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 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)nfs_vfsops.c 8.12 (Berkeley) 5/20/95
37 * $FreeBSD: src/sys/nfs/nfs_vfsops.c,v 1.91.2.7 2003/01/27 20:04:08 dillon Exp $
38 * $DragonFly: src/sys/vfs/nfs/nfs_vfsops.c,v 1.54 2008/07/31 20:23:40 swildner Exp $
41 #include "opt_bootp.h"
42 #include "opt_nfsroot.h"
44 #include <sys/param.h>
45 #include <sys/sockio.h>
47 #include <sys/vnode.h>
48 #include <sys/fcntl.h>
49 #include <sys/kernel.h>
50 #include <sys/sysctl.h>
51 #include <sys/malloc.h>
52 #include <sys/mount.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/systm.h>
59 #include <vm/vm_extern.h>
60 #include <vm/vm_zone.h>
63 #include <net/route.h>
64 #include <netinet/in.h>
66 #include <sys/thread2.h>
67 #include <sys/mutex2.h>
75 #include "nfsm_subs.h"
76 #include "nfsdiskless.h"
77 #include "nfsmountrpc.h"
79 extern int nfs_mountroot(struct mount *mp);
80 extern void bootpc_init(void);
83 extern struct vop_ops nfsv2_vnode_vops;
84 extern struct vop_ops nfsv2_fifo_vops;
85 extern struct vop_ops nfsv2_spec_vops;
87 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header");
88 MALLOC_DEFINE(M_NFSBIGFH, "NFSV3 bigfh", "NFS version 3 file handle");
89 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure");
90 MALLOC_DEFINE(M_NFSDIROFF, "NFSV3 diroff", "NFS directory offset data");
91 MALLOC_DEFINE(M_NFSRVDESC, "NFSV3 srvdesc", "NFS server socket descriptor");
92 MALLOC_DEFINE(M_NFSUID, "NFS uid", "Nfs uid mapping structure");
93 MALLOC_DEFINE(M_NFSHASH, "NFS hash", "NFS hash tables");
95 vm_zone_t nfsmount_zone;
97 struct nfsstats nfsstats;
98 SYSCTL_NODE(_vfs, OID_AUTO, nfs, CTLFLAG_RW, 0, "NFS filesystem");
99 SYSCTL_STRUCT(_vfs_nfs, NFS_NFSSTATS, nfsstats, CTLFLAG_RD,
100 &nfsstats, nfsstats, "");
101 static int nfs_ip_paranoia = 1;
102 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW,
103 &nfs_ip_paranoia, 0, "");
106 SYSCTL_INT(_vfs_nfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0, "");
110 * Tunable to determine the Read/Write unit size. Maximum value
111 * is NFS_MAXDATA. We also default to NFS_MAXDATA.
113 static int nfs_io_size = NFS_MAXDATA;
114 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_io_size, CTLFLAG_RW,
115 &nfs_io_size, 0, "NFS optimal I/O unit size");
117 static void nfs_decode_args (struct nfsmount *nmp,
118 struct nfs_args *argp);
119 static int mountnfs (struct nfs_args *,struct mount *,
120 struct sockaddr *,char *,char *,struct vnode **);
121 static int nfs_mount ( struct mount *mp, char *path, caddr_t data,
123 static int nfs_unmount ( struct mount *mp, int mntflags);
124 static int nfs_root ( struct mount *mp, struct vnode **vpp);
125 static int nfs_statfs ( struct mount *mp, struct statfs *sbp,
127 static int nfs_sync ( struct mount *mp, int waitfor);
130 * nfs vfs operations.
132 static struct vfsops nfs_vfsops = {
133 .vfs_mount = nfs_mount,
134 .vfs_unmount = nfs_unmount,
135 .vfs_root = nfs_root,
136 .vfs_statfs = nfs_statfs,
137 .vfs_sync = nfs_sync,
138 .vfs_init = nfs_init,
139 .vfs_uninit = nfs_uninit
141 VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK);
144 * This structure must be filled in by a primary bootstrap or bootstrap
145 * server for a diskless/dataless machine. It is initialized below just
146 * to ensure that it is allocated to initialized data (.data not .bss).
148 struct nfs_diskless nfs_diskless = { { { 0 } } };
149 struct nfsv3_diskless nfsv3_diskless = { { { 0 } } };
150 int nfs_diskless_valid = 0;
152 SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD,
153 &nfs_diskless_valid, 0, "");
155 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD,
156 nfsv3_diskless.root_hostnam, 0, "");
158 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD,
159 &nfsv3_diskless.root_saddr, sizeof nfsv3_diskless.root_saddr,
160 "%Ssockaddr_in", "");
162 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_swappath, CTLFLAG_RD,
163 nfsv3_diskless.swap_hostnam, 0, "");
165 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_swapaddr, CTLFLAG_RD,
166 &nfsv3_diskless.swap_saddr, sizeof nfsv3_diskless.swap_saddr,
170 void nfsargs_ntoh (struct nfs_args *);
171 static int nfs_mountdiskless (char *, char *, int,
172 struct sockaddr_in *, struct nfs_args *,
173 struct thread *, struct vnode **,
175 static void nfs_convert_diskless (void);
176 static void nfs_convert_oargs (struct nfs_args *args,
177 struct onfs_args *oargs);
180 * Calculate the buffer I/O block size to use. The maximum V2 block size
181 * is typically 8K, the maximum datagram size is typically 16K, and the
182 * maximum V3 block size is typically 32K. The buffer cache tends to work
183 * best with 16K blocks but we allow 32K for TCP connections.
185 * We force the block size to be at least a page for buffer cache efficiency.
188 nfs_iosize(int v3, int sotype)
194 if (sotype == SOCK_STREAM)
197 iomax = NFS_MAXDGRAMDATA;
199 iomax = NFS_V2MAXDATA;
201 if ((iosize = nfs_io_size) > iomax)
203 if (iosize < PAGE_SIZE)
207 * This is an aweful hack but until the buffer cache is rewritten
208 * we need it. The problem is that when you combine write() with
209 * mmap() the vm_page->valid bits can become weird looking
210 * (e.g. 0xfc). This occurs because NFS uses piecemeal buffers
211 * at the file EOF. To solve the problem the BIO system needs to
212 * be guarenteed that the NFS iosize for regular files will be a
213 * multiple of PAGE_SIZE so it can invalidate the whole page
214 * rather then just the piece of it owned by the buffer when
215 * NFS does vinvalbuf() calls.
217 if (iosize & PAGE_MASK)
218 iosize = (iosize & ~PAGE_MASK) + PAGE_SIZE;
223 nfs_convert_oargs(struct nfs_args *args, struct onfs_args *oargs)
225 args->version = NFS_ARGSVERSION;
226 args->addr = oargs->addr;
227 args->addrlen = oargs->addrlen;
228 args->sotype = oargs->sotype;
229 args->proto = oargs->proto;
230 args->fh = oargs->fh;
231 args->fhsize = oargs->fhsize;
232 args->flags = oargs->flags;
233 args->wsize = oargs->wsize;
234 args->rsize = oargs->rsize;
235 args->readdirsize = oargs->readdirsize;
236 args->timeo = oargs->timeo;
237 args->retrans = oargs->retrans;
238 args->maxgrouplist = oargs->maxgrouplist;
239 args->readahead = oargs->readahead;
240 args->deadthresh = oargs->deadthresh;
241 args->hostname = oargs->hostname;
245 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.swap_args,&nfs_diskless.swap_args);
255 bcopy(nfs_diskless.swap_fh,nfsv3_diskless.swap_fh,NFSX_V2FH);
256 nfsv3_diskless.swap_fhsize = NFSX_V2FH;
257 for (i = NFSX_V2FH - 1; i >= 0; --i) {
258 if (nfs_diskless.swap_fh[i])
262 nfsv3_diskless.swap_fhsize = 0;
264 bcopy(&nfs_diskless.swap_saddr,&nfsv3_diskless.swap_saddr,
265 sizeof(struct sockaddr_in));
266 bcopy(nfs_diskless.swap_hostnam,nfsv3_diskless.swap_hostnam, MNAMELEN);
267 nfsv3_diskless.swap_nblks = nfs_diskless.swap_nblks;
268 bcopy(&nfs_diskless.swap_ucred, &nfsv3_diskless.swap_ucred,
269 sizeof(struct ucred));
270 nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args);
272 bcopy(nfs_diskless.root_fh,nfsv3_diskless.root_fh,NFSX_V2FH);
273 nfsv3_diskless.root_fhsize = NFSX_V2FH;
274 for (i = NFSX_V2FH - 1; i >= 0; --i) {
275 if (nfs_diskless.root_fh[i])
279 nfsv3_diskless.root_fhsize = 0;
281 bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr,
282 sizeof(struct sockaddr_in));
283 bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN);
284 nfsv3_diskless.root_time = nfs_diskless.root_time;
285 bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam,
287 nfs_diskless_valid = 3;
294 nfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
297 struct nfs_statfs *sfp;
301 caddr_t bpos, dpos, cp2;
302 struct nfsmount *nmp = VFSTONFS(mp);
303 thread_t td = curthread;
304 int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr;
305 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
312 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
316 /* ignore the passed cred */
318 cred->cr_ngroups = 1;
319 if (v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
320 (void)nfs_fsinfo(nmp, vp, td);
321 nfsstats.rpccnt[NFSPROC_FSSTAT]++;
322 nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
324 nfsm_request(vp, NFSPROC_FSSTAT, td, cred);
326 nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK);
332 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
333 sbp->f_flags = nmp->nm_flag;
334 sbp->f_iosize = nfs_iosize(v3, nmp->nm_sotype);
337 sbp->f_bsize = NFS_FABLKSIZE;
338 tquad = fxdr_hyper(&sfp->sf_tbytes);
339 sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
340 tquad = fxdr_hyper(&sfp->sf_fbytes);
341 sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
342 tquad = fxdr_hyper(&sfp->sf_abytes);
343 sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
344 sbp->f_files = (fxdr_unsigned(int32_t,
345 sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
346 sbp->f_ffree = (fxdr_unsigned(int32_t,
347 sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
349 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
350 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
351 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
352 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
356 if (sbp != &mp->mnt_stat) {
357 sbp->f_type = mp->mnt_vfc->vfc_typenum;
358 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
368 * nfs version 3 fsinfo rpc call
371 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td)
373 struct nfsv3_fsinfo *fsp;
376 u_int32_t *tl, pref, max;
377 caddr_t bpos, dpos, cp2;
378 int error = 0, retattr;
379 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
382 nfsstats.rpccnt[NFSPROC_FSINFO]++;
383 nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
385 nfsm_request(vp, NFSPROC_FSINFO, td, nfs_vpcred(vp, ND_READ));
386 nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK);
388 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
389 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
390 if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
391 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
392 ~(NFS_FABLKSIZE - 1);
393 max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
394 if (max < nmp->nm_wsize && max > 0) {
395 nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
396 if (nmp->nm_wsize == 0)
399 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
400 if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
401 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
402 ~(NFS_FABLKSIZE - 1);
403 max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
404 if (max < nmp->nm_rsize && max > 0) {
405 nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
406 if (nmp->nm_rsize == 0)
409 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
410 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ)
411 nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) &
412 ~(NFS_DIRBLKSIZ - 1);
413 if (max < nmp->nm_readdirsize && max > 0) {
414 nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
415 if (nmp->nm_readdirsize == 0)
416 nmp->nm_readdirsize = max;
418 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
419 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
420 nmp->nm_maxfilesize = maxfsize;
421 nmp->nm_state |= NFSSTA_GOTFSINFO;
429 * Mount a remote root fs via. nfs. This depends on the info in the
430 * nfs_diskless structure that has been filled in properly by some primary
432 * It goes something like this:
433 * - do enough of "ifconfig" by calling ifioctl() so that the system
434 * can talk to the server
435 * - If nfs_diskless.mygateway is filled in, use that address as
437 * - build the rootfs mount point and call mountnfs() to do the rest.
440 nfs_mountroot(struct mount *mp)
442 struct mount *swap_mp;
443 struct nfsv3_diskless *nd = &nfsv3_diskless;
446 struct thread *td = curthread; /* XXX */
451 #if defined(BOOTP_NFSROOT) && defined(BOOTP)
452 bootpc_init(); /* use bootp to get nfs_diskless filled in */
456 * XXX time must be non-zero when we init the interface or else
457 * the arp code will wedge...
459 while (mycpu->gd_time_seconds == 0)
460 tsleep(mycpu, 0, "arpkludge", 10);
463 * The boot code may have passed us a diskless structure.
465 if (nfs_diskless_valid == 1)
466 nfs_convert_diskless();
468 #define SINP(sockaddr) ((struct sockaddr_in *)(sockaddr))
469 kprintf("nfs_mountroot: interface %s ip %s",
471 inet_ntoa(SINP(&nd->myif.ifra_addr)->sin_addr));
473 inet_ntoa(SINP(&nd->myif.ifra_broadaddr)->sin_addr));
474 kprintf(" mask %s\n",
475 inet_ntoa(SINP(&nd->myif.ifra_mask)->sin_addr));
479 * XXX splnet, so networks will receive...
484 * BOOTP does not necessarily have to be compiled into the kernel
485 * for an NFS root to work. If we inherited the network
486 * configuration for PXEBOOT then pxe_setup_nfsdiskless() has figured
487 * out our interface for us and all we need to do is ifconfig the
488 * interface. We only do this if the interface has not already been
489 * ifconfig'd by e.g. BOOTP.
491 error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td);
493 panic("nfs_mountroot: socreate(%04x): %d",
494 nd->myif.ifra_addr.sa_family, error);
497 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, proc0.p_ucred);
499 panic("nfs_mountroot: SIOCAIFADDR: %d", error);
501 soclose(so, FNONBLOCK);
504 * If the gateway field is filled in, set it as the default route.
506 if (nd->mygateway.sin_len != 0) {
507 struct sockaddr_in mask, sin;
509 bzero((caddr_t)&mask, sizeof(mask));
511 sin.sin_family = AF_INET;
512 sin.sin_len = sizeof(sin);
513 kprintf("nfs_mountroot: gateway %s\n",
514 inet_ntoa(nd->mygateway.sin_addr));
515 error = rtrequest_global(RTM_ADD, (struct sockaddr *)&sin,
516 (struct sockaddr *)&nd->mygateway,
517 (struct sockaddr *)&mask,
518 RTF_UP | RTF_GATEWAY);
520 kprintf("nfs_mountroot: unable to set gateway, error %d, continuing anyway\n", error);
524 * Create the rootfs mount point.
526 nd->root_args.fh = nd->root_fh;
527 nd->root_args.fhsize = nd->root_fhsize;
528 l = ntohl(nd->root_saddr.sin_addr.s_addr);
529 ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
530 (l >> 24) & 0xff, (l >> 16) & 0xff,
531 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->root_hostnam);
532 kprintf("NFS_ROOT: %s\n",buf);
533 if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY,
534 &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) {
535 mp->mnt_vfc->vfc_refcount--;
541 if (nd->swap_nblks) {
543 /* Convert to DEV_BSIZE instead of Kilobyte */
547 * Create a fake mount point just for the swap vnode so that the
548 * swap file can be on a different server from the rootfs.
550 nd->swap_args.fh = nd->swap_fh;
551 nd->swap_args.fhsize = nd->swap_fhsize;
552 l = ntohl(nd->swap_saddr.sin_addr.s_addr);
553 ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
554 (l >> 24) & 0xff, (l >> 16) & 0xff,
555 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->swap_hostnam);
556 kprintf("NFS SWAP: %s\n",buf);
557 if ((error = nfs_mountdiskless(buf, "/swap", 0,
558 &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0) {
564 VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size =
565 nd->swap_nblks * DEV_BSIZE ;
568 * Since the swap file is not the root dir of a file system,
569 * hack it to a regular file.
573 nfs_setvtype(vp, VREG);
574 swaponvp(td, vp, nd->swap_nblks);
577 mp->mnt_flag |= MNT_ROOTFS;
581 * This is not really an nfs issue, but it is much easier to
582 * set hostname here and then let the "/etc/rc.xxx" files
583 * mount the right /var based upon its preset value.
585 bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN);
586 hostname[MAXHOSTNAMELEN - 1] = '\0';
587 for (i = 0; i < MAXHOSTNAMELEN; i++)
588 if (hostname[i] == '\0')
590 inittodr(ntohl(nd->root_time));
596 * Internal version of mount system call for diskless setup.
599 nfs_mountdiskless(char *path, char *which, int mountflag,
600 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
601 struct vnode **vpp, struct mount **mpp)
604 struct sockaddr *nam;
611 if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) {
612 kprintf("nfs_mountroot: NFS not configured");
617 mp->mnt_kern_flag = 0;
618 mp->mnt_flag = mountflag;
619 nam = dup_sockaddr((struct sockaddr *)sin);
621 #if defined(BOOTP) || defined(NFS_ROOT)
622 if (args->fhsize == 0) {
625 kprintf("NFS_ROOT: No FH passed from loader, attempting mount rpc...");
626 while (*xpath && *xpath != ':')
631 error = md_mount(sin, xpath, args->fh, &args->fhsize, args, td);
633 kprintf("failed error %d.\n", error);
636 kprintf("success!\n");
640 if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) {
641 #if defined(BOOTP) || defined(NFS_ROOT)
644 kprintf("nfs_mountroot: mount %s on %s: %d", path, which, error);
645 mp->mnt_vfc->vfc_refcount--;
657 nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp)
664 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
665 * no sense in that context.
667 if (argp->sotype == SOCK_STREAM)
668 nmp->nm_flag &= ~NFSMNT_NOCONN;
670 /* Also clear RDIRPLUS if not NFSv3, it crashes some servers */
671 if ((argp->flags & NFSMNT_NFSV3) == 0)
672 nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
674 /* Re-bind if rsrvd port requested and wasn't on one */
675 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
676 && (argp->flags & NFSMNT_RESVPORT);
677 /* Also re-bind if we're switching to/from a connected UDP socket */
678 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
679 (argp->flags & NFSMNT_NOCONN));
681 /* Update flags atomically. Don't change the lock bits. */
682 nmp->nm_flag = argp->flags | nmp->nm_flag;
685 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
686 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
687 if (nmp->nm_timeo < NFS_MINTIMEO)
688 nmp->nm_timeo = NFS_MINTIMEO;
689 else if (nmp->nm_timeo > NFS_MAXTIMEO)
690 nmp->nm_timeo = NFS_MAXTIMEO;
693 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
694 nmp->nm_retry = argp->retrans;
695 if (nmp->nm_retry > NFS_MAXREXMIT)
696 nmp->nm_retry = NFS_MAXREXMIT;
699 maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
701 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
702 nmp->nm_wsize = argp->wsize;
703 /* Round down to multiple of blocksize */
704 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
705 if (nmp->nm_wsize <= 0)
706 nmp->nm_wsize = NFS_FABLKSIZE;
708 if (nmp->nm_wsize > maxio)
709 nmp->nm_wsize = maxio;
710 if (nmp->nm_wsize > MAXBSIZE)
711 nmp->nm_wsize = MAXBSIZE;
713 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
714 nmp->nm_rsize = argp->rsize;
715 /* Round down to multiple of blocksize */
716 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
717 if (nmp->nm_rsize <= 0)
718 nmp->nm_rsize = NFS_FABLKSIZE;
720 if (nmp->nm_rsize > maxio)
721 nmp->nm_rsize = maxio;
722 if (nmp->nm_rsize > MAXBSIZE)
723 nmp->nm_rsize = MAXBSIZE;
725 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
726 nmp->nm_readdirsize = argp->readdirsize;
728 if (nmp->nm_readdirsize > maxio)
729 nmp->nm_readdirsize = maxio;
730 if (nmp->nm_readdirsize > nmp->nm_rsize)
731 nmp->nm_readdirsize = nmp->nm_rsize;
733 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
734 nmp->nm_acregmin = argp->acregmin;
736 nmp->nm_acregmin = NFS_MINATTRTIMO;
737 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
738 nmp->nm_acregmax = argp->acregmax;
740 nmp->nm_acregmax = NFS_MAXATTRTIMO;
741 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
742 nmp->nm_acdirmin = argp->acdirmin;
744 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
745 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
746 nmp->nm_acdirmax = argp->acdirmax;
748 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
749 if (nmp->nm_acdirmin > nmp->nm_acdirmax)
750 nmp->nm_acdirmin = nmp->nm_acdirmax;
751 if (nmp->nm_acregmin > nmp->nm_acregmax)
752 nmp->nm_acregmin = nmp->nm_acregmax;
754 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) {
755 if (argp->maxgrouplist <= NFS_MAXGRPS)
756 nmp->nm_numgrps = argp->maxgrouplist;
758 nmp->nm_numgrps = NFS_MAXGRPS;
760 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
761 if (argp->readahead <= NFS_MAXRAHEAD)
762 nmp->nm_readahead = argp->readahead;
764 nmp->nm_readahead = NFS_MAXRAHEAD;
766 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) {
767 if (argp->deadthresh <= NFS_NEVERDEAD)
768 nmp->nm_deadthresh = argp->deadthresh;
770 nmp->nm_deadthresh = NFS_NEVERDEAD;
773 adjsock |= ((nmp->nm_sotype != argp->sotype) ||
774 (nmp->nm_soproto != argp->proto));
775 nmp->nm_sotype = argp->sotype;
776 nmp->nm_soproto = argp->proto;
778 if (nmp->nm_so && adjsock) {
779 nfs_safedisconnect(nmp);
780 if (nmp->nm_sotype == SOCK_DGRAM)
781 while (nfs_connect(nmp, NULL)) {
782 kprintf("nfs_args: retrying connect\n");
783 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
792 * It seems a bit dumb to copyinstr() the host and path here and then
793 * bcopy() them in mountnfs(), but I wanted to detect errors before
794 * doing the sockargs() call because sockargs() allocates an mbuf and
795 * an error after that means that I have to release the mbuf.
799 nfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred)
802 struct nfs_args args;
803 struct sockaddr *nam;
805 char pth[MNAMELEN], hst[MNAMELEN];
807 u_char nfh[NFSX_V3FHMAX];
813 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
816 if (args.version != NFS_ARGSVERSION) {
817 #ifdef COMPAT_PRELITE2
819 * If the argument version is unknown, then assume the
820 * caller is a pre-lite2 4.4BSD client and convert its
823 struct onfs_args oargs;
824 error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args));
827 nfs_convert_oargs(&args,&oargs);
828 #else /* !COMPAT_PRELITE2 */
829 return (EPROGMISMATCH);
830 #endif /* COMPAT_PRELITE2 */
832 if (mp->mnt_flag & MNT_UPDATE) {
833 struct nfsmount *nmp = VFSTONFS(mp);
838 * When doing an update, we can't change from or to
839 * v3, or change cookie translation
841 args.flags = (args.flags &
842 ~(NFSMNT_NFSV3/*|NFSMNT_XLATECOOKIE*/)) |
844 (NFSMNT_NFSV3/*|NFSMNT_XLATECOOKIE*/));
845 nfs_decode_args(nmp, &args);
850 * Make the nfs_ip_paranoia sysctl serve as the default connection
851 * or no-connection mode for those protocols that support
852 * no-connection mode (the flag will be cleared later for protocols
853 * that do not support no-connection mode). This will allow a client
854 * to receive replies from a different IP then the request was
855 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid),
858 if (nfs_ip_paranoia == 0)
859 args.flags |= NFSMNT_NOCONN;
860 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
862 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
865 error = copyinstr(path, pth, MNAMELEN-1, &len);
868 bzero(&pth[len], MNAMELEN - len);
869 error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
872 bzero(&hst[len], MNAMELEN - len);
873 /* sockargs() call must be after above copyin() calls */
874 error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen);
878 error = mountnfs(&args, mp, nam, pth, hst, &vp);
883 * Common code for mount and mountroot
886 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
887 char *pth, char *hst, struct vnode **vpp)
889 struct nfsmount *nmp;
893 if (mp->mnt_flag & MNT_UPDATE) {
895 /* update paths, file handles, etc, here XXX */
899 nmp = zalloc(nfsmount_zone);
900 bzero((caddr_t)nmp, sizeof (struct nfsmount));
901 mtx_init(&nmp->nm_rxlock);
902 mtx_init(&nmp->nm_txlock);
903 TAILQ_INIT(&nmp->nm_uidlruhead);
904 TAILQ_INIT(&nmp->nm_bioq);
905 TAILQ_INIT(&nmp->nm_reqq);
906 mp->mnt_data = (qaddr_t)nmp;
912 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too
913 * high, depending on whether we end up with negative offsets in
914 * the client or server somewhere. 2GB-1 may be safer.
916 * For V3, nfs_fsinfo will adjust this as necessary. Assume maximum
917 * that we can handle until we find out otherwise.
918 * XXX Our "safe" limit on the client is what we can store in our
919 * buffer cache using signed(!) block numbers.
921 if ((argp->flags & NFSMNT_NFSV3) == 0)
922 nmp->nm_maxfilesize = 0xffffffffLL;
924 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
926 nmp->nm_timeo = NFS_TIMEO;
927 nmp->nm_retry = NFS_RETRANS;
928 nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
929 nmp->nm_rsize = nmp->nm_wsize;
930 nmp->nm_readdirsize = NFS_READDIRSIZE;
931 nmp->nm_numgrps = NFS_MAXGRPS;
932 nmp->nm_readahead = NFS_DEFRAHEAD;
933 nmp->nm_deadthresh = NFS_DEADTHRESH;
934 nmp->nm_fhsize = argp->fhsize;
935 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
936 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
938 /* Set up the sockets and per-host congestion */
939 nmp->nm_sotype = argp->sotype;
940 nmp->nm_soproto = argp->proto;
941 nmp->nm_cred = crhold(proc0.p_ucred);
943 nfs_decode_args(nmp, argp);
946 * For Connection based sockets (TCP,...) defer the connect until
947 * the first request, in case the server is not responding.
949 if (nmp->nm_sotype == SOCK_DGRAM &&
950 (error = nfs_connect(nmp, NULL)))
954 * This is silly, but it has to be set so that vinifod() works.
955 * We do not want to do an nfs_statfs() here since we can get
956 * stuck on a dead server and we are holding a lock on the mount
959 mp->mnt_stat.f_iosize =
960 nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype);
963 * Install vop_ops for our vnops
965 vfs_add_vnodeops(mp, &nfsv2_vnode_vops, &mp->mnt_vn_norm_ops);
966 vfs_add_vnodeops(mp, &nfsv2_spec_vops, &mp->mnt_vn_spec_ops);
967 vfs_add_vnodeops(mp, &nfsv2_fifo_vops, &mp->mnt_vn_fifo_ops);
970 * A reference count is needed on the nfsnode representing the
971 * remote root. If this object is not persistent, then backward
972 * traversals of the mount point (i.e. "..") will not work if
973 * the nfsnode gets flushed out of the cache. Ufs does not have
974 * this problem, because one can identify root inodes by their
975 * number == ROOTINO (2).
977 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
983 * Retrieval of mountpoint attributes is delayed until nfs_rot
984 * or nfs_statfs are first called. This will happen either when
985 * we first traverse the mount point or if somebody does a df(1).
987 * NFSSTA_GOTFSINFO is used to flag if we have successfully
988 * retrieved mountpoint attributes. In the case of NFSv3 we
989 * also flag static fsinfo.
992 (*vpp)->v_type = VNON;
995 * Lose the lock but keep the ref.
998 TAILQ_INSERT_TAIL(&nfs_mountq, nmp, nm_entry);
1001 * Start the reader and writer threads.
1003 lwkt_create(nfssvc_iod_reader, nmp, &nmp->nm_rxthread,
1004 NULL, 0, -1, "nfsiod_rx");
1005 lwkt_create(nfssvc_iod_writer, nmp, &nmp->nm_txthread,
1006 NULL, 0, -1, "nfsiod_tx");
1010 nfs_disconnect(nmp);
1011 nfs_free_mount(nmp);
1012 FREE(nam, M_SONAME);
1017 * unmount system call
1020 nfs_unmount(struct mount *mp, int mntflags)
1022 struct nfsmount *nmp;
1023 int error, flags = 0;
1025 if (mntflags & MNT_FORCE)
1026 flags |= FORCECLOSE;
1029 * Goes something like this..
1030 * - Call vflush() to clear out vnodes for this file system
1031 * - Close the socket
1032 * - Free up the data structures
1034 /* In the forced case, cancel any outstanding requests. */
1035 if (flags & FORCECLOSE) {
1036 error = nfs_nmcancelreqs(nmp);
1041 * Must handshake with nfs_clientd() if it is active. XXX
1043 nmp->nm_state |= NFSSTA_DISMINPROG;
1045 /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */
1046 error = vflush(mp, 1, flags);
1048 nmp->nm_state &= ~NFSSTA_DISMINPROG;
1053 * We are now committed to the unmount.
1054 * For NQNFS, let the server daemon free the nfsmount structure.
1056 if (nmp->nm_flag & NFSMNT_KERB)
1057 nmp->nm_state |= NFSSTA_DISMNT;
1059 nfs_disconnect(nmp);
1060 FREE(nmp->nm_nam, M_SONAME);
1061 TAILQ_REMOVE(&nfs_mountq, nmp, nm_entry);
1063 if ((nmp->nm_flag & NFSMNT_KERB) == 0) {
1064 nfssvc_iod_stop(nmp);
1065 nfs_free_mount(nmp);
1071 nfs_free_mount(struct nfsmount *nmp)
1074 crfree(nmp->nm_cred);
1075 nmp->nm_cred = NULL;
1077 zfree(nfsmount_zone, nmp);
1081 * Return root of a filesystem
1084 nfs_root(struct mount *mp, struct vnode **vpp)
1087 struct nfsmount *nmp;
1093 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
1099 * Get transfer parameters and root vnode attributes
1101 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
1102 if (nmp->nm_flag & NFSMNT_NFSV3) {
1103 error = nfs_fsinfo(nmp, vp, curthread);
1104 mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype);
1106 if ((error = VOP_GETATTR(vp, &attrs)) == 0)
1107 nmp->nm_state |= NFSSTA_GOTFSINFO;
1112 * The root vnode is usually cached by the namecache so do not
1113 * try to avoid going over the wire even if we have previous
1114 * information cached. A stale NFS mount can loop
1115 * forever resolving the root vnode if we return no-error when
1116 * there is in fact an error.
1118 np->n_attrstamp = 0;
1119 error = VOP_GETATTR(vp, &attrs);
1121 if (vp->v_type == VNON)
1122 nfs_setvtype(vp, VDIR);
1139 static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
1140 static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
1143 * Flush out the buffer cache
1147 nfs_sync(struct mount *mp, int waitfor)
1149 struct scaninfo scaninfo;
1152 scaninfo.rescan = 1;
1153 scaninfo.waitfor = waitfor;
1154 scaninfo.allerror = 0;
1157 * Force stale buffer cache information to be flushed.
1160 while (error == 0 && scaninfo.rescan) {
1161 scaninfo.rescan = 0;
1162 error = vmntvnodescan(mp, VMSC_GETVP, nfs_sync_scan1,
1163 nfs_sync_scan2, &scaninfo);
1169 nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1171 struct scaninfo *info = data;
1173 if (vn_islocked(vp) || RB_EMPTY(&vp->v_rbdirty_tree))
1175 if (info->waitfor == MNT_LAZY)
1181 nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1183 struct scaninfo *info = data;
1186 error = VOP_FSYNC(vp, info->waitfor);
1188 info->allerror = error;