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 threads_timedout;
37 * RPC service thread pool.
39 * Pool of threads and temporary sockets. Generally there is only
40 * a single one of these per RPC service, but on NUMA machines those
41 * services that can benefit from it (i.e. nfs but not lockd) will
42 * have one pool per NUMA node. This optimisation reduces cross-
43 * node traffic on multi-node NUMA NFS servers.
46 unsigned int sp_id; /* pool id; also node id on NUMA */
47 spinlock_t sp_lock; /* protects all fields */
48 struct list_head sp_threads; /* idle server threads */
49 struct list_head sp_sockets; /* pending sockets */
50 unsigned int sp_nrthreads; /* # of threads in pool */
51 struct list_head sp_all_threads; /* all server threads */
52 struct svc_pool_stats sp_stats; /* statistics on pool operation */
53 } ____cacheline_aligned_in_smp;
58 * An RPC service is a ``daemon,'' possibly multithreaded, which
59 * receives and processes incoming RPC messages.
60 * It has one or more transport sockets associated with it, and maintains
61 * a list of idle threads waiting for input.
63 * We currently do not support more than one RPC program per daemon.
66 struct svc_program * sv_program; /* RPC program */
67 struct svc_stat * sv_stats; /* RPC statistics */
69 unsigned int sv_nrthreads; /* # of server threads */
70 unsigned int sv_maxconn; /* max connections allowed or
71 * '0' causing max to be based
72 * on number of threads. */
74 unsigned int sv_max_payload; /* datagram payload size */
75 unsigned int sv_max_mesg; /* max_payload + 1 page for overheads */
76 unsigned int sv_xdrsize; /* XDR buffer size */
77 struct list_head sv_permsocks; /* all permanent sockets */
78 struct list_head sv_tempsocks; /* all temporary sockets */
79 int sv_tmpcnt; /* count of temporary sockets */
80 struct timer_list sv_temptimer; /* timer for aging temporary sockets */
82 char * sv_name; /* service name */
84 unsigned int sv_nrpools; /* number of thread pools */
85 struct svc_pool * sv_pools; /* array of thread pools */
87 void (*sv_shutdown)(struct svc_serv *serv,
89 /* Callback to use when last thread
93 struct module * sv_module; /* optional module to count when
95 svc_thread_fn sv_function; /* main function for threads */
96 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
97 struct list_head sv_cb_list; /* queue for callback requests
98 * that arrive over the same
100 spinlock_t sv_cb_lock; /* protects the svc_cb_list */
101 wait_queue_head_t sv_cb_waitq; /* sleep here if there are no
102 * entries in the svc_cb_list */
103 struct svc_xprt *sv_bc_xprt; /* callback on fore channel */
104 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
108 * We use sv_nrthreads as a reference count. svc_destroy() drops
109 * this refcount, so we need to bump it up around operations that
110 * change the number of threads. Horrible, but there it is.
111 * Should be called with the BKL held.
113 static inline void svc_get(struct svc_serv *serv)
115 serv->sv_nrthreads++;
119 * Maximum payload size supported by a kernel RPC server.
120 * This is use to determine the max number of pages nfsd is
121 * willing to return in a single READ operation.
123 * These happen to all be powers of 2, which is not strictly
124 * necessary but helps enforce the real limitation, which is
125 * that they should be multiples of PAGE_CACHE_SIZE.
127 * For UDP transports, a block plus NFS,RPC, and UDP headers
128 * has to fit into the IP datagram limit of 64K. The largest
129 * feasible number for all known page sizes is probably 48K,
130 * but we choose 32K here. This is the same as the historical
131 * Linux limit; someone who cares more about NFS/UDP performance
132 * can test a larger number.
134 * For TCP transports we have more freedom. A size of 1MB is
135 * chosen to match the client limit. Other OSes are known to
136 * have larger limits, but those numbers are probably beyond
137 * the point of diminishing returns.
139 #define RPCSVC_MAXPAYLOAD (1*1024*1024u)
140 #define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD
141 #define RPCSVC_MAXPAYLOAD_UDP (32*1024u)
143 extern u32 svc_max_payload(const struct svc_rqst *rqstp);
146 * RPC Requsts and replies are stored in one or more pages.
147 * We maintain an array of pages for each server thread.
148 * Requests are copied into these pages as they arrive. Remaining
149 * pages are available to write the reply into.
151 * Pages are sent using ->sendpage so each server thread needs to
152 * allocate more to replace those used in sending. To help keep track
153 * of these pages we have a receive list where all pages initialy live,
154 * and a send list where pages are moved to when there are to be part
157 * We use xdr_buf for holding responses as it fits well with NFS
158 * read responses (that have a header, and some data pages, and possibly
159 * a tail) and means we can share some client side routines.
161 * The xdr_buf.head kvec always points to the first page in the rq_*pages
162 * list. The xdr_buf.pages pointer points to the second page on that
163 * list. xdr_buf.tail points to the end of the first page.
164 * This assumes that the non-page part of an rpc reply will fit
165 * in a page - NFSd ensures this. lockd also has no trouble.
167 * Each request/reply pair can have at most one "payload", plus two pages,
168 * one for the request, and one for the reply.
169 * We using ->sendfile to return read data, we might need one extra page
170 * if the request is not page-aligned. So add another '1'.
172 #define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
175 static inline u32 svc_getnl(struct kvec *iov)
180 iov->iov_base = (void*)vp;
181 iov->iov_len -= sizeof(__be32);
185 static inline void svc_putnl(struct kvec *iov, u32 val)
187 __be32 *vp = iov->iov_base + iov->iov_len;
189 iov->iov_len += sizeof(__be32);
192 static inline __be32 svc_getu32(struct kvec *iov)
197 iov->iov_base = (void*)vp;
198 iov->iov_len -= sizeof(__be32);
202 static inline void svc_ungetu32(struct kvec *iov)
204 __be32 *vp = (__be32 *)iov->iov_base;
205 iov->iov_base = (void *)(vp - 1);
206 iov->iov_len += sizeof(*vp);
209 static inline void svc_putu32(struct kvec *iov, __be32 val)
211 __be32 *vp = iov->iov_base + iov->iov_len;
213 iov->iov_len += sizeof(__be32);
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 */
225 struct sockaddr_storage rq_addr; /* peer address */
227 struct sockaddr_storage rq_daddr; /* dest addr of request
228 * - reply from here */
231 struct svc_serv * rq_server; /* RPC service definition */
232 struct svc_pool * rq_pool; /* thread pool */
233 struct svc_procedure * rq_procinfo; /* procedure info */
234 struct auth_ops * rq_authop; /* authentication flavour */
235 u32 rq_flavor; /* pseudoflavor */
236 struct svc_cred rq_cred; /* auth info */
237 void * rq_xprt_ctxt; /* transport specific context ptr */
238 struct svc_deferred_req*rq_deferred; /* deferred request we are replaying */
239 int rq_usedeferral; /* use deferral */
241 size_t rq_xprt_hlen; /* xprt header len */
242 struct xdr_buf rq_arg;
243 struct xdr_buf rq_res;
244 struct page * rq_pages[RPCSVC_MAXPAGES];
245 struct page * *rq_respages; /* points into rq_pages */
246 int rq_resused; /* number of pages used for result */
248 struct kvec rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
250 __be32 rq_xid; /* transmission id */
251 u32 rq_prog; /* program number */
252 u32 rq_vers; /* program version */
253 u32 rq_proc; /* procedure number */
254 u32 rq_prot; /* IP protocol */
256 rq_secure : 1; /* secure port */
258 void * rq_argp; /* decoded arguments */
259 void * rq_resp; /* xdr'd results */
260 void * rq_auth_data; /* flavor-specific data */
262 int rq_reserved; /* space on socket outq
263 * reserved for this request
266 struct cache_req rq_chandle; /* handle passed to caches for
270 /* Catering to nfsd */
271 struct auth_domain * rq_client; /* RPC peer info */
272 struct auth_domain * rq_gssclient; /* "gss/"-style peer info */
274 struct svc_cacherep * rq_cacherep; /* cache info */
275 int rq_splice_ok; /* turned off in gss privacy
276 * to prevent encrypting page
278 wait_queue_head_t rq_wait; /* synchronization */
279 struct task_struct *rq_task; /* service thread */
283 * Rigorous type checking on sockaddr type conversions
285 static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst)
287 return (struct sockaddr_in *) &rqst->rq_addr;
290 static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst)
292 return (struct sockaddr_in6 *) &rqst->rq_addr;
295 static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst)
297 return (struct sockaddr *) &rqst->rq_addr;
300 static inline struct sockaddr_in *svc_daddr_in(const struct svc_rqst *rqst)
302 return (struct sockaddr_in *) &rqst->rq_daddr;
305 static inline struct sockaddr_in6 *svc_daddr_in6(const struct svc_rqst *rqst)
307 return (struct sockaddr_in6 *) &rqst->rq_daddr;
310 static inline struct sockaddr *svc_daddr(const struct svc_rqst *rqst)
312 return (struct sockaddr *) &rqst->rq_daddr;
316 * Check buffer bounds after decoding arguments
319 xdr_argsize_check(struct svc_rqst *rqstp, __be32 *p)
321 char *cp = (char *)p;
322 struct kvec *vec = &rqstp->rq_arg.head[0];
323 return cp >= (char*)vec->iov_base
324 && cp <= (char*)vec->iov_base + vec->iov_len;
328 xdr_ressize_check(struct svc_rqst *rqstp, __be32 *p)
330 struct kvec *vec = &rqstp->rq_res.head[0];
333 vec->iov_len = cp - (char*)vec->iov_base;
335 return vec->iov_len <= PAGE_SIZE;
338 static inline void svc_free_res_pages(struct svc_rqst *rqstp)
340 while (rqstp->rq_resused) {
341 struct page **pp = (rqstp->rq_respages +
342 --rqstp->rq_resused);
350 struct svc_deferred_req {
351 u32 prot; /* protocol (UDP or TCP) */
352 struct svc_xprt *xprt;
353 struct sockaddr_storage addr; /* where reply must go */
355 struct sockaddr_storage daddr; /* where reply must come from */
357 struct cache_deferred_req handle;
364 * List of RPC programs on the same transport endpoint
367 struct svc_program * pg_next; /* other programs (same xprt) */
368 u32 pg_prog; /* program number */
369 unsigned int pg_lovers; /* lowest version */
370 unsigned int pg_hivers; /* lowest version */
371 unsigned int pg_nvers; /* number of versions */
372 struct svc_version ** pg_vers; /* version array */
373 char * pg_name; /* service name */
374 char * pg_class; /* class name: services sharing authentication */
375 struct svc_stat * pg_stats; /* rpc statistics */
376 int (*pg_authenticate)(struct svc_rqst *);
380 * RPC program version
383 u32 vs_vers; /* version number */
384 u32 vs_nproc; /* number of procedures */
385 struct svc_procedure * vs_proc; /* per-procedure info */
386 u32 vs_xdrsize; /* xdrsize needed for this version */
388 unsigned int vs_hidden : 1; /* Don't register with portmapper.
389 * Only used for nfsacl so far. */
391 /* Override dispatch function (e.g. when caching replies).
392 * A return value of 0 means drop the request.
393 * vs_dispatch == NULL means use default dispatcher.
395 int (*vs_dispatch)(struct svc_rqst *, __be32 *);
401 typedef __be32 (*svc_procfunc)(struct svc_rqst *, void *argp, void *resp);
402 struct svc_procedure {
403 svc_procfunc pc_func; /* process the request */
404 kxdrproc_t pc_decode; /* XDR decode args */
405 kxdrproc_t pc_encode; /* XDR encode result */
406 kxdrproc_t pc_release; /* XDR free result */
407 unsigned int pc_argsize; /* argument struct size */
408 unsigned int pc_ressize; /* result struct size */
409 unsigned int pc_count; /* call count */
410 unsigned int pc_cachetype; /* cache info (NFS) */
411 unsigned int pc_xdrressize; /* maximum size of XDR reply */
415 * Function prototypes.
417 int svc_rpcb_setup(struct svc_serv *serv, struct net *net);
418 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net);
419 int svc_bind(struct svc_serv *serv, struct net *net);
420 struct svc_serv *svc_create(struct svc_program *, unsigned int,
421 void (*shutdown)(struct svc_serv *, struct net *net));
422 struct svc_rqst *svc_prepare_thread(struct svc_serv *serv,
423 struct svc_pool *pool, int node);
424 void svc_exit_thread(struct svc_rqst *);
425 struct svc_serv * svc_create_pooled(struct svc_program *, unsigned int,
426 void (*shutdown)(struct svc_serv *, struct net *net),
427 svc_thread_fn, struct module *);
428 int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
429 int svc_pool_stats_open(struct svc_serv *serv, struct file *file);
430 void svc_destroy(struct svc_serv *);
431 void svc_shutdown_net(struct svc_serv *, struct net *);
432 int svc_process(struct svc_rqst *);
433 int bc_svc_process(struct svc_serv *, struct rpc_rqst *,
435 int svc_register(const struct svc_serv *, struct net *, const int,
436 const unsigned short, const unsigned short);
438 void svc_wake_up(struct svc_serv *);
439 void svc_reserve(struct svc_rqst *rqstp, int space);
440 struct svc_pool * svc_pool_for_cpu(struct svc_serv *serv, int cpu);
441 char * svc_print_addr(struct svc_rqst *, char *, size_t);
443 #define RPC_MAX_ADDRBUFLEN (63U)
446 * When we want to reduce the size of the reserved space in the response
447 * buffer, we need to take into account the size of any checksum data that
448 * may be at the end of the packet. This is difficult to determine exactly
449 * for all cases without actually generating the checksum, so we just use a
452 static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space)
456 if (rqstp->rq_authop->flavour)
457 added_space = RPC_MAX_AUTH_SIZE;
458 svc_reserve(rqstp, space + added_space);
461 #endif /* SUNRPC_SVC_H */