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);
82 extern struct vop_ops nfsv2_vnode_vops;
83 extern struct vop_ops nfsv2_fifo_vops;
84 extern struct vop_ops nfsv2_spec_vops;
86 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header");
87 MALLOC_DEFINE(M_NFSBIGFH, "NFSV3 bigfh", "NFS version 3 file handle");
88 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure");
89 MALLOC_DEFINE(M_NFSDIROFF, "NFSV3 diroff", "NFS directory offset data");
90 MALLOC_DEFINE(M_NFSRVDESC, "NFSV3 srvdesc", "NFS server socket descriptor");
91 MALLOC_DEFINE(M_NFSUID, "NFS uid", "Nfs uid mapping structure");
92 MALLOC_DEFINE(M_NFSHASH, "NFS hash", "NFS hash tables");
94 vm_zone_t nfsmount_zone;
96 struct nfsstats nfsstats;
97 SYSCTL_NODE(_vfs, OID_AUTO, nfs, CTLFLAG_RW, 0, "NFS filesystem");
98 SYSCTL_STRUCT(_vfs_nfs, NFS_NFSSTATS, nfsstats, CTLFLAG_RD,
99 &nfsstats, nfsstats, "");
100 static int nfs_ip_paranoia = 1;
101 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW,
102 &nfs_ip_paranoia, 0, "");
105 SYSCTL_INT(_vfs_nfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0, "");
109 * Tunable to determine the Read/Write unit size. Maximum value
110 * is NFS_MAXDATA. We also default to NFS_MAXDATA.
112 static int nfs_io_size = NFS_MAXDATA;
113 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_io_size, CTLFLAG_RW,
114 &nfs_io_size, 0, "NFS optimal I/O unit size");
116 static void nfs_decode_args (struct nfsmount *nmp,
117 struct nfs_args *argp);
118 static int mountnfs (struct nfs_args *,struct mount *,
119 struct sockaddr *,char *,char *,struct vnode **);
120 static int nfs_mount ( struct mount *mp, char *path, caddr_t data,
122 static int nfs_unmount ( struct mount *mp, int mntflags);
123 static int nfs_root ( struct mount *mp, struct vnode **vpp);
124 static int nfs_statfs ( struct mount *mp, struct statfs *sbp,
126 static int nfs_sync ( struct mount *mp, int waitfor);
129 * nfs vfs operations.
131 static struct vfsops nfs_vfsops = {
132 .vfs_mount = nfs_mount,
133 .vfs_unmount = nfs_unmount,
134 .vfs_root = nfs_root,
135 .vfs_statfs = nfs_statfs,
136 .vfs_sync = nfs_sync,
137 .vfs_init = nfs_init,
138 .vfs_uninit = nfs_uninit
140 VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK);
143 * This structure must be filled in by a primary bootstrap or bootstrap
144 * server for a diskless/dataless machine. It is initialized below just
145 * to ensure that it is allocated to initialized data (.data not .bss).
147 struct nfs_diskless nfs_diskless = { { { 0 } } };
148 struct nfsv3_diskless nfsv3_diskless = { { { 0 } } };
149 int nfs_diskless_valid = 0;
151 SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD,
152 &nfs_diskless_valid, 0, "");
154 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD,
155 nfsv3_diskless.root_hostnam, 0, "");
157 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD,
158 &nfsv3_diskless.root_saddr, sizeof nfsv3_diskless.root_saddr,
159 "%Ssockaddr_in", "");
161 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_swappath, CTLFLAG_RD,
162 nfsv3_diskless.swap_hostnam, 0, "");
164 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_swapaddr, CTLFLAG_RD,
165 &nfsv3_diskless.swap_saddr, sizeof nfsv3_diskless.swap_saddr,
169 void nfsargs_ntoh (struct nfs_args *);
170 static int nfs_mountdiskless (char *, char *, int,
171 struct sockaddr_in *, struct nfs_args *,
172 struct thread *, struct vnode **,
174 static void nfs_convert_diskless (void);
175 static void nfs_convert_oargs (struct nfs_args *args,
176 struct onfs_args *oargs);
179 * Calculate the buffer I/O block size to use. The maximum V2 block size
180 * is typically 8K, the maximum datagram size is typically 16K, and the
181 * maximum V3 block size is typically 32K. The buffer cache tends to work
182 * best with 16K blocks but we allow 32K for TCP connections.
184 * We force the block size to be at least a page for buffer cache efficiency.
187 nfs_iosize(int v3, int sotype)
193 if (sotype == SOCK_STREAM)
196 iomax = NFS_MAXDGRAMDATA;
198 iomax = NFS_V2MAXDATA;
200 if ((iosize = nfs_io_size) > iomax)
202 if (iosize < PAGE_SIZE)
206 * This is an aweful hack but until the buffer cache is rewritten
207 * we need it. The problem is that when you combine write() with
208 * mmap() the vm_page->valid bits can become weird looking
209 * (e.g. 0xfc). This occurs because NFS uses piecemeal buffers
210 * at the file EOF. To solve the problem the BIO system needs to
211 * be guarenteed that the NFS iosize for regular files will be a
212 * multiple of PAGE_SIZE so it can invalidate the whole page
213 * rather then just the piece of it owned by the buffer when
214 * NFS does vinvalbuf() calls.
216 if (iosize & PAGE_MASK)
217 iosize = (iosize & ~PAGE_MASK) + PAGE_SIZE;
222 nfs_convert_oargs(struct nfs_args *args, struct onfs_args *oargs)
224 args->version = NFS_ARGSVERSION;
225 args->addr = oargs->addr;
226 args->addrlen = oargs->addrlen;
227 args->sotype = oargs->sotype;
228 args->proto = oargs->proto;
229 args->fh = oargs->fh;
230 args->fhsize = oargs->fhsize;
231 args->flags = oargs->flags;
232 args->wsize = oargs->wsize;
233 args->rsize = oargs->rsize;
234 args->readdirsize = oargs->readdirsize;
235 args->timeo = oargs->timeo;
236 args->retrans = oargs->retrans;
237 args->maxgrouplist = oargs->maxgrouplist;
238 args->readahead = oargs->readahead;
239 args->deadthresh = oargs->deadthresh;
240 args->hostname = oargs->hostname;
244 nfs_convert_diskless(void)
248 bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif,
249 sizeof(struct ifaliasreq));
250 bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway,
251 sizeof(struct sockaddr_in));
252 nfs_convert_oargs(&nfsv3_diskless.swap_args,&nfs_diskless.swap_args);
254 bcopy(nfs_diskless.swap_fh,nfsv3_diskless.swap_fh,NFSX_V2FH);
255 nfsv3_diskless.swap_fhsize = NFSX_V2FH;
256 for (i = NFSX_V2FH - 1; i >= 0; --i) {
257 if (nfs_diskless.swap_fh[i])
261 nfsv3_diskless.swap_fhsize = 0;
263 bcopy(&nfs_diskless.swap_saddr,&nfsv3_diskless.swap_saddr,
264 sizeof(struct sockaddr_in));
265 bcopy(nfs_diskless.swap_hostnam,nfsv3_diskless.swap_hostnam, MNAMELEN);
266 nfsv3_diskless.swap_nblks = nfs_diskless.swap_nblks;
267 bcopy(&nfs_diskless.swap_ucred, &nfsv3_diskless.swap_ucred,
268 sizeof(struct ucred));
269 nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args);
271 bcopy(nfs_diskless.root_fh,nfsv3_diskless.root_fh,NFSX_V2FH);
272 nfsv3_diskless.root_fhsize = NFSX_V2FH;
273 for (i = NFSX_V2FH - 1; i >= 0; --i) {
274 if (nfs_diskless.root_fh[i])
278 nfsv3_diskless.root_fhsize = 0;
280 bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr,
281 sizeof(struct sockaddr_in));
282 bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN);
283 nfsv3_diskless.root_time = nfs_diskless.root_time;
284 bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam,
286 nfs_diskless_valid = 3;
293 nfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
296 struct nfs_statfs *sfp;
297 struct nfsmount *nmp = VFSTONFS(mp);
298 thread_t td = curthread;
299 int error = 0, retattr;
302 struct nfsm_info info;
305 info.v3 = (nmp->nm_flag & NFSMNT_NFSV3);
310 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
314 /* ignore the passed cred */
316 cred->cr_ngroups = 1;
317 if (info.v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
318 (void)nfs_fsinfo(nmp, vp, td);
319 nfsstats.rpccnt[NFSPROC_FSSTAT]++;
320 nfsm_reqhead(&info, vp, NFSPROC_FSSTAT, NFSX_FH(info.v3));
321 ERROROUT(nfsm_fhtom(&info, vp));
322 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSSTAT, td, cred, &error));
324 ERROROUT(nfsm_postop_attr(&info, vp, &retattr,
325 NFS_LATTR_NOSHRINK));
328 if (info.mrep != NULL)
332 NULLOUT(sfp = nfsm_dissect(&info, NFSX_STATFS(info.v3)));
333 sbp->f_flags = nmp->nm_flag;
334 sbp->f_iosize = nfs_iosize(info.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);
369 * nfs version 3 fsinfo rpc call
372 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td)
374 struct nfsv3_fsinfo *fsp;
376 int error = 0, retattr;
378 struct nfsm_info info;
381 nfsstats.rpccnt[NFSPROC_FSINFO]++;
382 nfsm_reqhead(&info, vp, NFSPROC_FSINFO, NFSX_FH(1));
383 ERROROUT(nfsm_fhtom(&info, vp));
384 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSINFO, td,
385 nfs_vpcred(vp, ND_READ), &error));
386 ERROROUT(nfsm_postop_attr(&info, vp, &retattr, NFS_LATTR_NOSHRINK));
388 NULLOUT(fsp = nfsm_dissect(&info, 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;
430 * Mount a remote root fs via. nfs. This depends on the info in the
431 * nfs_diskless structure that has been filled in properly by some primary
433 * It goes something like this:
434 * - do enough of "ifconfig" by calling ifioctl() so that the system
435 * can talk to the server
436 * - If nfs_diskless.mygateway is filled in, use that address as
438 * - build the rootfs mount point and call mountnfs() to do the rest.
441 nfs_mountroot(struct mount *mp)
443 struct mount *swap_mp;
444 struct nfsv3_diskless *nd = &nfsv3_diskless;
447 struct thread *td = curthread; /* XXX */
452 #if defined(BOOTP_NFSROOT) && defined(BOOTP)
453 bootpc_init(); /* use bootp to get nfs_diskless filled in */
457 * XXX time must be non-zero when we init the interface or else
458 * the arp code will wedge...
460 while (mycpu->gd_time_seconds == 0)
461 tsleep(mycpu, 0, "arpkludge", 10);
464 * The boot code may have passed us a diskless structure.
466 if (nfs_diskless_valid == 1)
467 nfs_convert_diskless();
469 #define SINP(sockaddr) ((struct sockaddr_in *)(sockaddr))
470 kprintf("nfs_mountroot: interface %s ip %s",
472 inet_ntoa(SINP(&nd->myif.ifra_addr)->sin_addr));
474 inet_ntoa(SINP(&nd->myif.ifra_broadaddr)->sin_addr));
475 kprintf(" mask %s\n",
476 inet_ntoa(SINP(&nd->myif.ifra_mask)->sin_addr));
480 * XXX splnet, so networks will receive...
485 * BOOTP does not necessarily have to be compiled into the kernel
486 * for an NFS root to work. If we inherited the network
487 * configuration for PXEBOOT then pxe_setup_nfsdiskless() has figured
488 * out our interface for us and all we need to do is ifconfig the
489 * interface. We only do this if the interface has not already been
490 * ifconfig'd by e.g. BOOTP.
492 error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td);
494 panic("nfs_mountroot: socreate(%04x): %d",
495 nd->myif.ifra_addr.sa_family, error);
498 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, proc0.p_ucred);
500 panic("nfs_mountroot: SIOCAIFADDR: %d", error);
502 soclose(so, FNONBLOCK);
505 * If the gateway field is filled in, set it as the default route.
507 if (nd->mygateway.sin_len != 0) {
508 struct sockaddr_in mask, sin;
510 bzero((caddr_t)&mask, sizeof(mask));
512 sin.sin_family = AF_INET;
513 sin.sin_len = sizeof(sin);
514 kprintf("nfs_mountroot: gateway %s\n",
515 inet_ntoa(nd->mygateway.sin_addr));
516 error = rtrequest_global(RTM_ADD, (struct sockaddr *)&sin,
517 (struct sockaddr *)&nd->mygateway,
518 (struct sockaddr *)&mask,
519 RTF_UP | RTF_GATEWAY);
521 kprintf("nfs_mountroot: unable to set gateway, error %d, continuing anyway\n", error);
525 * Create the rootfs mount point.
527 nd->root_args.fh = nd->root_fh;
528 nd->root_args.fhsize = nd->root_fhsize;
529 l = ntohl(nd->root_saddr.sin_addr.s_addr);
530 ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
531 (l >> 24) & 0xff, (l >> 16) & 0xff,
532 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->root_hostnam);
533 kprintf("NFS_ROOT: %s\n",buf);
534 if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY,
535 &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) {
536 mp->mnt_vfc->vfc_refcount--;
542 if (nd->swap_nblks) {
544 /* Convert to DEV_BSIZE instead of Kilobyte */
548 * Create a fake mount point just for the swap vnode so that the
549 * swap file can be on a different server from the rootfs.
551 nd->swap_args.fh = nd->swap_fh;
552 nd->swap_args.fhsize = nd->swap_fhsize;
553 l = ntohl(nd->swap_saddr.sin_addr.s_addr);
554 ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
555 (l >> 24) & 0xff, (l >> 16) & 0xff,
556 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->swap_hostnam);
557 kprintf("NFS SWAP: %s\n",buf);
558 if ((error = nfs_mountdiskless(buf, "/swap", 0,
559 &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0) {
565 VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size =
566 nd->swap_nblks * DEV_BSIZE ;
569 * Since the swap file is not the root dir of a file system,
570 * hack it to a regular file.
574 nfs_setvtype(vp, VREG);
575 swaponvp(td, vp, nd->swap_nblks);
578 mp->mnt_flag |= MNT_ROOTFS;
582 * This is not really an nfs issue, but it is much easier to
583 * set hostname here and then let the "/etc/rc.xxx" files
584 * mount the right /var based upon its preset value.
586 bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN);
587 hostname[MAXHOSTNAMELEN - 1] = '\0';
588 for (i = 0; i < MAXHOSTNAMELEN; i++)
589 if (hostname[i] == '\0')
591 inittodr(ntohl(nd->root_time));
597 * Internal version of mount system call for diskless setup.
600 nfs_mountdiskless(char *path, char *which, int mountflag,
601 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
602 struct vnode **vpp, struct mount **mpp)
605 struct sockaddr *nam;
612 if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) {
613 kprintf("nfs_mountroot: NFS not configured");
618 mp->mnt_kern_flag = 0;
619 mp->mnt_flag = mountflag;
620 nam = dup_sockaddr((struct sockaddr *)sin);
622 #if defined(BOOTP) || defined(NFS_ROOT)
623 if (args->fhsize == 0) {
626 kprintf("NFS_ROOT: No FH passed from loader, attempting mount rpc...");
627 while (*xpath && *xpath != ':')
632 error = md_mount(sin, xpath, args->fh, &args->fhsize, args, td);
634 kprintf("failed error %d.\n", error);
637 kprintf("success!\n");
641 if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) {
642 #if defined(BOOTP) || defined(NFS_ROOT)
645 kprintf("nfs_mountroot: mount %s on %s: %d", path, which, error);
646 mp->mnt_vfc->vfc_refcount--;
658 nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp)
665 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
666 * no sense in that context.
668 if (argp->sotype == SOCK_STREAM)
669 nmp->nm_flag &= ~NFSMNT_NOCONN;
671 /* Also clear RDIRPLUS if not NFSv3, it crashes some servers */
672 if ((argp->flags & NFSMNT_NFSV3) == 0)
673 nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
675 /* Re-bind if rsrvd port requested and wasn't on one */
676 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
677 && (argp->flags & NFSMNT_RESVPORT);
678 /* Also re-bind if we're switching to/from a connected UDP socket */
679 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
680 (argp->flags & NFSMNT_NOCONN));
682 /* Update flags atomically. Don't change the lock bits. */
683 nmp->nm_flag = argp->flags | nmp->nm_flag;
686 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
687 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
688 if (nmp->nm_timeo < NFS_MINTIMEO)
689 nmp->nm_timeo = NFS_MINTIMEO;
690 else if (nmp->nm_timeo > NFS_MAXTIMEO)
691 nmp->nm_timeo = NFS_MAXTIMEO;
694 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
695 nmp->nm_retry = argp->retrans;
696 if (nmp->nm_retry > NFS_MAXREXMIT)
697 nmp->nm_retry = NFS_MAXREXMIT;
700 maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
702 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
703 nmp->nm_wsize = argp->wsize;
704 /* Round down to multiple of blocksize */
705 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
706 if (nmp->nm_wsize <= 0)
707 nmp->nm_wsize = NFS_FABLKSIZE;
709 if (nmp->nm_wsize > maxio)
710 nmp->nm_wsize = maxio;
711 if (nmp->nm_wsize > MAXBSIZE)
712 nmp->nm_wsize = MAXBSIZE;
714 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
715 nmp->nm_rsize = argp->rsize;
716 /* Round down to multiple of blocksize */
717 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
718 if (nmp->nm_rsize <= 0)
719 nmp->nm_rsize = NFS_FABLKSIZE;
721 if (nmp->nm_rsize > maxio)
722 nmp->nm_rsize = maxio;
723 if (nmp->nm_rsize > MAXBSIZE)
724 nmp->nm_rsize = MAXBSIZE;
726 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
727 nmp->nm_readdirsize = argp->readdirsize;
729 if (nmp->nm_readdirsize > maxio)
730 nmp->nm_readdirsize = maxio;
731 if (nmp->nm_readdirsize > nmp->nm_rsize)
732 nmp->nm_readdirsize = nmp->nm_rsize;
734 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
735 nmp->nm_acregmin = argp->acregmin;
737 nmp->nm_acregmin = NFS_MINATTRTIMO;
738 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
739 nmp->nm_acregmax = argp->acregmax;
741 nmp->nm_acregmax = NFS_MAXATTRTIMO;
742 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
743 nmp->nm_acdirmin = argp->acdirmin;
745 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
746 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
747 nmp->nm_acdirmax = argp->acdirmax;
749 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
750 if (nmp->nm_acdirmin > nmp->nm_acdirmax)
751 nmp->nm_acdirmin = nmp->nm_acdirmax;
752 if (nmp->nm_acregmin > nmp->nm_acregmax)
753 nmp->nm_acregmin = nmp->nm_acregmax;
755 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) {
756 if (argp->maxgrouplist <= NFS_MAXGRPS)
757 nmp->nm_numgrps = argp->maxgrouplist;
759 nmp->nm_numgrps = NFS_MAXGRPS;
761 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
762 if (argp->readahead <= NFS_MAXRAHEAD)
763 nmp->nm_readahead = argp->readahead;
765 nmp->nm_readahead = NFS_MAXRAHEAD;
767 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) {
768 if (argp->deadthresh <= NFS_NEVERDEAD)
769 nmp->nm_deadthresh = argp->deadthresh;
771 nmp->nm_deadthresh = NFS_NEVERDEAD;
774 adjsock |= ((nmp->nm_sotype != argp->sotype) ||
775 (nmp->nm_soproto != argp->proto));
776 nmp->nm_sotype = argp->sotype;
777 nmp->nm_soproto = argp->proto;
779 if (nmp->nm_so && adjsock) {
780 nfs_safedisconnect(nmp);
781 if (nmp->nm_sotype == SOCK_DGRAM)
782 while (nfs_connect(nmp, NULL)) {
783 kprintf("nfs_args: retrying connect\n");
784 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
793 * It seems a bit dumb to copyinstr() the host and path here and then
794 * bcopy() them in mountnfs(), but I wanted to detect errors before
795 * doing the sockargs() call because sockargs() allocates an mbuf and
796 * an error after that means that I have to release the mbuf.
800 nfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred)
803 struct nfs_args args;
804 struct sockaddr *nam;
806 char pth[MNAMELEN], hst[MNAMELEN];
808 u_char nfh[NFSX_V3FHMAX];
814 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
817 if (args.version != NFS_ARGSVERSION) {
818 #ifdef COMPAT_PRELITE2
820 * If the argument version is unknown, then assume the
821 * caller is a pre-lite2 4.4BSD client and convert its
824 struct onfs_args oargs;
825 error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args));
828 nfs_convert_oargs(&args,&oargs);
829 #else /* !COMPAT_PRELITE2 */
830 return (EPROGMISMATCH);
831 #endif /* COMPAT_PRELITE2 */
833 if (mp->mnt_flag & MNT_UPDATE) {
834 struct nfsmount *nmp = VFSTONFS(mp);
839 * When doing an update, we can't change from or to
840 * v3, or change cookie translation
842 args.flags = (args.flags &
843 ~(NFSMNT_NFSV3/*|NFSMNT_XLATECOOKIE*/)) |
845 (NFSMNT_NFSV3/*|NFSMNT_XLATECOOKIE*/));
846 nfs_decode_args(nmp, &args);
851 * Make the nfs_ip_paranoia sysctl serve as the default connection
852 * or no-connection mode for those protocols that support
853 * no-connection mode (the flag will be cleared later for protocols
854 * that do not support no-connection mode). This will allow a client
855 * to receive replies from a different IP then the request was
856 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid),
859 if (nfs_ip_paranoia == 0)
860 args.flags |= NFSMNT_NOCONN;
861 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
863 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
866 error = copyinstr(path, pth, MNAMELEN-1, &len);
869 bzero(&pth[len], MNAMELEN - len);
870 error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
873 bzero(&hst[len], MNAMELEN - len);
874 /* sockargs() call must be after above copyin() calls */
875 error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen);
879 error = mountnfs(&args, mp, nam, pth, hst, &vp);
884 * Common code for mount and mountroot
887 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
888 char *pth, char *hst, struct vnode **vpp)
890 struct nfsmount *nmp;
896 if (mp->mnt_flag & MNT_UPDATE) {
898 /* update paths, file handles, etc, here XXX */
902 nmp = zalloc(nfsmount_zone);
903 bzero((caddr_t)nmp, sizeof (struct nfsmount));
904 mtx_init(&nmp->nm_rxlock);
905 mtx_init(&nmp->nm_txlock);
906 TAILQ_INIT(&nmp->nm_uidlruhead);
907 TAILQ_INIT(&nmp->nm_bioq);
908 TAILQ_INIT(&nmp->nm_reqq);
909 TAILQ_INIT(&nmp->nm_reqtxq);
910 TAILQ_INIT(&nmp->nm_reqrxq);
911 mp->mnt_data = (qaddr_t)nmp;
917 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too
918 * high, depending on whether we end up with negative offsets in
919 * the client or server somewhere. 2GB-1 may be safer.
921 * For V3, nfs_fsinfo will adjust this as necessary. Assume maximum
922 * that we can handle until we find out otherwise.
923 * XXX Our "safe" limit on the client is what we can store in our
924 * buffer cache using signed(!) block numbers.
926 if ((argp->flags & NFSMNT_NFSV3) == 0)
927 nmp->nm_maxfilesize = 0xffffffffLL;
929 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
931 nmp->nm_timeo = NFS_TIMEO;
932 nmp->nm_retry = NFS_RETRANS;
933 nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
934 nmp->nm_rsize = nmp->nm_wsize;
935 nmp->nm_readdirsize = NFS_READDIRSIZE;
936 nmp->nm_numgrps = NFS_MAXGRPS;
937 nmp->nm_readahead = NFS_DEFRAHEAD;
938 nmp->nm_deadthresh = NFS_DEADTHRESH;
939 nmp->nm_fhsize = argp->fhsize;
940 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
941 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
943 /* Set up the sockets and per-host congestion */
944 nmp->nm_sotype = argp->sotype;
945 nmp->nm_soproto = argp->proto;
946 nmp->nm_cred = crhold(proc0.p_ucred);
948 nfs_decode_args(nmp, argp);
951 * For Connection based sockets (TCP,...) defer the connect until
952 * the first request, in case the server is not responding.
954 if (nmp->nm_sotype == SOCK_DGRAM &&
955 (error = nfs_connect(nmp, NULL)))
959 * This is silly, but it has to be set so that vinifod() works.
960 * We do not want to do an nfs_statfs() here since we can get
961 * stuck on a dead server and we are holding a lock on the mount
964 mp->mnt_stat.f_iosize =
965 nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype);
968 * Install vop_ops for our vnops
970 vfs_add_vnodeops(mp, &nfsv2_vnode_vops, &mp->mnt_vn_norm_ops);
971 vfs_add_vnodeops(mp, &nfsv2_spec_vops, &mp->mnt_vn_spec_ops);
972 vfs_add_vnodeops(mp, &nfsv2_fifo_vops, &mp->mnt_vn_fifo_ops);
975 * A reference count is needed on the nfsnode representing the
976 * remote root. If this object is not persistent, then backward
977 * traversals of the mount point (i.e. "..") will not work if
978 * the nfsnode gets flushed out of the cache. Ufs does not have
979 * this problem, because one can identify root inodes by their
980 * number == ROOTINO (2).
982 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
988 * Retrieval of mountpoint attributes is delayed until nfs_rot
989 * or nfs_statfs are first called. This will happen either when
990 * we first traverse the mount point or if somebody does a df(1).
992 * NFSSTA_GOTFSINFO is used to flag if we have successfully
993 * retrieved mountpoint attributes. In the case of NFSv3 we
994 * also flag static fsinfo.
997 (*vpp)->v_type = VNON;
1000 * Lose the lock but keep the ref.
1003 TAILQ_INSERT_TAIL(&nfs_mountq, nmp, nm_entry);
1026 * Start the reader and writer threads.
1028 lwkt_create(nfssvc_iod_reader, nmp, &nmp->nm_rxthread,
1029 NULL, 0, rxcpu, "nfsiod_rx");
1030 lwkt_create(nfssvc_iod_writer, nmp, &nmp->nm_txthread,
1031 NULL, 0, txcpu, "nfsiod_tx");
1035 nfs_disconnect(nmp);
1036 nfs_free_mount(nmp);
1041 * unmount system call
1044 nfs_unmount(struct mount *mp, int mntflags)
1046 struct nfsmount *nmp;
1047 int error, flags = 0;
1049 if (mntflags & MNT_FORCE)
1050 flags |= FORCECLOSE;
1053 * Goes something like this..
1054 * - Call vflush() to clear out vnodes for this file system
1055 * - Close the socket
1056 * - Free up the data structures
1058 /* In the forced case, cancel any outstanding requests. */
1059 if (flags & FORCECLOSE) {
1060 error = nfs_nmcancelreqs(nmp);
1065 * Must handshake with nfs_clientd() if it is active. XXX
1067 nmp->nm_state |= NFSSTA_DISMINPROG;
1069 /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */
1070 error = vflush(mp, 1, flags);
1072 nmp->nm_state &= ~NFSSTA_DISMINPROG;
1077 * We are now committed to the unmount.
1078 * For NQNFS, let the server daemon free the nfsmount structure.
1080 if (nmp->nm_flag & NFSMNT_KERB)
1081 nmp->nm_state |= NFSSTA_DISMNT;
1082 nfssvc_iod_stop1(nmp);
1083 nfs_disconnect(nmp);
1084 nfssvc_iod_stop2(nmp);
1085 TAILQ_REMOVE(&nfs_mountq, nmp, nm_entry);
1087 if ((nmp->nm_flag & NFSMNT_KERB) == 0) {
1088 nfs_free_mount(nmp);
1094 nfs_free_mount(struct nfsmount *nmp)
1097 crfree(nmp->nm_cred);
1098 nmp->nm_cred = NULL;
1101 FREE(nmp->nm_nam, M_SONAME);
1104 zfree(nfsmount_zone, nmp);
1108 * Return root of a filesystem
1111 nfs_root(struct mount *mp, struct vnode **vpp)
1114 struct nfsmount *nmp;
1120 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
1126 * Get transfer parameters and root vnode attributes
1128 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
1129 if (nmp->nm_flag & NFSMNT_NFSV3) {
1130 error = nfs_fsinfo(nmp, vp, curthread);
1131 mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype);
1133 if ((error = VOP_GETATTR(vp, &attrs)) == 0)
1134 nmp->nm_state |= NFSSTA_GOTFSINFO;
1139 * The root vnode is usually cached by the namecache so do not
1140 * try to avoid going over the wire even if we have previous
1141 * information cached. A stale NFS mount can loop
1142 * forever resolving the root vnode if we return no-error when
1143 * there is in fact an error.
1145 np->n_attrstamp = 0;
1146 error = VOP_GETATTR(vp, &attrs);
1148 if (vp->v_type == VNON)
1149 nfs_setvtype(vp, VDIR);
1164 static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
1165 static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
1168 * Flush out the buffer cache
1172 nfs_sync(struct mount *mp, int waitfor)
1174 struct scaninfo scaninfo;
1177 scaninfo.rescan = 1;
1178 scaninfo.waitfor = waitfor;
1179 scaninfo.allerror = 0;
1182 * Force stale buffer cache information to be flushed.
1185 while (error == 0 && scaninfo.rescan) {
1186 scaninfo.rescan = 0;
1187 error = vmntvnodescan(mp, VMSC_GETVP, nfs_sync_scan1,
1188 nfs_sync_scan2, &scaninfo);
1194 nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1196 struct scaninfo *info = data;
1198 if (vn_islocked(vp) || RB_EMPTY(&vp->v_rbdirty_tree))
1200 if (info->waitfor == MNT_LAZY)
1206 nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1208 struct scaninfo *info = data;
1211 error = VOP_FSYNC(vp, info->waitfor);
1213 info->allerror = error;