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>
74 #include "nfsm_subs.h"
75 #include "nfsdiskless.h"
76 #include "nfsmountrpc.h"
78 extern int nfs_mountroot(struct mount *mp);
79 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;
300 caddr_t bpos, dpos, cp2;
301 struct nfsmount *nmp = VFSTONFS(mp);
302 thread_t td = curthread;
303 int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr;
304 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
311 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
315 /* ignore the passed cred */
317 cred->cr_ngroups = 1;
318 if (v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
319 (void)nfs_fsinfo(nmp, vp, td);
320 nfsstats.rpccnt[NFSPROC_FSSTAT]++;
321 nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
323 nfsm_request(vp, NFSPROC_FSSTAT, td, cred);
325 nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK);
331 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
332 sbp->f_flags = nmp->nm_flag;
333 sbp->f_iosize = nfs_iosize(v3, nmp->nm_sotype);
336 sbp->f_bsize = NFS_FABLKSIZE;
337 tquad = fxdr_hyper(&sfp->sf_tbytes);
338 sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
339 tquad = fxdr_hyper(&sfp->sf_fbytes);
340 sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
341 tquad = fxdr_hyper(&sfp->sf_abytes);
342 sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
343 sbp->f_files = (fxdr_unsigned(int32_t,
344 sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
345 sbp->f_ffree = (fxdr_unsigned(int32_t,
346 sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
348 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
349 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
350 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
351 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
355 if (sbp != &mp->mnt_stat) {
356 sbp->f_type = mp->mnt_vfc->vfc_typenum;
357 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
367 * nfs version 3 fsinfo rpc call
370 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td)
372 struct nfsv3_fsinfo *fsp;
375 u_int32_t *tl, pref, max;
376 caddr_t bpos, dpos, cp2;
377 int error = 0, retattr;
378 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
381 nfsstats.rpccnt[NFSPROC_FSINFO]++;
382 nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
384 nfsm_request(vp, NFSPROC_FSINFO, td, nfs_vpcred(vp, ND_READ));
385 nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK);
387 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
388 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
389 if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
390 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
391 ~(NFS_FABLKSIZE - 1);
392 max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
393 if (max < nmp->nm_wsize && max > 0) {
394 nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
395 if (nmp->nm_wsize == 0)
398 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
399 if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
400 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
401 ~(NFS_FABLKSIZE - 1);
402 max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
403 if (max < nmp->nm_rsize && max > 0) {
404 nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
405 if (nmp->nm_rsize == 0)
408 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
409 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ)
410 nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) &
411 ~(NFS_DIRBLKSIZ - 1);
412 if (max < nmp->nm_readdirsize && max > 0) {
413 nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
414 if (nmp->nm_readdirsize == 0)
415 nmp->nm_readdirsize = max;
417 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
418 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
419 nmp->nm_maxfilesize = maxfsize;
420 nmp->nm_state |= NFSSTA_GOTFSINFO;
428 * Mount a remote root fs via. nfs. This depends on the info in the
429 * nfs_diskless structure that has been filled in properly by some primary
431 * It goes something like this:
432 * - do enough of "ifconfig" by calling ifioctl() so that the system
433 * can talk to the server
434 * - If nfs_diskless.mygateway is filled in, use that address as
436 * - build the rootfs mount point and call mountnfs() to do the rest.
439 nfs_mountroot(struct mount *mp)
441 struct mount *swap_mp;
442 struct nfsv3_diskless *nd = &nfsv3_diskless;
445 struct thread *td = curthread; /* XXX */
450 #if defined(BOOTP_NFSROOT) && defined(BOOTP)
451 bootpc_init(); /* use bootp to get nfs_diskless filled in */
455 * XXX time must be non-zero when we init the interface or else
456 * the arp code will wedge...
458 while (mycpu->gd_time_seconds == 0)
459 tsleep(mycpu, 0, "arpkludge", 10);
462 * The boot code may have passed us a diskless structure.
464 if (nfs_diskless_valid == 1)
465 nfs_convert_diskless();
467 #define SINP(sockaddr) ((struct sockaddr_in *)(sockaddr))
468 kprintf("nfs_mountroot: interface %s ip %s",
470 inet_ntoa(SINP(&nd->myif.ifra_addr)->sin_addr));
472 inet_ntoa(SINP(&nd->myif.ifra_broadaddr)->sin_addr));
473 kprintf(" mask %s\n",
474 inet_ntoa(SINP(&nd->myif.ifra_mask)->sin_addr));
478 * XXX splnet, so networks will receive...
483 * BOOTP does not necessarily have to be compiled into the kernel
484 * for an NFS root to work. If we inherited the network
485 * configuration for PXEBOOT then pxe_setup_nfsdiskless() has figured
486 * out our interface for us and all we need to do is ifconfig the
487 * interface. We only do this if the interface has not already been
488 * ifconfig'd by e.g. BOOTP.
490 error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td);
492 panic("nfs_mountroot: socreate(%04x): %d",
493 nd->myif.ifra_addr.sa_family, error);
496 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, proc0.p_ucred);
498 panic("nfs_mountroot: SIOCAIFADDR: %d", error);
500 soclose(so, FNONBLOCK);
503 * If the gateway field is filled in, set it as the default route.
505 if (nd->mygateway.sin_len != 0) {
506 struct sockaddr_in mask, sin;
508 bzero((caddr_t)&mask, sizeof(mask));
510 sin.sin_family = AF_INET;
511 sin.sin_len = sizeof(sin);
512 kprintf("nfs_mountroot: gateway %s\n",
513 inet_ntoa(nd->mygateway.sin_addr));
514 error = rtrequest_global(RTM_ADD, (struct sockaddr *)&sin,
515 (struct sockaddr *)&nd->mygateway,
516 (struct sockaddr *)&mask,
517 RTF_UP | RTF_GATEWAY);
519 kprintf("nfs_mountroot: unable to set gateway, error %d, continuing anyway\n", error);
523 * Create the rootfs mount point.
525 nd->root_args.fh = nd->root_fh;
526 nd->root_args.fhsize = nd->root_fhsize;
527 l = ntohl(nd->root_saddr.sin_addr.s_addr);
528 ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
529 (l >> 24) & 0xff, (l >> 16) & 0xff,
530 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->root_hostnam);
531 kprintf("NFS_ROOT: %s\n",buf);
532 if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY,
533 &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) {
534 mp->mnt_vfc->vfc_refcount--;
540 if (nd->swap_nblks) {
542 /* Convert to DEV_BSIZE instead of Kilobyte */
546 * Create a fake mount point just for the swap vnode so that the
547 * swap file can be on a different server from the rootfs.
549 nd->swap_args.fh = nd->swap_fh;
550 nd->swap_args.fhsize = nd->swap_fhsize;
551 l = ntohl(nd->swap_saddr.sin_addr.s_addr);
552 ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
553 (l >> 24) & 0xff, (l >> 16) & 0xff,
554 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->swap_hostnam);
555 kprintf("NFS SWAP: %s\n",buf);
556 if ((error = nfs_mountdiskless(buf, "/swap", 0,
557 &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0) {
563 VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size =
564 nd->swap_nblks * DEV_BSIZE ;
567 * Since the swap file is not the root dir of a file system,
568 * hack it to a regular file.
572 nfs_setvtype(vp, VREG);
573 swaponvp(td, vp, nd->swap_nblks);
576 mp->mnt_flag |= MNT_ROOTFS;
580 * This is not really an nfs issue, but it is much easier to
581 * set hostname here and then let the "/etc/rc.xxx" files
582 * mount the right /var based upon its preset value.
584 bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN);
585 hostname[MAXHOSTNAMELEN - 1] = '\0';
586 for (i = 0; i < MAXHOSTNAMELEN; i++)
587 if (hostname[i] == '\0')
589 inittodr(ntohl(nd->root_time));
595 * Internal version of mount system call for diskless setup.
598 nfs_mountdiskless(char *path, char *which, int mountflag,
599 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
600 struct vnode **vpp, struct mount **mpp)
603 struct sockaddr *nam;
610 if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) {
611 kprintf("nfs_mountroot: NFS not configured");
616 mp->mnt_kern_flag = 0;
617 mp->mnt_flag = mountflag;
618 nam = dup_sockaddr((struct sockaddr *)sin);
620 #if defined(BOOTP) || defined(NFS_ROOT)
621 if (args->fhsize == 0) {
624 kprintf("NFS_ROOT: No FH passed from loader, attempting mount rpc...");
625 while (*xpath && *xpath != ':')
630 error = md_mount(sin, xpath, args->fh, &args->fhsize, args, td);
632 kprintf("failed error %d.\n", error);
635 kprintf("success!\n");
639 if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) {
640 #if defined(BOOTP) || defined(NFS_ROOT)
643 kprintf("nfs_mountroot: mount %s on %s: %d", path, which, error);
644 mp->mnt_vfc->vfc_refcount--;
656 nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp)
663 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
664 * no sense in that context.
666 if (argp->sotype == SOCK_STREAM)
667 nmp->nm_flag &= ~NFSMNT_NOCONN;
669 /* Also clear RDIRPLUS if not NFSv3, it crashes some servers */
670 if ((argp->flags & NFSMNT_NFSV3) == 0)
671 nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
673 /* Re-bind if rsrvd port requested and wasn't on one */
674 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
675 && (argp->flags & NFSMNT_RESVPORT);
676 /* Also re-bind if we're switching to/from a connected UDP socket */
677 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
678 (argp->flags & NFSMNT_NOCONN));
680 /* Update flags atomically. Don't change the lock bits. */
681 nmp->nm_flag = argp->flags | nmp->nm_flag;
684 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
685 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
686 if (nmp->nm_timeo < NFS_MINTIMEO)
687 nmp->nm_timeo = NFS_MINTIMEO;
688 else if (nmp->nm_timeo > NFS_MAXTIMEO)
689 nmp->nm_timeo = NFS_MAXTIMEO;
692 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
693 nmp->nm_retry = argp->retrans;
694 if (nmp->nm_retry > NFS_MAXREXMIT)
695 nmp->nm_retry = NFS_MAXREXMIT;
698 maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
700 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
701 nmp->nm_wsize = argp->wsize;
702 /* Round down to multiple of blocksize */
703 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
704 if (nmp->nm_wsize <= 0)
705 nmp->nm_wsize = NFS_FABLKSIZE;
707 if (nmp->nm_wsize > maxio)
708 nmp->nm_wsize = maxio;
709 if (nmp->nm_wsize > MAXBSIZE)
710 nmp->nm_wsize = MAXBSIZE;
712 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
713 nmp->nm_rsize = argp->rsize;
714 /* Round down to multiple of blocksize */
715 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
716 if (nmp->nm_rsize <= 0)
717 nmp->nm_rsize = NFS_FABLKSIZE;
719 if (nmp->nm_rsize > maxio)
720 nmp->nm_rsize = maxio;
721 if (nmp->nm_rsize > MAXBSIZE)
722 nmp->nm_rsize = MAXBSIZE;
724 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
725 nmp->nm_readdirsize = argp->readdirsize;
727 if (nmp->nm_readdirsize > maxio)
728 nmp->nm_readdirsize = maxio;
729 if (nmp->nm_readdirsize > nmp->nm_rsize)
730 nmp->nm_readdirsize = nmp->nm_rsize;
732 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
733 nmp->nm_acregmin = argp->acregmin;
735 nmp->nm_acregmin = NFS_MINATTRTIMO;
736 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
737 nmp->nm_acregmax = argp->acregmax;
739 nmp->nm_acregmax = NFS_MAXATTRTIMO;
740 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
741 nmp->nm_acdirmin = argp->acdirmin;
743 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
744 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
745 nmp->nm_acdirmax = argp->acdirmax;
747 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
748 if (nmp->nm_acdirmin > nmp->nm_acdirmax)
749 nmp->nm_acdirmin = nmp->nm_acdirmax;
750 if (nmp->nm_acregmin > nmp->nm_acregmax)
751 nmp->nm_acregmin = nmp->nm_acregmax;
753 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) {
754 if (argp->maxgrouplist <= NFS_MAXGRPS)
755 nmp->nm_numgrps = argp->maxgrouplist;
757 nmp->nm_numgrps = NFS_MAXGRPS;
759 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
760 if (argp->readahead <= NFS_MAXRAHEAD)
761 nmp->nm_readahead = argp->readahead;
763 nmp->nm_readahead = NFS_MAXRAHEAD;
765 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) {
766 if (argp->deadthresh <= NFS_NEVERDEAD)
767 nmp->nm_deadthresh = argp->deadthresh;
769 nmp->nm_deadthresh = NFS_NEVERDEAD;
772 adjsock |= ((nmp->nm_sotype != argp->sotype) ||
773 (nmp->nm_soproto != argp->proto));
774 nmp->nm_sotype = argp->sotype;
775 nmp->nm_soproto = argp->proto;
777 if (nmp->nm_so && adjsock) {
778 nfs_safedisconnect(nmp);
779 if (nmp->nm_sotype == SOCK_DGRAM)
780 while (nfs_connect(nmp, NULL)) {
781 kprintf("nfs_args: retrying connect\n");
782 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
791 * It seems a bit dumb to copyinstr() the host and path here and then
792 * bcopy() them in mountnfs(), but I wanted to detect errors before
793 * doing the sockargs() call because sockargs() allocates an mbuf and
794 * an error after that means that I have to release the mbuf.
798 nfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred)
801 struct nfs_args args;
802 struct sockaddr *nam;
804 char pth[MNAMELEN], hst[MNAMELEN];
806 u_char nfh[NFSX_V3FHMAX];
812 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
815 if (args.version != NFS_ARGSVERSION) {
816 #ifdef COMPAT_PRELITE2
818 * If the argument version is unknown, then assume the
819 * caller is a pre-lite2 4.4BSD client and convert its
822 struct onfs_args oargs;
823 error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args));
826 nfs_convert_oargs(&args,&oargs);
827 #else /* !COMPAT_PRELITE2 */
828 return (EPROGMISMATCH);
829 #endif /* COMPAT_PRELITE2 */
831 if (mp->mnt_flag & MNT_UPDATE) {
832 struct nfsmount *nmp = VFSTONFS(mp);
837 * When doing an update, we can't change from or to
838 * v3, or change cookie translation
840 args.flags = (args.flags &
841 ~(NFSMNT_NFSV3/*|NFSMNT_XLATECOOKIE*/)) |
843 (NFSMNT_NFSV3/*|NFSMNT_XLATECOOKIE*/));
844 nfs_decode_args(nmp, &args);
849 * Make the nfs_ip_paranoia sysctl serve as the default connection
850 * or no-connection mode for those protocols that support
851 * no-connection mode (the flag will be cleared later for protocols
852 * that do not support no-connection mode). This will allow a client
853 * to receive replies from a different IP then the request was
854 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid),
857 if (nfs_ip_paranoia == 0)
858 args.flags |= NFSMNT_NOCONN;
859 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
861 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
864 error = copyinstr(path, pth, MNAMELEN-1, &len);
867 bzero(&pth[len], MNAMELEN - len);
868 error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
871 bzero(&hst[len], MNAMELEN - len);
872 /* sockargs() call must be after above copyin() calls */
873 error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen);
877 error = mountnfs(&args, mp, nam, pth, hst, &vp);
882 * Common code for mount and mountroot
885 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
886 char *pth, char *hst, struct vnode **vpp)
888 struct nfsmount *nmp;
892 if (mp->mnt_flag & MNT_UPDATE) {
894 /* update paths, file handles, etc, here XXX */
898 nmp = zalloc(nfsmount_zone);
899 bzero((caddr_t)nmp, sizeof (struct nfsmount));
900 TAILQ_INIT(&nmp->nm_uidlruhead);
901 TAILQ_INIT(&nmp->nm_bioq);
902 mp->mnt_data = (qaddr_t)nmp;
908 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too
909 * high, depending on whether we end up with negative offsets in
910 * the client or server somewhere. 2GB-1 may be safer.
912 * For V3, nfs_fsinfo will adjust this as necessary. Assume maximum
913 * that we can handle until we find out otherwise.
914 * XXX Our "safe" limit on the client is what we can store in our
915 * buffer cache using signed(!) block numbers.
917 if ((argp->flags & NFSMNT_NFSV3) == 0)
918 nmp->nm_maxfilesize = 0xffffffffLL;
920 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
922 nmp->nm_timeo = NFS_TIMEO;
923 nmp->nm_retry = NFS_RETRANS;
924 nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
925 nmp->nm_rsize = nmp->nm_wsize;
926 nmp->nm_readdirsize = NFS_READDIRSIZE;
927 nmp->nm_numgrps = NFS_MAXGRPS;
928 nmp->nm_readahead = NFS_DEFRAHEAD;
929 nmp->nm_deadthresh = NFS_DEADTHRESH;
930 nmp->nm_fhsize = argp->fhsize;
931 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
932 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
934 /* Set up the sockets and per-host congestion */
935 nmp->nm_sotype = argp->sotype;
936 nmp->nm_soproto = argp->proto;
937 nmp->nm_cred = crhold(proc0.p_ucred);
939 nfs_decode_args(nmp, argp);
942 * For Connection based sockets (TCP,...) defer the connect until
943 * the first request, in case the server is not responding.
945 if (nmp->nm_sotype == SOCK_DGRAM &&
946 (error = nfs_connect(nmp, NULL)))
950 * This is silly, but it has to be set so that vinifod() works.
951 * We do not want to do an nfs_statfs() here since we can get
952 * stuck on a dead server and we are holding a lock on the mount
955 mp->mnt_stat.f_iosize =
956 nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype);
959 * Install vop_ops for our vnops
961 vfs_add_vnodeops(mp, &nfsv2_vnode_vops, &mp->mnt_vn_norm_ops);
962 vfs_add_vnodeops(mp, &nfsv2_spec_vops, &mp->mnt_vn_spec_ops);
963 vfs_add_vnodeops(mp, &nfsv2_fifo_vops, &mp->mnt_vn_fifo_ops);
966 * A reference count is needed on the nfsnode representing the
967 * remote root. If this object is not persistent, then backward
968 * traversals of the mount point (i.e. "..") will not work if
969 * the nfsnode gets flushed out of the cache. Ufs does not have
970 * this problem, because one can identify root inodes by their
971 * number == ROOTINO (2).
973 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
979 * Retrieval of mountpoint attributes is delayed until nfs_rot
980 * or nfs_statfs are first called. This will happen either when
981 * we first traverse the mount point or if somebody does a df(1).
983 * NFSSTA_GOTFSINFO is used to flag if we have successfully
984 * retrieved mountpoint attributes. In the case of NFSv3 we
985 * also flag static fsinfo.
988 (*vpp)->v_type = VNON;
991 * Lose the lock but keep the ref.
1004 * unmount system call
1007 nfs_unmount(struct mount *mp, int mntflags)
1009 struct nfsmount *nmp;
1010 int error, flags = 0;
1012 if (mntflags & MNT_FORCE)
1013 flags |= FORCECLOSE;
1016 * Goes something like this..
1017 * - Call vflush() to clear out vnodes for this file system
1018 * - Close the socket
1019 * - Free up the data structures
1021 /* In the forced case, cancel any outstanding requests. */
1022 if (flags & FORCECLOSE) {
1023 error = nfs_nmcancelreqs(nmp);
1028 * Must handshake with nfs_clientd() if it is active. XXX
1030 nmp->nm_state |= NFSSTA_DISMINPROG;
1032 /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */
1033 error = vflush(mp, 1, flags);
1035 nmp->nm_state &= ~NFSSTA_DISMINPROG;
1040 * We are now committed to the unmount.
1041 * For NQNFS, let the server daemon free the nfsmount structure.
1043 if (nmp->nm_flag & NFSMNT_KERB)
1044 nmp->nm_state |= NFSSTA_DISMNT;
1046 nfs_disconnect(nmp);
1047 FREE(nmp->nm_nam, M_SONAME);
1049 if ((nmp->nm_flag & NFSMNT_KERB) == 0)
1050 nfs_free_mount(nmp);
1055 nfs_free_mount(struct nfsmount *nmp)
1058 crfree(nmp->nm_cred);
1059 nmp->nm_cred = NULL;
1061 zfree(nfsmount_zone, nmp);
1065 * Return root of a filesystem
1068 nfs_root(struct mount *mp, struct vnode **vpp)
1071 struct nfsmount *nmp;
1077 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
1083 * Get transfer parameters and root vnode attributes
1085 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
1086 if (nmp->nm_flag & NFSMNT_NFSV3) {
1087 error = nfs_fsinfo(nmp, vp, curthread);
1088 mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype);
1090 if ((error = VOP_GETATTR(vp, &attrs)) == 0)
1091 nmp->nm_state |= NFSSTA_GOTFSINFO;
1096 * The root vnode is usually cached by the namecache so do not
1097 * try to avoid going over the wire even if we have previous
1098 * information cached. A stale NFS mount can loop
1099 * forever resolving the root vnode if we return no-error when
1100 * there is in fact an error.
1102 np->n_attrstamp = 0;
1103 error = VOP_GETATTR(vp, &attrs);
1105 if (vp->v_type == VNON)
1106 nfs_setvtype(vp, VDIR);
1123 static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
1124 static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
1127 * Flush out the buffer cache
1131 nfs_sync(struct mount *mp, int waitfor)
1133 struct scaninfo scaninfo;
1136 scaninfo.rescan = 1;
1137 scaninfo.waitfor = waitfor;
1138 scaninfo.allerror = 0;
1141 * Force stale buffer cache information to be flushed.
1144 while (error == 0 && scaninfo.rescan) {
1145 scaninfo.rescan = 0;
1146 error = vmntvnodescan(mp, VMSC_GETVP, nfs_sync_scan1,
1147 nfs_sync_scan2, &scaninfo);
1153 nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1155 struct scaninfo *info = data;
1157 if (vn_islocked(vp) || RB_EMPTY(&vp->v_rbdirty_tree))
1159 if (info->waitfor == MNT_LAZY)
1165 nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1167 struct scaninfo *info = data;
1170 error = VOP_FSYNC(vp, info->waitfor);
1172 info->allerror = error;