1 /*
2  * TO DO:
3  *	- gc of file systems (not going to do just yet?)
4  *	- statistics file
5  *	- configure on amsterdam
6  */
7 
8 #include <u.h>
9 #include <libc.h>
10 #include <bio.h>
11 #include <ip.h>
12 #include <thread.h>
13 #include <libsec.h>
14 #include <sunrpc.h>
15 #include <nfs3.h>
16 #include <diskfs.h>
17 #include <venti.h>
18 #include "nfs3srv.h"
19 
20 #define trace if(!tracecalls){}else print
21 
22 typedef struct Ipokay Ipokay;
23 typedef struct Config Config;
24 typedef struct Ctree Ctree;
25 typedef struct Cnode Cnode;
26 
27 struct Ipokay
28 {
29 	int okay;
30 	uchar ip[IPaddrlen];
31 	uchar mask[IPaddrlen];
32 };
33 
34 struct Config
35 {
36 	Ipokay *ok;
37 	uint nok;
38 	ulong mtime;
39 	Ctree *ctree;
40 };
41 
42 char 		*addr;
43 int			blocksize;
44 int			cachesize;
45 Config		config;
46 char			*configfile;
47 int			encryptedhandles = 1;
48 Channel		*nfschan;
49 Channel		*mountchan;
50 Channel		*timerchan;
51 Nfs3Handle	root;
52 SunSrv		*srv;
53 int			tracecalls;
54 VtCache		*vcache;
55 VtConn		*z;
56 
57 void			cryptinit(void);
58 void			timerthread(void*);
59 void			timerproc(void*);
60 
61 extern	void			handleunparse(Fsys*, Nfs3Handle*, Nfs3Handle*, int);
62 extern	Nfs3Status	handleparse(Nfs3Handle*, Fsys**, Nfs3Handle*, int);
63 
64 Nfs3Status	logread(Cnode*, u32int, u64int, uchar**, u32int*, u1int*);
65 Nfs3Status	refreshdiskread(Cnode*, u32int, u64int, uchar**, u32int*, u1int*);
66 Nfs3Status	refreshconfigread(Cnode*, u32int, u64int, uchar**, u32int*, u1int*);
67 
68 int readconfigfile(Config *cp);
69 void setrootfid(void);
70 int ipokay(uchar *ip, ushort port);
71 
72 u64int	unittoull(char*);
73 
74 void
75 usage(void)
76 {
77 	fprint(2, "usage: vnfs [-LLRVr] [-a addr] [-b blocksize] [-c cachesize] configfile\n");
78 	threadexitsall("usage");
79 }
80 
81 void
82 threadmain(int argc, char **argv)
83 {
84 	fmtinstall('B', sunrpcfmt);
85 	fmtinstall('C', suncallfmt);
86 	fmtinstall('F', vtfcallfmt);
87 	fmtinstall('H', encodefmt);
88 	fmtinstall('I', eipfmt);
89 	fmtinstall('V', vtscorefmt);
90 	sunfmtinstall(&nfs3prog);
91 	sunfmtinstall(&nfsmount3prog);
92 
93 	addr = "udp!*!2049";
94 	blocksize = 8192;
95 	cachesize = 400;
96 	srv = sunsrv();
97 	srv->ipokay = ipokay;
98 	cryptinit();
99 
100 	ARGBEGIN{
101 	default:
102 		usage();
103 	case 'E':
104 		encryptedhandles = 0;
105 		break;
106 	case 'L':
107 		if(srv->localonly == 0)
108 			srv->localonly = 1;
109 		else
110 			srv->localparanoia = 1;
111 		break;
112 	case 'R':
113 		srv->chatty++;
114 		break;
115 	case 'T':
116 		tracecalls = 1;
117 		break;
118 	case 'V':
119 		if(chattyventi++)
120 			vttracelevel++;
121 		break;
122 	case 'a':
123 		addr = EARGF(usage());
124 		break;
125 	case 'b':
126 		blocksize = unittoull(EARGF(usage()));
127 		break;
128 	case 'c':
129 		cachesize = unittoull(EARGF(usage()));
130 		break;
131 	case 'r':
132 		srv->alwaysreject++;
133 		break;
134 	}ARGEND
135 
136 	if(argc != 1)
137 		usage();
138 
139 	if((z = vtdial(nil)) == nil)
140 		sysfatal("vtdial: %r");
141 	if(vtconnect(z) < 0)
142 		sysfatal("vtconnect: %r");
143 	if((vcache = vtcachealloc(z, blocksize, cachesize)) == nil)
144 		sysfatal("vtcache: %r");
145 
146 	configfile = argv[0];
147 	if(readconfigfile(&config) < 0)
148 		sysfatal("readConfig: %r");
149 	setrootfid();
150 
151 	nfschan = chancreate(sizeof(SunMsg*), 0);
152 	mountchan = chancreate(sizeof(SunMsg*), 0);
153 	timerchan = chancreate(sizeof(void*), 0);
154 	
155 	if(sunsrvudp(srv, addr) < 0)
156 		sysfatal("starting server: %r");
157 
158 	sunsrvthreadcreate(srv, nfs3proc, nfschan);
159 	sunsrvthreadcreate(srv, mount3proc, mountchan);
160 	sunsrvthreadcreate(srv, timerthread, nil);
161 	proccreate(timerproc, nil, 32768);
162 	
163 	sunsrvprog(srv, &nfs3prog, nfschan);
164 	sunsrvprog(srv, &nfsmount3prog, mountchan);
165 
166 	threadexits(nil);
167 }
168 
169 #define TWID64	((u64int)~(u64int)0)
170 
171 u64int
172 unittoull(char *s)
173 {
174 	char *es;
175 	u64int n;
176 
177 	if(s == nil)
178 		return TWID64;
179 	n = strtoul(s, &es, 0);
180 	if(*es == 'k' || *es == 'K'){
181 		n *= 1024;
182 		es++;
183 	}else if(*es == 'm' || *es == 'M'){
184 		n *= 1024*1024;
185 		es++;
186 	}else if(*es == 'g' || *es == 'G'){
187 		n *= 1024*1024*1024;
188 		es++;
189 	}else if(*es == 't' || *es == 'T'){
190 		n *= 1024*1024;
191 		n *= 1024*1024;
192 	}
193 	if(*es != '\0')
194 		return TWID64;
195 	return n;
196 }
197 
198 /*
199  * Handles.
200  * 
201  * We store all the state about which file a client is accessing in
202  * the handle, so that we don't have to maintain any per-client state
203  * ourselves.  In order to avoid leaking handles or letting clients 
204  * create arbitrary handles, we sign and encrypt each handle with
205  * AES using a key selected randomly when the server starts.
206  * Thus, handles cannot be used across sessions.  
207  *
208  * The decrypted handles begin with the following header:
209  *
210  *	sessid[8]		random session id chosen at start time
211  *	len[4]		length of handle that follows
212  *
213  * If we're pressed for space in the rest of the handle, we could
214  * probably reduce the amount of sessid bytes.  Note that the sessid
215  * bytes must be consistent during a run of vnfs, or else some 
216  * clients (e.g., Linux 2.4) eventually notice that successive TLookups
217  * return different handles, and they return "Stale NFS file handle"
218  * errors to system calls in response (even though we never sent
219  * that error!).
220  *
221  * Security woes aside, the fact that we have to shove everything
222  * into the handles is quite annoying.  We have to encode, in 40 bytes:
223  *
224  *	- position in the synthesized config tree
225  *	- enough of the path to do glob matching
226  *	- position in an archived file system image
227  *
228  * and the handles need to be stable across changes in the config file
229  * (though not across server restarts since encryption screws
230  * that up nicely).
231  * 
232  * We encode each of the first two as a 10-byte hash that is 
233  * the first half of a SHA1 hash.  
234  */
235 
236 enum
237 {
238 	SessidSize = 8,
239 	HeaderSize = SessidSize+4,
240 	MaxHandleSize = Nfs3MaxHandleSize - HeaderSize
241 };
242 
243 AESstate		aesstate;
244 uchar		sessid[SessidSize];
245 
246 static void
247 hencrypt(Nfs3Handle *h)
248 {
249 	uchar *p;
250 	AESstate aes;
251 
252 	/*
253 	 * root handle has special encryption - a single 0 byte - so that it
254 	 * never goes stale.
255 	 */
256 	if(h->len == root.len && memcmp(h->h, root.h, root.len) == 0){
257 		h->h[0] = 0;
258 		h->len = 1;
259 		return;
260 	}
261 
262 	if(!encryptedhandles)
263 		return;
264 
265 	if(h->len > MaxHandleSize){
266 		/* oops */
267 		fprint(2, "handle too long: %.*lH\n", h->len, h->h);
268 		memset(h->h, 'X', Nfs3MaxHandleSize);
269 		h->len = Nfs3MaxHandleSize;
270 		return;
271 	}
272 
273 	p = h->h;
274 	memmove(p+HeaderSize, p, h->len);
275 	memmove(p, sessid, SessidSize);
276 	*(u32int*)(p+SessidSize) = h->len;
277 	h->len += HeaderSize;
278 
279 	if(encryptedhandles){
280 		while(h->len < MaxHandleSize)
281 			h->h[h->len++] = 0;
282 		aes = aesstate;
283 		aesCBCencrypt(h->h, MaxHandleSize, &aes);
284 	}
285 }
286 
287 static Nfs3Status
288 hdecrypt(Nfs3Handle *h)
289 {
290 	AESstate aes;
291 	
292 	if(h->len == 1 && h->h[0] == 0){	/* single 0 byte is root */
293 		*h = root;
294 		return Nfs3Ok;
295 	}
296 
297 	if(!encryptedhandles)
298 		return Nfs3Ok;
299 
300 	if(h->len <= HeaderSize)
301 		return Nfs3ErrBadHandle;
302 	if(encryptedhandles){
303 		if(h->len != MaxHandleSize)
304 			return Nfs3ErrBadHandle;
305 		aes = aesstate;
306 		aesCBCdecrypt(h->h, h->len, &aes);
307 	}
308 	if(memcmp(h->h, sessid, SessidSize) != 0)
309 		return Nfs3ErrStale;	/* give benefit of doubt */
310 	h->len = *(u32int*)(h->h+SessidSize);
311 	if(h->len >= MaxHandleSize-HeaderSize)
312 		return Nfs3ErrBadHandle;
313 	memmove(h->h, h->h+HeaderSize, h->len);
314 	return Nfs3Ok;
315 }
316 
317 void
318 cryptinit(void)
319 {
320 	uchar key[32], ivec[AESbsize];
321 	int i;
322 	
323 	*(u32int*)sessid = truerand();
324 	for(i=0; i<nelem(key); i+=4)
325 		*(u32int*)&key[i] = truerand();
326 	for(i=0; i<nelem(ivec); i++)
327 		ivec[i] = fastrand();
328 	setupAESstate(&aesstate, key, sizeof key, ivec);
329 }
330 
331 /*
332  * Config file.
333  *
334  * The main purpose of the configuration file is to define a tree
335  * in which the archived file system images are mounted.
336  * The tree is stored as Entry structures, defined below.
337  *
338  * The configuration file also allows one to define shell-like
339  * glob expressions matching paths that are not to be displayed.
340  * The matched files or directories are shown in directory listings
341  * (could suppress these if we cared) but they cannot be opened,
342  * read, or written, and getattr returns zeroed data.
343  */
344 enum
345 {
346 	/* sizes used in handles; see nfs server below */
347 	CnodeHandleSize = 8,
348 	FsysHandleOffset = CnodeHandleSize
349 };
350 
351 /*
352  * Config file tree.
353  */
354 struct Ctree
355 {
356 	Cnode *root;
357 	Cnode *hash[1024];
358 };
359 
360 struct Cnode
361 {
362 	char *name;	/* path element */
363 	Cnode *parent;	/* in tree */
364 	Cnode *nextsib;	/* in tree */
365 	Cnode *kidlist;	/* in tree */
366 	Cnode *nexthash;	/* in hash list */
367 	
368 	Nfs3Status (*read)(Cnode*, u32int, u64int, uchar**, u32int*, u1int*);	/* synthesized read fn */
369 
370 	uchar handle[VtScoreSize];	/* sha1(path to here) */
371 	ulong mtime;	/* mtime for this directory entry */
372 	
373 	/* fsys overlay on this node */
374 	Fsys *fsys;	/* cache of memory structure */
375 	Nfs3Handle fsyshandle;
376 	int isblackhole;	/* walking down keeps you here */
377 
378 	/*
379 	 * mount point info.
380 	 * if a mount point is inside another file system,
381 	 * the fsys and fsyshandle above have the old fs info,
382 	 * the mfsys and mfsyshandle below have the new one.
383 	 * getattrs must use the old info for consistency.
384 	 */
385 	int ismtpt;	/* whether there is an fsys mounted here */
386 	uchar fsysscore[VtScoreSize];	/* score of fsys image on venti */
387 	char *fsysimage;	/* raw disk image */
388 	Fsys *mfsys;	/* mounted file system (nil until walked) */
389 	Nfs3Handle mfsyshandle;	/* handle to root of mounted fsys */
390 	
391 	int mark;	/* gc */
392 };
393 
394 static uint
395 dumbhash(uchar *s)
396 {
397 	return (s[0]<<2)|(s[1]>>6);	/* first 10 bits */
398 }
399 
400 static Cnode*
401 mkcnode(Ctree *t, Cnode *parent, char *elem, uint elen, char *path, uint plen)
402 {
403 	uint h;
404 	Cnode *n;
405 	
406 	n = emalloc(sizeof *n + elen+1);
407 	n->name = (char*)(n+1);
408 	memmove(n->name, elem, elen);
409 	n->name[elen] = 0;
410 	n->parent = parent;
411 	if(parent){
412 		n->nextsib = parent->kidlist;
413 		parent->kidlist = n;
414 	}
415 	n->kidlist = nil;
416 	sha1((uchar*)path, plen, n->handle, nil);
417 	h = dumbhash(n->handle);
418 	n->nexthash = t->hash[h];
419 	t->hash[h] = n;
420 	
421 	return n;
422 }
423 
424 void
425 markctree(Ctree *t)
426 {
427 	int i;
428 	Cnode *n;
429 
430 	for(i=0; i<nelem(t->hash); i++)
431 		for(n=t->hash[i]; n; n=n->nexthash)
432 			if(n->name[0] != '+')
433 				n->mark = 1;
434 }
435 
436 int
437 refreshdisk(void)
438 {
439 	int i;
440 	Cnode *n;
441 	Ctree *t;
442 	
443 	t = config.ctree;
444 	for(i=0; i<nelem(t->hash); i++)
445 		for(n=t->hash[i]; n; n=n->nexthash){
446 			if(n->mfsys)
447 				disksync(n->mfsys->disk);
448 			if(n->fsys)
449 				disksync(n->fsys->disk);
450 		}
451 	return 0;
452 }
453 
454 void
455 sweepctree(Ctree *t)
456 {
457 	int i;
458 	Cnode *n;
459 
460 	/* just zero all the garbage and leave it linked into the tree */
461 	for(i=0; i<nelem(t->hash); i++){
462 		for(n=t->hash[i]; n; n=n->nexthash){
463 			if(!n->mark)
464 				continue;
465 			n->fsys = nil;
466 			free(n->fsysimage);
467 			n->fsysimage = nil;
468 			memset(n->fsysscore, 0, sizeof n->fsysscore);
469 			n->mfsys = nil;
470 			n->ismtpt = 0;
471 			memset(&n->fsyshandle, 0, sizeof n->fsyshandle);
472 			memset(&n->mfsyshandle, 0, sizeof n->mfsyshandle);
473 		}
474 	}
475 }
476 
477 static Cnode*
478 cnodewalk(Cnode *n, char *name, uint len, int markokay)
479 {
480 	Cnode *nn;
481 	
482 	for(nn=n->kidlist; nn; nn=nn->nextsib)
483 		if(strncmp(nn->name, name, len) == 0 && nn->name[len] == 0)
484 		if(!nn->mark || markokay)
485 			return nn;
486 	return nil;
487 }
488 
489 Cnode*
490 ctreewalkpath(Ctree *t, char *name, ulong createmtime)
491 {
492 	Cnode *n, *nn;
493 	char *p, *nextp;
494 	
495 	n = t->root;
496 	p = name;
497 	for(; *p; p=nextp){
498 		n->mark = 0;
499 		assert(*p == '/');
500 		p++;
501 		nextp = strchr(p, '/');
502 		if(nextp == nil)
503 			nextp = p+strlen(p);
504 		if((nn = cnodewalk(n, p, nextp-p, 1)) == nil){
505 			if(createmtime == 0)
506 				return nil;
507 			nn = mkcnode(t, n, p, nextp-p, name, nextp-name);
508 			nn->mtime = createmtime;
509 		}
510 		if(nn->mark)
511 			nn->mark = 0;
512 		n = nn;
513 	}
514 	n->mark = 0;
515 	return n;
516 }
517 
518 Ctree*
519 mkctree(void)
520 {
521 	Ctree *t;
522 	
523 	t = emalloc(sizeof *t);
524 	t->root = mkcnode(t, nil, "", 0, "", 0);
525 	
526 	ctreewalkpath(t, "/+log", time(0))->read = logread;
527 	ctreewalkpath(t, "/+refreshdisk", time(0))->read = refreshdiskread;
528 	ctreewalkpath(t, "/+refreshconfig", time(0))->read = refreshconfigread;
529 
530 	return t;
531 }
532 
533 Cnode*
534 ctreemountfsys(Ctree *t, char *path, ulong time, uchar *score, char *file)
535 {
536 	Cnode *n;
537 	
538 	if(time == 0)
539 		time = 1;
540 	n = ctreewalkpath(t, path, time);
541 	if(score){
542 		if(n->ismtpt && (n->fsysimage || memcmp(n->fsysscore, score, VtScoreSize) != 0)){
543 			free(n->fsysimage);
544 			n->fsysimage = nil;
545 			n->fsys = nil;	/* leak (might be other refs) */
546 		}
547 		memmove(n->fsysscore, score, VtScoreSize);
548 	}else{
549 		if(n->ismtpt && (n->fsysimage==nil || strcmp(n->fsysimage, file) != 0)){
550 			free(n->fsysimage);
551 			n->fsysimage = nil;
552 			n->fsys = nil;	/* leak (might be other refs) */
553 		}
554 		n->fsysimage = emalloc(strlen(file)+1);
555 		strcpy(n->fsysimage, file);
556 	}
557 	n->ismtpt = 1;
558 	return n;
559 }
560 
561 Cnode*
562 cnodebyhandle(Ctree *t, uchar *p)
563 {
564 	int h;
565 	Cnode *n;
566 	
567 	h = dumbhash(p);
568 	for(n=t->hash[h]; n; n=n->nexthash)
569 		if(memcmp(n->handle, p, CnodeHandleSize) == 0)
570 			return n;
571 	return nil;
572 }
573 
574 static int
575 parseipandmask(char *s, uchar *ip, uchar *mask)
576 {
577 	char *p, *q;
578 	
579 	p = strchr(s, '/');
580 	if(p)
581 		*p++ = 0;
582 	if(parseip(ip, s) == ~0UL)
583 		return -1;
584 	if(p == nil)
585 		memset(mask, 0xFF, IPaddrlen);
586 	else{
587 		if(isdigit((uchar)*p) && strtol(p, &q, 10)>=0 && *q==0)
588 			*--p = '/';
589 		if(parseipmask(mask, p) == ~0UL)
590 			return -1;
591 		if(*p != '/')
592 			*--p = '/';
593 	}
594 /*fprint(2, "parseipandmask %s => %I %I\n", s, ip, mask); */
595 	return 0;
596 }
597 
598 static int
599 parsetime(char *s, ulong *time)
600 {
601 	ulong x;
602 	char *p;
603 	int i;
604 	Tm tm;
605 	
606 	/* decimal integer is seconds since 1970 */
607 	x = strtoul(s, &p, 10);
608 	if(x > 0 && *p == 0){
609 		*time = x;
610 		return 0;
611 	}
612 	
613 	/* otherwise expect yyyy/mmdd/hhmm */
614 	if(strlen(s) != 14 || s[4] != '/' || s[9] != '/')
615 		return -1;
616 	for(i=0; i<4; i++)
617 		if(!isdigit((uchar)s[i]) || !isdigit((uchar)s[i+5]) || !isdigit((uchar)s[i+10]))
618 			return -1;
619 	memset(&tm, 0, sizeof tm);
620 	tm.year = atoi(s)-1900;
621 	if(tm.year < 0 || tm.year > 200)
622 		return -1;
623 	tm.mon = (s[5]-'0')*10+s[6]-'0' - 1;
624 	if(tm.mon < 0 || tm.mon > 11)
625 		return -1; 
626 	tm.mday = (s[7]-'0')*10+s[8]-'0';
627 	if(tm.mday < 0 || tm.mday > 31)
628 		return -1;
629 	tm.hour = (s[10]-'0')*10+s[11]-'0';
630 	if(tm.hour < 0 || tm.hour > 23)
631 		return -1;
632 	tm.min = (s[12]-'0')*10+s[13]-'0';
633 	if(tm.min < 0 || tm.min > 59)
634 		return -1;
635 	strcpy(tm.zone, "XXX");	/* anything but GMT */
636 if(0){
637 print("tm2sec %d/%d/%d/%d/%d\n",
638 	tm.year, tm.mon, tm.mday, tm.hour, tm.min);
639 }
640 	*time = tm2sec(&tm);
641 if(0) print("time %lud\n", *time);
642 	return 0;
643 }
644 
645 
646 int
647 readconfigfile(Config *cp)
648 {
649 	char *f[10], *image, *p, *pref, *q, *name;
650 	int nf, line;
651 	uchar scorebuf[VtScoreSize], *score;
652 	ulong time;
653 	Biobuf *b;
654 	Config c;
655 	Dir *dir;
656 
657 	name = configfile;
658 	c = *cp;
659 	if((dir = dirstat(name)) == nil)
660 		return -1;
661 	if(c.mtime == dir->mtime){
662 		free(dir);
663 		return 0;
664 	}
665 	c.mtime = dir->mtime;
666 	free(dir);
667 	if((b = Bopen(name, OREAD)) == nil){
668 		free(dir);
669 		return -1;
670 	}
671 	
672 	/*
673 	 * Reuse old tree, garbage collecting entries that
674 	 * are not mentioned in the new config file.
675 	 */
676 	if(c.ctree == nil)
677 		c.ctree = mkctree();
678 
679 	markctree(c.ctree);
680 	c.ok = nil;
681 	c.nok = 0;
682 	
683 	line = 0;
684 	for(; (p=Brdstr(b, '\n', 1)) != nil; free(p)){
685 		line++;
686 		if((q = strchr(p, '#')) != nil)
687 			*q = 0;
688 		nf = tokenize(p, f, nelem(f));
689 		if(nf == 0)
690 			continue;
691 		if(strcmp(f[0], "mount") == 0){
692 			if(nf != 4){
693 				werrstr("syntax error: mount /path /dev|score mtime");
694 				goto badline;
695 			}
696 			if(f[1][0] != '/'){
697 				werrstr("unrooted path %s", f[1]);
698 				goto badline;
699 			}
700 			score = nil;
701 			image = nil;
702 			if(f[2][0] == '/'){
703 				if(access(f[2], AEXIST) < 0){
704 					werrstr("image %s does not exist", f[2]);
705 					goto badline;
706 				}
707 				image = f[2];
708 			}else{
709 				if(vtparsescore(f[2], &pref, scorebuf) < 0){
710 					werrstr("bad score %s", f[2]);
711 					goto badline;
712 				}
713 				score = scorebuf;
714 			}
715 			if(parsetime(f[3], &time) < 0){
716 				fprint(2, "%s:%d: bad time %s\n", name, line, f[3]);
717 				time = 1;
718 			}
719 			ctreemountfsys(c.ctree, f[1], time, score, image);
720 			continue;
721 		}
722 		if(strcmp(f[0], "allow") == 0 || strcmp(f[0], "deny") == 0){
723 			if(nf != 2){
724 				werrstr("syntax error: allow|deny ip[/mask]"); 
725 				goto badline;
726 			}
727 			c.ok = erealloc(c.ok, (c.nok+1)*sizeof(c.ok[0]));
728 			if(parseipandmask(f[1], c.ok[c.nok].ip, c.ok[c.nok].mask) < 0){
729 				werrstr("bad ip[/mask]: %s", f[1]);
730 				goto badline;
731 			}
732 			c.ok[c.nok].okay = (strcmp(f[0], "allow") == 0);
733 			c.nok++;
734 			continue;
735 		}
736 		werrstr("unknown verb '%s'", f[0]);
737 	badline:
738 		fprint(2, "%s:%d: %r\n", name, line);
739 	}
740 	Bterm(b);
741 
742 	sweepctree(c.ctree);
743 	free(cp->ok);
744 	*cp = c;
745 	return 0;
746 }
747 
748 int
749 ipokay(uchar *ip, ushort port)
750 {
751 	int i;
752 	uchar ipx[IPaddrlen];
753 	Ipokay *ok; 
754 	
755 	for(i=0; i<config.nok; i++){
756 		ok = &config.ok[i];
757 		maskip(ip, ok->mask, ipx);
758 if(0) fprint(2, "%I & %I = %I (== %I?)\n",
759 	ip, ok->mask, ipx, ok->ip);
760 		if(memcmp(ipx, ok->ip, IPaddrlen) == 0)
761 			return ok->okay;
762 	}
763 	if(config.nok == 0)	/* all is permitted */
764 		return 1;
765 	/* otherwise default is none allowed */
766 	return 0;
767 }
768 
769 Nfs3Status
770 cnodelookup(Ctree *t, Cnode **np, char *name)
771 {
772 	Cnode *n, *nn;
773 	
774 	n = *np;
775 	if(n->isblackhole)
776 		return Nfs3Ok;
777 	if((nn = cnodewalk(n, name, strlen(name), 0)) == nil){
778 		if(n->ismtpt || n->fsys){
779 			if((nn = cnodewalk(n, "", 0, 1)) == nil){
780 				nn = mkcnode(t, n, "", 0, (char*)n->handle, SHA1dlen);
781 				nn->isblackhole = 1;
782 			}
783 			nn->mark = 0;
784 		}
785 	}
786 	if(nn == nil)
787 		return Nfs3ErrNoEnt;
788 	*np = nn;
789 	return Nfs3Ok;
790 }
791 
792 Nfs3Status
793 cnodegetattr(Cnode *n, Nfs3Attr *attr)
794 {
795 	memset(attr, 0, sizeof *attr);
796 	if(n->read){
797 		attr->type = Nfs3FileReg;
798 		attr->mode = 0444;
799 		attr->size = 512;
800 		attr->nlink = 1;
801 	}else{
802 		attr->type = Nfs3FileDir;
803 		attr->mode = 0555;
804 		attr->size = 1024;
805 		attr->nlink = 10;
806 	}
807 	attr->fileid = *(u64int*)n->handle;
808 	attr->atime.sec = n->mtime;
809 	attr->mtime.sec = n->mtime;
810 	attr->ctime.sec = n->mtime;
811 	return Nfs3Ok;
812 }
813 
814 Nfs3Status
815 cnodereaddir(Cnode *n, u32int count, u64int cookie, uchar **pdata, u32int *pcount, u1int *peof)
816 {
817 	uchar *data, *p, *ep, *np;
818 	u64int c;
819 	Nfs3Entry ne;
820 	
821 	n = n->kidlist;
822 	c = cookie;
823 	for(; c && n; c--)
824 		n = n->nextsib;
825 	if(n == nil){
826 		*pdata = 0;
827 		*pcount = 0;
828 		*peof = 1;
829 		return Nfs3Ok;
830 	}
831 	
832 	data = emalloc(count);
833 	p = data;
834 	ep = data+count;
835 	while(n && p < ep){
836 		if(n->mark || n->name[0] == '+'){
837 			n = n->nextsib;
838 			++cookie;
839 			continue;
840 		}
841 		ne.name = n->name;
842 		ne.namelen = strlen(n->name);
843 		ne.cookie = ++cookie;
844 		ne.fileid = *(u64int*)n->handle;
845 		if(nfs3entrypack(p, ep, &np, &ne) < 0)
846 			break;
847 		p = np;
848 		n = n->nextsib;
849 	}
850 	*pdata = data;
851 	*pcount = p - data;
852 	*peof = n==nil;
853 	return Nfs3Ok;
854 }
855 
856 void
857 timerproc(void *v)
858 {
859 	for(;;){
860 		sleep(60*1000);
861 		sendp(timerchan, 0);
862 	}
863 }
864 
865 void
866 timerthread(void *v)
867 {
868 	for(;;){
869 		recvp(timerchan);
870 	/*	refreshconfig(); */
871 	}
872 }
873 
874 /*
875  * Actually serve the NFS requests.  Called from nfs3srv.c.  
876  * Each request runs in its own thread (coroutine).
877  *
878  * Decrypted handles have the form:
879  *
880  *	config[20] - SHA1 hash identifying a config tree node
881  *	glob[10] - SHA1 hash prefix identifying a glob state
882  *	fsyshandle[<=10] - disk file system handle (usually 4 bytes)
883  */
884  
885 /*
886  * A fid represents a point in the file tree.
887  * There are three components, all derived from the handle:
888  *
889  *	- config tree position (also used to find fsys)
890  *	- glob state for exclusions
891  *	- file system position
892  */
893 enum
894 {
895 	HAccess,
896 	HAttr,
897 	HWalk,
898 	HDotdot,
899 	HRead
900 };
901 typedef struct Fid Fid;
902 struct Fid
903 {
904 	Cnode *cnode;
905 	Fsys *fsys;
906 	Nfs3Handle fsyshandle;
907 };
908 
909 int
910 handlecmp(Nfs3Handle *h, Nfs3Handle *h1)
911 {
912 	if(h->len != h1->len)
913 		return h->len - h1->len;
914 	return memcmp(h->h, h1->h, h->len);
915 }
916 
917 Nfs3Status
918 handletofid(Nfs3Handle *eh, Fid *fid, int mode)
919 {
920 	int domount;
921 	Cnode *n;
922 	Disk *disk, *cdisk;
923 	Fsys *fsys;
924 	Nfs3Status ok;
925 	Nfs3Handle h2, *h, *fh;
926 	
927 	memset(fid, 0, sizeof *fid);
928 
929 	domount = 1;
930 	if(mode == HDotdot)
931 		domount = 0;
932 	/*
933 	 * Not necessary, but speeds up ls -l /dump/2005
934 	 * HAttr and HAccess must be handled the same way
935 	 * because both can be used to fetch attributes.
936 	 * Acting differently yields inconsistencies at mount points,
937 	 * and causes FreeBSD ls -l to fail.
938 	 */
939 	if(mode == HAttr || mode == HAccess)
940 		domount = 0;
941 
942 	/*
943 	 * Decrypt handle.
944 	 */
945 	h2 = *eh;
946 	h = &h2;
947 	if((ok = hdecrypt(h)) != Nfs3Ok)
948 		return ok;
949 	trace("handletofid: decrypted %.*lH\n", h->len, h->h);
950 	if(h->len < FsysHandleOffset)
951 		return Nfs3ErrBadHandle;
952 
953 	/*
954 	 * Find place in config tree.
955 	 */
956 	if((n = cnodebyhandle(config.ctree, h->h)) == nil)
957 		return Nfs3ErrStale;
958 	fid->cnode = n;
959 
960 	if(n->ismtpt && domount){
961 		/*
962 		 * Open fsys for mount point if needed.
963 		 */
964 		if(n->mfsys == nil){
965 			trace("handletofid: mounting %V/%s\n", n->fsysscore, n->fsysimage);
966 			if(n->fsysimage){
967 				if(strcmp(n->fsysimage, "/dev/null") == 0)
968 					return Nfs3ErrAcces;
969 				if((disk = diskopenfile(n->fsysimage)) == nil){
970 					fprint(2, "cannot open disk %s: %r\n", n->fsysimage);
971 					return Nfs3ErrIo;
972 				}
973 				if((cdisk = diskcache(disk, blocksize, 64)) == nil){
974 					fprint(2, "cannot cache disk %s: %r\n", n->fsysimage);
975 					diskclose(disk);
976 				}
977 				disk = cdisk;
978 			}else{
979 				if((disk = diskopenventi(vcache, n->fsysscore)) == nil){
980 					fprint(2, "cannot open venti disk %V: %r\n", n->fsysscore);
981 					return Nfs3ErrIo;
982 				}
983 			}
984 			if((fsys = fsysopen(disk)) == nil){
985 				fprint(2, "cannot open fsys on %V: %r\n", n->fsysscore);
986 				diskclose(disk);
987 				return Nfs3ErrIo;
988 			}
989 			n->mfsys = fsys;
990 			fsysroot(fsys, &n->mfsyshandle);
991 		}
992 		
993 		/*
994 		 * Use inner handle.
995 		 */
996 		fid->fsys = n->mfsys;
997 		fid->fsyshandle = n->mfsyshandle;
998 	}else{
999 		/*
1000 		 * Use fsys handle from tree or from handle.
1001 		 * This assumes that fsyshandle was set by fidtohandle
1002 		 * earlier, so it's not okay to reuse handles (except the root)
1003 		 * across sessions.  The encryption above makes and 
1004 		 * enforces the same restriction, so this is okay.
1005 		 */
1006 		fid->fsys = n->fsys;
1007 		fh = &fid->fsyshandle;
1008 		if(n->isblackhole){
1009 			fh->len = h->len-FsysHandleOffset;
1010 			memmove(fh->h, h->h+FsysHandleOffset, fh->len);
1011 		}else
1012 			*fh = n->fsyshandle;
1013 		trace("handletofid: fsyshandle %.*lH\n", fh->len, fh->h);
1014 	}
1015 
1016 	/*
1017 	 * TO DO (maybe): some sort of path restriction here.
1018 	 */
1019 	trace("handletofid: cnode %s fsys %p fsyshandle %.*lH\n",
1020 		n->name, fid->fsys, fid->fsyshandle.len, fid->fsyshandle.h);
1021 	return Nfs3Ok;
1022 }
1023 
1024 void
1025 _fidtohandle(Fid *fid, Nfs3Handle *h)
1026 {
1027 	Cnode *n;
1028 	
1029 	n = fid->cnode;
1030 	/*
1031 	 * Record fsys handle in n, don't bother sending it to client
1032 	 * for black holes.
1033 	 */
1034 	n->fsys = fid->fsys;
1035 	if(!n->isblackhole){
1036 		n->fsyshandle = fid->fsyshandle;
1037 		fid->fsyshandle.len = 0;
1038 	}
1039 	memmove(h->h, n->handle, CnodeHandleSize);
1040 	memmove(h->h+FsysHandleOffset, fid->fsyshandle.h, fid->fsyshandle.len);
1041 	h->len = FsysHandleOffset+fid->fsyshandle.len;
1042 }
1043 
1044 void
1045 fidtohandle(Fid *fid, Nfs3Handle *h)
1046 {
1047 	_fidtohandle(fid, h);
1048 	hencrypt(h);
1049 }
1050 
1051 void
1052 setrootfid(void)
1053 {
1054 	Fid fid;
1055 	
1056 	memset(&fid, 0, sizeof fid);
1057 	fid.cnode = config.ctree->root;
1058 	_fidtohandle(&fid, &root);
1059 fprint(2, "handle %.*lH\n", root.len, root.h);
1060 }
1061 
1062 void
1063 fsgetroot(Nfs3Handle *h)
1064 {
1065 	*h = root;
1066 	hencrypt(h);
1067 }
1068 
1069 Nfs3Status
1070 fsgetattr(SunAuthUnix *au, Nfs3Handle *h, Nfs3Attr *attr)
1071 {
1072 	Fid fid;
1073 	Nfs3Status ok;
1074 
1075 	trace("getattr %.*lH\n", h->len, h->h);
1076 	if((ok = handletofid(h, &fid, HAttr)) != Nfs3Ok)
1077 		return ok;
1078 	if(fid.fsys)
1079 		return fsysgetattr(fid.fsys, au, &fid.fsyshandle, attr);
1080 	else
1081 		return cnodegetattr(fid.cnode, attr);
1082 }
1083 
1084 /*
1085  * Lookup is always the hard part.
1086  */
1087 Nfs3Status
1088 fslookup(SunAuthUnix *au, Nfs3Handle *h, char *name, Nfs3Handle *nh)
1089 {
1090 	Fid fid;
1091 	Cnode *n;
1092 	Nfs3Status ok;
1093 	Nfs3Handle xh;
1094 	int mode;
1095 	
1096 	trace("lookup %.*lH %s\n", h->len, h->h, name);
1097 
1098 	mode = HWalk;
1099 	if(strcmp(name, "..") == 0 || strcmp(name, ".") == 0)
1100 		mode = HDotdot;
1101 	if((ok = handletofid(h, &fid, mode)) != Nfs3Ok){
1102 		nfs3errstr(ok);
1103 		trace("lookup: handletofid %r\n");
1104 		return ok;
1105 	}
1106 	
1107 	if(strcmp(name, ".") == 0){
1108 		fidtohandle(&fid, nh);
1109 		return Nfs3Ok;
1110 	}
1111 
1112 	/*
1113 	 * Walk down file system and cnode simultaneously.
1114 	 * If dotdot and file system doesn't move, need to walk
1115 	 * up cnode.  Save the corresponding fsys handles in
1116 	 * the cnode as we walk down so that we'll have them 
1117 	 * for dotdotting back up.
1118 	 */
1119 	n = fid.cnode;
1120 	if(mode == HWalk){
1121 		/*
1122 		 * Walk down config tree and file system simultaneously.
1123 		 */
1124 		if((ok = cnodelookup(config.ctree, &n, name)) != Nfs3Ok){
1125 			nfs3errstr(ok);
1126 			trace("lookup: cnodelookup: %r\n");
1127 			return ok;
1128 		}
1129 		fid.cnode = n;
1130 		if(fid.fsys){
1131 			if((ok = fsyslookup(fid.fsys, au, &fid.fsyshandle, name, &xh)) != Nfs3Ok){
1132 				nfs3errstr(ok);
1133 				trace("lookup: fsyslookup: %r\n");
1134 				return ok;
1135 			}
1136 			fid.fsyshandle = xh;
1137 		}		
1138 	}else{
1139 		/*
1140 		 * Walking dotdot.  Ick.
1141 		 */
1142 		trace("lookup dotdot fsys=%p\n", fid.fsys);
1143 		if(fid.fsys){
1144 			/*
1145 			 * Walk up file system, then try up config tree.
1146 			 */
1147 			if((ok = fsyslookup(fid.fsys, au, &fid.fsyshandle, "..", &xh)) != Nfs3Ok){
1148 				nfs3errstr(ok);
1149 				trace("lookup fsyslookup: %r\n");
1150 				return ok;
1151 			}
1152 			fid.fsyshandle = xh;
1153 
1154 			/*
1155 			 * Usually just go to n->parent.
1156 			 * 
1157 			 * If we're in a subtree of the mounted file system that
1158 			 * isn't represented explicitly by the config tree (instead
1159 			 * the black hole node represents the entire file tree),
1160 			 * then we only go to n->parent when we've dotdotted back
1161 			 * to the right handle.
1162 			 */
1163 			if(n->parent == nil)
1164 				trace("lookup dotdot: no parent\n");
1165 			else{
1166 				trace("lookup dotdot: parent %.*lH, have %.*lH\n",
1167 					n->parent->fsyshandle.len, n->parent->fsyshandle.h,
1168 					xh.len, xh.h);
1169 			}
1170 			
1171 			if(n->isblackhole){
1172 				if(handlecmp(&n->parent->mfsyshandle, &xh) == 0)
1173 					n = n->parent;
1174 			}else{
1175 				if(n->parent)
1176 					n = n->parent;
1177 			}
1178 		}else{
1179 			/*
1180 			 * No file system, just walk up.
1181 			 */
1182 			if(n->parent)
1183 				n = n->parent;
1184 		}
1185 		fid.fsys = n->fsys;
1186 		fid.fsyshandle = n->fsyshandle;
1187 		fid.cnode = n;
1188 	}
1189 	fidtohandle(&fid, nh);
1190 	return Nfs3Ok;
1191 }
1192 
1193 Nfs3Status
1194 fsaccess(SunAuthUnix *au, Nfs3Handle *h, u32int want, u32int *got, Nfs3Attr *attr)
1195 {
1196 	Fid fid;
1197 	Nfs3Status ok;
1198 	
1199 	trace("access %.*lH 0x%ux\n", h->len, h->h, want);
1200 	if((ok = handletofid(h, &fid, HAccess)) != Nfs3Ok)
1201 		return ok;
1202 	if(fid.fsys)
1203 		return fsysaccess(fid.fsys, au, &fid.fsyshandle, want, got, attr);
1204 	*got = want & (Nfs3AccessRead|Nfs3AccessLookup|Nfs3AccessExecute);
1205 	return cnodegetattr(fid.cnode, attr);
1206 }
1207 
1208 Nfs3Status
1209 fsreadlink(SunAuthUnix *au, Nfs3Handle *h, char **link)
1210 {
1211 	Fid fid;
1212 	Nfs3Status ok;
1213 	
1214 	trace("readlink %.*lH\n", h->len, h->h);
1215 	if((ok = handletofid(h, &fid, HRead)) != Nfs3Ok)
1216 		return ok;
1217 	if(fid.fsys)
1218 		return fsysreadlink(fid.fsys, au, &fid.fsyshandle, link);
1219 	*link = 0;
1220 	return Nfs3ErrNotSupp;
1221 }
1222 
1223 Nfs3Status
1224 fsreadfile(SunAuthUnix *au, Nfs3Handle *h, u32int count, u64int offset, uchar **data, u32int *pcount, u1int *peof)
1225 {
1226 	Fid fid;
1227 	Nfs3Status ok;
1228 	
1229 	trace("readfile %.*lH\n", h->len, h->h);
1230 	if((ok = handletofid(h, &fid, HRead)) != Nfs3Ok)
1231 		return ok;
1232 	if(fid.cnode->read)
1233 		return fid.cnode->read(fid.cnode, count, offset, data, pcount, peof);
1234 	if(fid.fsys)
1235 		return fsysreadfile(fid.fsys, au, &fid.fsyshandle, count, offset, data, pcount, peof);
1236 	return Nfs3ErrNotSupp;
1237 }
1238 
1239 Nfs3Status
1240 fsreaddir(SunAuthUnix *au, Nfs3Handle *h, u32int len, u64int cookie, uchar **pdata, u32int *pcount, u1int *peof)
1241 {
1242 	Fid fid;
1243 	Nfs3Status ok;
1244 
1245 	trace("readdir %.*lH\n", h->len, h->h);	
1246 	if((ok = handletofid(h, &fid, HRead)) != Nfs3Ok)
1247 		return ok;
1248 	if(fid.fsys)
1249 		return fsysreaddir(fid.fsys, au, &fid.fsyshandle, len, cookie, pdata, pcount, peof);
1250 	return cnodereaddir(fid.cnode, len, cookie, pdata, pcount, peof);
1251 }
1252 
1253 Nfs3Status
1254 logread(Cnode *n, u32int count, u64int offset, uchar **data, u32int *pcount, u1int *peof)
1255 {
1256 	*pcount = 0;
1257 	*peof = 1;
1258 	return Nfs3Ok;
1259 }
1260 
1261 Nfs3Status
1262 refreshdiskread(Cnode *n, u32int count, u64int offset, uchar **data, u32int *pcount, u1int *peof)
1263 {
1264 	char buf[128];
1265 	
1266 	if(offset != 0){
1267 		*pcount = 0;
1268 		*peof = 1;
1269 		return Nfs3Ok;
1270 	}
1271 	if(refreshdisk() < 0)
1272 		snprint(buf, sizeof buf, "refreshdisk: %r\n");
1273 	else
1274 		strcpy(buf, "ok\n");
1275 	*data = emalloc(strlen(buf));
1276 	strcpy((char*)*data, buf);
1277 	*pcount = strlen(buf);
1278 	*peof = 1;
1279 	return Nfs3Ok;
1280 }
1281 
1282 Nfs3Status
1283 refreshconfigread(Cnode *n, u32int count, u64int offset, uchar **data, u32int *pcount, u1int *peof)
1284 {
1285 	char buf[128];
1286 	
1287 	if(offset != 0){
1288 		*pcount = 0;
1289 		*peof = 1;
1290 		return Nfs3Ok;
1291 	}
1292 	if(readconfigfile(&config) < 0)
1293 		snprint(buf, sizeof buf, "readconfig: %r\n");
1294 	else
1295 		strcpy(buf, "ok\n");
1296 	*data = emalloc(strlen(buf));
1297 	strcpy((char*)*data, buf);
1298 	*pcount = strlen(buf);
1299 	*peof = 1;
1300 	return Nfs3Ok;
1301 }
1302