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/in6.h>
15 #include <linux/sunrpc/types.h>
16 #include <linux/sunrpc/xdr.h>
17 #include <linux/sunrpc/auth.h>
18 #include <linux/sunrpc/svcauth.h>
19 #include <linux/wait.h>
20 #include <linux/mm.h>
21
22 /*
23 * This is the RPC server thread function prototype
24 */
25 typedef int (*svc_thread_fn)(void *);
26
27 /* statistics for svc_pool structures */
28 struct svc_pool_stats {
29 unsigned long packets;
30 unsigned long sockets_queued;
31 unsigned long threads_woken;
32 unsigned long overloads_avoided;
33 unsigned long threads_timedout;
34 };
35
36 /*
37 *
38 * RPC service thread pool.
39 *
40 * Pool of threads and temporary sockets. Generally there is only
41 * a single one of these per RPC service, but on NUMA machines those
42 * services that can benefit from it (i.e. nfs but not lockd) will
43 * have one pool per NUMA node. This optimisation reduces cross-
44 * node traffic on multi-node NUMA NFS servers.
45 */
46 struct svc_pool {
47 unsigned int sp_id; /* pool id; also node id on NUMA */
48 spinlock_t sp_lock; /* protects all fields */
49 struct list_head sp_threads; /* idle server threads */
50 struct list_head sp_sockets; /* pending sockets */
51 unsigned int sp_nrthreads; /* # of threads in pool */
52 struct list_head sp_all_threads; /* all server threads */
53 int sp_nwaking; /* number of threads woken but not yet active */
54 struct svc_pool_stats sp_stats; /* statistics on pool operation */
55 } ____cacheline_aligned_in_smp;
56
57 /*
58 * RPC service.
59 *
60 * An RPC service is a ``daemon,'' possibly multithreaded, which
61 * receives and processes incoming RPC messages.
62 * It has one or more transport sockets associated with it, and maintains
63 * a list of idle threads waiting for input.
64 *
65 * We currently do not support more than one RPC program per daemon.
66 */
67 struct svc_serv {
68 struct svc_program * sv_program; /* RPC program */
69 struct svc_stat * sv_stats; /* RPC statistics */
70 spinlock_t sv_lock;
71 unsigned int sv_nrthreads; /* # of server threads */
72 unsigned int sv_maxconn; /* max connections allowed or
73 * '' causing max to be based
74 * on number of threads. */
75
76 unsigned int sv_max_payload; /* datagram payload size */
77 unsigned int sv_max_mesg; /* max_payload + 1 page for overheads */
78 unsigned int sv_xdrsize; /* XDR buffer size */
79 struct list_head sv_permsocks; /* all permanent sockets */
80 struct list_head sv_tempsocks; /* all temporary sockets */
81 int sv_tmpcnt; /* count of temporary sockets */
82 struct timer_list sv_temptimer; /* timer for aging temporary sockets */
83
84 char * sv_name; /* service name */
85
86 unsigned int sv_nrpools; /* number of thread pools */
87 struct svc_pool * sv_pools; /* array of thread pools */
88
89 void (*sv_shutdown)(struct svc_serv *serv);
90 /* Callback to use when last thread
91 * exits.
92 */
93
94 struct module * sv_module; /* optional module to count when
95 * adding threads */
96 svc_thread_fn sv_function; /* main function for threads */
97 #if defined(CONFIG_NFS_V4_1)
98 struct list_head sv_cb_list; /* queue for callback requests
99 * that arrive over the same
100 * connection */
101 spinlock_t sv_cb_lock; /* protects the svc_cb_list */
102 wait_queue_head_t sv_cb_waitq; /* sleep here if there are no
103 * entries in the svc_cb_list */
104 struct svc_xprt *bc_xprt;
105 #endif /* CONFIG_NFS_V4_1 */
106 };
107
108 /*
109 * We use sv_nrthreads as a reference count. svc_destroy() drops
110 * this refcount, so we need to bump it up around operations that
111 * change the number of threads. Horrible, but there it is.
112 * Should be called with the BKL held.
113 */
114 static inline void svc_get(struct svc_serv *serv)
115 {
116 serv->sv_nrthreads++;
117 }
118
119 /*
120 * Maximum payload size supported by a kernel RPC server.
121 * This is use to determine the max number of pages nfsd is
122 * willing to return in a single READ operation.
123 *
124 * These happen to all be powers of 2, which is not strictly
125 * necessary but helps enforce the real limitation, which is
126 * that they should be multiples of PAGE_CACHE_SIZE.
127 *
128 * For UDP transports, a block plus NFS,RPC, and UDP headers
129 * has to fit into the IP datagram limit of 64K. The largest
130 * feasible number for all known page sizes is probably 48K,
131 * but we choose 32K here. This is the same as the historical
132 * Linux limit; someone who cares more about NFS/UDP performance
133 * can test a larger number.
134 *
135 * For TCP transports we have more freedom. A size of 1MB is
136 * chosen to match the client limit. Other OSes are known to
137 * have larger limits, but those numbers are probably beyond
138 * the point of diminishing returns.
139 */
140 #define RPCSVC_MAXPAYLOAD (1*1024*1024u)
141 #define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD
142 #define RPCSVC_MAXPAYLOAD_UDP (32*1024u)
143
144 extern u32 svc_max_payload(const struct svc_rqst *rqstp);
145
146 /*
147 * RPC Requsts and replies are stored in one or more pages.
148 * We maintain an array of pages for each server thread.
149 * Requests are copied into these pages as they arrive. Remaining
150 * pages are available to write the reply into.
151 *
152 * Pages are sent using ->sendpage so each server thread needs to
153 * allocate more to replace those used in sending. To help keep track
154 * of these pages we have a receive list where all pages initialy live,
155 * and a send list where pages are moved to when there are to be part
156 * of a reply.
157 *
158 * We use xdr_buf for holding responses as it fits well with NFS
159 * read responses (that have a header, and some data pages, and possibly
160 * a tail) and means we can share some client side routines.
161 *
162 * The xdr_buf.head kvec always points to the first page in the rq_*pages
163 * list. The xdr_buf.pages pointer points to the second page on that
164 * list. xdr_buf.tail points to the end of the first page.
165 * This assumes that the non-page part of an rpc reply will fit
166 * in a page - NFSd ensures this. lockd also has no trouble.
167 *
168 * Each request/reply pair can have at most one "payload", plus two pages,
169 * one for the request, and one for the reply.
170 * We using ->sendfile to return read data, we might need one extra page
171 * if the request is not page-aligned. So add another '1'.
172 */
173 #define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
174 + 2 + 1)
175
176 static inline u32 svc_getnl(struct kvec *iov)
177 {
178 __be32 val, *vp;
179 vp = iov->iov_base;
180 val = *vp++;
181 iov->iov_base = (void*)vp;
182 iov->iov_len -= sizeof(__be32);
183 return ntohl(val);
184 }
185
186 static inline void svc_putnl(struct kvec *iov, u32 val)
187 {
188 __be32 *vp = iov->iov_base + iov->iov_len;
189 *vp = htonl(val);
190 iov->iov_len += sizeof(__be32);
191 }
192
193 static inline __be32 svc_getu32(struct kvec *iov)
194 {
195 __be32 val, *vp;
196 vp = iov->iov_base;
197 val = *vp++;
198 iov->iov_base = (void*)vp;
199 iov->iov_len -= sizeof(__be32);
200 return val;
201 }
202
203 static inline void svc_ungetu32(struct kvec *iov)
204 {
205 __be32 *vp = (__be32 *)iov->iov_base;
206 iov->iov_base = (void *)(vp - 1);
207 iov->iov_len += sizeof(*vp);
208 }
209
210 static inline void svc_putu32(struct kvec *iov, __be32 val)
211 {
212 __be32 *vp = iov->iov_base + iov->iov_len;
213 *vp = val;
214 iov->iov_len += sizeof(__be32);
215 }
216
217 union svc_addr_u {
218 struct in_addr addr;
219 struct in6_addr addr6;
220 };
221
222 /*
223 * The context of a single thread, including the request currently being
224 * processed.
225 */
226 struct svc_rqst {
227 struct list_head rq_list; /* idle list */
228 struct list_head rq_all; /* all threads list */
229 struct svc_xprt * rq_xprt; /* transport ptr */
230 struct sockaddr_storage rq_addr; /* peer address */
231 size_t rq_addrlen;
232
233 struct svc_serv * rq_server; /* RPC service definition */
234 struct svc_pool * rq_pool; /* thread pool */
235 struct svc_procedure * rq_procinfo; /* procedure info */
236 struct auth_ops * rq_authop; /* authentication flavour */
237 u32 rq_flavor; /* pseudoflavor */
238 struct svc_cred rq_cred; /* auth info */
239 void * rq_xprt_ctxt; /* transport specific context ptr */
240 struct svc_deferred_req*rq_deferred; /* deferred request we are replaying */
241 int rq_usedeferral; /* use deferral */
242
243 size_t rq_xprt_hlen; /* xprt header len */
244 struct xdr_buf rq_arg;
245 struct xdr_buf rq_res;
246 struct page * rq_pages[RPCSVC_MAXPAGES];
247 struct page * *rq_respages; /* points into rq_pages */
248 int rq_resused; /* number of pages used for result */
249
250 struct kvec rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
251
252 __be32 rq_xid; /* transmission id */
253 u32 rq_prog; /* program number */
254 u32 rq_vers; /* program version */
255 u32 rq_proc; /* procedure number */
256 u32 rq_prot; /* IP protocol */
257 unsigned short
258 rq_secure : 1; /* secure port */
259
260 union svc_addr_u rq_daddr; /* dest addr of request
261 * - reply from here */
262
263 void * rq_argp; /* decoded arguments */
264 void * rq_resp; /* xdr'd results */
265 void * rq_auth_data; /* flavor-specific data */
266
267 int rq_reserved; /* space on socket outq
268 * reserved for this request
269 */
270
271 struct cache_req rq_chandle; /* handle passed to caches for
272 * request delaying
273 */
274 /* Catering to nfsd */
275 struct auth_domain * rq_client; /* RPC peer info */
276 struct auth_domain * rq_gssclient; /* "gss/"-style peer info */
277 struct svc_cacherep * rq_cacherep; /* cache info */
278 struct knfsd_fh * rq_reffh; /* Referrence filehandle, used to
279 * determine what device number
280 * to report (real or virtual)
281 */
282 int rq_splice_ok; /* turned off in gss privacy
283 * to prevent encrypting page
284 * cache pages */
285 wait_queue_head_t rq_wait; /* synchronization */
286 struct task_struct *rq_task; /* service thread */
287 int rq_waking; /* 1 if thread is being woken */
288 };
289
290 /*
291 * Rigorous type checking on sockaddr type conversions
292 */
293 static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst)
294 {
295 return (struct sockaddr_in *) &rqst->rq_addr;
296 }
297
298 static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst)
299 {
300 return (struct sockaddr_in6 *) &rqst->rq_addr;
301 }
302
303 static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst)
304 {
305 return (struct sockaddr *) &rqst->rq_addr;
306 }
307
308 /*
309 * Check buffer bounds after decoding arguments
310 */
311 static inline int
312 xdr_argsize_check(struct svc_rqst *rqstp, __be32 *p)
313 {
314 char *cp = (char *)p;
315 struct kvec *vec = &rqstp->rq_arg.head[0];
316 return cp >= (char*)vec->iov_base
317 && cp <= (char*)vec->iov_base + vec->iov_len;
318 }
319
320 static inline int
321 xdr_ressize_check(struct svc_rqst *rqstp, __be32 *p)
322 {
323 struct kvec *vec = &rqstp->rq_res.head[0];
324 char *cp = (char*)p;
325
326 vec->iov_len = cp - (char*)vec->iov_base;
327
328 return vec->iov_len <= PAGE_SIZE;
329 }
330
331 static inline void svc_free_res_pages(struct svc_rqst *rqstp)
332 {
333 while (rqstp->rq_resused) {
334 struct page **pp = (rqstp->rq_respages +
335 --rqstp->rq_resused);
336 if (*pp) {
337 put_page(*pp);
338 *pp = NULL;
339 }
340 }
341 }
342
343 struct svc_deferred_req {
344 u32 prot; /* protocol (UDP or TCP) */
345 struct svc_xprt *xprt;
346 struct sockaddr_storage addr; /* where reply must go */
347 size_t addrlen;
348 union svc_addr_u daddr; /* where reply must come from */
349 struct cache_deferred_req handle;
350 size_t xprt_hlen;
351 int argslen;
352 __be32 args[0];
353 };
354
355 /*
356 * List of RPC programs on the same transport endpoint
357 */
358 struct svc_program {
359 struct svc_program * pg_next; /* other programs (same xprt) */
360 u32 pg_prog; /* program number */
361 unsigned int pg_lovers; /* lowest version */
362 unsigned int pg_hivers; /* lowest version */
363 unsigned int pg_nvers; /* number of versions */
364 struct svc_version ** pg_vers; /* version array */
365 char * pg_name; /* service name */
366 char * pg_class; /* class name: services sharing authentication */
367 struct svc_stat * pg_stats; /* rpc statistics */
368 int (*pg_authenticate)(struct svc_rqst *);
369 };
370
371 /*
372 * RPC program version
373 */
374 struct svc_version {
375 u32 vs_vers; /* version number */
376 u32 vs_nproc; /* number of procedures */
377 struct svc_procedure * vs_proc; /* per-procedure info */
378 u32 vs_xdrsize; /* xdrsize needed for this version */
379
380 unsigned int vs_hidden : 1; /* Don't register with portmapper.
381 * Only used for nfsacl so far. */
382
383 /* Override dispatch function (e.g. when caching replies).
384 * A return value of 0 means drop the request.
385 * vs_dispatch == NULL means use default dispatcher.
386 */
387 int (*vs_dispatch)(struct svc_rqst *, __be32 *);
388 };
389
390 /*
391 * RPC procedure info
392 */
393 typedef __be32 (*svc_procfunc)(struct svc_rqst *, void *argp, void *resp);
394 struct svc_procedure {
395 svc_procfunc pc_func; /* process the request */
396 kxdrproc_t pc_decode; /* XDR decode args */
397 kxdrproc_t pc_encode; /* XDR encode result */
398 kxdrproc_t pc_release; /* XDR free result */
399 unsigned int pc_argsize; /* argument struct size */
400 unsigned int pc_ressize; /* result struct size */
401 unsigned int pc_count; /* call count */
402 unsigned int pc_cachetype; /* cache info (NFS) */
403 unsigned int pc_xdrressize; /* maximum size of XDR reply */
404 };
405
406 /*
407 * Function prototypes.
408 */
409 struct svc_serv *svc_create(struct svc_program *, unsigned int,
410 void (*shutdown)(struct svc_serv *));
411 struct svc_rqst *svc_prepare_thread(struct svc_serv *serv,
412 struct svc_pool *pool);
413 void svc_exit_thread(struct svc_rqst *);
414 struct svc_serv * svc_create_pooled(struct svc_program *, unsigned int,
415 void (*shutdown)(struct svc_serv *),
416 svc_thread_fn, struct module *);
417 int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
418 int svc_pool_stats_open(struct svc_serv *serv, struct file *file);
419 void svc_destroy(struct svc_serv *);
420 int svc_process(struct svc_rqst *);
421 int bc_svc_process(struct svc_serv *, struct rpc_rqst *,
422 struct svc_rqst *);
423 int svc_register(const struct svc_serv *, const int,
424 const unsigned short, const unsigned short);
425
426 void svc_wake_up(struct svc_serv *);
427 void svc_reserve(struct svc_rqst *rqstp, int space);
428 struct svc_pool * svc_pool_for_cpu(struct svc_serv *serv, int cpu);
429 char * svc_print_addr(struct svc_rqst *, char *, size_t);
430
431 #define RPC_MAX_ADDRBUFLEN (63U)
432
433 /*
434 * When we want to reduce the size of the reserved space in the response
435 * buffer, we need to take into account the size of any checksum data that
436 * may be at the end of the packet. This is difficult to determine exactly
437 * for all cases without actually generating the checksum, so we just use a
438 * static value.
439 */
440 static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space)
441 {
442 int added_space = 0;
443
444 if (rqstp->rq_authop->flavour)
445 added_space = RPC_MAX_AUTH_SIZE;
446 svc_reserve(rqstp, space + added_space);
447 }
448
449 #endif /* SUNRPC_SVC_H */
450
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.