client");
78
79extern int nfscl_ticks;
80extern struct timeval nfsboottime;
81extern struct nfsstats	newnfsstats;
82extern int nfsrv_useacl;
83extern int nfscl_debuglevel;
84extern enum nfsiod_state ncl_iodwant[NFS_MAXASYNCDAEMON];
85extern struct nfsmount *ncl_iodmount[NFS_MAXASYNCDAEMON];
86extern struct mtx ncl_iod_mutex;
87NFSCLSTATEMUTEX;
88
89MALLOC_DEFINE(M_NEWNFSREQ, "newnfsclient_req", "New NFS request header");
90MALLOC_DEFINE(M_NEWNFSMNT, "newnfsmnt", "New NFS mount struct");
91
92SYSCTL_DECL(_vfs_nfs);
93static int nfs_ip_paranoia = 1;
94SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW,
95    &nfs_ip_paranoia, 0, "");
96static int nfs_tprintf_initial_delay = NFS_TPRINTF_INITIAL_DELAY;
97SYSCTL_INT(_vfs_nfs, NFS_TPRINTF_INITIAL_DELAY,
98        downdelayinitial, CTLFLAG_RW, &nfs_tprintf_initial_delay, 0, "");
99/* how long between console messages "nfs server foo not responding" */
100static int nfs_tprintf_delay = NFS_TPRINTF_DELAY;
101SYSCTL_INT(_vfs_nfs, NFS_TPRINTF_DELAY,
102        downdelayinterval, CTLFLAG_RW, &nfs_tprintf_delay, 0, "");
103
104static int	nfs_mountroot(struct mount *);
105static void	nfs_sec_name(char *, int *);
106static void	nfs_decode_args(struct mount *mp, struct nfsmount *nmp,
107		    struct nfs_args *argp, const char *, struct ucred *,
108		    struct thread *);
109static int	mountnfs(struct nfs_args *, struct mount *,
110		    struct sockaddr *, char *, u_char *, int, u_char *, int,
111		    u_char *, int, struct vnode **, struct ucred *,
112		    struct thread *, int, int, int);
113static void	nfs_getnlminfo(struct vnode *, uint8_t *, size_t *,
114		    struct sockaddr_storage *, int *, off_t *,
115		    struct timeval *);
116static vfs_mount_t nfs_mount;
117static vfs_cmount_t nfs_cmount;
118static vfs_unmount_t nfs_unmount;
119static vfs_root_t nfs_root;
120static vfs_statfs_t nfs_statfs;
121static vfs_sync_t nfs_sync;
122static vfs_sysctl_t nfs_sysctl;
123static vfs_purge_t nfs_purge;
124
125/*
126 * nfs vfs operations.
127 */
128static struct vfsops nfs_vfsops = {
129	.vfs_init =		ncl_init,
130	.vfs_mount =		nfs_mount,
131	.vfs_cmount =		nfs_cmount,
132	.vfs_root =		nfs_root,
133	.vfs_statfs =		nfs_statfs,
134	.vfs_sync =		nfs_sync,
135	.vfs_uninit =		ncl_uninit,
136	.vfs_unmount =		nfs_unmount,
137	.vfs_sysctl =		nfs_sysctl,
138	.vfs_purge =		nfs_purge,
139};
140VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK | VFCF_SBDRY);
141
142/* So that loader and kldload(2) can find us, wherever we are.. */
143MODULE_VERSION(nfs, 1);
144MODULE_DEPEND(nfs, nfscommon, 1, 1, 1);
145MODULE_DEPEND(nfs, krpc, 1, 1, 1);
146MODULE_DEPEND(nfs, nfssvc, 1, 1, 1);
147MODULE_DEPEND(nfs, nfslock, 1, 1, 1);
148
149/*
150 * This structure is now defined in sys/nfs/nfs_diskless.c so that it
151 * can be shared by both NFS clients. It is declared here so that it
152 * will be defined for kernels built without NFS_ROOT, although it
153 * isn't used in that case.
154 */
155#if !defined(NFS_ROOT) && !defined(NFSCLIENT)
156struct nfs_diskless	nfs_diskless = { { { 0 } } };
157struct nfsv3_diskless	nfsv3_diskless = { { { 0 } } };
158int			nfs_diskless_valid = 0;
159#endif
160
161SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD,
162    &nfs_diskless_valid, 0,
163    "Has the diskless struct been filled correctly");
164
165SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD,
166    nfsv3_diskless.root_hostnam, 0, "Path to nfs root");
167
168SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD,
169    &nfsv3_diskless.root_saddr, sizeof(nfsv3_diskless.root_saddr),
170    "%Ssockaddr_in", "Diskless root nfs address");
171
172
173void		newnfsargs_ntoh(struct nfs_args *);
174static int	nfs_mountdiskless(char *,
175		    struct sockaddr_in *, struct nfs_args *,
176		    struct thread *, struct vnode **, struct mount *);
177static void	nfs_convert_diskless(void);
178static void	nfs_convert_oargs(struct nfs_args *args,
179		    struct onfs_args *oargs);
180
181int
182newnfs_iosize(struct nfsmount *nmp)
183{
184	int iosize, maxio;
185
186	/* First, set the upper limit for iosize */
187	if (nmp->nm_flag & NFSMNT_NFSV4) {
188		maxio = NFS_MAXBSIZE;
189	} else if (nmp->nm_flag & NFSMNT_NFSV3) {
190		if (nmp->nm_sotype == SOCK_DGRAM)
191			maxio = NFS_MAXDGRAMDATA;
192		else
193			maxio = NFS_MAXBSIZE;
194	} else {
195		maxio = NFS_V2MAXDATA;
196	}
197	if (nmp->nm_rsize > maxio || nmp->nm_rsize == 0)
198		nmp->nm_rsize = maxio;
199	if (nmp->nm_rsize > NFS_MAXBSIZE)
200		nmp->nm_rsize = NFS_MAXBSIZE;
201	if (nmp->nm_readdirsize > maxio || nmp->nm_readdirsize == 0)
202		nmp->nm_readdirsize = maxio;
203	if (nmp->nm_readdirsize > nmp->nm_rsize)
204		nmp->nm_readdirsize = nmp->nm_rsize;
205	if (nmp->nm_wsize > maxio || nmp->nm_wsize == 0)
206		nmp->nm_wsize = maxio;
207	if (nmp->nm_wsize > NFS_MAXBSIZE)
208		nmp->nm_wsize = NFS_MAXBSIZE;
209
210	/*
211	 * Calculate the size used for io buffers.  Use the larger
212	 * of the two sizes to minimise nfs requests but make sure
213	 * that it is at least one VM page to avoid wasting buffer
214	 * space.  It must also be at least NFS_DIRBLKSIZ, since
215	 * that is the buffer size used for directories.
216	 */
217	iosize = imax(nmp->nm_rsize, nmp->nm_wsize);
218	iosize = imax(iosize, PAGE_SIZE);
219	iosize = imax(iosize, NFS_DIRBLKSIZ);
220	nmp->nm_mountp->mnt_stat.f_iosize = iosize;
221	return (iosize);
222}
223
224static void
225nfs_convert_oargs(struct nfs_args *args, struct onfs_args *oargs)
226{
227
228	args->version = NFS_ARGSVERSION;
229	args->addr = oargs->addr;
230	args->addrlen = oargs->addrlen;
231	args->sotype = oargs->sotype;
232	args->proto = oargs->proto;
233	args->fh = oargs->fh;
234	args->fhsize = oargs->fhsize;
235	args->flags = oargs->flags;
236	args->wsize = oargs->wsize;
237	args->rsize = oargs->rsize;
238	args->readdirsize = oargs->readdirsize;
239	args->timeo = oargs->timeo;
240	args->retrans = oargs->retrans;
241	args->readahead = oargs->readahead;
242	args->hostname = oargs->hostname;
243}
244
245static void
246nfs_convert_diskless(void)
247{
248
249	bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif,
250		sizeof(struct ifaliasreq));
251	bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway,
252		sizeof(struct sockaddr_in));
253	nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args);
254	if (nfsv3_diskless.root_args.flags & NFSMNT_NFSV3) {
255		nfsv3_diskless.root_fhsize = NFSX_MYFH;
256		bcopy(nfs_diskless.root_fh, nfsv3_diskless.root_fh, NFSX_MYFH);
257	} else {
258		nfsv3_diskless.root_fhsize = NFSX_V2FH;
259		bcopy(nfs_diskless.root_fh, nfsv3_diskless.root_fh, NFSX_V2FH);
260	}
261	bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr,
262		sizeof(struct sockaddr_in));
263	bcopy(nfs_diskless.root_hostnam, nfsv3_diskless.root_hostnam, MNAMELEN);
264	nfsv3_diskless.root_time = nfs_diskless.root_time;
265	bcopy(nfs_diskless.my_hostnam, nfsv3_diskless.my_hostnam,
266		MAXHOSTNAMELEN);
267	nfs_diskless_valid = 3;
268}
269
270/*
271 * nfs statfs call
272 */
273static int
274nfs_statfs(struct mount *mp, struct statfs *sbp)
275{
276	struct vnode *vp;
277	struct thread *td;
278	struct nfsmount *nmp = VFSTONFS(mp);
279	struct nfsvattr nfsva;
280	struct nfsfsinfo fs;
281	struct nfsstatfs sb;
282	int error = 0, attrflag, gotfsinfo = 0, ret;
283	struct nfsnode *np;
284
285	td = curthread;
286
287	error = vfs_busy(mp, MBF_NOWAIT);
288	if (error)
289		return (error);
290	error = ncl_nget(mp, nmp->nm_fh, nmp->nm_fhsize, &np, LK_EXCLUSIVE);
291	if (error) {
292		vfs_unbusy(mp);
293		return (error);
294	}
295	vp = NFSTOV(np);
296	mtx_lock(&nmp->nm_mtx);
297	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) {
298		mtx_unlock(&nmp->nm_mtx);
299		error = nfsrpc_fsinfo(vp, &fs, td->td_ucred, td, &nfsva,
300		    &attrflag, NULL);
301		if (!error)
302			gotfsinfo = 1;
303	} else
304		mtx_unlock(&nmp->nm_mtx);
305	if (!error)
306		error = nfsrpc_statfs(vp, &sb, &fs, td->td_ucred, td, &nfsva,
307		    &attrflag, NULL);
308	if (error != 0)
309		NFSCL_DEBUG(2, "statfs=%d\n", error);
310	if (attrflag == 0) {
311		ret = nfsrpc_getattrnovp(nmp, nmp->nm_fh, nmp->nm_fhsize, 1,
312		    td->td_ucred, td, &nfsva, NULL, NULL);
313		if (ret) {
314			/*
315			 * Just set default values to get things going.
316			 */
317			NFSBZERO((caddr_t)&nfsva, sizeof (struct nfsvattr));
318			nfsva.na_vattr.va_type = VDIR;
319			nfsva.na_vattr.va_mode = 0777;
320			nfsva.na_vattr.va_nlink = 100;
321			nfsva.na_vattr.va_uid = (uid_t)0;
322			nfsva.na_vattr.va_gid = (gid_t)0;
323			nfsva.na_vattr.va_fileid = 2;
324			nfsva.na_vattr.va_gen = 1;
325			nfsva.na_vattr.va_blocksize = NFS_FABLKSIZE;
326			nfsva.na_vattr.va_size = 512 * 1024;
327		}
328	}
329	(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
330	if (!error) {
331	    mtx_lock(&nmp->nm_mtx);
332	    if (gotfsinfo || (nmp->nm_flag & NFSMNT_NFSV4))
333		nfscl_loadfsinfo(nmp, &fs);
334	    nfscl_loadsbinfo(nmp, &sb, sbp);
335	    sbp->f_iosize = newnfs_iosize(nmp);
336	    mtx_unlock(&nmp->nm_mtx);
337	    if (sbp != &mp->mnt_stat) {
338		bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN);
339		bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
340	    }
341	    strncpy(&sbp->f_fstypename[0], mp->mnt_vfc->vfc_name, MFSNAMELEN);
342	} else if (NFS_ISV4(vp)) {
343		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
344	}
345	vput(vp);
346	vfs_unbusy(mp);
347	return (error);
348}
349
350/*
351 * nfs version 3 fsinfo rpc call
352 */
353int
354ncl_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct ucred *cred,
355    struct thread *td)
356{
357	struct nfsfsinfo fs;
358	struct nfsvattr nfsva;
359	int error, attrflag;
360
361	error = nfsrpc_fsinfo(vp, &fs, cred, td, &nfsva, &attrflag, NULL);
362	if (!error) {
363		if (attrflag)
364			(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
365			    1);
366		mtx_lock(&nmp->nm_mtx);
367		nfscl_loadfsinfo(nmp, &fs);
368		mtx_unlock(&nmp->nm_mtx);
369	}
370	return (error);
371}
372
373/*
374 * Mount a remote root fs via. nfs. This depends on the info in the
375 * nfs_diskless structure that has been filled in properly by some primary
376 * bootstrap.
377 * It goes something like this:
378 * - do enough of "ifconfig" by calling ifioctl() so that the system
379 *   can talk to the server
380 * - If nfs_diskless.mygateway is filled in, use that address as
381 *   a default gateway.
382 * - build the rootfs mount point and call mountnfs() to do the rest.
383 *
384 * It is assumed to be safe to read, modify, and write the nfsv3_diskless
385 * structure, as well as other global NFS client variables here, as
386 * nfs_mountroot() will be called once in the boot before any other NFS
387 * client activity occurs.
388 */
389static int
390nfs_mountroot(struct mount *mp)
391{
392	struct thread *td = curthread;
393	struct nfsv3_diskless *nd = &nfsv3_diskless;
394	struct socket *so;
395	struct vnode *vp;
396	struct ifreq ir;
397	int error;
398	u_long l;
399	char buf[128];
400	char *cp;
401
402#if defined(BOOTP_NFSROOT) && defined(BOOTP)
403	bootpc_init();		/* use bootp to get nfs_diskless filled in */
404#elif defined(NFS_ROOT)
405	nfs_setup_diskless();
406#endif
407
408	if (nfs_diskless_valid == 0)
409		return (-1);
410	if (nfs_diskless_valid == 1)
411		nfs_convert_diskless();
412
413	/*
414	 * XXX splnet, so networks will receive...
415	 */
416	splnet();
417
418	/*
419	 * Do enough of ifconfig(8) so that the critical net interface can
420	 * talk to the server.
421	 */
422	error = socreate(nd->myif.ifra_addr.sa_family, &so, nd->root_args.sotype, 0,
423	    td->td_ucred, td);
424	if (error)
425		panic("nfs_mountroot: socreate(%04x): %d",
426			nd->myif.ifra_addr.sa_family, error);
427
428#if 0 /* XXX Bad idea */
429	/*
430	 * We might not have been told the right interface, so we pass
431	 * over the first ten interfaces of the same kind, until we get
432	 * one of them configured.
433	 */
434
435	for (i = strlen(nd->myif.ifra_name) - 1;
436		nd->myif.ifra_name[i] >= '0' &&
437		nd->myif.ifra_name[i] <= '9';
438		nd->myif.ifra_name[i] ++) {
439		error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td);
440		if(!error)
441			break;
442	}
443#endif
444	error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td);
445	if (error)
446		panic("nfs_mountroot: SIOCAIFADDR: %d", error);
447	if ((cp = getenv("boot.netif.mtu")) != NULL) {
448		ir.ifr_mtu = strtol(cp, NULL, 10);
449		bcopy(nd->myif.ifra_name, ir.ifr_name, IFNAMSIZ);
450		freeenv(cp);
451		error = ifioctl(so, SIOCSIFMTU, (caddr_t)&ir, td);
452		if (error)
453			printf("nfs_mountroot: SIOCSIFMTU: %d", error);
454	}
455	soclose(so);
456
457	/*
458	 * If the gateway field is filled in, set it as the default route.
459	 * Note that pxeboot will set a default route of 0 if the route
460	 * is not set by the DHCP server.  Check also for a value of 0
461	 * to avoid panicking inappropriately in that situation.
462	 */
463	if (nd->mygateway.sin_len != 0 &&
464	    nd->mygateway.sin_addr.s_addr != 0) {
465		struct sockaddr_in mask, sin;
466
467		bzero((caddr_t)&mask, sizeof(mask));
468		sin = mask;
469		sin.sin_family = AF_INET;
470		sin.sin_len = sizeof(sin);
471                /* XXX MRT use table 0 for this sort of thing */
472		CURVNET_SET(TD_TO_VNET(td));
473		error = rtrequest_fib(RTM_ADD, (struct sockaddr *)&sin,
474		    (struct sockaddr *)&nd->mygateway,
475		    (struct sockaddr *)&mask,
476		    RTF_UP | RTF_GATEWAY, NULL, RT_DEFAULT_FIB);
477		CURVNET_RESTORE();
478		if (error)
479			panic("nfs_mountroot: RTM_ADD: %d", error);
480	}
481
482	/*
483	 * Create the rootfs mount point.
484	 */
485	nd->root_args.fh = nd->root_fh;
486	nd->root_args.fhsize = nd->root_fhsize;
487	l = ntohl(nd->root_saddr.sin_addr.s_addr);
488	snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
489		(l >> 24) & 0xff, (l >> 16) & 0xff,
490		(l >>  8) & 0xff, (l >>  0) & 0xff, nd->root_hostnam);
491	printf("NFS ROOT: %s\n", buf);
492	nd->root_args.hostname = buf;
493	if ((error = nfs_mountdiskless(buf,
494	    &nd->root_saddr, &nd->root_args, td, &vp, mp)) != 0) {
495		return (error);
496	}
497
498	/*
499	 * This is not really an nfs issue, but it is much easier to
500	 * set hostname here and then let the "/etc/rc.xxx" files
501	 * mount the right /var based upon its preset value.
502	 */
503	mtx_lock(&prison0.pr_mtx);
504	strlcpy(prison0.pr_hostname, nd->my_hostnam,
505	    sizeof(prison0.pr_hostname));
506	mtx_unlock(&prison0.pr_mtx);
507	inittodr(ntohl(nd->root_time));
508	return (0);
509}
510
511/*
512 * Internal version of mount system call for diskless setup.
513 */
514static int
515nfs_mountdiskless(char *path,
516    struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
517    struct vnode **vpp, struct mount *mp)
518{
519	struct sockaddr *nam;
520	int dirlen, error;
521	char *dirpath;
522
523	/*
524	 * Find the directory path in "path", which also has the server's
525	 * name/ip address in it.
526	 */
527	dirpath = strchr(path, ':');
528	if (dirpath != NULL)
529		dirlen = strlen(++dirpath);
530	else
531		dirlen = 0;
532	nam = sodupsockaddr((struct sockaddr *)sin, M_WAITOK);
533	if ((error = mountnfs(args, mp, nam, path, NULL, 0, dirpath, dirlen,
534	    NULL, 0, vpp, td->td_ucred, td, NFS_DEFAULT_NAMETIMEO,
535	    NFS_DEFAULT_NEGNAMETIMEO, 0)) != 0) {
536		printf("nfs_mountroot: mount %s on /: %d\n", path, error);
537		return (error);
538	}
539	return (0);
540}
541
542static void
543nfs_sec_name(char *sec, int *flagsp)
544{
545	if (!strcmp(sec, "krb5"))
546		*flagsp |= NFSMNT_KERB;
547	else if (!strcmp(sec, "krb5i"))
548		*flagsp |= (NFSMNT_KERB | NFSMNT_INTEGRITY);
549	else if (!strcmp(sec, "krb5p"))
550		*flagsp |= (NFSMNT_KERB | NFSMNT_PRIVACY);
551}
552
553static void
554nfs_decode_args(struct mount *mp, struct nfsmount *nmp, struct nfs_args *argp,
555    const char *hostname, struct ucred *cred, struct thread *td)
556{
557	int s;
558	int adjsock;
559	char *p;
560
561	s = splnet();
562
563	/*
564	 * Set read-only flag if requested; otherwise, clear it if this is
565	 * an update.  If this is not an update, then either the read-only
566	 * flag is already clear, or this is a root mount and it was set
567	 * intentionally at some previous point.
568	 */
569	if (vfs_getopt(mp->mnt_optnew, "ro", NULL, NULL) == 0) {
570		MNT_ILOCK(mp);
571		mp->mnt_flag |= MNT_RDONLY;
572		MNT_IUNLOCK(mp);
573	} else if (mp->mnt_flag & MNT_UPDATE) {
574		MNT_ILOCK(mp);
575		mp->mnt_flag &= ~MNT_RDONLY;
576		MNT_IUNLOCK(mp);
577	}
578
579	/*
580	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
581	 * no sense in that context.  Also, set up appropriate retransmit
582	 * and soft timeout behavior.
583	 */
584	if (argp->sotype == SOCK_STREAM) {
585		nmp->nm_flag &= ~NFSMNT_NOCONN;
586		nmp->nm_timeo = NFS_MAXTIMEO;
587		if ((argp->flags & NFSMNT_NFSV4) != 0)
588			nmp->nm_retry = INT_MAX;
589		else
590			nmp->nm_retry = NFS_RETRANS_TCP;
591	}
592
593	/* Also clear RDIRPLUS if NFSv2, it crashes some servers */
594	if ((argp->flags & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0) {
595		argp->flags &= ~NFSMNT_RDIRPLUS;
596		nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
597	}
598
599	/* Re-bind if rsrvd port requested and wasn't on one */
600	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
601		  && (argp->flags & NFSMNT_RESVPORT);
602	/* Also re-bind if we're switching to/from a connected UDP socket */
603	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
604		    (argp->flags & NFSMNT_NOCONN));
605
606	/* Update flags atomically.  Don't change the lock bits. */
607	nmp->nm_flag = argp->flags | nmp->nm_flag;
608	splx(s);
609
610	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
611		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
612		if (nmp->nm_timeo < NFS_MINTIMEO)
613			nmp->nm_timeo = NFS_MINTIMEO;
614		else if (nmp->nm_timeo > NFS_MAXTIMEO)
615			nmp->nm_timeo = NFS_MAXTIMEO;
616	}
617
618	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
619		nmp->nm_retry = argp->retrans;
620		if (nmp->nm_retry > NFS_MAXREXMIT)
621			nmp->nm_retry = NFS_MAXREXMIT;
622	}
623
624	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
625		nmp->nm_wsize = argp->wsize;
626		/*
627		 * Clip at the power of 2 below the size. There is an
628		 * issue (not isolated) that causes intermittent page
629		 * faults if this is not done.
630		 */
631		if (nmp->nm_wsize > NFS_FABLKSIZE)
632			nmp->nm_wsize = 1 << (fls(nmp->nm_wsize) - 1);
633		else
634			nmp->nm_wsize = NFS_FABLKSIZE;
635	}
636
637	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
638		nmp->nm_rsize = argp->rsize;
639		/*
640		 * Clip at the power of 2 below the size. There is an
641		 * issue (not isolated) that causes intermittent page
642		 * faults if this is not done.
643		 */
644		if (nmp->nm_rsize > NFS_FABLKSIZE)
645			nmp->nm_rsize = 1 << (fls(nmp->nm_rsize) - 1);
646		else
647			nmp->nm_rsize = NFS_FABLKSIZE;
648	}
649
650	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
651		nmp->nm_readdirsize = argp->readdirsize;
652	}
653
654	if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
655		nmp->nm_acregmin = argp->acregmin;
656	else
657		nmp->nm_acregmin = NFS_MINATTRTIMO;
658	if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
659		nmp->nm_acregmax = argp->acregmax;
660	else
661		nmp->nm_acregmax = NFS_MAXATTRTIMO;
662	if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
663		nmp->nm_acdirmin = argp->acdirmin;
664	else
665		nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
666	if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
667		nmp->nm_acdirmax = argp->acdirmax;
668	else
669		nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
670	if (nmp->nm_acdirmin > nmp->nm_acdirmax)
671		nmp->nm_acdirmin = nmp->nm_acdirmax;
672	if (nmp->nm_acregmin > nmp->nm_acregmax)
673		nmp->nm_acregmin = nmp->nm_acregmax;
674
675	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
676		if (argp->readahead <= NFS_MAXRAHEAD)
677			nmp->nm_readahead = argp->readahead;
678		else
679			nmp->nm_readahead = NFS_MAXRAHEAD;
680	}
681	if ((argp->flags & NFSMNT_WCOMMITSIZE) && argp->wcommitsize >= 0) {
682		if (argp->wcommitsize < nmp->nm_wsize)
683			nmp->nm_wcommitsize = nmp->nm_wsize;
684		else
685			nmp->nm_wcommitsize = argp->wcommitsize;
686	}
687
688	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
689		    (nmp->nm_soproto != argp->proto));
690
691	if (nmp->nm_client != NULL && adjsock) {
692		int haslock = 0, error = 0;
693
694		if (nmp->nm_sotype == SOCK_STREAM) {
695			error = newnfs_sndlock(&nmp->nm_sockreq.nr_lock);
696			if (!error)
697				haslock = 1;
698		}
699		if (!error) {
700		    newnfs_disconnect(&nmp->nm_sockreq);
701		    if (haslock)
702			newnfs_sndunlock(&nmp->nm_sockreq.nr_lock);
703		    nmp->nm_sotype = argp->sotype;
704		    nmp->nm_soproto = argp->proto;
705		    if (nmp->nm_sotype == SOCK_DGRAM)
706			while (newnfs_connect(nmp, &nmp->nm_sockreq,
707			    cred, td, 0)) {
708				printf("newnfs_args: retrying connect\n");
709				(void) nfs_catnap(PSOCK, 0, "newnfscon");
710			}
711		}
712	} else {
713		nmp->nm_sotype = argp->sotype;
714		nmp->nm_soproto = argp->proto;
715	}
716
717	if (hostname != NULL) {
718		strlcpy(nmp->nm_hostname, hostname,
719		    sizeof(nmp->nm_hostname));
720		p = strchr(nmp->nm_hostname, ':');
721		if (p != NULL)
722			*p = '\0';
723	}
724}
725
726static const char *nfs_opts[] = { "from", "nfs_args",
727    "noac", "noatime", "noexec", "suiddir", "nosuid", "nosymfollow", "union",
728    "noclusterr", "noclusterw", "multilabel", "acls", "force", "update",
729    "async", "noconn", "nolockd", "conn", "lockd", "intr", "rdirplus",
730    "readdirsize", "soft", "hard", "mntudp", "tcp", "udp", "wsize", "rsize",
731    "retrans", "actimeo", "acregmin", "acregmax", "acdirmin", "acdirmax",
732    "resvport", "readahead", "hostname", "timeo", "timeout", "addr", "fh",
733    "nfsv3", "sec", "principal", "nfsv4", "gssname", "allgssname", "dirpath",
734    "minorversion", "nametimeo", "negnametimeo", "nocto", "noncontigwr",
735    "pnfs", "wcommitsize",
736    NULL };
737
738/*
739 * VFS Operations.
740 *
741 * mount system call
742 * It seems a bit dumb to copyinstr() the host and path here and then
743 * bcopy() them in mountnfs(), but I wanted to detect errors before
744 * doing the sockargs() call because sockargs() allocates an mbuf and
745 * an error after that means that I have to release the mbuf.
746 */
747/* ARGSUSED */
748static int
749nfs_mount(struct mount *mp)
750{
751	struct nfs_args args = {
752	    .version = NFS_ARGSVERSION,
753	    .addr = NULL,
754	    .addrlen = sizeof (struct sockaddr_in),
755	    .sotype = SOCK_STREAM,
756	    .proto = 0,
757	    .fh = NULL,
758	    .fhsize = 0,
759	    .flags = NFSMNT_RESVPORT,
760	    .wsize = NFS_WSIZE,
761	    .rsize = NFS_RSIZE,
762	    .readdirsize = NFS_READDIRSIZE,
763	    .timeo = 10,
764	    .retrans = NFS_RETRANS,
765	    .readahead = NFS_DEFRAHEAD,
766	    .wcommitsize = 0,			/* was: NQ_DEFLEASE */
767	    .hostname = NULL,
768	    .acregmin = NFS_MINATTRTIMO,
769	    .acregmax = NFS_MAXATTRTIMO,
770	    .acdirmin = NFS_MINDIRATTRTIMO,
771	    .acdirmax = NFS_MAXDIRATTRTIMO,
772	};
773	int error = 0, ret, len;
774	struct sockaddr *nam = NULL;
775	struct vnode *vp;
776	struct thread *td;
777	char hst[MNAMELEN];
778	u_char nfh[NFSX_FHMAX], krbname[100], dirpath[100], srvkrbname[100];
779	char *cp, *opt, *name, *secname;
780	int nametimeo = NFS_DEFAULT_NAMETIMEO;
781	int negnametimeo = NFS_DEFAULT_NEGNAMETIMEO;
782	int minvers = 0;
783	int dirlen, has_nfs_args_opt, krbnamelen, srvkrbnamelen;
784	size_t hstlen;
785
786	has_nfs_args_opt = 0;
787	if (vfs_filteropt(mp->mnt_optnew, nfs_opts)) {
788		error = EINVAL;
789		goto out;
790	}
791
792	td = curthread;
793	if ((mp->mnt_flag & (MNT_ROOTFS | MNT_UPDATE)) == MNT_ROOTFS) {
794		error = nfs_mountroot(mp);
795		goto out;
796	}
797
798	nfscl_init();
799
800	/*
801	 * The old mount_nfs program passed the struct nfs_args
802	 * from userspace to kernel.  The new mount_nfs program
803	 * passes string options via nmount() from userspace to kernel
804	 * and we populate the struct nfs_args in the kernel.
805	 */
806	if (vfs_getopt(mp->mnt_optnew, "nfs_args", NULL, NULL) == 0) {
807		error = vfs_copyopt(mp->mnt_optnew, "nfs_args", &args,
808		    sizeof(args));
809		if (error != 0)
810			goto out;
811
812		if (args.version != NFS_ARGSVERSION) {
813			error = EPROGMISMATCH;
814			goto out;
815		}
816		has_nfs_args_opt = 1;
817	}
818
819	/* Handle the new style options. */
820	if (vfs_getopt(mp->mnt_optnew, "noac", NULL, NULL) == 0) {
821		args.acdirmin = args.acdirmax =
822		    args.acregmin = args.acregmax = 0;
823		args.flags |= NFSMNT_ACDIRMIN | NFSMNT_ACDIRMAX |
824		    NFSMNT_ACREGMIN | NFSMNT_ACREGMAX;
825	}
826	if (vfs_getopt(mp->mnt_optnew, "noconn", NULL, NULL) == 0)
827		args.flags |= NFSMNT_NOCONN;
828	if (vfs_getopt(mp->mnt_optnew, "conn", NULL, NULL) == 0)
829		args.flags &= ~NFSMNT_NOCONN;
830	if (vfs_getopt(mp->mnt_optnew, "nolockd", NULL, NULL) == 0)
831		args.flags |= NFSMNT_NOLOCKD;
832	if (vfs_getopt(mp->mnt_optnew, "lockd", NULL, NULL) == 0)
833		args.flags &= ~NFSMNT_NOLOCKD;
834	if (vfs_getopt(mp->mnt_optnew, "intr", NULL, NULL) == 0)
835		args.flags |= NFSMNT_INT;
836	if (vfs_getopt(mp->mnt_optnew, "rdirplus", NULL, NULL) == 0)
837		args.flags |= NFSMNT_RDIRPLUS;
838	if (vfs_getopt(mp->mnt_optnew, "resvport", NULL, NULL) == 0)
839		args.flags |= NFSMNT_RESVPORT;
840	if (vfs_getopt(mp->mnt_optnew, "noresvport", NULL, NULL) == 0)
841		args.flags &= ~NFSMNT_RESVPORT;
842	if (vfs_getopt(mp->mnt_optnew, "soft", NULL, NULL) == 0)
843		args.flags |= NFSMNT_SOFT;
844	if (vfs_getopt(mp->mnt_optnew, "hard", NULL, NULL) == 0)
845		args.flags &= ~NFSMNT_SOFT;
846	if (vfs_getopt(mp->mnt_optnew, "mntudp", NULL, NULL) == 0)
847		args.sotype = SOCK_DGRAM;
848	if (vfs_getopt(mp->mnt_optnew, "udp", NULL, NULL) == 0)
849		args.sotype = SOCK_DGRAM;
850	if (vfs_getopt(mp->mnt_optnew, "tcp", NULL, NULL) == 0)
851		args.sotype = SOCK_STREAM;
852	if (vfs_getopt(mp->mnt_optnew, "nfsv3", NULL, NULL) == 0)
853		args.flags |= NFSMNT_NFSV3;
854	if (vfs_getopt(mp->mnt_optnew, "nfsv4", NULL, NULL) == 0) {
855		args.flags |= NFSMNT_NFSV4;
856		args.sotype = SOCK_STREAM;
857	}
858	if (vfs_getopt(mp->mnt_optnew, "allgssname", NULL, NULL) == 0)
859		args.flags |= NFSMNT_ALLGSSNAME;
860	if (vfs_getopt(mp->mnt_optnew, "nocto", NULL, NULL) == 0)
861		args.flags |= NFSMNT_NOCTO;
862	if (vfs_getopt(mp->mnt_optnew, "noncontigwr", NULL, NULL) == 0)
863		args.flags |= NFSMNT_NONCONTIGWR;
864	if (vfs_getopt(mp->mnt_optnew, "pnfs", NULL, NULL) == 0)
865		args.flags |= NFSMNT_PNFS;
866	if (vfs_getopt(mp->mnt_optnew, "readdirsize", (void **)&opt, NULL) == 0) {
867		if (opt == NULL) {
868			vfs_mount_error(mp, "illegal readdirsize");
869			error = EINVAL;
870			goto out;
871		}
872		ret = sscanf(opt, "%d", &args.readdirsize);
873		if (ret != 1 || args.readdirsize <= 0) {
874			vfs_mount_error(mp, "illegal readdirsize: %s",
875			    opt);
876			error = EINVAL;
877			goto out;
878		}
879		args.flags |= NFSMNT_READDIRSIZE;
880	}
881	if (vfs_getopt(mp->mnt_optnew, "readahead", (void **)&opt, NULL) == 0) {
882		if (opt == NULL) {
883			vfs_mount_error(mp, "illegal readahead");
884			error = EINVAL;
885			goto out;
886		}
887		ret = sscanf(opt, "%d", &args.readahead);
888		if (ret != 1 || args.readahead <= 0) {
889			vfs_mount_error(mp, "illegal readahead: %s",
890			    opt);
891			error = EINVAL;
892			goto out;
893		}
894		args.flags |= NFSMNT_READAHEAD;
895	}
896	if (vfs_getopt(mp->mnt_optnew, "wsize", (void **)&opt, NULL) == 0) {
897		if (opt == NULL) {
898			vfs_mount_error(mp, "illegal wsize");
899			error = EINVAL;
900			goto out;
901		}
902		ret = sscanf(opt, "%d", &args.wsize);
903		if (ret != 1 || args.wsize <= 0) {
904			vfs_mount_error(mp, "illegal wsize: %s",
905			    opt);
906			error = EINVAL;
907			goto out;
908		}
909		args.flags |= NFSMNT_WSIZE;
910	}
911	if (vfs_getopt(mp->mnt_optnew, "rsize", (void **)&opt, NULL) == 0) {
912		if (opt == NULL) {
913			vfs_mount_error(mp, "illegal rsize");
914			error = EINVAL;
915			goto out;
916		}
917		ret = sscanf(opt, "%d", &args.rsize);
918		if (ret != 1 || args.rsize <= 0) {
919			vfs_mount_error(mp, "illegal wsize: %s",
920			    opt);
921			error = EINVAL;
922			goto out;
923		}
924		args.flags |= NFSMNT_RSIZE;
925	}
926	if (vfs_getopt(mp->mnt_optnew, "retrans", (void **)&opt, NULL) == 0) {
927		if (opt == NULL) {
928			vfs_mount_error(mp, "illegal retrans");
929			error = EINVAL;
930			goto out;
931		}
932		ret = sscanf(opt, "%d", &args.retrans);
933		if (ret != 1 || args.retrans <= 0) {
934			vfs_mount_error(mp, "illegal retrans: %s",
935			    opt);
936			error = EINVAL;
937			goto out;
938		}
939		args.flags |= NFSMNT_RETRANS;
940	}
941	if (vfs_getopt(mp->mnt_optnew, "actimeo", (void **)&opt, NULL) == 0) {
942		ret = sscanf(opt, "%d", &args.acregmin);
943		if (ret != 1 || args.acregmin < 0) {
944			vfs_mount_error(mp, "illegal actimeo: %s",
945			    opt);
946			error = EINVAL;
947			goto out;
948		}
949		args.acdirmin = args.acdirmax = args.acregmax = args.acregmin;
950		args.flags |= NFSMNT_ACDIRMIN | NFSMNT_ACDIRMAX |
951		    NFSMNT_ACREGMIN | NFSMNT_ACREGMAX;
952	}
953	if (vfs_getopt(mp->mnt_optnew, "acregmin", (void **)&opt, NULL) == 0) {
954		ret = sscanf(opt, "%d", &args.acregmin);
955		if (ret != 1 || args.acregmin < 0) {
956			vfs_mount_error(mp, "illegal acregmin: %s",
957			    opt);
958			error = EINVAL;
959			goto out;
960		}
961		args.flags |= NFSMNT_ACREGMIN;
962	}
963	if (vfs_getopt(mp->mnt_optnew, "acregmax", (void **)&opt, NULL) == 0) {
964		ret = sscanf(opt, "%d", &args.acregmax);
965		if (ret != 1 || args.acregmax < 0) {
966			vfs_mount_error(mp, "illegal acregmax: %s",
967			    opt);
968			error = EINVAL;
969			goto out;
970		}
971		args.flags |= NFSMNT_ACREGMAX;
972	}
973	if (vfs_getopt(mp->mnt_optnew, "acdirmin", (void **)&opt, NULL) == 0) {
974		ret = sscanf(opt, "%d", &args.acdirmin);
975		if (ret != 1 || args.acdirmin < 0) {
976			vfs_mount_error(mp, "illegal acdirmin: %s",
977			    opt);
978			error = EINVAL;
979			goto out;
980		}
981		args.flags |= NFSMNT_ACDIRMIN;
982	}
983	if (vfs_getopt(mp->mnt_optnew, "acdirmax", (void **)&opt, NULL) == 0) {
984		ret = sscanf(opt, "%d", &args.acdirmax);
985		if (ret != 1 || args.acdirmax < 0) {
986			vfs_mount_error(mp, "illegal acdirmax: %s",
987			    opt);
988			error = EINVAL;
989			goto out;
990		}
991		args.flags |= NFSMNT_ACDIRMAX;
992	}
993	if (vfs_getopt(mp->mnt_optnew, "wcommitsize", (void **)&opt, NULL) == 0) {
994		ret = sscanf(opt, "%d", &args.wcommitsize);
995		if (ret != 1 || args.wcommitsize < 0) {
996			vfs_mount_error(mp, "illegal wcommitsize: %s", opt);
997			error = EINVAL;
998			goto out;
999		}
1000		args.flags |= NFSMNT_WCOMMITSIZE;
1001	}
1002	if (vfs_getopt(mp->mnt_optnew, "timeo", (void **)&opt, NULL) == 0) {
1003		ret = sscanf(opt, "%d", &args.timeo);
1004		if (ret != 1 || args.timeo <= 0) {
1005			vfs_mount_error(mp, "illegal timeo: %s",
1006			    opt);
1007			error = EINVAL;
1008			goto out;
1009		}
1010		args.flags |= NFSMNT_TIMEO;
1011	}
1012	if (vfs_getopt(mp->mnt_optnew, "timeout", (void **)&opt, NULL) == 0) {
1013		ret = sscanf(opt, "%d", &args.timeo);
1014		if (ret != 1 || args.timeo <= 0) {
1015			vfs_mount_error(mp, "illegal timeout: %s",
1016			    opt);
1017			error = EINVAL;
1018			goto out;
1019		}
1020		args.flags |= NFSMNT_TIMEO;
1021	}
1022	if (vfs_getopt(mp->mnt_optnew, "nametimeo", (void **)&opt, NULL) == 0) {
1023		ret = sscanf(opt, "%d", &nametimeo);
1024		if (ret != 1 || nametimeo < 0) {
1025			vfs_mount_error(mp, "illegal nametimeo: %s", opt);
1026			error = EINVAL;
1027			goto out;
1028		}
1029	}
1030	if (vfs_getopt(mp->mnt_optnew, "negnametimeo", (void **)&opt, NULL)
1031	    == 0) {
1032		ret = sscanf(opt, "%d", &negnametimeo);
1033		if (ret != 1 || negnametimeo < 0) {
1034			vfs_mount_error(mp, "illegal negnametimeo: %s",
1035			    opt);
1036			error = EINVAL;
1037			goto out;
1038		}
1039	}
1040	if (vfs_getopt(mp->mnt_optnew, "minorversion", (void **)&opt, NULL) ==
1041	    0) {
1042		ret = sscanf(opt, "%d", &minvers);
1043		if (ret != 1 || minvers < 0 || minvers > 1 ||
1044		    (args.flags & NFSMNT_NFSV4) == 0) {
1045			vfs_mount_error(mp, "illegal minorversion: %s", opt);
1046			error = EINVAL;
1047			goto out;
1048		}
1049	}
1050	if (vfs_getopt(mp->mnt_optnew, "sec",
1051		(void **) &secname, NULL) == 0)
1052		nfs_sec_name(secname, &args.flags);
1053
1054	if (mp->mnt_flag & MNT_UPDATE) {
1055		struct nfsmount *nmp = VFSTONFS(mp);
1056
1057		if (nmp == NULL) {
1058			error = EIO;
1059			goto out;
1060		}
1061
1062		/*
1063		 * If a change from TCP->UDP is done and there are thread(s)
1064		 * that have I/O RPC(s) in progress with a tranfer size
1065		 * greater than NFS_MAXDGRAMDATA, those thread(s) will be
1066		 * hung, retrying the RPC(s) forever. Usually these threads
1067		 * will be seen doing an uninterruptible sleep on wait channel
1068		 * "newnfsreq" (truncated to "newnfsre" by procstat).
1069		 */
1070		if (args.sotype == SOCK_DGRAM && nmp->nm_sotype == SOCK_STREAM)
1071			tprintf(td->td_proc, LOG_WARNING,
1072	"Warning: mount -u that changes TCP->UDP can result in hung threads\n");
1073
1074		/*
1075		 * When doing an update, we can't change version,
1076		 * security, switch lockd strategies or change cookie
1077		 * translation
1078		 */
1079		args.flags = (args.flags &
1080		    ~(NFSMNT_NFSV3 |
1081		      NFSMNT_NFSV4 |
1082		      NFSMNT_KERB |
1083		      NFSMNT_INTEGRITY |
1084		      NFSMNT_PRIVACY |
1085		      NFSMNT_NOLOCKD /*|NFSMNT_XLATECOOKIE*/)) |
1086		    (nmp->nm_flag &
1087			(NFSMNT_NFSV3 |
1088			 NFSMNT_NFSV4 |
1089			 NFSMNT_KERB |
1090			 NFSMNT_INTEGRITY |
1091			 NFSMNT_PRIVACY |
1092			 NFSMNT_NOLOCKD /*|NFSMNT_XLATECOOKIE*/));
1093		nfs_decode_args(mp, nmp, &args, NULL, td->td_ucred, td);
1094		goto out;
1095	}
1096
1097	/*
1098	 * Make the nfs_ip_paranoia sysctl serve as the default connection
1099	 * or no-connection mode for those protocols that support
1100	 * no-connection mode (the flag will be cleared later for protocols
1101	 * that do not support no-connection mode).  This will allow a client
1102	 * to receive replies from a different IP then the request was
1103	 * sent to.  Note: default value for nfs_ip_paranoia is 1 (paranoid),
1104	 * not 0.
1105	 */
1106	if (nfs_ip_paranoia == 0)
1107		args.flags |= NFSMNT_NOCONN;
1108
1109	if (has_nfs_args_opt != 0) {
1110		/*
1111		 * In the 'nfs_args' case, the pointers in the args
1112		 * structure are in userland - we copy them in here.
1113		 */
1114		if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX) {
1115			vfs_mount_error(mp, "Bad file handle");
1116			error = EINVAL;
1117			goto out;
1118		}
1119		error = copyin((caddr_t)args.fh, (caddr_t)nfh,
1120		    args.fhsize);
1121		if (error != 0)
1122			goto out;
1123		error = copyinstr(args.hostname, hst, MNAMELEN - 1, &hstlen);
1124		if (error != 0)
1125			goto out;
1126		bzero(&hst[hstlen], MNAMELEN - hstlen);
1127		args.hostname = hst;
1128		/* sockargs() call must be after above copyin() calls */
1129		error = getsockaddr(&nam, (caddr_t)args.addr,
1130		    args.addrlen);
1131		if (error != 0)
1132			goto out;
1133	} else {
1134		if (vfs_getopt(mp->mnt_optnew, "fh", (void **)&args.fh,
1135		    &args.fhsize) == 0) {
1136			if (args.fhsize < 0 || args.fhsize > NFSX_FHMAX) {
1137				vfs_mount_error(mp, "Bad file handle");
1138				error = EINVAL;
1139				goto out;
1140			}
1141			bcopy(args.fh, nfh, args.fhsize);
1142		} else {
1143			args.fhsize = 0;
1144		}
1145		(void) vfs_getopt(mp->mnt_optnew, "hostname",
1146		    (void **)&args.hostname, &len);
1147		if (args.hostname == NULL) {
1148			vfs_mount_error(mp, "Invalid hostname");
1149			error = EINVAL;
1150			goto out;
1151		}
1152		if (len >= MNAMELEN) {
1153			vfs_mount_error(mp, "Hostname too long");
1154			error = EINVAL;
1155			goto out;
1156		}
1157		bcopy(args.hostname, hst, len);
1158		hst[len] = '\0';
1159	}
1160
1161	if (vfs_getopt(mp->mnt_optnew, "principal", (void **)&name, NULL) == 0)
1162		strlcpy(srvkrbname, name, sizeof (srvkrbname));
1163	else {
1164		snprintf(srvkrbname, sizeof (srvkrbname), "nfs@%s", hst);
1165		cp = strchr(srvkrbname, ':');
1166		if (cp != NULL)
1167			*cp = '\0';
1168	}
1169	srvkrbnamelen = strlen(srvkrbname);
1170
1171	if (vfs_getopt(mp->mnt_optnew, "gssname", (void **)&name, NULL) == 0)
1172		strlcpy(krbname, name, sizeof (krbname));
1173	else
1174		krbname[0] = '\0';
1175	krbnamelen = strlen(krbname);
1176
1177	if (vfs_getopt(mp->mnt_optnew, "dirpath", (void **)&name, NULL) == 0)
1178		strlcpy(dirpath, name, sizeof (dirpath));
1179	else
1180		dirpath[0] = '\0';
1181	dirlen = strlen(dirpath);
1182
1183	if (has_nfs_args_opt == 0) {
1184		if (vfs_getopt(mp->mnt_optnew, "addr",
1185		    (void **)&args.addr, &args.addrlen) == 0) {
1186			if (args.addrlen > SOCK_MAXADDRLEN) {
1187				error = ENAMETOOLONG;
1188				goto out;
1189			}
1190			nam = malloc(args.addrlen, M_SONAME, M_WAITOK);
1191			bcopy(args.addr, nam, args.addrlen);
1192			nam->sa_len = args.addrlen;
1193		} else {
1194			vfs_mount_error(mp, "No server address");
1195			error = EINVAL;
1196			goto out;
1197		}
1198	}
1199
1200	args.fh = nfh;
1201	error = mountnfs(&args, mp, nam, hst, krbname, krbnamelen, dirpath,
1202	    dirlen, srvkrbname, srvkrbnamelen, &vp, td->td_ucred, td,
1203	    nametimeo, negnametimeo, minvers);
1204out:
1205	if (!error) {
1206		MNT_ILOCK(mp);
1207		mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_NO_IOPF |
1208		    MNTK_USES_BCACHE;
1209		if ((VFSTONFS(mp)->nm_flag & NFSMNT_NFSV4) != 0)
1210			mp->mnt_kern_flag |= MNTK_NULL_NOCACHE;
1211		MNT_IUNLOCK(mp);
1212	}
1213	return (error);
1214}
1215
1216
1217/*
1218 * VFS Operations.
1219 *
1220 * mount system call
1221 * It seems a bit dumb to copyinstr() the host and path here and then
1222 * bcopy() them in mountnfs(), but I wanted to detect errors before
1223 * doing the sockargs() call because sockargs() allocates an mbuf and
1224 * an error after that means that I have to release the mbuf.
1225 */
1226/* ARGSUSED */
1227static int
1228nfs_cmount(struct mntarg *ma, void *data, uint64_t flags)
1229{
1230	int error;
1231	struct nfs_args args;
1232
1233	error = copyin(data, &args, sizeof (struct nfs_args));
1234	if (error)
1235		return error;
1236
1237	ma = mount_arg(ma, "nfs_args", &args, sizeof args);
1238
1239	error = kernel_mount(ma, flags);
1240	return (error);
1241}
1242
1243/*
1244 * Common code for mount and mountroot
1245 */
1246static int
1247mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
1248    char *hst, u_char *krbname, int krbnamelen, u_char *dirpath, int dirlen,
1249    u_char *srvkrbname, int srvkrbnamelen, struct vnode **vpp,
1250    struct ucred *cred, struct thread *td, int nametimeo, int negnametimeo,
1251    int minvers)
1252{
1253	struct nfsmount *nmp;
1254	struct nfsnode *np;
1255	int error, trycnt, ret;
1256	struct nfsvattr nfsva;
1257	struct nfsclclient *clp;
1258	struct nfsclds *dsp, *tdsp;
1259	uint32_t lease;
1260	static u_int64_t clval = 0;
1261
1262	NFSCL_DEBUG(3, "in mnt\n");
1263	clp = NULL;
1264	if (mp->mnt_flag & MNT_UPDATE) {
1265		nmp = VFSTONFS(mp);
1266		printf("%s: MNT_UPDATE is no longer handled here\n", __func__);
1267		FREE(nam, M_SONAME);
1268		return (0);
1269	} else {
1270		MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount) +
1271		    krbnamelen + dirlen + srvkrbnamelen + 2,
1272		    M_NEWNFSMNT, M_WAITOK | M_ZERO);
1273		TAILQ_INIT(&nmp->nm_bufq);
1274		TAILQ_INIT(&nmp->nm_sess);
1275		if (clval == 0)
1276			clval = (u_int64_t)nfsboottime.tv_sec;
1277		nmp->nm_clval = clval++;
1278		nmp->nm_krbnamelen = krbnamelen;
1279		nmp->nm_dirpathlen = dirlen;
1280		nmp->nm_srvkrbnamelen = srvkrbnamelen;
1281		if (td->td_ucred->cr_uid != (uid_t)0) {
1282			/*
1283			 * nm_uid is used to get KerberosV credentials for
1284			 * the nfsv4 state handling operations if there is
1285			 * no host based principal set. Use the uid of
1286			 * this user if not root, since they are doing the
1287			 * mount. I don't think setting this for root will
1288			 * work, since root normally does not have user
1289			 * credentials in a credentials cache.
1290			 */
1291			nmp->nm_uid = td->td_ucred->cr_uid;
1292		} else {
1293			/*
1294			 * Just set to -1, so it won't be used.
1295			 */
1296			nmp->nm_uid = (uid_t)-1;
1297		}
1298
1299		/* Copy and null terminate all the names */
1300		if (nmp->nm_krbnamelen > 0) {
1301			bcopy(krbname, nmp->nm_krbname, nmp->nm_krbnamelen);
1302			nmp->nm_name[nmp->nm_krbnamelen] = '\0';
1303		}
1304		if (nmp->nm_dirpathlen > 0) {
1305			bcopy(dirpath, NFSMNT_DIRPATH(nmp),
1306			    nmp->nm_dirpathlen);
1307			nmp->nm_name[nmp->nm_krbnamelen + nmp->nm_dirpathlen
1308			    + 1] = '\0';
1309		}
1310		if (nmp->nm_srvkrbnamelen > 0) {
1311			bcopy(srvkrbname, NFSMNT_SRVKRBNAME(nmp),
1312			    nmp->nm_srvkrbnamelen);
1313			nmp->nm_name[nmp->nm_krbnamelen + nmp->nm_dirpathlen
1314			    + nmp->nm_srvkrbnamelen + 2] = '\0';
1315		}
1316		nmp->nm_sockreq.nr_cred = crhold(cred);
1317		mtx_init(&nmp->nm_sockreq.nr_mtx, "nfssock", NULL, MTX_DEF);
1318		mp->mnt_data = nmp;
1319		nmp->nm_getinfo = nfs_getnlminfo;
1320		nmp->nm_vinvalbuf = ncl_vinvalbuf;
1321	}
1322	vfs_getnewfsid(mp);
1323	nmp->nm_mountp = mp;
1324	mtx_init(&nmp->nm_mtx, "NFSmount lock", NULL, MTX_DEF | MTX_DUPOK);
1325
1326	/*
1327	 * Since nfs_decode_args() might optionally set them, these
1328	 * need to be set to defaults before the call, so that the
1329	 * optional settings aren't overwritten.
1330	 */
1331	nmp->nm_nametimeo = nametimeo;
1332	nmp->nm_negnametimeo = negnametimeo;
1333	nmp->nm_timeo = NFS_TIMEO;
1334	nmp->nm_retry = NFS_RETRANS;
1335	nmp->nm_readahead = NFS_DEFRAHEAD;
1336
1337	/* This is empirical approximation of sqrt(hibufspace) * 256. */
1338	nmp->nm_wcommitsize = NFS_MAXBSIZE / 256;
1339	while ((long)nmp->nm_wcommitsize * nmp->nm_wcommitsize < hibufspace)
1340		nmp->nm_wcommitsize *= 2;
1341	nmp->nm_wcommitsize *= 256;
1342
1343	if ((argp->flags & NFSMNT_NFSV4) != 0)
1344		nmp->nm_minorvers = minvers;
1345	else
1346		nmp->nm_minorvers = 0;
1347
1348	nfs_decode_args(mp, nmp, argp, hst, cred, td);
1349
1350	/*
1351	 * V2 can only handle 32 bit filesizes.  A 4GB-1 limit may be too
1352	 * high, depending on whether we end up with negative offsets in
1353	 * the client or server somewhere.  2GB-1 may be safer.
1354	 *
1355	 * For V3, ncl_fsinfo will adjust this as necessary.  Assume maximum
1356	 * that we can handle until we find out otherwise.
1357	 */
1358	if ((argp->flags & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0)
1359		nmp->nm_maxfilesize = 0xffffffffLL;
1360	else
1361		nmp->nm_maxfilesize = OFF_MAX;
1362
1363	if ((argp->flags & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0) {
1364		nmp->nm_wsize = NFS_WSIZE;
1365		nmp->nm_rsize = NFS_RSIZE;
1366		nmp->nm_readdirsize = NFS_READDIRSIZE;
1367	}
1368	nmp->nm_numgrps = NFS_MAXGRPS;
1369	nmp->nm_tprintf_delay = nfs_tprintf_delay;
1370	if (nmp->nm_tprintf_delay < 0)
1371		nmp->nm_tprintf_delay = 0;
1372	nmp->nm_tprintf_initial_delay = nfs_tprintf_initial_delay;
1373	if (nmp->nm_tprintf_initial_delay < 0)
1374		nmp->nm_tprintf_initial_delay = 0;
1375	nmp->nm_fhsize = argp->fhsize;
1376	if (nmp->nm_fhsize > 0)
1377		bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
1378	bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
1379	nmp->nm_nam = nam;
1380	/* Set up the sockets and per-host congestion */
1381	nmp->nm_sotype = argp->sotype;
1382	nmp->nm_soproto = argp->proto;
1383	nmp->nm_sockreq.nr_prog = NFS_PROG;
1384	if ((argp->flags & NFSMNT_NFSV4))
1385		nmp->nm_sockreq.nr_vers = NFS_VER4;
1386	else if ((argp->flags & NFSMNT_NFSV3))
1387		nmp->nm_sockreq.nr_vers = NFS_VER3;
1388	else
1389		nmp->nm_sockreq.nr_vers = NFS_VER2;
1390
1391
1392	if ((error = newnfs_connect(nmp, &nmp->nm_sockreq, cred, td, 0)))
1393		goto bad;
1394	/* For NFSv4.1, get the clientid now. */
1395	if (nmp->nm_minorvers > 0) {
1396		NFSCL_DEBUG(3, "at getcl\n");
1397		error = nfscl_getcl(mp, cred, td, 0, &clp);
1398		NFSCL_DEBUG(3, "aft getcl=%d\n", error);
1399		if (error != 0)
1400			goto bad;
1401	}
1402
1403	if (nmp->nm_fhsize == 0 && (nmp->nm_flag & NFSMNT_NFSV4) &&
1404	    nmp->nm_dirpathlen > 0) {
1405		NFSCL_DEBUG(3, "in dirp\n");
1406		/*
1407		 * If the fhsize on the mount point == 0 for V4, the mount
1408		 * path needs to be looked up.
1409		 */
1410		trycnt = 3;
1411		do {
1412			error = nfsrpc_getdirpath(nmp, NFSMNT_DIRPATH(nmp),
1413			    cred, td);
1414			NFSCL_DEBUG(3, "aft dirp=%d\n", error);
1415			if (error)
1416				(void) nfs_catnap(PZERO, error, "nfsgetdirp");
1417		} while (error && --trycnt > 0);
1418		if (error) {
1419			error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1420			goto bad;
1421		}
1422	}
1423
1424	/*
1425	 * A reference count is needed on the nfsnode representing the
1426	 * remote root.  If this object is not persistent, then backward
1427	 * traversals of the mount point (i.e. "..") will not work if
1428	 * the nfsnode gets flushed out of the cache. Ufs does not have
1429	 * this problem, because one can identify root inodes by their
1430	 * number == ROOTINO (2).
1431	 */
1432	if (nmp->nm_fhsize > 0) {
1433		/*
1434		 * Set f_iosize to NFS_DIRBLKSIZ so that bo_bsize gets set
1435		 * non-zero for the root vnode. f_iosize will be set correctly
1436		 * by nfs_statfs() before any I/O occurs.
1437		 */
1438		mp->mnt_stat.f_iosize = NFS_DIRBLKSIZ;
1439		error = ncl_nget(mp, nmp->nm_fh, nmp->nm_fhsize, &np,
1440		    LK_EXCLUSIVE);
1441		if (error)
1442			goto bad;
1443		*vpp = NFSTOV(np);
1444
1445		/*
1446		 * Get file attributes and transfer parameters for the
1447		 * mountpoint.  This has the side effect of filling in
1448		 * (*vpp)->v_type with the correct value.
1449		 */
1450		ret = nfsrpc_getattrnovp(nmp, nmp->nm_fh, nmp->nm_fhsize, 1,
1451		    cred, td, &nfsva, NULL, &lease);
1452		if (ret) {
1453			/*
1454			 * Just set default values to get things going.
1455			 */
1456			NFSBZERO((caddr_t)&nfsva, sizeof (struct nfsvattr));
1457			nfsva.na_vattr.va_type = VDIR;
1458			nfsva.na_vattr.va_mode = 0777;
1459			nfsva.na_vattr.va_nlink = 100;
1460			nfsva.na_vattr.va_uid = (uid_t)0;
1461			nfsva.na_vattr.va_gid = (gid_t)0;
1462			nfsva.na_vattr.va_fileid = 2;
1463			nfsva.na_vattr.va_gen = 1;
1464			nfsva.na_vattr.va_blocksize = NFS_FABLKSIZE;
1465			nfsva.na_vattr.va_size = 512 * 1024;
1466			lease = 60;
1467		}
1468		(void) nfscl_loadattrcache(vpp, &nfsva, NULL, NULL, 0, 1);
1469		if (nmp->nm_minorvers > 0) {
1470			NFSCL_DEBUG(3, "lease=%d\n", (int)lease);
1471			NFSLOCKCLSTATE();
1472			clp->nfsc_renew = NFSCL_RENEW(lease);
1473			clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
1474			clp->nfsc_clientidrev++;
1475			if (clp->nfsc_clientidrev == 0)
1476				clp->nfsc_clientidrev++;
1477			NFSUNLOCKCLSTATE();
1478			/*
1479			 * Mount will succeed, so the renew thread can be
1480			 * started now.
1481			 */
1482			nfscl_start_renewthread(clp);
1483			nfscl_clientrelease(clp);
1484		}
1485		if (argp->flags & NFSMNT_NFSV3)
1486			ncl_fsinfo(nmp, *vpp, cred, td);
1487
1488		/* Mark if the mount point supports NFSv4 ACLs. */
1489		if ((argp->flags & NFSMNT_NFSV4) != 0 && nfsrv_useacl != 0 &&
1490		    ret == 0 &&
1491		    NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) {
1492			MNT_ILOCK(mp);
1493			mp->mnt_flag |= MNT_NFS4ACLS;
1494			MNT_IUNLOCK(mp);
1495		}
1496
1497		/*
1498		 * Lose the lock but keep the ref.
1499		 */
1500		NFSVOPUNLOCK(*vpp, 0);
1501		return (0);
1502	}
1503	error = EIO;
1504
1505bad:
1506	if (clp != NULL)
1507		nfscl_clientrelease(clp);
1508	newnfs_disconnect(&nmp->nm_sockreq);
1509	crfree(nmp->nm_sockreq.nr_cred);
1510	if (nmp->nm_sockreq.nr_auth != NULL)
1511		AUTH_DESTROY(nmp->nm_sockreq.nr_auth);
1512	mtx_destroy(&nmp->nm_sockreq.nr_mtx);
1513	mtx_destroy(&nmp->nm_mtx);
1514	if (nmp->nm_clp != NULL) {
1515		NFSLOCKCLSTATE();
1516		LIST_REMOVE(nmp->nm_clp, nfsc_list);
1517		NFSUNLOCKCLSTATE();
1518		free(nmp->nm_clp, M_NFSCLCLIENT);
1519	}
1520	TAILQ_FOREACH_SAFE(dsp, &nmp->nm_sess, nfsclds_list, tdsp)
1521		nfscl_freenfsclds(dsp);
1522	FREE(nmp, M_NEWNFSMNT);
1523	FREE(nam, M_SONAME);
1524	return (error);
1525}
1526
1527/*
1528 * unmount system call
1529 */
1530static int
1531nfs_unmount(struct mount *mp, int mntflags)
1532{
1533	struct thread *td;
1534	struct nfsmount *nmp;
1535	int error, flags = 0, i, trycnt = 0;
1536	struct nfsclds *dsp, *tdsp;
1537
1538	td = curthread;
1539
1540	if (mntflags & MNT_FORCE)
1541		flags |= FORCECLOSE;
1542	nmp = VFSTONFS(mp);
1543	/*
1544	 * Goes something like this..
1545	 * - Call vflush() to clear out vnodes for this filesystem
1546	 * - Close the socket
1547	 * - Free up the data structures
1548	 */
1549	/* In the forced case, cancel any outstanding requests. */
1550	if (mntflags & MNT_FORCE) {
1551		error = newnfs_nmcancelreqs(nmp);
1552		if (error)
1553			goto out;
1554		/* For a forced close, get rid of the renew thread now */
1555		nfscl_umount(nmp, td);
1556	}
1557	/* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */
1558	do {
1559		error = vflush(mp, 1, flags, td);
1560		if ((mntflags & MNT_FORCE) && error != 0 && ++trycnt < 30)
1561			(void) nfs_catnap(PSOCK, error, "newndm");
1562	} while ((mntflags & MNT_FORCE) && error != 0 && trycnt < 30);
1563	if (error)
1564		goto out;
1565
1566	/*
1567	 * We are now committed to the unmount.
1568	 */
1569	if ((mntflags & MNT_FORCE) == 0)
1570		nfscl_umount(nmp, td);
1571	/* Make sure no nfsiods are assigned to this mount. */
1572	mtx_lock(&ncl_iod_mutex);
1573	for (i = 0; i < NFS_MAXASYNCDAEMON; i++)
1574		if (ncl_iodmount[i] == nmp) {
1575			ncl_iodwant[i] = NFSIOD_AVAILABLE;
1576			ncl_iodmount[i] = NULL;
1577		}
1578	mtx_unlock(&ncl_iod_mutex);
1579	newnfs_disconnect(&nmp->nm_sockreq);
1580	crfree(nmp->nm_sockreq.nr_cred);
1581	FREE(nmp->nm_nam, M_SONAME);
1582	if (nmp->nm_sockreq.nr_auth != NULL)
1583		AUTH_DESTROY(nmp->nm_sockreq.nr_auth);
1584	mtx_destroy(&nmp->nm_sockreq.nr_mtx);
1585	mtx_destroy(&nmp->nm_mtx);
1586	TAILQ_FOREACH_SAFE(dsp, &nmp->nm_sess, nfsclds_list, tdsp)
1587		nfscl_freenfsclds(dsp);
1588	FREE(nmp, M_NEWNFSMNT);
1589out:
1590	return (error);
1591}
1592
1593/*
1594 * Return root of a filesystem
1595 */
1596static int
1597nfs_root(struct mount *mp, int flags, struct vnode **vpp)
1598{
1599	struct vnode *vp;
1600	struct nfsmount *nmp;
1601	struct nfsnode *np;
1602	int error;
1603
1604	nmp = VFSTONFS(mp);
1605	error = ncl_nget(mp, nmp->nm_fh, nmp->nm_fhsize, &np, flags);
1606	if (error)
1607		return error;
1608	vp = NFSTOV(np);
1609	/*
1610	 * Get transfer parameters and attributes for root vnode once.
1611	 */
1612	mtx_lock(&nmp->nm_mtx);
1613	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) {
1614		mtx_unlock(&nmp->nm_mtx);
1615		ncl_fsinfo(nmp, vp, curthread->td_ucred, curthread);
1616	} else
1617		mtx_unlock(&nmp->nm_mtx);
1618	if (vp->v_type == VNON)
1619	    vp->v_type = VDIR;
1620	vp->v_vflag |= VV_ROOT;
1621	*vpp = vp;
1622	return (0);
1623}
1624
1625/*
1626 * Flush out the buffer cache
1627 */
1628/* ARGSUSED */
1629static int
1630nfs_sync(struct mount *mp, int waitfor)
1631{
1632	struct vnode *vp, *mvp;
1633	struct thread *td;
1634	int error, allerror = 0;
1635
1636	td = curthread;
1637
1638	MNT_ILOCK(mp);
1639	/*
1640	 * If a forced dismount is in progress, return from here so that
1641	 * the umount(2) syscall doesn't get stuck in VFS_SYNC() before
1642	 * calling VFS_UNMOUNT().
1643	 */
1644	if ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
1645		MNT_IUNLOCK(mp);
1646		return (EBADF);
1647	}
1648	MNT_IUNLOCK(mp);
1649
1650	/*
1651	 * Force stale buffer cache information to be flushed.
1652	 */
1653loop:
1654	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1655		/* XXX Racy bv_cnt check. */
1656		if (NFSVOPISLOCKED(vp) || vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
1657		    waitfor == MNT_LAZY) {
1658			VI_UNLOCK(vp);
1659			continue;
1660		}
1661		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
1662			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1663			goto loop;
1664		}
1665		error = VOP_FSYNC(vp, waitfor, td);
1666		if (error)
1667			allerror = error;
1668		NFSVOPUNLOCK(vp, 0);
1669		vrele(vp);
1670	}
1671	return (allerror);
1672}
1673
1674static int
1675nfs_sysctl(struct mount *mp, fsctlop_t op, struct sysctl_req *req)
1676{
1677	struct nfsmount *nmp = VFSTONFS(mp);
1678	struct vfsquery vq;
1679	int error;
1680
1681	bzero(&vq, sizeof(vq));
1682	switch (op) {
1683#if 0
1684	case VFS_CTL_NOLOCKS:
1685		val = (nmp->nm_flag & NFSMNT_NOLOCKS) ? 1 : 0;
1686 		if (req->oldptr != NULL) {
1687 			error = SYSCTL_OUT(req, &val, sizeof(val));
1688 			if (error)
1689 				return (error);
1690 		}
1691 		if (req->newptr != NULL) {
1692 			error = SYSCTL_IN(req, &val, sizeof(val));
1693 			if (error)
1694 				return (error);
1695			if (val)
1696				nmp->nm_flag |= NFSMNT_NOLOCKS;
1697			else
1698				nmp->nm_flag &= ~NFSMNT_NOLOCKS;
1699 		}
1700		break;
1701#endif
1702	case VFS_CTL_QUERY:
1703		mtx_lock(&nmp->nm_mtx);
1704		if (nmp->nm_state & NFSSTA_TIMEO)
1705			vq.vq_flags |= VQ_NOTRESP;
1706		mtx_unlock(&nmp->nm_mtx);
1707#if 0
1708		if (!(nmp->nm_flag & NFSMNT_NOLOCKS) &&
1709		    (nmp->nm_state & NFSSTA_LOCKTIMEO))
1710			vq.vq_flags |= VQ_NOTRESPLOCK;
1711#endif
1712		error = SYSCTL_OUT(req, &vq, sizeof(vq));
1713		break;
1714 	case VFS_CTL_TIMEO:
1715 		if (req->oldptr != NULL) {
1716 			error = SYSCTL_OUT(req, &nmp->nm_tprintf_initial_delay,
1717 			    sizeof(nmp->nm_tprintf_initial_delay));
1718 			if (error)
1719 				return (error);
1720 		}
1721 		if (req->newptr != NULL) {
1722			error = vfs_suser(mp, req->td);
1723			if (error)
1724				return (error);
1725 			error = SYSCTL_IN(req, &nmp->nm_tprintf_initial_delay,
1726 			    sizeof(nmp->nm_tprintf_initial_delay));
1727 			if (error)
1728 				return (error);
1729 			if (nmp->nm_tprintf_initial_delay < 0)
1730 				nmp->nm_tprintf_initial_delay = 0;
1731 		}
1732		break;
1733	default:
1734		return (ENOTSUP);
1735	}
1736	return (0);
1737}
1738
1739/*
1740 * Purge any RPCs in progress, so that they will all return errors.
1741 * This allows dounmount() to continue as far as VFS_UNMOUNT() for a
1742 * forced dismount.
1743 */
1744static void
1745nfs_purge(struct mount *mp)
1746{
1747	struct nfsmount *nmp = VFSTONFS(mp);
1748
1749	newnfs_nmcancelreqs(nmp);
1750}
1751
1752/*
1753 * Extract the information needed by the nlm from the nfs vnode.
1754 */
1755static void
1756nfs_getnlminfo(struct vnode *vp, uint8_t *fhp, size_t *fhlenp,
1757    struct sockaddr_storage *sp, int *is_v3p, off_t *sizep,
1758    struct timeval *timeop)
1759{
1760	struct nfsmount *nmp;
1761	struct nfsnode *np = VTONFS(vp);
1762
1763	nmp = VFSTONFS(vp->v_mount);
1764	if (fhlenp != NULL)
1765		*fhlenp = (size_t)np->n_fhp->nfh_len;
1766	if (fhp != NULL)
1767		bcopy(np->n_fhp->nfh_fh, fhp, np->n_fhp->nfh_len);
1768	if (sp != NULL)
1769		bcopy(nmp->nm_nam, sp, min(nmp->nm_nam->sa_len, sizeof(*sp)));
1770	if (is_v3p != NULL)
1771		*is_v3p = NFS_ISV3(vp);
1772	if (sizep != NULL)
1773		*sizep = np->n_size;
1774	if (timeop != NULL) {
1775		timeop->tv_sec = nmp->nm_timeo / NFS_HZ;
1776		timeop->tv_usec = (nmp->nm_timeo % NFS_HZ) * (1000000 / NFS_HZ);
1777	}
1778}
1779
1780/*
1781 * This function prints out an option name, based on the conditional
1782 * argument.
1783 */
1784static __inline void nfscl_printopt(struct nfsmount *nmp, int testval,
1785    char *opt, char **buf, size_t *blen)
1786{
1787	int len;
1788
1789	if (testval != 0 && *blen > strlen(opt)) {
1790		len = snprintf(*buf, *blen, "%s", opt);
1791		if (len != strlen(opt))
1792			printf("EEK!!\n");
1793		*buf += len;
1794		*blen -= len;
1795	}
1796}
1797
1798/*
1799 * This function printf out an options integer value.
1800 */
1801static __inline void nfscl_printoptval(struct nfsmount *nmp, int optval,
1802    char *opt, char **buf, size_t *blen)
1803{
1804	int len;
1805
1806	if (*blen > strlen(opt) + 1) {
1807		/* Could result in truncated output string. */
1808		len = snprintf(*buf, *blen, "%s=%d", opt, optval);
1809		if (len < *blen) {
1810			*buf += len;
1811			*blen -= len;
1812		}
1813	}
1814}
1815
1816/*
1817 * Load the option flags and values into the buffer.
1818 */
1819void nfscl_retopts(struct nfsmount *nmp, char *buffer, size_t buflen)
1820{
1821	char *buf;
1822	size_t blen;
1823
1824	buf = buffer;
1825	blen = buflen;
1826	nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NFSV4) != 0, "nfsv4", &buf,
1827	    &blen);
1828	if ((nmp->nm_flag & NFSMNT_NFSV4) != 0) {
1829		nfscl_printoptval(nmp, nmp->nm_minorvers, ",minorversion", &buf,
1830		    &blen);
1831		nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_PNFS) != 0, ",pnfs",
1832		    &buf, &blen);
1833	}
1834	nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NFSV3) != 0, "nfsv3", &buf,
1835	    &blen);
1836	nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0,
1837	    "nfsv2", &buf, &blen);
1838	nfscl_printopt(nmp, nmp->nm_sotype == SOCK_STREAM, ",tcp", &buf, &blen);
1839	nfscl_printopt(nmp, nmp->nm_sotype != SOCK_STREAM, ",udp", &buf, &blen);
1840	nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_RESVPORT) != 0, ",resvport",
1841	    &buf, &blen);
1842	nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NOCONN) != 0, ",noconn",
1843	    &buf, &blen);
1844	nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_SOFT) == 0, ",hard", &buf,
1845	    &blen);
1846	nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_SOFT) != 0, ",soft", &buf,
1847	    &blen);
1848	nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_INT) != 0, ",intr", &buf,
1849	    &blen);
1850	nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NOCTO) == 0, ",cto", &buf,
1851	    &blen);
1852	nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NOCTO) != 0, ",nocto", &buf,
1853	    &blen);
1854	nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NONCONTIGWR) != 0,
1855	    ",noncontigwr", &buf, &blen);
1856	nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_NOLOCKD | NFSMNT_NFSV4)) ==
1857	    0, ",lockd", &buf, &blen);
1858	nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_NOLOCKD | NFSMNT_NFSV4)) ==
1859	    NFSMNT_NOLOCKD, ",nolockd", &buf, &blen);
1860	nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_RDIRPLUS) != 0, ",rdirplus",
1861	    &buf, &blen);
1862	nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_KERB) == 0, ",sec=sys",
1863	    &buf, &blen);
1864	nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_KERB | NFSMNT_INTEGRITY |
1865	    NFSMNT_PRIVACY)) == NFSMNT_KERB, ",sec=krb5", &buf, &blen);
1866	nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_KERB | NFSMNT_INTEGRITY |
1867	    NFSMNT_PRIVACY)) == (NFSMNT_KERB | NFSMNT_INTEGRITY), ",sec=krb5i",
1868	    &buf, &blen);
1869	nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_KERB | NFSMNT_INTEGRITY |
1870	    NFSMNT_PRIVACY)) == (NFSMNT_KERB | NFSMNT_PRIVACY), ",sec=krb5p",
1871	    &buf, &blen);
1872	nfscl_printoptval(nmp, nmp->nm_acdirmin, ",acdirmin", &buf, &blen);
1873	nfscl_printoptval(nmp, nmp->nm_acdirmax, ",acdirmax", &buf, &blen);
1874	nfscl_printoptval(nmp, nmp->nm_acregmin, ",acregmin", &buf, &blen);
1875	nfscl_printoptval(nmp, nmp->nm_acregmax, ",acregmax", &buf, &blen);
1876	nfscl_printoptval(nmp, nmp->nm_nametimeo, ",nametimeo", &buf, &blen);
1877	nfscl_printoptval(nmp, nmp->nm_negnametimeo, ",negnametimeo", &buf,
1878	    &blen);
1879	nfscl_printoptval(nmp, nmp->nm_rsize, ",rsize", &buf, &blen);
1880	nfscl_printoptval(nmp, nmp->nm_wsize, ",wsize", &buf, &blen);
1881	nfscl_printoptval(nmp, nmp->nm_readdirsize, ",readdirsize", &buf,
1882	    &blen);
1883	nfscl_printoptval(nmp, nmp->nm_readahead, ",readahead", &buf, &blen);
1884	nfscl_printoptval(nmp, nmp->nm_wcommitsize, ",wcommitsize", &buf,
1885	    &blen);
1886	nfscl_printoptval(nmp, nmp->nm_timeo, ",timeout", &buf, &blen);
1887	nfscl_printoptval(nmp, nmp->nm_retry, ",retrans", &buf, &blen);
1888}
1889
1890