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 *);
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;
38 * RPC service thread pool.
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.
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;
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.
65 * We currently do not support more than one RPC program per daemon.
68 struct svc_program * sv_program; /* RPC program */
69 struct svc_stat * sv_stats; /* RPC statistics */
71 unsigned int sv_nrthreads; /* # of server threads */
72 unsigned int sv_maxconn; /* max connections allowed or
73 * '0' causing max to be based
74 * on number of threads. */
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 sa_family_t sv_family; /* listener's address family */
85 char * sv_name; /* service name */
87 unsigned int sv_nrpools; /* number of thread pools */
88 struct svc_pool * sv_pools; /* array of thread pools */
90 void (*sv_shutdown)(struct svc_serv *serv);
91 /* Callback to use when last thread
95 struct module * sv_module; /* optional module to count when
97 svc_thread_fn sv_function; /* main function for threads */
98 unsigned int sv_drc_max_pages; /* Total pages for DRC */
99 unsigned int sv_drc_pages_used;/* DRC pages used */
103 * We use sv_nrthreads as a reference count. svc_destroy() drops
104 * this refcount, so we need to bump it up around operations that
105 * change the number of threads. Horrible, but there it is.
106 * Should be called with the BKL held.
108 static inline void svc_get(struct svc_serv *serv)
110 serv->sv_nrthreads++;
114 * Maximum payload size supported by a kernel RPC server.
115 * This is use to determine the max number of pages nfsd is
116 * willing to return in a single READ operation.
118 * These happen to all be powers of 2, which is not strictly
119 * necessary but helps enforce the real limitation, which is
120 * that they should be multiples of PAGE_CACHE_SIZE.
122 * For UDP transports, a block plus NFS,RPC, and UDP headers
123 * has to fit into the IP datagram limit of 64K. The largest
124 * feasible number for all known page sizes is probably 48K,
125 * but we choose 32K here. This is the same as the historical
126 * Linux limit; someone who cares more about NFS/UDP performance
127 * can test a larger number.
129 * For TCP transports we have more freedom. A size of 1MB is
130 * chosen to match the client limit. Other OSes are known to
131 * have larger limits, but those numbers are probably beyond
132 * the point of diminishing returns.
134 #define RPCSVC_MAXPAYLOAD (1*1024*1024u)
135 #define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD
136 #define RPCSVC_MAXPAYLOAD_UDP (32*1024u)
138 extern u32 svc_max_payload(const struct svc_rqst *rqstp);
141 * RPC Requsts and replies are stored in one or more pages.
142 * We maintain an array of pages for each server thread.
143 * Requests are copied into these pages as they arrive. Remaining
144 * pages are available to write the reply into.
146 * Pages are sent using ->sendpage so each server thread needs to
147 * allocate more to replace those used in sending. To help keep track
148 * of these pages we have a receive list where all pages initialy live,
149 * and a send list where pages are moved to when there are to be part
152 * We use xdr_buf for holding responses as it fits well with NFS
153 * read responses (that have a header, and some data pages, and possibly
154 * a tail) and means we can share some client side routines.
156 * The xdr_buf.head kvec always points to the first page in the rq_*pages
157 * list. The xdr_buf.pages pointer points to the second page on that
158 * list. xdr_buf.tail points to the end of the first page.
159 * This assumes that the non-page part of an rpc reply will fit
160 * in a page - NFSd ensures this. lockd also has no trouble.
162 * Each request/reply pair can have at most one "payload", plus two pages,
163 * one for the request, and one for the reply.
164 * We using ->sendfile to return read data, we might need one extra page
165 * if the request is not page-aligned. So add another '1'.
167 #define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
170 static inline u32 svc_getnl(struct kvec *iov)
175 iov->iov_base = (void*)vp;
176 iov->iov_len -= sizeof(__be32);
180 static inline void svc_putnl(struct kvec *iov, u32 val)
182 __be32 *vp = iov->iov_base + iov->iov_len;
184 iov->iov_len += sizeof(__be32);
187 static inline __be32 svc_getu32(struct kvec *iov)
192 iov->iov_base = (void*)vp;
193 iov->iov_len -= sizeof(__be32);
197 static inline void svc_ungetu32(struct kvec *iov)
199 __be32 *vp = (__be32 *)iov->iov_base;
200 iov->iov_base = (void *)(vp - 1);
201 iov->iov_len += sizeof(*vp);
204 static inline void svc_putu32(struct kvec *iov, __be32 val)
206 __be32 *vp = iov->iov_base + iov->iov_len;
208 iov->iov_len += sizeof(__be32);
213 struct in6_addr addr6;
217 * The context of a single thread, including the request currently being
221 struct list_head rq_list; /* idle list */
222 struct list_head rq_all; /* all threads list */
223 struct svc_xprt * rq_xprt; /* transport ptr */
224 struct sockaddr_storage rq_addr; /* peer address */
227 struct svc_serv * rq_server; /* RPC service definition */
228 struct svc_pool * rq_pool; /* thread pool */
229 struct svc_procedure * rq_procinfo; /* procedure info */
230 struct auth_ops * rq_authop; /* authentication flavour */
231 u32 rq_flavor; /* pseudoflavor */
232 struct svc_cred rq_cred; /* auth info */
233 void * rq_xprt_ctxt; /* transport specific context ptr */
234 struct svc_deferred_req*rq_deferred; /* deferred request we are replaying */
235 int rq_usedeferral; /* use deferral */
237 size_t rq_xprt_hlen; /* xprt header len */
238 struct xdr_buf rq_arg;
239 struct xdr_buf rq_res;
240 struct page * rq_pages[RPCSVC_MAXPAGES];
241 struct page * *rq_respages; /* points into rq_pages */
242 int rq_resused; /* number of pages used for result */
244 struct kvec rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
246 __be32 rq_xid; /* transmission id */
247 u32 rq_prog; /* program number */
248 u32 rq_vers; /* program version */
249 u32 rq_proc; /* procedure number */
250 u32 rq_prot; /* IP protocol */
252 rq_secure : 1; /* secure port */
254 union svc_addr_u rq_daddr; /* dest addr of request
255 * - reply from here */
257 void * rq_argp; /* decoded arguments */
258 void * rq_resp; /* xdr'd results */
259 void * rq_auth_data; /* flavor-specific data */
261 int rq_reserved; /* space on socket outq
262 * reserved for this request
265 struct cache_req rq_chandle; /* handle passed to caches for
268 /* Catering to nfsd */
269 struct auth_domain * rq_client; /* RPC peer info */
270 struct auth_domain * rq_gssclient; /* "gss/"-style peer info */
271 struct svc_cacherep * rq_cacherep; /* cache info */
272 struct knfsd_fh * rq_reffh; /* Referrence filehandle, used to
273 * determine what device number
274 * to report (real or virtual)
276 int rq_splice_ok; /* turned off in gss privacy
277 * to prevent encrypting page
279 wait_queue_head_t rq_wait; /* synchronization */
280 struct task_struct *rq_task; /* service thread */
281 int rq_waking; /* 1 if thread is being woken */
285 * Rigorous type checking on sockaddr type conversions
287 static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst)
289 return (struct sockaddr_in *) &rqst->rq_addr;
292 static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst)
294 return (struct sockaddr_in6 *) &rqst->rq_addr;
297 static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst)
299 return (struct sockaddr *) &rqst->rq_addr;
303 * Check buffer bounds after decoding arguments
306 xdr_argsize_check(struct svc_rqst *rqstp, __be32 *p)
308 char *cp = (char *)p;
309 struct kvec *vec = &rqstp->rq_arg.head[0];
310 return cp >= (char*)vec->iov_base
311 && cp <= (char*)vec->iov_base + vec->iov_len;
315 xdr_ressize_check(struct svc_rqst *rqstp, __be32 *p)
317 struct kvec *vec = &rqstp->rq_res.head[0];
320 vec->iov_len = cp - (char*)vec->iov_base;
322 return vec->iov_len <= PAGE_SIZE;
325 static inline void svc_free_res_pages(struct svc_rqst *rqstp)
327 while (rqstp->rq_resused) {
328 struct page **pp = (rqstp->rq_respages +
329 --rqstp->rq_resused);
337 struct svc_deferred_req {
338 u32 prot; /* protocol (UDP or TCP) */
339 struct svc_xprt *xprt;
340 struct sockaddr_storage addr; /* where reply must go */
342 union svc_addr_u daddr; /* where reply must come from */
343 struct cache_deferred_req handle;
350 * List of RPC programs on the same transport endpoint
353 struct svc_program * pg_next; /* other programs (same xprt) */
354 u32 pg_prog; /* program number */
355 unsigned int pg_lovers; /* lowest version */
356 unsigned int pg_hivers; /* lowest version */
357 unsigned int pg_nvers; /* number of versions */
358 struct svc_version ** pg_vers; /* version array */
359 char * pg_name; /* service name */
360 char * pg_class; /* class name: services sharing authentication */
361 struct svc_stat * pg_stats; /* rpc statistics */
362 int (*pg_authenticate)(struct svc_rqst *);
366 * RPC program version
369 u32 vs_vers; /* version number */
370 u32 vs_nproc; /* number of procedures */
371 struct svc_procedure * vs_proc; /* per-procedure info */
372 u32 vs_xdrsize; /* xdrsize needed for this version */
374 unsigned int vs_hidden : 1; /* Don't register with portmapper.
375 * Only used for nfsacl so far. */
377 /* Override dispatch function (e.g. when caching replies).
378 * A return value of 0 means drop the request.
379 * vs_dispatch == NULL means use default dispatcher.
381 int (*vs_dispatch)(struct svc_rqst *, __be32 *);
387 typedef __be32 (*svc_procfunc)(struct svc_rqst *, void *argp, void *resp);
388 struct svc_procedure {
389 svc_procfunc pc_func; /* process the request */
390 kxdrproc_t pc_decode; /* XDR decode args */
391 kxdrproc_t pc_encode; /* XDR encode result */
392 kxdrproc_t pc_release; /* XDR free result */
393 unsigned int pc_argsize; /* argument struct size */
394 unsigned int pc_ressize; /* result struct size */
395 unsigned int pc_count; /* call count */
396 unsigned int pc_cachetype; /* cache info (NFS) */
397 unsigned int pc_xdrressize; /* maximum size of XDR reply */
401 * Function prototypes.
403 struct svc_serv *svc_create(struct svc_program *, unsigned int, sa_family_t,
404 void (*shutdown)(struct svc_serv *));
405 struct svc_rqst *svc_prepare_thread(struct svc_serv *serv,
406 struct svc_pool *pool);
407 void svc_exit_thread(struct svc_rqst *);
408 struct svc_serv * svc_create_pooled(struct svc_program *, unsigned int,
409 sa_family_t, void (*shutdown)(struct svc_serv *),
410 svc_thread_fn, struct module *);
411 int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
412 int svc_pool_stats_open(struct svc_serv *serv, struct file *file);
413 void svc_destroy(struct svc_serv *);
414 int svc_process(struct svc_rqst *);
415 int svc_register(const struct svc_serv *, const unsigned short,
416 const unsigned short);
418 void svc_wake_up(struct svc_serv *);
419 void svc_reserve(struct svc_rqst *rqstp, int space);
420 struct svc_pool * svc_pool_for_cpu(struct svc_serv *serv, int cpu);
421 char * svc_print_addr(struct svc_rqst *, char *, size_t);
423 #define RPC_MAX_ADDRBUFLEN (63U)
426 * When we want to reduce the size of the reserved space in the response
427 * buffer, we need to take into account the size of any checksum data that
428 * may be at the end of the packet. This is difficult to determine exactly
429 * for all cases without actually generating the checksum, so we just use a
432 static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space)
436 if (rqstp->rq_authop->flavour)
437 added_space = RPC_MAX_AUTH_SIZE;
438 svc_reserve(rqstp, space + added_space);
441 #endif /* SUNRPC_SVC_H */