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", "New NFS request header");
90 MALLOC_DEFINE(M_NEWNFSMNT, "newnfsmnt", "New 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, "");
104 static int nfs_mountroot(struct mount *);
105 static void nfs_sec_name(char *, int *);
106 static void nfs_decode_args(struct mount *mp, struct nfsmount *nmp,
107 struct nfs_args *argp, const char *, struct ucred *,
109 static 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);
113 static void nfs_getnlminfo(struct vnode *, uint8_t *, size_t *,
114 struct sockaddr_storage *, int *, off_t *,
116 static vfs_mount_t nfs_mount;
117 static vfs_cmount_t nfs_cmount;
118 static vfs_unmount_t nfs_unmount;
119 static vfs_root_t nfs_root;
120 static vfs_statfs_t nfs_statfs;
121 static vfs_sync_t nfs_sync;
122 static vfs_sysctl_t nfs_sysctl;
123 static vfs_purge_t nfs_purge;
126 * nfs vfs operations.
128 static 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,
140 VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK | VFCF_SBDRY);
142 /* So that loader and kldload(2) can find us, wherever we are.. */
143 MODULE_VERSION(nfs, 1);
144 MODULE_DEPEND(nfs, nfscommon, 1, 1, 1);
145 MODULE_DEPEND(nfs, krpc, 1, 1, 1);
146 MODULE_DEPEND(nfs, nfssvc, 1, 1, 1);
147 MODULE_DEPEND(nfs, nfslock, 1, 1, 1);
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.
155 #if !defined(NFS_ROOT) && !defined(NFSCLIENT)
156 struct nfs_diskless nfs_diskless = { { { 0 } } };
157 struct nfsv3_diskless nfsv3_diskless = { { { 0 } } };
158 int nfs_diskless_valid = 0;
161 SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD,
162 &nfs_diskless_valid, 0,
163 "Has the diskless struct been filled correctly");
165 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD,
166 nfsv3_diskless.root_hostnam, 0, "Path to nfs root");
168 SYSCTL_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");
173 void newnfsargs_ntoh(struct nfs_args *);
174 static int nfs_mountdiskless(char *,
175 struct sockaddr_in *, struct nfs_args *,
176 struct thread *, struct vnode **, struct mount *);
177 static void nfs_convert_diskless(void);
178 static void nfs_convert_oargs(struct nfs_args *args,
179 struct onfs_args *oargs);
182 newnfs_iosize(struct nfsmount *nmp)
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;
193 maxio = NFS_MAXBSIZE;
195 maxio = NFS_V2MAXDATA;
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;
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.
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;
225 nfs_convert_oargs(struct nfs_args *args, struct onfs_args *oargs)
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;
246 nfs_convert_diskless(void)
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);
258 nfsv3_diskless.root_fhsize = NFSX_V2FH;
259 bcopy(nfs_diskless.root_fh, nfsv3_diskless.root_fh, NFSX_V2FH);
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,
267 nfs_diskless_valid = 3;
274 nfs_statfs(struct mount *mp, struct statfs *sbp)
278 struct nfsmount *nmp = VFSTONFS(mp);
279 struct nfsvattr nfsva;
282 int error = 0, attrflag, gotfsinfo = 0, ret;
287 error = vfs_busy(mp, MBF_NOWAIT);
290 error = ncl_nget(mp, nmp->nm_fh, nmp->nm_fhsize, &np, LK_EXCLUSIVE);
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,
304 mtx_unlock(&nmp->nm_mtx);
306 error = nfsrpc_statfs(vp, &sb, &fs, td->td_ucred, td, &nfsva,
309 NFSCL_DEBUG(2, "statfs=%d\n", error);
311 ret = nfsrpc_getattrnovp(nmp, nmp->nm_fh, nmp->nm_fhsize, 1,
312 td->td_ucred, td, &nfsva, NULL, NULL);
315 * Just set default values to get things going.
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;
329 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
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);
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);
351 * nfs version 3 fsinfo rpc call
354 ncl_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct ucred *cred,
358 struct nfsvattr nfsva;
361 error = nfsrpc_fsinfo(vp, &fs, cred, td, &nfsva, &attrflag, NULL);
364 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
366 mtx_lock(&nmp->nm_mtx);
367 nfscl_loadfsinfo(nmp, &fs);
368 mtx_unlock(&nmp->nm_mtx);
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
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
382 * - build the rootfs mount point and call mountnfs() to do the rest.
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.
390 nfs_mountroot(struct mount *mp)
392 struct thread *td = curthread;
393 struct nfsv3_diskless *nd = &nfsv3_diskless;
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();
408 if (nfs_diskless_valid == 0)
410 if (nfs_diskless_valid == 1)
411 nfs_convert_diskless();
414 * XXX splnet, so networks will receive...
419 * Do enough of ifconfig(8) so that the critical net interface can
420 * talk to the server.
422 error = socreate(nd->myif.ifra_addr.sa_family, &so, nd->root_args.sotype, 0,
425 panic("nfs_mountroot: socreate(%04x): %d",
426 nd->myif.ifra_addr.sa_family, error);
428 #if 0 /* XXX Bad idea */
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.
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);
444 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td);
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);
451 error = ifioctl(so, SIOCSIFMTU, (caddr_t)&ir, td);
453 printf("nfs_mountroot: SIOCSIFMTU: %d", error);
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.
463 if (nd->mygateway.sin_len != 0 &&
464 nd->mygateway.sin_addr.s_addr != 0) {
465 struct sockaddr_in mask, sin;
467 bzero((caddr_t)&mask, sizeof(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);
479 panic("nfs_mountroot: RTM_ADD: %d", error);
483 * Create the rootfs mount point.
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) {
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.
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));
512 * Internal version of mount system call for diskless setup.
515 nfs_mountdiskless(char *path,
516 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
517 struct vnode **vpp, struct mount *mp)
519 struct sockaddr *nam;
524 * Find the directory path in "path", which also has the server's
525 * name/ip address in it.
527 dirpath = strchr(path, ':');
529 dirlen = strlen(++dirpath);
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);
543 nfs_sec_name(char *sec, int *flagsp)
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);
554 nfs_decode_args(struct mount *mp, struct nfsmount *nmp, struct nfs_args *argp,
555 const char *hostname, struct ucred *cred, struct thread *td)
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.
569 if (vfs_getopt(mp->mnt_optnew, "ro", NULL, NULL) == 0) {
571 mp->mnt_flag |= MNT_RDONLY;
573 } else if (mp->mnt_flag & MNT_UPDATE) {
575 mp->mnt_flag &= ~MNT_RDONLY;
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.
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;
590 nmp->nm_retry = NFS_RETRANS_TCP;
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;
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));
606 /* Update flags atomically. Don't change the lock bits. */
607 nmp->nm_flag = argp->flags | nmp->nm_flag;
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;
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;
624 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
625 nmp->nm_wsize = argp->wsize;
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.
631 if (nmp->nm_wsize > NFS_FABLKSIZE)
632 nmp->nm_wsize = 1 << (fls(nmp->nm_wsize) - 1);
634 nmp->nm_wsize = NFS_FABLKSIZE;
637 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
638 nmp->nm_rsize = argp->rsize;
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.
644 if (nmp->nm_rsize > NFS_FABLKSIZE)
645 nmp->nm_rsize = 1 << (fls(nmp->nm_rsize) - 1);
647 nmp->nm_rsize = NFS_FABLKSIZE;
650 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
651 nmp->nm_readdirsize = argp->readdirsize;
654 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
655 nmp->nm_acregmin = argp->acregmin;
657 nmp->nm_acregmin = NFS_MINATTRTIMO;
658 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
659 nmp->nm_acregmax = argp->acregmax;
661 nmp->nm_acregmax = NFS_MAXATTRTIMO;
662 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
663 nmp->nm_acdirmin = argp->acdirmin;
665 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
666 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
667 nmp->nm_acdirmax = argp->acdirmax;
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;
675 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
676 if (argp->readahead <= NFS_MAXRAHEAD)
677 nmp->nm_readahead = argp->readahead;
679 nmp->nm_readahead = NFS_MAXRAHEAD;
681 if ((argp->flags & NFSMNT_WCOMMITSIZE) && argp->wcommitsize >= 0) {
682 if (argp->wcommitsize < nmp->nm_wsize)
683 nmp->nm_wcommitsize = nmp->nm_wsize;
685 nmp->nm_wcommitsize = argp->wcommitsize;
688 adjsock |= ((nmp->nm_sotype != argp->sotype) ||
689 (nmp->nm_soproto != argp->proto));
691 if (nmp->nm_client != NULL && adjsock) {
692 int haslock = 0, error = 0;
694 if (nmp->nm_sotype == SOCK_STREAM) {
695 error = newnfs_sndlock(&nmp->nm_sockreq.nr_lock);
700 newnfs_disconnect(&nmp->nm_sockreq);
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,
708 printf("newnfs_args: retrying connect\n");
709 (void) nfs_catnap(PSOCK, 0, "newnfscon");
713 nmp->nm_sotype = argp->sotype;
714 nmp->nm_soproto = argp->proto;
717 if (hostname != NULL) {
718 strlcpy(nmp->nm_hostname, hostname,
719 sizeof(nmp->nm_hostname));
720 p = strchr(nmp->nm_hostname, ':');
726 static 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",
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.
749 nfs_mount(struct mount *mp)
751 struct nfs_args args = {
752 .version = NFS_ARGSVERSION,
754 .addrlen = sizeof (struct sockaddr_in),
755 .sotype = SOCK_STREAM,
759 .flags = NFSMNT_RESVPORT,
762 .readdirsize = NFS_READDIRSIZE,
764 .retrans = NFS_RETRANS,
765 .readahead = NFS_DEFRAHEAD,
766 .wcommitsize = 0, /* was: NQ_DEFLEASE */
768 .acregmin = NFS_MINATTRTIMO,
769 .acregmax = NFS_MAXATTRTIMO,
770 .acdirmin = NFS_MINDIRATTRTIMO,
771 .acdirmax = NFS_MAXDIRATTRTIMO,
773 int error = 0, ret, len;
774 struct sockaddr *nam = NULL;
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;
783 int dirlen, has_nfs_args_opt, krbnamelen, srvkrbnamelen;
786 has_nfs_args_opt = 0;
787 if (vfs_filteropt(mp->mnt_optnew, nfs_opts)) {
793 if ((mp->mnt_flag & (MNT_ROOTFS | MNT_UPDATE)) == MNT_ROOTFS) {
794 error = nfs_mountroot(mp);
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.
806 if (vfs_getopt(mp->mnt_optnew, "nfs_args", NULL, NULL) == 0) {
807 error = vfs_copyopt(mp->mnt_optnew, "nfs_args", &args,
812 if (args.version != NFS_ARGSVERSION) {
813 error = EPROGMISMATCH;
816 has_nfs_args_opt = 1;
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;
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;
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) {
868 vfs_mount_error(mp, "illegal readdirsize");
872 ret = sscanf(opt, "%d", &args.readdirsize);
873 if (ret != 1 || args.readdirsize <= 0) {
874 vfs_mount_error(mp, "illegal readdirsize: %s",
879 args.flags |= NFSMNT_READDIRSIZE;
881 if (vfs_getopt(mp->mnt_optnew, "readahead", (void **)&opt, NULL) == 0) {
883 vfs_mount_error(mp, "illegal readahead");
887 ret = sscanf(opt, "%d", &args.readahead);
888 if (ret != 1 || args.readahead <= 0) {
889 vfs_mount_error(mp, "illegal readahead: %s",
894 args.flags |= NFSMNT_READAHEAD;
896 if (vfs_getopt(mp->mnt_optnew, "wsize", (void **)&opt, NULL) == 0) {
898 vfs_mount_error(mp, "illegal wsize");
902 ret = sscanf(opt, "%d", &args.wsize);
903 if (ret != 1 || args.wsize <= 0) {
904 vfs_mount_error(mp, "illegal wsize: %s",
909 args.flags |= NFSMNT_WSIZE;
911 if (vfs_getopt(mp->mnt_optnew, "rsize", (void **)&opt, NULL) == 0) {
913 vfs_mount_error(mp, "illegal rsize");
917 ret = sscanf(opt, "%d", &args.rsize);
918 if (ret != 1 || args.rsize <= 0) {
919 vfs_mount_error(mp, "illegal wsize: %s",
924 args.flags |= NFSMNT_RSIZE;
926 if (vfs_getopt(mp->mnt_optnew, "retrans", (void **)&opt, NULL) == 0) {
928 vfs_mount_error(mp, "illegal retrans");
932 ret = sscanf(opt, "%d", &args.retrans);
933 if (ret != 1 || args.retrans <= 0) {
934 vfs_mount_error(mp, "illegal retrans: %s",
939 args.flags |= NFSMNT_RETRANS;
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",
949 args.acdirmin = args.acdirmax = args.acregmax = args.acregmin;
950 args.flags |= NFSMNT_ACDIRMIN | NFSMNT_ACDIRMAX |
951 NFSMNT_ACREGMIN | NFSMNT_ACREGMAX;
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",
961 args.flags |= NFSMNT_ACREGMIN;
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",
971 args.flags |= NFSMNT_ACREGMAX;
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",
981 args.flags |= NFSMNT_ACDIRMIN;
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",
991 args.flags |= NFSMNT_ACDIRMAX;
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);
1000 args.flags |= NFSMNT_WCOMMITSIZE;
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",
1010 args.flags |= NFSMNT_TIMEO;
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",
1020 args.flags |= NFSMNT_TIMEO;
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);
1030 if (vfs_getopt(mp->mnt_optnew, "negnametimeo", (void **)&opt, NULL)
1032 ret = sscanf(opt, "%d", &negnametimeo);
1033 if (ret != 1 || negnametimeo < 0) {
1034 vfs_mount_error(mp, "illegal negnametimeo: %s",
1040 if (vfs_getopt(mp->mnt_optnew, "minorversion", (void **)&opt, NULL) ==
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);
1050 if (vfs_getopt(mp->mnt_optnew, "sec",
1051 (void **) &secname, NULL) == 0)
1052 nfs_sec_name(secname, &args.flags);
1054 if (mp->mnt_flag & MNT_UPDATE) {
1055 struct nfsmount *nmp = VFSTONFS(mp);
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).
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");
1075 * When doing an update, we can't change version,
1076 * security, switch lockd strategies or change cookie
1079 args.flags = (args.flags &
1085 NFSMNT_NOLOCKD /*|NFSMNT_XLATECOOKIE*/)) |
1092 NFSMNT_NOLOCKD /*|NFSMNT_XLATECOOKIE*/));
1093 nfs_decode_args(mp, nmp, &args, NULL, td->td_ucred, td);
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),
1106 if (nfs_ip_paranoia == 0)
1107 args.flags |= NFSMNT_NOCONN;
1109 if (has_nfs_args_opt != 0) {
1111 * In the 'nfs_args' case, the pointers in the args
1112 * structure are in userland - we copy them in here.
1114 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX) {
1115 vfs_mount_error(mp, "Bad file handle");
1119 error = copyin((caddr_t)args.fh, (caddr_t)nfh,
1123 error = copyinstr(args.hostname, hst, MNAMELEN - 1, &hstlen);
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,
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");
1141 bcopy(args.fh, nfh, args.fhsize);
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");
1152 bcopy(args.hostname, hst, MNAMELEN);
1153 hst[MNAMELEN - 1] = '\0';
1156 if (vfs_getopt(mp->mnt_optnew, "principal", (void **)&name, NULL) == 0)
1157 strlcpy(srvkrbname, name, sizeof (srvkrbname));
1159 snprintf(srvkrbname, sizeof (srvkrbname), "nfs@%s", hst);
1160 cp = strchr(srvkrbname, ':');
1164 srvkrbnamelen = strlen(srvkrbname);
1166 if (vfs_getopt(mp->mnt_optnew, "gssname", (void **)&name, NULL) == 0)
1167 strlcpy(krbname, name, sizeof (krbname));
1170 krbnamelen = strlen(krbname);
1172 if (vfs_getopt(mp->mnt_optnew, "dirpath", (void **)&name, NULL) == 0)
1173 strlcpy(dirpath, name, sizeof (dirpath));
1176 dirlen = strlen(dirpath);
1178 if (has_nfs_args_opt == 0) {
1179 if (vfs_getopt(mp->mnt_optnew, "addr",
1180 (void **)&args.addr, &args.addrlen) == 0) {
1181 if (args.addrlen > SOCK_MAXADDRLEN) {
1182 error = ENAMETOOLONG;
1185 nam = malloc(args.addrlen, M_SONAME, M_WAITOK);
1186 bcopy(args.addr, nam, args.addrlen);
1187 nam->sa_len = args.addrlen;
1189 vfs_mount_error(mp, "No server address");
1196 error = mountnfs(&args, mp, nam, hst, krbname, krbnamelen, dirpath,
1197 dirlen, srvkrbname, srvkrbnamelen, &vp, td->td_ucred, td,
1198 nametimeo, negnametimeo, minvers);
1202 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_NO_IOPF |
1214 * It seems a bit dumb to copyinstr() the host and path here and then
1215 * bcopy() them in mountnfs(), but I wanted to detect errors before
1216 * doing the sockargs() call because sockargs() allocates an mbuf and
1217 * an error after that means that I have to release the mbuf.
1221 nfs_cmount(struct mntarg *ma, void *data, uint64_t flags)
1224 struct nfs_args args;
1226 error = copyin(data, &args, sizeof (struct nfs_args));
1230 ma = mount_arg(ma, "nfs_args", &args, sizeof args);
1232 error = kernel_mount(ma, flags);
1237 * Common code for mount and mountroot
1240 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
1241 char *hst, u_char *krbname, int krbnamelen, u_char *dirpath, int dirlen,
1242 u_char *srvkrbname, int srvkrbnamelen, struct vnode **vpp,
1243 struct ucred *cred, struct thread *td, int nametimeo, int negnametimeo,
1246 struct nfsmount *nmp;
1248 int error, trycnt, ret;
1249 struct nfsvattr nfsva;
1250 struct nfsclclient *clp;
1251 struct nfsclds *dsp, *tdsp;
1253 static u_int64_t clval = 0;
1255 NFSCL_DEBUG(3, "in mnt\n");
1257 if (mp->mnt_flag & MNT_UPDATE) {
1259 printf("%s: MNT_UPDATE is no longer handled here\n", __func__);
1260 FREE(nam, M_SONAME);
1263 MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount) +
1264 krbnamelen + dirlen + srvkrbnamelen + 2,
1265 M_NEWNFSMNT, M_WAITOK | M_ZERO);
1266 TAILQ_INIT(&nmp->nm_bufq);
1268 clval = (u_int64_t)nfsboottime.tv_sec;
1269 nmp->nm_clval = clval++;
1270 nmp->nm_krbnamelen = krbnamelen;
1271 nmp->nm_dirpathlen = dirlen;
1272 nmp->nm_srvkrbnamelen = srvkrbnamelen;
1273 if (td->td_ucred->cr_uid != (uid_t)0) {
1275 * nm_uid is used to get KerberosV credentials for
1276 * the nfsv4 state handling operations if there is
1277 * no host based principal set. Use the uid of
1278 * this user if not root, since they are doing the
1279 * mount. I don't think setting this for root will
1280 * work, since root normally does not have user
1281 * credentials in a credentials cache.
1283 nmp->nm_uid = td->td_ucred->cr_uid;
1286 * Just set to -1, so it won't be used.
1288 nmp->nm_uid = (uid_t)-1;
1291 /* Copy and null terminate all the names */
1292 if (nmp->nm_krbnamelen > 0) {
1293 bcopy(krbname, nmp->nm_krbname, nmp->nm_krbnamelen);
1294 nmp->nm_name[nmp->nm_krbnamelen] = '\0';
1296 if (nmp->nm_dirpathlen > 0) {
1297 bcopy(dirpath, NFSMNT_DIRPATH(nmp),
1298 nmp->nm_dirpathlen);
1299 nmp->nm_name[nmp->nm_krbnamelen + nmp->nm_dirpathlen
1302 if (nmp->nm_srvkrbnamelen > 0) {
1303 bcopy(srvkrbname, NFSMNT_SRVKRBNAME(nmp),
1304 nmp->nm_srvkrbnamelen);
1305 nmp->nm_name[nmp->nm_krbnamelen + nmp->nm_dirpathlen
1306 + nmp->nm_srvkrbnamelen + 2] = '\0';
1308 nmp->nm_sockreq.nr_cred = crhold(cred);
1309 mtx_init(&nmp->nm_sockreq.nr_mtx, "nfssock", NULL, MTX_DEF);
1311 nmp->nm_getinfo = nfs_getnlminfo;
1312 nmp->nm_vinvalbuf = ncl_vinvalbuf;
1315 nmp->nm_mountp = mp;
1316 mtx_init(&nmp->nm_mtx, "NFSmount lock", NULL, MTX_DEF | MTX_DUPOK);
1319 * Since nfs_decode_args() might optionally set them, these
1320 * need to be set to defaults before the call, so that the
1321 * optional settings aren't overwritten.
1323 nmp->nm_nametimeo = nametimeo;
1324 nmp->nm_negnametimeo = negnametimeo;
1325 nmp->nm_timeo = NFS_TIMEO;
1326 nmp->nm_retry = NFS_RETRANS;
1327 nmp->nm_readahead = NFS_DEFRAHEAD;
1329 /* This is empirical approximation of sqrt(hibufspace) * 256. */
1330 nmp->nm_wcommitsize = NFS_MAXBSIZE / 256;
1331 while ((long)nmp->nm_wcommitsize * nmp->nm_wcommitsize < hibufspace)
1332 nmp->nm_wcommitsize *= 2;
1333 nmp->nm_wcommitsize *= 256;
1335 if ((argp->flags & NFSMNT_NFSV4) != 0)
1336 nmp->nm_minorvers = minvers;
1338 nmp->nm_minorvers = 0;
1340 nfs_decode_args(mp, nmp, argp, hst, cred, td);
1343 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too
1344 * high, depending on whether we end up with negative offsets in
1345 * the client or server somewhere. 2GB-1 may be safer.
1347 * For V3, ncl_fsinfo will adjust this as necessary. Assume maximum
1348 * that we can handle until we find out otherwise.
1350 if ((argp->flags & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0)
1351 nmp->nm_maxfilesize = 0xffffffffLL;
1353 nmp->nm_maxfilesize = OFF_MAX;
1355 if ((argp->flags & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0) {
1356 nmp->nm_wsize = NFS_WSIZE;
1357 nmp->nm_rsize = NFS_RSIZE;
1358 nmp->nm_readdirsize = NFS_READDIRSIZE;
1360 nmp->nm_numgrps = NFS_MAXGRPS;
1361 nmp->nm_tprintf_delay = nfs_tprintf_delay;
1362 if (nmp->nm_tprintf_delay < 0)
1363 nmp->nm_tprintf_delay = 0;
1364 nmp->nm_tprintf_initial_delay = nfs_tprintf_initial_delay;
1365 if (nmp->nm_tprintf_initial_delay < 0)
1366 nmp->nm_tprintf_initial_delay = 0;
1367 nmp->nm_fhsize = argp->fhsize;
1368 if (nmp->nm_fhsize > 0)
1369 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
1370 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
1372 /* Set up the sockets and per-host congestion */
1373 nmp->nm_sotype = argp->sotype;
1374 nmp->nm_soproto = argp->proto;
1375 nmp->nm_sockreq.nr_prog = NFS_PROG;
1376 if ((argp->flags & NFSMNT_NFSV4))
1377 nmp->nm_sockreq.nr_vers = NFS_VER4;
1378 else if ((argp->flags & NFSMNT_NFSV3))
1379 nmp->nm_sockreq.nr_vers = NFS_VER3;
1381 nmp->nm_sockreq.nr_vers = NFS_VER2;
1384 if ((error = newnfs_connect(nmp, &nmp->nm_sockreq, cred, td, 0)))
1386 /* For NFSv4.1, get the clientid now. */
1387 if (nmp->nm_minorvers > 0) {
1388 NFSCL_DEBUG(3, "at getcl\n");
1389 error = nfscl_getcl(mp, cred, td, 0, &clp);
1390 NFSCL_DEBUG(3, "aft getcl=%d\n", error);
1395 if (nmp->nm_fhsize == 0 && (nmp->nm_flag & NFSMNT_NFSV4) &&
1396 nmp->nm_dirpathlen > 0) {
1397 NFSCL_DEBUG(3, "in dirp\n");
1399 * If the fhsize on the mount point == 0 for V4, the mount
1400 * path needs to be looked up.
1404 error = nfsrpc_getdirpath(nmp, NFSMNT_DIRPATH(nmp),
1406 NFSCL_DEBUG(3, "aft dirp=%d\n", error);
1408 (void) nfs_catnap(PZERO, error, "nfsgetdirp");
1409 } while (error && --trycnt > 0);
1411 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1417 * A reference count is needed on the nfsnode representing the
1418 * remote root. If this object is not persistent, then backward
1419 * traversals of the mount point (i.e. "..") will not work if
1420 * the nfsnode gets flushed out of the cache. Ufs does not have
1421 * this problem, because one can identify root inodes by their
1422 * number == ROOTINO (2).
1424 if (nmp->nm_fhsize > 0) {
1426 * Set f_iosize to NFS_DIRBLKSIZ so that bo_bsize gets set
1427 * non-zero for the root vnode. f_iosize will be set correctly
1428 * by nfs_statfs() before any I/O occurs.
1430 mp->mnt_stat.f_iosize = NFS_DIRBLKSIZ;
1431 error = ncl_nget(mp, nmp->nm_fh, nmp->nm_fhsize, &np,
1438 * Get file attributes and transfer parameters for the
1439 * mountpoint. This has the side effect of filling in
1440 * (*vpp)->v_type with the correct value.
1442 ret = nfsrpc_getattrnovp(nmp, nmp->nm_fh, nmp->nm_fhsize, 1,
1443 cred, td, &nfsva, NULL, &lease);
1446 * Just set default values to get things going.
1448 NFSBZERO((caddr_t)&nfsva, sizeof (struct nfsvattr));
1449 nfsva.na_vattr.va_type = VDIR;
1450 nfsva.na_vattr.va_mode = 0777;
1451 nfsva.na_vattr.va_nlink = 100;
1452 nfsva.na_vattr.va_uid = (uid_t)0;
1453 nfsva.na_vattr.va_gid = (gid_t)0;
1454 nfsva.na_vattr.va_fileid = 2;
1455 nfsva.na_vattr.va_gen = 1;
1456 nfsva.na_vattr.va_blocksize = NFS_FABLKSIZE;
1457 nfsva.na_vattr.va_size = 512 * 1024;
1460 (void) nfscl_loadattrcache(vpp, &nfsva, NULL, NULL, 0, 1);
1461 if (nmp->nm_minorvers > 0) {
1462 NFSCL_DEBUG(3, "lease=%d\n", (int)lease);
1464 clp->nfsc_renew = NFSCL_RENEW(lease);
1465 clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
1466 clp->nfsc_clientidrev++;
1467 if (clp->nfsc_clientidrev == 0)
1468 clp->nfsc_clientidrev++;
1471 * Mount will succeed, so the renew thread can be
1474 nfscl_start_renewthread(clp);
1475 nfscl_clientrelease(clp);
1477 if (argp->flags & NFSMNT_NFSV3)
1478 ncl_fsinfo(nmp, *vpp, cred, td);
1480 /* Mark if the mount point supports NFSv4 ACLs. */
1481 if ((argp->flags & NFSMNT_NFSV4) != 0 && nfsrv_useacl != 0 &&
1483 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) {
1485 mp->mnt_flag |= MNT_NFS4ACLS;
1490 * Lose the lock but keep the ref.
1492 NFSVOPUNLOCK(*vpp, 0);
1499 nfscl_clientrelease(clp);
1500 newnfs_disconnect(&nmp->nm_sockreq);
1501 crfree(nmp->nm_sockreq.nr_cred);
1502 if (nmp->nm_sockreq.nr_auth != NULL)
1503 AUTH_DESTROY(nmp->nm_sockreq.nr_auth);
1504 mtx_destroy(&nmp->nm_sockreq.nr_mtx);
1505 mtx_destroy(&nmp->nm_mtx);
1506 if (nmp->nm_clp != NULL) {
1508 LIST_REMOVE(nmp->nm_clp, nfsc_list);
1510 free(nmp->nm_clp, M_NFSCLCLIENT);
1512 TAILQ_FOREACH_SAFE(dsp, &nmp->nm_sess, nfsclds_list, tdsp)
1513 nfscl_freenfsclds(dsp);
1514 FREE(nmp, M_NEWNFSMNT);
1515 FREE(nam, M_SONAME);
1520 * unmount system call
1523 nfs_unmount(struct mount *mp, int mntflags)
1526 struct nfsmount *nmp;
1527 int error, flags = 0, i, trycnt = 0;
1528 struct nfsclds *dsp, *tdsp;
1532 if (mntflags & MNT_FORCE)
1533 flags |= FORCECLOSE;
1536 * Goes something like this..
1537 * - Call vflush() to clear out vnodes for this filesystem
1538 * - Close the socket
1539 * - Free up the data structures
1541 /* In the forced case, cancel any outstanding requests. */
1542 if (mntflags & MNT_FORCE) {
1543 error = newnfs_nmcancelreqs(nmp);
1546 /* For a forced close, get rid of the renew thread now */
1547 nfscl_umount(nmp, td);
1549 /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */
1551 error = vflush(mp, 1, flags, td);
1552 if ((mntflags & MNT_FORCE) && error != 0 && ++trycnt < 30)
1553 (void) nfs_catnap(PSOCK, error, "newndm");
1554 } while ((mntflags & MNT_FORCE) && error != 0 && trycnt < 30);
1559 * We are now committed to the unmount.
1561 if ((mntflags & MNT_FORCE) == 0)
1562 nfscl_umount(nmp, td);
1563 /* Make sure no nfsiods are assigned to this mount. */
1564 mtx_lock(&ncl_iod_mutex);
1565 for (i = 0; i < NFS_MAXASYNCDAEMON; i++)
1566 if (ncl_iodmount[i] == nmp) {
1567 ncl_iodwant[i] = NFSIOD_AVAILABLE;
1568 ncl_iodmount[i] = NULL;
1570 mtx_unlock(&ncl_iod_mutex);
1571 newnfs_disconnect(&nmp->nm_sockreq);
1572 crfree(nmp->nm_sockreq.nr_cred);
1573 FREE(nmp->nm_nam, M_SONAME);
1574 if (nmp->nm_sockreq.nr_auth != NULL)
1575 AUTH_DESTROY(nmp->nm_sockreq.nr_auth);
1576 mtx_destroy(&nmp->nm_sockreq.nr_mtx);
1577 mtx_destroy(&nmp->nm_mtx);
1578 TAILQ_FOREACH_SAFE(dsp, &nmp->nm_sess, nfsclds_list, tdsp)
1579 nfscl_freenfsclds(dsp);
1580 FREE(nmp, M_NEWNFSMNT);
1586 * Return root of a filesystem
1589 nfs_root(struct mount *mp, int flags, struct vnode **vpp)
1592 struct nfsmount *nmp;
1597 error = ncl_nget(mp, nmp->nm_fh, nmp->nm_fhsize, &np, flags);
1602 * Get transfer parameters and attributes for root vnode once.
1604 mtx_lock(&nmp->nm_mtx);
1605 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) {
1606 mtx_unlock(&nmp->nm_mtx);
1607 ncl_fsinfo(nmp, vp, curthread->td_ucred, curthread);
1609 mtx_unlock(&nmp->nm_mtx);
1610 if (vp->v_type == VNON)
1612 vp->v_vflag |= VV_ROOT;
1618 * Flush out the buffer cache
1622 nfs_sync(struct mount *mp, int waitfor)
1624 struct vnode *vp, *mvp;
1626 int error, allerror = 0;
1632 * If a forced dismount is in progress, return from here so that
1633 * the umount(2) syscall doesn't get stuck in VFS_SYNC() before
1634 * calling VFS_UNMOUNT().
1636 if ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
1643 * Force stale buffer cache information to be flushed.
1646 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1647 /* XXX Racy bv_cnt check. */
1648 if (NFSVOPISLOCKED(vp) || vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
1649 waitfor == MNT_LAZY) {
1653 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
1654 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1657 error = VOP_FSYNC(vp, waitfor, td);
1660 NFSVOPUNLOCK(vp, 0);
1667 nfs_sysctl(struct mount *mp, fsctlop_t op, struct sysctl_req *req)
1669 struct nfsmount *nmp = VFSTONFS(mp);
1673 bzero(&vq, sizeof(vq));
1676 case VFS_CTL_NOLOCKS:
1677 val = (nmp->nm_flag & NFSMNT_NOLOCKS) ? 1 : 0;
1678 if (req->oldptr != NULL) {
1679 error = SYSCTL_OUT(req, &val, sizeof(val));
1683 if (req->newptr != NULL) {
1684 error = SYSCTL_IN(req, &val, sizeof(val));
1688 nmp->nm_flag |= NFSMNT_NOLOCKS;
1690 nmp->nm_flag &= ~NFSMNT_NOLOCKS;
1695 mtx_lock(&nmp->nm_mtx);
1696 if (nmp->nm_state & NFSSTA_TIMEO)
1697 vq.vq_flags |= VQ_NOTRESP;
1698 mtx_unlock(&nmp->nm_mtx);
1700 if (!(nmp->nm_flag & NFSMNT_NOLOCKS) &&
1701 (nmp->nm_state & NFSSTA_LOCKTIMEO))
1702 vq.vq_flags |= VQ_NOTRESPLOCK;
1704 error = SYSCTL_OUT(req, &vq, sizeof(vq));
1707 if (req->oldptr != NULL) {
1708 error = SYSCTL_OUT(req, &nmp->nm_tprintf_initial_delay,
1709 sizeof(nmp->nm_tprintf_initial_delay));
1713 if (req->newptr != NULL) {
1714 error = vfs_suser(mp, req->td);
1717 error = SYSCTL_IN(req, &nmp->nm_tprintf_initial_delay,
1718 sizeof(nmp->nm_tprintf_initial_delay));
1721 if (nmp->nm_tprintf_initial_delay < 0)
1722 nmp->nm_tprintf_initial_delay = 0;
1732 * Purge any RPCs in progress, so that they will all return errors.
1733 * This allows dounmount() to continue as far as VFS_UNMOUNT() for a
1737 nfs_purge(struct mount *mp)
1739 struct nfsmount *nmp = VFSTONFS(mp);
1741 newnfs_nmcancelreqs(nmp);
1745 * Extract the information needed by the nlm from the nfs vnode.
1748 nfs_getnlminfo(struct vnode *vp, uint8_t *fhp, size_t *fhlenp,
1749 struct sockaddr_storage *sp, int *is_v3p, off_t *sizep,
1750 struct timeval *timeop)
1752 struct nfsmount *nmp;
1753 struct nfsnode *np = VTONFS(vp);
1755 nmp = VFSTONFS(vp->v_mount);
1757 *fhlenp = (size_t)np->n_fhp->nfh_len;
1759 bcopy(np->n_fhp->nfh_fh, fhp, np->n_fhp->nfh_len);
1761 bcopy(nmp->nm_nam, sp, min(nmp->nm_nam->sa_len, sizeof(*sp)));
1763 *is_v3p = NFS_ISV3(vp);
1765 *sizep = np->n_size;
1766 if (timeop != NULL) {
1767 timeop->tv_sec = nmp->nm_timeo / NFS_HZ;
1768 timeop->tv_usec = (nmp->nm_timeo % NFS_HZ) * (1000000 / NFS_HZ);
1773 * This function prints out an option name, based on the conditional
1776 static __inline void nfscl_printopt(struct nfsmount *nmp, int testval,
1777 char *opt, char **buf, size_t *blen)
1781 if (testval != 0 && *blen > strlen(opt)) {
1782 len = snprintf(*buf, *blen, "%s", opt);
1783 if (len != strlen(opt))
1791 * This function printf out an options integer value.
1793 static __inline void nfscl_printoptval(struct nfsmount *nmp, int optval,
1794 char *opt, char **buf, size_t *blen)
1798 if (*blen > strlen(opt) + 1) {
1799 /* Could result in truncated output string. */
1800 len = snprintf(*buf, *blen, "%s=%d", opt, optval);
1809 * Load the option flags and values into the buffer.
1811 void nfscl_retopts(struct nfsmount *nmp, char *buffer, size_t buflen)
1818 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NFSV4) != 0, "nfsv4", &buf,
1820 if ((nmp->nm_flag & NFSMNT_NFSV4) != 0) {
1821 nfscl_printoptval(nmp, nmp->nm_minorvers, ",minorversion", &buf,
1823 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_PNFS) != 0, ",pnfs",
1826 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NFSV3) != 0, "nfsv3", &buf,
1828 nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0,
1829 "nfsv2", &buf, &blen);
1830 nfscl_printopt(nmp, nmp->nm_sotype == SOCK_STREAM, ",tcp", &buf, &blen);
1831 nfscl_printopt(nmp, nmp->nm_sotype != SOCK_STREAM, ",udp", &buf, &blen);
1832 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_RESVPORT) != 0, ",resvport",
1834 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NOCONN) != 0, ",noconn",
1836 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_SOFT) == 0, ",hard", &buf,
1838 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_SOFT) != 0, ",soft", &buf,
1840 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_INT) != 0, ",intr", &buf,
1842 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NOCTO) == 0, ",cto", &buf,
1844 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NOCTO) != 0, ",nocto", &buf,
1846 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NONCONTIGWR) != 0,
1847 ",noncontigwr", &buf, &blen);
1848 nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_NOLOCKD | NFSMNT_NFSV4)) ==
1849 0, ",lockd", &buf, &blen);
1850 nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_NOLOCKD | NFSMNT_NFSV4)) ==
1851 NFSMNT_NOLOCKD, ",nolockd", &buf, &blen);
1852 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_RDIRPLUS) != 0, ",rdirplus",
1854 nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_KERB) == 0, ",sec=sys",
1856 nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_KERB | NFSMNT_INTEGRITY |
1857 NFSMNT_PRIVACY)) == NFSMNT_KERB, ",sec=krb5", &buf, &blen);
1858 nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_KERB | NFSMNT_INTEGRITY |
1859 NFSMNT_PRIVACY)) == (NFSMNT_KERB | NFSMNT_INTEGRITY), ",sec=krb5i",
1861 nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_KERB | NFSMNT_INTEGRITY |
1862 NFSMNT_PRIVACY)) == (NFSMNT_KERB | NFSMNT_PRIVACY), ",sec=krb5p",
1864 nfscl_printoptval(nmp, nmp->nm_acdirmin, ",acdirmin", &buf, &blen);
1865 nfscl_printoptval(nmp, nmp->nm_acdirmax, ",acdirmax", &buf, &blen);
1866 nfscl_printoptval(nmp, nmp->nm_acregmin, ",acregmin", &buf, &blen);
1867 nfscl_printoptval(nmp, nmp->nm_acregmax, ",acregmax", &buf, &blen);
1868 nfscl_printoptval(nmp, nmp->nm_nametimeo, ",nametimeo", &buf, &blen);
1869 nfscl_printoptval(nmp, nmp->nm_negnametimeo, ",negnametimeo", &buf,
1871 nfscl_printoptval(nmp, nmp->nm_rsize, ",rsize", &buf, &blen);
1872 nfscl_printoptval(nmp, nmp->nm_wsize, ",wsize", &buf, &blen);
1873 nfscl_printoptval(nmp, nmp->nm_readdirsize, ",readdirsize", &buf,
1875 nfscl_printoptval(nmp, nmp->nm_readahead, ",readahead", &buf, &blen);
1876 nfscl_printoptval(nmp, nmp->nm_wcommitsize, ",wcommitsize", &buf,
1878 nfscl_printoptval(nmp, nmp->nm_timeo, ",timeout", &buf, &blen);
1879 nfscl_printoptval(nmp, nmp->nm_retry, ",retrans", &buf, &blen);