blob: 96c1b6ae7d96db95bfc5299f70545cec3c5fd571 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/include/linux/sunrpc/svc.h
3 *
4 * RPC server declarations.
5 *
6 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
7 */
8
9
10#ifndef SUNRPC_SVC_H
11#define SUNRPC_SVC_H
12
13#include <linux/in.h>
14#include <linux/sunrpc/types.h>
15#include <linux/sunrpc/xdr.h>
Greg Banks7adae482006-10-04 02:15:47 -070016#include <linux/sunrpc/auth.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/sunrpc/svcauth.h>
18#include <linux/wait.h>
19#include <linux/mm.h>
20
Greg Banksa7455442006-10-02 02:17:59 -070021/*
22 * This is the RPC server thread function prototype
23 */
24typedef void (*svc_thread_fn)(struct svc_rqst *);
Greg Banks3262c812006-10-02 02:17:58 -070025
26/*
27 *
28 * RPC service thread pool.
29 *
30 * Pool of threads and temporary sockets. Generally there is only
31 * a single one of these per RPC service, but on NUMA machines those
32 * services that can benefit from it (i.e. nfs but not lockd) will
33 * have one pool per NUMA node. This optimisation reduces cross-
34 * node traffic on multi-node NUMA NFS servers.
35 */
36struct svc_pool {
37 unsigned int sp_id; /* pool id; also node id on NUMA */
38 spinlock_t sp_lock; /* protects all fields */
39 struct list_head sp_threads; /* idle server threads */
40 struct list_head sp_sockets; /* pending sockets */
41 unsigned int sp_nrthreads; /* # of threads in pool */
Greg Banksa7455442006-10-02 02:17:59 -070042 struct list_head sp_all_threads; /* all server threads */
Greg Banks3262c812006-10-02 02:17:58 -070043} ____cacheline_aligned_in_smp;
44
Linus Torvalds1da177e2005-04-16 15:20:36 -070045/*
46 * RPC service.
47 *
48 * An RPC service is a ``daemon,'' possibly multithreaded, which
49 * receives and processes incoming RPC messages.
50 * It has one or more transport sockets associated with it, and maintains
51 * a list of idle threads waiting for input.
52 *
53 * We currently do not support more than one RPC program per daemon.
54 */
55struct svc_serv {
Linus Torvalds1da177e2005-04-16 15:20:36 -070056 struct svc_program * sv_program; /* RPC program */
57 struct svc_stat * sv_stats; /* RPC statistics */
58 spinlock_t sv_lock;
59 unsigned int sv_nrthreads; /* # of server threads */
NeilBrownc6b0a9f2006-10-06 00:44:05 -070060 unsigned int sv_max_payload; /* datagram payload size */
61 unsigned int sv_max_mesg; /* max_payload + 1 page for overheads */
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 unsigned int sv_xdrsize; /* XDR buffer size */
63
64 struct list_head sv_permsocks; /* all permanent sockets */
65 struct list_head sv_tempsocks; /* all temporary sockets */
66 int sv_tmpcnt; /* count of temporary sockets */
Greg Banks36bdfc82006-10-02 02:17:54 -070067 struct timer_list sv_temptimer; /* timer for aging temporary sockets */
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
69 char * sv_name; /* service name */
NeilBrownbc591cc2006-10-02 02:17:44 -070070
Greg Banks3262c812006-10-02 02:17:58 -070071 unsigned int sv_nrpools; /* number of thread pools */
72 struct svc_pool * sv_pools; /* array of thread pools */
73
NeilBrownbc591cc2006-10-02 02:17:44 -070074 void (*sv_shutdown)(struct svc_serv *serv);
75 /* Callback to use when last thread
76 * exits.
77 */
Greg Banksa7455442006-10-02 02:17:59 -070078
79 struct module * sv_module; /* optional module to count when
80 * adding threads */
81 svc_thread_fn sv_function; /* main function for threads */
82 int sv_kill_signal; /* signal to kill threads */
Linus Torvalds1da177e2005-04-16 15:20:36 -070083};
84
85/*
Greg Banks9a24ab52006-10-02 02:17:58 -070086 * We use sv_nrthreads as a reference count. svc_destroy() drops
87 * this refcount, so we need to bump it up around operations that
88 * change the number of threads. Horrible, but there it is.
89 * Should be called with the BKL held.
90 */
91static inline void svc_get(struct svc_serv *serv)
92{
93 serv->sv_nrthreads++;
94}
95
96/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 * Maximum payload size supported by a kernel RPC server.
98 * This is use to determine the max number of pages nfsd is
99 * willing to return in a single READ operation.
Greg Banks7adae482006-10-04 02:15:47 -0700100 *
101 * These happen to all be powers of 2, which is not strictly
102 * necessary but helps enforce the real limitation, which is
103 * that they should be multiples of PAGE_CACHE_SIZE.
104 *
105 * For UDP transports, a block plus NFS,RPC, and UDP headers
106 * has to fit into the IP datagram limit of 64K. The largest
107 * feasible number for all known page sizes is probably 48K,
108 * but we choose 32K here. This is the same as the historical
109 * Linux limit; someone who cares more about NFS/UDP performance
110 * can test a larger number.
111 *
112 * For TCP transports we have more freedom. A size of 1MB is
113 * chosen to match the client limit. Other OSes are known to
114 * have larger limits, but those numbers are probably beyond
115 * the point of diminishing returns.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 */
Greg Banks7adae482006-10-04 02:15:47 -0700117#define RPCSVC_MAXPAYLOAD (1*1024*1024u)
118#define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD
119#define RPCSVC_MAXPAYLOAD_UDP (32*1024u)
120
121extern u32 svc_max_payload(const struct svc_rqst *rqstp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
123/*
124 * RPC Requsts and replies are stored in one or more pages.
125 * We maintain an array of pages for each server thread.
126 * Requests are copied into these pages as they arrive. Remaining
127 * pages are available to write the reply into.
128 *
129 * Pages are sent using ->sendpage so each server thread needs to
130 * allocate more to replace those used in sending. To help keep track
131 * of these pages we have a receive list where all pages initialy live,
132 * and a send list where pages are moved to when there are to be part
133 * of a reply.
134 *
135 * We use xdr_buf for holding responses as it fits well with NFS
136 * read responses (that have a header, and some data pages, and possibly
137 * a tail) and means we can share some client side routines.
138 *
139 * The xdr_buf.head kvec always points to the first page in the rq_*pages
140 * list. The xdr_buf.pages pointer points to the second page on that
141 * list. xdr_buf.tail points to the end of the first page.
142 * This assumes that the non-page part of an rpc reply will fit
143 * in a page - NFSd ensures this. lockd also has no trouble.
144 *
145 * Each request/reply pair can have at most one "payload", plus two pages,
146 * one for the request, and one for the reply.
NeilBrown250f3912007-01-26 00:56:59 -0800147 * We using ->sendfile to return read data, we might need one extra page
148 * if the request is not page-aligned. So add another '1'.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 */
NeilBrown250f3912007-01-26 00:56:59 -0800150#define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
151 + 2 + 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
Alexey Dobriyan76994312006-09-26 22:28:46 -0700153static inline u32 svc_getnl(struct kvec *iov)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154{
Alexey Dobriyan76994312006-09-26 22:28:46 -0700155 __be32 val, *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 vp = iov->iov_base;
157 val = *vp++;
158 iov->iov_base = (void*)vp;
Alexey Dobriyan76994312006-09-26 22:28:46 -0700159 iov->iov_len -= sizeof(__be32);
160 return ntohl(val);
161}
162
163static inline void svc_putnl(struct kvec *iov, u32 val)
164{
165 __be32 *vp = iov->iov_base + iov->iov_len;
166 *vp = htonl(val);
167 iov->iov_len += sizeof(__be32);
168}
169
170static inline __be32 svc_getu32(struct kvec *iov)
171{
172 __be32 val, *vp;
173 vp = iov->iov_base;
174 val = *vp++;
175 iov->iov_base = (void*)vp;
176 iov->iov_len -= sizeof(__be32);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 return val;
178}
179
180static inline void svc_ungetu32(struct kvec *iov)
181{
Alexey Dobriyan76994312006-09-26 22:28:46 -0700182 __be32 *vp = (__be32 *)iov->iov_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 iov->iov_base = (void *)(vp - 1);
184 iov->iov_len += sizeof(*vp);
185}
186
Alexey Dobriyan76994312006-09-26 22:28:46 -0700187static inline void svc_putu32(struct kvec *iov, __be32 val)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188{
Alexey Dobriyan76994312006-09-26 22:28:46 -0700189 __be32 *vp = iov->iov_base + iov->iov_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 *vp = val;
Alexey Dobriyan76994312006-09-26 22:28:46 -0700191 iov->iov_len += sizeof(__be32);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192}
193
194
195/*
196 * The context of a single thread, including the request currently being
197 * processed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 */
199struct svc_rqst {
200 struct list_head rq_list; /* idle list */
Greg Banksa7455442006-10-02 02:17:59 -0700201 struct list_head rq_all; /* all threads list */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 struct svc_sock * rq_sock; /* socket */
Chuck Lever27459f02007-02-12 00:53:34 -0800203 struct sockaddr_storage rq_addr; /* peer address */
204 size_t rq_addrlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205
206 struct svc_serv * rq_server; /* RPC service definition */
Greg Banks3262c812006-10-02 02:17:58 -0700207 struct svc_pool * rq_pool; /* thread pool */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 struct svc_procedure * rq_procinfo; /* procedure info */
209 struct auth_ops * rq_authop; /* authentication flavour */
210 struct svc_cred rq_cred; /* auth info */
211 struct sk_buff * rq_skbuff; /* fast recv inet buffer */
212 struct svc_deferred_req*rq_deferred; /* deferred request we are replaying */
213
214 struct xdr_buf rq_arg;
215 struct xdr_buf rq_res;
NeilBrown44524352006-10-04 02:15:46 -0700216 struct page * rq_pages[RPCSVC_MAXPAGES];
217 struct page * *rq_respages; /* points into rq_pages */
218 int rq_resused; /* number of pages used for result */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
NeilBrown3cc03b12006-10-04 02:15:47 -0700220 struct kvec rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
221
Alexey Dobriyand8ed0292006-09-26 22:29:38 -0700222 __be32 rq_xid; /* transmission id */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 u32 rq_prog; /* program number */
224 u32 rq_vers; /* program version */
225 u32 rq_proc; /* procedure number */
226 u32 rq_prot; /* IP protocol */
227 unsigned short
228 rq_secure : 1; /* secure port */
229
230
Alexey Dobriyand8ed0292006-09-26 22:29:38 -0700231 __be32 rq_daddr; /* dest addr of request - reply from here */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
233 void * rq_argp; /* decoded arguments */
234 void * rq_resp; /* xdr'd results */
235 void * rq_auth_data; /* flavor-specific data */
236
237 int rq_reserved; /* space on socket outq
238 * reserved for this request
239 */
240
241 struct cache_req rq_chandle; /* handle passed to caches for
242 * request delaying
243 */
244 /* Catering to nfsd */
245 struct auth_domain * rq_client; /* RPC peer info */
246 struct svc_cacherep * rq_cacherep; /* cache info */
247 struct knfsd_fh * rq_reffh; /* Referrence filehandle, used to
248 * determine what device number
249 * to report (real or virtual)
250 */
J. Bruce Fields5c04c462006-06-30 01:56:19 -0700251 int rq_sendfile_ok; /* turned off in gss privacy
252 * to prevent encrypting page
253 * cache pages */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 wait_queue_head_t rq_wait; /* synchronization */
Greg Banksa7455442006-10-02 02:17:59 -0700255 struct task_struct *rq_task; /* service thread */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256};
257
258/*
Chuck Lever27459f02007-02-12 00:53:34 -0800259 * Rigorous type checking on sockaddr type conversions
260 */
261static inline struct sockaddr_in *svc_addr_in(struct svc_rqst *rqst)
262{
263 return (struct sockaddr_in *) &rqst->rq_addr;
264}
265
266static inline struct sockaddr *svc_addr(struct svc_rqst *rqst)
267{
268 return (struct sockaddr *) &rqst->rq_addr;
269}
270
271/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 * Check buffer bounds after decoding arguments
273 */
274static inline int
Alexey Dobriyand8ed0292006-09-26 22:29:38 -0700275xdr_argsize_check(struct svc_rqst *rqstp, __be32 *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276{
277 char *cp = (char *)p;
278 struct kvec *vec = &rqstp->rq_arg.head[0];
NeilBrown0ba75362005-11-07 01:00:26 -0800279 return cp >= (char*)vec->iov_base
280 && cp <= (char*)vec->iov_base + vec->iov_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281}
282
283static inline int
Alexey Dobriyand8ed0292006-09-26 22:29:38 -0700284xdr_ressize_check(struct svc_rqst *rqstp, __be32 *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285{
286 struct kvec *vec = &rqstp->rq_res.head[0];
287 char *cp = (char*)p;
288
289 vec->iov_len = cp - (char*)vec->iov_base;
290
291 return vec->iov_len <= PAGE_SIZE;
292}
293
NeilBrown44524352006-10-04 02:15:46 -0700294static inline void svc_free_res_pages(struct svc_rqst *rqstp)
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +0000295{
NeilBrown44524352006-10-04 02:15:46 -0700296 while (rqstp->rq_resused) {
297 struct page **pp = (rqstp->rq_respages +
298 --rqstp->rq_resused);
299 if (*pp) {
300 put_page(*pp);
301 *pp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 }
303 }
304}
305
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306struct svc_deferred_req {
307 u32 prot; /* protocol (UDP or TCP) */
Chuck Lever24422222007-02-12 00:53:33 -0800308 struct svc_sock *svsk;
309 struct sockaddr_storage addr; /* where reply must go */
310 size_t addrlen;
Alexey Dobriyand8ed0292006-09-26 22:29:38 -0700311 __be32 daddr; /* where reply must come from */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 struct cache_deferred_req handle;
313 int argslen;
Alexey Dobriyand8ed0292006-09-26 22:29:38 -0700314 __be32 args[0];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315};
316
317/*
Andreas Gruenbacher9ba02632005-06-22 17:16:24 +0000318 * List of RPC programs on the same transport endpoint
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 */
320struct svc_program {
Andreas Gruenbacher9ba02632005-06-22 17:16:24 +0000321 struct svc_program * pg_next; /* other programs (same xprt) */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 u32 pg_prog; /* program number */
323 unsigned int pg_lovers; /* lowest version */
324 unsigned int pg_hivers; /* lowest version */
325 unsigned int pg_nvers; /* number of versions */
326 struct svc_version ** pg_vers; /* version array */
327 char * pg_name; /* service name */
328 char * pg_class; /* class name: services sharing authentication */
329 struct svc_stat * pg_stats; /* rpc statistics */
330 int (*pg_authenticate)(struct svc_rqst *);
331};
332
333/*
334 * RPC program version
335 */
336struct svc_version {
337 u32 vs_vers; /* version number */
338 u32 vs_nproc; /* number of procedures */
339 struct svc_procedure * vs_proc; /* per-procedure info */
340 u32 vs_xdrsize; /* xdrsize needed for this version */
341
Olaf Kirchbc5fea42006-10-04 02:16:05 -0700342 unsigned int vs_hidden : 1; /* Don't register with portmapper.
343 * Only used for nfsacl so far. */
344
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 /* Override dispatch function (e.g. when caching replies).
346 * A return value of 0 means drop the request.
347 * vs_dispatch == NULL means use default dispatcher.
348 */
Alexey Dobriyand8ed0292006-09-26 22:29:38 -0700349 int (*vs_dispatch)(struct svc_rqst *, __be32 *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350};
351
352/*
353 * RPC procedure info
354 */
Al Viro7111c662006-10-19 23:28:45 -0700355typedef __be32 (*svc_procfunc)(struct svc_rqst *, void *argp, void *resp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356struct svc_procedure {
357 svc_procfunc pc_func; /* process the request */
358 kxdrproc_t pc_decode; /* XDR decode args */
359 kxdrproc_t pc_encode; /* XDR encode result */
360 kxdrproc_t pc_release; /* XDR free result */
361 unsigned int pc_argsize; /* argument struct size */
362 unsigned int pc_ressize; /* result struct size */
363 unsigned int pc_count; /* call count */
364 unsigned int pc_cachetype; /* cache info (NFS) */
365 unsigned int pc_xdrressize; /* maximum size of XDR reply */
366};
367
368/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 * Function prototypes.
370 */
NeilBrownbc591cc2006-10-02 02:17:44 -0700371struct svc_serv * svc_create(struct svc_program *, unsigned int,
372 void (*shutdown)(struct svc_serv*));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373int svc_create_thread(svc_thread_fn, struct svc_serv *);
374void svc_exit_thread(struct svc_rqst *);
Greg Banksa7455442006-10-02 02:17:59 -0700375struct svc_serv * svc_create_pooled(struct svc_program *, unsigned int,
376 void (*shutdown)(struct svc_serv*),
377 svc_thread_fn, int sig, struct module *);
378int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379void svc_destroy(struct svc_serv *);
NeilBrown6fb2b472006-10-02 02:17:50 -0700380int svc_process(struct svc_rqst *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381int svc_register(struct svc_serv *, int, unsigned short);
382void svc_wake_up(struct svc_serv *);
383void svc_reserve(struct svc_rqst *rqstp, int space);
Greg Banksbfd24162006-10-02 02:18:01 -0700384struct svc_pool * svc_pool_for_cpu(struct svc_serv *serv, int cpu);
Chuck Leverad06e4b2007-02-12 00:53:32 -0800385char * svc_print_addr(struct svc_rqst *, char *, size_t);
386
387#define RPC_MAX_ADDRBUFLEN (63U)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
389#endif /* SUNRPC_SVC_H */