xref: /illumos-gate/usr/src/lib/libnsl/rpc/svc.c (revision b1cdc720)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright 2014 Nexenta Systems, Inc.  All rights reserved.
25  * Copyright (c) 2016 by Delphix. All rights reserved.
26  * Copyright 2017 Joyent Inc
27  */
28 /*
29  * Copyright 1993 OpenVision Technologies, Inc., All Rights Reserved.
30  */
31 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
32 /* All Rights Reserved */
33 /*
34  * Portions of this source code were derived from Berkeley
35  * 4.3 BSD under license from the Regents of the University of
36  * California.
37  */
38 
39 /*
40  * svc.c, Server-side remote procedure call interface.
41  *
42  * There are two sets of procedures here.  The xprt routines are
43  * for handling transport handles.  The svc routines handle the
44  * list of service routines.
45  *
46  */
47 
48 #include "mt.h"
49 #include "rpc_mt.h"
50 #include <assert.h>
51 #include <errno.h>
52 #include <sys/types.h>
53 #include <stropts.h>
54 #include <sys/conf.h>
55 #include <rpc/rpc.h>
56 #include <rpc/auth.h>
57 #ifdef PORTMAP
58 #include <rpc/pmap_clnt.h>
59 #endif
60 #include <sys/poll.h>
61 #include <netconfig.h>
62 #include <syslog.h>
63 #include <stdlib.h>
64 #include <unistd.h>
65 #include <string.h>
66 #include <limits.h>
67 
68 extern bool_t __svc_get_door_cred();
69 extern bool_t __rpc_get_local_cred();
70 
71 SVCXPRT **svc_xports;
72 static int nsvc_xports; 	/* total number of svc_xports allocated */
73 
74 XDR **svc_xdrs;		/* common XDR receive area */
75 int nsvc_xdrs;		/* total number of svc_xdrs allocated */
76 
77 int __rpc_use_pollfd_done;	/* to unlimit the number of connections */
78 
79 #define	NULL_SVC ((struct svc_callout *)0)
80 
81 /*
82  * The services list
83  * Each entry represents a set of procedures (an rpc program).
84  * The dispatch routine takes request structs and runs the
85  * appropriate procedure.
86  */
87 static struct svc_callout {
88 	struct svc_callout *sc_next;
89 	rpcprog_t	    sc_prog;
90 	rpcvers_t	    sc_vers;
91 	char		   *sc_netid;
92 	void		    (*sc_dispatch)();
93 } *svc_head;
94 extern rwlock_t	svc_lock;
95 
96 static struct svc_callout *svc_find();
97 int _svc_prog_dispatch();
98 void svc_getreq_common();
99 char *strdup();
100 
101 extern mutex_t	svc_door_mutex;
102 extern cond_t	svc_door_waitcv;
103 extern int	svc_ndoorfds;
104 extern SVCXPRT_LIST *_svc_xprtlist;
105 extern mutex_t xprtlist_lock;
106 extern void __svc_rm_from_xlist();
107 
108 #if !defined(_LP64)
109 extern fd_set _new_svc_fdset;
110 #endif
111 
112 /*
113  * If the allocated array of reactor is too small, this value is used as a
114  * margin. This reduces the number of allocations.
115  */
116 #define	USER_FD_INCREMENT 5
117 
118 static void add_pollfd(int fd, short events);
119 static void remove_pollfd(int fd);
120 static void __svc_remove_input_of_fd(int fd);
121 
122 /*
123  * Data used to handle reactor:
124  * 	- one file descriptor we listen to,
125  *	- one callback we call if the fd pops,
126  *	- and a cookie passed as a parameter to the callback.
127  *
128  * The structure is an array indexed on the file descriptor. Each entry is
129  * pointing to the first element of a double-linked list of callback.
130  * only one callback may be associated to a couple (fd, event).
131  */
132 
133 struct _svc_user_fd_head;
134 
135 typedef struct {
136 	struct _svc_user_fd_node *next;
137 	struct _svc_user_fd_node *previous;
138 } _svc_user_link;
139 
140 typedef struct _svc_user_fd_node {
141 	_svc_user_link lnk;
142 	svc_input_id_t id;
143 	int	    fd;
144 	unsigned int   events;
145 	svc_callback_t callback;
146 	void*	  cookie;
147 } _svc_user_fd_node;
148 
149 typedef struct _svc_user_fd_head {
150 	struct _svc_user_fd_node *list;
151 	unsigned int mask;    /* logical OR of all sub-masks */
152 } _svc_user_fd_head;
153 
154 
155 /* Array of defined reactor - indexed on file descriptor */
156 static _svc_user_fd_head *svc_userfds = NULL;
157 
158 /* current size of file descriptor */
159 static int svc_nuserfds = 0;
160 
161 /* Mutex to ensure MT safe operations for user fds callbacks. */
162 static mutex_t svc_userfds_lock = DEFAULTMUTEX;
163 
164 
165 /*
166  * This structure is used to have constant time alogrithms. There is an array
167  * of this structure as large as svc_nuserfds. When the user is registering a
168  * new callback, the address of the created structure is stored in a cell of
169  * this array. The address of this cell is the returned unique identifier.
170  *
171  * On removing, the id is given by the user, then we know if this cell is
172  * filled or not (with free). If it is free, we return an error. Otherwise,
173  * we can free the structure pointed by fd_node.
174  *
175  * On insertion, we use the linked list created by (first_free,
176  * next_free). In this way with a constant time computation, we can give a
177  * correct index to the user.
178  */
179 
180 typedef struct _svc_management_user_fd {
181 	bool_t free;
182 	union {
183 		svc_input_id_t next_free;
184 		_svc_user_fd_node *fd_node;
185 	} data;
186 } _svc_management_user_fd;
187 
188 /* index to the first free elem */
189 static svc_input_id_t first_free = (svc_input_id_t)-1;
190 /* the size of this array is the same as svc_nuserfds */
191 static _svc_management_user_fd* user_fd_mgt_array = NULL;
192 
193 /* current size of user_fd_mgt_array */
194 static int svc_nmgtuserfds = 0;
195 
196 
197 /* Define some macros to access data associated to registration ids. */
198 #define	node_from_id(id) (user_fd_mgt_array[(int)id].data.fd_node)
199 #define	is_free_id(id) (user_fd_mgt_array[(int)id].free)
200 
201 #ifndef POLLSTANDARD
202 #define	POLLSTANDARD \
203 	(POLLIN|POLLPRI|POLLOUT|POLLRDNORM|POLLRDBAND| \
204 	POLLWRBAND|POLLERR|POLLHUP|POLLNVAL)
205 #endif
206 
207 /*
208  * To free an Id, we set the cell as free and insert its address in the list
209  * of free cell.
210  */
211 
212 static void
_svc_free_id(const svc_input_id_t id)213 _svc_free_id(const svc_input_id_t id)
214 {
215 	assert(((int)id >= 0) && ((int)id < svc_nmgtuserfds));
216 	user_fd_mgt_array[(int)id].free = TRUE;
217 	user_fd_mgt_array[(int)id].data.next_free = first_free;
218 	first_free = id;
219 }
220 
221 /*
222  * To get a free cell, we just have to take it from the free linked list and
223  * set the flag to "not free". This function also allocates new memory if
224  * necessary
225  */
226 static svc_input_id_t
_svc_attribute_new_id(_svc_user_fd_node * node)227 _svc_attribute_new_id(_svc_user_fd_node *node)
228 {
229 	int selected_index = (int)first_free;
230 	assert(node != NULL);
231 
232 	if (selected_index == -1) {
233 		/* Allocate new entries */
234 		int L_inOldSize = svc_nmgtuserfds;
235 		int i;
236 		_svc_management_user_fd *tmp;
237 
238 		svc_nmgtuserfds += USER_FD_INCREMENT;
239 
240 		tmp = realloc(user_fd_mgt_array,
241 		    svc_nmgtuserfds * sizeof (_svc_management_user_fd));
242 
243 		if (tmp == NULL) {
244 			syslog(LOG_ERR, "_svc_attribute_new_id: out of memory");
245 			svc_nmgtuserfds = L_inOldSize;
246 			errno = ENOMEM;
247 			return ((svc_input_id_t)-1);
248 		}
249 
250 		user_fd_mgt_array = tmp;
251 
252 		for (i = svc_nmgtuserfds - 1; i >= L_inOldSize; i--)
253 			_svc_free_id((svc_input_id_t)i);
254 		selected_index = (int)first_free;
255 	}
256 
257 	node->id = (svc_input_id_t)selected_index;
258 	first_free = user_fd_mgt_array[selected_index].data.next_free;
259 
260 	user_fd_mgt_array[selected_index].data.fd_node = node;
261 	user_fd_mgt_array[selected_index].free = FALSE;
262 
263 	return ((svc_input_id_t)selected_index);
264 }
265 
266 /*
267  * Access to a pollfd treatment. Scan all the associated callbacks that have
268  * at least one bit in their mask that masks a received event.
269  *
270  * If event POLLNVAL is received, we check that one callback processes it, if
271  * not, then remove the file descriptor from the poll. If there is one, let
272  * the user do the work.
273  */
274 void
__svc_getreq_user(struct pollfd * pfd)275 __svc_getreq_user(struct pollfd *pfd)
276 {
277 	int fd = pfd->fd;
278 	short revents = pfd->revents;
279 	bool_t invalHandled = FALSE;
280 	_svc_user_fd_node *node;
281 
282 	(void) mutex_lock(&svc_userfds_lock);
283 
284 	if ((fd < 0) || (fd >= svc_nuserfds)) {
285 		(void) mutex_unlock(&svc_userfds_lock);
286 		return;
287 	}
288 
289 	node = svc_userfds[fd].list;
290 
291 	/* check if at least one mask fits */
292 	if (0 == (revents & svc_userfds[fd].mask)) {
293 		(void) mutex_unlock(&svc_userfds_lock);
294 		return;
295 	}
296 
297 	while ((svc_userfds[fd].mask != 0) && (node != NULL)) {
298 		/*
299 		 * If one of the received events maps the ones the node listens
300 		 * to
301 		 */
302 		_svc_user_fd_node *next = node->lnk.next;
303 
304 		if (node->callback != NULL) {
305 			if (node->events & revents) {
306 				if (revents & POLLNVAL) {
307 					invalHandled = TRUE;
308 				}
309 
310 				/*
311 				 * The lock must be released before calling the
312 				 * user function, as this function can call
313 				 * svc_remove_input() for example.
314 				 */
315 				(void) mutex_unlock(&svc_userfds_lock);
316 				node->callback(node->id, node->fd,
317 				    node->events & revents, node->cookie);
318 				/*
319 				 * Do not use the node structure anymore, as it
320 				 * could have been deallocated by the previous
321 				 * callback.
322 				 */
323 				(void) mutex_lock(&svc_userfds_lock);
324 			}
325 		}
326 		node = next;
327 	}
328 
329 	if ((revents & POLLNVAL) && !invalHandled)
330 		__svc_remove_input_of_fd(fd);
331 	(void) mutex_unlock(&svc_userfds_lock);
332 }
333 
334 
335 /*
336  * Check if a file descriptor is associated with a user reactor.
337  * To do this, just check that the array indexed on fd has a non-void linked
338  * list (ie. first element is not NULL)
339  */
340 bool_t
__is_a_userfd(int fd)341 __is_a_userfd(int fd)
342 {
343 	/* Checks argument */
344 	if ((fd < 0) || (fd >= svc_nuserfds))
345 		return (FALSE);
346 	return ((svc_userfds[fd].mask == 0x0000)? FALSE:TRUE);
347 }
348 
349 /* free everything concerning user fd */
350 /* used in svc_run.c => no static */
351 
352 void
__destroy_userfd(void)353 __destroy_userfd(void)
354 {
355 	int one_fd;
356 	/* Clean user fd */
357 	if (svc_userfds != NULL) {
358 		for (one_fd = 0; one_fd < svc_nuserfds; one_fd++) {
359 			_svc_user_fd_node *node;
360 
361 			node = svc_userfds[one_fd].list;
362 			while (node != NULL) {
363 				_svc_user_fd_node *tmp = node;
364 				_svc_free_id(node->id);
365 				node = node->lnk.next;
366 				free(tmp);
367 			}
368 		}
369 
370 		free(user_fd_mgt_array);
371 		user_fd_mgt_array = NULL;
372 		first_free = (svc_input_id_t)-1;
373 
374 		free(svc_userfds);
375 		svc_userfds = NULL;
376 		svc_nuserfds = 0;
377 	}
378 }
379 
380 /*
381  * Remove all the callback associated with a fd => useful when the fd is
382  * closed for instance
383  */
384 static void
__svc_remove_input_of_fd(int fd)385 __svc_remove_input_of_fd(int fd)
386 {
387 	_svc_user_fd_node **pnode;
388 	_svc_user_fd_node *tmp;
389 
390 	if ((fd < 0) || (fd >= svc_nuserfds))
391 		return;
392 
393 	pnode = &svc_userfds[fd].list;
394 	while ((tmp = *pnode) != NULL) {
395 		*pnode = tmp->lnk.next;
396 
397 		_svc_free_id(tmp->id);
398 		free(tmp);
399 	}
400 
401 	svc_userfds[fd].mask = 0;
402 }
403 
404 /*
405  * Allow user to add an fd in the poll list. If it does not succeed, return
406  * -1. Otherwise, return a svc_id
407  */
408 
409 svc_input_id_t
svc_add_input(int user_fd,unsigned int events,svc_callback_t user_callback,void * cookie)410 svc_add_input(int user_fd, unsigned int events,
411     svc_callback_t user_callback, void *cookie)
412 {
413 	_svc_user_fd_node *new_node;
414 
415 	if (user_fd < 0) {
416 		errno = EINVAL;
417 		return ((svc_input_id_t)-1);
418 	}
419 
420 	if ((events == 0x0000) ||
421 	    (events & ~(POLLIN|POLLPRI|POLLOUT|POLLRDNORM|POLLRDBAND|\
422 	    POLLWRBAND|POLLERR|POLLHUP|POLLNVAL))) {
423 		errno = EINVAL;
424 		return ((svc_input_id_t)-1);
425 	}
426 
427 	(void) mutex_lock(&svc_userfds_lock);
428 
429 	if ((user_fd < svc_nuserfds) &&
430 	    (svc_userfds[user_fd].mask & events) != 0) {
431 		/* Already registrated call-back */
432 		errno = EEXIST;
433 		(void) mutex_unlock(&svc_userfds_lock);
434 		return ((svc_input_id_t)-1);
435 	}
436 
437 	/* Handle memory allocation. */
438 	if (user_fd >= svc_nuserfds) {
439 		int oldSize = svc_nuserfds;
440 		int i;
441 		_svc_user_fd_head *tmp;
442 
443 		svc_nuserfds = (user_fd + 1) + USER_FD_INCREMENT;
444 
445 		tmp = realloc(svc_userfds,
446 		    svc_nuserfds * sizeof (_svc_user_fd_head));
447 
448 		if (tmp == NULL) {
449 			syslog(LOG_ERR, "svc_add_input: out of memory");
450 			svc_nuserfds = oldSize;
451 			errno = ENOMEM;
452 			(void) mutex_unlock(&svc_userfds_lock);
453 			return ((svc_input_id_t)-1);
454 		}
455 
456 		svc_userfds = tmp;
457 
458 		for (i = oldSize; i < svc_nuserfds; i++) {
459 			svc_userfds[i].list = NULL;
460 			svc_userfds[i].mask = 0;
461 		}
462 	}
463 
464 	new_node = malloc(sizeof (_svc_user_fd_node));
465 	if (new_node == NULL) {
466 		syslog(LOG_ERR, "svc_add_input: out of memory");
467 		errno = ENOMEM;
468 		(void) mutex_unlock(&svc_userfds_lock);
469 		return ((svc_input_id_t)-1);
470 	}
471 
472 	/* create a new node */
473 	new_node->fd		= user_fd;
474 	new_node->events	= events;
475 	new_node->callback	= user_callback;
476 	new_node->cookie	= cookie;
477 
478 	if (_svc_attribute_new_id(new_node) == -1) {
479 		(void) mutex_unlock(&svc_userfds_lock);
480 		free(new_node);
481 		return ((svc_input_id_t)-1);
482 	}
483 
484 	/* Add the new element at the beginning of the list. */
485 	if (svc_userfds[user_fd].list != NULL)
486 		svc_userfds[user_fd].list->lnk.previous = new_node;
487 	new_node->lnk.next = svc_userfds[user_fd].list;
488 	new_node->lnk.previous = NULL;
489 
490 	svc_userfds[user_fd].list = new_node;
491 
492 	/* refresh global mask for this file desciptor */
493 	svc_userfds[user_fd].mask |= events;
494 
495 	/* refresh mask for the poll */
496 	add_pollfd(user_fd, (svc_userfds[user_fd].mask));
497 
498 	(void) mutex_unlock(&svc_userfds_lock);
499 	return (new_node->id);
500 }
501 
502 int
svc_remove_input(svc_input_id_t id)503 svc_remove_input(svc_input_id_t id)
504 {
505 	_svc_user_fd_node* node;
506 	_svc_user_fd_node* next;
507 	_svc_user_fd_node* previous;
508 	int fd;		/* caching optim */
509 
510 	(void) mutex_lock(&svc_userfds_lock);
511 
512 	/* Immediately update data for id management */
513 	if (user_fd_mgt_array == NULL || id >= svc_nmgtuserfds ||
514 	    is_free_id(id)) {
515 		errno = EINVAL;
516 		(void) mutex_unlock(&svc_userfds_lock);
517 		return (-1);
518 	}
519 
520 	node = node_from_id(id);
521 	assert(node != NULL);
522 
523 	_svc_free_id(id);
524 	next		= node->lnk.next;
525 	previous	= node->lnk.previous;
526 	fd		= node->fd; /* caching optim */
527 
528 	/* Remove this node from the list. */
529 	if (previous != NULL) {
530 		previous->lnk.next = next;
531 	} else {
532 		assert(svc_userfds[fd].list == node);
533 		svc_userfds[fd].list = next;
534 	}
535 	if (next != NULL)
536 		next->lnk.previous = previous;
537 
538 	/* Remove the node flags from the global mask */
539 	svc_userfds[fd].mask ^= node->events;
540 
541 	free(node);
542 	if (svc_userfds[fd].mask == 0) {
543 		assert(svc_userfds[fd].list == NULL);
544 		remove_pollfd(fd);
545 	} else {
546 		assert(svc_userfds[fd].list != NULL);
547 	}
548 	/* <=> CLEAN NEEDED TO SHRINK MEMORY USAGE */
549 
550 	(void) mutex_unlock(&svc_userfds_lock);
551 	return (0);
552 }
553 
554 /*
555  * Provides default service-side functions for authentication flavors
556  * that do not use all the fields in struct svc_auth_ops.
557  */
558 
559 /*ARGSUSED*/
560 static int
authany_wrap(AUTH * auth,XDR * xdrs,xdrproc_t xfunc,caddr_t xwhere)561 authany_wrap(AUTH *auth, XDR *xdrs, xdrproc_t xfunc, caddr_t xwhere)
562 {
563 	return (*xfunc)(xdrs, xwhere);
564 }
565 
566 struct svc_auth_ops svc_auth_any_ops = {
567 	authany_wrap,
568 	authany_wrap,
569 };
570 
571 /*
572  * Return pointer to server authentication structure.
573  */
574 SVCAUTH *
__svc_get_svcauth(SVCXPRT * xprt)575 __svc_get_svcauth(SVCXPRT *xprt)
576 {
577 /* LINTED pointer alignment */
578 	return (&SVC_XP_AUTH(xprt));
579 }
580 
581 /*
582  * A callback routine to cleanup after a procedure is executed.
583  */
584 void (*__proc_cleanup_cb)() = NULL;
585 
586 void *
__svc_set_proc_cleanup_cb(void * cb)587 __svc_set_proc_cleanup_cb(void *cb)
588 {
589 	void	*tmp = (void *)__proc_cleanup_cb;
590 
591 	__proc_cleanup_cb = (void (*)())cb;
592 	return (tmp);
593 }
594 
595 /* ***************  SVCXPRT related stuff **************** */
596 
597 
598 static int pollfd_shrinking = 1;
599 
600 
601 /*
602  * Add fd to svc_pollfd
603  */
604 static void
add_pollfd(int fd,short events)605 add_pollfd(int fd, short events)
606 {
607 	if (fd < FD_SETSIZE) {
608 		FD_SET(fd, &svc_fdset);
609 #if !defined(_LP64)
610 		FD_SET(fd, &_new_svc_fdset);
611 #endif
612 		svc_nfds++;
613 		svc_nfds_set++;
614 		if (fd >= svc_max_fd)
615 			svc_max_fd = fd + 1;
616 	}
617 	if (fd >= svc_max_pollfd)
618 		svc_max_pollfd = fd + 1;
619 	if (svc_max_pollfd > svc_pollfd_allocd) {
620 		int i = svc_pollfd_allocd;
621 		pollfd_t *tmp;
622 		do {
623 			svc_pollfd_allocd += POLLFD_EXTEND;
624 		} while (svc_max_pollfd > svc_pollfd_allocd);
625 		tmp = realloc(svc_pollfd,
626 		    sizeof (pollfd_t) * svc_pollfd_allocd);
627 		if (tmp != NULL) {
628 			svc_pollfd = tmp;
629 			for (; i < svc_pollfd_allocd; i++)
630 				POLLFD_CLR(i, tmp);
631 		} else {
632 			/*
633 			 * give an error message; undo fdset setting
634 			 * above;  reset the pollfd_shrinking flag.
635 			 * because of this poll will not be done
636 			 * on these fds.
637 			 */
638 			if (fd < FD_SETSIZE) {
639 				FD_CLR(fd, &svc_fdset);
640 #if !defined(_LP64)
641 				FD_CLR(fd, &_new_svc_fdset);
642 #endif
643 				svc_nfds--;
644 				svc_nfds_set--;
645 				if (fd == (svc_max_fd - 1))
646 					svc_max_fd--;
647 			}
648 			if (fd == (svc_max_pollfd - 1))
649 				svc_max_pollfd--;
650 			pollfd_shrinking = 0;
651 			syslog(LOG_ERR, "add_pollfd: out of memory");
652 			_exit(1);
653 		}
654 	}
655 	svc_pollfd[fd].fd	= fd;
656 	svc_pollfd[fd].events	= events;
657 	svc_npollfds++;
658 	svc_npollfds_set++;
659 }
660 
661 /*
662  * the fd is still active but only the bit in fdset is cleared.
663  * do not subtract svc_nfds or svc_npollfds
664  */
665 void
clear_pollfd(int fd)666 clear_pollfd(int fd)
667 {
668 	if (fd < FD_SETSIZE && FD_ISSET(fd, &svc_fdset)) {
669 		FD_CLR(fd, &svc_fdset);
670 #if !defined(_LP64)
671 		FD_CLR(fd, &_new_svc_fdset);
672 #endif
673 		svc_nfds_set--;
674 	}
675 	if (fd < svc_pollfd_allocd && POLLFD_ISSET(fd, svc_pollfd)) {
676 		POLLFD_CLR(fd, svc_pollfd);
677 		svc_npollfds_set--;
678 	}
679 }
680 
681 /*
682  * sets the bit in fdset for an active fd so that poll() is done for that
683  */
684 void
set_pollfd(int fd,short events)685 set_pollfd(int fd, short events)
686 {
687 	if (fd < FD_SETSIZE) {
688 		FD_SET(fd, &svc_fdset);
689 #if !defined(_LP64)
690 		FD_SET(fd, &_new_svc_fdset);
691 #endif
692 		svc_nfds_set++;
693 	}
694 	if (fd < svc_pollfd_allocd) {
695 		svc_pollfd[fd].fd	= fd;
696 		svc_pollfd[fd].events	= events;
697 		svc_npollfds_set++;
698 	}
699 }
700 
701 /*
702  * remove a svc_pollfd entry; it does not shrink the memory
703  */
704 static void
remove_pollfd(int fd)705 remove_pollfd(int fd)
706 {
707 	clear_pollfd(fd);
708 	if (fd == (svc_max_fd - 1))
709 		svc_max_fd--;
710 	svc_nfds--;
711 	if (fd == (svc_max_pollfd - 1))
712 		svc_max_pollfd--;
713 	svc_npollfds--;
714 }
715 
716 /*
717  * delete a svc_pollfd entry; it shrinks the memory
718  * use remove_pollfd if you do not want to shrink
719  */
720 static void
delete_pollfd(int fd)721 delete_pollfd(int fd)
722 {
723 	remove_pollfd(fd);
724 	if (pollfd_shrinking && svc_max_pollfd <
725 	    (svc_pollfd_allocd - POLLFD_SHRINK)) {
726 		do {
727 			svc_pollfd_allocd -= POLLFD_SHRINK;
728 		} while (svc_max_pollfd < (svc_pollfd_allocd - POLLFD_SHRINK));
729 		svc_pollfd = realloc(svc_pollfd,
730 		    sizeof (pollfd_t) * svc_pollfd_allocd);
731 		if (svc_pollfd == NULL) {
732 			syslog(LOG_ERR, "delete_pollfd: out of memory");
733 			_exit(1);
734 		}
735 	}
736 }
737 
738 
739 /*
740  * Activate a transport handle.
741  */
742 void
xprt_register(const SVCXPRT * xprt)743 xprt_register(const SVCXPRT *xprt)
744 {
745 	int fd = xprt->xp_fd;
746 #ifdef CALLBACK
747 	extern void (*_svc_getreqset_proc)();
748 #endif
749 /* VARIABLES PROTECTED BY svc_fd_lock: svc_xports, svc_fdset */
750 
751 	(void) rw_wrlock(&svc_fd_lock);
752 	if (svc_xports == NULL) {
753 		/* allocate some small amount first */
754 		svc_xports = calloc(FD_INCREMENT,  sizeof (SVCXPRT *));
755 		if (svc_xports == NULL) {
756 			syslog(LOG_ERR, "xprt_register: out of memory");
757 			_exit(1);
758 		}
759 		nsvc_xports = FD_INCREMENT;
760 
761 #ifdef CALLBACK
762 		/*
763 		 * XXX: This code does not keep track of the server state.
764 		 *
765 		 * This provides for callback support.	When a client
766 		 * recv's a call from another client on the server fd's,
767 		 * it calls _svc_getreqset_proc() which would return
768 		 * after serving all the server requests.  Also look under
769 		 * clnt_dg.c and clnt_vc.c  (clnt_call part of it)
770 		 */
771 		_svc_getreqset_proc = svc_getreq_poll;
772 #endif
773 	}
774 
775 	while (fd >= nsvc_xports) {
776 		SVCXPRT **tmp_xprts = svc_xports;
777 
778 		/* time to expand svc_xprts */
779 		tmp_xprts = realloc(svc_xports,
780 		    sizeof (SVCXPRT *) * (nsvc_xports + FD_INCREMENT));
781 		if (tmp_xprts == NULL) {
782 			syslog(LOG_ERR, "xprt_register : out of memory.");
783 			_exit(1);
784 		}
785 
786 		svc_xports = tmp_xprts;
787 		(void) memset(&svc_xports[nsvc_xports], 0,
788 		    sizeof (SVCXPRT *) * FD_INCREMENT);
789 		nsvc_xports += FD_INCREMENT;
790 	}
791 
792 	svc_xports[fd] = (SVCXPRT *)xprt;
793 
794 	add_pollfd(fd, MASKVAL);
795 
796 	if (svc_polling) {
797 		char dummy;
798 
799 		/*
800 		 * This happens only in one of the MT modes.
801 		 * Wake up poller.
802 		 */
803 		(void) write(svc_pipe[1], &dummy, sizeof (dummy));
804 	}
805 	/*
806 	 * If already dispatching door based services, start
807 	 * dispatching TLI based services now.
808 	 */
809 	(void) mutex_lock(&svc_door_mutex);
810 	if (svc_ndoorfds > 0)
811 		(void) cond_signal(&svc_door_waitcv);
812 	(void) mutex_unlock(&svc_door_mutex);
813 
814 	if (svc_xdrs == NULL) {
815 		/* allocate initial chunk */
816 		svc_xdrs = calloc(FD_INCREMENT, sizeof (XDR *));
817 		if (svc_xdrs != NULL)
818 			nsvc_xdrs = FD_INCREMENT;
819 		else {
820 			syslog(LOG_ERR, "xprt_register : out of memory.");
821 			_exit(1);
822 		}
823 	}
824 	(void) rw_unlock(&svc_fd_lock);
825 }
826 
827 /*
828  * De-activate a transport handle.
829  */
830 void
__xprt_unregister_private(const SVCXPRT * xprt,bool_t lock_not_held)831 __xprt_unregister_private(const SVCXPRT *xprt, bool_t lock_not_held)
832 {
833 	int fd = xprt->xp_fd;
834 
835 	if (lock_not_held)
836 		(void) rw_wrlock(&svc_fd_lock);
837 	if ((fd < nsvc_xports) && (svc_xports[fd] == xprt)) {
838 		svc_xports[fd] = NULL;
839 		delete_pollfd(fd);
840 	}
841 	if (lock_not_held)
842 		(void) rw_unlock(&svc_fd_lock);
843 	__svc_rm_from_xlist(&_svc_xprtlist, xprt, &xprtlist_lock);
844 }
845 
846 void
xprt_unregister(const SVCXPRT * xprt)847 xprt_unregister(const SVCXPRT *xprt)
848 {
849 	__xprt_unregister_private(xprt, TRUE);
850 }
851 
852 /* ********************** CALLOUT list related stuff ************* */
853 
854 /*
855  * Add a service program to the callout list.
856  * The dispatch routine will be called when a rpc request for this
857  * program number comes in.
858  */
859 bool_t
svc_reg(const SVCXPRT * xprt,const rpcprog_t prog,const rpcvers_t vers,void (* dispatch)(),const struct netconfig * nconf)860 svc_reg(const SVCXPRT *xprt, const rpcprog_t prog, const rpcvers_t vers,
861     void (*dispatch)(), const struct netconfig *nconf)
862 {
863 	struct svc_callout *prev;
864 	struct svc_callout *s, **s2;
865 	struct netconfig *tnconf;
866 	char *netid = NULL;
867 	int flag = 0;
868 
869 /* VARIABLES PROTECTED BY svc_lock: s, prev, svc_head */
870 
871 	if (xprt->xp_netid) {
872 		netid = strdup(xprt->xp_netid);
873 		flag = 1;
874 	} else if (nconf && nconf->nc_netid) {
875 		netid = strdup(nconf->nc_netid);
876 		flag = 1;
877 	} else if ((tnconf = __rpcfd_to_nconf(xprt->xp_fd, xprt->xp_type))
878 	    != NULL) {
879 		netid = strdup(tnconf->nc_netid);
880 		flag = 1;
881 		freenetconfigent(tnconf);
882 	} /* must have been created with svc_raw_create */
883 	if ((netid == NULL) && (flag == 1))
884 		return (FALSE);
885 
886 	(void) rw_wrlock(&svc_lock);
887 	if ((s = svc_find(prog, vers, &prev, netid)) != NULL_SVC) {
888 		if (netid)
889 			free(netid);
890 		if (s->sc_dispatch == dispatch)
891 			goto rpcb_it; /* it is registering another xptr */
892 		(void) rw_unlock(&svc_lock);
893 		return (FALSE);
894 	}
895 	s = malloc(sizeof (struct svc_callout));
896 	if (s == NULL) {
897 		if (netid)
898 			free(netid);
899 		(void) rw_unlock(&svc_lock);
900 		return (FALSE);
901 	}
902 
903 	s->sc_prog = prog;
904 	s->sc_vers = vers;
905 	s->sc_dispatch = dispatch;
906 	s->sc_netid = netid;
907 	s->sc_next = NULL;
908 
909 	/*
910 	 * The ordering of transports is such that the most frequently used
911 	 * one appears first.  So add the new entry to the end of the list.
912 	 */
913 	for (s2 = &svc_head; *s2 != NULL; s2 = &(*s2)->sc_next)
914 		;
915 	*s2 = s;
916 
917 	if ((xprt->xp_netid == NULL) && (flag == 1) && netid)
918 		if ((((SVCXPRT *)xprt)->xp_netid = strdup(netid)) == NULL) {
919 			syslog(LOG_ERR, "svc_reg : strdup failed.");
920 			free(netid);
921 			free(s);
922 			*s2 = NULL;
923 			(void) rw_unlock(&svc_lock);
924 			return (FALSE);
925 		}
926 
927 rpcb_it:
928 	(void) rw_unlock(&svc_lock);
929 
930 	/* now register the information with the local binder service */
931 	if (nconf)
932 		return (rpcb_set(prog, vers, nconf, &xprt->xp_ltaddr));
933 	return (TRUE);
934 	/*NOTREACHED*/
935 }
936 
937 /*
938  * Remove a service program from the callout list.
939  */
940 void
svc_unreg(const rpcprog_t prog,const rpcvers_t vers)941 svc_unreg(const rpcprog_t prog, const rpcvers_t vers)
942 {
943 	struct svc_callout *prev;
944 	struct svc_callout *s;
945 
946 	/* unregister the information anyway */
947 	(void) rpcb_unset(prog, vers, NULL);
948 
949 	(void) rw_wrlock(&svc_lock);
950 	while ((s = svc_find(prog, vers, &prev, NULL)) != NULL_SVC) {
951 		if (prev == NULL_SVC) {
952 			svc_head = s->sc_next;
953 		} else {
954 			prev->sc_next = s->sc_next;
955 		}
956 		s->sc_next = NULL_SVC;
957 		if (s->sc_netid)
958 			free(s->sc_netid);
959 		free(s);
960 	}
961 	(void) rw_unlock(&svc_lock);
962 }
963 
964 #ifdef PORTMAP
965 /*
966  * Add a service program to the callout list.
967  * The dispatch routine will be called when a rpc request for this
968  * program number comes in.
969  * For version 2 portmappers.
970  */
971 bool_t
svc_register(SVCXPRT * xprt,rpcprog_t prog,rpcvers_t vers,void (* dispatch)(),int protocol)972 svc_register(SVCXPRT *xprt, rpcprog_t prog, rpcvers_t vers,
973     void (*dispatch)(), int protocol)
974 {
975 	struct svc_callout *prev;
976 	struct svc_callout *s;
977 	struct netconfig *nconf;
978 	char *netid = NULL;
979 	int flag = 0;
980 
981 	if (xprt->xp_netid) {
982 		netid = strdup(xprt->xp_netid);
983 		flag = 1;
984 	} else if ((ioctl(xprt->xp_fd, I_FIND, "timod") > 0) && ((nconf =
985 	    __rpcfd_to_nconf(xprt->xp_fd, xprt->xp_type)) != NULL)) {
986 		/* fill in missing netid field in SVCXPRT */
987 		netid = strdup(nconf->nc_netid);
988 		flag = 1;
989 		freenetconfigent(nconf);
990 	} /* must be svc_raw_create */
991 
992 	if ((netid == NULL) && (flag == 1))
993 		return (FALSE);
994 
995 	(void) rw_wrlock(&svc_lock);
996 	if ((s = svc_find(prog, vers, &prev, netid)) != NULL_SVC) {
997 		if (netid)
998 			free(netid);
999 		if (s->sc_dispatch == dispatch)
1000 			goto pmap_it;  /* it is registering another xptr */
1001 		(void) rw_unlock(&svc_lock);
1002 		return (FALSE);
1003 	}
1004 	s = malloc(sizeof (struct svc_callout));
1005 	if (s == (struct svc_callout *)0) {
1006 		if (netid)
1007 			free(netid);
1008 		(void) rw_unlock(&svc_lock);
1009 		return (FALSE);
1010 	}
1011 	s->sc_prog = prog;
1012 	s->sc_vers = vers;
1013 	s->sc_dispatch = dispatch;
1014 	s->sc_netid = netid;
1015 	s->sc_next = svc_head;
1016 	svc_head = s;
1017 
1018 	if ((xprt->xp_netid == NULL) && (flag == 1) && netid)
1019 		if ((xprt->xp_netid = strdup(netid)) == NULL) {
1020 			syslog(LOG_ERR, "svc_register : strdup failed.");
1021 			free(netid);
1022 			svc_head = s->sc_next;
1023 			free(s);
1024 			(void) rw_unlock(&svc_lock);
1025 			return (FALSE);
1026 		}
1027 
1028 pmap_it:
1029 	(void) rw_unlock(&svc_lock);
1030 	/* now register the information with the local binder service */
1031 	if (protocol)
1032 		return (pmap_set(prog, vers, protocol, xprt->xp_port));
1033 	return (TRUE);
1034 }
1035 
1036 /*
1037  * Remove a service program from the callout list.
1038  * For version 2 portmappers.
1039  */
1040 void
svc_unregister(rpcprog_t prog,rpcvers_t vers)1041 svc_unregister(rpcprog_t prog, rpcvers_t vers)
1042 {
1043 	struct svc_callout *prev;
1044 	struct svc_callout *s;
1045 
1046 	(void) rw_wrlock(&svc_lock);
1047 	while ((s = svc_find(prog, vers, &prev, NULL)) != NULL_SVC) {
1048 		if (prev == NULL_SVC) {
1049 			svc_head = s->sc_next;
1050 		} else {
1051 			prev->sc_next = s->sc_next;
1052 		}
1053 		s->sc_next = NULL_SVC;
1054 		if (s->sc_netid)
1055 			free(s->sc_netid);
1056 		free(s);
1057 		/* unregister the information with the local binder service */
1058 		(void) pmap_unset(prog, vers);
1059 	}
1060 	(void) rw_unlock(&svc_lock);
1061 }
1062 #endif /* PORTMAP */
1063 
1064 /*
1065  * Search the callout list for a program number, return the callout
1066  * struct.
1067  * Also check for transport as well.  Many routines such as svc_unreg
1068  * dont give any corresponding transport, so dont check for transport if
1069  * netid == NULL
1070  */
1071 static struct svc_callout *
svc_find(rpcprog_t prog,rpcvers_t vers,struct svc_callout ** prev,char * netid)1072 svc_find(rpcprog_t prog, rpcvers_t vers, struct svc_callout **prev, char *netid)
1073 {
1074 	struct svc_callout *s, *p;
1075 
1076 /* WRITE LOCK HELD ON ENTRY: svc_lock */
1077 
1078 /*	assert(RW_WRITE_HELD(&svc_lock)); */
1079 	p = NULL_SVC;
1080 	for (s = svc_head; s != NULL_SVC; s = s->sc_next) {
1081 		if (((s->sc_prog == prog) && (s->sc_vers == vers)) &&
1082 		    ((netid == NULL) || (s->sc_netid == NULL) ||
1083 		    (strcmp(netid, s->sc_netid) == 0)))
1084 			break;
1085 		p = s;
1086 	}
1087 	*prev = p;
1088 	return (s);
1089 }
1090 
1091 
1092 /* ******************* REPLY GENERATION ROUTINES  ************ */
1093 
1094 /*
1095  * Send a reply to an rpc request
1096  */
1097 bool_t
svc_sendreply(const SVCXPRT * xprt,const xdrproc_t xdr_results,const caddr_t xdr_location)1098 svc_sendreply(const SVCXPRT *xprt, const xdrproc_t xdr_results,
1099     const caddr_t xdr_location)
1100 {
1101 	struct rpc_msg rply;
1102 
1103 	rply.rm_direction = REPLY;
1104 	rply.rm_reply.rp_stat = MSG_ACCEPTED;
1105 	rply.acpted_rply.ar_verf = xprt->xp_verf;
1106 	rply.acpted_rply.ar_stat = SUCCESS;
1107 	rply.acpted_rply.ar_results.where = xdr_location;
1108 	rply.acpted_rply.ar_results.proc = xdr_results;
1109 	return (SVC_REPLY((SVCXPRT *)xprt, &rply));
1110 }
1111 
1112 /*
1113  * No procedure error reply
1114  */
1115 void
svcerr_noproc(const SVCXPRT * xprt)1116 svcerr_noproc(const SVCXPRT *xprt)
1117 {
1118 	struct rpc_msg rply;
1119 
1120 	rply.rm_direction = REPLY;
1121 	rply.rm_reply.rp_stat = MSG_ACCEPTED;
1122 	rply.acpted_rply.ar_verf = xprt->xp_verf;
1123 	rply.acpted_rply.ar_stat = PROC_UNAVAIL;
1124 	SVC_REPLY((SVCXPRT *)xprt, &rply);
1125 }
1126 
1127 /*
1128  * Can't decode args error reply
1129  */
1130 void
svcerr_decode(const SVCXPRT * xprt)1131 svcerr_decode(const SVCXPRT *xprt)
1132 {
1133 	struct rpc_msg rply;
1134 
1135 	rply.rm_direction = REPLY;
1136 	rply.rm_reply.rp_stat = MSG_ACCEPTED;
1137 	rply.acpted_rply.ar_verf = xprt->xp_verf;
1138 	rply.acpted_rply.ar_stat = GARBAGE_ARGS;
1139 	SVC_REPLY((SVCXPRT *)xprt, &rply);
1140 }
1141 
1142 /*
1143  * Some system error
1144  */
1145 void
svcerr_systemerr(const SVCXPRT * xprt)1146 svcerr_systemerr(const SVCXPRT *xprt)
1147 {
1148 	struct rpc_msg rply;
1149 
1150 	rply.rm_direction = REPLY;
1151 	rply.rm_reply.rp_stat = MSG_ACCEPTED;
1152 	rply.acpted_rply.ar_verf = xprt->xp_verf;
1153 	rply.acpted_rply.ar_stat = SYSTEM_ERR;
1154 	SVC_REPLY((SVCXPRT *)xprt, &rply);
1155 }
1156 
1157 /*
1158  * Tell RPC package to not complain about version errors to the client.	 This
1159  * is useful when revving broadcast protocols that sit on a fixed address.
1160  * There is really one (or should be only one) example of this kind of
1161  * protocol: the portmapper (or rpc binder).
1162  */
1163 void
__svc_versquiet_on(const SVCXPRT * xprt)1164 __svc_versquiet_on(const SVCXPRT *xprt)
1165 {
1166 /* LINTED pointer alignment */
1167 	svc_flags(xprt) |= SVC_VERSQUIET;
1168 }
1169 
1170 void
__svc_versquiet_off(const SVCXPRT * xprt)1171 __svc_versquiet_off(const SVCXPRT *xprt)
1172 {
1173 /* LINTED pointer alignment */
1174 	svc_flags(xprt) &= ~SVC_VERSQUIET;
1175 }
1176 
1177 void
svc_versquiet(const SVCXPRT * xprt)1178 svc_versquiet(const SVCXPRT *xprt)
1179 {
1180 	__svc_versquiet_on(xprt);
1181 }
1182 
1183 int
__svc_versquiet_get(const SVCXPRT * xprt)1184 __svc_versquiet_get(const SVCXPRT *xprt)
1185 {
1186 /* LINTED pointer alignment */
1187 	return (svc_flags(xprt) & SVC_VERSQUIET);
1188 }
1189 
1190 /*
1191  * Authentication error reply
1192  */
1193 void
svcerr_auth(const SVCXPRT * xprt,const enum auth_stat why)1194 svcerr_auth(const SVCXPRT *xprt, const enum auth_stat why)
1195 {
1196 	struct rpc_msg rply;
1197 
1198 	rply.rm_direction = REPLY;
1199 	rply.rm_reply.rp_stat = MSG_DENIED;
1200 	rply.rjcted_rply.rj_stat = AUTH_ERROR;
1201 	rply.rjcted_rply.rj_why = why;
1202 	SVC_REPLY((SVCXPRT *)xprt, &rply);
1203 }
1204 
1205 /*
1206  * Auth too weak error reply
1207  */
1208 void
svcerr_weakauth(const SVCXPRT * xprt)1209 svcerr_weakauth(const SVCXPRT *xprt)
1210 {
1211 	svcerr_auth(xprt, AUTH_TOOWEAK);
1212 }
1213 
1214 /*
1215  * Program unavailable error reply
1216  */
1217 void
svcerr_noprog(const SVCXPRT * xprt)1218 svcerr_noprog(const SVCXPRT *xprt)
1219 {
1220 	struct rpc_msg rply;
1221 
1222 	rply.rm_direction = REPLY;
1223 	rply.rm_reply.rp_stat = MSG_ACCEPTED;
1224 	rply.acpted_rply.ar_verf = xprt->xp_verf;
1225 	rply.acpted_rply.ar_stat = PROG_UNAVAIL;
1226 	SVC_REPLY((SVCXPRT *)xprt, &rply);
1227 }
1228 
1229 /*
1230  * Program version mismatch error reply
1231  */
1232 void
svcerr_progvers(const SVCXPRT * xprt,const rpcvers_t low_vers,const rpcvers_t high_vers)1233 svcerr_progvers(const SVCXPRT *xprt, const rpcvers_t low_vers,
1234     const rpcvers_t high_vers)
1235 {
1236 	struct rpc_msg rply;
1237 
1238 	rply.rm_direction = REPLY;
1239 	rply.rm_reply.rp_stat = MSG_ACCEPTED;
1240 	rply.acpted_rply.ar_verf = xprt->xp_verf;
1241 	rply.acpted_rply.ar_stat = PROG_MISMATCH;
1242 	rply.acpted_rply.ar_vers.low = low_vers;
1243 	rply.acpted_rply.ar_vers.high = high_vers;
1244 	SVC_REPLY((SVCXPRT *)xprt, &rply);
1245 }
1246 
1247 /* ******************* SERVER INPUT STUFF ******************* */
1248 
1249 /*
1250  * Get server side input from some transport.
1251  *
1252  * Statement of authentication parameters management:
1253  * This function owns and manages all authentication parameters, specifically
1254  * the "raw" parameters (msg.rm_call.cb_cred and msg.rm_call.cb_verf) and
1255  * the "cooked" credentials (rqst->rq_clntcred).
1256  * However, this function does not know the structure of the cooked
1257  * credentials, so it make the following assumptions:
1258  *   a) the structure is contiguous (no pointers), and
1259  *   b) the cred structure size does not exceed RQCRED_SIZE bytes.
1260  * In all events, all three parameters are freed upon exit from this routine.
1261  * The storage is trivially management on the call stack in user land, but
1262  * is mallocated in kernel land.
1263  */
1264 
1265 void
svc_getreq(int rdfds)1266 svc_getreq(int rdfds)
1267 {
1268 	fd_set readfds;
1269 
1270 	FD_ZERO(&readfds);
1271 	readfds.fds_bits[0] = rdfds;
1272 	svc_getreqset(&readfds);
1273 }
1274 
1275 void
svc_getreqset(fd_set * readfds)1276 svc_getreqset(fd_set *readfds)
1277 {
1278 	int i;
1279 
1280 	for (i = 0; i < svc_max_fd; i++) {
1281 		/* fd has input waiting */
1282 		if (FD_ISSET(i, readfds))
1283 			svc_getreq_common(i);
1284 	}
1285 }
1286 
1287 void
svc_getreq_poll(struct pollfd * pfdp,const int pollretval)1288 svc_getreq_poll(struct pollfd *pfdp, const int pollretval)
1289 {
1290 	int i;
1291 	int fds_found;
1292 
1293 	for (i = fds_found = 0; fds_found < pollretval; i++) {
1294 		struct pollfd *p = &pfdp[i];
1295 
1296 		if (p->revents) {
1297 			/* fd has input waiting */
1298 			fds_found++;
1299 			/*
1300 			 *	We assume that this function is only called
1301 			 *	via someone select()ing from svc_fdset or
1302 			 *	poll()ing from svc_pollset[].  Thus it's safe
1303 			 *	to handle the POLLNVAL event by simply turning
1304 			 *	the corresponding bit off in svc_fdset.  The
1305 			 *	svc_pollset[] array is derived from svc_fdset
1306 			 *	and so will also be updated eventually.
1307 			 *
1308 			 *	XXX Should we do an xprt_unregister() instead?
1309 			 */
1310 			/* Handle user callback */
1311 			if (__is_a_userfd(p->fd) == TRUE) {
1312 				(void) rw_rdlock(&svc_fd_lock);
1313 				__svc_getreq_user(p);
1314 				(void) rw_unlock(&svc_fd_lock);
1315 			} else {
1316 				if (p->revents & POLLNVAL) {
1317 					(void) rw_wrlock(&svc_fd_lock);
1318 					remove_pollfd(p->fd);	/* XXX */
1319 					(void) rw_unlock(&svc_fd_lock);
1320 				} else {
1321 					svc_getreq_common(p->fd);
1322 				}
1323 			}
1324 		}
1325 	}
1326 }
1327 
1328 void
svc_getreq_common(const int fd)1329 svc_getreq_common(const int fd)
1330 {
1331 	SVCXPRT *xprt;
1332 	enum xprt_stat stat;
1333 	struct rpc_msg *msg;
1334 	struct svc_req *r;
1335 	char *cred_area;
1336 
1337 	(void) rw_rdlock(&svc_fd_lock);
1338 
1339 	/* HANDLE USER CALLBACK */
1340 	if (__is_a_userfd(fd) == TRUE) {
1341 		struct pollfd virtual_fd;
1342 
1343 		virtual_fd.events = virtual_fd.revents = (short)0xFFFF;
1344 		virtual_fd.fd = fd;
1345 		__svc_getreq_user(&virtual_fd);
1346 		(void) rw_unlock(&svc_fd_lock);
1347 		return;
1348 	}
1349 
1350 	/*
1351 	 * The transport associated with this fd could have been
1352 	 * removed from svc_timeout_nonblock_xprt_and_LRU, for instance.
1353 	 * This can happen if two or more fds get read events and are
1354 	 * passed to svc_getreq_poll/set, the first fd is seviced by
1355 	 * the dispatch routine and cleans up any dead transports.  If
1356 	 * one of the dead transports removed is the other fd that
1357 	 * had a read event then svc_getreq_common() will be called with no
1358 	 * xprt associated with the fd that had the original read event.
1359 	 */
1360 	if ((fd >= nsvc_xports) || (xprt = svc_xports[fd]) == NULL) {
1361 		(void) rw_unlock(&svc_fd_lock);
1362 		return;
1363 	}
1364 	(void) rw_unlock(&svc_fd_lock);
1365 /* LINTED pointer alignment */
1366 	msg = SVCEXT(xprt)->msg;
1367 /* LINTED pointer alignment */
1368 	r = SVCEXT(xprt)->req;
1369 /* LINTED pointer alignment */
1370 	cred_area = SVCEXT(xprt)->cred_area;
1371 	msg->rm_call.cb_cred.oa_base = cred_area;
1372 	msg->rm_call.cb_verf.oa_base = &(cred_area[MAX_AUTH_BYTES]);
1373 	r->rq_clntcred = &(cred_area[2 * MAX_AUTH_BYTES]);
1374 
1375 	/* receive msgs from xprtprt (support batch calls) */
1376 	do {
1377 		bool_t dispatch;
1378 
1379 		if (dispatch = SVC_RECV(xprt, msg))
1380 			(void) _svc_prog_dispatch(xprt, msg, r);
1381 		/*
1382 		 * Check if the xprt has been disconnected in a recursive call
1383 		 * in the service dispatch routine. If so, then break
1384 		 */
1385 		(void) rw_rdlock(&svc_fd_lock);
1386 		if (xprt != svc_xports[fd]) {
1387 			(void) rw_unlock(&svc_fd_lock);
1388 			break;
1389 		}
1390 		(void) rw_unlock(&svc_fd_lock);
1391 
1392 		/*
1393 		 * Call cleanup procedure if set.
1394 		 */
1395 		if (__proc_cleanup_cb != NULL && dispatch)
1396 			(*__proc_cleanup_cb)(xprt);
1397 
1398 		if ((stat = SVC_STAT(xprt)) == XPRT_DIED) {
1399 			SVC_DESTROY(xprt);
1400 			break;
1401 		}
1402 	} while (stat == XPRT_MOREREQS);
1403 }
1404 
1405 int
_svc_prog_dispatch(SVCXPRT * xprt,struct rpc_msg * msg,struct svc_req * r)1406 _svc_prog_dispatch(SVCXPRT *xprt, struct rpc_msg *msg, struct svc_req *r)
1407 {
1408 	struct svc_callout *s;
1409 	enum auth_stat why;
1410 	int prog_found;
1411 	rpcvers_t low_vers;
1412 	rpcvers_t high_vers;
1413 	void (*disp_fn)();
1414 
1415 	r->rq_xprt = xprt;
1416 	r->rq_prog = msg->rm_call.cb_prog;
1417 	r->rq_vers = msg->rm_call.cb_vers;
1418 	r->rq_proc = msg->rm_call.cb_proc;
1419 	r->rq_cred = msg->rm_call.cb_cred;
1420 /* LINTED pointer alignment */
1421 	SVC_XP_AUTH(r->rq_xprt).svc_ah_ops = svc_auth_any_ops;
1422 /* LINTED pointer alignment */
1423 	SVC_XP_AUTH(r->rq_xprt).svc_ah_private = NULL;
1424 
1425 	/* first authenticate the message */
1426 	/* Check for null flavor and bypass these calls if possible */
1427 
1428 	if (msg->rm_call.cb_cred.oa_flavor == AUTH_NULL) {
1429 		r->rq_xprt->xp_verf.oa_flavor = _null_auth.oa_flavor;
1430 		r->rq_xprt->xp_verf.oa_length = 0;
1431 	} else {
1432 		bool_t no_dispatch;
1433 
1434 		if ((why = __gss_authenticate(r, msg,
1435 		    &no_dispatch)) != AUTH_OK) {
1436 			svcerr_auth(xprt, why);
1437 			return (0);
1438 		}
1439 		if (no_dispatch)
1440 			return (0);
1441 	}
1442 	/* match message with a registered service */
1443 	prog_found = FALSE;
1444 	low_vers = (rpcvers_t)(0 - 1);
1445 	high_vers = 0;
1446 	(void) rw_rdlock(&svc_lock);
1447 	for (s = svc_head; s != NULL_SVC; s = s->sc_next) {
1448 		if (s->sc_prog == r->rq_prog) {
1449 			prog_found = TRUE;
1450 			if (s->sc_vers == r->rq_vers) {
1451 				if ((xprt->xp_netid == NULL) ||
1452 				    (s->sc_netid == NULL) ||
1453 				    (strcmp(xprt->xp_netid,
1454 				    s->sc_netid) == 0)) {
1455 					disp_fn = (*s->sc_dispatch);
1456 					(void) rw_unlock(&svc_lock);
1457 					disp_fn(r, xprt);
1458 					return (1);
1459 				}
1460 				prog_found = FALSE;
1461 			}
1462 			if (s->sc_vers < low_vers)
1463 				low_vers = s->sc_vers;
1464 			if (s->sc_vers > high_vers)
1465 				high_vers = s->sc_vers;
1466 		}		/* found correct program */
1467 	}
1468 	(void) rw_unlock(&svc_lock);
1469 
1470 	/*
1471 	 * if we got here, the program or version
1472 	 * is not served ...
1473 	 */
1474 	if (prog_found) {
1475 /* LINTED pointer alignment */
1476 		if (!version_keepquiet(xprt))
1477 			svcerr_progvers(xprt, low_vers, high_vers);
1478 	} else {
1479 		svcerr_noprog(xprt);
1480 	}
1481 	return (0);
1482 }
1483 
1484 /* ******************* SVCXPRT allocation and deallocation ***************** */
1485 
1486 /*
1487  * svc_xprt_alloc() - allocate a service transport handle
1488  */
1489 SVCXPRT *
svc_xprt_alloc(void)1490 svc_xprt_alloc(void)
1491 {
1492 	SVCXPRT		*xprt = NULL;
1493 	SVCXPRT_EXT	*xt = NULL;
1494 	SVCXPRT_LIST	*xlist = NULL;
1495 	struct rpc_msg	*msg = NULL;
1496 	struct svc_req	*req = NULL;
1497 	char		*cred_area = NULL;
1498 
1499 	if ((xprt = calloc(1, sizeof (SVCXPRT))) == NULL)
1500 		goto err_exit;
1501 
1502 	if ((xt = calloc(1, sizeof (SVCXPRT_EXT))) == NULL)
1503 		goto err_exit;
1504 	xprt->xp_p3 = (caddr_t)xt; /* SVCEXT(xprt) = xt */
1505 
1506 	if ((xlist = calloc(1, sizeof (SVCXPRT_LIST))) == NULL)
1507 		goto err_exit;
1508 	xt->my_xlist = xlist;
1509 	xlist->xprt = xprt;
1510 
1511 	if ((msg = malloc(sizeof (struct rpc_msg))) == NULL)
1512 		goto err_exit;
1513 	xt->msg = msg;
1514 
1515 	if ((req = malloc(sizeof (struct svc_req))) == NULL)
1516 		goto err_exit;
1517 	xt->req = req;
1518 
1519 	if ((cred_area = malloc(2*MAX_AUTH_BYTES + RQCRED_SIZE)) == NULL)
1520 		goto err_exit;
1521 	xt->cred_area = cred_area;
1522 
1523 /* LINTED pointer alignment */
1524 	(void) mutex_init(&svc_send_mutex(xprt), USYNC_THREAD, (void *)0);
1525 	return (xprt);
1526 
1527 err_exit:
1528 	svc_xprt_free(xprt);
1529 	return (NULL);
1530 }
1531 
1532 
1533 /*
1534  * svc_xprt_free() - free a service handle
1535  */
1536 void
svc_xprt_free(SVCXPRT * xprt)1537 svc_xprt_free(SVCXPRT *xprt)
1538 {
1539 /* LINTED pointer alignment */
1540 	SVCXPRT_EXT	*xt = xprt ? SVCEXT(xprt) : NULL;
1541 	SVCXPRT_LIST	*my_xlist = xt ? xt->my_xlist: NULL;
1542 	struct rpc_msg	*msg = xt ? xt->msg : NULL;
1543 	struct svc_req	*req = xt ? xt->req : NULL;
1544 	char		*cred_area = xt ? xt->cred_area : NULL;
1545 
1546 	if (xprt)
1547 		free(xprt);
1548 	if (xt)
1549 		free(xt);
1550 	if (my_xlist)
1551 		free(my_xlist);
1552 	if (msg)
1553 		free(msg);
1554 	if (req)
1555 		free(req);
1556 	if (cred_area)
1557 		free(cred_area);
1558 }
1559 
1560 
1561 /*
1562  * svc_xprt_destroy() - free parent and child xprt list
1563  */
1564 void
svc_xprt_destroy(SVCXPRT * xprt)1565 svc_xprt_destroy(SVCXPRT *xprt)
1566 {
1567 	SVCXPRT_LIST	*xlist, *xnext = NULL;
1568 	int		type;
1569 
1570 /* LINTED pointer alignment */
1571 	if (SVCEXT(xprt)->parent)
1572 /* LINTED pointer alignment */
1573 		xprt = SVCEXT(xprt)->parent;
1574 /* LINTED pointer alignment */
1575 	type = svc_type(xprt);
1576 /* LINTED pointer alignment */
1577 	for (xlist = SVCEXT(xprt)->my_xlist; xlist != NULL; xlist = xnext) {
1578 		xnext = xlist->next;
1579 		xprt = xlist->xprt;
1580 		switch (type) {
1581 		case SVC_DGRAM:
1582 			svc_dg_xprtfree(xprt);
1583 			break;
1584 		case SVC_RENDEZVOUS:
1585 			svc_vc_xprtfree(xprt);
1586 			break;
1587 		case SVC_CONNECTION:
1588 			svc_fd_xprtfree(xprt);
1589 			break;
1590 		case SVC_DOOR:
1591 			svc_door_xprtfree(xprt);
1592 			break;
1593 		}
1594 	}
1595 }
1596 
1597 
1598 /*
1599  * svc_copy() - make a copy of parent
1600  */
1601 SVCXPRT *
svc_copy(SVCXPRT * xprt)1602 svc_copy(SVCXPRT *xprt)
1603 {
1604 /* LINTED pointer alignment */
1605 	switch (svc_type(xprt)) {
1606 	case SVC_DGRAM:
1607 		return (svc_dg_xprtcopy(xprt));
1608 	case SVC_RENDEZVOUS:
1609 		return (svc_vc_xprtcopy(xprt));
1610 	case SVC_CONNECTION:
1611 		return (svc_fd_xprtcopy(xprt));
1612 	}
1613 	return (NULL);
1614 }
1615 
1616 
1617 /*
1618  * _svc_destroy_private() - private SVC_DESTROY interface
1619  */
1620 void
_svc_destroy_private(SVCXPRT * xprt)1621 _svc_destroy_private(SVCXPRT *xprt)
1622 {
1623 /* LINTED pointer alignment */
1624 	switch (svc_type(xprt)) {
1625 	case SVC_DGRAM:
1626 		_svc_dg_destroy_private(xprt);
1627 		break;
1628 	case SVC_RENDEZVOUS:
1629 	case SVC_CONNECTION:
1630 		_svc_vc_destroy_private(xprt, TRUE);
1631 		break;
1632 	}
1633 }
1634 
1635 /*
1636  * svc_get_local_cred() - fetch local user credentials.  This always
1637  * works over doors based transports.  For local transports, this
1638  * does not yield correct results unless the __rpc_negotiate_uid()
1639  * call has been invoked to enable this feature.
1640  */
1641 bool_t
svc_get_local_cred(SVCXPRT * xprt,svc_local_cred_t * lcred)1642 svc_get_local_cred(SVCXPRT *xprt, svc_local_cred_t *lcred)
1643 {
1644 	/* LINTED pointer alignment */
1645 	if (svc_type(xprt) == SVC_DOOR)
1646 		return (__svc_get_door_cred(xprt, lcred));
1647 	return (__rpc_get_local_cred(xprt, lcred));
1648 }
1649 
1650 
1651 /* ******************* DUPLICATE ENTRY HANDLING ROUTINES ************** */
1652 
1653 /*
1654  * the dup cacheing routines below provide a cache of received
1655  * transactions. rpc service routines can use this to detect
1656  * retransmissions and re-send a non-failure response. Uses a
1657  * lru scheme to find entries to get rid of entries in the cache,
1658  * though only DUP_DONE entries are placed on the lru list.
1659  * the routines were written towards development of a generic
1660  * SVC_DUP() interface, which can be expanded to encompass the
1661  * svc_dg_enablecache() routines as well. the cache is currently
1662  * private to the automounter.
1663  */
1664 
1665 
1666 /* dupcache header contains xprt specific information */
1667 struct dupcache {
1668 	rwlock_t	dc_lock;
1669 	time_t		dc_time;
1670 	int		dc_buckets;
1671 	int		dc_maxsz;
1672 	int		dc_basis;
1673 	struct dupreq 	*dc_mru;
1674 	struct dupreq	**dc_hashtbl;
1675 };
1676 
1677 /*
1678  * private duplicate cache request routines
1679  */
1680 static int __svc_dupcache_check(struct svc_req *, caddr_t *, uint_t *,
1681 		struct dupcache *, uint32_t, uint32_t);
1682 static struct dupreq *__svc_dupcache_victim(struct dupcache *, time_t);
1683 static int __svc_dupcache_enter(struct svc_req *, struct dupreq *,
1684 		struct dupcache *, uint32_t, uint32_t, time_t);
1685 static int __svc_dupcache_update(struct svc_req *, caddr_t, uint_t, int,
1686 		struct dupcache *, uint32_t, uint32_t);
1687 #ifdef DUP_DEBUG
1688 static void __svc_dupcache_debug(struct dupcache *);
1689 #endif /* DUP_DEBUG */
1690 
1691 /* default parameters for the dupcache */
1692 #define	DUPCACHE_BUCKETS	257
1693 #define	DUPCACHE_TIME		900
1694 #define	DUPCACHE_MAXSZ		INT_MAX
1695 
1696 /*
1697  * __svc_dupcache_init(void *condition, int basis, char *xprt_cache)
1698  * initialize the duprequest cache and assign it to the xprt_cache
1699  * Use default values depending on the cache condition and basis.
1700  * return TRUE on success and FALSE on failure
1701  */
1702 bool_t
__svc_dupcache_init(void * condition,int basis,char ** xprt_cache)1703 __svc_dupcache_init(void *condition, int basis, char **xprt_cache)
1704 {
1705 	static mutex_t initdc_lock = DEFAULTMUTEX;
1706 	int i;
1707 	struct dupcache *dc;
1708 
1709 	(void) mutex_lock(&initdc_lock);
1710 	if (*xprt_cache != NULL) { /* do only once per xprt */
1711 		(void) mutex_unlock(&initdc_lock);
1712 		syslog(LOG_ERR,
1713 		    "__svc_dupcache_init: multiply defined dup cache");
1714 		return (FALSE);
1715 	}
1716 
1717 	switch (basis) {
1718 	case DUPCACHE_FIXEDTIME:
1719 		dc = malloc(sizeof (struct dupcache));
1720 		if (dc == NULL) {
1721 			(void) mutex_unlock(&initdc_lock);
1722 			syslog(LOG_ERR,
1723 			    "__svc_dupcache_init: memory alloc failed");
1724 			return (FALSE);
1725 		}
1726 		(void) rwlock_init(&(dc->dc_lock), USYNC_THREAD, NULL);
1727 		if (condition != NULL)
1728 			dc->dc_time = *((time_t *)condition);
1729 		else
1730 			dc->dc_time = DUPCACHE_TIME;
1731 		dc->dc_buckets = DUPCACHE_BUCKETS;
1732 		dc->dc_maxsz = DUPCACHE_MAXSZ;
1733 		dc->dc_basis = basis;
1734 		dc->dc_mru = NULL;
1735 		dc->dc_hashtbl = malloc(dc->dc_buckets *
1736 		    sizeof (struct dupreq *));
1737 		if (dc->dc_hashtbl == NULL) {
1738 			free(dc);
1739 			(void) mutex_unlock(&initdc_lock);
1740 			syslog(LOG_ERR,
1741 			    "__svc_dupcache_init: memory alloc failed");
1742 			return (FALSE);
1743 		}
1744 		for (i = 0; i < DUPCACHE_BUCKETS; i++)
1745 			dc->dc_hashtbl[i] = NULL;
1746 		*xprt_cache = (char *)dc;
1747 		break;
1748 	default:
1749 		(void) mutex_unlock(&initdc_lock);
1750 		syslog(LOG_ERR,
1751 		    "__svc_dupcache_init: undefined dup cache basis");
1752 		return (FALSE);
1753 	}
1754 
1755 	(void) mutex_unlock(&initdc_lock);
1756 
1757 	return (TRUE);
1758 }
1759 
1760 /*
1761  * __svc_dup(struct svc_req *req, caddr_t *resp_buf, uint_t *resp_bufsz,
1762  *	char *xprt_cache)
1763  * searches the request cache. Creates an entry and returns DUP_NEW if
1764  * the request is not found in the cache.  If it is found, then it
1765  * returns the state of the request (in progress, drop, or done) and
1766  * also allocates, and passes back results to the user (if any) in
1767  * resp_buf, and its length in resp_bufsz. DUP_ERROR is returned on error.
1768  */
1769 int
__svc_dup(struct svc_req * req,caddr_t * resp_buf,uint_t * resp_bufsz,char * xprt_cache)1770 __svc_dup(struct svc_req *req, caddr_t *resp_buf, uint_t *resp_bufsz,
1771     char *xprt_cache)
1772 {
1773 	uint32_t drxid, drhash;
1774 	int rc;
1775 	struct dupreq *dr = NULL;
1776 	time_t timenow = time(NULL);
1777 
1778 	/* LINTED pointer alignment */
1779 	struct dupcache *dc = (struct dupcache *)xprt_cache;
1780 
1781 	if (dc == NULL) {
1782 		syslog(LOG_ERR, "__svc_dup: undefined cache");
1783 		return (DUP_ERROR);
1784 	}
1785 
1786 	/* get the xid of the request */
1787 	if (SVC_CONTROL(req->rq_xprt, SVCGET_XID, (void*)&drxid) == FALSE) {
1788 		syslog(LOG_ERR, "__svc_dup: xid error");
1789 		return (DUP_ERROR);
1790 	}
1791 	drhash = drxid % dc->dc_buckets;
1792 
1793 	if ((rc = __svc_dupcache_check(req, resp_buf, resp_bufsz, dc, drxid,
1794 	    drhash)) != DUP_NEW)
1795 		return (rc);
1796 
1797 	if ((dr = __svc_dupcache_victim(dc, timenow)) == NULL)
1798 		return (DUP_ERROR);
1799 
1800 	if ((rc = __svc_dupcache_enter(req, dr, dc, drxid, drhash, timenow))
1801 	    == DUP_ERROR)
1802 		return (rc);
1803 
1804 	return (DUP_NEW);
1805 }
1806 
1807 
1808 
1809 /*
1810  * __svc_dupcache_check(struct svc_req *req, caddr_t *resp_buf,
1811  *		uint_t *resp_bufsz,truct dupcache *dc, uint32_t drxid,
1812  * 		uint32_t drhash)
1813  * Checks to see whether an entry already exists in the cache. If it does
1814  * copy back into the resp_buf, if appropriate. Return the status of
1815  * the request, or DUP_NEW if the entry is not in the cache
1816  */
1817 static int
__svc_dupcache_check(struct svc_req * req,caddr_t * resp_buf,uint_t * resp_bufsz,struct dupcache * dc,uint32_t drxid,uint32_t drhash)1818 __svc_dupcache_check(struct svc_req *req, caddr_t *resp_buf, uint_t *resp_bufsz,
1819     struct dupcache *dc, uint32_t drxid, uint32_t drhash)
1820 {
1821 	struct dupreq *dr = NULL;
1822 
1823 	(void) rw_rdlock(&(dc->dc_lock));
1824 	dr = dc->dc_hashtbl[drhash];
1825 	while (dr != NULL) {
1826 		if (dr->dr_xid == drxid &&
1827 		    dr->dr_proc == req->rq_proc &&
1828 		    dr->dr_prog == req->rq_prog &&
1829 		    dr->dr_vers == req->rq_vers &&
1830 		    dr->dr_addr.len == req->rq_xprt->xp_rtaddr.len &&
1831 		    memcmp(dr->dr_addr.buf, req->rq_xprt->xp_rtaddr.buf,
1832 		    dr->dr_addr.len) == 0) { /* entry found */
1833 			if (dr->dr_hash != drhash) {
1834 				/* sanity check */
1835 				(void) rw_unlock((&dc->dc_lock));
1836 				syslog(LOG_ERR,
1837 				    "\n__svc_dupdone: hashing error");
1838 				return (DUP_ERROR);
1839 			}
1840 
1841 			/*
1842 			 * return results for requests on lru list, if
1843 			 * appropriate requests must be DUP_DROP or DUP_DONE
1844 			 * to have a result. A NULL buffer in the cache
1845 			 * implies no results were sent during dupdone.
1846 			 * A NULL buffer in the call implies not interested
1847 			 * in results.
1848 			 */
1849 			if (((dr->dr_status == DUP_DONE) ||
1850 			    (dr->dr_status == DUP_DROP)) &&
1851 			    resp_buf != NULL &&
1852 			    dr->dr_resp.buf != NULL) {
1853 				*resp_buf = malloc(dr->dr_resp.len);
1854 				if (*resp_buf == NULL) {
1855 					syslog(LOG_ERR,
1856 					"__svc_dupcache_check: malloc failed");
1857 					(void) rw_unlock(&(dc->dc_lock));
1858 					return (DUP_ERROR);
1859 				}
1860 				(void) memset(*resp_buf, 0, dr->dr_resp.len);
1861 				(void) memcpy(*resp_buf, dr->dr_resp.buf,
1862 				    dr->dr_resp.len);
1863 				*resp_bufsz = dr->dr_resp.len;
1864 			} else {
1865 				/* no result */
1866 				if (resp_buf)
1867 					*resp_buf = NULL;
1868 				if (resp_bufsz)
1869 					*resp_bufsz = 0;
1870 			}
1871 			(void) rw_unlock(&(dc->dc_lock));
1872 			return (dr->dr_status);
1873 		}
1874 		dr = dr->dr_chain;
1875 	}
1876 	(void) rw_unlock(&(dc->dc_lock));
1877 	return (DUP_NEW);
1878 }
1879 
1880 /*
1881  * __svc_dupcache_victim(struct dupcache *dc, time_t timenow)
1882  * Return a victim dupreq entry to the caller, depending on cache policy.
1883  */
1884 static struct dupreq *
__svc_dupcache_victim(struct dupcache * dc,time_t timenow)1885 __svc_dupcache_victim(struct dupcache *dc, time_t timenow)
1886 {
1887 	struct dupreq *dr = NULL;
1888 
1889 	switch (dc->dc_basis) {
1890 	case DUPCACHE_FIXEDTIME:
1891 		/*
1892 		 * The hash policy is to free up a bit of the hash
1893 		 * table before allocating a new entry as the victim.
1894 		 * Freeing up the hash table each time should split
1895 		 * the cost of keeping the hash table clean among threads.
1896 		 * Note that only DONE or DROPPED entries are on the lru
1897 		 * list but we do a sanity check anyway.
1898 		 */
1899 		(void) rw_wrlock(&(dc->dc_lock));
1900 		while ((dc->dc_mru) && (dr = dc->dc_mru->dr_next) &&
1901 		    ((timenow - dr->dr_time) > dc->dc_time)) {
1902 			/* clean and then free the entry */
1903 			if (dr->dr_status != DUP_DONE &&
1904 			    dr->dr_status != DUP_DROP) {
1905 				/*
1906 				 * The LRU list can't contain an
1907 				 * entry where the status is other than
1908 				 * DUP_DONE or DUP_DROP.
1909 				 */
1910 				syslog(LOG_ERR,
1911 				    "__svc_dupcache_victim: bad victim");
1912 #ifdef DUP_DEBUG
1913 				/*
1914 				 * Need to hold the reader/writers lock to
1915 				 * print the cache info, since we already
1916 				 * hold the writers lock, we shall continue
1917 				 * calling __svc_dupcache_debug()
1918 				 */
1919 				__svc_dupcache_debug(dc);
1920 #endif /* DUP_DEBUG */
1921 				(void) rw_unlock(&(dc->dc_lock));
1922 				return (NULL);
1923 			}
1924 			/* free buffers */
1925 			if (dr->dr_resp.buf) {
1926 				free(dr->dr_resp.buf);
1927 				dr->dr_resp.buf = NULL;
1928 			}
1929 			if (dr->dr_addr.buf) {
1930 				free(dr->dr_addr.buf);
1931 				dr->dr_addr.buf = NULL;
1932 			}
1933 
1934 			/* unhash the entry */
1935 			if (dr->dr_chain)
1936 				dr->dr_chain->dr_prevchain = dr->dr_prevchain;
1937 			if (dr->dr_prevchain)
1938 				dr->dr_prevchain->dr_chain = dr->dr_chain;
1939 			if (dc->dc_hashtbl[dr->dr_hash] == dr)
1940 				dc->dc_hashtbl[dr->dr_hash] = dr->dr_chain;
1941 
1942 			/* modify the lru pointers */
1943 			if (dc->dc_mru == dr) {
1944 				dc->dc_mru = NULL;
1945 			} else {
1946 				dc->dc_mru->dr_next = dr->dr_next;
1947 				dr->dr_next->dr_prev = dc->dc_mru;
1948 			}
1949 			free(dr);
1950 			dr = NULL;
1951 		}
1952 		(void) rw_unlock(&(dc->dc_lock));
1953 
1954 		/*
1955 		 * Allocate and return new clean entry as victim
1956 		 */
1957 		if ((dr = malloc(sizeof (*dr))) == NULL) {
1958 			syslog(LOG_ERR,
1959 			    "__svc_dupcache_victim: malloc failed");
1960 			return (NULL);
1961 		}
1962 		(void) memset(dr, 0, sizeof (*dr));
1963 		return (dr);
1964 	default:
1965 		syslog(LOG_ERR,
1966 		    "__svc_dupcache_victim: undefined dup cache_basis");
1967 		return (NULL);
1968 	}
1969 }
1970 
1971 /*
1972  * __svc_dupcache_enter(struct svc_req *req, struct dupreq *dr,
1973  *	struct dupcache *dc, uint32_t drxid, uint32_t drhash, time_t timenow)
1974  * build new duprequest entry and then insert into the cache
1975  */
1976 static int
__svc_dupcache_enter(struct svc_req * req,struct dupreq * dr,struct dupcache * dc,uint32_t drxid,uint32_t drhash,time_t timenow)1977 __svc_dupcache_enter(struct svc_req *req, struct dupreq *dr,
1978     struct dupcache *dc, uint32_t drxid, uint32_t drhash, time_t timenow)
1979 {
1980 	dr->dr_xid = drxid;
1981 	dr->dr_prog = req->rq_prog;
1982 	dr->dr_vers = req->rq_vers;
1983 	dr->dr_proc = req->rq_proc;
1984 	dr->dr_addr.maxlen = req->rq_xprt->xp_rtaddr.len;
1985 	dr->dr_addr.len = dr->dr_addr.maxlen;
1986 	if ((dr->dr_addr.buf = malloc(dr->dr_addr.maxlen)) == NULL) {
1987 		syslog(LOG_ERR, "__svc_dupcache_enter: malloc failed");
1988 		free(dr);
1989 		return (DUP_ERROR);
1990 	}
1991 	(void) memset(dr->dr_addr.buf, 0, dr->dr_addr.len);
1992 	(void) memcpy(dr->dr_addr.buf, req->rq_xprt->xp_rtaddr.buf,
1993 	    dr->dr_addr.len);
1994 	dr->dr_resp.buf = NULL;
1995 	dr->dr_resp.maxlen = 0;
1996 	dr->dr_resp.len = 0;
1997 	dr->dr_status = DUP_INPROGRESS;
1998 	dr->dr_time = timenow;
1999 	dr->dr_hash = drhash;	/* needed for efficient victim cleanup */
2000 
2001 	/* place entry at head of hash table */
2002 	(void) rw_wrlock(&(dc->dc_lock));
2003 	dr->dr_chain = dc->dc_hashtbl[drhash];
2004 	dr->dr_prevchain = NULL;
2005 	if (dc->dc_hashtbl[drhash] != NULL)
2006 		dc->dc_hashtbl[drhash]->dr_prevchain = dr;
2007 	dc->dc_hashtbl[drhash] = dr;
2008 	(void) rw_unlock(&(dc->dc_lock));
2009 	return (DUP_NEW);
2010 }
2011 
2012 /*
2013  * __svc_dupdone(struct svc_req *req, caddr_t resp_buf, uint_t resp_bufsz,
2014  *		int status, char *xprt_cache)
2015  * Marks the request done (DUP_DONE or DUP_DROP) and stores the response.
2016  * Only DONE and DROP requests can be marked as done. Sets the lru pointers
2017  * to make the entry the most recently used. Returns DUP_ERROR or status.
2018  */
2019 int
__svc_dupdone(struct svc_req * req,caddr_t resp_buf,uint_t resp_bufsz,int status,char * xprt_cache)2020 __svc_dupdone(struct svc_req *req, caddr_t resp_buf, uint_t resp_bufsz,
2021     int status, char *xprt_cache)
2022 {
2023 	uint32_t drxid, drhash;
2024 	int rc;
2025 
2026 	/* LINTED pointer alignment */
2027 	struct dupcache *dc = (struct dupcache *)xprt_cache;
2028 
2029 	if (dc == NULL) {
2030 		syslog(LOG_ERR, "__svc_dupdone: undefined cache");
2031 		return (DUP_ERROR);
2032 	}
2033 
2034 	if (status != DUP_DONE && status != DUP_DROP) {
2035 		syslog(LOG_ERR, "__svc_dupdone: invalid dupdone status");
2036 		syslog(LOG_ERR, "	 must be DUP_DONE or DUP_DROP");
2037 		return (DUP_ERROR);
2038 	}
2039 
2040 	/* find the xid of the entry in the cache */
2041 	if (SVC_CONTROL(req->rq_xprt, SVCGET_XID, (void*)&drxid) == FALSE) {
2042 		syslog(LOG_ERR, "__svc_dup: xid error");
2043 		return (DUP_ERROR);
2044 	}
2045 	drhash = drxid % dc->dc_buckets;
2046 
2047 	/* update the status of the entry and result buffers, if required */
2048 	if ((rc = __svc_dupcache_update(req, resp_buf, resp_bufsz, status,
2049 	    dc, drxid, drhash)) == DUP_ERROR) {
2050 		syslog(LOG_ERR, "__svc_dupdone: cache entry error");
2051 		return (DUP_ERROR);
2052 	}
2053 
2054 	return (rc);
2055 }
2056 
2057 /*
2058  * __svc_dupcache_update(struct svc_req *req, caddr_t resp_buf,
2059  * 	uint_t resp_bufsz, int status, struct dupcache *dc, uint32_t drxid,
2060  * 	uint32_t drhash)
2061  * Check if entry exists in the dupcacache. If it does, update its status
2062  * and time and also its buffer, if appropriate. Its possible, but unlikely
2063  * for DONE requests to not exist in the cache. Return DUP_ERROR or status.
2064  */
2065 static int
__svc_dupcache_update(struct svc_req * req,caddr_t resp_buf,uint_t resp_bufsz,int status,struct dupcache * dc,uint32_t drxid,uint32_t drhash)2066 __svc_dupcache_update(struct svc_req *req, caddr_t resp_buf, uint_t resp_bufsz,
2067     int status, struct dupcache *dc, uint32_t drxid, uint32_t drhash)
2068 {
2069 	struct dupreq *dr = NULL;
2070 	time_t timenow = time(NULL);
2071 
2072 	(void) rw_wrlock(&(dc->dc_lock));
2073 	dr = dc->dc_hashtbl[drhash];
2074 	while (dr != NULL) {
2075 		if (dr->dr_xid == drxid &&
2076 		    dr->dr_proc == req->rq_proc &&
2077 		    dr->dr_prog == req->rq_prog &&
2078 		    dr->dr_vers == req->rq_vers &&
2079 		    dr->dr_addr.len == req->rq_xprt->xp_rtaddr.len &&
2080 		    memcmp(dr->dr_addr.buf, req->rq_xprt->xp_rtaddr.buf,
2081 		    dr->dr_addr.len) == 0) { /* entry found */
2082 			if (dr->dr_hash != drhash) {
2083 				/* sanity check */
2084 				(void) rw_unlock(&(dc->dc_lock));
2085 				syslog(LOG_ERR,
2086 				"\n__svc_dupdone: hashing error");
2087 				return (DUP_ERROR);
2088 			}
2089 
2090 			/* store the results if bufer is not NULL */
2091 			if (resp_buf != NULL) {
2092 				if ((dr->dr_resp.buf =
2093 				    malloc(resp_bufsz)) == NULL) {
2094 					(void) rw_unlock(&(dc->dc_lock));
2095 					syslog(LOG_ERR,
2096 					    "__svc_dupdone: malloc failed");
2097 					return (DUP_ERROR);
2098 				}
2099 				(void) memset(dr->dr_resp.buf, 0, resp_bufsz);
2100 				(void) memcpy(dr->dr_resp.buf, resp_buf,
2101 				    (uint_t)resp_bufsz);
2102 				dr->dr_resp.len = resp_bufsz;
2103 			}
2104 
2105 			/* update status and done time */
2106 			dr->dr_status = status;
2107 			dr->dr_time = timenow;
2108 
2109 			/* move the entry to the mru position */
2110 			if (dc->dc_mru == NULL) {
2111 				dr->dr_next = dr;
2112 				dr->dr_prev = dr;
2113 			} else {
2114 				dr->dr_next = dc->dc_mru->dr_next;
2115 				dc->dc_mru->dr_next->dr_prev = dr;
2116 				dr->dr_prev = dc->dc_mru;
2117 				dc->dc_mru->dr_next = dr;
2118 			}
2119 			dc->dc_mru = dr;
2120 
2121 			(void) rw_unlock(&(dc->dc_lock));
2122 			return (status);
2123 		}
2124 		dr = dr->dr_chain;
2125 	}
2126 	(void) rw_unlock(&(dc->dc_lock));
2127 	syslog(LOG_ERR, "__svc_dupdone: entry not in dup cache");
2128 	return (DUP_ERROR);
2129 }
2130 
2131 #ifdef DUP_DEBUG
2132 /*
2133  * __svc_dupcache_debug(struct dupcache *dc)
2134  * print out the hash table stuff
2135  *
2136  * This function requires the caller to hold the reader
2137  * or writer version of the duplicate request cache lock (dc_lock).
2138  */
2139 static void
__svc_dupcache_debug(struct dupcache * dc)2140 __svc_dupcache_debug(struct dupcache *dc)
2141 {
2142 	struct dupreq *dr = NULL;
2143 	int i;
2144 	bool_t bval;
2145 
2146 	fprintf(stderr, "   HASHTABLE\n");
2147 	for (i = 0; i < dc->dc_buckets; i++) {
2148 		bval = FALSE;
2149 		dr = dc->dc_hashtbl[i];
2150 		while (dr != NULL) {
2151 			if (!bval) {	/* ensures bucket printed only once */
2152 				fprintf(stderr, "    bucket : %d\n", i);
2153 				bval = TRUE;
2154 			}
2155 			fprintf(stderr, "\txid: %u status: %d time: %ld",
2156 			    dr->dr_xid, dr->dr_status, dr->dr_time);
2157 			fprintf(stderr, " dr: %x chain: %x prevchain: %x\n",
2158 			    dr, dr->dr_chain, dr->dr_prevchain);
2159 			dr = dr->dr_chain;
2160 		}
2161 	}
2162 
2163 	fprintf(stderr, "   LRU\n");
2164 	if (dc->dc_mru) {
2165 		dr = dc->dc_mru->dr_next;	/* lru */
2166 		while (dr != dc->dc_mru) {
2167 			fprintf(stderr, "\txid: %u status : %d time : %ld",
2168 			    dr->dr_xid, dr->dr_status, dr->dr_time);
2169 			fprintf(stderr, " dr: %x next: %x prev: %x\n",
2170 			    dr, dr->dr_next, dr->dr_prev);
2171 			dr = dr->dr_next;
2172 		}
2173 		fprintf(stderr, "\txid: %u status: %d time: %ld",
2174 		    dr->dr_xid, dr->dr_status, dr->dr_time);
2175 		fprintf(stderr, " dr: %x next: %x prev: %x\n",
2176 		    dr, dr->dr_next, dr->dr_prev);
2177 	}
2178 }
2179 #endif /* DUP_DEBUG */
2180