]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - include/linux/sunrpc/svc.h
[PATCH] knfsd: Prepare knfsd for support of rsize/wsize of up to 1MB, over TCP
[mirror_ubuntu-artful-kernel.git] / include / linux / sunrpc / svc.h
1 /*
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>
16 #include <linux/sunrpc/auth.h>
17 #include <linux/sunrpc/svcauth.h>
18 #include <linux/wait.h>
19 #include <linux/mm.h>
20
21 /*
22 * This is the RPC server thread function prototype
23 */
24 typedef void (*svc_thread_fn)(struct svc_rqst *);
25
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 */
36 struct 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 */
42 struct list_head sp_all_threads; /* all server threads */
43 } ____cacheline_aligned_in_smp;
44
45 /*
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 */
55 struct svc_serv {
56 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 */
60 unsigned int sv_bufsz; /* datagram buffer size */
61 unsigned int sv_xdrsize; /* XDR buffer size */
62
63 struct list_head sv_permsocks; /* all permanent sockets */
64 struct list_head sv_tempsocks; /* all temporary sockets */
65 int sv_tmpcnt; /* count of temporary sockets */
66 struct timer_list sv_temptimer; /* timer for aging temporary sockets */
67
68 char * sv_name; /* service name */
69
70 unsigned int sv_nrpools; /* number of thread pools */
71 struct svc_pool * sv_pools; /* array of thread pools */
72
73 void (*sv_shutdown)(struct svc_serv *serv);
74 /* Callback to use when last thread
75 * exits.
76 */
77
78 struct module * sv_module; /* optional module to count when
79 * adding threads */
80 svc_thread_fn sv_function; /* main function for threads */
81 int sv_kill_signal; /* signal to kill threads */
82 };
83
84 /*
85 * We use sv_nrthreads as a reference count. svc_destroy() drops
86 * this refcount, so we need to bump it up around operations that
87 * change the number of threads. Horrible, but there it is.
88 * Should be called with the BKL held.
89 */
90 static inline void svc_get(struct svc_serv *serv)
91 {
92 serv->sv_nrthreads++;
93 }
94
95 /*
96 * Maximum payload size supported by a kernel RPC server.
97 * This is use to determine the max number of pages nfsd is
98 * willing to return in a single READ operation.
99 *
100 * These happen to all be powers of 2, which is not strictly
101 * necessary but helps enforce the real limitation, which is
102 * that they should be multiples of PAGE_CACHE_SIZE.
103 *
104 * For UDP transports, a block plus NFS,RPC, and UDP headers
105 * has to fit into the IP datagram limit of 64K. The largest
106 * feasible number for all known page sizes is probably 48K,
107 * but we choose 32K here. This is the same as the historical
108 * Linux limit; someone who cares more about NFS/UDP performance
109 * can test a larger number.
110 *
111 * For TCP transports we have more freedom. A size of 1MB is
112 * chosen to match the client limit. Other OSes are known to
113 * have larger limits, but those numbers are probably beyond
114 * the point of diminishing returns.
115 */
116 #define RPCSVC_MAXPAYLOAD (1*1024*1024u)
117 #define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD
118 #define RPCSVC_MAXPAYLOAD_UDP (32*1024u)
119
120 extern u32 svc_max_payload(const struct svc_rqst *rqstp);
121
122 /*
123 * RPC Requsts and replies are stored in one or more pages.
124 * We maintain an array of pages for each server thread.
125 * Requests are copied into these pages as they arrive. Remaining
126 * pages are available to write the reply into.
127 *
128 * Pages are sent using ->sendpage so each server thread needs to
129 * allocate more to replace those used in sending. To help keep track
130 * of these pages we have a receive list where all pages initialy live,
131 * and a send list where pages are moved to when there are to be part
132 * of a reply.
133 *
134 * We use xdr_buf for holding responses as it fits well with NFS
135 * read responses (that have a header, and some data pages, and possibly
136 * a tail) and means we can share some client side routines.
137 *
138 * The xdr_buf.head kvec always points to the first page in the rq_*pages
139 * list. The xdr_buf.pages pointer points to the second page on that
140 * list. xdr_buf.tail points to the end of the first page.
141 * This assumes that the non-page part of an rpc reply will fit
142 * in a page - NFSd ensures this. lockd also has no trouble.
143 *
144 * Each request/reply pair can have at most one "payload", plus two pages,
145 * one for the request, and one for the reply.
146 */
147 #define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE + 2)
148
149 static inline u32 svc_getnl(struct kvec *iov)
150 {
151 __be32 val, *vp;
152 vp = iov->iov_base;
153 val = *vp++;
154 iov->iov_base = (void*)vp;
155 iov->iov_len -= sizeof(__be32);
156 return ntohl(val);
157 }
158
159 static inline void svc_putnl(struct kvec *iov, u32 val)
160 {
161 __be32 *vp = iov->iov_base + iov->iov_len;
162 *vp = htonl(val);
163 iov->iov_len += sizeof(__be32);
164 }
165
166 static inline __be32 svc_getu32(struct kvec *iov)
167 {
168 __be32 val, *vp;
169 vp = iov->iov_base;
170 val = *vp++;
171 iov->iov_base = (void*)vp;
172 iov->iov_len -= sizeof(__be32);
173 return val;
174 }
175
176 static inline void svc_ungetu32(struct kvec *iov)
177 {
178 __be32 *vp = (__be32 *)iov->iov_base;
179 iov->iov_base = (void *)(vp - 1);
180 iov->iov_len += sizeof(*vp);
181 }
182
183 static inline void svc_putu32(struct kvec *iov, __be32 val)
184 {
185 __be32 *vp = iov->iov_base + iov->iov_len;
186 *vp = val;
187 iov->iov_len += sizeof(__be32);
188 }
189
190
191 /*
192 * The context of a single thread, including the request currently being
193 * processed.
194 */
195 struct svc_rqst {
196 struct list_head rq_list; /* idle list */
197 struct list_head rq_all; /* all threads list */
198 struct svc_sock * rq_sock; /* socket */
199 struct sockaddr_in rq_addr; /* peer address */
200 int rq_addrlen;
201
202 struct svc_serv * rq_server; /* RPC service definition */
203 struct svc_pool * rq_pool; /* thread pool */
204 struct svc_procedure * rq_procinfo; /* procedure info */
205 struct auth_ops * rq_authop; /* authentication flavour */
206 struct svc_cred rq_cred; /* auth info */
207 struct sk_buff * rq_skbuff; /* fast recv inet buffer */
208 struct svc_deferred_req*rq_deferred; /* deferred request we are replaying */
209
210 struct xdr_buf rq_arg;
211 struct xdr_buf rq_res;
212 struct page * rq_pages[RPCSVC_MAXPAGES];
213 struct page * *rq_respages; /* points into rq_pages */
214 int rq_resused; /* number of pages used for result */
215
216 struct kvec rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
217
218 __be32 rq_xid; /* transmission id */
219 u32 rq_prog; /* program number */
220 u32 rq_vers; /* program version */
221 u32 rq_proc; /* procedure number */
222 u32 rq_prot; /* IP protocol */
223 unsigned short
224 rq_secure : 1; /* secure port */
225
226
227 __be32 rq_daddr; /* dest addr of request - reply from here */
228
229 void * rq_argp; /* decoded arguments */
230 void * rq_resp; /* xdr'd results */
231 void * rq_auth_data; /* flavor-specific data */
232
233 int rq_reserved; /* space on socket outq
234 * reserved for this request
235 */
236
237 struct cache_req rq_chandle; /* handle passed to caches for
238 * request delaying
239 */
240 /* Catering to nfsd */
241 struct auth_domain * rq_client; /* RPC peer info */
242 struct svc_cacherep * rq_cacherep; /* cache info */
243 struct knfsd_fh * rq_reffh; /* Referrence filehandle, used to
244 * determine what device number
245 * to report (real or virtual)
246 */
247 int rq_sendfile_ok; /* turned off in gss privacy
248 * to prevent encrypting page
249 * cache pages */
250 wait_queue_head_t rq_wait; /* synchronization */
251 struct task_struct *rq_task; /* service thread */
252 };
253
254 /*
255 * Check buffer bounds after decoding arguments
256 */
257 static inline int
258 xdr_argsize_check(struct svc_rqst *rqstp, __be32 *p)
259 {
260 char *cp = (char *)p;
261 struct kvec *vec = &rqstp->rq_arg.head[0];
262 return cp >= (char*)vec->iov_base
263 && cp <= (char*)vec->iov_base + vec->iov_len;
264 }
265
266 static inline int
267 xdr_ressize_check(struct svc_rqst *rqstp, __be32 *p)
268 {
269 struct kvec *vec = &rqstp->rq_res.head[0];
270 char *cp = (char*)p;
271
272 vec->iov_len = cp - (char*)vec->iov_base;
273
274 return vec->iov_len <= PAGE_SIZE;
275 }
276
277 static inline void svc_free_res_pages(struct svc_rqst *rqstp)
278 {
279 while (rqstp->rq_resused) {
280 struct page **pp = (rqstp->rq_respages +
281 --rqstp->rq_resused);
282 if (*pp) {
283 put_page(*pp);
284 *pp = NULL;
285 }
286 }
287 }
288
289 struct svc_deferred_req {
290 u32 prot; /* protocol (UDP or TCP) */
291 struct sockaddr_in addr;
292 struct svc_sock *svsk; /* where reply must go */
293 __be32 daddr; /* where reply must come from */
294 struct cache_deferred_req handle;
295 int argslen;
296 __be32 args[0];
297 };
298
299 /*
300 * List of RPC programs on the same transport endpoint
301 */
302 struct svc_program {
303 struct svc_program * pg_next; /* other programs (same xprt) */
304 u32 pg_prog; /* program number */
305 unsigned int pg_lovers; /* lowest version */
306 unsigned int pg_hivers; /* lowest version */
307 unsigned int pg_nvers; /* number of versions */
308 struct svc_version ** pg_vers; /* version array */
309 char * pg_name; /* service name */
310 char * pg_class; /* class name: services sharing authentication */
311 struct svc_stat * pg_stats; /* rpc statistics */
312 int (*pg_authenticate)(struct svc_rqst *);
313 };
314
315 /*
316 * RPC program version
317 */
318 struct svc_version {
319 u32 vs_vers; /* version number */
320 u32 vs_nproc; /* number of procedures */
321 struct svc_procedure * vs_proc; /* per-procedure info */
322 u32 vs_xdrsize; /* xdrsize needed for this version */
323
324 /* Override dispatch function (e.g. when caching replies).
325 * A return value of 0 means drop the request.
326 * vs_dispatch == NULL means use default dispatcher.
327 */
328 int (*vs_dispatch)(struct svc_rqst *, __be32 *);
329 };
330
331 /*
332 * RPC procedure info
333 */
334 typedef int (*svc_procfunc)(struct svc_rqst *, void *argp, void *resp);
335 struct svc_procedure {
336 svc_procfunc pc_func; /* process the request */
337 kxdrproc_t pc_decode; /* XDR decode args */
338 kxdrproc_t pc_encode; /* XDR encode result */
339 kxdrproc_t pc_release; /* XDR free result */
340 unsigned int pc_argsize; /* argument struct size */
341 unsigned int pc_ressize; /* result struct size */
342 unsigned int pc_count; /* call count */
343 unsigned int pc_cachetype; /* cache info (NFS) */
344 unsigned int pc_xdrressize; /* maximum size of XDR reply */
345 };
346
347 /*
348 * Function prototypes.
349 */
350 struct svc_serv * svc_create(struct svc_program *, unsigned int,
351 void (*shutdown)(struct svc_serv*));
352 int svc_create_thread(svc_thread_fn, struct svc_serv *);
353 void svc_exit_thread(struct svc_rqst *);
354 struct svc_serv * svc_create_pooled(struct svc_program *, unsigned int,
355 void (*shutdown)(struct svc_serv*),
356 svc_thread_fn, int sig, struct module *);
357 int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
358 void svc_destroy(struct svc_serv *);
359 int svc_process(struct svc_rqst *);
360 int svc_register(struct svc_serv *, int, unsigned short);
361 void svc_wake_up(struct svc_serv *);
362 void svc_reserve(struct svc_rqst *rqstp, int space);
363 struct svc_pool * svc_pool_for_cpu(struct svc_serv *serv, int cpu);
364
365 #endif /* SUNRPC_SVC_H */