2 * Copyright (c) 1989, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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.
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
32 * from nfs_vfsops.c 8.12 (Berkeley) 5/20/95
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
39 #include "opt_bootp.h"
40 #include "opt_nfsroot.h"
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/kernel.h>
47 #include <sys/clock.h>
49 #include <sys/limits.h>
51 #include <sys/malloc.h>
53 #include <sys/module.h>
54 #include <sys/mount.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>
64 #include <vm/vm_extern.h>
68 #include <net/route.h>
69 #include <netinet/in.h>
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>
77 FEATURE(nfscl, "NFSv4 client");
79 extern int nfscl_ticks;
80 extern struct timeval nfsboottime;
81 extern struct nfsstats newnfsstats;
82 extern int nfsrv_useacl;
83 extern int nfscl_debuglevel;
84 extern enum nfsiod_state ncl_iodwant[NFS_MAXASYNCDAEMON];
85 extern struct nfsmount *ncl_iodmount[NFS_MAXASYNCDAEMON];
86 extern struct mtx ncl_iod_mutex;
89 MALLOC_DEFINE(M_NEWNFSREQ, "newnfsclient_req", "NFS request header");
90 MALLOC_DEFINE(M_NEWNFSMNT, "newnfsmnt", "NFS mount struct");
92 SYSCTL_DECL(_vfs_nfs);
93 static int nfs_ip_paranoia = 1;
94 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW,
95 &nfs_ip_paranoia, 0, "");
96 static int nfs_tprintf_initial_delay = NFS_TPRINTF_INITIAL_DELAY;
97 SYSCTL_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" */
100 static int nfs_tprintf_delay = NFS_TPRINTF_DELAY;
101 SYSCTL_INT(_vfs_nfs, NFS_TPRINTF_DELAY,
102 downdelayinterval, CTLFLAG_RW, &nfs_tprintf_delay, 0, "");
105 SYSCTL_INT(_vfs_nfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0,
106 "Toggle debug flag");
109 static int nfs_mountroot(struct mount *);
110 static void nfs_sec_name(char *, int *);
111 static void nfs_decode_args(struct mount *mp, struct nfsmount *nmp,
112 struct nfs_args *argp, const char *, struct ucred *,
114 static int mountnfs(struct nfs_args *, struct mount *,
115 struct sockaddr *, char *, u_char *, int, u_char *, int,
116 u_char *, int, struct vnode **, struct ucred *,
117 struct thread *, int, int, int);
118 static void nfs_getnlminfo(struct vnode *, uint8_t *, size_t *,
119 struct sockaddr_storage *, int *, off_t *,
121 static vfs_mount_t nfs_mount;
122 static vfs_cmount_t nfs_cmount;
123 static vfs_unmount_t nfs_unmount;
124 static vfs_root_t nfs_root;
125 static vfs_statfs_t nfs_statfs;
126 static vfs_sync_t nfs_sync;
127 static vfs_sysctl_t nfs_sysctl;
128 static vfs_purge_t nfs_purge;
131 * nfs vfs operations.
133 static struct vfsops nfs_vfsops = {
134 .vfs_init = ncl_init,
135 .vfs_mount = nfs_mount,
136 .vfs_cmount = nfs_cmount,
137 .vfs_root = nfs_root,
138 .vfs_statfs = nfs_statfs,
139 .vfs_sync = nfs_sync,
140 .vfs_uninit = ncl_uninit,
141 .vfs_unmount = nfs_unmount,
142 .vfs_sysctl = nfs_sysctl,
143 .vfs_purge = nfs_purge,
145 VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK | VFCF_SBDRY);
147 /* So that loader and kldload(2) can find us, wherever we are.. */
148 MODULE_VERSION(nfs, 1);
149 MODULE_DEPEND(nfs, nfscommon, 1, 1, 1);
150 MODULE_DEPEND(nfs, krpc, 1, 1, 1);
151 MODULE_DEPEND(nfs, nfssvc, 1, 1, 1);
152 MODULE_DEPEND(nfs, nfslock, 1, 1, 1);
155 * This structure is now defined in sys/nfs/nfs_diskless.c so that it
156 * can be shared by both NFS clients. It is declared here so that it
157 * will be defined for kernels built without NFS_ROOT, although it
158 * isn't used in that case.
160 #if !defined(NFS_ROOT)
161 struct nfs_diskless nfs_diskless = { { { 0 } } };
162 struct nfsv3_diskless nfsv3_diskless = { { { 0 } } };
163 int nfs_diskless_valid = 0;
166 SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD,
167 &nfs_diskless_valid, 0,
168 "Has the diskless struct been filled correctly");
170 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD,
171 nfsv3_diskless.root_hostnam, 0, "Path to nfs root");
173 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD,
174 &nfsv3_diskless.root_saddr, sizeof(nfsv3_diskless.root_saddr),
175 "%Ssockaddr_in", "Diskless root nfs address");
178 void newnfsargs_ntoh(struct nfs_args *);
179 static int nfs_mountdiskless(char *,
180 struct sockaddr_in *, struct nfs_args *,
181 struct thread *, struct vnode **, struct mount *);
182 static void nfs_convert_diskless(void);
183 static void nfs_convert_oargs(struct nfs_args *args,
184 struct onfs_args *oargs);
187 newnfs_iosize(struct nfsmount *nmp)
191 /* First, set the upper limit for iosize */
192 if (nmp->nm_flag & NFSMNT_NFSV4) {
193 maxio = NFS_MAXBSIZE;
194 } else if (nmp->nm_flag & NFSMNT_NFSV3) {
195 if (nmp->nm_sotype == SOCK_DGRAM)
196 maxio = NFS_MAXDGRAMDATA;
198 maxio = NFS_MAXBSIZE;
200 maxio = NFS_V2MAXDATA;
202 if (nmp->nm_rsize > maxio || nmp->nm_rsize == 0)
203 nmp->nm_rsize = maxio;
204 if (nmp->nm_rsize > NFS_MAXBSIZE)
205 nmp->nm_rsize = NFS_MAXBSIZE;
206 if (nmp->nm_readdirsize > maxio || nmp->nm_readdirsize == 0)
207 nmp->nm_readdirsize = maxio;
208 if (nmp->nm_readdirsize > nmp->nm_rsize)
209 nmp->nm_readdirsize = nmp->nm_rsize;
210 if (nmp->nm_wsize > maxio || nmp->nm_wsize == 0)
211 nmp->nm_wsize = maxio;
212 if (nmp->nm_wsize > NFS_MAXBSIZE)
213 nmp->nm_wsize = NFS_MAXBSIZE;
216 * Calculate the size used for io buffers. Use the larger
217 * of the two sizes to minimise nfs requests but make sure
218 * that it is at least one VM page to avoid wasting buffer
219 * space. It must also be at least NFS_DIRBLKSIZ, since
220 * that is the buffer size used for directories.
222 iosize = imax(nmp->nm_rsize, nmp->nm_wsize);
223 iosize = imax(iosize, PAGE_SIZE);
224 iosize = imax(iosize, NFS_DIRBLKSIZ);
225 nmp->nm_mountp->mnt_stat.f_iosize = iosize;
230 nfs_convert_oargs(struct nfs_args *args, struct onfs_args *oargs)
233 args->version = NFS_ARGSVERSION;
234 args->addr = oargs->addr;
235 args->addrlen = oargs->addrlen;
236 args->sotype = oargs->sotype;
237 args->proto = oargs->proto;
238 args->fh = oargs->fh;
239 args->fhsize = oargs->fhsize;
240 args->flags = oargs->flags;
241 args->wsize = oargs->wsize;
242 args->rsize = oargs->rsize;
243 args->readdirsize = oargs->readdirsize;
244 args->timeo = oargs->timeo;
245 args->retrans = oargs->retrans;
246 args->readahead = oargs->readahead;
247 args->hostname = oargs->hostname;
251 nfs_convert_diskless(void)
254 bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif,
255 sizeof(struct ifaliasreq));
256 bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway,
257 sizeof(struct sockaddr_in));
258 nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args);
259 if (nfsv3_diskless.root_args.flags & NFSMNT_NFSV3) {
260 nfsv3_diskless.root_fhsize = NFSX_MYFH;
261 bcopy(nfs_diskless.root_fh, nfsv3_diskless.root_fh, NFSX_MYFH);
263 nfsv3_diskless.root_fhsize = NFSX_V2FH;
264 bcopy(nfs_diskless.root_fh, nfsv3_diskless.root_fh, NFSX_V2FH);
266 bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr,
267 sizeof(struct sockaddr_in));
268 bcopy(nfs_diskless.root_hostnam, nfsv3_diskless.root_hostnam, MNAMELEN);
269 nfsv3_diskless.root_time = nfs_diskless.root_time;
270 bcopy(nfs_diskless.my_hostnam, nfsv3_diskless.my_hostnam,
272 nfs_diskless_valid = 3;
279 nfs_statfs(struct mount *mp, struct statfs *sbp)
283 struct nfsmount *nmp = VFSTONFS(mp);
284 struct nfsvattr nfsva;
287 int error = 0, attrflag, gotfsinfo = 0, ret;
292 error = vfs_busy(mp, MBF_NOWAIT);
295 error = ncl_nget(mp, nmp->nm_fh, nmp->nm_fhsize, &np, LK_EXCLUSIVE);
301 mtx_lock(&nmp->nm_mtx);
302 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) {
303 mtx_unlock(&nmp->nm_mtx);
304 error = nfsrpc_fsinfo(vp, &fs, td->td_ucred, td, &nfsva,
309 mtx_unlock(&nmp->nm_mtx);
311 error = nfsrpc_statfs(vp, &sb, &fs, td->td_ucred, td, &nfsva,
314 NFSCL_DEBUG(2, "statfs=%d\n", error);
316 ret = nfsrpc_getattrnovp(nmp, nmp->nm_fh, nmp->nm_fhsize, 1,
317 td->td_ucred, td, &nfsva, NULL, NULL);
320 * Just set default values to get things going.
322 NFSBZERO((caddr_t)&nfsva, sizeof (struct nfsvattr));
323 nfsva.na_vattr.va_type = VDIR;
324 nfsva.na_vattr.va_mode = 0777;
325 nfsva.na_vattr.va_nlink = 100;
326 nfsva.na_vattr.va_uid = (uid_t)0;
327 nfsva.na_vattr.va_gid = (gid_t)0;
328 nfsva.na_vattr.va_fileid = 2;
329 nfsva.na_vattr.va_gen = 1;
330 nfsva.na_vattr.va_blocksize = NFS_FABLKSIZE;
331 nfsva.na_vattr.va_size = 512 * 1024;
334 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
336 mtx_lock(&nmp->nm_mtx);
337 if (gotfsinfo || (nmp->nm_flag & NFSMNT_NFSV4))
338 nfscl_loadfsinfo(nmp, &fs);
339 nfscl_loadsbinfo(nmp, &sb, sbp);
340 sbp->f_iosize = newnfs_iosize(nmp);
341 mtx_unlock(&nmp->nm_mtx);
342 if (sbp != &mp->mnt_stat) {
343 bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN);
344 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
346 strncpy(&sbp->f_fstypename[0], mp->mnt_vfc->vfc_name, MFSNAMELEN);
347 } else if (NFS_ISV4(vp)) {
348 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
356 * nfs version 3 fsinfo rpc call
359 ncl_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct ucred *cred,
363 struct nfsvattr nfsva;
366 error = nfsrpc_fsinfo(vp, &fs, cred, td, &nfsva, &attrflag, NULL);
369 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
371 mtx_lock(&nmp->nm_mtx);
372 nfscl_loadfsinfo(nmp, &fs);
373 mtx_unlock(&nmp->nm_mtx);
379 * Mount a remote root fs via. nfs. This depends on the info in the
380 * nfs_diskless structure that has been filled in properly by some primary
382 * It goes something like this:
383 * - do enough of "ifconfig" by calling ifioctl() so that the system
384 * can talk to the server
385 * - If nfs_diskless.mygateway is filled in, use that address as
387 * - build the rootfs mount point and call mountnfs() to do the rest.
389 * It is assumed to be safe to read, modify, and write the nfsv3_diskless
390 * structure, as well as other global NFS client variables here, as
391 * nfs_mountroot() will be called once in the boot before any other NFS
392 * client activity occurs.
395 nfs_mountroot(struct mount *mp)
397 struct thread *td = curthread;
398 struct nfsv3_diskless *nd = &nfsv3_diskless;
407 #if defined(BOOTP_NFSROOT) && defined(BOOTP)
408 bootpc_init(); /* use bootp to get nfs_diskless filled in */
409 #elif defined(NFS_ROOT)
410 nfs_setup_diskless();
413 if (nfs_diskless_valid == 0)
415 if (nfs_diskless_valid == 1)
416 nfs_convert_diskless();
419 * XXX splnet, so networks will receive...
424 * Do enough of ifconfig(8) so that the critical net interface can
425 * talk to the server.
427 error = socreate(nd->myif.ifra_addr.sa_family, &so, nd->root_args.sotype, 0,
430 panic("nfs_mountroot: socreate(%04x): %d",
431 nd->myif.ifra_addr.sa_family, error);
433 #if 0 /* XXX Bad idea */
435 * We might not have been told the right interface, so we pass
436 * over the first ten interfaces of the same kind, until we get
437 * one of them configured.
440 for (i = strlen(nd->myif.ifra_name) - 1;
441 nd->myif.ifra_name[i] >= '0' &&
442 nd->myif.ifra_name[i] <= '9';
443 nd->myif.ifra_name[i] ++) {
444 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td);
449 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td);
451 panic("nfs_mountroot: SIOCAIFADDR: %d", error);
452 if ((cp = kern_getenv("boot.netif.mtu")) != NULL) {
453 ir.ifr_mtu = strtol(cp, NULL, 10);
454 bcopy(nd->myif.ifra_name, ir.ifr_name, IFNAMSIZ);
456 error = ifioctl(so, SIOCSIFMTU, (caddr_t)&ir, td);
458 printf("nfs_mountroot: SIOCSIFMTU: %d", error);
463 * If the gateway field is filled in, set it as the default route.
464 * Note that pxeboot will set a default route of 0 if the route
465 * is not set by the DHCP server. Check also for a value of 0
466 * to avoid panicking inappropriately in that situation.
468 if (nd->mygateway.sin_len != 0 &&
469 nd->mygateway.sin_addr.s_addr != 0) {
470 struct sockaddr_in mask, sin;
472 bzero((caddr_t)&mask, sizeof(mask));
474 sin.sin_family = AF_INET;
475 sin.sin_len = sizeof(sin);
476 /* XXX MRT use table 0 for this sort of thing */
477 CURVNET_SET(TD_TO_VNET(td));
478 error = rtrequest_fib(RTM_ADD, (struct sockaddr *)&sin,
479 (struct sockaddr *)&nd->mygateway,
480 (struct sockaddr *)&mask,
481 RTF_UP | RTF_GATEWAY, NULL, RT_DEFAULT_FIB);
484 panic("nfs_mountroot: RTM_ADD: %d", error);
488 * Create the rootfs mount point.
490 nd->root_args.fh = nd->root_fh;
491 nd->root_args.fhsize = nd->root_fhsize;
492 l = ntohl(nd->root_saddr.sin_addr.s_addr);
493 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
494 (l >> 24) & 0xff, (l >> 16) & 0xff,
495 (l >> 8) & 0xff, (l >> 0) & 0xff, nd->root_hostnam);
496 printf("NFS ROOT: %s\n", buf);
497 nd->root_args.hostname = buf;
498 if ((error = nfs_mountdiskless(buf,
499 &nd->root_saddr, &nd->root_args, td, &vp, mp)) != 0) {
504 * This is not really an nfs issue, but it is much easier to
505 * set hostname here and then let the "/etc/rc.xxx" files
506 * mount the right /var based upon its preset value.
508 mtx_lock(&prison0.pr_mtx);
509 strlcpy(prison0.pr_hostname, nd->my_hostnam,
510 sizeof(prison0.pr_hostname));
511 mtx_unlock(&prison0.pr_mtx);
512 inittodr(ntohl(nd->root_time));
517 * Internal version of mount system call for diskless setup.
520 nfs_mountdiskless(char *path,
521 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
522 struct vnode **vpp, struct mount *mp)
524 struct sockaddr *nam;
529 * Find the directory path in "path", which also has the server's
530 * name/ip address in it.
532 dirpath = strchr(path, ':');
534 dirlen = strlen(++dirpath);
537 nam = sodupsockaddr((struct sockaddr *)sin, M_WAITOK);
538 if ((error = mountnfs(args, mp, nam, path, NULL, 0, dirpath, dirlen,
539 NULL, 0, vpp, td->td_ucred, td, NFS_DEFAULT_NAMETIMEO,
540 NFS_DEFAULT_NEGNAMETIMEO, 0)) != 0) {
541 printf("nfs_mountroot: mount %s on /: %d\n", path, error);
548 nfs_sec_name(char *sec, int *flagsp)
550 if (!strcmp(sec, "krb5"))
551 *flagsp |= NFSMNT_KERB;
552 else if (!strcmp(sec, "krb5i"))
553 *flagsp |= (NFSMNT_KERB | NFSMNT_INTEGRITY);
554 else if (!strcmp(sec, "krb5p"))
555 *flagsp |= (NFSMNT_KERB | NFSMNT_PRIVACY);
559 nfs_decode_args(struct mount *mp, struct nfsmount *nmp, struct nfs_args *argp,
560 const char *hostname, struct ucred *cred, struct thread *td)
569 * Set read-only flag if requested; otherwise, clear it if this is
570 * an update. If this is not an update, then either the read-only
571 * flag is already clear, or this is a root mount and it was set
572 * intentionally at some previous point.
574 if (vfs_getopt(mp->mnt_optnew, "ro", NULL, NULL) == 0) {
576 mp->mnt_flag |= MNT_RDONLY;
578 } else if (mp->mnt_flag & MNT_UPDATE) {
580 mp->mnt_flag &= ~MNT_RDONLY;
585 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
586 * no sense in that context. Also, set up appropriate retransmit
587 * and soft timeout behavior.
589 if (argp->sotype == SOCK_STREAM) {
590 nmp->nm_flag &= ~NFSMNT_NOCONN;
591 nmp->nm_timeo = NFS_MAXTIMEO;
592 if ((argp->flags & NFSMNT_NFSV4) != 0)
593 nmp->nm_retry = INT_MAX;
595 nmp->nm_retry = NFS_RETRANS_TCP;
598 /* Also clear RDIRPLUS if NFSv2, it crashes some servers */
599 if ((argp->flags & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0) {
600 argp->flags &= ~NFSMNT_RDIRPLUS;
601 nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
604 /* Re-bind if rsrvd port requested and wasn't on one */
605 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
606 && (argp->flags & NFSMNT_RESVPORT);
607 /* Also re-bind if we're switching to/from a connected UDP socket */
608 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
609 (argp->flags & NFSMNT_NOCONN));
611 /* Update flags atomically. Don't change the lock bits. */
612 nmp->nm_flag = argp->flags | nmp->nm_flag;
615 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
616 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
617 if (nmp->nm_timeo < NFS_MINTIMEO)
618 nmp->nm_timeo = NFS_MINTIMEO;
619 else if (nmp->nm_timeo > NFS_MAXTIMEO)
620 nmp->nm_timeo = NFS_MAXTIMEO;
623 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
624 nmp->nm_retry = argp->retrans;
625 if (nmp->nm_retry > NFS_MAXREXMIT)
626 nmp->nm_retry = NFS_MAXREXMIT;
629 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
630 nmp->nm_wsize = argp->wsize;
632 * Clip at the power of 2 below the size. There is an
633 * issue (not isolated) that causes intermittent page
634 * faults if this is not done.
636 if (nmp->nm_wsize > NFS_FABLKSIZE)
637 nmp->nm_wsize = 1 << (fls(nmp->nm_wsize) - 1);
639 nmp->nm_wsize = NFS_FABLKSIZE;
642 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
643 nmp->nm_rsize = argp->rsize;
645 * Clip at the power of 2 below the size. There is an
646 * issue (not isolated) that causes intermittent page
647 * faults if this is not done.
649 if (nmp->nm_rsize > NFS_FABLKSIZE)
650 nmp->nm_rsize = 1 << (fls(nmp->nm_rsize) - 1);
652 nmp->nm_rsize = NFS_FABLKSIZE;
655 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
656 nmp->nm_readdirsize = argp->readdirsize;
659 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
660 nmp->nm_acregmin = argp->acregmin;
662 nmp->nm_acregmin = NFS_MINATTRTIMO;
663 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
664 nmp->nm_acregmax = argp->acregmax;
666 nmp->nm_acregmax = NFS_MAXATTRTIMO;
667 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
668 nmp->nm_acdirmin = argp->acdirmin;
670 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
671 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
672 nmp->nm_acdirmax = argp->acdirmax;
674 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
675 if (nmp->nm_acdirmin > nmp->nm_acdirmax)
676 nmp->nm_acdirmin = nmp->nm_acdirmax;
677 if (nmp->nm_acregmin > nmp->nm_acregmax)
678 nmp->nm_acregmin = nmp->nm_acregmax;
680 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
681 if (argp->readahead <= NFS_MAXRAHEAD)
682 nmp->nm_readahead = argp->readahead;
684 nmp->nm_readahead = NFS_MAXRAHEAD;
686 if ((argp->flags & NFSMNT_WCOMMITSIZE) && argp->wcommitsize >= 0) {
687 if (argp->wcommitsize < nmp->nm_wsize)
688 nmp->nm_wcommitsize = nmp->nm_wsize;
690 nmp->nm_wcommitsize = argp->wcommitsize;
693 adjsock |= ((nmp->nm_sotype != argp->sotype) ||
694 (nmp->nm_soproto != argp->proto));
696 if (nmp->nm_client != NULL && adjsock) {
697 int haslock = 0, error = 0;
699 if (nmp->nm_sotype == SOCK_STREAM) {
700 error = newnfs_sndlock(&nmp->nm_sockreq.nr_lock);
705 newnfs_disconnect(&nmp->nm_sockreq);
707 newnfs_sndunlock(&nmp->nm_sockreq.nr_lock);
708 nmp->nm_sotype = argp->sotype;
709 nmp->nm_soproto = argp->proto;
710 if (nmp->nm_sotype == SOCK_DGRAM)
711 while (newnfs_connect(nmp, &nmp->nm_sockreq,
713 printf("newnfs_args: retrying connect\n");
714 (void) nfs_catnap(PSOCK, 0, "nfscon");
718 nmp->nm_sotype = argp->sotype;
719 nmp->nm_soproto = argp->proto;
722 if (hostname != NULL) {
723 strlcpy(nmp->nm_hostname, hostname,
724 sizeof(nmp->nm_hostname));
725 p = strchr(nmp->nm_hostname, ':');
731 static const char *nfs_opts[] = { "from", "nfs_args",
732 "noac", "noatime", "noexec", "suiddir", "nosuid", "nosymfollow", "union",
733 "noclusterr", "noclusterw", "multilabel", "acls", "force", "update",
734 "async", "noconn", "nolockd", "conn", "lockd", "intr", "rdirplus",
735 "readdirsize", "soft", "hard", "mntudp", "tcp", "udp", "wsize", "rsize",
736 "retrans", "actimeo", "acregmin", "acregmax", "acdirmin", "acdirmax",
737 "resvport", "readahead", "hostname", "timeo", "timeout", "addr", "fh",
738 "nfsv3", "sec", "principal", "nfsv4", "gssname", "allgssname", "dirpath",
739 "minorversion", "nametimeo", "negnametimeo", "nocto", "noncontigwr",
740 "pnfs", "wcommitsize",
747 * It seems a bit dumb to copyinstr() the host and path here and then
748 * bcopy() them in mountnfs(), but I wanted to detect errors before
749 * doing the sockargs() call because sockargs() allocates an mbuf and
750 * an error after that means that I have to release the mbuf.
754 nfs_mount(struct mount *mp)
756 struct nfs_args args = {
757 .version = NFS_ARGSVERSION,
759 .addrlen = sizeof (struct sockaddr_in),
760 .sotype = SOCK_STREAM,
764 .flags = NFSMNT_RESVPORT,
767 .readdirsize = NFS_READDIRSIZE,
769 .retrans = NFS_RETRANS,
770 .readahead = NFS_DEFRAHEAD,
771 .wcommitsize = 0, /* was: NQ_DEFLEASE */
773 .acregmin = NFS_MINATTRTIMO,
774 .acregmax = NFS_MAXATTRTIMO,
775 .acdirmin = NFS_MINDIRATTRTIMO,
776 .acdirmax = NFS_MAXDIRATTRTIMO,
778 int error = 0, ret, len;
779 struct sockaddr *nam = NULL;
783 u_char nfh[NFSX_FHMAX], krbname[100], dirpath[100], srvkrbname[100];
784 char *cp, *opt, *name, *secname;
785 int nametimeo = NFS_DEFAULT_NAMETIMEO;
786 int negnametimeo = NFS_DEFAULT_NEGNAMETIMEO;
788 int dirlen, has_nfs_args_opt, krbnamelen, srvkrbnamelen;
791 has_nfs_args_opt = 0;
792 if (vfs_filteropt(mp->mnt_optnew, nfs_opts)) {
798 if ((mp->mnt_flag & (MNT_ROOTFS | MNT_UPDATE)) == MNT_ROOTFS) {
799 error = nfs_mountroot(mp);
806 * The old mount_nfs program passed the struct nfs_args
807 * from userspace to kernel. The new mount_nfs program
808 * passes string options via nmount() from userspace to kernel
809 * and we populate the struct nfs_args in the kernel.
811 if (vfs_getopt(mp->mnt_optnew, "nfs_args", NULL, NULL) == 0) {
812 error = vfs_copyopt(mp->mnt_optnew, "nfs_args", &args,
817 if (args.version != NFS_ARGSVERSION) {
818 error = EPROGMISMATCH;
821 has_nfs_args_opt = 1;
824 /* Handle the new style options. */
825 if (vfs_getopt(mp->mnt_optnew, "noac", NULL, NULL) == 0) {
826 args.acdirmin = args.acdirmax =
827 args.acregmin = args.acregmax = 0;
828 args.flags |= NFSMNT_ACDIRMIN | NFSMNT_ACDIRMAX |
829 NFSMNT_ACREGMIN | NFSMNT_ACREGMAX;
831 if (vfs_getopt(mp->mnt_optnew, "noconn", NULL, NULL) == 0)
832 args.flags |= NFSMNT_NOCONN;
833 if (vfs_getopt(mp->mnt_optnew, "conn", NULL, NULL) == 0)
834 args.flags &= ~NFSMNT_NOCONN;
835 if (vfs_getopt(mp->mnt_optnew, "nolockd", NULL, NULL) == 0)
836 args.flags |= NFSMNT_NOLOCKD;
837 if (vfs_getopt(mp->mnt_optnew, "lockd", NULL, NULL) == 0)
838 args.flags &= ~NFSMNT_NOLOCKD;
839 if (vfs_getopt(mp->mnt_optnew, "intr", NULL, NULL) == 0)
840 args.flags |= NFSMNT_INT;
841 if (vfs_getopt(mp->mnt_optnew, "rdirplus", NULL, NULL) == 0)
842 args.flags |= NFSMNT_RDIRPLUS;
843 if (vfs_getopt(mp->mnt_optnew, "resvport", NULL, NULL) == 0)
844 args.flags |= NFSMNT_RESVPORT;
845 if (vfs_getopt(mp->mnt_optnew, "noresvport", NULL, NULL) == 0)
846 args.flags &= ~NFSMNT_RESVPORT;
847 if (vfs_getopt(mp->mnt_optnew, "soft", NULL, NULL) == 0)
848 args.flags |= NFSMNT_SOFT;
849 if (vfs_getopt(mp->mnt_optnew, "hard", NULL, NULL) == 0)
850 args.flags &= ~NFSMNT_SOFT;
851 if (vfs_getopt(mp->mnt_optnew, "mntudp", NULL, NULL) == 0)
852 args.sotype = SOCK_DGRAM;
853 if (vfs_getopt(mp->mnt_optnew, "udp", NULL, NULL) == 0)
854 args.sotype = SOCK_DGRAM;
855 if (vfs_getopt(mp->mnt_optnew, "tcp", NULL, NULL) == 0)
856 args.sotype = SOCK_STREAM;
857 if (vfs_getopt(mp->mnt_optnew, "nfsv3", NULL, NULL) == 0)
858 args.flags |= NFSMNT_NFSV3;
859 if (vfs_getopt(mp->mnt_optnew, "nfsv4", NULL, NULL) == 0) {
860 args.flags |= NFSMNT_NFSV4;
861 args.sotype = SOCK_STREAM;
863 if (vfs_getopt(mp->mnt_optnew, "allgssname", NULL, NULL) == 0)
864 args.flags |= NFSMNT_ALLGSSNAME;
865 if (vfs_getopt(mp->mnt_optnew, "nocto", NULL, NULL) == 0)
866 args.flags |= NFSMNT_NOCTO;
867 if (vfs_getopt(mp->mnt_optnew, "noncontigwr", NULL, NULL) == 0)
868 args.flags |= NFSMNT_NONCONTIGWR;
869 if (vfs_getopt(mp->mnt_optnew, "pnfs", NULL, NULL) == 0)
870 args.flags |= NFSMNT_PNFS;
871 if (vfs_getopt(mp->mnt_optnew, "readdirsize", (void **)&opt, NULL) == 0) {
873 vfs_mount_error(mp, "illegal readdirsize");
877 ret = sscanf(opt, "%d", &args.readdirsize);
878 if (ret != 1 || args.readdirsize <= 0) {
879 vfs_mount_error(mp, "illegal readdirsize: %s",
884 args.flags |= NFSMNT_READDIRSIZE;
886 if (vfs_getopt(mp->mnt_optnew, "readahead", (void **)&opt, NULL) == 0) {
888 vfs_mount_error(mp, "illegal readahead");
892 ret = sscanf(opt, "%d", &args.readahead);
893 if (ret != 1 || args.readahead <= 0) {
894 vfs_mount_error(mp, "illegal readahead: %s",
899 args.flags |= NFSMNT_READAHEAD;
901 if (vfs_getopt(mp->mnt_optnew, "wsize", (void **)&opt, NULL) == 0) {
903 vfs_mount_error(mp, "illegal wsize");
907 ret = sscanf(opt, "%d", &args.wsize);
908 if (ret != 1 || args.wsize <= 0) {
909 vfs_mount_error(mp, "illegal wsize: %s",
914 args.flags |= NFSMNT_WSIZE;
916 if (vfs_getopt(mp->mnt_optnew, "rsize", (void **)&opt, NULL) == 0) {
918 vfs_mount_error(mp, "illegal rsize");
922 ret = sscanf(opt, "%d", &args.rsize);
923 if (ret != 1 || args.rsize <= 0) {
924 vfs_mount_error(mp, "illegal wsize: %s",
929 args.flags |= NFSMNT_RSIZE;
931 if (vfs_getopt(mp->mnt_optnew, "retrans", (void **)&opt, NULL) == 0) {
933 vfs_mount_error(mp, "illegal retrans");
937 ret = sscanf(opt, "%d", &args.retrans);
938 if (ret != 1 || args.retrans <= 0) {
939 vfs_mount_error(mp, "illegal retrans: %s",
944 args.flags |= NFSMNT_RETRANS;
946 if (vfs_getopt(mp->mnt_optnew, "actimeo", (void **)&opt, NULL) == 0) {
947 ret = sscanf(opt, "%d", &args.acregmin);
948 if (ret != 1 || args.acregmin < 0) {
949 vfs_mount_error(mp, "illegal actimeo: %s",
954 args.acdirmin = args.acdirmax = args.acregmax = args.acregmin;
955 args.flags |= NFSMNT_ACDIRMIN | NFSMNT_ACDIRMAX |
956 NFSMNT_ACREGMIN | NFSMNT_ACREGMAX;
958 if (vfs_getopt(mp->mnt_optnew, "acregmin", (void **)&opt, NULL) == 0) {
959 ret = sscanf(opt, "%d", &args.acregmin);
960 if (ret != 1 || args.acregmin < 0) {
961 vfs_mount_error(mp, "illegal acregmin: %s",
966 args.flags |= NFSMNT_ACREGMIN;
968 if (vfs_getopt(mp->mnt_optnew, "acregmax", (void **)&opt, NULL) == 0) {
969 ret = sscanf(opt, "%d", &args.acregmax);
970 if (ret != 1 || args.acregmax < 0) {
971 vfs_mount_error(mp, "illegal acregmax: %s",
976 args.flags |= NFSMNT_ACREGMAX;
978 if (vfs_getopt(mp->mnt_optnew, "acdirmin", (void **)&opt, NULL) == 0) {
979 ret = sscanf(opt, "%d", &args.acdirmin);
980 if (ret != 1 || args.acdirmin < 0) {
981 vfs_mount_error(mp, "illegal acdirmin: %s",
986 args.flags |= NFSMNT_ACDIRMIN;
988 if (vfs_getopt(mp->mnt_optnew, "acdirmax", (void **)&opt, NULL) == 0) {
989 ret = sscanf(opt, "%d", &args.acdirmax);
990 if (ret != 1 || args.acdirmax < 0) {
991 vfs_mount_error(mp, "illegal acdirmax: %s",
996 args.flags |= NFSMNT_ACDIRMAX;
998 if (vfs_getopt(mp->mnt_optnew, "wcommitsize", (void **)&opt, NULL) == 0) {
999 ret = sscanf(opt, "%d", &args.wcommitsize);
1000 if (ret != 1 || args.wcommitsize < 0) {
1001 vfs_mount_error(mp, "illegal wcommitsize: %s", opt);
1005 args.flags |= NFSMNT_WCOMMITSIZE;
1007 if (vfs_getopt(mp->mnt_optnew, "timeo", (void **)&opt, NULL) == 0) {
1008 ret = sscanf(opt, "%d", &args.timeo);
1009 if (ret != 1 || args.timeo <= 0) {
1010 vfs_mount_error(mp, "illegal timeo: %s",
1015 args.flags |= NFSMNT_TIMEO;
1017 if (vfs_getopt(mp->mnt_optnew, "timeout", (void **)&opt, NULL) == 0) {
1018 ret = sscanf(opt, "%d", &args.timeo);
1019 if (ret != 1 || args.timeo <= 0) {
1020 vfs_mount_error(mp, "illegal timeout: %s",
1025 args.flags |= NFSMNT_TIMEO;
1027 if (vfs_getopt(mp->mnt_optnew, "nametimeo", (void **)&opt, NULL) == 0) {
1028 ret = sscanf(opt, "%d", &nametimeo);
1029 if (ret != 1 || nametimeo < 0) {
1030 vfs_mount_error(mp, "illegal nametimeo: %s", opt);
1035 if (vfs_getopt(mp->mnt_optnew, "negnametimeo", (void **)&opt, NULL)
1037 ret = sscanf(opt, "%d", &negnametimeo);
1038 if (ret != 1 || negnametimeo < 0) {
1039 vfs_mount_error(mp, "illegal negnametimeo: %s",
1045 if (vfs_getopt(mp->mnt_optnew, "minorversion", (void **)&opt, NULL) ==
1047 ret = sscanf(opt, "%d", &minvers);
1048 if (ret != 1 || minvers < 0 || minvers > 1 ||
1049 (args.flags & NFSMNT_NFSV4) == 0) {
1050 vfs_mount_error(mp, "illegal minorversion: %s", opt);
1055 if (vfs_getopt(mp->mnt_optnew, "sec",
1056 (void **) &secname, NULL) == 0)
1057 nfs_sec_name(secname, &args.flags);
1059 if (mp->mnt_flag & MNT_UPDATE) {
1060 struct nfsmount *nmp = VFSTONFS(mp);
1068 * If a change from TCP->UDP is done and there are thread(s)
1069 * that have I/O RPC(s) in progress with a tranfer size
1070 * greater than NFS_MAXDGRAMDATA, those thread(s) will be
1071 * hung, retrying the RPC(s) forever. Usually these threads
1072 * will be seen doing an uninterruptible sleep on wait channel
1075 if (args.sotype == SOCK_DGRAM && nmp->nm_sotype == SOCK_STREAM)
1076 tprintf(td->td_proc, LOG_WARNING,
1077 "Warning: mount -u that changes TCP->UDP can result in hung threads\n");
1080 * When doing an update, we can't change version,
1081 * security, switch lockd strategies or change cookie
1084 args.flags = (args.flags &
1090 NFSMNT_NOLOCKD /*|NFSMNT_XLATECOOKIE*/)) |
1097 NFSMNT_NOLOCKD /*|NFSMNT_XLATECOOKIE*/));
1098 nfs_decode_args(mp, nmp, &args, NULL, td->td_ucred, td);
1103 * Make the nfs_ip_paranoia sysctl serve as the default connection
1104 * or no-connection mode for those protocols that support
1105 * no-connection mode (the flag will be cleared later for protocols
1106 * that do not support no-connection mode). This will allow a client
1107 * to receive replies from a different IP then the request was
1108 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid),
1111 if (nfs_ip_paranoia == 0)
1112 args.flags |= NFSMNT_NOCONN;
1114 if (has_nfs_args_opt != 0) {
1116 * In the 'nfs_args' case, the pointers in the args
1117 * structure are in userland - we copy them in here.
1119 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX) {
1120 vfs_mount_error(mp, "Bad file handle");
1124 error = copyin((caddr_t)args.fh, (caddr_t)nfh,
1128 error = copyinstr(args.hostname, hst, MNAMELEN - 1, &hstlen);
1131 bzero(&hst[hstlen], MNAMELEN - hstlen);
1132 args.hostname = hst;
1133 /* sockargs() call must be after above copyin() calls */
1134 error = getsockaddr(&nam, (caddr_t)args.addr,
1139 if (vfs_getopt(mp->mnt_optnew, "fh", (void **)&args.fh,
1140 &args.fhsize) == 0) {
1141 if (args.fhsize < 0 || args.fhsize > NFSX_FHMAX) {
1142 vfs_mount_error(mp, "Bad file handle");
1146 bcopy(args.fh, nfh, args.fhsize);
1150 (void) vfs_getopt(mp->mnt_optnew, "hostname",
1151 (void **)&args.hostname, &len);
1152 if (args.hostname == NULL) {
1153 vfs_mount_error(mp, "Invalid hostname");
1157 bcopy(args.hostname, hst, MNAMELEN);
1158 hst[MNAMELEN - 1] = '\0';
1161 if (vfs_getopt(mp->mnt_optnew, "principal", (void **)&name, NULL) == 0)
1162 strlcpy(srvkrbname, name, sizeof (srvkrbname));
1164 snprintf(srvkrbname, sizeof (srvkrbname), "nfs@%s", hst);
1165 cp = strchr(srvkrbname, ':');
1169 srvkrbnamelen = strlen(srvkrbname);
1171 if (vfs_getopt(mp->mnt_optnew, "gssname", (void **)&name, NULL) == 0)
1172 strlcpy(krbname, name, sizeof (krbname));
1175 krbnamelen = strlen(krbname);
1177 if (vfs_getopt(mp->mnt_optnew, "dirpath", (void **)&name, NULL) == 0)
1178 strlcpy(dirpath, name, sizeof (dirpath));
1181 dirlen = strlen(dirpath);
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;
1190 nam = malloc(args.addrlen, M_SONAME, M_WAITOK);
1191 bcopy(args.addr, nam, args.addrlen);
1192 nam->sa_len = args.addrlen;
1194 vfs_mount_error(mp, "No server address");
1201 error = mountnfs(&args, mp, nam, hst, krbname, krbnamelen, dirpath,
1202 dirlen, srvkrbname, srvkrbnamelen, &vp, td->td_ucred, td,
1203 nametimeo, negnametimeo, minvers);
1207 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_NO_IOPF |
1219 * It seems a bit dumb to copyinstr() the host and path here and then
1220 * bcopy() them in mountnfs(), but I wanted to detect errors before
1221 * doing the sockargs() call because sockargs() allocates an mbuf and
1222 * an error after that means that I have to release the mbuf.
1226 nfs_cmount(struct mntarg *ma, void *data, uint64_t flags)
1229 struct nfs_args args;
1231 error = copyin(data, &args, sizeof (struct nfs_args));
1235 ma = mount_arg(ma, "nfs_args", &args, sizeof args);
1237 error = kernel_mount(ma, flags);
1242 * Common code for mount and mountroot
1245 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
1246 char *hst, u_char *krbname, int krbnamelen, u_char *dirpath, int dirlen,
1247 u_char *srvkrbname, int srvkrbnamelen, struct vnode **vpp,
1248 struct ucred *cred, struct thread *td, int nametimeo, int negnametimeo,
1251 struct nfsmount *nmp;
1253 int error, trycnt, ret;
1254 struct nfsvattr nfsva;
1255 struct nfsclclient *clp;
1256 struct nfsclds *dsp, *tdsp;
1258 static u_int64_t clval = 0;
1260 NFSCL_DEBUG(3, "in mnt\n");
1262 if (mp->mnt_flag & MNT_UPDATE) {
1264 printf("%s: MNT_UPDATE is no longer handled here\n", __func__);
1265 FREE(nam, M_SONAME);
1268 MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount) +
1269 krbnamelen + dirlen + srvkrbnamelen + 2,
1270 M_NEWNFSMNT, M_WAITOK | M_ZERO);
1271 TAILQ_INIT(&nmp->nm_bufq);
1273 clval = (u_int64_t)nfsboottime.tv_sec;
1274 nmp->nm_clval = clval++;
1275 nmp->nm_krbnamelen = krbnamelen;
1276 nmp->nm_dirpathlen = dirlen;
1277 nmp->nm_srvkrbnamelen = srvkrbnamelen;
1278 if (td->td_ucred->cr_uid != (uid_t)0) {
1280 * nm_uid is used to get KerberosV credentials for
1281 * the nfsv4 state handling operations if there is
1282 * no host based principal set. Use the uid of
1283 * this user if not root, since they are doing the
1284 * mount. I don't think setting this for root will
1285 * work, since root normally does not have user
1286 * credentials in a credentials cache.
1288 nmp->nm_uid = td->td_ucred->cr_uid;
1291 * Just set to -1, so it won't be used.
1293 nmp->nm_uid = (uid_t)-1;
1296 /* Copy and null terminate all the names */
1297 if (nmp->nm_krbnamelen > 0) {
1298 bcopy(krbname, nmp->nm_krbname, nmp->nm_krbnamelen);
1299 nmp->nm_name[nmp->nm_krbnamelen] = '\0';
1301 if (nmp->nm_dirpathlen > 0) {
1302 bcopy(dirpath, NFSMNT_DIRPATH(nmp),
1303 nmp->nm_dirpathlen);
1304 nmp->nm_name[nmp->nm_krbnamelen + nmp->nm_dirpathlen
1307 if (nmp->nm_srvkrbnamelen > 0) {
1308 bcopy(srvkrbname, NFSMNT_SRVKRBNAME(nmp),
1309 nmp->nm_srvkrbnamelen);
1310 nmp->nm_name[nmp->nm_krbnamelen + nmp->nm_dirpathlen
1311 + nmp->nm_srvkrbnamelen + 2] = '\0';
1313 nmp->nm_sockreq.nr_cred = crhold(cred);
1314 mtx_init(&nmp->nm_sockreq.nr_mtx, "nfssock", NULL, MTX_DEF);
1316 nmp->nm_getinfo = nfs_getnlminfo;
1317 nmp->nm_vinvalbuf = ncl_vinvalbuf;
1320 nmp->nm_mountp = mp;
1321 mtx_init(&nmp->nm_mtx, "NFSmount lock", NULL, MTX_DEF | MTX_DUPOK);
1324 * Since nfs_decode_args() might optionally set them, these
1325 * need to be set to defaults before the call, so that the
1326 * optional settings aren't overwritten.
1328 nmp->nm_nametimeo = nametimeo;
1329 nmp->nm_negnametimeo = negnametimeo;
1330 nmp->nm_timeo = NFS_TIMEO;
1331 nmp->nm_retry = NFS_RETRANS;
1332 nmp->nm_readahead = NFS_DEFRAHEAD;
1334 /* This is empirical approximation of sqrt(hibufspace) * 256. */
1335 nmp->nm_wcommitsize = NFS_MAXBSIZE / 256;
1336 while ((long)nmp->nm_wcommitsize * nmp->nm_wcommitsize < hibufspace)
1337 nmp->nm_wcommitsize *= 2;
1338 nmp->nm_wcommitsize *= 256;
1340 if ((argp->flags & NFSMNT_NFSV4) != 0)
1341 nmp->nm_minorvers = minvers;
1343 nmp->nm_minorvers = 0;
1345 nfs_decode_args(mp, nmp, argp, hst, cred, td);
1348 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too
1349 * high, depending on whether we end up with negative offsets in
1350 * the client or server somewhere. 2GB-1 may be safer.
1352 * For V3, ncl_fsinfo will adjust this as necessary. Assume maximum
1353 * that we can handle until we find out otherwise.
1355 if ((argp->flags & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0)
1356 nmp->nm_maxfilesize = 0xffffffffLL;
1358 nmp->nm_maxfilesize = OFF_MAX;
1360 if ((argp->flags & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0) {
1361 nmp->nm_wsize = NFS_WSIZE;
1362 nmp->nm_rsize = NFS_RSIZE;
1363 nmp->nm_readdirsize = NFS_READDIRSIZE;
1365 nmp->nm_numgrps = NFS_MAXGRPS;
1366 nmp->nm_tprintf_delay = nfs_tprintf_delay;
1367 if (nmp->nm_tprintf_delay < 0)
1368 nmp->nm_tprintf_delay = 0;
1369 nmp->nm_tprintf_initial_delay = nfs_tprintf_initial_delay;
1370 if (nmp->nm_tprintf_initial_delay < 0)
1371 nmp->nm_tprintf_initial_delay = 0;
1372 nmp->nm_fhsize = argp->fhsize;
1373 if (nmp->nm_fhsize > 0)
1374 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
1375 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
1377 /* Set up the sockets and per-host congestion */
1378 nmp->nm_sotype = argp->sotype;
1379 nmp->nm_soproto = argp->proto;
1380 nmp->nm_sockreq.nr_prog = NFS_PROG;
1381 if ((argp->flags & NFSMNT_NFSV4))
1382 nmp->nm_sockreq.nr_vers = NFS_VER4;
1383 else if ((argp->flags & NFSMNT_NFSV3))
1384 nmp->nm_sockreq.nr_vers = NFS_VER3;
1386 nmp->nm_sockreq.nr_vers = NFS_VER2;
1389 if ((error = newnfs_connect(nmp, &nmp->nm_sockreq, cred, td, 0)))
1391 /* For NFSv4.1, get the clientid now. */
1392 if (nmp->nm_minorvers > 0) {
1393 NFSCL_DEBUG(3, "at getcl\n");
1394 error = nfscl_getcl(mp, cred, td, 0, &clp);
1395 NFSCL_DEBUG(3, "aft getcl=%d\n", error);
1400 if (nmp->nm_fhsize == 0 && (nmp->nm_flag & NFSMNT_NFSV4) &&
1401 nmp->nm_dirpathlen > 0) {
1402 NFSCL_DEBUG(3, "in dirp\n");
1404 * If the fhsize on the mount point == 0 for V4, the mount
1405 * path needs to be looked up.
1409 error = nfsrpc_getdirpath(nmp, NFSMNT_DIRPATH(nmp),
1411 NFSCL_DEBUG(3, "aft dirp=%d\n", error);
1413 (void) nfs_catnap(PZERO, error, "nfsgetdirp");
1414 } while (error && --trycnt > 0);
1416 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1422 * A reference count is needed on the nfsnode representing the
1423 * remote root. If this object is not persistent, then backward
1424 * traversals of the mount point (i.e. "..") will not work if
1425 * the nfsnode gets flushed out of the cache. Ufs does not have
1426 * this problem, because one can identify root inodes by their
1427 * number == ROOTINO (2).
1429 if (nmp->nm_fhsize > 0) {
1431 * Set f_iosize to NFS_DIRBLKSIZ so that bo_bsize gets set
1432 * non-zero for the root vnode. f_iosize will be set correctly
1433 * by nfs_statfs() before any I/O occurs.
1435 mp->mnt_stat.f_iosize = NFS_DIRBLKSIZ;
1436 error = ncl_nget(mp, nmp->nm_fh, nmp->nm_fhsize, &np,
1443 * Get file attributes and transfer parameters for the
1444 * mountpoint. This has the side effect of filling in
1445 * (*vpp)->v_type with the correct value.
1447 ret = nfsrpc_getattrnovp(nmp, nmp->nm_fh, nmp->nm_fhsize, 1,
1448 cred, td, &nfsva, NULL, &lease);
1451 * Just set default values to get things going.
1453 NFSBZERO((caddr_t)&nfsva, sizeof (struct nfsvattr));
1454 nfsva.na_vattr.va_type = VDIR;
1455 nfsva.na_vattr.va_mode = 0777;
1456 nfsva.na_vattr.va_nlink = 100;
1457 nfsva.na_vattr.va_uid = (uid_t)0;
1458 nfsva.na_vattr.va_gid = (gid_t)0;
1459 nfsva.na_vattr.va_fileid = 2;
1460 nfsva.na_vattr.va_gen = 1;
1461 nfsva.na_vattr.va_blocksize = NFS_FABLKSIZE;
1462 nfsva.na_vattr.va_size = 512 * 1024;
1465 (void) nfscl_loadattrcache(vpp, &nfsva, NULL, NULL, 0, 1);
1466 if (nmp->nm_minorvers > 0) {
1467 NFSCL_DEBUG(3, "lease=%d\n", (int)lease);
1469 clp->nfsc_renew = NFSCL_RENEW(lease);
1470 clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
1471 clp->nfsc_clientidrev++;
1472 if (clp->nfsc_clientidrev == 0)
1473 clp->nfsc_clientidrev++;
1476 * Mount will succeed, so the renew thread can be
1479 nfscl_start_renewthread(clp);
1480 nfscl_clientrelease(clp);
1482 if (argp->flags & NFSMNT_NFSV3)
1483 ncl_fsinfo(nmp, *vpp, cred, td);
1485 /* Mark if the mount point supports NFSv4 ACLs. */
1486 if ((argp->flags & NFSMNT_NFSV4) != 0 && nfsrv_useacl != 0 &&
1488 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) {
1490 mp->mnt_flag |= MNT_NFS4ACLS;
1495 * Lose the lock but keep the ref.
1497 NFSVOPUNLOCK(*vpp, 0);
1504 nfscl_clientrelease(clp);
1505 newnfs_disconnect(&nmp->nm_sockreq);
1506 crfree(nmp->nm_sockreq.nr_cred);
1507 if (nmp->nm_sockreq.nr_auth != NULL)
1508 AUTH_DESTROY(nmp->nm_sockreq.nr_auth);
1509 mtx_destroy(&nmp->nm_sockreq.nr_mtx);
1510 mtx_destroy(&nmp->nm_mtx);
1511 if (nmp->nm_clp != NULL) {
1513 LIST_REMOVE(nmp->nm_clp, nfsc_list);
1515 free(nmp->nm_clp, M_NFSCLCLIENT);
1517 TAILQ_FOREACH_SAFE(dsp, &nmp->nm_sess, nfsclds_list, tdsp)
1518 nfscl_freenfsclds(dsp);
1519 FREE(nmp, M_NEWNFSMNT);
1520 FREE(nam, M_SONAME);
1525 * unmount system call
1528 nfs_unmount(struct mount *mp, int mntflags)
1531 struct nfsmount *nmp;
1532 int error, flags = 0, i, trycnt = 0;
1533 struct nfsclds *dsp, *tdsp;
1537 if (mntflags & MNT_FORCE)
1538 flags |= FORCECLOSE;
1541 * Goes something like this..
1542 * - Call vflush() to clear out vnodes for this filesystem
1543 * - Close the socket
1544 * - Free up the data structures
1546 /* In the forced case, cancel any outstanding requests. */
1547 if (mntflags & MNT_FORCE) {
1548 error = newnfs_nmcancelreqs(nmp);
1551 /* For a forced close, get rid of the renew thread now */
1552 nfscl_umount(nmp, td);
1554 /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */
1556 error = vflush(mp, 1, flags, td);
1557 if ((mntflags & MNT_FORCE) && error != 0 && ++trycnt < 30)
1558 (void) nfs_catnap(PSOCK, error, "newndm");
1559 } while ((mntflags & MNT_FORCE) && error != 0 && trycnt < 30);
1564 * We are now committed to the unmount.
1566 if ((mntflags & MNT_FORCE) == 0)
1567 nfscl_umount(nmp, td);
1568 /* Make sure no nfsiods are assigned to this mount. */
1569 mtx_lock(&ncl_iod_mutex);
1570 for (i = 0; i < NFS_MAXASYNCDAEMON; i++)
1571 if (ncl_iodmount[i] == nmp) {
1572 ncl_iodwant[i] = NFSIOD_AVAILABLE;
1573 ncl_iodmount[i] = NULL;
1575 mtx_unlock(&ncl_iod_mutex);
1576 newnfs_disconnect(&nmp->nm_sockreq);
1577 crfree(nmp->nm_sockreq.nr_cred);
1578 FREE(nmp->nm_nam, M_SONAME);
1579 if (nmp->nm_sockreq.nr_auth != NULL)
1580 AUTH_DESTROY(nmp->nm_sockreq.nr_auth);
1581 mtx_destroy(&nmp->nm_sockreq.nr_mtx);
1582 mtx_destroy(&nmp->nm_mtx);
1583 TAILQ_FOREACH_SAFE(dsp, &nmp->nm_sess, nfsclds_list, tdsp)
1584 nfscl_freenfsclds(dsp);
1585 FREE(nmp, M_NEWNFSMNT);
1591 * Return root of a filesystem
1594 nfs_root(struct mount *mp, int flags, struct vnode **vpp)
1597 struct nfsmount *nmp;
1602 error = ncl_nget(mp, nmp->nm_fh, nmp->nm_fhsize, &np, flags);
1607 * Get transfer parameters and attributes for root vnode once.
1609 mtx_lock(&nmp->nm_mtx);
1610 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) {
1611 mtx_unlock(&nmp->nm_mtx);
1612 ncl_fsinfo(nmp, vp, curthread->td_ucred, curthread);
1614 mtx_unlock(&nmp->nm_mtx);
1615 if (vp->v_type == VNON)
1617 vp->v_vflag |= VV_ROOT;
1623 * Flush out the buffer cache
1627 nfs_sync(struct mount *mp, int waitfor)
1629 struct vnode *vp, *mvp;
1631 int error, allerror = 0;
1637 * If a forced dismount is in progress, return from here so that
1638 * the umount(2) syscall doesn't get stuck in VFS_SYNC() before
1639 * calling VFS_UNMOUNT().
1641 if ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
1648 * Force stale buffer cache information to be flushed.
1651 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1652 /* XXX Racy bv_cnt check. */
1653 if (NFSVOPISLOCKED(vp) || vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
1654 waitfor == MNT_LAZY) {
1658 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
1659 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1662 error = VOP_FSYNC(vp, waitfor, td);
1665 NFSVOPUNLOCK(vp, 0);
1672 nfs_sysctl(struct mount *mp, fsctlop_t op, struct sysctl_req *req)
1674 struct nfsmount *nmp = VFSTONFS(mp);
1678 bzero(&vq, sizeof(vq));
1681 case VFS_CTL_NOLOCKS:
1682 val = (nmp->nm_flag & NFSMNT_NOLOCKS) ? 1 : 0;
1683 if (req->oldptr != NULL) {
1684 error = SYSCTL_OUT(req, &val, sizeof(val));
1688 if (req->newptr != NULL) {
1689 error = SYSCTL_IN(req, &val, sizeof(val));
1693 nmp->nm_flag |= NFSMNT_NOLOCKS;
1695 nmp->nm_flag &= ~NFSMNT_NOLOCKS;
1700 mtx_lock(&nmp->nm_mtx);
1701 if (nmp->nm_state & NFSSTA_TIMEO)
1702 vq.vq_flags |= VQ_NOTRESP;
1703 mtx_unlock(&nmp->nm_mtx);
1705 if (!(nmp->nm_flag & NFSMNT_NOLOCKS) &&
1706 (nmp->nm_state & NFSSTA_LOCKTIMEO))
1707 vq.vq_flags |= VQ_NOTRESPLOCK;
1709 error = SYSCTL_OUT(req, &vq, sizeof(vq));
1712 if (req->oldptr != NULL) {
1713 error = SYSCTL_OUT(req, &nmp->nm_tprintf_initial_delay,
1714 sizeof(nmp->nm_tprintf_initial_delay));
1718 if (req->newptr != NULL) {
1719 error = vfs_suser(mp, req->td);
1722 error = SYSCTL_IN(req, &nmp->nm_tprintf_initial_delay,
1723 sizeof(nmp->nm_tprintf_initial_delay));
1726 if (nmp->nm_tprintf_initial_delay < 0)
1727 nmp->nm_tprintf_initial_delay = 0;
1737 * Purge any RPCs in progress, so that they will all return errors.
1738 * This allows dounmount() to continue as far as VFS_UNMOUNT() for a
1742 nfs_purge(struct mount *mp)
1744 struct nfsmount *nmp = VFSTONFS(mp);
1746 newnfs_nmcancelreqs(nmp);
1750 * Extract the information needed by the nlm from the nfs vnode.
1753 nfs_getnlminfo(struct vnode *vp, uint8_t *fhp, size_t *fhlenp,
1754 struct sockaddr_storage *sp, int *is_v3p, off_t *sizep,
1755 struct timeval *timeop)
1757 struct nfsmount *nmp;
1758 struct nfsnode *np = VTONFS(vp);
1760 nmp = VFSTONFS(vp->v_mount);
1762 *fhlenp = (size_t)np->n_fhp->nfh_len;
1764 bcopy(np->n_fhp->nfh_fh, fhp, np->n_fhp->nfh_len);
1766 bcopy(nmp->nm_nam, sp, min(nmp->nm_nam->sa_len, sizeof(*sp)));
1768 *is_v3p = NFS_ISV3(vp);
1770 *sizep = np->n_size;
1771 if (timeop != NULL) {
1772 timeop->tv_sec = nmp->nm_timeo / NFS_HZ;
1773 timeop->tv_usec = (nmp->nm_timeo % NFS_HZ) * (1000000 / NFS_HZ);
1778 * This function prints out an option name, based on the conditional
1781 static __inline void nfscl_printopt(struct nfsmount *nmp, int testval,
1782 char *opt, char **buf, size_t *blen)
1786 if (testval != 0 && *blen > strlen(opt)) {
1787 len = snprintf(*buf, *blen, "%s", opt);
1788 if (len != strlen(opt))
1796 * This function printf out an options integer value.
1798 static __inline void nfscl_printoptval(struct nfsmount *nmp, int optval,
1799 char *opt, char **buf, size_t *blen)
1803 if (*blen > strlen(opt) + 1) {
1804 /* Could result in truncated output string. */
1805 len = snprintf(*buf, *blen, "%s=%d", opt, optval);
1814 * Load the option flags and values into the buffer.
1816 void nfscl_retopts(struct nfsmount *nmp, char *buffer, size_t buflen)
1823 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NFSV4) != 0, "nfsv4", &buf,
1825 if ((nmp->nm_flag & NFSMNT_NFSV4) != 0) {
1826 nfscl_printoptval(nmp, nmp->nm_minorvers, ",minorversion", &buf,
1828 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_PNFS) != 0, ",pnfs",
1831 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NFSV3) != 0, "nfsv3", &buf,
1833 nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0,
1834 "nfsv2", &buf, &blen);
1835 nfscl_printopt(nmp, nmp->nm_sotype == SOCK_STREAM, ",tcp", &buf, &blen);
1836 nfscl_printopt(nmp, nmp->nm_sotype != SOCK_STREAM, ",udp", &buf, &blen);
1837 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_RESVPORT) != 0, ",resvport",
1839 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NOCONN) != 0, ",noconn",
1841 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_SOFT) == 0, ",hard", &buf,
1843 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_SOFT) != 0, ",soft", &buf,
1845 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_INT) != 0, ",intr", &buf,
1847 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NOCTO) == 0, ",cto", &buf,
1849 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NOCTO) != 0, ",nocto", &buf,
1851 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NONCONTIGWR) != 0,
1852 ",noncontigwr", &buf, &blen);
1853 nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_NOLOCKD | NFSMNT_NFSV4)) ==
1854 0, ",lockd", &buf, &blen);
1855 nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_NOLOCKD | NFSMNT_NFSV4)) ==
1856 NFSMNT_NOLOCKD, ",nolockd", &buf, &blen);
1857 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_RDIRPLUS) != 0, ",rdirplus",
1859 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_KERB) == 0, ",sec=sys",
1861 nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_KERB | NFSMNT_INTEGRITY |
1862 NFSMNT_PRIVACY)) == NFSMNT_KERB, ",sec=krb5", &buf, &blen);
1863 nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_KERB | NFSMNT_INTEGRITY |
1864 NFSMNT_PRIVACY)) == (NFSMNT_KERB | NFSMNT_INTEGRITY), ",sec=krb5i",
1866 nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_KERB | NFSMNT_INTEGRITY |
1867 NFSMNT_PRIVACY)) == (NFSMNT_KERB | NFSMNT_PRIVACY), ",sec=krb5p",
1869 nfscl_printoptval(nmp, nmp->nm_acdirmin, ",acdirmin", &buf, &blen);
1870 nfscl_printoptval(nmp, nmp->nm_acdirmax, ",acdirmax", &buf, &blen);
1871 nfscl_printoptval(nmp, nmp->nm_acregmin, ",acregmin", &buf, &blen);
1872 nfscl_printoptval(nmp, nmp->nm_acregmax, ",acregmax", &buf, &blen);
1873 nfscl_printoptval(nmp, nmp->nm_nametimeo, ",nametimeo", &buf, &blen);
1874 nfscl_printoptval(nmp, nmp->nm_negnametimeo, ",negnametimeo", &buf,
1876 nfscl_printoptval(nmp, nmp->nm_rsize, ",rsize", &buf, &blen);
1877 nfscl_printoptval(nmp, nmp->nm_wsize, ",wsize", &buf, &blen);
1878 nfscl_printoptval(nmp, nmp->nm_readdirsize, ",readdirsize", &buf,
1880 nfscl_printoptval(nmp, nmp->nm_readahead, ",readahead", &buf, &blen);
1881 nfscl_printoptval(nmp, nmp->nm_wcommitsize, ",wcommitsize", &buf,
1883 nfscl_printoptval(nmp, nmp->nm_timeo, ",timeout", &buf, &blen);
1884 nfscl_printoptval(nmp, nmp->nm_retry, ",retrans", &buf, &blen);