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_statvfs(struct mount *mp, struct statvfs *sbp,
128 static int nfs_sync ( struct mount *mp, int waitfor);
131 * nfs vfs operations.
133 static struct vfsops nfs_vfsops = {
134 .vfs_mount = nfs_mount,
135 .vfs_unmount = nfs_unmount,
136 .vfs_root = nfs_root,
137 .vfs_statfs = nfs_statfs,
138 .vfs_statvfs = nfs_statvfs,
139 .vfs_sync = nfs_sync,
140 .vfs_init = nfs_init,
141 .vfs_uninit = nfs_uninit
143 VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK);
146 * This structure must be filled in by a primary bootstrap or bootstrap
147 * server for a diskless/dataless machine. It is initialized below just
148 * to ensure that it is allocated to initialized data (.data not .bss).
150 struct nfs_diskless nfs_diskless = { { { 0 } } };
151 struct nfsv3_diskless nfsv3_diskless = { { { 0 } } };
152 int nfs_diskless_valid = 0;
154 SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD,
155 &nfs_diskless_valid, 0, "");
157 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD,
158 nfsv3_diskless.root_hostnam, 0, "");
160 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD,
161 &nfsv3_diskless.root_saddr, sizeof nfsv3_diskless.root_saddr,
162 "%Ssockaddr_in", "");
164 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_swappath, CTLFLAG_RD,
165 nfsv3_diskless.swap_hostnam, 0, "");
167 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_swapaddr, CTLFLAG_RD,
168 &nfsv3_diskless.swap_saddr, sizeof nfsv3_diskless.swap_saddr,
172 void nfsargs_ntoh (struct nfs_args *);
173 static int nfs_mountdiskless (char *, char *, int,
174 struct sockaddr_in *, struct nfs_args *,
175 struct thread *, struct vnode **,
177 static void nfs_convert_diskless (void);
178 static void nfs_convert_oargs (struct nfs_args *args,
179 struct onfs_args *oargs);
182 * Calculate the buffer I/O block size to use. The maximum V2 block size
183 * is typically 8K, the maximum datagram size is typically 16K, and the
184 * maximum V3 block size is typically 32K. The buffer cache tends to work
185 * best with 16K blocks but we allow 32K for TCP connections.
187 * We force the block size to be at least a page for buffer cache efficiency.
190 nfs_iosize(int v3, int sotype)
196 if (sotype == SOCK_STREAM)
199 iomax = NFS_MAXDGRAMDATA;
201 iomax = NFS_V2MAXDATA;
203 if ((iosize = nfs_io_size) > iomax)
205 if (iosize < PAGE_SIZE)
209 * This is an aweful hack but until the buffer cache is rewritten
210 * we need it. The problem is that when you combine write() with
211 * mmap() the vm_page->valid bits can become weird looking
212 * (e.g. 0xfc). This occurs because NFS uses piecemeal buffers
213 * at the file EOF. To solve the problem the BIO system needs to
214 * be guarenteed that the NFS iosize for regular files will be a
215 * multiple of PAGE_SIZE so it can invalidate the whole page
216 * rather then just the piece of it owned by the buffer when
217 * NFS does vinvalbuf() calls.
219 if (iosize & PAGE_MASK)
220 iosize = (iosize & ~PAGE_MASK) + PAGE_SIZE;
225 nfs_convert_oargs(struct nfs_args *args, struct onfs_args *oargs)
227 args->version = NFS_ARGSVERSION;
228 args->addr = oargs->addr;
229 args->addrlen = oargs->addrlen;
230 args->sotype = oargs->sotype;
231 args->proto = oargs->proto;
232 args->fh = oargs->fh;
233 args->fhsize = oargs->fhsize;
234 args->flags = oargs->flags;
235 args->wsize = oargs->wsize;
236 args->rsize = oargs->rsize;
237 args->readdirsize = oargs->readdirsize;
238 args->timeo = oargs->timeo;
239 args->retrans = oargs->retrans;
240 args->maxgrouplist = oargs->maxgrouplist;
241 args->readahead = oargs->readahead;
242 args->deadthresh = oargs->deadthresh;
243 args->hostname = oargs->hostname;
247 nfs_convert_diskless(void)
251 bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif,
252 sizeof(struct ifaliasreq));
253 bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway,
254 sizeof(struct sockaddr_in));
255 nfs_convert_oargs(&nfsv3_diskless.swap_args,&nfs_diskless.swap_args);
257 bcopy(nfs_diskless.swap_fh,nfsv3_diskless.swap_fh,NFSX_V2FH);
258 nfsv3_diskless.swap_fhsize = NFSX_V2FH;
259 for (i = NFSX_V2FH - 1; i >= 0; --i) {
260 if (nfs_diskless.swap_fh[i])
264 nfsv3_diskless.swap_fhsize = 0;
266 bcopy(&nfs_diskless.swap_saddr,&nfsv3_diskless.swap_saddr,
267 sizeof(struct sockaddr_in));
268 bcopy(nfs_diskless.swap_hostnam,nfsv3_diskless.swap_hostnam, MNAMELEN);
269 nfsv3_diskless.swap_nblks = nfs_diskless.swap_nblks;
270 bcopy(&nfs_diskless.swap_ucred, &nfsv3_diskless.swap_ucred,
271 sizeof(struct ucred));
272 nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args);
274 bcopy(nfs_diskless.root_fh,nfsv3_diskless.root_fh,NFSX_V2FH);
275 nfsv3_diskless.root_fhsize = NFSX_V2FH;
276 for (i = NFSX_V2FH - 1; i >= 0; --i) {
277 if (nfs_diskless.root_fh[i])
281 nfsv3_diskless.root_fhsize = 0;
283 bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr,
284 sizeof(struct sockaddr_in));
285 bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN);
286 nfsv3_diskless.root_time = nfs_diskless.root_time;
287 bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam,
289 nfs_diskless_valid = 3;
296 nfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
299 struct nfs_statfs *sfp;
300 struct nfsmount *nmp = VFSTONFS(mp);
301 thread_t td = curthread;
302 int error = 0, retattr;
305 struct nfsm_info info;
308 info.v3 = (nmp->nm_flag & NFSMNT_NFSV3);
310 lwkt_gettoken(&nmp->nm_token);
315 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
317 lwkt_reltoken(&nmp->nm_token);
321 /* ignore the passed cred */
323 cred->cr_ngroups = 1;
324 if (info.v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
325 (void)nfs_fsinfo(nmp, vp, td);
326 nfsstats.rpccnt[NFSPROC_FSSTAT]++;
327 nfsm_reqhead(&info, vp, NFSPROC_FSSTAT, NFSX_FH(info.v3));
328 ERROROUT(nfsm_fhtom(&info, vp));
329 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSSTAT, td, cred, &error));
331 ERROROUT(nfsm_postop_attr(&info, vp, &retattr,
332 NFS_LATTR_NOSHRINK));
335 if (info.mrep != NULL)
339 NULLOUT(sfp = nfsm_dissect(&info, NFSX_STATFS(info.v3)));
340 sbp->f_flags = nmp->nm_flag;
343 sbp->f_bsize = NFS_FABLKSIZE;
344 tquad = fxdr_hyper(&sfp->sf_tbytes);
345 sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
346 tquad = fxdr_hyper(&sfp->sf_fbytes);
347 sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
348 tquad = fxdr_hyper(&sfp->sf_abytes);
349 sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
350 sbp->f_files = (fxdr_unsigned(int32_t,
351 sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
352 sbp->f_ffree = (fxdr_unsigned(int32_t,
353 sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
355 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
356 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
357 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
358 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
364 * Some values are pre-set in mnt_stat. Note in particular f_iosize
365 * cannot be changed once the filesystem is mounted as it is used
366 * as the basis for BIOs.
368 if (sbp != &mp->mnt_stat) {
369 sbp->f_type = mp->mnt_vfc->vfc_typenum;
370 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
371 sbp->f_iosize = mp->mnt_stat.f_iosize;
378 lwkt_reltoken(&nmp->nm_token);
383 nfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
386 struct nfs_statfs *sfp;
387 struct nfsmount *nmp = VFSTONFS(mp);
388 thread_t td = curthread;
389 int error = 0, retattr;
391 struct nfsm_info info;
394 info.v3 = (nmp->nm_flag & NFSMNT_NFSV3);
395 lwkt_gettoken(&nmp->nm_token);
400 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
402 lwkt_reltoken(&nmp->nm_token);
406 /* ignore the passed cred */
408 cred->cr_ngroups = 1;
409 if (info.v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
410 (void)nfs_fsinfo(nmp, vp, td);
411 nfsstats.rpccnt[NFSPROC_FSSTAT]++;
412 nfsm_reqhead(&info, vp, NFSPROC_FSSTAT, NFSX_FH(info.v3));
413 ERROROUT(nfsm_fhtom(&info, vp));
414 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSSTAT, td, cred, &error));
416 ERROROUT(nfsm_postop_attr(&info, vp, &retattr,
417 NFS_LATTR_NOSHRINK));
420 if (info.mrep != NULL)
424 NULLOUT(sfp = nfsm_dissect(&info, NFSX_STATFS(info.v3)));
425 sbp->f_flag = nmp->nm_flag;
426 sbp->f_owner = nmp->nm_cred->cr_ruid;
429 sbp->f_bsize = NFS_FABLKSIZE;
430 sbp->f_frsize = NFS_FABLKSIZE;
431 sbp->f_blocks = (fxdr_hyper(&sfp->sf_tbytes) /
432 ((u_quad_t)NFS_FABLKSIZE));
433 sbp->f_bfree = (fxdr_hyper(&sfp->sf_fbytes) /
434 ((u_quad_t)NFS_FABLKSIZE));
435 sbp->f_bavail = (fxdr_hyper(&sfp->sf_abytes) /
436 ((u_quad_t)NFS_FABLKSIZE));
437 sbp->f_files = fxdr_hyper(&sfp->sf_tfiles);
438 sbp->f_ffree = fxdr_hyper(&sfp->sf_ffiles);
439 sbp->f_favail = fxdr_hyper(&sfp->sf_afiles);
441 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
442 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
443 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
444 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
449 sbp->f_syncreads = 0;
450 sbp->f_syncwrites = 0;
451 sbp->f_asyncreads = 0;
452 sbp->f_asyncwrites = 0;
453 sbp->f_type = mp->mnt_vfc->vfc_typenum;
460 lwkt_reltoken(&nmp->nm_token);
465 * nfs version 3 fsinfo rpc call
468 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td)
470 struct nfsv3_fsinfo *fsp;
472 int error = 0, retattr;
474 struct nfsm_info info;
477 nfsstats.rpccnt[NFSPROC_FSINFO]++;
478 nfsm_reqhead(&info, vp, NFSPROC_FSINFO, NFSX_FH(1));
479 ERROROUT(nfsm_fhtom(&info, vp));
480 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSINFO, td,
481 nfs_vpcred(vp, ND_READ), &error));
482 ERROROUT(nfsm_postop_attr(&info, vp, &retattr, NFS_LATTR_NOSHRINK));
484 NULLOUT(fsp = nfsm_dissect(&info, NFSX_V3FSINFO));
485 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
486 if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
487 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
488 ~(NFS_FABLKSIZE - 1);
489 max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
490 if (max < nmp->nm_wsize && max > 0) {
491 nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
492 if (nmp->nm_wsize == 0)
495 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
496 if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
497 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
498 ~(NFS_FABLKSIZE - 1);
499 max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
500 if (max < nmp->nm_rsize && max > 0) {
501 nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
502 if (nmp->nm_rsize == 0)
505 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
506 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ)
507 nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) &
508 ~(NFS_DIRBLKSIZ - 1);
509 if (max < nmp->nm_readdirsize && max > 0) {
510 nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
511 if (nmp->nm_readdirsize == 0)
512 nmp->nm_readdirsize = max;
514 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
515 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
516 nmp->nm_maxfilesize = maxfsize;
517 nmp->nm_state |= NFSSTA_GOTFSINFO;
520 * Use the smaller of rsize/wsize for the biosize.
522 if (nmp->nm_rsize < nmp->nm_wsize)
523 nmp->nm_mountp->mnt_stat.f_iosize = nmp->nm_rsize;
525 nmp->nm_mountp->mnt_stat.f_iosize = nmp->nm_wsize;
534 * Mount a remote root fs via. nfs. This depends on the info in the
535 * nfs_diskless structure that has been filled in properly by some primary
537 * It goes something like this:
538 * - do enough of "ifconfig" by calling ifioctl() so that the system
539 * can talk to the server
540 * - If nfs_diskless.mygateway is filled in, use that address as
542 * - build the rootfs mount point and call mountnfs() to do the rest.
545 nfs_mountroot(struct mount *mp)
547 struct mount *swap_mp;
548 struct nfsv3_diskless *nd = &nfsv3_diskless;
551 struct thread *td = curthread; /* XXX */
556 #if defined(BOOTP_NFSROOT) && defined(BOOTP)
557 bootpc_init(); /* use bootp to get nfs_diskless filled in */
561 * XXX time must be non-zero when we init the interface or else
562 * the arp code will wedge...
564 while (mycpu->gd_time_seconds == 0)
565 tsleep(mycpu, 0, "arpkludge", 10);
568 * The boot code may have passed us a diskless structure.
570 if (nfs_diskless_valid == 1)
571 nfs_convert_diskless();
573 #define SINP(sockaddr) ((struct sockaddr_in *)(sockaddr))
574 kprintf("nfs_mountroot: interface %s ip %s",
576 inet_ntoa(SINP(&nd->myif.ifra_addr)->sin_addr));
578 inet_ntoa(SINP(&nd->myif.ifra_broadaddr)->sin_addr));
579 kprintf(" mask %s\n",
580 inet_ntoa(SINP(&nd->myif.ifra_mask)->sin_addr));
584 * XXX splnet, so networks will receive...
589 * BOOTP does not necessarily have to be compiled into the kernel
590 * for an NFS root to work. If we inherited the network
591 * configuration for PXEBOOT then pxe_setup_nfsdiskless() has figured
592 * out our interface for us and all we need to do is ifconfig the
593 * interface. We only do this if the interface has not already been
594 * ifconfig'd by e.g. BOOTP.
596 error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td);
598 panic("nfs_mountroot: socreate(%04x): %d",
599 nd->myif.ifra_addr.sa_family, error);
602 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, proc0.p_ucred);
604 panic("nfs_mountroot: SIOCAIFADDR: %d", error);
606 soclose(so, FNONBLOCK);
609 * If the gateway field is filled in, set it as the default route.
611 if (nd->mygateway.sin_len != 0) {
612 struct sockaddr_in mask, sin;
614 bzero((caddr_t)&mask, sizeof(mask));
616 sin.sin_family = AF_INET;
617 sin.sin_len = sizeof(sin);
618 kprintf("nfs_mountroot: gateway %s\n",
619 inet_ntoa(nd->mygateway.sin_addr));
620 error = rtrequest_global(RTM_ADD, (struct sockaddr *)&sin,
621 (struct sockaddr *)&nd->mygateway,
622 (struct sockaddr *)&mask,
623 RTF_UP | RTF_GATEWAY);
625 kprintf("nfs_mountroot: unable to set gateway, error %d, continuing anyway\n", error);
629 * Create the rootfs mount point.
631 nd->root_args.fh = nd->root_fh;
632 nd->root_args.fhsize = nd->root_fhsize;
633 l = ntohl(nd->root_saddr.sin_addr.s_addr);
634 ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
635 (l >> 24) & 0xff, (l >> 16) & 0xff,
636 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->root_hostnam);
637 kprintf("NFS_ROOT: %s\n",buf);
638 if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY,
639 &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) {
640 mp->mnt_vfc->vfc_refcount--;
646 if (nd->swap_nblks) {
648 /* Convert to DEV_BSIZE instead of Kilobyte */
652 * Create a fake mount point just for the swap vnode so that the
653 * swap file can be on a different server from the rootfs.
655 nd->swap_args.fh = nd->swap_fh;
656 nd->swap_args.fhsize = nd->swap_fhsize;
657 l = ntohl(nd->swap_saddr.sin_addr.s_addr);
658 ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
659 (l >> 24) & 0xff, (l >> 16) & 0xff,
660 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->swap_hostnam);
661 kprintf("NFS SWAP: %s\n",buf);
662 if ((error = nfs_mountdiskless(buf, "/swap", 0,
663 &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0) {
669 VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size =
670 nd->swap_nblks * DEV_BSIZE ;
673 * Since the swap file is not the root dir of a file system,
674 * hack it to a regular file.
676 vclrflags(vp, VROOT);
678 nfs_setvtype(vp, VREG);
679 swaponvp(td, vp, nd->swap_nblks);
682 mp->mnt_flag |= MNT_ROOTFS;
686 * This is not really an nfs issue, but it is much easier to
687 * set hostname here and then let the "/etc/rc.xxx" files
688 * mount the right /var based upon its preset value.
690 bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN);
691 hostname[MAXHOSTNAMELEN - 1] = '\0';
692 for (i = 0; i < MAXHOSTNAMELEN; i++)
693 if (hostname[i] == '\0')
695 inittodr(ntohl(nd->root_time));
701 * Internal version of mount system call for diskless setup.
704 nfs_mountdiskless(char *path, char *which, int mountflag,
705 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
706 struct vnode **vpp, struct mount **mpp)
709 struct sockaddr *nam;
716 if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) {
717 kprintf("nfs_mountroot: NFS not configured");
722 mp->mnt_kern_flag = 0;
723 mp->mnt_flag = mountflag;
724 nam = dup_sockaddr((struct sockaddr *)sin);
726 #if defined(BOOTP) || defined(NFS_ROOT)
727 if (args->fhsize == 0) {
730 kprintf("NFS_ROOT: No FH passed from loader, attempting mount rpc...");
731 while (*xpath && *xpath != ':')
736 error = md_mount(sin, xpath, args->fh, &args->fhsize, args, td);
738 kprintf("failed error %d.\n", error);
741 kprintf("success!\n");
745 if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) {
746 #if defined(BOOTP) || defined(NFS_ROOT)
749 kprintf("nfs_mountroot: mount %s on %s: %d", path, which, error);
750 mp->mnt_vfc->vfc_refcount--;
762 nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp)
769 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
770 * no sense in that context.
772 if (nmp->nm_sotype == SOCK_STREAM)
773 nmp->nm_flag &= ~NFSMNT_NOCONN;
775 /* Also clear RDIRPLUS if not NFSv3, it crashes some servers */
776 if ((argp->flags & NFSMNT_NFSV3) == 0)
777 nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
780 * Re-bind if rsrvd port flag has changed
782 adjsock = (nmp->nm_flag & NFSMNT_RESVPORT) !=
783 (argp->flags & NFSMNT_RESVPORT);
785 /* Update flags atomically. Don't change the lock bits. */
786 nmp->nm_flag = argp->flags | nmp->nm_flag;
789 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
790 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
791 if (nmp->nm_timeo < NFS_MINTIMEO)
792 nmp->nm_timeo = NFS_MINTIMEO;
793 else if (nmp->nm_timeo > NFS_MAXTIMEO)
794 nmp->nm_timeo = NFS_MAXTIMEO;
797 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
798 nmp->nm_retry = argp->retrans;
799 if (nmp->nm_retry > NFS_MAXREXMIT)
800 nmp->nm_retry = NFS_MAXREXMIT;
804 * These parameters effect the buffer cache and cannot be changed
805 * once we've successfully mounted.
807 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
808 maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, nmp->nm_sotype);
810 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
811 nmp->nm_wsize = argp->wsize;
812 /* Round down to multiple of blocksize */
813 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
814 if (nmp->nm_wsize <= 0)
815 nmp->nm_wsize = NFS_FABLKSIZE;
817 if (nmp->nm_wsize > maxio)
818 nmp->nm_wsize = maxio;
819 if (nmp->nm_wsize > MAXBSIZE)
820 nmp->nm_wsize = MAXBSIZE;
822 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
823 nmp->nm_rsize = argp->rsize;
824 /* Round down to multiple of blocksize */
825 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
826 if (nmp->nm_rsize <= 0)
827 nmp->nm_rsize = NFS_FABLKSIZE;
829 if (nmp->nm_rsize > maxio)
830 nmp->nm_rsize = maxio;
831 if (nmp->nm_rsize > MAXBSIZE)
832 nmp->nm_rsize = MAXBSIZE;
834 if ((argp->flags & NFSMNT_READDIRSIZE) &&
835 argp->readdirsize > 0) {
836 nmp->nm_readdirsize = argp->readdirsize;
838 if (nmp->nm_readdirsize > maxio)
839 nmp->nm_readdirsize = maxio;
840 if (nmp->nm_readdirsize > nmp->nm_rsize)
841 nmp->nm_readdirsize = nmp->nm_rsize;
844 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
845 nmp->nm_acregmin = argp->acregmin;
847 nmp->nm_acregmin = NFS_MINATTRTIMO;
848 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
849 nmp->nm_acregmax = argp->acregmax;
851 nmp->nm_acregmax = NFS_MAXATTRTIMO;
852 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
853 nmp->nm_acdirmin = argp->acdirmin;
855 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
856 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
857 nmp->nm_acdirmax = argp->acdirmax;
859 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
860 if (nmp->nm_acdirmin > nmp->nm_acdirmax)
861 nmp->nm_acdirmin = nmp->nm_acdirmax;
862 if (nmp->nm_acregmin > nmp->nm_acregmax)
863 nmp->nm_acregmin = nmp->nm_acregmax;
865 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) {
866 if (argp->maxgrouplist <= NFS_MAXGRPS)
867 nmp->nm_numgrps = argp->maxgrouplist;
869 nmp->nm_numgrps = NFS_MAXGRPS;
871 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
872 if (argp->readahead <= NFS_MAXRAHEAD)
873 nmp->nm_readahead = argp->readahead;
875 nmp->nm_readahead = NFS_MAXRAHEAD;
877 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) {
878 if (argp->deadthresh <= NFS_NEVERDEAD)
879 nmp->nm_deadthresh = argp->deadthresh;
881 nmp->nm_deadthresh = NFS_NEVERDEAD;
884 if (nmp->nm_so && adjsock) {
885 nfs_safedisconnect(nmp);
886 if (nmp->nm_sotype == SOCK_DGRAM)
887 while (nfs_connect(nmp, NULL)) {
888 kprintf("nfs_args: retrying connect\n");
889 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
898 * It seems a bit dumb to copyinstr() the host and path here and then
899 * bcopy() them in mountnfs(), but I wanted to detect errors before
900 * doing the sockargs() call because sockargs() allocates an mbuf and
901 * an error after that means that I have to release the mbuf.
905 nfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred)
908 struct nfs_args args;
909 struct sockaddr *nam;
911 char pth[MNAMELEN], hst[MNAMELEN];
913 u_char nfh[NFSX_V3FHMAX];
919 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
922 if (args.version != NFS_ARGSVERSION) {
923 #ifdef COMPAT_PRELITE2
925 * If the argument version is unknown, then assume the
926 * caller is a pre-lite2 4.4BSD client and convert its
929 struct onfs_args oargs;
930 error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args));
933 nfs_convert_oargs(&args,&oargs);
934 #else /* !COMPAT_PRELITE2 */
935 return (EPROGMISMATCH);
936 #endif /* COMPAT_PRELITE2 */
938 if (mp->mnt_flag & MNT_UPDATE) {
939 struct nfsmount *nmp = VFSTONFS(mp);
944 * When doing an update, we can't change from or to
945 * v3, or change cookie translation, or rsize or wsize.
947 args.flags &= ~(NFSMNT_NFSV3 | NFSMNT_RSIZE | NFSMNT_WSIZE);
948 args.flags |= nmp->nm_flag & (NFSMNT_NFSV3);
949 nfs_decode_args(nmp, &args);
954 * Make the nfs_ip_paranoia sysctl serve as the default connection
955 * or no-connection mode for those protocols that support
956 * no-connection mode (the flag will be cleared later for protocols
957 * that do not support no-connection mode). This will allow a client
958 * to receive replies from a different IP then the request was
959 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid),
962 if (nfs_ip_paranoia == 0)
963 args.flags |= NFSMNT_NOCONN;
964 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
966 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
969 error = copyinstr(path, pth, MNAMELEN-1, &len);
972 bzero(&pth[len], MNAMELEN - len);
973 error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
976 bzero(&hst[len], MNAMELEN - len);
977 /* sockargs() call must be after above copyin() calls */
978 error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen);
982 error = mountnfs(&args, mp, nam, pth, hst, &vp);
987 * Common code for mount and mountroot
990 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
991 char *pth, char *hst, struct vnode **vpp)
993 struct nfsmount *nmp;
999 if (mp->mnt_flag & MNT_UPDATE) {
1001 /* update paths, file handles, etc, here XXX */
1002 FREE(nam, M_SONAME);
1005 nmp = zalloc(nfsmount_zone);
1006 bzero((caddr_t)nmp, sizeof (struct nfsmount));
1007 mtx_init(&nmp->nm_rxlock);
1008 mtx_init(&nmp->nm_txlock);
1009 TAILQ_INIT(&nmp->nm_uidlruhead);
1010 TAILQ_INIT(&nmp->nm_bioq);
1011 TAILQ_INIT(&nmp->nm_reqq);
1012 TAILQ_INIT(&nmp->nm_reqtxq);
1013 TAILQ_INIT(&nmp->nm_reqrxq);
1014 mp->mnt_data = (qaddr_t)nmp;
1015 lwkt_token_init(&nmp->nm_token, 1, "nfs_token");
1018 nmp->nm_mountp = mp;
1019 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
1021 lwkt_gettoken(&nmp->nm_token);
1024 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too
1025 * high, depending on whether we end up with negative offsets in
1026 * the client or server somewhere. 2GB-1 may be safer.
1028 * For V3, nfs_fsinfo will adjust this as necessary. Assume maximum
1029 * that we can handle until we find out otherwise.
1030 * XXX Our "safe" limit on the client is what we can store in our
1031 * buffer cache using signed(!) block numbers.
1033 if ((argp->flags & NFSMNT_NFSV3) == 0)
1034 nmp->nm_maxfilesize = 0xffffffffLL;
1036 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
1038 nmp->nm_timeo = NFS_TIMEO;
1039 nmp->nm_retry = NFS_RETRANS;
1040 nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
1041 nmp->nm_rsize = nmp->nm_wsize;
1042 nmp->nm_readdirsize = NFS_READDIRSIZE;
1043 nmp->nm_numgrps = NFS_MAXGRPS;
1044 nmp->nm_readahead = NFS_DEFRAHEAD;
1045 nmp->nm_deadthresh = NFS_DEADTHRESH;
1046 nmp->nm_fhsize = argp->fhsize;
1047 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
1048 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
1050 /* Set up the sockets and per-host congestion */
1051 nmp->nm_sotype = argp->sotype;
1052 nmp->nm_soproto = argp->proto;
1053 nmp->nm_cred = crhold(proc0.p_ucred);
1055 nfs_decode_args(nmp, argp);
1058 * For Connection based sockets (TCP,...) defer the connect until
1059 * the first request, in case the server is not responding.
1061 if (nmp->nm_sotype == SOCK_DGRAM &&
1062 (error = nfs_connect(nmp, NULL)))
1066 * This is silly, but it has to be set so that vinifod() works.
1067 * We do not want to do an nfs_statfs() here since we can get
1068 * stuck on a dead server and we are holding a lock on the mount
1071 mp->mnt_stat.f_iosize =
1072 nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype);
1075 * Install vop_ops for our vnops
1077 vfs_add_vnodeops(mp, &nfsv2_vnode_vops, &mp->mnt_vn_norm_ops);
1078 vfs_add_vnodeops(mp, &nfsv2_spec_vops, &mp->mnt_vn_spec_ops);
1079 vfs_add_vnodeops(mp, &nfsv2_fifo_vops, &mp->mnt_vn_fifo_ops);
1082 * A reference count is needed on the nfsnode representing the
1083 * remote root. If this object is not persistent, then backward
1084 * traversals of the mount point (i.e. "..") will not work if
1085 * the nfsnode gets flushed out of the cache. Ufs does not have
1086 * this problem, because one can identify root inodes by their
1087 * number == ROOTINO (2).
1089 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
1095 * Retrieval of mountpoint attributes is delayed until nfs_rot
1096 * or nfs_statfs are first called. This will happen either when
1097 * we first traverse the mount point or if somebody does a df(1).
1099 * NFSSTA_GOTFSINFO is used to flag if we have successfully
1100 * retrieved mountpoint attributes. In the case of NFSv3 we
1101 * also flag static fsinfo.
1104 (*vpp)->v_type = VNON;
1107 * Lose the lock but keep the ref.
1110 lwkt_gettoken(&nfs_token);
1111 TAILQ_INSERT_TAIL(&nfs_mountq, nmp, nm_entry);
1112 lwkt_reltoken(&nfs_token);
1136 * Start the reader and writer threads.
1138 lwkt_create(nfssvc_iod_reader, nmp, &nmp->nm_rxthread,
1139 NULL, 0, rxcpu, "nfsiod_rx");
1140 lwkt_create(nfssvc_iod_writer, nmp, &nmp->nm_txthread,
1141 NULL, 0, txcpu, "nfsiod_tx");
1142 lwkt_reltoken(&nmp->nm_token);
1145 nfs_disconnect(nmp);
1146 lwkt_reltoken(&nmp->nm_token);
1147 nfs_free_mount(nmp);
1152 * unmount system call
1155 nfs_unmount(struct mount *mp, int mntflags)
1157 struct nfsmount *nmp;
1158 int error, flags = 0;
1161 lwkt_gettoken(&nmp->nm_token);
1162 if (mntflags & MNT_FORCE) {
1163 flags |= FORCECLOSE;
1164 nmp->nm_flag |= NFSMNT_FORCE;
1168 * Goes something like this..
1169 * - Call vflush() to clear out vnodes for this file system
1170 * - Close the socket
1171 * - Free up the data structures
1173 /* In the forced case, cancel any outstanding requests. */
1174 if (flags & FORCECLOSE) {
1175 error = nfs_nmcancelreqs(nmp);
1177 kprintf("NFS: %s: Unable to cancel all requests\n",
1178 mp->mnt_stat.f_mntfromname);
1179 /* continue anyway */
1184 * Must handshake with nfs_clientd() if it is active. XXX
1186 nmp->nm_state |= NFSSTA_DISMINPROG;
1189 * We hold 1 extra ref on the root vnode; see comment in mountnfs().
1191 * If this doesn't work and we are doing a forced unmount we continue
1194 error = vflush(mp, 1, flags);
1196 nmp->nm_state &= ~NFSSTA_DISMINPROG;
1197 if ((flags & FORCECLOSE) == 0) {
1198 lwkt_reltoken(&nmp->nm_token);
1204 * We are now committed to the unmount.
1205 * For NQNFS, let the server daemon free the nfsmount structure.
1207 if (nmp->nm_flag & NFSMNT_KERB)
1208 nmp->nm_state |= NFSSTA_DISMNT;
1209 nfssvc_iod_stop1(nmp);
1210 nfs_disconnect(nmp);
1211 nfssvc_iod_stop2(nmp);
1213 lwkt_gettoken(&nfs_token);
1214 TAILQ_REMOVE(&nfs_mountq, nmp, nm_entry);
1215 lwkt_reltoken(&nfs_token);
1217 lwkt_reltoken(&nmp->nm_token);
1219 if ((nmp->nm_flag & NFSMNT_KERB) == 0) {
1220 nfs_free_mount(nmp);
1226 nfs_free_mount(struct nfsmount *nmp)
1229 crfree(nmp->nm_cred);
1230 nmp->nm_cred = NULL;
1233 FREE(nmp->nm_nam, M_SONAME);
1236 zfree(nfsmount_zone, nmp);
1240 * Return root of a filesystem
1243 nfs_root(struct mount *mp, struct vnode **vpp)
1246 struct nfsmount *nmp;
1252 lwkt_gettoken(&nmp->nm_token);
1253 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
1255 lwkt_reltoken(&nmp->nm_token);
1261 * Get transfer parameters and root vnode attributes
1263 * NOTE: nfs_fsinfo() is expected to override the default
1266 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
1267 if (nmp->nm_flag & NFSMNT_NFSV3) {
1268 mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype);
1269 error = nfs_fsinfo(nmp, vp, curthread);
1271 if ((error = VOP_GETATTR(vp, &attrs)) == 0)
1272 nmp->nm_state |= NFSSTA_GOTFSINFO;
1277 * The root vnode is usually cached by the namecache so do not
1278 * try to avoid going over the wire even if we have previous
1279 * information cached. A stale NFS mount can loop
1280 * forever resolving the root vnode if we return no-error when
1281 * there is in fact an error.
1283 np->n_attrstamp = 0;
1284 error = VOP_GETATTR(vp, &attrs);
1286 if (vp->v_type == VNON)
1287 nfs_setvtype(vp, VDIR);
1288 vsetflags(vp, VROOT);
1293 lwkt_reltoken(&nmp->nm_token);
1303 static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
1304 static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
1307 * Flush out the buffer cache
1311 nfs_sync(struct mount *mp, int waitfor)
1313 struct nfsmount *nmp = VFSTONFS(mp);
1314 struct scaninfo scaninfo;
1317 scaninfo.rescan = 1;
1318 scaninfo.waitfor = waitfor;
1319 scaninfo.allerror = 0;
1322 * Force stale buffer cache information to be flushed.
1324 lwkt_gettoken(&nmp->nm_token);
1326 while (error == 0 && scaninfo.rescan) {
1327 scaninfo.rescan = 0;
1328 error = vmntvnodescan(mp, VMSC_GETVP, nfs_sync_scan1,
1329 nfs_sync_scan2, &scaninfo);
1331 lwkt_reltoken(&nmp->nm_token);
1336 nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1338 struct scaninfo *info = data;
1340 if (vn_islocked(vp) || RB_EMPTY(&vp->v_rbdirty_tree))
1342 if (info->waitfor == MNT_LAZY)
1348 nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1350 struct scaninfo *info = data;
1353 error = VOP_FSYNC(vp, info->waitfor, 0);
1355 info->allerror = error;