xref: /netbsd-src/sys/nfs/nfs_iod.c (revision 313c6c94c424eed90c7b7e494aa83308a0a5d0ce)
1 /*	$NetBSD: nfs_iod.c,v 1.3 2009/03/14 21:04:25 dsl Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)nfs_syscalls.c	8.5 (Berkeley) 3/30/95
35  */
36 
37 #include <sys/cdefs.h>
38 __KERNEL_RCSID(0, "$NetBSD: nfs_iod.c,v 1.3 2009/03/14 21:04:25 dsl Exp $");
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/file.h>
44 #include <sys/stat.h>
45 #include <sys/vnode.h>
46 #include <sys/mount.h>
47 #include <sys/proc.h>
48 #include <sys/uio.h>
49 #include <sys/malloc.h>
50 #include <sys/kmem.h>
51 #include <sys/buf.h>
52 #include <sys/mbuf.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/signalvar.h>
56 #include <sys/domain.h>
57 #include <sys/protosw.h>
58 #include <sys/namei.h>
59 #include <sys/syslog.h>
60 #include <sys/filedesc.h>
61 #include <sys/kthread.h>
62 #include <sys/kauth.h>
63 #include <sys/syscallargs.h>
64 
65 #include <netinet/in.h>
66 #include <netinet/tcp.h>
67 #include <nfs/xdr_subs.h>
68 #include <nfs/rpcv2.h>
69 #include <nfs/nfsproto.h>
70 #include <nfs/nfs.h>
71 #include <nfs/nfsm_subs.h>
72 #include <nfs/nfsrvcache.h>
73 #include <nfs/nfsmount.h>
74 #include <nfs/nfsnode.h>
75 #include <nfs/nfsrtt.h>
76 #include <nfs/nfs_var.h>
77 
78 int nuidhash_max = NFS_MAXUIDHASH;
79 
80 /*
81  * locking order:
82  *	nfs_iodlist_lock -> nid_lock -> nm_lock
83  */
84 kmutex_t nfs_iodlist_lock;
85 struct nfs_iodlist nfs_iodlist_idle;
86 struct nfs_iodlist nfs_iodlist_all;
87 int nfs_niothreads = -1; /* == "0, and has never been set" */
88 int nfs_defect = 0;
89 
90 /*
91  * Asynchronous I/O threads for client nfs.
92  * They do read-ahead and write-behind operations on the block I/O cache.
93  * Never returns unless it fails or gets killed.
94  */
95 
96 static void
97 nfssvc_iod(void *arg)
98 {
99 	struct buf *bp;
100 	struct nfs_iod *myiod;
101 	struct nfsmount *nmp;
102 
103 	myiod = kmem_alloc(sizeof(*myiod), KM_SLEEP);
104 	mutex_init(&myiod->nid_lock, MUTEX_DEFAULT, IPL_NONE);
105 	cv_init(&myiod->nid_cv, "nfsiod");
106 	myiod->nid_exiting = false;
107 	myiod->nid_mount = NULL;
108 	mutex_enter(&nfs_iodlist_lock);
109 	LIST_INSERT_HEAD(&nfs_iodlist_all, myiod, nid_all);
110 	mutex_exit(&nfs_iodlist_lock);
111 
112 	for (;;) {
113 		mutex_enter(&nfs_iodlist_lock);
114 		LIST_INSERT_HEAD(&nfs_iodlist_idle, myiod, nid_idle);
115 		mutex_exit(&nfs_iodlist_lock);
116 
117 		mutex_enter(&myiod->nid_lock);
118 		while (/*CONSTCOND*/ true) {
119 			nmp = myiod->nid_mount;
120 			if (nmp) {
121 				myiod->nid_mount = NULL;
122 				break;
123 			}
124 			if (__predict_false(myiod->nid_exiting)) {
125 				/*
126 				 * drop nid_lock to preserve locking order.
127 				 */
128 				mutex_exit(&myiod->nid_lock);
129 				mutex_enter(&nfs_iodlist_lock);
130 				mutex_enter(&myiod->nid_lock);
131 				/*
132 				 * recheck nid_mount because nfs_asyncio can
133 				 * pick us in the meantime as we are still on
134 				 * nfs_iodlist_lock.
135 				 */
136 				if (myiod->nid_mount != NULL) {
137 					mutex_exit(&nfs_iodlist_lock);
138 					continue;
139 				}
140 				LIST_REMOVE(myiod, nid_idle);
141 				mutex_exit(&nfs_iodlist_lock);
142 				goto quit;
143 			}
144 			cv_wait(&myiod->nid_cv, &myiod->nid_lock);
145 		}
146 		mutex_exit(&myiod->nid_lock);
147 
148 		mutex_enter(&nmp->nm_lock);
149 		while ((bp = TAILQ_FIRST(&nmp->nm_bufq)) != NULL) {
150 			/* Take one off the front of the list */
151 			TAILQ_REMOVE(&nmp->nm_bufq, bp, b_freelist);
152 			nmp->nm_bufqlen--;
153 			if (nmp->nm_bufqlen < 2 * nmp->nm_bufqiods) {
154 				cv_broadcast(&nmp->nm_aiocv);
155 			}
156 			mutex_exit(&nmp->nm_lock);
157 			KERNEL_LOCK(1, curlwp);
158 			(void)nfs_doio(bp);
159 			KERNEL_UNLOCK_LAST(curlwp);
160 			mutex_enter(&nmp->nm_lock);
161 			/*
162 			 * If there are more than one iod on this mount,
163 			 * then defect so that the iods can be shared out
164 			 * fairly between the mounts
165 			 */
166 			if (nfs_defect && nmp->nm_bufqiods > 1) {
167 				break;
168 			}
169 		}
170 		KASSERT(nmp->nm_bufqiods > 0);
171 		nmp->nm_bufqiods--;
172 		mutex_exit(&nmp->nm_lock);
173 	}
174 quit:
175 	KASSERT(myiod->nid_mount == NULL);
176 	mutex_exit(&myiod->nid_lock);
177 
178 	cv_destroy(&myiod->nid_cv);
179 	mutex_destroy(&myiod->nid_lock);
180 	kmem_free(myiod, sizeof(*myiod));
181 
182 	kthread_exit(0);
183 }
184 
185 void
186 nfs_iodinit(void)
187 {
188 
189 	mutex_init(&nfs_iodlist_lock, MUTEX_DEFAULT, IPL_NONE);
190 	LIST_INIT(&nfs_iodlist_all);
191 	LIST_INIT(&nfs_iodlist_idle);
192 }
193 
194 void
195 nfs_iodfini(void)
196 {
197 	int error;
198 
199 	error = nfs_set_niothreads(0);
200 	KASSERT(error == 0);
201 	mutex_destroy(&nfs_iodlist_lock);
202 }
203 
204 int
205 nfs_set_niothreads(int newval)
206 {
207 	struct nfs_iod *nid;
208 	int error = 0;
209         int hold_count;
210 
211 	KERNEL_UNLOCK_ALL(curlwp, &hold_count);
212 
213 	mutex_enter(&nfs_iodlist_lock);
214 	/* clamp to sane range */
215 	nfs_niothreads = max(0, min(newval, NFS_MAXASYNCDAEMON));
216 
217 	while (nfs_numasync != nfs_niothreads && error == 0) {
218 		while (nfs_numasync < nfs_niothreads) {
219 
220 			/*
221 			 * kthread_create can wait for pagedaemon and
222 			 * pagedaemon can wait for nfsiod which needs to acquire
223 			 * nfs_iodlist_lock.
224 			 */
225 
226 			mutex_exit(&nfs_iodlist_lock);
227 			error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
228 			    nfssvc_iod, NULL, NULL, "nfsio");
229 			mutex_enter(&nfs_iodlist_lock);
230 			if (error) {
231 				/* give up */
232 				nfs_niothreads = nfs_numasync;
233 				break;
234 			}
235 			nfs_numasync++;
236 		}
237 		while (nfs_numasync > nfs_niothreads) {
238 			nid = LIST_FIRST(&nfs_iodlist_all);
239 			if (nid == NULL) {
240 				/* iod has not started yet. */
241 				kpause("nfsiorm", false, hz, &nfs_iodlist_lock);
242 				continue;
243 			}
244 			LIST_REMOVE(nid, nid_all);
245 			mutex_enter(&nid->nid_lock);
246 			KASSERT(!nid->nid_exiting);
247 			nid->nid_exiting = true;
248 			cv_signal(&nid->nid_cv);
249 			mutex_exit(&nid->nid_lock);
250 			nfs_numasync--;
251 		}
252 	}
253 	mutex_exit(&nfs_iodlist_lock);
254 
255 	KERNEL_LOCK(hold_count, curlwp);
256 	return error;
257 }
258 
259 /*
260  * Get an authorization string for the uid by having the mount_nfs sitting
261  * on this mount point porpous out of the kernel and do it.
262  */
263 int
264 nfs_getauth(struct nfsmount *nmp, struct nfsreq *rep, kauth_cred_t cred, char **auth_str, int *auth_len, char *verf_str, int *verf_len, NFSKERBKEY_T key)
265 	/* key:		 return session key */
266 {
267 	int error = 0;
268 
269 	while ((nmp->nm_iflag & NFSMNT_WAITAUTH) == 0) {
270 		nmp->nm_iflag |= NFSMNT_WANTAUTH;
271 		(void) tsleep((void *)&nmp->nm_authtype, PSOCK,
272 			"nfsauth1", 2 * hz);
273 		error = nfs_sigintr(nmp, rep, rep->r_lwp);
274 		if (error) {
275 			nmp->nm_iflag &= ~NFSMNT_WANTAUTH;
276 			return (error);
277 		}
278 	}
279 	nmp->nm_iflag &= ~(NFSMNT_WAITAUTH | NFSMNT_WANTAUTH);
280 	nmp->nm_authstr = *auth_str = (char *)malloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK);
281 	nmp->nm_authlen = RPCAUTH_MAXSIZ;
282 	nmp->nm_verfstr = verf_str;
283 	nmp->nm_verflen = *verf_len;
284 	nmp->nm_authuid = kauth_cred_geteuid(cred);
285 	wakeup((void *)&nmp->nm_authstr);
286 
287 	/*
288 	 * And wait for mount_nfs to do its stuff.
289 	 */
290 	while ((nmp->nm_iflag & NFSMNT_HASAUTH) == 0 && error == 0) {
291 		(void) tsleep((void *)&nmp->nm_authlen, PSOCK,
292 			"nfsauth2", 2 * hz);
293 		error = nfs_sigintr(nmp, rep, rep->r_lwp);
294 	}
295 	if (nmp->nm_iflag & NFSMNT_AUTHERR) {
296 		nmp->nm_iflag &= ~NFSMNT_AUTHERR;
297 		error = EAUTH;
298 	}
299 	if (error)
300 		free((void *)*auth_str, M_TEMP);
301 	else {
302 		*auth_len = nmp->nm_authlen;
303 		*verf_len = nmp->nm_verflen;
304 		memcpy(key, nmp->nm_key, sizeof (NFSKERBKEY_T));
305 	}
306 	nmp->nm_iflag &= ~NFSMNT_HASAUTH;
307 	nmp->nm_iflag |= NFSMNT_WAITAUTH;
308 	if (nmp->nm_iflag & NFSMNT_WANTAUTH) {
309 		nmp->nm_iflag &= ~NFSMNT_WANTAUTH;
310 		wakeup((void *)&nmp->nm_authtype);
311 	}
312 	return (error);
313 }
314 
315 /*
316  * Get a nickname authenticator and verifier.
317  */
318 int
319 nfs_getnickauth(struct nfsmount *nmp, kauth_cred_t cred, char **auth_str,
320     int *auth_len, char *verf_str, int verf_len)
321 {
322 	struct timeval ktvin, ktvout, tv;
323 	struct nfsuid *nuidp;
324 	u_int32_t *nickp, *verfp;
325 
326 	memset(&ktvout, 0, sizeof ktvout);	/* XXX gcc */
327 
328 #ifdef DIAGNOSTIC
329 	if (verf_len < (4 * NFSX_UNSIGNED))
330 		panic("nfs_getnickauth verf too small");
331 #endif
332 	LIST_FOREACH(nuidp, NMUIDHASH(nmp, kauth_cred_geteuid(cred)), nu_hash) {
333 		if (kauth_cred_geteuid(nuidp->nu_cr) == kauth_cred_geteuid(cred))
334 			break;
335 	}
336 	if (!nuidp || nuidp->nu_expire < time_second)
337 		return (EACCES);
338 
339 	/*
340 	 * Move to the end of the lru list (end of lru == most recently used).
341 	 */
342 	TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru);
343 	TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru);
344 
345 	nickp = (u_int32_t *)malloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK);
346 	*nickp++ = txdr_unsigned(RPCAKN_NICKNAME);
347 	*nickp = txdr_unsigned(nuidp->nu_nickname);
348 	*auth_str = (char *)nickp;
349 	*auth_len = 2 * NFSX_UNSIGNED;
350 
351 	/*
352 	 * Now we must encrypt the verifier and package it up.
353 	 */
354 	verfp = (u_int32_t *)verf_str;
355 	*verfp++ = txdr_unsigned(RPCAKN_NICKNAME);
356 	getmicrotime(&tv);
357 	if (tv.tv_sec > nuidp->nu_timestamp.tv_sec ||
358 	    (tv.tv_sec == nuidp->nu_timestamp.tv_sec &&
359 	     tv.tv_usec > nuidp->nu_timestamp.tv_usec))
360 		nuidp->nu_timestamp = tv;
361 	else
362 		nuidp->nu_timestamp.tv_usec++;
363 	ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec);
364 	ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec);
365 
366 	/*
367 	 * Now encrypt the timestamp verifier in ecb mode using the session
368 	 * key.
369 	 */
370 #ifdef NFSKERB
371 	XXX
372 #endif
373 
374 	*verfp++ = ktvout.tv_sec;
375 	*verfp++ = ktvout.tv_usec;
376 	*verfp = 0;
377 	return (0);
378 }
379 
380 /*
381  * Save the current nickname in a hash list entry on the mount point.
382  */
383 int
384 nfs_savenickauth(struct nfsmount *nmp, kauth_cred_t cred, int len, NFSKERBKEY_T key, struct mbuf **mdp, char **dposp, struct mbuf *mrep)
385 {
386 	struct nfsuid *nuidp;
387 	u_int32_t *tl;
388 	int32_t t1;
389 	struct mbuf *md = *mdp;
390 	struct timeval ktvin, ktvout;
391 	u_int32_t nick;
392 	char *dpos = *dposp, *cp2;
393 	int deltasec, error = 0;
394 
395 	memset(&ktvout, 0, sizeof ktvout);	 /* XXX gcc */
396 
397 	if (len == (3 * NFSX_UNSIGNED)) {
398 		nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
399 		ktvin.tv_sec = *tl++;
400 		ktvin.tv_usec = *tl++;
401 		nick = fxdr_unsigned(u_int32_t, *tl);
402 
403 		/*
404 		 * Decrypt the timestamp in ecb mode.
405 		 */
406 #ifdef NFSKERB
407 		XXX
408 #endif
409 		ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec);
410 		ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec);
411 		deltasec = time_second - ktvout.tv_sec;
412 		if (deltasec < 0)
413 			deltasec = -deltasec;
414 		/*
415 		 * If ok, add it to the hash list for the mount point.
416 		 */
417 		if (deltasec <= NFS_KERBCLOCKSKEW) {
418 			if (nmp->nm_numuids < nuidhash_max) {
419 				nmp->nm_numuids++;
420 				nuidp = kmem_alloc(sizeof(*nuidp), KM_SLEEP);
421 			} else {
422 				nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead);
423 				LIST_REMOVE(nuidp, nu_hash);
424 				TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp,
425 					nu_lru);
426 			}
427 			nuidp->nu_flag = 0;
428 			kauth_cred_seteuid(nuidp->nu_cr, kauth_cred_geteuid(cred));
429 			nuidp->nu_expire = time_second + NFS_KERBTTL;
430 			nuidp->nu_timestamp = ktvout;
431 			nuidp->nu_nickname = nick;
432 			memcpy(nuidp->nu_key, key, sizeof (NFSKERBKEY_T));
433 			TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp,
434 				nu_lru);
435 			LIST_INSERT_HEAD(NMUIDHASH(nmp, kauth_cred_geteuid(cred)),
436 				nuidp, nu_hash);
437 		}
438 	} else
439 		nfsm_adv(nfsm_rndup(len));
440 nfsmout:
441 	*mdp = md;
442 	*dposp = dpos;
443 	return (error);
444 }
445