new retransmit part 3
[cor_2_6_31.git] / include / linux / sunrpc / svc.h
blobea8009695c69b1ea437b255fd0a141ebf21feaa7
1 /*
2 * linux/include/linux/sunrpc/svc.h
4 * RPC server declarations.
6 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
7 */
10 #ifndef SUNRPC_SVC_H
11 #define SUNRPC_SVC_H
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>
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.
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;
58 * RPC service.
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.
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 * '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 */
84 char * sv_name; /* service name */
86 unsigned int sv_nrpools; /* number of thread pools */
87 struct svc_pool * sv_pools; /* array of thread pools */
89 void (*sv_shutdown)(struct svc_serv *serv);
90 /* Callback to use when last thread
91 * exits.
94 struct module * sv_module; /* optional module to count when
95 * adding threads */
96 svc_thread_fn sv_function; /* main function for threads */
97 unsigned int sv_drc_max_pages; /* Total pages for DRC */
98 unsigned int sv_drc_pages_used;/* DRC pages used */
99 #if defined(CONFIG_NFS_V4_1)
100 struct list_head sv_cb_list; /* queue for callback requests
101 * that arrive over the same
102 * connection */
103 spinlock_t sv_cb_lock; /* protects the svc_cb_list */
104 wait_queue_head_t sv_cb_waitq; /* sleep here if there are no
105 * entries in the svc_cb_list */
106 struct svc_xprt *bc_xprt;
107 #endif /* CONFIG_NFS_V4_1 */
111 * We use sv_nrthreads as a reference count. svc_destroy() drops
112 * this refcount, so we need to bump it up around operations that
113 * change the number of threads. Horrible, but there it is.
114 * Should be called with the BKL held.
116 static inline void svc_get(struct svc_serv *serv)
118 serv->sv_nrthreads++;
122 * Maximum payload size supported by a kernel RPC server.
123 * This is use to determine the max number of pages nfsd is
124 * willing to return in a single READ operation.
126 * These happen to all be powers of 2, which is not strictly
127 * necessary but helps enforce the real limitation, which is
128 * that they should be multiples of PAGE_CACHE_SIZE.
130 * For UDP transports, a block plus NFS,RPC, and UDP headers
131 * has to fit into the IP datagram limit of 64K. The largest
132 * feasible number for all known page sizes is probably 48K,
133 * but we choose 32K here. This is the same as the historical
134 * Linux limit; someone who cares more about NFS/UDP performance
135 * can test a larger number.
137 * For TCP transports we have more freedom. A size of 1MB is
138 * chosen to match the client limit. Other OSes are known to
139 * have larger limits, but those numbers are probably beyond
140 * the point of diminishing returns.
142 #define RPCSVC_MAXPAYLOAD (1*1024*1024u)
143 #define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD
144 #define RPCSVC_MAXPAYLOAD_UDP (32*1024u)
146 extern u32 svc_max_payload(const struct svc_rqst *rqstp);
149 * RPC Requsts and replies are stored in one or more pages.
150 * We maintain an array of pages for each server thread.
151 * Requests are copied into these pages as they arrive. Remaining
152 * pages are available to write the reply into.
154 * Pages are sent using ->sendpage so each server thread needs to
155 * allocate more to replace those used in sending. To help keep track
156 * of these pages we have a receive list where all pages initialy live,
157 * and a send list where pages are moved to when there are to be part
158 * of a reply.
160 * We use xdr_buf for holding responses as it fits well with NFS
161 * read responses (that have a header, and some data pages, and possibly
162 * a tail) and means we can share some client side routines.
164 * The xdr_buf.head kvec always points to the first page in the rq_*pages
165 * list. The xdr_buf.pages pointer points to the second page on that
166 * list. xdr_buf.tail points to the end of the first page.
167 * This assumes that the non-page part of an rpc reply will fit
168 * in a page - NFSd ensures this. lockd also has no trouble.
170 * Each request/reply pair can have at most one "payload", plus two pages,
171 * one for the request, and one for the reply.
172 * We using ->sendfile to return read data, we might need one extra page
173 * if the request is not page-aligned. So add another '1'.
175 #define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
176 + 2 + 1)
178 static inline u32 svc_getnl(struct kvec *iov)
180 __be32 val, *vp;
181 vp = iov->iov_base;
182 val = *vp++;
183 iov->iov_base = (void*)vp;
184 iov->iov_len -= sizeof(__be32);
185 return ntohl(val);
188 static inline void svc_putnl(struct kvec *iov, u32 val)
190 __be32 *vp = iov->iov_base + iov->iov_len;
191 *vp = htonl(val);
192 iov->iov_len += sizeof(__be32);
195 static inline __be32 svc_getu32(struct kvec *iov)
197 __be32 val, *vp;
198 vp = iov->iov_base;
199 val = *vp++;
200 iov->iov_base = (void*)vp;
201 iov->iov_len -= sizeof(__be32);
202 return val;
205 static inline void svc_ungetu32(struct kvec *iov)
207 __be32 *vp = (__be32 *)iov->iov_base;
208 iov->iov_base = (void *)(vp - 1);
209 iov->iov_len += sizeof(*vp);
212 static inline void svc_putu32(struct kvec *iov, __be32 val)
214 __be32 *vp = iov->iov_base + iov->iov_len;
215 *vp = val;
216 iov->iov_len += sizeof(__be32);
219 union svc_addr_u {
220 struct in_addr addr;
221 struct in6_addr addr6;
225 * The context of a single thread, including the request currently being
226 * processed.
228 struct svc_rqst {
229 struct list_head rq_list; /* idle list */
230 struct list_head rq_all; /* all threads list */
231 struct svc_xprt * rq_xprt; /* transport ptr */
232 struct sockaddr_storage rq_addr; /* peer address */
233 size_t rq_addrlen;
235 struct svc_serv * rq_server; /* RPC service definition */
236 struct svc_pool * rq_pool; /* thread pool */
237 struct svc_procedure * rq_procinfo; /* procedure info */
238 struct auth_ops * rq_authop; /* authentication flavour */
239 u32 rq_flavor; /* pseudoflavor */
240 struct svc_cred rq_cred; /* auth info */
241 void * rq_xprt_ctxt; /* transport specific context ptr */
242 struct svc_deferred_req*rq_deferred; /* deferred request we are replaying */
243 int rq_usedeferral; /* use deferral */
245 size_t rq_xprt_hlen; /* xprt header len */
246 struct xdr_buf rq_arg;
247 struct xdr_buf rq_res;
248 struct page * rq_pages[RPCSVC_MAXPAGES];
249 struct page * *rq_respages; /* points into rq_pages */
250 int rq_resused; /* number of pages used for result */
252 struct kvec rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
254 __be32 rq_xid; /* transmission id */
255 u32 rq_prog; /* program number */
256 u32 rq_vers; /* program version */
257 u32 rq_proc; /* procedure number */
258 u32 rq_prot; /* IP protocol */
259 unsigned short
260 rq_secure : 1; /* secure port */
262 union svc_addr_u rq_daddr; /* dest addr of request
263 * - reply from here */
265 void * rq_argp; /* decoded arguments */
266 void * rq_resp; /* xdr'd results */
267 void * rq_auth_data; /* flavor-specific data */
269 int rq_reserved; /* space on socket outq
270 * reserved for this request
273 struct cache_req rq_chandle; /* handle passed to caches for
274 * request delaying
276 /* Catering to nfsd */
277 struct auth_domain * rq_client; /* RPC peer info */
278 struct auth_domain * rq_gssclient; /* "gss/"-style peer info */
279 struct svc_cacherep * rq_cacherep; /* cache info */
280 struct knfsd_fh * rq_reffh; /* Referrence filehandle, used to
281 * determine what device number
282 * to report (real or virtual)
284 int rq_splice_ok; /* turned off in gss privacy
285 * to prevent encrypting page
286 * cache pages */
287 wait_queue_head_t rq_wait; /* synchronization */
288 struct task_struct *rq_task; /* service thread */
289 int rq_waking; /* 1 if thread is being woken */
293 * Rigorous type checking on sockaddr type conversions
295 static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst)
297 return (struct sockaddr_in *) &rqst->rq_addr;
300 static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst)
302 return (struct sockaddr_in6 *) &rqst->rq_addr;
305 static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst)
307 return (struct sockaddr *) &rqst->rq_addr;
311 * Check buffer bounds after decoding arguments
313 static inline int
314 xdr_argsize_check(struct svc_rqst *rqstp, __be32 *p)
316 char *cp = (char *)p;
317 struct kvec *vec = &rqstp->rq_arg.head[0];
318 return cp >= (char*)vec->iov_base
319 && cp <= (char*)vec->iov_base + vec->iov_len;
322 static inline int
323 xdr_ressize_check(struct svc_rqst *rqstp, __be32 *p)
325 struct kvec *vec = &rqstp->rq_res.head[0];
326 char *cp = (char*)p;
328 vec->iov_len = cp - (char*)vec->iov_base;
330 return vec->iov_len <= PAGE_SIZE;
333 static inline void svc_free_res_pages(struct svc_rqst *rqstp)
335 while (rqstp->rq_resused) {
336 struct page **pp = (rqstp->rq_respages +
337 --rqstp->rq_resused);
338 if (*pp) {
339 put_page(*pp);
340 *pp = NULL;
345 struct svc_deferred_req {
346 u32 prot; /* protocol (UDP or TCP) */
347 struct svc_xprt *xprt;
348 struct sockaddr_storage addr; /* where reply must go */
349 size_t addrlen;
350 union svc_addr_u daddr; /* where reply must come from */
351 struct cache_deferred_req handle;
352 size_t xprt_hlen;
353 int argslen;
354 __be32 args[0];
358 * List of RPC programs on the same transport endpoint
360 struct svc_program {
361 struct svc_program * pg_next; /* other programs (same xprt) */
362 u32 pg_prog; /* program number */
363 unsigned int pg_lovers; /* lowest version */
364 unsigned int pg_hivers; /* lowest version */
365 unsigned int pg_nvers; /* number of versions */
366 struct svc_version ** pg_vers; /* version array */
367 char * pg_name; /* service name */
368 char * pg_class; /* class name: services sharing authentication */
369 struct svc_stat * pg_stats; /* rpc statistics */
370 int (*pg_authenticate)(struct svc_rqst *);
374 * RPC program version
376 struct svc_version {
377 u32 vs_vers; /* version number */
378 u32 vs_nproc; /* number of procedures */
379 struct svc_procedure * vs_proc; /* per-procedure info */
380 u32 vs_xdrsize; /* xdrsize needed for this version */
382 unsigned int vs_hidden : 1; /* Don't register with portmapper.
383 * Only used for nfsacl so far. */
385 /* Override dispatch function (e.g. when caching replies).
386 * A return value of 0 means drop the request.
387 * vs_dispatch == NULL means use default dispatcher.
389 int (*vs_dispatch)(struct svc_rqst *, __be32 *);
393 * RPC procedure info
395 typedef __be32 (*svc_procfunc)(struct svc_rqst *, void *argp, void *resp);
396 struct svc_procedure {
397 svc_procfunc pc_func; /* process the request */
398 kxdrproc_t pc_decode; /* XDR decode args */
399 kxdrproc_t pc_encode; /* XDR encode result */
400 kxdrproc_t pc_release; /* XDR free result */
401 unsigned int pc_argsize; /* argument struct size */
402 unsigned int pc_ressize; /* result struct size */
403 unsigned int pc_count; /* call count */
404 unsigned int pc_cachetype; /* cache info (NFS) */
405 unsigned int pc_xdrressize; /* maximum size of XDR reply */
409 * Function prototypes.
411 struct svc_serv *svc_create(struct svc_program *, unsigned int,
412 void (*shutdown)(struct svc_serv *));
413 struct svc_rqst *svc_prepare_thread(struct svc_serv *serv,
414 struct svc_pool *pool);
415 void svc_exit_thread(struct svc_rqst *);
416 struct svc_serv * svc_create_pooled(struct svc_program *, unsigned int,
417 void (*shutdown)(struct svc_serv *),
418 svc_thread_fn, struct module *);
419 int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
420 int svc_pool_stats_open(struct svc_serv *serv, struct file *file);
421 void svc_destroy(struct svc_serv *);
422 int svc_process(struct svc_rqst *);
423 int bc_svc_process(struct svc_serv *, struct rpc_rqst *,
424 struct svc_rqst *);
425 int svc_register(const struct svc_serv *, const int,
426 const unsigned short, const unsigned short);
428 void svc_wake_up(struct svc_serv *);
429 void svc_reserve(struct svc_rqst *rqstp, int space);
430 struct svc_pool * svc_pool_for_cpu(struct svc_serv *serv, int cpu);
431 char * svc_print_addr(struct svc_rqst *, char *, size_t);
433 #define RPC_MAX_ADDRBUFLEN (63U)
436 * When we want to reduce the size of the reserved space in the response
437 * buffer, we need to take into account the size of any checksum data that
438 * may be at the end of the packet. This is difficult to determine exactly
439 * for all cases without actually generating the checksum, so we just use a
440 * static value.
442 static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space)
444 int added_space = 0;
446 if (rqstp->rq_authop->flavour)
447 added_space = RPC_MAX_AUTH_SIZE;
448 svc_reserve(rqstp, space + added_space);
451 #endif /* SUNRPC_SVC_H */