xref: /illumos-gate/usr/src/uts/common/klm/nlm_rpc_handle.c (revision 45ede40b2394db7967e59f19288fae9b62efd4aa)
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 2019 Nexenta by DDN, Inc. All rights reserved.
24  * Copyright (c) 2012 by Delphix. All rights reserved.
25  */
26 
27 #include <sys/param.h>
28 #include <sys/systm.h>
29 #include <sys/socket.h>
30 #include <sys/syslog.h>
31 #include <sys/systm.h>
32 #include <sys/unistd.h>
33 #include <sys/queue.h>
34 #include <sys/sdt.h>
35 #include <netinet/in.h>
36 
37 #include <rpc/rpc.h>
38 #include <rpc/xdr.h>
39 #include <rpc/pmap_prot.h>
40 #include <rpc/pmap_clnt.h>
41 #include <rpc/rpcb_prot.h>
42 
43 #include <rpcsvc/nlm_prot.h>
44 #include <rpcsvc/sm_inter.h>
45 
46 #include "nlm_impl.h"
47 
48 /*
49  * The following errors codes from nlm_null_rpc indicate that the port we have
50  * cached for the client's NLM service is stale and that we need to establish
51  * a new RPC client.
52  */
53 #define	NLM_STALE_CLNT(_status)			\
54 	((_status) == RPC_PROGUNAVAIL ||	\
55 	(_status) == RPC_PROGVERSMISMATCH ||	\
56 	(_status) == RPC_PROCUNAVAIL ||		\
57 	(_status) == RPC_CANTCONNECT ||		\
58 	(_status) == RPC_XPRTFAILED)
59 
60 static struct kmem_cache *nlm_rpch_cache = NULL;
61 
62 static int nlm_rpch_ctor(void *, void *, int);
63 static void nlm_rpch_dtor(void *, void *);
64 static void destroy_rpch(nlm_rpc_t *);
65 static nlm_rpc_t *get_nlm_rpc_fromcache(struct nlm_host *, int);
66 static void update_host_rpcbinding(struct nlm_host *, int);
67 static int refresh_nlm_rpc(struct nlm_host *, nlm_rpc_t *);
68 static void nlm_host_rele_rpc_locked(struct nlm_host *, nlm_rpc_t *);
69 
70 static nlm_rpc_t *
71 get_nlm_rpc_fromcache(struct nlm_host *hostp, int vers)
72 {
73 	nlm_rpc_t *rpcp;
74 	bool_t found = FALSE;
75 
76 	ASSERT(MUTEX_HELD(&hostp->nh_lock));
77 	if (TAILQ_EMPTY(&hostp->nh_rpchc))
78 		return (NULL);
79 
80 	TAILQ_FOREACH(rpcp, &hostp->nh_rpchc, nr_link) {
81 		if (rpcp->nr_vers == vers) {
82 			found = TRUE;
83 			break;
84 		}
85 	}
86 
87 	if (!found)
88 		return (NULL);
89 
90 	TAILQ_REMOVE(&hostp->nh_rpchc, rpcp, nr_link);
91 	return (rpcp);
92 }
93 
94 /*
95  * Update host's RPC binding (host->nh_addr).
96  * The function is executed by only one thread at time.
97  */
98 static void
99 update_host_rpcbinding(struct nlm_host *hostp, int vers)
100 {
101 	enum clnt_stat stat;
102 
103 	ASSERT(MUTEX_HELD(&hostp->nh_lock));
104 
105 	/*
106 	 * Mark RPC binding state as "update in progress" in order
107 	 * to say other threads that they need to wait until binding
108 	 * is fully updated.
109 	 */
110 	hostp->nh_rpcb_state = NRPCB_UPDATE_INPROGRESS;
111 	hostp->nh_rpcb_ustat = RPC_SUCCESS;
112 	mutex_exit(&hostp->nh_lock);
113 
114 	stat = rpcbind_getaddr(&hostp->nh_knc, NLM_PROG, vers, &hostp->nh_addr);
115 	mutex_enter(&hostp->nh_lock);
116 
117 	hostp->nh_rpcb_state = ((stat == RPC_SUCCESS) ?
118 	    NRPCB_UPDATED : NRPCB_NEED_UPDATE);
119 
120 	hostp->nh_rpcb_ustat = stat;
121 	cv_broadcast(&hostp->nh_rpcb_cv);
122 }
123 
124 /*
125  * Refresh RPC handle taken from host handles cache.
126  * This function is called when an RPC handle is either
127  * uninitialized or was initialized using a binding that's
128  * no longer current.
129  */
130 static int
131 refresh_nlm_rpc(struct nlm_host *hostp, nlm_rpc_t *rpcp)
132 {
133 	uint32_t zero = 0;
134 	int ret;
135 
136 	if (rpcp->nr_handle == NULL) {
137 		bool_t clset = TRUE;
138 
139 		ret = clnt_tli_kcreate(&hostp->nh_knc, &hostp->nh_addr,
140 		    NLM_PROG, rpcp->nr_vers, 0, NLM_RPC_RETRIES,
141 		    CRED(), &rpcp->nr_handle);
142 
143 		/*
144 		 * Set the client's CLSET_NODELAYONERR option to true. The
145 		 * RPC clnt_call interface creates an artificial delay for
146 		 * certain call errors in order to prevent RPC consumers
147 		 * from getting into tight retry loops. Since this function is
148 		 * called by the NLM service routines we would like to avoid
149 		 * this artificial delay when possible. We do not retry if the
150 		 * NULL request fails so it is safe for us to turn this option
151 		 * on.
152 		 */
153 		if (clnt_control(rpcp->nr_handle, CLSET_NODELAYONERR,
154 		    (char *)&clset) == FALSE) {
155 			NLM_ERR("Unable to set CLSET_NODELAYONERR\n");
156 		}
157 	} else {
158 		ret = clnt_tli_kinit(rpcp->nr_handle, &hostp->nh_knc,
159 		    &hostp->nh_addr, 0, NLM_RPC_RETRIES, CRED());
160 		if (ret == 0) {
161 			enum clnt_stat stat;
162 
163 			/*
164 			 * Check whether host's RPC binding is still
165 			 * fresh, i.e. if remote program is still sits
166 			 * on the same port we assume. Call NULL proc
167 			 * to do it.
168 			 *
169 			 * Note: Even though we set no delay on error on the
170 			 * client handle the call to nlm_null_rpc can still
171 			 * delay for 10 seconds before returning an error. For
172 			 * example the no delay on error option is not honored
173 			 * for RPC_XPRTFAILED errors (see clnt_cots_kcallit).
174 			 */
175 			stat = nlm_null_rpc(rpcp->nr_handle, rpcp->nr_vers);
176 			if (NLM_STALE_CLNT(stat)) {
177 				ret = ESTALE;
178 			}
179 			/*
180 			 * Need to reset the XID after the null call above,
181 			 * otherwise we'll reuse the XID from that call.
182 			 */
183 			(void) CLNT_CONTROL(rpcp->nr_handle, CLSET_XID,
184 			    (char *)&zero);
185 		}
186 	}
187 
188 	return (ret);
189 }
190 
191 /*
192  * Get RPC handle that can be used to talk to the NLM
193  * of given version running on given host.
194  * Saves obtained RPC handle to rpcpp argument.
195  *
196  * If error occures, return nonzero error code.
197  */
198 int
199 nlm_host_get_rpc(struct nlm_host *hostp, int vers, nlm_rpc_t **rpcpp)
200 {
201 	nlm_rpc_t *rpcp = NULL;
202 	int rc;
203 
204 	mutex_enter(&hostp->nh_lock);
205 
206 	/*
207 	 * If this handle is either uninitialized, or was
208 	 * initialized using binding that's now stale
209 	 * do the init or re-init.
210 	 * See comments to enum nlm_rpcb_state for more
211 	 * details.
212 	 */
213 again:
214 	while (hostp->nh_rpcb_state != NRPCB_UPDATED) {
215 		if (hostp->nh_rpcb_state == NRPCB_UPDATE_INPROGRESS) {
216 			rc = cv_wait_sig(&hostp->nh_rpcb_cv, &hostp->nh_lock);
217 			if (rc == 0) {
218 				mutex_exit(&hostp->nh_lock);
219 				rc = EINTR;
220 				goto errout;
221 			}
222 		}
223 
224 		/*
225 		 * Check if RPC binding was marked for update.
226 		 * If so, start RPC binding update operation.
227 		 * NOTE: the operation can be executed by only
228 		 * one thread at time.
229 		 */
230 		if (hostp->nh_rpcb_state == NRPCB_NEED_UPDATE)
231 			update_host_rpcbinding(hostp, vers);
232 
233 		/*
234 		 * Check if RPC error occured during RPC binding
235 		 * update operation. If so, report a correspoding
236 		 * error.
237 		 */
238 		if (hostp->nh_rpcb_ustat != RPC_SUCCESS) {
239 			mutex_exit(&hostp->nh_lock);
240 			rc = ENOENT;
241 			goto errout;
242 		}
243 	}
244 
245 	rpcp = get_nlm_rpc_fromcache(hostp, vers);
246 	mutex_exit(&hostp->nh_lock);
247 	if (rpcp == NULL) {
248 		/*
249 		 * There weren't any RPC handles in a host
250 		 * cache. No luck, just create a new one.
251 		 */
252 		rpcp = kmem_cache_alloc(nlm_rpch_cache, KM_SLEEP);
253 		rpcp->nr_vers = vers;
254 	}
255 
256 	/*
257 	 * Refresh RPC binding
258 	 */
259 	rc = refresh_nlm_rpc(hostp, rpcp);
260 	if (rc != 0) {
261 		if (rc == ESTALE) {
262 			/*
263 			 * Host's RPC binding is stale, we have
264 			 * to update it. Put the RPC handle back
265 			 * to the cache and mark the host as
266 			 * "need update".
267 			 */
268 			mutex_enter(&hostp->nh_lock);
269 			hostp->nh_rpcb_state = NRPCB_NEED_UPDATE;
270 			nlm_host_rele_rpc_locked(hostp, rpcp);
271 			goto again;
272 		}
273 
274 		destroy_rpch(rpcp);
275 		goto errout;
276 	}
277 
278 	DTRACE_PROBE2(end, struct nlm_host *, hostp,
279 	    nlm_rpc_t *, rpcp);
280 
281 	*rpcpp = rpcp;
282 	return (0);
283 
284 errout:
285 	NLM_ERR("Can't get RPC client handle for: %s", hostp->nh_name);
286 	return (rc);
287 }
288 
289 void
290 nlm_host_rele_rpc(struct nlm_host *hostp, nlm_rpc_t *rpcp)
291 {
292 	mutex_enter(&hostp->nh_lock);
293 	nlm_host_rele_rpc_locked(hostp, rpcp);
294 	mutex_exit(&hostp->nh_lock);
295 }
296 
297 static void
298 nlm_host_rele_rpc_locked(struct nlm_host *hostp, nlm_rpc_t *rpcp)
299 {
300 	ASSERT(mutex_owned(&hostp->nh_lock));
301 	TAILQ_INSERT_HEAD(&hostp->nh_rpchc, rpcp, nr_link);
302 }
303 
304 /*
305  * The function invalidates host's RPC binding by marking it
306  * as not fresh. In this case another time thread tries to
307  * get RPC handle from host's handles cache, host's RPC binding
308  * will be updated.
309  *
310  * The function should be executed when RPC call invoked via
311  * handle taken from RPC cache returns RPC_PROCUNAVAIL.
312  */
313 void
314 nlm_host_invalidate_binding(struct nlm_host *hostp)
315 {
316 	mutex_enter(&hostp->nh_lock);
317 	hostp->nh_rpcb_state = NRPCB_NEED_UPDATE;
318 	mutex_exit(&hostp->nh_lock);
319 }
320 
321 void
322 nlm_rpc_init(void)
323 {
324 	nlm_rpch_cache = kmem_cache_create("nlm_rpch_cache",
325 	    sizeof (nlm_rpc_t), 0, nlm_rpch_ctor, nlm_rpch_dtor,
326 	    NULL, NULL, NULL, 0);
327 }
328 
329 void
330 nlm_rpc_cache_destroy(struct nlm_host *hostp)
331 {
332 	nlm_rpc_t *rpcp;
333 
334 	/*
335 	 * There's no need to lock host's mutex here,
336 	 * nlm_rpc_cache_destroy() should be called from
337 	 * only one place: nlm_host_destroy, when all
338 	 * resources host owns are already cleaned up.
339 	 * So there shouldn't be any raises.
340 	 */
341 	while ((rpcp = TAILQ_FIRST(&hostp->nh_rpchc)) != NULL) {
342 		TAILQ_REMOVE(&hostp->nh_rpchc, rpcp, nr_link);
343 		destroy_rpch(rpcp);
344 	}
345 }
346 
347 /* ARGSUSED */
348 static int
349 nlm_rpch_ctor(void *datap, void *cdrarg, int kmflags)
350 {
351 	nlm_rpc_t *rpcp = (nlm_rpc_t *)datap;
352 
353 	bzero(rpcp, sizeof (*rpcp));
354 	return (0);
355 }
356 
357 /* ARGSUSED */
358 static void
359 nlm_rpch_dtor(void *datap, void *cdrarg)
360 {
361 	nlm_rpc_t *rpcp = (nlm_rpc_t *)datap;
362 	ASSERT(rpcp->nr_handle == NULL);
363 }
364 
365 static void
366 destroy_rpch(nlm_rpc_t *rpcp)
367 {
368 	if (rpcp->nr_handle != NULL) {
369 		AUTH_DESTROY(rpcp->nr_handle->cl_auth);
370 		CLNT_DESTROY(rpcp->nr_handle);
371 		rpcp->nr_handle = NULL;
372 	}
373 
374 	kmem_cache_free(nlm_rpch_cache, rpcp);
375 }
376