xref: /csrg-svn/sys/nfs/nfs_syscalls.c (revision 47573)
1 /*
2  * Copyright (c) 1989 The Regents of the University of California.
3  * All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * %sccs.include.redist.c%
9  *
10  *	@(#)nfs_syscalls.c	7.25 (Berkeley) 03/19/91
11  */
12 
13 #include "param.h"
14 #include "systm.h"
15 #include "kernel.h"
16 #include "file.h"
17 #include "stat.h"
18 #include "vnode.h"
19 #include "mount.h"
20 #include "proc.h"
21 #include "malloc.h"
22 #include "buf.h"
23 #include "mbuf.h"
24 #include "socket.h"
25 #include "socketvar.h"
26 #include "domain.h"
27 #include "protosw.h"
28 
29 #include "../netinet/in.h"
30 #include "../netinet/tcp.h"
31 
32 #include "nfsv2.h"
33 #include "nfs.h"
34 #include "nfsrvcache.h"
35 
36 /* Global defs. */
37 extern u_long nfs_prog, nfs_vers;
38 extern int (*nfsrv_procs[NFS_NPROCS])();
39 extern struct buf nfs_bqueue;
40 extern int nfs_numasync;
41 extern struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON];
42 extern int nfs_tcpnodelay;
43 struct mbuf *nfs_compress();
44 
45 #define	TRUE	1
46 #define	FALSE	0
47 
48 static int nfs_asyncdaemon[NFS_MAXASYNCDAEMON];
49 static int compressreply[NFS_NPROCS] = {
50 	FALSE,
51 	TRUE,
52 	TRUE,
53 	FALSE,
54 	TRUE,
55 	TRUE,
56 	FALSE,
57 	FALSE,
58 	TRUE,
59 	TRUE,
60 	TRUE,
61 	TRUE,
62 	TRUE,
63 	TRUE,
64 	TRUE,
65 	TRUE,
66 	TRUE,
67 	TRUE,
68 };
69 /*
70  * NFS server system calls
71  * getfh() lives here too, but maybe should move to kern/vfs_syscalls.c
72  */
73 
74 /*
75  * Get file handle system call
76  */
77 /* ARGSUSED */
78 getfh(p, uap, retval)
79 	struct proc *p;
80 	register struct args {
81 		char	*fname;
82 		fhandle_t *fhp;
83 	} *uap;
84 	int *retval;
85 {
86 	register struct nameidata *ndp;
87 	register struct vnode *vp;
88 	fhandle_t fh;
89 	int error;
90 	struct nameidata nd;
91 
92 	/*
93 	 * Must be super user
94 	 */
95 	if (error = suser(p->p_ucred, &p->p_acflag))
96 		return (error);
97 	ndp = &nd;
98 	ndp->ni_nameiop = LOOKUP | LOCKLEAF | FOLLOW;
99 	ndp->ni_segflg = UIO_USERSPACE;
100 	ndp->ni_dirp = uap->fname;
101 	if (error = namei(ndp, p))
102 		return (error);
103 	vp = ndp->ni_vp;
104 	bzero((caddr_t)&fh, sizeof(fh));
105 	fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
106 	error = VFS_VPTOFH(vp, &fh.fh_fid);
107 	vput(vp);
108 	if (error)
109 		return (error);
110 	error = copyout((caddr_t)&fh, (caddr_t)uap->fhp, sizeof (fh));
111 	return (error);
112 }
113 
114 /*
115  * Nfs server psuedo system call for the nfsd's
116  * Never returns unless it fails or gets killed
117  */
118 /* ARGSUSED */
119 nfssvc(p, uap, retval)
120 	struct proc *p;
121 	register struct args {
122 		int s;
123 		caddr_t mskval;
124 		int msklen;
125 		caddr_t mtchval;
126 		int mtchlen;
127 	} *uap;
128 	int *retval;
129 {
130 	register struct mbuf *m;
131 	register int siz;
132 	register struct ucred *cr;
133 	struct file *fp;
134 	struct mbuf *mreq, *mrep, *nam, *md;
135 	struct mbuf msk, mtch;
136 	struct socket *so;
137 	caddr_t dpos;
138 	int procid, repstat, error, cacherep, wascomp;
139 	u_long retxid;
140 
141 	/*
142 	 * Must be super user
143 	 */
144 	if (error = suser(p->p_ucred, &p->p_acflag))
145 		return (error);
146 	if (error = getsock(p->p_fd, uap->s, &fp))
147 		return (error);
148 	so = (struct socket *)fp->f_data;
149 	if (sosendallatonce(so))
150 		siz = NFS_MAXPACKET;
151 	else
152 		siz = NFS_MAXPACKET + sizeof(u_long);
153 	if (error = soreserve(so, siz, siz))
154 		goto bad;
155 	if (error = sockargs(&nam, uap->mskval, uap->msklen, MT_SONAME))
156 		goto bad;
157 	bcopy((caddr_t)nam, (caddr_t)&msk, sizeof (struct mbuf));
158 	msk.m_data = msk.m_dat;
159 	m_freem(nam);
160 	if (error = sockargs(&nam, uap->mtchval, uap->mtchlen, MT_SONAME))
161 		goto bad;
162 	bcopy((caddr_t)nam, (caddr_t)&mtch, sizeof (struct mbuf));
163 	mtch.m_data = mtch.m_dat;
164 	m_freem(nam);
165 
166 	/* Copy the cred so others don't see changes */
167 	cr = p->p_ucred = crcopy(p->p_ucred);
168 
169 	/*
170 	 * Set protocol specific options { for now TCP only } and
171 	 * reserve some space. For datagram sockets, this can get called
172 	 * repeatedly for the same socket, but that isn't harmful.
173 	 */
174 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
175 		MGET(m, M_WAIT, MT_SOOPTS);
176 		*mtod(m, int *) = 1;
177 		m->m_len = sizeof(int);
178 		sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m);
179 	}
180 	if (so->so_proto->pr_domain->dom_family == AF_INET &&
181 	    so->so_proto->pr_protocol == IPPROTO_TCP &&
182 	    nfs_tcpnodelay) {
183 		MGET(m, M_WAIT, MT_SOOPTS);
184 		*mtod(m, int *) = 1;
185 		m->m_len = sizeof(int);
186 		sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m);
187 	}
188 	so->so_rcv.sb_flags &= ~SB_NOINTR;
189 	so->so_rcv.sb_timeo = 0;
190 	so->so_snd.sb_flags &= ~SB_NOINTR;
191 	so->so_snd.sb_timeo = 0;
192 
193 	/*
194 	 * Just loop around doin our stuff until SIGKILL
195 	 */
196 	for (;;) {
197 		if (error = nfs_getreq(so, nfs_prog, nfs_vers, NFS_NPROCS-1,
198 		   &nam, &mrep, &md, &dpos, &retxid, &procid, cr,
199 		   &msk, &mtch, &wascomp)) {
200 			if (nam)
201 				m_freem(nam);
202 			if (error == EPIPE || error == EINTR ||
203 			    error == ERESTART) {
204 				error = 0;
205 				goto bad;
206 			}
207 			so->so_error = 0;
208 			continue;
209 		}
210 
211 		if (nam)
212 			cacherep = nfsrv_getcache(nam, retxid, procid, &mreq);
213 		else
214 			cacherep = RC_DOIT;
215 		switch (cacherep) {
216 		case RC_DOIT:
217 			if (error = (*(nfsrv_procs[procid]))(mrep, md, dpos,
218 				cr, retxid, &mreq, &repstat)) {
219 				nfsstats.srv_errs++;
220 				if (nam) {
221 					nfsrv_updatecache(nam, retxid, procid,
222 						FALSE, repstat, mreq);
223 					m_freem(nam);
224 				}
225 				break;
226 			}
227 			nfsstats.srvrpccnt[procid]++;
228 			if (nam)
229 				nfsrv_updatecache(nam, retxid, procid, TRUE,
230 					repstat, mreq);
231 			mrep = (struct mbuf *)0;
232 		case RC_REPLY:
233 			m = mreq;
234 			siz = 0;
235 			while (m) {
236 				siz += m->m_len;
237 				m = m->m_next;
238 			}
239 			if (siz <= 0 || siz > NFS_MAXPACKET) {
240 				printf("mbuf siz=%d\n",siz);
241 				panic("Bad nfs svc reply");
242 			}
243 			mreq->m_pkthdr.len = siz;
244 			mreq->m_pkthdr.rcvif = (struct ifnet *)0;
245 			if (wascomp && compressreply[procid]) {
246 				mreq = nfs_compress(mreq);
247 				siz = mreq->m_pkthdr.len;
248 			}
249 			/*
250 			 * For non-atomic protocols, prepend a Sun RPC
251 			 * Record Mark.
252 			 */
253 			if (!sosendallatonce(so)) {
254 				M_PREPEND(mreq, sizeof(u_long), M_WAIT);
255 				*mtod(mreq, u_long *) = htonl(0x80000000 | siz);
256 			}
257 			error = nfs_send(so, nam, mreq, (struct nfsreq *)0);
258 			if (nam)
259 				m_freem(nam);
260 			if (mrep)
261 				m_freem(mrep);
262 			if (error) {
263 				if (error == EPIPE || error == EINTR ||
264 				    error == ERESTART)
265 					goto bad;
266 				so->so_error = 0;
267 			}
268 			break;
269 		case RC_DROPIT:
270 			m_freem(mrep);
271 			m_freem(nam);
272 			break;
273 		};
274 	}
275 bad:
276 	return (error);
277 }
278 
279 /*
280  * Nfs pseudo system call for asynchronous i/o daemons.
281  * These babies just pretend to be disk interrupt service routines
282  * for client nfs. They are mainly here for read ahead/write behind.
283  * Never returns unless it fails or gets killed
284  */
285 /* ARGSUSED */
286 async_daemon(p, uap, retval)
287 	struct proc *p;
288 	struct args *uap;
289 	int *retval;
290 {
291 	register struct buf *bp, *dp;
292 	register int i, myiod;
293 	int error;
294 
295 	/*
296 	 * Must be super user
297 	 */
298 	if (error = suser(p->p_ucred, &p->p_acflag))
299 		return (error);
300 	/*
301 	 * Assign my position or return error if too many already running
302 	 */
303 	myiod = -1;
304 	for (i = 0; i < NFS_MAXASYNCDAEMON; i++)
305 		if (nfs_asyncdaemon[i] == 0) {
306 			nfs_asyncdaemon[i]++;
307 			myiod = i;
308 			break;
309 		}
310 	if (myiod == -1)
311 		return (EBUSY);
312 	nfs_numasync++;
313 	dp = &nfs_bqueue;
314 	/*
315 	 * Just loop around doin our stuff until SIGKILL
316 	 */
317 	for (;;) {
318 		while (dp->b_actf == NULL && error == 0) {
319 			nfs_iodwant[myiod] = curproc;
320 			error = tsleep((caddr_t)&nfs_iodwant[myiod],
321 				PWAIT | PCATCH, "nfsidl", 0);
322 			nfs_iodwant[myiod] = (struct proc *)0;
323 		}
324 		while (dp->b_actf != NULL) {
325 			/* Take one off the end of the list */
326 			bp = dp->b_actl;
327 			if (bp->b_actl == dp) {
328 				dp->b_actf = dp->b_actl = (struct buf *)0;
329 			} else {
330 				dp->b_actl = bp->b_actl;
331 				bp->b_actl->b_actf = dp;
332 			}
333 			(void) nfs_doio(bp);
334 		}
335 		if (error) {
336 			nfs_asyncdaemon[myiod] = 0;
337 			nfs_numasync--;
338 			return (error);
339 		}
340 	}
341 }
342