xref: /csrg-svn/sys/nfs/nfs_vnops.c (revision 46981)
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_vnops.c	7.53 (Berkeley) 03/04/91
11  */
12 
13 /*
14  * vnode op calls for sun nfs version 2
15  */
16 
17 #include "machine/mtpr.h"
18 #include "param.h"
19 #include "user.h"
20 #include "proc.h"
21 #include "kernel.h"
22 #include "systm.h"
23 #include "mount.h"
24 #include "buf.h"
25 #include "malloc.h"
26 #include "mbuf.h"
27 #include "errno.h"
28 #include "file.h"
29 #include "conf.h"
30 #include "vnode.h"
31 #include "map.h"
32 #include "../ufs/quota.h"
33 #include "../ufs/inode.h"
34 #include "nfsv2.h"
35 #include "nfs.h"
36 #include "nfsnode.h"
37 #include "nfsmount.h"
38 #include "xdr_subs.h"
39 #include "nfsm_subs.h"
40 #include "nfsiom.h"
41 
42 /* Defs */
43 #define	TRUE	1
44 #define	FALSE	0
45 
46 /* Global vars */
47 int	nfs_lookup(),
48 	nfs_create(),
49 	nfs_mknod(),
50 	nfs_open(),
51 	nfs_close(),
52 	nfs_access(),
53 	nfs_getattr(),
54 	nfs_setattr(),
55 	nfs_read(),
56 	nfs_write(),
57 	nfs_remove(),
58 	nfs_link(),
59 	nfs_rename(),
60 	nfs_mkdir(),
61 	nfs_rmdir(),
62 	nfs_symlink(),
63 	nfs_readdir(),
64 	nfs_readlink(),
65 	nfs_abortop(),
66 	nfs_lock(),
67 	nfs_unlock(),
68 	nfs_bmap(),
69 	nfs_strategy(),
70 	nfs_fsync(),
71 	nfs_inactive(),
72 	nfs_reclaim(),
73 	nfs_print(),
74 	nfs_islocked(),
75 	nfs_advlock(),
76 	eopnotsupp(),
77 	seltrue();
78 
79 struct vnodeops nfsv2_vnodeops = {
80 	nfs_lookup,		/* lookup */
81 	nfs_create,		/* create */
82 	nfs_mknod,		/* mknod */
83 	nfs_open,		/* open */
84 	nfs_close,		/* close */
85 	nfs_access,		/* access */
86 	nfs_getattr,		/* getattr */
87 	nfs_setattr,		/* setattr */
88 	nfs_read,		/* read */
89 	nfs_write,		/* write */
90 	eopnotsupp,		/* ioctl */
91 	seltrue,		/* select */
92 	eopnotsupp,		/* mmap */
93 	nfs_fsync,		/* fsync */
94 	nullop,			/* seek */
95 	nfs_remove,		/* remove */
96 	nfs_link,		/* link */
97 	nfs_rename,		/* rename */
98 	nfs_mkdir,		/* mkdir */
99 	nfs_rmdir,		/* rmdir */
100 	nfs_symlink,		/* symlink */
101 	nfs_readdir,		/* readdir */
102 	nfs_readlink,		/* readlink */
103 	nfs_abortop,		/* abortop */
104 	nfs_inactive,		/* inactive */
105 	nfs_reclaim,		/* reclaim */
106 	nfs_lock,		/* lock */
107 	nfs_unlock,		/* unlock */
108 	nfs_bmap,		/* bmap */
109 	nfs_strategy,		/* strategy */
110 	nfs_print,		/* print */
111 	nfs_islocked,		/* islocked */
112 	nfs_advlock,		/* advlock */
113 };
114 
115 /* Special device vnode ops */
116 int	spec_lookup(),
117 	spec_open(),
118 	spec_read(),
119 	spec_write(),
120 	spec_strategy(),
121 	spec_bmap(),
122 	spec_ioctl(),
123 	spec_select(),
124 	spec_close(),
125 	spec_badop(),
126 	spec_advlock();
127 
128 struct vnodeops spec_nfsv2nodeops = {
129 	spec_lookup,		/* lookup */
130 	spec_badop,		/* create */
131 	spec_badop,		/* mknod */
132 	spec_open,		/* open */
133 	spec_close,		/* close */
134 	nfs_access,		/* access */
135 	nfs_getattr,		/* getattr */
136 	nfs_setattr,		/* setattr */
137 	spec_read,		/* read */
138 	spec_write,		/* write */
139 	spec_ioctl,		/* ioctl */
140 	spec_select,		/* select */
141 	spec_badop,		/* mmap */
142 	nullop,			/* fsync */
143 	spec_badop,		/* seek */
144 	spec_badop,		/* remove */
145 	spec_badop,		/* link */
146 	spec_badop,		/* rename */
147 	spec_badop,		/* mkdir */
148 	spec_badop,		/* rmdir */
149 	spec_badop,		/* symlink */
150 	spec_badop,		/* readdir */
151 	spec_badop,		/* readlink */
152 	spec_badop,		/* abortop */
153 	nfs_inactive,		/* inactive */
154 	nfs_reclaim,		/* reclaim */
155 	nfs_lock,		/* lock */
156 	nfs_unlock,		/* unlock */
157 	spec_bmap,		/* bmap */
158 	spec_strategy,		/* strategy */
159 	nfs_print,		/* print */
160 	nfs_islocked,		/* islocked */
161 	spec_advlock,		/* advlock */
162 };
163 
164 #ifdef FIFO
165 int	fifo_lookup(),
166 	fifo_open(),
167 	fifo_read(),
168 	fifo_write(),
169 	fifo_bmap(),
170 	fifo_ioctl(),
171 	fifo_select(),
172 	fifo_close(),
173 	fifo_print(),
174 	fifo_badop(),
175 	fifo_advlock();
176 
177 struct vnodeops fifo_nfsv2nodeops = {
178 	fifo_lookup,		/* lookup */
179 	fifo_badop,		/* create */
180 	fifo_badop,		/* mknod */
181 	fifo_open,		/* open */
182 	fifo_close,		/* close */
183 	nfs_access,		/* access */
184 	nfs_getattr,		/* getattr */
185 	nfs_setattr,		/* setattr */
186 	fifo_read,		/* read */
187 	fifo_write,		/* write */
188 	fifo_ioctl,		/* ioctl */
189 	fifo_select,		/* select */
190 	fifo_badop,		/* mmap */
191 	nullop,			/* fsync */
192 	fifo_badop,		/* seek */
193 	fifo_badop,		/* remove */
194 	fifo_badop,		/* link */
195 	fifo_badop,		/* rename */
196 	fifo_badop,		/* mkdir */
197 	fifo_badop,		/* rmdir */
198 	fifo_badop,		/* symlink */
199 	fifo_badop,		/* readdir */
200 	fifo_badop,		/* readlink */
201 	fifo_badop,		/* abortop */
202 	nfs_inactive,		/* inactive */
203 	nfs_reclaim,		/* reclaim */
204 	nfs_lock,		/* lock */
205 	nfs_unlock,		/* unlock */
206 	fifo_bmap,		/* bmap */
207 	fifo_badop,		/* strategy */
208 	nfs_print,		/* print */
209 	nfs_islocked,		/* islocked */
210 	fifo_advlock,		/* advlock */
211 };
212 #endif /* FIFO */
213 
214 extern u_long nfs_procids[NFS_NPROCS];
215 extern u_long nfs_prog, nfs_vers;
216 extern char nfsiobuf[MAXPHYS+NBPG];
217 struct map nfsmap[NFS_MSIZ];
218 struct buf nfs_bqueue;		/* Queue head for nfsiod's */
219 int nfs_asyncdaemons = 0;
220 struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON];
221 static int nfsmap_want = 0;
222 
223 /*
224  * nfs null call from vfs.
225  */
226 nfs_null(vp, cred)
227 	struct vnode *vp;
228 	struct ucred *cred;
229 {
230 	caddr_t bpos, dpos;
231 	u_long xid;
232 	int error = 0;
233 	struct mbuf *mreq, *mrep, *md, *mb;
234 
235 	nfsm_reqhead(nfs_procids[NFSPROC_NULL], cred, 0);
236 	nfsm_request(vp, NFSPROC_NULL, u.u_procp, 0);
237 	nfsm_reqdone;
238 	return (error);
239 }
240 
241 /*
242  * nfs access vnode op.
243  * Essentially just get vattr and then imitate iaccess()
244  */
245 nfs_access(vp, mode, cred)
246 	struct vnode *vp;
247 	int mode;
248 	register struct ucred *cred;
249 {
250 	register struct vattr *vap;
251 	register gid_t *gp;
252 	struct vattr vattr;
253 	register int i;
254 	int error;
255 
256 	/*
257 	 * If you're the super-user,
258 	 * you always get access.
259 	 */
260 	if (cred->cr_uid == 0)
261 		return (0);
262 	vap = &vattr;
263 	if (error = nfs_dogetattr(vp, vap, cred, 0))
264 		return (error);
265 	/*
266 	 * Access check is based on only one of owner, group, public.
267 	 * If not owner, then check group. If not a member of the
268 	 * group, then check public access.
269 	 */
270 	if (cred->cr_uid != vap->va_uid) {
271 		mode >>= 3;
272 		gp = cred->cr_groups;
273 		for (i = 0; i < cred->cr_ngroups; i++, gp++)
274 			if (vap->va_gid == *gp)
275 				goto found;
276 		mode >>= 3;
277 found:
278 		;
279 	}
280 	if ((vap->va_mode & mode) != 0)
281 		return (0);
282 	return (EACCES);
283 }
284 
285 /*
286  * nfs open vnode op
287  * Just check to see if the type is ok
288  */
289 /* ARGSUSED */
290 nfs_open(vp, mode, cred)
291 	struct vnode *vp;
292 	int mode;
293 	struct ucred *cred;
294 {
295 	register enum vtype vtyp;
296 
297 	vtyp = vp->v_type;
298 	if (vtyp == VREG || vtyp == VDIR || vtyp == VLNK)
299 		return (0);
300 	else
301 		return (EACCES);
302 }
303 
304 /*
305  * nfs close vnode op
306  * For reg files, invalidate any buffer cache entries.
307  */
308 /* ARGSUSED */
309 nfs_close(vp, fflags, cred)
310 	register struct vnode *vp;
311 	int fflags;
312 	struct ucred *cred;
313 {
314 	register struct nfsnode *np = VTONFS(vp);
315 	int error = 0;
316 
317 	if (vp->v_type == VREG && (np->n_flag & NMODIFIED)) {
318 		nfs_lock(vp);
319 		np->n_flag &= ~NMODIFIED;
320 		vinvalbuf(vp, TRUE);
321 		np->n_attrstamp = 0;
322 		if (np->n_flag & NWRITEERR) {
323 			np->n_flag &= ~NWRITEERR;
324 			error = np->n_error;
325 		}
326 		nfs_unlock(vp);
327 	}
328 	return (error);
329 }
330 
331 /*
332  * nfs getattr call from vfs.
333  */
334 nfs_getattr(vp, vap, cred)
335 	register struct vnode *vp;
336 	struct vattr *vap;
337 	struct ucred *cred;
338 {
339 	return (nfs_dogetattr(vp, vap, cred, 0));
340 }
341 
342 nfs_dogetattr(vp, vap, cred, tryhard)
343 	register struct vnode *vp;
344 	struct vattr *vap;
345 	struct ucred *cred;
346 	int tryhard;
347 {
348 	register caddr_t cp;
349 	register long t1;
350 	caddr_t bpos, dpos;
351 	u_long xid;
352 	int error = 0;
353 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
354 
355 	/* First look in the cache.. */
356 	if (nfs_getattrcache(vp, vap) == 0)
357 		return (0);
358 	nfsstats.rpccnt[NFSPROC_GETATTR]++;
359 	nfsm_reqhead(nfs_procids[NFSPROC_GETATTR], cred, NFSX_FH);
360 	nfsm_fhtom(vp);
361 	nfsm_request(vp, NFSPROC_GETATTR, u.u_procp, tryhard);
362 	nfsm_loadattr(vp, vap);
363 	nfsm_reqdone;
364 	return (error);
365 }
366 
367 /*
368  * nfs setattr call.
369  */
370 nfs_setattr(vp, vap, cred)
371 	register struct vnode *vp;
372 	register struct vattr *vap;
373 	struct ucred *cred;
374 {
375 	register struct nfsv2_sattr *sp;
376 	register caddr_t cp;
377 	register long t1;
378 	caddr_t bpos, dpos;
379 	u_long xid;
380 	int error = 0;
381 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
382 	struct nfsnode *np;
383 
384 	nfsstats.rpccnt[NFSPROC_SETATTR]++;
385 	nfsm_reqhead(nfs_procids[NFSPROC_SETATTR], cred, NFSX_FH+NFSX_SATTR);
386 	nfsm_fhtom(vp);
387 	nfsm_build(sp, struct nfsv2_sattr *, NFSX_SATTR);
388 	if (vap->va_mode == 0xffff)
389 		sp->sa_mode = VNOVAL;
390 	else
391 		sp->sa_mode = vtonfs_mode(vp->v_type, vap->va_mode);
392 	if (vap->va_uid == 0xffff)
393 		sp->sa_uid = VNOVAL;
394 	else
395 		sp->sa_uid = txdr_unsigned(vap->va_uid);
396 	if (vap->va_gid == 0xffff)
397 		sp->sa_gid = VNOVAL;
398 	else
399 		sp->sa_gid = txdr_unsigned(vap->va_gid);
400 	sp->sa_size = txdr_unsigned(vap->va_size);
401 	sp->sa_atime.tv_sec = txdr_unsigned(vap->va_atime.tv_sec);
402 	sp->sa_atime.tv_usec = txdr_unsigned(vap->va_flags);
403 	txdr_time(&vap->va_mtime, &sp->sa_mtime);
404 	if (vap->va_size != VNOVAL || vap->va_mtime.tv_sec != VNOVAL ||
405 	    vap->va_atime.tv_sec != VNOVAL) {
406 		np = VTONFS(vp);
407 		if (np->n_flag & NMODIFIED) {
408 			np->n_flag &= ~NMODIFIED;
409 			vinvalbuf(vp, TRUE);
410 			np->n_attrstamp = 0;
411 		}
412 	}
413 	nfsm_request(vp, NFSPROC_SETATTR, u.u_procp, 1);
414 	nfsm_loadattr(vp, (struct vattr *)0);
415 	/* should we fill in any vap fields ?? */
416 	nfsm_reqdone;
417 	return (error);
418 }
419 
420 /*
421  * nfs lookup call, one step at a time...
422  * First look in cache
423  * If not found, unlock the directory nfsnode and do the rpc
424  */
425 nfs_lookup(vp, ndp)
426 	register struct vnode *vp;
427 	register struct nameidata *ndp;
428 {
429 	register struct vnode *vdp;
430 	register u_long *p;
431 	register caddr_t cp;
432 	register long t1, t2;
433 	caddr_t bpos, dpos, cp2;
434 	u_long xid;
435 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
436 	struct vnode *newvp;
437 	long len;
438 	nfsv2fh_t *fhp;
439 	struct nfsnode *np;
440 	int lockparent, wantparent, flag, error = 0;
441 
442 	ndp->ni_dvp = vp;
443 	ndp->ni_vp = NULL;
444 	if (vp->v_type != VDIR)
445 		return (ENOTDIR);
446 	lockparent = ndp->ni_nameiop & LOCKPARENT;
447 	flag = ndp->ni_nameiop & OPMASK;
448 	wantparent = ndp->ni_nameiop & (LOCKPARENT|WANTPARENT);
449 	if ((error = cache_lookup(ndp)) && error != ENOENT) {
450 		struct vattr vattr;
451 		int vpid;
452 
453 #ifdef PARANOID
454 		if (vp == ndp->ni_rdir && ndp->ni_isdotdot)
455 			panic("nfs_lookup: .. through root");
456 #endif
457 		vdp = ndp->ni_vp;
458 		vpid = vdp->v_id;
459 		/*
460 		 * See the comment starting `Step through' in ufs/ufs_lookup.c
461 		 * for an explanation of the locking protocol
462 		 */
463 		if (vp == vdp) {
464 			VREF(vdp);
465 			error = 0;
466 		} else if (ndp->ni_isdotdot) {
467 			nfs_unlock(vp);
468 			error = vget(vdp);
469 		} else {
470 			error = vget(vdp);
471 			nfs_unlock(vp);
472 		}
473 		if (!error) {
474 			if (vpid == vdp->v_id) {
475 			   if (!nfs_dogetattr(vdp, &vattr, ndp->ni_cred, 0) &&
476 			       vattr.va_ctime.tv_sec == VTONFS(vdp)->n_ctime) {
477 				nfsstats.lookupcache_hits++;
478 				return (0);
479 			   } else {
480 				cache_purge(vdp);
481 				nfs_nput(vdp);
482 			   }
483 			} else {
484 				nfs_nput(vdp);
485 			}
486 		}
487 		ndp->ni_vp = NULLVP;
488 	} else
489 		nfs_unlock(vp);
490 	error = 0;
491 	nfsstats.lookupcache_misses++;
492 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
493 	len = ndp->ni_namelen;
494 	nfsm_reqhead(nfs_procids[NFSPROC_LOOKUP], ndp->ni_cred, NFSX_FH+NFSX_UNSIGNED+nfsm_rndup(len));
495 	nfsm_fhtom(vp);
496 	nfsm_strtom(ndp->ni_ptr, len, NFS_MAXNAMLEN);
497 	nfsm_request(vp, NFSPROC_LOOKUP, u.u_procp, 0);
498 nfsmout:
499 	if (error) {
500 		if (lockparent || (flag != CREATE && flag != RENAME) ||
501 		    *ndp->ni_next != 0)
502 			nfs_lock(vp);
503 		return (error);
504 	}
505 	nfsm_disect(fhp,nfsv2fh_t *,NFSX_FH);
506 
507 	/*
508 	 * Handle DELETE and RENAME cases...
509 	 */
510 	if (flag == DELETE && *ndp->ni_next == 0) {
511 		if (!bcmp(VTONFS(vp)->n_fh.fh_bytes, (caddr_t)fhp, NFSX_FH)) {
512 			VREF(vp);
513 			newvp = vp;
514 			np = VTONFS(vp);
515 		} else {
516 			if (error = nfs_nget(vp->v_mount, fhp, &np)) {
517 				nfs_lock(vp);
518 				m_freem(mrep);
519 				return (error);
520 			}
521 			newvp = NFSTOV(np);
522 		}
523 		if (error =
524 		    nfs_loadattrcache(&newvp, &md, &dpos, (struct vattr *)0)) {
525 			nfs_lock(vp);
526 			if (newvp != vp)
527 				nfs_nput(newvp);
528 			else
529 				vrele(vp);
530 			m_freem(mrep);
531 			return (error);
532 		}
533 		ndp->ni_vp = newvp;
534 		if (lockparent || vp == newvp)
535 			nfs_lock(vp);
536 		m_freem(mrep);
537 		return (0);
538 	}
539 
540 	if (flag == RENAME && wantparent && *ndp->ni_next == 0) {
541 		if (!bcmp(VTONFS(vp)->n_fh.fh_bytes, (caddr_t)fhp, NFSX_FH)) {
542 			nfs_lock(vp);
543 			m_freem(mrep);
544 			return (EISDIR);
545 		}
546 		if (error = nfs_nget(vp->v_mount, fhp, &np)) {
547 			nfs_lock(vp);
548 			m_freem(mrep);
549 			return (error);
550 		}
551 		newvp = NFSTOV(np);
552 		if (error =
553 		    nfs_loadattrcache(&newvp, &md, &dpos, (struct vattr *)0)) {
554 			nfs_lock(vp);
555 			nfs_nput(newvp);
556 			m_freem(mrep);
557 			return (error);
558 		}
559 		ndp->ni_vp = newvp;
560 		if (lockparent)
561 			nfs_lock(vp);
562 		m_freem(mrep);
563 		return (0);
564 	}
565 
566 	if (!bcmp(VTONFS(vp)->n_fh.fh_bytes, (caddr_t)fhp, NFSX_FH)) {
567 		VREF(vp);
568 		newvp = vp;
569 		np = VTONFS(vp);
570 	} else if (ndp->ni_isdotdot) {
571 		if (error = nfs_nget(vp->v_mount, fhp, &np)) {
572 			nfs_lock(vp);
573 			m_freem(mrep);
574 			return (error);
575 		}
576 		newvp = NFSTOV(np);
577 	} else {
578 		if (error = nfs_nget(vp->v_mount, fhp, &np)) {
579 			nfs_lock(vp);
580 			m_freem(mrep);
581 			return (error);
582 		}
583 		newvp = NFSTOV(np);
584 	}
585 	if (error = nfs_loadattrcache(&newvp, &md, &dpos, (struct vattr *)0)) {
586 		nfs_lock(vp);
587 		if (newvp != vp)
588 			nfs_nput(newvp);
589 		else
590 			vrele(vp);
591 		m_freem(mrep);
592 		return (error);
593 	}
594 	m_freem(mrep);
595 
596 	if (vp == newvp || (lockparent && *ndp->ni_next == '\0'))
597 		nfs_lock(vp);
598 	ndp->ni_vp = newvp;
599 	if (error == 0 && ndp->ni_makeentry) {
600 		np->n_ctime = np->n_vattr.va_ctime.tv_sec;
601 		cache_enter(ndp);
602 	}
603 	return (error);
604 }
605 
606 /*
607  * nfs read call.
608  * Just call nfs_bioread() to do the work.
609  */
610 nfs_read(vp, uiop, ioflag, cred)
611 	register struct vnode *vp;
612 	struct uio *uiop;
613 	int ioflag;
614 	struct ucred *cred;
615 {
616 	if (vp->v_type != VREG)
617 		return (EPERM);
618 	return (nfs_bioread(vp, uiop, ioflag, cred));
619 }
620 
621 /*
622  * nfs readlink call
623  */
624 nfs_readlink(vp, uiop, cred)
625 	struct vnode *vp;
626 	struct uio *uiop;
627 	struct ucred *cred;
628 {
629 	if (vp->v_type != VLNK)
630 		return (EPERM);
631 	return (nfs_bioread(vp, uiop, 0, cred));
632 }
633 
634 /*
635  * Do a readlink rpc.
636  * Called by nfs_doio() from below the buffer cache.
637  */
638 nfs_readlinkrpc(vp, uiop, cred, procp)
639 	register struct vnode *vp;
640 	struct uio *uiop;
641 	struct ucred *cred;
642 	struct proc *procp;
643 {
644 	register u_long *p;
645 	register caddr_t cp;
646 	register long t1;
647 	caddr_t bpos, dpos, cp2;
648 	u_long xid;
649 	int error = 0;
650 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
651 	long len;
652 
653 	nfsstats.rpccnt[NFSPROC_READLINK]++;
654 	nfs_unlock(vp);
655 	nfsm_reqhead(nfs_procids[NFSPROC_READLINK], cred, NFSX_FH);
656 	nfsm_fhtom(vp);
657 	nfsm_request(vp, NFSPROC_READLINK, procp, 0);
658 	nfsm_strsiz(len, NFS_MAXPATHLEN);
659 	nfsm_mtouio(uiop, len);
660 	nfsm_reqdone;
661 	nfs_lock(vp);
662 	return (error);
663 }
664 
665 /*
666  * nfs read rpc call
667  * Ditto above
668  */
669 nfs_readrpc(vp, uiop, cred, procp)
670 	register struct vnode *vp;
671 	struct uio *uiop;
672 	struct ucred *cred;
673 	struct proc *procp;
674 {
675 	register u_long *p;
676 	register caddr_t cp;
677 	register long t1;
678 	caddr_t bpos, dpos, cp2;
679 	u_long xid;
680 	int error = 0;
681 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
682 	struct nfsmount *nmp;
683 	long len, retlen, tsiz;
684 
685 	nmp = VFSTONFS(vp->v_mount);
686 	tsiz = uiop->uio_resid;
687 	while (tsiz > 0) {
688 		nfsstats.rpccnt[NFSPROC_READ]++;
689 		len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
690 		nfsm_reqhead(nfs_procids[NFSPROC_READ], cred, NFSX_FH+NFSX_UNSIGNED*3);
691 		nfsm_fhtom(vp);
692 		nfsm_build(p, u_long *, NFSX_UNSIGNED*3);
693 		*p++ = txdr_unsigned(uiop->uio_offset);
694 		*p++ = txdr_unsigned(len);
695 		*p = 0;
696 		nfsm_request(vp, NFSPROC_READ, procp, 1);
697 		nfsm_loadattr(vp, (struct vattr *)0);
698 		nfsm_strsiz(retlen, nmp->nm_rsize);
699 		nfsm_mtouio(uiop, retlen);
700 		m_freem(mrep);
701 		if (retlen < len)
702 			tsiz = 0;
703 		else
704 			tsiz -= len;
705 	}
706 nfsmout:
707 	return (error);
708 }
709 
710 /*
711  * nfs write call
712  */
713 nfs_writerpc(vp, uiop, cred, procp)
714 	register struct vnode *vp;
715 	struct uio *uiop;
716 	struct ucred *cred;
717 	struct proc *procp;
718 {
719 	register u_long *p;
720 	register caddr_t cp;
721 	register long t1;
722 	caddr_t bpos, dpos;
723 	u_long xid;
724 	int error = 0;
725 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
726 	struct nfsmount *nmp;
727 	long len, tsiz;
728 
729 	nmp = VFSTONFS(vp->v_mount);
730 	tsiz = uiop->uio_resid;
731 	while (tsiz > 0) {
732 		nfsstats.rpccnt[NFSPROC_WRITE]++;
733 		len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
734 		nfsm_reqhead(nfs_procids[NFSPROC_WRITE], cred,
735 			NFSX_FH+NFSX_UNSIGNED*4);
736 		nfsm_fhtom(vp);
737 		nfsm_build(p, u_long *, NFSX_UNSIGNED*4);
738 		*(p+1) = txdr_unsigned(uiop->uio_offset);
739 		*(p+3) = txdr_unsigned(len);
740 		nfsm_uiotom(uiop, len);
741 		nfsm_request(vp, NFSPROC_WRITE, procp, 1);
742 		nfsm_loadattr(vp, (struct vattr *)0);
743 		m_freem(mrep);
744 		tsiz -= len;
745 	}
746 nfsmout:
747 	return (error);
748 }
749 
750 /*
751  * nfs mknod call
752  * This is a kludge. Use a create rpc but with the IFMT bits of the mode
753  * set to specify the file type and the size field for rdev.
754  */
755 /* ARGSUSED */
756 nfs_mknod(ndp, vap, cred)
757 	struct nameidata *ndp;
758 	struct ucred *cred;
759 	register struct vattr *vap;
760 {
761 	register struct nfsv2_sattr *sp;
762 	register u_long *p;
763 	register caddr_t cp;
764 	register long t1, t2;
765 	caddr_t bpos, dpos;
766 	u_long xid;
767 	int error = 0;
768 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
769 	u_long rdev;
770 
771 	if (vap->va_type == VCHR || vap->va_type == VBLK)
772 		rdev = txdr_unsigned(vap->va_rdev);
773 #ifdef FIFO
774 	else if (vap->va_type == VFIFO)
775 		rdev = 0xffffffff;
776 #endif /* FIFO */
777 	else {
778 		VOP_ABORTOP(ndp);
779 		vput(ndp->ni_dvp);
780 		return (EOPNOTSUPP);
781 	}
782 	nfsstats.rpccnt[NFSPROC_CREATE]++;
783 	nfsm_reqhead(nfs_procids[NFSPROC_CREATE], ndp->ni_cred,
784 	  NFSX_FH+NFSX_UNSIGNED+nfsm_rndup(ndp->ni_dent.d_namlen)+NFSX_SATTR);
785 	nfsm_fhtom(ndp->ni_dvp);
786 	nfsm_strtom(ndp->ni_dent.d_name, ndp->ni_dent.d_namlen, NFS_MAXNAMLEN);
787 	nfsm_build(sp, struct nfsv2_sattr *, NFSX_SATTR);
788 	sp->sa_mode = vtonfs_mode(vap->va_type, vap->va_mode);
789 	sp->sa_uid = txdr_unsigned(ndp->ni_cred->cr_uid);
790 	sp->sa_gid = txdr_unsigned(ndp->ni_cred->cr_gid);
791 	sp->sa_size = rdev;
792 	/* or should these be VNOVAL ?? */
793 	txdr_time(&vap->va_atime, &sp->sa_atime);
794 	txdr_time(&vap->va_mtime, &sp->sa_mtime);
795 	nfsm_request(ndp->ni_dvp, NFSPROC_CREATE, u.u_procp, 1);
796 	nfsm_reqdone;
797 	VTONFS(ndp->ni_dvp)->n_flag |= NMODIFIED;
798 	nfs_nput(ndp->ni_dvp);
799 	return (error);
800 }
801 
802 /*
803  * nfs file create call
804  */
805 nfs_create(ndp, vap)
806 	register struct nameidata *ndp;
807 	register struct vattr *vap;
808 {
809 	register struct nfsv2_sattr *sp;
810 	register u_long *p;
811 	register caddr_t cp;
812 	register long t1, t2;
813 	caddr_t bpos, dpos, cp2;
814 	u_long xid;
815 	int error = 0;
816 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
817 
818 	nfsstats.rpccnt[NFSPROC_CREATE]++;
819 	nfsm_reqhead(nfs_procids[NFSPROC_CREATE], ndp->ni_cred,
820 	  NFSX_FH+NFSX_UNSIGNED+nfsm_rndup(ndp->ni_dent.d_namlen)+NFSX_SATTR);
821 	nfsm_fhtom(ndp->ni_dvp);
822 	nfsm_strtom(ndp->ni_dent.d_name, ndp->ni_dent.d_namlen, NFS_MAXNAMLEN);
823 	nfsm_build(sp, struct nfsv2_sattr *, NFSX_SATTR);
824 	sp->sa_mode = vtonfs_mode(VREG, vap->va_mode);
825 	sp->sa_uid = txdr_unsigned(ndp->ni_cred->cr_uid);
826 	sp->sa_gid = txdr_unsigned(ndp->ni_cred->cr_gid);
827 	sp->sa_size = txdr_unsigned(0);
828 	/* or should these be VNOVAL ?? */
829 	txdr_time(&vap->va_atime, &sp->sa_atime);
830 	txdr_time(&vap->va_mtime, &sp->sa_mtime);
831 	nfsm_request(ndp->ni_dvp, NFSPROC_CREATE, u.u_procp, 1);
832 	nfsm_mtofh(ndp->ni_dvp, ndp->ni_vp);
833 	nfsm_reqdone;
834 	VTONFS(ndp->ni_dvp)->n_flag |= NMODIFIED;
835 	nfs_nput(ndp->ni_dvp);
836 	return (error);
837 }
838 
839 /*
840  * nfs file remove call
841  * To try and make nfs semantics closer to ufs semantics, a file that has
842  * other processes using the vnode is renamed instead of removed and then
843  * removed later on the last close.
844  * - If v_usecount > 1
845  *	  If a rename is not already in the works
846  *	     call nfs_sillyrename() to set it up
847  *     else
848  *	  do the remove rpc
849  */
850 nfs_remove(ndp)
851 	register struct nameidata *ndp;
852 {
853 	register struct vnode *vp = ndp->ni_vp;
854 	register struct nfsnode *np = VTONFS(ndp->ni_vp);
855 	register u_long *p;
856 	register caddr_t cp;
857 	register long t1, t2;
858 	caddr_t bpos, dpos;
859 	u_long xid;
860 	int error = 0;
861 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
862 
863 	if (vp->v_usecount > 1) {
864 		if (!np->n_sillyrename)
865 			error = nfs_sillyrename(ndp, REMOVE);
866 	} else {
867 		nfsstats.rpccnt[NFSPROC_REMOVE]++;
868 		nfsm_reqhead(nfs_procids[NFSPROC_REMOVE], ndp->ni_cred,
869 			NFSX_FH+NFSX_UNSIGNED+nfsm_rndup(ndp->ni_dent.d_namlen));
870 		nfsm_fhtom(ndp->ni_dvp);
871 		nfsm_strtom(ndp->ni_dent.d_name, ndp->ni_dent.d_namlen, NFS_MAXNAMLEN);
872 		nfsm_request(ndp->ni_dvp, NFSPROC_REMOVE, u.u_procp, 1);
873 		nfsm_reqdone;
874 		VTONFS(ndp->ni_dvp)->n_flag |= NMODIFIED;
875 		/*
876 		 * Kludge City: If the first reply to the remove rpc is lost..
877 		 *   the reply to the retransmitted request will be ENOENT
878 		 *   since the file was in fact removed
879 		 *   Therefore, we cheat and return success.
880 		 */
881 		if (error == ENOENT)
882 			error = 0;
883 	}
884 	np->n_attrstamp = 0;
885 	if (ndp->ni_dvp == vp)
886 		vrele(vp);
887 	else
888 		nfs_nput(ndp->ni_dvp);
889 	nfs_nput(vp);
890 	return (error);
891 }
892 
893 /*
894  * nfs file remove rpc called from nfs_inactive
895  */
896 nfs_removeit(ndp)
897 	register struct nameidata *ndp;
898 {
899 	register u_long *p;
900 	register caddr_t cp;
901 	register long t1, t2;
902 	caddr_t bpos, dpos;
903 	u_long xid;
904 	int error = 0;
905 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
906 
907 	nfsstats.rpccnt[NFSPROC_REMOVE]++;
908 	nfsm_reqhead(nfs_procids[NFSPROC_REMOVE], ndp->ni_cred,
909 		NFSX_FH+NFSX_UNSIGNED+nfsm_rndup(ndp->ni_dent.d_namlen));
910 	nfsm_fhtom(ndp->ni_dvp);
911 	nfsm_strtom(ndp->ni_dent.d_name, ndp->ni_dent.d_namlen, NFS_MAXNAMLEN);
912 	nfsm_request(ndp->ni_dvp, NFSPROC_REMOVE, u.u_procp, 1);
913 	nfsm_reqdone;
914 	VTONFS(ndp->ni_dvp)->n_flag |= NMODIFIED;
915 	return (error);
916 }
917 
918 /*
919  * nfs file rename call
920  */
921 nfs_rename(sndp, tndp)
922 	register struct nameidata *sndp, *tndp;
923 {
924 	register u_long *p;
925 	register caddr_t cp;
926 	register long t1, t2;
927 	caddr_t bpos, dpos;
928 	u_long xid;
929 	int error = 0;
930 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
931 
932 	nfsstats.rpccnt[NFSPROC_RENAME]++;
933 	nfsm_reqhead(nfs_procids[NFSPROC_RENAME], tndp->ni_cred,
934 		(NFSX_FH+NFSX_UNSIGNED)*2+nfsm_rndup(sndp->ni_dent.d_namlen)+
935 		nfsm_rndup(tndp->ni_dent.d_namlen)); /* or sndp->ni_cred?*/
936 	nfsm_fhtom(sndp->ni_dvp);
937 	nfsm_strtom(sndp->ni_dent.d_name,sndp->ni_dent.d_namlen,NFS_MAXNAMLEN);
938 	nfsm_fhtom(tndp->ni_dvp);
939 	nfsm_strtom(tndp->ni_dent.d_name,tndp->ni_dent.d_namlen,NFS_MAXNAMLEN);
940 	nfsm_request(sndp->ni_dvp, NFSPROC_RENAME, u.u_procp, 1);
941 	nfsm_reqdone;
942 	VTONFS(sndp->ni_dvp)->n_flag |= NMODIFIED;
943 	VTONFS(tndp->ni_dvp)->n_flag |= NMODIFIED;
944 	if (sndp->ni_vp->v_type == VDIR) {
945 		if (tndp->ni_vp != NULL && tndp->ni_vp->v_type == VDIR)
946 			cache_purge(tndp->ni_dvp);
947 		cache_purge(sndp->ni_dvp);
948 	}
949 	VOP_ABORTOP(tndp);
950 	if (tndp->ni_dvp == tndp->ni_vp)
951 		vrele(tndp->ni_dvp);
952 	else
953 		vput(tndp->ni_dvp);
954 	if (tndp->ni_vp)
955 		vput(tndp->ni_vp);
956 	VOP_ABORTOP(sndp);
957 	vrele(sndp->ni_dvp);
958 	vrele(sndp->ni_vp);
959 	/*
960 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
961 	 */
962 	if (error == ENOENT)
963 		error = 0;
964 	return (error);
965 }
966 
967 /*
968  * nfs file rename rpc called from nfs_remove() above
969  */
970 nfs_renameit(sndp, tndp)
971 	register struct nameidata *sndp, *tndp;
972 {
973 	register u_long *p;
974 	register caddr_t cp;
975 	register long t1, t2;
976 	caddr_t bpos, dpos;
977 	u_long xid;
978 	int error = 0;
979 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
980 
981 	nfsstats.rpccnt[NFSPROC_RENAME]++;
982 	nfsm_reqhead(nfs_procids[NFSPROC_RENAME], tndp->ni_cred,
983 		(NFSX_FH+NFSX_UNSIGNED)*2+nfsm_rndup(sndp->ni_dent.d_namlen)+
984 		nfsm_rndup(tndp->ni_dent.d_namlen)); /* or sndp->ni_cred?*/
985 	nfsm_fhtom(sndp->ni_dvp);
986 	nfsm_strtom(sndp->ni_dent.d_name,sndp->ni_dent.d_namlen,NFS_MAXNAMLEN);
987 	nfsm_fhtom(tndp->ni_dvp);
988 	nfsm_strtom(tndp->ni_dent.d_name,tndp->ni_dent.d_namlen,NFS_MAXNAMLEN);
989 	nfsm_request(sndp->ni_dvp, NFSPROC_RENAME, u.u_procp, 1);
990 	nfsm_reqdone;
991 	VTONFS(sndp->ni_dvp)->n_flag |= NMODIFIED;
992 	VTONFS(tndp->ni_dvp)->n_flag |= NMODIFIED;
993 	return (error);
994 }
995 
996 /*
997  * nfs hard link create call
998  */
999 nfs_link(vp, ndp)
1000 	register struct vnode *vp;
1001 	register struct nameidata *ndp;
1002 {
1003 	register u_long *p;
1004 	register caddr_t cp;
1005 	register long t1, t2;
1006 	caddr_t bpos, dpos;
1007 	u_long xid;
1008 	int error = 0;
1009 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1010 
1011 	if (ndp->ni_dvp != vp)
1012 		nfs_lock(vp);
1013 	nfsstats.rpccnt[NFSPROC_LINK]++;
1014 	nfsm_reqhead(nfs_procids[NFSPROC_LINK], ndp->ni_cred,
1015 		NFSX_FH*2+NFSX_UNSIGNED+nfsm_rndup(ndp->ni_dent.d_namlen));
1016 	nfsm_fhtom(vp);
1017 	nfsm_fhtom(ndp->ni_dvp);
1018 	nfsm_strtom(ndp->ni_dent.d_name, ndp->ni_dent.d_namlen, NFS_MAXNAMLEN);
1019 	nfsm_request(vp, NFSPROC_LINK, u.u_procp, 1);
1020 	nfsm_reqdone;
1021 	VTONFS(vp)->n_attrstamp = 0;
1022 	VTONFS(ndp->ni_dvp)->n_flag |= NMODIFIED;
1023 	if (ndp->ni_dvp != vp)
1024 		nfs_unlock(vp);
1025 	nfs_nput(ndp->ni_dvp);
1026 	/*
1027 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1028 	 */
1029 	if (error == EEXIST)
1030 		error = 0;
1031 	return (error);
1032 }
1033 
1034 /*
1035  * nfs symbolic link create call
1036  */
1037 nfs_symlink(ndp, vap, nm)
1038 	struct nameidata *ndp;
1039 	struct vattr *vap;
1040 	char *nm;		/* is this the path ?? */
1041 {
1042 	register struct nfsv2_sattr *sp;
1043 	register u_long *p;
1044 	register caddr_t cp;
1045 	register long t1, t2;
1046 	caddr_t bpos, dpos;
1047 	u_long xid;
1048 	int error = 0;
1049 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1050 
1051 	nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1052 	nfsm_reqhead(nfs_procids[NFSPROC_SYMLINK], ndp->ni_cred,
1053 	NFSX_FH+NFSX_UNSIGNED+nfsm_rndup(ndp->ni_dent.d_namlen)+NFSX_UNSIGNED);
1054 	nfsm_fhtom(ndp->ni_dvp);
1055 	nfsm_strtom(ndp->ni_dent.d_name, ndp->ni_dent.d_namlen, NFS_MAXNAMLEN);
1056 	nfsm_strtom(nm, strlen(nm), NFS_MAXPATHLEN);
1057 	nfsm_build(sp, struct nfsv2_sattr *, NFSX_SATTR);
1058 	sp->sa_mode = vtonfs_mode(VLNK, vap->va_mode);
1059 	sp->sa_uid = txdr_unsigned(ndp->ni_cred->cr_uid);
1060 	sp->sa_gid = txdr_unsigned(ndp->ni_cred->cr_gid);
1061 	sp->sa_size = txdr_unsigned(VNOVAL);
1062 	txdr_time(&vap->va_atime, &sp->sa_atime);	/* or VNOVAL ?? */
1063 	txdr_time(&vap->va_mtime, &sp->sa_mtime);	/* or VNOVAL ?? */
1064 	nfsm_request(ndp->ni_dvp, NFSPROC_SYMLINK, u.u_procp, 1);
1065 	nfsm_reqdone;
1066 	VTONFS(ndp->ni_dvp)->n_flag |= NMODIFIED;
1067 	nfs_nput(ndp->ni_dvp);
1068 	/*
1069 	 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1070 	 */
1071 	if (error == EEXIST)
1072 		error = 0;
1073 	return (error);
1074 }
1075 
1076 /*
1077  * nfs make dir call
1078  */
1079 nfs_mkdir(ndp, vap)
1080 	register struct nameidata *ndp;
1081 	struct vattr *vap;
1082 {
1083 	register struct nfsv2_sattr *sp;
1084 	register u_long *p;
1085 	register caddr_t cp;
1086 	register long t1, t2;
1087 	register int len;
1088 	caddr_t bpos, dpos, cp2;
1089 	u_long xid;
1090 	int error = 0, firsttry = 1;
1091 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1092 
1093 	len = ndp->ni_dent.d_namlen;
1094 	nfsstats.rpccnt[NFSPROC_MKDIR]++;
1095 	nfsm_reqhead(nfs_procids[NFSPROC_MKDIR], ndp->ni_cred,
1096 	  NFSX_FH+NFSX_UNSIGNED+nfsm_rndup(len)+NFSX_SATTR);
1097 	nfsm_fhtom(ndp->ni_dvp);
1098 	nfsm_strtom(ndp->ni_dent.d_name, len, NFS_MAXNAMLEN);
1099 	nfsm_build(sp, struct nfsv2_sattr *, NFSX_SATTR);
1100 	sp->sa_mode = vtonfs_mode(VDIR, vap->va_mode);
1101 	sp->sa_uid = txdr_unsigned(ndp->ni_cred->cr_uid);
1102 	sp->sa_gid = txdr_unsigned(ndp->ni_cred->cr_gid);
1103 	sp->sa_size = txdr_unsigned(VNOVAL);
1104 	txdr_time(&vap->va_atime, &sp->sa_atime);	/* or VNOVAL ?? */
1105 	txdr_time(&vap->va_mtime, &sp->sa_mtime);	/* or VNOVAL ?? */
1106 	nfsm_request(ndp->ni_dvp, NFSPROC_MKDIR, u.u_procp, 1);
1107 	nfsm_mtofh(ndp->ni_dvp, ndp->ni_vp);
1108 	nfsm_reqdone;
1109 	VTONFS(ndp->ni_dvp)->n_flag |= NMODIFIED;
1110 	/*
1111 	 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
1112 	 * if we can succeed in looking up the directory.
1113 	 * "firsttry" is necessary since the macros may "goto nfsmout" which
1114 	 * is above the if on errors. (Ugh)
1115 	 */
1116 	if (error == EEXIST && firsttry) {
1117 		firsttry = 0;
1118 		error = 0;
1119 		nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1120 		ndp->ni_vp = NULL;
1121 		nfsm_reqhead(nfs_procids[NFSPROC_LOOKUP], ndp->ni_cred,
1122 		    NFSX_FH+NFSX_UNSIGNED+nfsm_rndup(len));
1123 		nfsm_fhtom(ndp->ni_dvp);
1124 		nfsm_strtom(ndp->ni_dent.d_name, len, NFS_MAXNAMLEN);
1125 		nfsm_request(ndp->ni_dvp, NFSPROC_LOOKUP, u.u_procp, 1);
1126 		nfsm_mtofh(ndp->ni_dvp, ndp->ni_vp);
1127 		if (ndp->ni_vp->v_type != VDIR) {
1128 			vput(ndp->ni_vp);
1129 			error = EEXIST;
1130 		}
1131 		m_freem(mrep);
1132 	}
1133 	nfs_nput(ndp->ni_dvp);
1134 	return (error);
1135 }
1136 
1137 /*
1138  * nfs remove directory call
1139  */
1140 nfs_rmdir(ndp)
1141 	register struct nameidata *ndp;
1142 {
1143 	register u_long *p;
1144 	register caddr_t cp;
1145 	register long t1, t2;
1146 	caddr_t bpos, dpos;
1147 	u_long xid;
1148 	int error = 0;
1149 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1150 
1151 	if (ndp->ni_dvp == ndp->ni_vp) {
1152 		vrele(ndp->ni_dvp);
1153 		nfs_nput(ndp->ni_dvp);
1154 		return (EINVAL);
1155 	}
1156 	nfsstats.rpccnt[NFSPROC_RMDIR]++;
1157 	nfsm_reqhead(nfs_procids[NFSPROC_RMDIR], ndp->ni_cred,
1158 		NFSX_FH+NFSX_UNSIGNED+nfsm_rndup(ndp->ni_dent.d_namlen));
1159 	nfsm_fhtom(ndp->ni_dvp);
1160 	nfsm_strtom(ndp->ni_dent.d_name, ndp->ni_dent.d_namlen, NFS_MAXNAMLEN);
1161 	nfsm_request(ndp->ni_dvp, NFSPROC_RMDIR, u.u_procp, 1);
1162 	nfsm_reqdone;
1163 	VTONFS(ndp->ni_dvp)->n_flag |= NMODIFIED;
1164 	cache_purge(ndp->ni_dvp);
1165 	cache_purge(ndp->ni_vp);
1166 	nfs_nput(ndp->ni_vp);
1167 	nfs_nput(ndp->ni_dvp);
1168 	/*
1169 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
1170 	 */
1171 	if (error == ENOENT)
1172 		error = 0;
1173 	return (error);
1174 }
1175 
1176 /*
1177  * nfs readdir call
1178  * Although cookie is defined as opaque, I translate it to/from net byte
1179  * order so that it looks more sensible. This appears consistent with the
1180  * Ultrix implementation of NFS.
1181  */
1182 nfs_readdir(vp, uiop, cred, eofflagp)
1183 	register struct vnode *vp;
1184 	struct uio *uiop;
1185 	struct ucred *cred;
1186 	int *eofflagp;
1187 {
1188 	register struct nfsnode *np = VTONFS(vp);
1189 	int tresid, error;
1190 	struct vattr vattr;
1191 
1192 	if (vp->v_type != VDIR)
1193 		return (EPERM);
1194 	/*
1195 	 * First, check for hit on the EOF offset cache
1196 	 */
1197 	if (uiop->uio_offset != 0 && uiop->uio_offset == np->n_direofoffset &&
1198 	    (np->n_flag & NMODIFIED) == 0 &&
1199 	    nfs_dogetattr(vp, &vattr, cred, 0) == 0 &&
1200 	    np->n_mtime == vattr.va_mtime.tv_sec) {
1201 		*eofflagp = 1;
1202 		nfsstats.direofcache_hits++;
1203 		return (0);
1204 	}
1205 
1206 	/*
1207 	 * Call nfs_bioread() to do the real work.
1208 	 */
1209 	tresid = uiop->uio_resid;
1210 	error = nfs_bioread(vp, uiop, 0, cred);
1211 
1212 	if (!error && uiop->uio_resid == tresid) {
1213 		*eofflagp = 1;
1214 		nfsstats.direofcache_misses++;
1215 	} else
1216 		*eofflagp = 0;
1217 	return (error);
1218 }
1219 
1220 /*
1221  * Readdir rpc call.
1222  * Called from below the buffer cache by nfs_doio().
1223  */
1224 nfs_readdirrpc(vp, uiop, cred, procp)
1225 	register struct vnode *vp;
1226 	struct uio *uiop;
1227 	struct ucred *cred;
1228 	struct proc *procp;
1229 {
1230 	register long len;
1231 	register struct direct *dp;
1232 	register u_long *p;
1233 	register caddr_t cp;
1234 	register long t1;
1235 	long tlen, lastlen;
1236 	caddr_t bpos, dpos, cp2;
1237 	u_long xid;
1238 	int error = 0;
1239 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1240 	struct mbuf *md2;
1241 	caddr_t dpos2;
1242 	int siz;
1243 	int more_dirs = 1;
1244 	off_t off, savoff;
1245 	struct direct *savdp;
1246 	struct nfsmount *nmp;
1247 	struct nfsnode *np = VTONFS(vp);
1248 	long tresid;
1249 
1250 	nmp = VFSTONFS(vp->v_mount);
1251 	tresid = uiop->uio_resid;
1252 	/*
1253 	 * Loop around doing readdir rpc's of size uio_resid or nm_rsize,
1254 	 * whichever is smaller, truncated to a multiple of DIRBLKSIZ.
1255 	 * The stopping criteria is EOF or buffer full.
1256 	 */
1257 	while (more_dirs && uiop->uio_resid >= DIRBLKSIZ) {
1258 		nfsstats.rpccnt[NFSPROC_READDIR]++;
1259 		nfsm_reqhead(nfs_procids[NFSPROC_READDIR], cred, xid);
1260 		nfsm_fhtom(vp);
1261 		nfsm_build(p, u_long *, 2*NFSX_UNSIGNED);
1262 		*p++ = txdr_unsigned(uiop->uio_offset);
1263 		*p = txdr_unsigned(((uiop->uio_resid > nmp->nm_rsize) ?
1264 			nmp->nm_rsize : uiop->uio_resid) & ~(DIRBLKSIZ-1));
1265 		nfsm_request(vp, NFSPROC_READDIR, procp, 0);
1266 		siz = 0;
1267 		nfsm_disect(p, u_long *, NFSX_UNSIGNED);
1268 		more_dirs = fxdr_unsigned(int, *p);
1269 
1270 		/* Save the position so that we can do nfsm_mtouio() later */
1271 		dpos2 = dpos;
1272 		md2 = md;
1273 
1274 		/* loop thru the dir entries, doctoring them to 4bsd form */
1275 		off = uiop->uio_offset;
1276 #ifdef lint
1277 		dp = (struct direct *)0;
1278 #endif /* lint */
1279 		while (more_dirs && siz < uiop->uio_resid) {
1280 			savoff = off;		/* Hold onto offset and dp */
1281 			savdp = dp;
1282 			nfsm_disecton(p, u_long *, 2*NFSX_UNSIGNED);
1283 			dp = (struct direct *)p;
1284 			dp->d_ino = fxdr_unsigned(u_long, *p++);
1285 			len = fxdr_unsigned(int, *p);
1286 			if (len <= 0 || len > NFS_MAXNAMLEN) {
1287 				error = EBADRPC;
1288 				m_freem(mrep);
1289 				goto nfsmout;
1290 			}
1291 			dp->d_namlen = (u_short)len;
1292 			nfsm_adv(len);		/* Point past name */
1293 			tlen = nfsm_rndup(len);
1294 			/*
1295 			 * This should not be necessary, but some servers have
1296 			 * broken XDR such that these bytes are not null filled.
1297 			 */
1298 			if (tlen != len) {
1299 				*dpos = '\0';	/* Null-terminate */
1300 				nfsm_adv(tlen - len);
1301 				len = tlen;
1302 			}
1303 			nfsm_disecton(p, u_long *, 2*NFSX_UNSIGNED);
1304 			off = fxdr_unsigned(off_t, *p);
1305 			*p++ = 0;	/* Ensures null termination of name */
1306 			more_dirs = fxdr_unsigned(int, *p);
1307 			dp->d_reclen = len+4*NFSX_UNSIGNED;
1308 			siz += dp->d_reclen;
1309 		}
1310 		/*
1311 		 * If at end of rpc data, get the eof boolean
1312 		 */
1313 		if (!more_dirs) {
1314 			nfsm_disecton(p, u_long *, NFSX_UNSIGNED);
1315 			more_dirs = (fxdr_unsigned(int, *p) == 0);
1316 
1317 			/*
1318 			 * If at EOF, cache directory offset
1319 			 */
1320 			if (!more_dirs)
1321 				np->n_direofoffset = off;
1322 		}
1323 		/*
1324 		 * If there is too much to fit in the data buffer, use savoff and
1325 		 * savdp to trim off the last record.
1326 		 * --> we are not at eof
1327 		 */
1328 		if (siz > uiop->uio_resid) {
1329 			off = savoff;
1330 			siz -= dp->d_reclen;
1331 			dp = savdp;
1332 			more_dirs = 0;	/* Paranoia */
1333 		}
1334 		if (siz > 0) {
1335 			lastlen = dp->d_reclen;
1336 			md = md2;
1337 			dpos = dpos2;
1338 			nfsm_mtouio(uiop, siz);
1339 			uiop->uio_offset = off;
1340 		} else
1341 			more_dirs = 0;	/* Ugh, never happens, but in case.. */
1342 		m_freem(mrep);
1343 	}
1344 	/*
1345 	 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
1346 	 * by increasing d_reclen for the last record.
1347 	 */
1348 	if (uiop->uio_resid < tresid) {
1349 		len = uiop->uio_resid & (DIRBLKSIZ - 1);
1350 		if (len > 0) {
1351 			dp = (struct direct *)
1352 				(uiop->uio_iov->iov_base - lastlen);
1353 			dp->d_reclen += len;
1354 			uiop->uio_iov->iov_base += len;
1355 			uiop->uio_iov->iov_len -= len;
1356 			uiop->uio_resid -= len;
1357 		}
1358 	}
1359 nfsmout:
1360 	return (error);
1361 }
1362 
1363 static char hextoasc[] = "0123456789abcdef";
1364 
1365 /*
1366  * Silly rename. To make the NFS filesystem that is stateless look a little
1367  * more like the "ufs" a remove of an active vnode is translated to a rename
1368  * to a funny looking filename that is removed by nfs_inactive on the
1369  * nfsnode. There is the potential for another process on a different client
1370  * to create the same funny name between the nfs_lookitup() fails and the
1371  * nfs_rename() completes, but...
1372  */
1373 nfs_sillyrename(ndp, flag)
1374 	register struct nameidata *ndp;
1375 	int flag;
1376 {
1377 	register struct nfsnode *np;
1378 	register struct sillyrename *sp;
1379 	register struct nameidata *tndp;
1380 	int error;
1381 	short pid;
1382 
1383 	np = VTONFS(ndp->ni_dvp);
1384 	cache_purge(ndp->ni_dvp);
1385 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
1386 		M_TEMP, M_WAITOK);
1387 	sp->s_flag = flag;
1388 	bcopy((caddr_t)&np->n_fh, (caddr_t)&sp->s_fh, NFSX_FH);
1389 	np = VTONFS(ndp->ni_vp);
1390 	tndp = &sp->s_namei;
1391 	tndp->ni_cred = crdup(ndp->ni_cred);
1392 
1393 	/* Fudge together a funny name */
1394 	pid = u.u_procp->p_pid;
1395 	bcopy(".nfsAxxxx4.4", tndp->ni_dent.d_name, 13);
1396 	tndp->ni_dent.d_namlen = 12;
1397 	tndp->ni_dent.d_name[8] = hextoasc[pid & 0xf];
1398 	tndp->ni_dent.d_name[7] = hextoasc[(pid >> 4) & 0xf];
1399 	tndp->ni_dent.d_name[6] = hextoasc[(pid >> 8) & 0xf];
1400 	tndp->ni_dent.d_name[5] = hextoasc[(pid >> 12) & 0xf];
1401 
1402 	/* Try lookitups until we get one that isn't there */
1403 	while (nfs_lookitup(ndp->ni_dvp, tndp, (nfsv2fh_t *)0) == 0) {
1404 		tndp->ni_dent.d_name[4]++;
1405 		if (tndp->ni_dent.d_name[4] > 'z') {
1406 			error = EINVAL;
1407 			goto bad;
1408 		}
1409 	}
1410 	if (error = nfs_renameit(ndp, tndp))
1411 		goto bad;
1412 	nfs_lookitup(ndp->ni_dvp, tndp, &np->n_fh);
1413 	np->n_sillyrename = sp;
1414 	return (0);
1415 bad:
1416 	crfree(tndp->ni_cred);
1417 	free((caddr_t)sp, M_TEMP);
1418 	return (error);
1419 }
1420 
1421 /*
1422  * Look up a file name for silly rename stuff.
1423  * Just like nfs_lookup() except that it doesn't load returned values
1424  * into the nfsnode table.
1425  * If fhp != NULL it copies the returned file handle out
1426  */
1427 nfs_lookitup(vp, ndp, fhp)
1428 	register struct vnode *vp;
1429 	register struct nameidata *ndp;
1430 	nfsv2fh_t *fhp;
1431 {
1432 	register u_long *p;
1433 	register caddr_t cp;
1434 	register long t1, t2;
1435 	caddr_t bpos, dpos, cp2;
1436 	u_long xid;
1437 	int error = 0;
1438 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1439 	long len;
1440 
1441 	nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1442 	ndp->ni_dvp = vp;
1443 	ndp->ni_vp = NULL;
1444 	len = ndp->ni_dent.d_namlen;
1445 	nfsm_reqhead(nfs_procids[NFSPROC_LOOKUP], ndp->ni_cred, NFSX_FH+NFSX_UNSIGNED+nfsm_rndup(len));
1446 	nfsm_fhtom(vp);
1447 	nfsm_strtom(ndp->ni_dent.d_name, len, NFS_MAXNAMLEN);
1448 	nfsm_request(vp, NFSPROC_LOOKUP, u.u_procp, 1);
1449 	if (fhp != NULL) {
1450 		nfsm_disect(cp, caddr_t, NFSX_FH);
1451 		bcopy(cp, (caddr_t)fhp, NFSX_FH);
1452 	}
1453 	nfsm_reqdone;
1454 	return (error);
1455 }
1456 
1457 /*
1458  * Kludge City..
1459  * - make nfs_bmap() essentially a no-op that does no translation
1460  * - do nfs_strategy() by faking physical I/O with nfs_readrpc/nfs_writerpc
1461  *   after mapping the physical addresses into Kernel Virtual space in the
1462  *   nfsiobuf area.
1463  *   (Maybe I could use the process's page mapping, but I was concerned that
1464  *    Kernel Write might not be enabled and also figured copyout() would do
1465  *    a lot more work than bcopy() and also it currently happens in the
1466  *    context of the swapper process (2).
1467  */
1468 nfs_bmap(vp, bn, vpp, bnp)
1469 	struct vnode *vp;
1470 	daddr_t bn;
1471 	struct vnode **vpp;
1472 	daddr_t *bnp;
1473 {
1474 	if (vpp != NULL)
1475 		*vpp = vp;
1476 	if (bnp != NULL)
1477 		*bnp = bn * btodb(vp->v_mount->mnt_stat.f_bsize);
1478 	return (0);
1479 }
1480 
1481 /*
1482  * Strategy routine for phys. i/o
1483  * If the biod's are running, queue a request
1484  * otherwise just call nfs_doio() to get it done
1485  */
1486 nfs_strategy(bp)
1487 	register struct buf *bp;
1488 {
1489 	register struct buf *dp;
1490 	register int i;
1491 	struct proc *rp;
1492 	int error = 0;
1493 	int fnd = 0;
1494 
1495 	/*
1496 	 * Set b_proc. It seems a bit silly to do it here, but since bread()
1497 	 * doesn't set it, I will.
1498 	 * Set b_proc == NULL for asynchronous ops, since these may still
1499 	 * be hanging about after the process terminates.
1500 	 */
1501 	if (bp->b_flags & B_ASYNC)
1502 		bp->b_proc = (struct proc *)0;
1503 	else
1504 		bp->b_proc = u.u_procp;
1505 
1506 	/*
1507 	 * If the op is asynchronous and an i/o daemon is waiting
1508 	 * queue the request, wake it up and wait for completion
1509 	 * otherwise just do it ourselves.
1510 	 */
1511 	if (bp->b_proc == (struct proc *)NULL)
1512 	    for (i = 0; i < nfs_asyncdaemons; i++) {
1513 		if (rp = nfs_iodwant[i]) {
1514 			/*
1515 			 * Ensure that the async_daemon is still waiting here
1516 			 */
1517 			if (rp->p_stat != SSLEEP ||
1518 			    rp->p_wchan != ((caddr_t)&nfs_iodwant[i])) {
1519 				nfs_iodwant[i] = (struct proc *)0;
1520 				continue;
1521 			}
1522 			dp = &nfs_bqueue;
1523 			if (dp->b_actf == NULL) {
1524 				dp->b_actl = bp;
1525 				bp->b_actf = dp;
1526 			} else {
1527 				dp->b_actf->b_actl = bp;
1528 				bp->b_actf = dp->b_actf;
1529 			}
1530 			dp->b_actf = bp;
1531 			bp->b_actl = dp;
1532 			fnd++;
1533 			nfs_iodwant[i] = (struct proc *)0;
1534 			wakeup((caddr_t)&nfs_iodwant[i]);
1535 			break;
1536 		}
1537 	}
1538 	if (!fnd)
1539 		error = nfs_doio(bp);
1540 	return (error);
1541 }
1542 
1543 /*
1544  * Fun and games with i/o
1545  * Essentially play ubasetup() and disk interrupt service routine by
1546  * mapping the data buffer into kernel virtual space and doing the
1547  * nfs read or write rpc's from it.
1548  * If the nfsiod's are not running, this is just called from nfs_strategy(),
1549  * otherwise it is called by the nfsiods to do what would normally be
1550  * partially disk interrupt driven.
1551  */
1552 nfs_doio(bp)
1553 	register struct buf *bp;
1554 {
1555 	register struct uio *uiop;
1556 	register struct vnode *vp;
1557 	struct nfsnode *np;
1558 	struct ucred *cr;
1559 	struct proc *rp;
1560 	int error;
1561 	struct uio uio;
1562 	struct iovec io;
1563 #if !defined(hp300) && !defined(i386)
1564 	register struct pte *pte, *ppte;
1565 	register caddr_t vaddr;
1566 	int npf, npf2;
1567 	int reg, o;
1568 	caddr_t vbase;
1569 	unsigned v;
1570 #endif
1571 
1572 	vp = bp->b_vp;
1573 	np = VTONFS(vp);
1574 	uiop = &uio;
1575 	uiop->uio_iov = &io;
1576 	uiop->uio_iovcnt = 1;
1577 	uiop->uio_segflg = UIO_SYSSPACE;
1578 
1579 	/*
1580 	 * For phys i/o, map the b_addr into kernel virtual space using
1581 	 * the Nfsiomap pte's
1582 	 * Also, add a temporary b_rcred for reading using the process's uid
1583 	 * and a guess at a group
1584 	 */
1585 	if (bp->b_flags & B_PHYS) {
1586 		rp = (bp->b_flags & B_DIRTY) ? pageproc : bp->b_proc;
1587 		bp->b_rcred = cr = crcopy(rp->p_ucred);
1588 #if defined(hp300) || defined(i386)
1589 		/* mapping was already done by vmapbuf */
1590 		io.iov_base = bp->b_un.b_addr;
1591 #else
1592 		o = (int)bp->b_un.b_addr & PGOFSET;
1593 		npf2 = npf = btoc(bp->b_bcount + o);
1594 
1595 		/*
1596 		 * Get some mapping page table entries
1597 		 */
1598 		while ((reg = rmalloc(nfsmap, (long)npf)) == 0) {
1599 			nfsmap_want++;
1600 			(void) tsleep((caddr_t)&nfsmap_want, PZERO-1, "nfsmap",
1601 					0);
1602 		}
1603 		reg--;
1604 		if (bp->b_flags & B_PAGET)
1605 			pte = &Usrptmap[btokmx((struct pte *)bp->b_un.b_addr)];
1606 		else {
1607 			v = btop(bp->b_un.b_addr);
1608 			if (bp->b_flags & B_UAREA)
1609 				pte = &rp->p_addr[v];
1610 			else
1611 				pte = vtopte(rp, v);
1612 		}
1613 
1614 		/*
1615 		 * Play vmaccess() but with the Nfsiomap page table
1616 		 */
1617 		ppte = &Nfsiomap[reg];
1618 		vbase = vaddr = &nfsiobuf[reg*NBPG];
1619 		while (npf != 0) {
1620 			mapin(ppte, (u_int)vaddr, pte->pg_pfnum, (int)(PG_V|PG_KW));
1621 #if defined(tahoe)
1622 			mtpr(P1DC, vaddr);
1623 #endif
1624 			ppte++;
1625 			pte++;
1626 			vaddr += NBPG;
1627 			--npf;
1628 		}
1629 		io.iov_base = vbase+o;
1630 #endif /* !defined(hp300) */
1631 
1632 		/*
1633 		 * And do the i/o rpc
1634 		 */
1635 		io.iov_len = uiop->uio_resid = bp->b_bcount;
1636 		uiop->uio_offset = bp->b_blkno * DEV_BSIZE;
1637 		if (bp->b_flags & B_READ) {
1638 			uiop->uio_rw = UIO_READ;
1639 			nfsstats.read_physios++;
1640 			bp->b_error = error = nfs_readrpc(vp, uiop,
1641 				bp->b_rcred, bp->b_proc);
1642 			(void) vnode_pager_uncache(vp);
1643 		} else {
1644 			uiop->uio_rw = UIO_WRITE;
1645 			nfsstats.write_physios++;
1646 			bp->b_error = error = nfs_writerpc(vp, uiop,
1647 				bp->b_wcred, bp->b_proc);
1648 		}
1649 
1650 		/*
1651 		 * Finally, release pte's used by physical i/o
1652 		 */
1653 		crfree(cr);
1654 #if !defined(hp300) && !defined(i386)
1655 		rmfree(nfsmap, (long)npf2, (long)++reg);
1656 		if (nfsmap_want) {
1657 			nfsmap_want = 0;
1658 			wakeup((caddr_t)&nfsmap_want);
1659 		}
1660 #endif
1661 	} else {
1662 		if (bp->b_flags & B_READ) {
1663 			io.iov_len = uiop->uio_resid = bp->b_bcount;
1664 			io.iov_base = bp->b_un.b_addr;
1665 			uiop->uio_rw = UIO_READ;
1666 			switch (vp->v_type) {
1667 			case VREG:
1668 				uiop->uio_offset = bp->b_blkno * DEV_BSIZE;
1669 				nfsstats.read_bios++;
1670 				error = nfs_readrpc(vp, uiop, bp->b_rcred,
1671 					bp->b_proc);
1672 				break;
1673 			case VLNK:
1674 				uiop->uio_offset = 0;
1675 				nfsstats.readlink_bios++;
1676 				error = nfs_readlinkrpc(vp, uiop, bp->b_rcred,
1677 						bp->b_proc);
1678 				break;
1679 			case VDIR:
1680 				uiop->uio_offset = bp->b_lblkno;
1681 				nfsstats.readdir_bios++;
1682 				error = nfs_readdirrpc(vp, uiop, bp->b_rcred,
1683 					    bp->b_proc);
1684 				/*
1685 				 * Save offset cookie in b_blkno.
1686 				 */
1687 				bp->b_blkno = uiop->uio_offset;
1688 				break;
1689 			};
1690 			bp->b_error = error;
1691 		} else {
1692 			io.iov_len = uiop->uio_resid = bp->b_dirtyend
1693 				- bp->b_dirtyoff;
1694 			uiop->uio_offset = (bp->b_blkno * DEV_BSIZE)
1695 				+ bp->b_dirtyoff;
1696 			io.iov_base = bp->b_un.b_addr + bp->b_dirtyoff;
1697 			uiop->uio_rw = UIO_WRITE;
1698 			nfsstats.write_bios++;
1699 			bp->b_error = error = nfs_writerpc(vp, uiop,
1700 				bp->b_wcred, bp->b_proc);
1701 			if (error) {
1702 				np->n_error = error;
1703 				np->n_flag |= NWRITEERR;
1704 			}
1705 			bp->b_dirtyoff = bp->b_dirtyend = 0;
1706 		}
1707 	}
1708 	if (error)
1709 		bp->b_flags |= B_ERROR;
1710 	bp->b_resid = uiop->uio_resid;
1711 	biodone(bp);
1712 	return (error);
1713 }
1714 
1715 /*
1716  * Flush all the blocks associated with a vnode.
1717  * 	Walk through the buffer pool and push any dirty pages
1718  *	associated with the vnode.
1719  */
1720 /* ARGSUSED */
1721 nfs_fsync(vp, fflags, cred, waitfor)
1722 	register struct vnode *vp;
1723 	int fflags;
1724 	struct ucred *cred;
1725 	int waitfor;
1726 {
1727 	register struct nfsnode *np = VTONFS(vp);
1728 	int error = 0;
1729 
1730 	if (np->n_flag & NMODIFIED) {
1731 		np->n_flag &= ~NMODIFIED;
1732 		vflushbuf(vp, waitfor == MNT_WAIT ? B_SYNC : 0);
1733 	}
1734 	if (!error && (np->n_flag & NWRITEERR))
1735 		error = np->n_error;
1736 	return (error);
1737 }
1738 
1739 /*
1740  * NFS advisory byte-level locks.
1741  * Currently unsupported.
1742  */
1743 nfs_advlock(vp, id, op, fl, flags)
1744 	struct vnode *vp;
1745 	caddr_t id;
1746 	int op;
1747 	struct flock *fl;
1748 	int flags;
1749 {
1750 
1751 	return (EOPNOTSUPP);
1752 }
1753 
1754 /*
1755  * Print out the contents of an nfsnode.
1756  */
1757 nfs_print(vp)
1758 	struct vnode *vp;
1759 {
1760 	register struct nfsnode *np = VTONFS(vp);
1761 
1762 	printf("tag VT_NFS, fileid %d fsid 0x%x",
1763 		np->n_vattr.va_fileid, np->n_vattr.va_fsid);
1764 #ifdef FIFO
1765 	if (vp->v_type == VFIFO)
1766 		fifo_printinfo(vp);
1767 #endif /* FIFO */
1768 	printf("%s\n", (np->n_flag & NLOCKED) ? " (LOCKED)" : "");
1769 	if (np->n_lockholder == 0)
1770 		return;
1771 	printf("\towner pid %d", np->n_lockholder);
1772 	if (np->n_lockwaiter)
1773 		printf(" waiting pid %d", np->n_lockwaiter);
1774 	printf("\n");
1775 }
1776