2 * linux/include/linux/sunrpc/svc.h
4 * RPC server declarations.
6 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
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>
23 * This is the RPC server thread function prototype
25 typedef int (*svc_thread_fn)(void *);
29 * RPC service thread pool.
31 * Pool of threads and temporary sockets. Generally there is only
32 * a single one of these per RPC service, but on NUMA machines those
33 * services that can benefit from it (i.e. nfs but not lockd) will
34 * have one pool per NUMA node. This optimisation reduces cross-
35 * node traffic on multi-node NUMA NFS servers.
38 unsigned int sp_id; /* pool id; also node id on NUMA */
39 spinlock_t sp_lock; /* protects all fields */
40 struct list_head sp_threads; /* idle server threads */
41 struct list_head sp_sockets; /* pending sockets */
42 unsigned int sp_nrthreads; /* # of threads in pool */
43 struct list_head sp_all_threads; /* all server threads */
44 } ____cacheline_aligned_in_smp;
49 * An RPC service is a ``daemon,'' possibly multithreaded, which
50 * receives and processes incoming RPC messages.
51 * It has one or more transport sockets associated with it, and maintains
52 * a list of idle threads waiting for input.
54 * We currently do not support more than one RPC program per daemon.
57 struct svc_program * sv_program; /* RPC program */
58 struct svc_stat * sv_stats; /* RPC statistics */
60 unsigned int sv_nrthreads; /* # of server threads */
61 unsigned int sv_maxconn; /* max connections allowed or
62 * '0' causing max to be based
63 * on number of threads. */
65 unsigned int sv_max_payload; /* datagram payload size */
66 unsigned int sv_max_mesg; /* max_payload + 1 page for overheads */
67 unsigned int sv_xdrsize; /* XDR buffer size */
68 struct list_head sv_permsocks; /* all permanent sockets */
69 struct list_head sv_tempsocks; /* all temporary sockets */
70 int sv_tmpcnt; /* count of temporary sockets */
71 struct timer_list sv_temptimer; /* timer for aging temporary sockets */
73 char * sv_name; /* service name */
75 unsigned int sv_nrpools; /* number of thread pools */
76 struct svc_pool * sv_pools; /* array of thread pools */
78 void (*sv_shutdown)(struct svc_serv *serv);
79 /* Callback to use when last thread
83 struct module * sv_module; /* optional module to count when
85 svc_thread_fn sv_function; /* main function for threads */
89 * We use sv_nrthreads as a reference count. svc_destroy() drops
90 * this refcount, so we need to bump it up around operations that
91 * change the number of threads. Horrible, but there it is.
92 * Should be called with the BKL held.
94 static inline void svc_get(struct svc_serv *serv)
100 * Maximum payload size supported by a kernel RPC server.
101 * This is use to determine the max number of pages nfsd is
102 * willing to return in a single READ operation.
104 * These happen to all be powers of 2, which is not strictly
105 * necessary but helps enforce the real limitation, which is
106 * that they should be multiples of PAGE_CACHE_SIZE.
108 * For UDP transports, a block plus NFS,RPC, and UDP headers
109 * has to fit into the IP datagram limit of 64K. The largest
110 * feasible number for all known page sizes is probably 48K,
111 * but we choose 32K here. This is the same as the historical
112 * Linux limit; someone who cares more about NFS/UDP performance
113 * can test a larger number.
115 * For TCP transports we have more freedom. A size of 1MB is
116 * chosen to match the client limit. Other OSes are known to
117 * have larger limits, but those numbers are probably beyond
118 * the point of diminishing returns.
120 #define RPCSVC_MAXPAYLOAD (1*1024*1024u)
121 #define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD
122 #define RPCSVC_MAXPAYLOAD_UDP (32*1024u)
124 extern u32 svc_max_payload(const struct svc_rqst *rqstp);
127 * RPC Requsts and replies are stored in one or more pages.
128 * We maintain an array of pages for each server thread.
129 * Requests are copied into these pages as they arrive. Remaining
130 * pages are available to write the reply into.
132 * Pages are sent using ->sendpage so each server thread needs to
133 * allocate more to replace those used in sending. To help keep track
134 * of these pages we have a receive list where all pages initialy live,
135 * and a send list where pages are moved to when there are to be part
138 * We use xdr_buf for holding responses as it fits well with NFS
139 * read responses (that have a header, and some data pages, and possibly
140 * a tail) and means we can share some client side routines.
142 * The xdr_buf.head kvec always points to the first page in the rq_*pages
143 * list. The xdr_buf.pages pointer points to the second page on that
144 * list. xdr_buf.tail points to the end of the first page.
145 * This assumes that the non-page part of an rpc reply will fit
146 * in a page - NFSd ensures this. lockd also has no trouble.
148 * Each request/reply pair can have at most one "payload", plus two pages,
149 * one for the request, and one for the reply.
150 * We using ->sendfile to return read data, we might need one extra page
151 * if the request is not page-aligned. So add another '1'.
153 #define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
156 static inline u32 svc_getnl(struct kvec *iov)
161 iov->iov_base = (void*)vp;
162 iov->iov_len -= sizeof(__be32);
166 static inline void svc_putnl(struct kvec *iov, u32 val)
168 __be32 *vp = iov->iov_base + iov->iov_len;
170 iov->iov_len += sizeof(__be32);
173 static inline __be32 svc_getu32(struct kvec *iov)
178 iov->iov_base = (void*)vp;
179 iov->iov_len -= sizeof(__be32);
183 static inline void svc_ungetu32(struct kvec *iov)
185 __be32 *vp = (__be32 *)iov->iov_base;
186 iov->iov_base = (void *)(vp - 1);
187 iov->iov_len += sizeof(*vp);
190 static inline void svc_putu32(struct kvec *iov, __be32 val)
192 __be32 *vp = iov->iov_base + iov->iov_len;
194 iov->iov_len += sizeof(__be32);
199 struct in6_addr addr6;
203 * The context of a single thread, including the request currently being
207 struct list_head rq_list; /* idle list */
208 struct list_head rq_all; /* all threads list */
209 struct svc_xprt * rq_xprt; /* transport ptr */
210 struct sockaddr_storage rq_addr; /* peer address */
213 struct svc_serv * rq_server; /* RPC service definition */
214 struct svc_pool * rq_pool; /* thread pool */
215 struct svc_procedure * rq_procinfo; /* procedure info */
216 struct auth_ops * rq_authop; /* authentication flavour */
217 u32 rq_flavor; /* pseudoflavor */
218 struct svc_cred rq_cred; /* auth info */
219 void * rq_xprt_ctxt; /* transport specific context ptr */
220 struct svc_deferred_req*rq_deferred; /* deferred request we are replaying */
222 size_t rq_xprt_hlen; /* xprt header len */
223 struct xdr_buf rq_arg;
224 struct xdr_buf rq_res;
225 struct page * rq_pages[RPCSVC_MAXPAGES];
226 struct page * *rq_respages; /* points into rq_pages */
227 int rq_resused; /* number of pages used for result */
229 struct kvec rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
231 __be32 rq_xid; /* transmission id */
232 u32 rq_prog; /* program number */
233 u32 rq_vers; /* program version */
234 u32 rq_proc; /* procedure number */
235 u32 rq_prot; /* IP protocol */
237 rq_secure : 1; /* secure port */
239 union svc_addr_u rq_daddr; /* dest addr of request
240 * - reply from here */
242 void * rq_argp; /* decoded arguments */
243 void * rq_resp; /* xdr'd results */
244 void * rq_auth_data; /* flavor-specific data */
246 int rq_reserved; /* space on socket outq
247 * reserved for this request
250 struct cache_req rq_chandle; /* handle passed to caches for
253 /* Catering to nfsd */
254 struct auth_domain * rq_client; /* RPC peer info */
255 struct auth_domain * rq_gssclient; /* "gss/"-style peer info */
256 struct svc_cacherep * rq_cacherep; /* cache info */
257 struct knfsd_fh * rq_reffh; /* Referrence filehandle, used to
258 * determine what device number
259 * to report (real or virtual)
261 int rq_splice_ok; /* turned off in gss privacy
262 * to prevent encrypting page
264 wait_queue_head_t rq_wait; /* synchronization */
265 struct task_struct *rq_task; /* service thread */
269 * Rigorous type checking on sockaddr type conversions
271 static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst)
273 return (struct sockaddr_in *) &rqst->rq_addr;
276 static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst)
278 return (struct sockaddr_in6 *) &rqst->rq_addr;
281 static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst)
283 return (struct sockaddr *) &rqst->rq_addr;
287 * Check buffer bounds after decoding arguments
290 xdr_argsize_check(struct svc_rqst *rqstp, __be32 *p)
292 char *cp = (char *)p;
293 struct kvec *vec = &rqstp->rq_arg.head[0];
294 return cp >= (char*)vec->iov_base
295 && cp <= (char*)vec->iov_base + vec->iov_len;
299 xdr_ressize_check(struct svc_rqst *rqstp, __be32 *p)
301 struct kvec *vec = &rqstp->rq_res.head[0];
304 vec->iov_len = cp - (char*)vec->iov_base;
306 return vec->iov_len <= PAGE_SIZE;
309 static inline void svc_free_res_pages(struct svc_rqst *rqstp)
311 while (rqstp->rq_resused) {
312 struct page **pp = (rqstp->rq_respages +
313 --rqstp->rq_resused);
321 struct svc_deferred_req {
322 u32 prot; /* protocol (UDP or TCP) */
323 struct svc_xprt *xprt;
324 struct sockaddr_storage addr; /* where reply must go */
326 union svc_addr_u daddr; /* where reply must come from */
327 struct cache_deferred_req handle;
334 * List of RPC programs on the same transport endpoint
337 struct svc_program * pg_next; /* other programs (same xprt) */
338 u32 pg_prog; /* program number */
339 unsigned int pg_lovers; /* lowest version */
340 unsigned int pg_hivers; /* lowest version */
341 unsigned int pg_nvers; /* number of versions */
342 struct svc_version ** pg_vers; /* version array */
343 char * pg_name; /* service name */
344 char * pg_class; /* class name: services sharing authentication */
345 struct svc_stat * pg_stats; /* rpc statistics */
346 int (*pg_authenticate)(struct svc_rqst *);
350 * RPC program version
353 u32 vs_vers; /* version number */
354 u32 vs_nproc; /* number of procedures */
355 struct svc_procedure * vs_proc; /* per-procedure info */
356 u32 vs_xdrsize; /* xdrsize needed for this version */
358 unsigned int vs_hidden : 1; /* Don't register with portmapper.
359 * Only used for nfsacl so far. */
361 /* Override dispatch function (e.g. when caching replies).
362 * A return value of 0 means drop the request.
363 * vs_dispatch == NULL means use default dispatcher.
365 int (*vs_dispatch)(struct svc_rqst *, __be32 *);
371 typedef __be32 (*svc_procfunc)(struct svc_rqst *, void *argp, void *resp);
372 struct svc_procedure {
373 svc_procfunc pc_func; /* process the request */
374 kxdrproc_t pc_decode; /* XDR decode args */
375 kxdrproc_t pc_encode; /* XDR encode result */
376 kxdrproc_t pc_release; /* XDR free result */
377 unsigned int pc_argsize; /* argument struct size */
378 unsigned int pc_ressize; /* result struct size */
379 unsigned int pc_count; /* call count */
380 unsigned int pc_cachetype; /* cache info (NFS) */
381 unsigned int pc_xdrressize; /* maximum size of XDR reply */
385 * Function prototypes.
387 struct svc_serv *svc_create(struct svc_program *, unsigned int,
388 void (*shutdown)(struct svc_serv *));
389 struct svc_rqst *svc_prepare_thread(struct svc_serv *serv,
390 struct svc_pool *pool);
391 void svc_exit_thread(struct svc_rqst *);
392 struct svc_serv * svc_create_pooled(struct svc_program *, unsigned int,
393 void (*shutdown)(struct svc_serv *),
394 svc_thread_fn, struct module *);
395 int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
396 void svc_destroy(struct svc_serv *);
397 int svc_process(struct svc_rqst *);
398 int svc_register(const struct svc_serv *, const int,
399 const unsigned short, const unsigned short);
401 void svc_wake_up(struct svc_serv *);
402 void svc_reserve(struct svc_rqst *rqstp, int space);
403 struct svc_pool * svc_pool_for_cpu(struct svc_serv *serv, int cpu);
404 char * svc_print_addr(struct svc_rqst *, char *, size_t);
406 #define RPC_MAX_ADDRBUFLEN (63U)
409 * When we want to reduce the size of the reserved space in the response
410 * buffer, we need to take into account the size of any checksum data that
411 * may be at the end of the packet. This is difficult to determine exactly
412 * for all cases without actually generating the checksum, so we just use a
415 static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space)
419 if (rqstp->rq_authop->flavour)
420 added_space = RPC_MAX_AUTH_SIZE;
421 svc_reserve(rqstp, space + added_space);
424 #endif /* SUNRPC_SVC_H */