xref: /dragonfly/sys/vfs/nfs/nfs_iod.c (revision 8af44722)
1 /*
2  * Copyright (c) 2009 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in
15  *    the documentation and/or other materials provided with the
16  *    distribution.
17  * 3. Neither the name of The DragonFly Project nor the names of its
18  *    contributors may be used to endorse or promote products derived
19  *    from this software without specific, prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 /*
35  * NFSIOD operations - now built into the kernel.
36  */
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/proc.h>
40 #include <sys/malloc.h>
41 #include <sys/mount.h>
42 #include <sys/kernel.h>
43 #include <sys/mbuf.h>
44 #include <sys/vnode.h>
45 #include <sys/fcntl.h>
46 #include <sys/protosw.h>
47 #include <sys/resourcevar.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
50 #include <sys/socketops.h>
51 #include <sys/syslog.h>
52 #include <sys/thread.h>
53 #include <sys/tprintf.h>
54 #include <sys/sysctl.h>
55 #include <sys/signalvar.h>
56 
57 #include <sys/signal2.h>
58 
59 #include <netinet/in.h>
60 #include <netinet/tcp.h>
61 
62 #include "rpcv2.h"
63 #include "nfsproto.h"
64 #include "nfs.h"
65 #include "xdr_subs.h"
66 #include "nfsm_subs.h"
67 #include "nfsmount.h"
68 #include "nfsnode.h"
69 #include "nfsrtt.h"
70 
71 /*
72  * nfs service connection reader thread
73  */
74 void
75 nfssvc_iod_reader(void *arg)
76 {
77 	struct nfsmount *nmp = arg;
78 	struct nfsm_info *info;
79 	struct nfsreq *req;
80 	int error;
81 
82 	lwkt_gettoken(&nmp->nm_token);
83 
84 	if (nmp->nm_rxstate == NFSSVC_INIT)
85 		nmp->nm_rxstate = NFSSVC_PENDING;
86 	for (;;) {
87 		if (nmp->nm_rxstate == NFSSVC_WAITING) {
88 			if (TAILQ_FIRST(&nmp->nm_reqq) == NULL &&
89 			    TAILQ_FIRST(&nmp->nm_reqrxq) == NULL) {
90 				tsleep(&nmp->nm_rxstate, 0, "nfsidl", 0);
91 			} else {
92 				/*
93 				 * This can happen during shutdown, we don't
94 				 * want to hardloop.
95 				 */
96 				error = nfs_reply(nmp, NULL);
97 				if (error && error != EWOULDBLOCK) {
98 					tsleep(&nmp->nm_rxstate, 0,
99 						"nfsxxx", hz / 10);
100 				}
101 			}
102 			continue;
103 		}
104 		if (nmp->nm_rxstate != NFSSVC_PENDING)
105 			break;
106 		nmp->nm_rxstate = NFSSVC_WAITING;
107 
108 		/*
109 		 * Process requests which have received replies.  Only
110 		 * process the post-reply states.  If we get EINPROGRESS
111 		 * it means the request went back to an auth or retransmit
112 		 * state and we let the iod_writer thread deal with it.
113 		 *
114 		 * Any lock on the request is strictly temporary due to
115 		 * MP races (XXX).
116 		 *
117 		 * If the request completes we run the info->done call
118 		 * to finish up the I/O.
119 		 */
120 		while ((req = TAILQ_FIRST(&nmp->nm_reqrxq)) != NULL) {
121 			if (req->r_flags & R_LOCKED) {
122 				while (req->r_flags & R_LOCKED) {
123 					req->r_flags |= R_WANTED;
124 					tsleep(req, 0, "nfstrac", 0);
125 				}
126 				continue;
127 			}
128 			TAILQ_REMOVE(&nmp->nm_reqrxq, req, r_chain);
129 			info = req->r_info;
130 			KKASSERT(info);
131 			info->error = nfs_request(info,
132 						  NFSM_STATE_PROCESSREPLY,
133 						  NFSM_STATE_DONE);
134 			if (info->error == EINPROGRESS) {
135 				kprintf("rxq: move info %p back to txq\n", info);
136 				TAILQ_INSERT_TAIL(&nmp->nm_reqtxq, req, r_chain);
137 				nfssvc_iod_writer_wakeup(nmp);
138 			} else {
139 				atomic_subtract_int(&nmp->nm_bioqlen, 1);
140 				info->done(info);
141 			}
142 		}
143 	}
144 	nmp->nm_rxthread = NULL;
145 	nmp->nm_rxstate = NFSSVC_DONE;
146 
147 	lwkt_reltoken(&nmp->nm_token);
148 	wakeup(&nmp->nm_rxthread);
149 }
150 
151 /*
152  * nfs service connection writer thread
153  *
154  * The writer sits on the send side of the client's socket and
155  * does both the initial processing of BIOs and also transmission
156  * and retransmission of nfsreq's.
157  *
158  * The writer processes both new BIOs from nm_bioq and retransmit
159  * or state machine jumpbacks from nm_reqtxq
160  */
161 void
162 nfssvc_iod_writer(void *arg)
163 {
164 	struct nfsmount *nmp = arg;
165 	struct bio *bio;
166 	struct nfsreq *req;
167 	struct vnode *vp;
168 	nfsm_info_t info;
169 
170 	lwkt_gettoken(&nmp->nm_token);
171 
172 	if (nmp->nm_txstate == NFSSVC_INIT)
173 		nmp->nm_txstate = NFSSVC_PENDING;
174 
175 	for (;;) {
176 		if (nmp->nm_txstate == NFSSVC_WAITING) {
177 			tsleep(&nmp->nm_txstate, 0, "nfsidl", 0);
178 			continue;
179 		}
180 		if (nmp->nm_txstate != NFSSVC_PENDING)
181 			break;
182 		nmp->nm_txstate = NFSSVC_WAITING;
183 
184 		/*
185 		 * Eep, we could blow out the mbuf allocator if we just
186 		 * did everything the kernel wanted us to do.
187 		 */
188 		while ((bio = TAILQ_FIRST(&nmp->nm_bioq)) != NULL) {
189 			if (nmp->nm_reqqlen > nfs_maxasyncbio)
190 				break;
191 			TAILQ_REMOVE(&nmp->nm_bioq, bio, bio_act);
192 			vp = bio->bio_driver_info;
193 			nfs_startio(vp, bio, NULL);
194 		}
195 
196 		/*
197 		 * Process reauths & retransmits.  If we get an EINPROGRESS
198 		 * it means the state transitioned to WAITREPLY or later.
199 		 * Otherwise the request completed (probably with an error
200 		 * since we didn't get to a replied state).
201 		 */
202 		while ((req = TAILQ_FIRST(&nmp->nm_reqtxq)) != NULL) {
203 			TAILQ_REMOVE(&nmp->nm_reqtxq, req, r_chain);
204 			info = req->r_info;
205 			KKASSERT(info);
206 			info->error = nfs_request(info,
207 						  NFSM_STATE_AUTH,
208 						  NFSM_STATE_WAITREPLY);
209 			if (info->error == EINPROGRESS) {
210 				;
211 			} else {
212 				atomic_subtract_int(&nmp->nm_bioqlen, 1);
213 				info->done(info);
214 			}
215 		}
216 	}
217 	nmp->nm_txthread = NULL;
218 	nmp->nm_txstate = NFSSVC_DONE;
219 	lwkt_reltoken(&nmp->nm_token);
220 	wakeup(&nmp->nm_txthread);
221 }
222 
223 void
224 nfssvc_iod_stop1(struct nfsmount *nmp)
225 {
226 	nmp->nm_txstate = NFSSVC_STOPPING;
227 	nmp->nm_rxstate = NFSSVC_STOPPING;
228 }
229 
230 void
231 nfssvc_iod_stop2(struct nfsmount *nmp)
232 {
233 	wakeup(&nmp->nm_txstate);
234 	while (nmp->nm_txthread)
235 		tsleep(&nmp->nm_txthread, 0, "nfssttx", hz*2);
236 	wakeup(&nmp->nm_rxstate);
237 	while (nmp->nm_rxthread)
238 		tsleep(&nmp->nm_rxthread, 0, "nfsstrx", hz*2);
239 }
240 
241 void
242 nfssvc_iod_writer_wakeup(struct nfsmount *nmp)
243 {
244 	if (nmp->nm_txstate == NFSSVC_WAITING) {
245 		nmp->nm_txstate = NFSSVC_PENDING;
246 		wakeup(&nmp->nm_txstate);
247 	}
248 }
249 
250 void
251 nfssvc_iod_reader_wakeup(struct nfsmount *nmp)
252 {
253 	if (nmp->nm_rxstate == NFSSVC_WAITING) {
254 		nmp->nm_rxstate = NFSSVC_PENDING;
255 		wakeup(&nmp->nm_rxstate);
256 	}
257 }
258