nfs41: Refactor svc_process()
[deliverable/linux.git] / include / linux / sunrpc / svc.h
CommitLineData
1da177e4
LT
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>
73df0dba 14#include <linux/in6.h>
1da177e4
LT
15#include <linux/sunrpc/types.h>
16#include <linux/sunrpc/xdr.h>
7adae489 17#include <linux/sunrpc/auth.h>
1da177e4
LT
18#include <linux/sunrpc/svcauth.h>
19#include <linux/wait.h>
20#include <linux/mm.h>
21
a7455442
GB
22/*
23 * This is the RPC server thread function prototype
24 */
9867d76c 25typedef int (*svc_thread_fn)(void *);
3262c816 26
03cf6c9f
GB
27/* statistics for svc_pool structures */
28struct 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
3262c816
GB
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 */
46struct 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 */
a7455442 52 struct list_head sp_all_threads; /* all server threads */
59a252ff 53 int sp_nwaking; /* number of threads woken but not yet active */
03cf6c9f 54 struct svc_pool_stats sp_stats; /* statistics on pool operation */
3262c816
GB
55} ____cacheline_aligned_in_smp;
56
1da177e4
LT
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 */
67struct svc_serv {
1da177e4
LT
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 */
c9233eb7
JL
72 unsigned int sv_maxconn; /* max connections allowed or
73 * '0' causing max to be based
74 * on number of threads. */
75
c6b0a9f8
N
76 unsigned int sv_max_payload; /* datagram payload size */
77 unsigned int sv_max_mesg; /* max_payload + 1 page for overheads */
1da177e4 78 unsigned int sv_xdrsize; /* XDR buffer size */
1da177e4
LT
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 */
36bdfc8b 82 struct timer_list sv_temptimer; /* timer for aging temporary sockets */
1da177e4
LT
83
84 char * sv_name; /* service name */
bc591ccf 85
3262c816
GB
86 unsigned int sv_nrpools; /* number of thread pools */
87 struct svc_pool * sv_pools; /* array of thread pools */
88
bc591ccf
N
89 void (*sv_shutdown)(struct svc_serv *serv);
90 /* Callback to use when last thread
91 * exits.
92 */
a7455442
GB
93
94 struct module * sv_module; /* optional module to count when
95 * adding threads */
96 svc_thread_fn sv_function; /* main function for threads */
c3d06f9c
AA
97 unsigned int sv_drc_max_pages; /* Total pages for DRC */
98 unsigned int sv_drc_pages_used;/* DRC pages used */
56632b5b
RL
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#endif /* CONFIG_NFS_V4_1 */
1da177e4
LT
107};
108
9a24ab57
GB
109/*
110 * We use sv_nrthreads as a reference count. svc_destroy() drops
111 * this refcount, so we need to bump it up around operations that
112 * change the number of threads. Horrible, but there it is.
113 * Should be called with the BKL held.
114 */
115static inline void svc_get(struct svc_serv *serv)
116{
117 serv->sv_nrthreads++;
118}
119
1da177e4
LT
120/*
121 * Maximum payload size supported by a kernel RPC server.
122 * This is use to determine the max number of pages nfsd is
123 * willing to return in a single READ operation.
7adae489
GB
124 *
125 * These happen to all be powers of 2, which is not strictly
126 * necessary but helps enforce the real limitation, which is
127 * that they should be multiples of PAGE_CACHE_SIZE.
128 *
129 * For UDP transports, a block plus NFS,RPC, and UDP headers
130 * has to fit into the IP datagram limit of 64K. The largest
131 * feasible number for all known page sizes is probably 48K,
132 * but we choose 32K here. This is the same as the historical
133 * Linux limit; someone who cares more about NFS/UDP performance
134 * can test a larger number.
135 *
136 * For TCP transports we have more freedom. A size of 1MB is
137 * chosen to match the client limit. Other OSes are known to
138 * have larger limits, but those numbers are probably beyond
139 * the point of diminishing returns.
1da177e4 140 */
7adae489
GB
141#define RPCSVC_MAXPAYLOAD (1*1024*1024u)
142#define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD
143#define RPCSVC_MAXPAYLOAD_UDP (32*1024u)
144
145extern u32 svc_max_payload(const struct svc_rqst *rqstp);
1da177e4
LT
146
147/*
148 * RPC Requsts and replies are stored in one or more pages.
149 * We maintain an array of pages for each server thread.
150 * Requests are copied into these pages as they arrive. Remaining
151 * pages are available to write the reply into.
152 *
153 * Pages are sent using ->sendpage so each server thread needs to
154 * allocate more to replace those used in sending. To help keep track
155 * of these pages we have a receive list where all pages initialy live,
156 * and a send list where pages are moved to when there are to be part
157 * of a reply.
158 *
159 * We use xdr_buf for holding responses as it fits well with NFS
160 * read responses (that have a header, and some data pages, and possibly
161 * a tail) and means we can share some client side routines.
162 *
163 * The xdr_buf.head kvec always points to the first page in the rq_*pages
164 * list. The xdr_buf.pages pointer points to the second page on that
165 * list. xdr_buf.tail points to the end of the first page.
166 * This assumes that the non-page part of an rpc reply will fit
167 * in a page - NFSd ensures this. lockd also has no trouble.
168 *
169 * Each request/reply pair can have at most one "payload", plus two pages,
170 * one for the request, and one for the reply.
250f3915
N
171 * We using ->sendfile to return read data, we might need one extra page
172 * if the request is not page-aligned. So add another '1'.
1da177e4 173 */
250f3915
N
174#define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
175 + 2 + 1)
1da177e4 176
76994313 177static inline u32 svc_getnl(struct kvec *iov)
1da177e4 178{
76994313 179 __be32 val, *vp;
1da177e4
LT
180 vp = iov->iov_base;
181 val = *vp++;
182 iov->iov_base = (void*)vp;
76994313
AD
183 iov->iov_len -= sizeof(__be32);
184 return ntohl(val);
185}
186
187static inline void svc_putnl(struct kvec *iov, u32 val)
188{
189 __be32 *vp = iov->iov_base + iov->iov_len;
190 *vp = htonl(val);
191 iov->iov_len += sizeof(__be32);
192}
193
194static inline __be32 svc_getu32(struct kvec *iov)
195{
196 __be32 val, *vp;
197 vp = iov->iov_base;
198 val = *vp++;
199 iov->iov_base = (void*)vp;
200 iov->iov_len -= sizeof(__be32);
1da177e4
LT
201 return val;
202}
203
204static inline void svc_ungetu32(struct kvec *iov)
205{
76994313 206 __be32 *vp = (__be32 *)iov->iov_base;
1da177e4
LT
207 iov->iov_base = (void *)(vp - 1);
208 iov->iov_len += sizeof(*vp);
209}
210
76994313 211static inline void svc_putu32(struct kvec *iov, __be32 val)
1da177e4 212{
76994313 213 __be32 *vp = iov->iov_base + iov->iov_len;
1da177e4 214 *vp = val;
76994313 215 iov->iov_len += sizeof(__be32);
1da177e4
LT
216}
217
73df0dba
CL
218union svc_addr_u {
219 struct in_addr addr;
73df0dba 220 struct in6_addr addr6;
73df0dba
CL
221};
222
1da177e4
LT
223/*
224 * The context of a single thread, including the request currently being
225 * processed.
1da177e4
LT
226 */
227struct svc_rqst {
228 struct list_head rq_list; /* idle list */
a7455442 229 struct list_head rq_all; /* all threads list */
57b1d3ba 230 struct svc_xprt * rq_xprt; /* transport ptr */
27459f09
CL
231 struct sockaddr_storage rq_addr; /* peer address */
232 size_t rq_addrlen;
1da177e4
LT
233
234 struct svc_serv * rq_server; /* RPC service definition */
3262c816 235 struct svc_pool * rq_pool; /* thread pool */
1da177e4
LT
236 struct svc_procedure * rq_procinfo; /* procedure info */
237 struct auth_ops * rq_authop; /* authentication flavour */
c4170583 238 u32 rq_flavor; /* pseudoflavor */
1da177e4 239 struct svc_cred rq_cred; /* auth info */
5148bf4e 240 void * rq_xprt_ctxt; /* transport specific context ptr */
1da177e4 241 struct svc_deferred_req*rq_deferred; /* deferred request we are replaying */
2f425878 242 int rq_usedeferral; /* use deferral */
1da177e4 243
260c1d12 244 size_t rq_xprt_hlen; /* xprt header len */
1da177e4
LT
245 struct xdr_buf rq_arg;
246 struct xdr_buf rq_res;
44524359
N
247 struct page * rq_pages[RPCSVC_MAXPAGES];
248 struct page * *rq_respages; /* points into rq_pages */
249 int rq_resused; /* number of pages used for result */
1da177e4 250
3cc03b16
N
251 struct kvec rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
252
d8ed029d 253 __be32 rq_xid; /* transmission id */
1da177e4
LT
254 u32 rq_prog; /* program number */
255 u32 rq_vers; /* program version */
256 u32 rq_proc; /* procedure number */
257 u32 rq_prot; /* IP protocol */
258 unsigned short
259 rq_secure : 1; /* secure port */
260
73df0dba
CL
261 union svc_addr_u rq_daddr; /* dest addr of request
262 * - reply from here */
1da177e4
LT
263
264 void * rq_argp; /* decoded arguments */
265 void * rq_resp; /* xdr'd results */
266 void * rq_auth_data; /* flavor-specific data */
267
268 int rq_reserved; /* space on socket outq
269 * reserved for this request
270 */
271
272 struct cache_req rq_chandle; /* handle passed to caches for
273 * request delaying
274 */
275 /* Catering to nfsd */
276 struct auth_domain * rq_client; /* RPC peer info */
3ab4d8b1 277 struct auth_domain * rq_gssclient; /* "gss/"-style peer info */
1da177e4
LT
278 struct svc_cacherep * rq_cacherep; /* cache info */
279 struct knfsd_fh * rq_reffh; /* Referrence filehandle, used to
280 * determine what device number
281 * to report (real or virtual)
282 */
cf8208d0 283 int rq_splice_ok; /* turned off in gss privacy
5c04c46a
BF
284 * to prevent encrypting page
285 * cache pages */
1da177e4 286 wait_queue_head_t rq_wait; /* synchronization */
a7455442 287 struct task_struct *rq_task; /* service thread */
59a252ff 288 int rq_waking; /* 1 if thread is being woken */
1da177e4
LT
289};
290
27459f09
CL
291/*
292 * Rigorous type checking on sockaddr type conversions
293 */
5344b12d 294static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst)
27459f09
CL
295{
296 return (struct sockaddr_in *) &rqst->rq_addr;
297}
298
5344b12d 299static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst)
95756482
CL
300{
301 return (struct sockaddr_in6 *) &rqst->rq_addr;
302}
303
5344b12d 304static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst)
27459f09
CL
305{
306 return (struct sockaddr *) &rqst->rq_addr;
307}
308
1da177e4
LT
309/*
310 * Check buffer bounds after decoding arguments
311 */
312static inline int
d8ed029d 313xdr_argsize_check(struct svc_rqst *rqstp, __be32 *p)
1da177e4
LT
314{
315 char *cp = (char *)p;
316 struct kvec *vec = &rqstp->rq_arg.head[0];
0ba7536d
N
317 return cp >= (char*)vec->iov_base
318 && cp <= (char*)vec->iov_base + vec->iov_len;
1da177e4
LT
319}
320
321static inline int
d8ed029d 322xdr_ressize_check(struct svc_rqst *rqstp, __be32 *p)
1da177e4
LT
323{
324 struct kvec *vec = &rqstp->rq_res.head[0];
325 char *cp = (char*)p;
326
327 vec->iov_len = cp - (char*)vec->iov_base;
328
329 return vec->iov_len <= PAGE_SIZE;
330}
331
44524359 332static inline void svc_free_res_pages(struct svc_rqst *rqstp)
a257cdd0 333{
44524359
N
334 while (rqstp->rq_resused) {
335 struct page **pp = (rqstp->rq_respages +
336 --rqstp->rq_resused);
337 if (*pp) {
338 put_page(*pp);
339 *pp = NULL;
1da177e4
LT
340 }
341 }
342}
343
1da177e4
LT
344struct svc_deferred_req {
345 u32 prot; /* protocol (UDP or TCP) */
8c7b0172 346 struct svc_xprt *xprt;
24422222
CL
347 struct sockaddr_storage addr; /* where reply must go */
348 size_t addrlen;
73df0dba 349 union svc_addr_u daddr; /* where reply must come from */
1da177e4 350 struct cache_deferred_req handle;
260c1d12 351 size_t xprt_hlen;
1da177e4 352 int argslen;
d8ed029d 353 __be32 args[0];
1da177e4
LT
354};
355
356/*
9ba02638 357 * List of RPC programs on the same transport endpoint
1da177e4
LT
358 */
359struct svc_program {
9ba02638 360 struct svc_program * pg_next; /* other programs (same xprt) */
1da177e4
LT
361 u32 pg_prog; /* program number */
362 unsigned int pg_lovers; /* lowest version */
363 unsigned int pg_hivers; /* lowest version */
364 unsigned int pg_nvers; /* number of versions */
365 struct svc_version ** pg_vers; /* version array */
366 char * pg_name; /* service name */
367 char * pg_class; /* class name: services sharing authentication */
368 struct svc_stat * pg_stats; /* rpc statistics */
369 int (*pg_authenticate)(struct svc_rqst *);
370};
371
372/*
373 * RPC program version
374 */
375struct svc_version {
376 u32 vs_vers; /* version number */
377 u32 vs_nproc; /* number of procedures */
378 struct svc_procedure * vs_proc; /* per-procedure info */
379 u32 vs_xdrsize; /* xdrsize needed for this version */
380
bc5fea42
OK
381 unsigned int vs_hidden : 1; /* Don't register with portmapper.
382 * Only used for nfsacl so far. */
383
1da177e4
LT
384 /* Override dispatch function (e.g. when caching replies).
385 * A return value of 0 means drop the request.
386 * vs_dispatch == NULL means use default dispatcher.
387 */
d8ed029d 388 int (*vs_dispatch)(struct svc_rqst *, __be32 *);
1da177e4
LT
389};
390
391/*
392 * RPC procedure info
393 */
7111c66e 394typedef __be32 (*svc_procfunc)(struct svc_rqst *, void *argp, void *resp);
1da177e4
LT
395struct svc_procedure {
396 svc_procfunc pc_func; /* process the request */
397 kxdrproc_t pc_decode; /* XDR decode args */
398 kxdrproc_t pc_encode; /* XDR encode result */
399 kxdrproc_t pc_release; /* XDR free result */
400 unsigned int pc_argsize; /* argument struct size */
401 unsigned int pc_ressize; /* result struct size */
402 unsigned int pc_count; /* call count */
403 unsigned int pc_cachetype; /* cache info (NFS) */
404 unsigned int pc_xdrressize; /* maximum size of XDR reply */
405};
406
1da177e4
LT
407/*
408 * Function prototypes.
409 */
49a9072f 410struct svc_serv *svc_create(struct svc_program *, unsigned int,
e851db5b 411 void (*shutdown)(struct svc_serv *));
0113ab34
JL
412struct svc_rqst *svc_prepare_thread(struct svc_serv *serv,
413 struct svc_pool *pool);
1da177e4 414void svc_exit_thread(struct svc_rqst *);
a7455442 415struct svc_serv * svc_create_pooled(struct svc_program *, unsigned int,
49a9072f 416 void (*shutdown)(struct svc_serv *),
e851db5b 417 svc_thread_fn, struct module *);
a7455442 418int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
03cf6c9f 419int svc_pool_stats_open(struct svc_serv *serv, struct file *file);
1da177e4 420void svc_destroy(struct svc_serv *);
6fb2b47f 421int svc_process(struct svc_rqst *);
4b62e58c
CL
422int svc_register(const struct svc_serv *, const int,
423 const unsigned short, const unsigned short);
a26cfad6 424
1da177e4
LT
425void svc_wake_up(struct svc_serv *);
426void svc_reserve(struct svc_rqst *rqstp, int space);
bfd24160 427struct svc_pool * svc_pool_for_cpu(struct svc_serv *serv, int cpu);
ad06e4bd
CL
428char * svc_print_addr(struct svc_rqst *, char *, size_t);
429
430#define RPC_MAX_ADDRBUFLEN (63U)
1da177e4 431
cd123012
JL
432/*
433 * When we want to reduce the size of the reserved space in the response
434 * buffer, we need to take into account the size of any checksum data that
435 * may be at the end of the packet. This is difficult to determine exactly
436 * for all cases without actually generating the checksum, so we just use a
437 * static value.
438 */
fbb7878c 439static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space)
cd123012 440{
fbb7878c 441 int added_space = 0;
cd123012 442
fbb7878c
BF
443 if (rqstp->rq_authop->flavour)
444 added_space = RPC_MAX_AUTH_SIZE;
445 svc_reserve(rqstp, space + added_space);
cd123012
JL
446}
447
1da177e4 448#endif /* SUNRPC_SVC_H */
This page took 0.655211 seconds and 5 git commands to generate.