Commit | Line | Data |
---|---|---|
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> | |
14 | #include <linux/sunrpc/types.h> | |
15 | #include <linux/sunrpc/xdr.h> | |
7adae489 | 16 | #include <linux/sunrpc/auth.h> |
1da177e4 LT |
17 | #include <linux/sunrpc/svcauth.h> |
18 | #include <linux/wait.h> | |
19 | #include <linux/mm.h> | |
20 | ||
a7455442 GB |
21 | /* |
22 | * This is the RPC server thread function prototype | |
23 | */ | |
24 | typedef void (*svc_thread_fn)(struct svc_rqst *); | |
3262c816 GB |
25 | |
26 | /* | |
27 | * | |
28 | * RPC service thread pool. | |
29 | * | |
30 | * Pool of threads and temporary sockets. Generally there is only | |
31 | * a single one of these per RPC service, but on NUMA machines those | |
32 | * services that can benefit from it (i.e. nfs but not lockd) will | |
33 | * have one pool per NUMA node. This optimisation reduces cross- | |
34 | * node traffic on multi-node NUMA NFS servers. | |
35 | */ | |
36 | struct svc_pool { | |
37 | unsigned int sp_id; /* pool id; also node id on NUMA */ | |
38 | spinlock_t sp_lock; /* protects all fields */ | |
39 | struct list_head sp_threads; /* idle server threads */ | |
40 | struct list_head sp_sockets; /* pending sockets */ | |
41 | unsigned int sp_nrthreads; /* # of threads in pool */ | |
a7455442 | 42 | struct list_head sp_all_threads; /* all server threads */ |
3262c816 GB |
43 | } ____cacheline_aligned_in_smp; |
44 | ||
1da177e4 LT |
45 | /* |
46 | * RPC service. | |
47 | * | |
48 | * An RPC service is a ``daemon,'' possibly multithreaded, which | |
49 | * receives and processes incoming RPC messages. | |
50 | * It has one or more transport sockets associated with it, and maintains | |
51 | * a list of idle threads waiting for input. | |
52 | * | |
53 | * We currently do not support more than one RPC program per daemon. | |
54 | */ | |
55 | struct svc_serv { | |
1da177e4 LT |
56 | struct svc_program * sv_program; /* RPC program */ |
57 | struct svc_stat * sv_stats; /* RPC statistics */ | |
58 | spinlock_t sv_lock; | |
59 | unsigned int sv_nrthreads; /* # of server threads */ | |
c6b0a9f8 N |
60 | unsigned int sv_max_payload; /* datagram payload size */ |
61 | unsigned int sv_max_mesg; /* max_payload + 1 page for overheads */ | |
1da177e4 LT |
62 | unsigned int sv_xdrsize; /* XDR buffer size */ |
63 | ||
64 | struct list_head sv_permsocks; /* all permanent sockets */ | |
65 | struct list_head sv_tempsocks; /* all temporary sockets */ | |
66 | int sv_tmpcnt; /* count of temporary sockets */ | |
36bdfc8b | 67 | struct timer_list sv_temptimer; /* timer for aging temporary sockets */ |
1da177e4 LT |
68 | |
69 | char * sv_name; /* service name */ | |
bc591ccf | 70 | |
3262c816 GB |
71 | unsigned int sv_nrpools; /* number of thread pools */ |
72 | struct svc_pool * sv_pools; /* array of thread pools */ | |
73 | ||
bc591ccf N |
74 | void (*sv_shutdown)(struct svc_serv *serv); |
75 | /* Callback to use when last thread | |
76 | * exits. | |
77 | */ | |
a7455442 GB |
78 | |
79 | struct module * sv_module; /* optional module to count when | |
80 | * adding threads */ | |
81 | svc_thread_fn sv_function; /* main function for threads */ | |
82 | int sv_kill_signal; /* signal to kill threads */ | |
1da177e4 LT |
83 | }; |
84 | ||
9a24ab57 GB |
85 | /* |
86 | * We use sv_nrthreads as a reference count. svc_destroy() drops | |
87 | * this refcount, so we need to bump it up around operations that | |
88 | * change the number of threads. Horrible, but there it is. | |
89 | * Should be called with the BKL held. | |
90 | */ | |
91 | static inline void svc_get(struct svc_serv *serv) | |
92 | { | |
93 | serv->sv_nrthreads++; | |
94 | } | |
95 | ||
1da177e4 LT |
96 | /* |
97 | * Maximum payload size supported by a kernel RPC server. | |
98 | * This is use to determine the max number of pages nfsd is | |
99 | * willing to return in a single READ operation. | |
7adae489 GB |
100 | * |
101 | * These happen to all be powers of 2, which is not strictly | |
102 | * necessary but helps enforce the real limitation, which is | |
103 | * that they should be multiples of PAGE_CACHE_SIZE. | |
104 | * | |
105 | * For UDP transports, a block plus NFS,RPC, and UDP headers | |
106 | * has to fit into the IP datagram limit of 64K. The largest | |
107 | * feasible number for all known page sizes is probably 48K, | |
108 | * but we choose 32K here. This is the same as the historical | |
109 | * Linux limit; someone who cares more about NFS/UDP performance | |
110 | * can test a larger number. | |
111 | * | |
112 | * For TCP transports we have more freedom. A size of 1MB is | |
113 | * chosen to match the client limit. Other OSes are known to | |
114 | * have larger limits, but those numbers are probably beyond | |
115 | * the point of diminishing returns. | |
1da177e4 | 116 | */ |
7adae489 GB |
117 | #define RPCSVC_MAXPAYLOAD (1*1024*1024u) |
118 | #define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD | |
119 | #define RPCSVC_MAXPAYLOAD_UDP (32*1024u) | |
120 | ||
121 | extern u32 svc_max_payload(const struct svc_rqst *rqstp); | |
1da177e4 LT |
122 | |
123 | /* | |
124 | * RPC Requsts and replies are stored in one or more pages. | |
125 | * We maintain an array of pages for each server thread. | |
126 | * Requests are copied into these pages as they arrive. Remaining | |
127 | * pages are available to write the reply into. | |
128 | * | |
129 | * Pages are sent using ->sendpage so each server thread needs to | |
130 | * allocate more to replace those used in sending. To help keep track | |
131 | * of these pages we have a receive list where all pages initialy live, | |
132 | * and a send list where pages are moved to when there are to be part | |
133 | * of a reply. | |
134 | * | |
135 | * We use xdr_buf for holding responses as it fits well with NFS | |
136 | * read responses (that have a header, and some data pages, and possibly | |
137 | * a tail) and means we can share some client side routines. | |
138 | * | |
139 | * The xdr_buf.head kvec always points to the first page in the rq_*pages | |
140 | * list. The xdr_buf.pages pointer points to the second page on that | |
141 | * list. xdr_buf.tail points to the end of the first page. | |
142 | * This assumes that the non-page part of an rpc reply will fit | |
143 | * in a page - NFSd ensures this. lockd also has no trouble. | |
144 | * | |
145 | * Each request/reply pair can have at most one "payload", plus two pages, | |
146 | * one for the request, and one for the reply. | |
147 | */ | |
148 | #define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE + 2) | |
149 | ||
76994313 | 150 | static inline u32 svc_getnl(struct kvec *iov) |
1da177e4 | 151 | { |
76994313 | 152 | __be32 val, *vp; |
1da177e4 LT |
153 | vp = iov->iov_base; |
154 | val = *vp++; | |
155 | iov->iov_base = (void*)vp; | |
76994313 AD |
156 | iov->iov_len -= sizeof(__be32); |
157 | return ntohl(val); | |
158 | } | |
159 | ||
160 | static inline void svc_putnl(struct kvec *iov, u32 val) | |
161 | { | |
162 | __be32 *vp = iov->iov_base + iov->iov_len; | |
163 | *vp = htonl(val); | |
164 | iov->iov_len += sizeof(__be32); | |
165 | } | |
166 | ||
167 | static inline __be32 svc_getu32(struct kvec *iov) | |
168 | { | |
169 | __be32 val, *vp; | |
170 | vp = iov->iov_base; | |
171 | val = *vp++; | |
172 | iov->iov_base = (void*)vp; | |
173 | iov->iov_len -= sizeof(__be32); | |
1da177e4 LT |
174 | return val; |
175 | } | |
176 | ||
177 | static inline void svc_ungetu32(struct kvec *iov) | |
178 | { | |
76994313 | 179 | __be32 *vp = (__be32 *)iov->iov_base; |
1da177e4 LT |
180 | iov->iov_base = (void *)(vp - 1); |
181 | iov->iov_len += sizeof(*vp); | |
182 | } | |
183 | ||
76994313 | 184 | static inline void svc_putu32(struct kvec *iov, __be32 val) |
1da177e4 | 185 | { |
76994313 | 186 | __be32 *vp = iov->iov_base + iov->iov_len; |
1da177e4 | 187 | *vp = val; |
76994313 | 188 | iov->iov_len += sizeof(__be32); |
1da177e4 LT |
189 | } |
190 | ||
191 | ||
192 | /* | |
193 | * The context of a single thread, including the request currently being | |
194 | * processed. | |
1da177e4 LT |
195 | */ |
196 | struct svc_rqst { | |
197 | struct list_head rq_list; /* idle list */ | |
a7455442 | 198 | struct list_head rq_all; /* all threads list */ |
1da177e4 LT |
199 | struct svc_sock * rq_sock; /* socket */ |
200 | struct sockaddr_in rq_addr; /* peer address */ | |
201 | int rq_addrlen; | |
202 | ||
203 | struct svc_serv * rq_server; /* RPC service definition */ | |
3262c816 | 204 | struct svc_pool * rq_pool; /* thread pool */ |
1da177e4 LT |
205 | struct svc_procedure * rq_procinfo; /* procedure info */ |
206 | struct auth_ops * rq_authop; /* authentication flavour */ | |
207 | struct svc_cred rq_cred; /* auth info */ | |
208 | struct sk_buff * rq_skbuff; /* fast recv inet buffer */ | |
209 | struct svc_deferred_req*rq_deferred; /* deferred request we are replaying */ | |
210 | ||
211 | struct xdr_buf rq_arg; | |
212 | struct xdr_buf rq_res; | |
44524359 N |
213 | struct page * rq_pages[RPCSVC_MAXPAGES]; |
214 | struct page * *rq_respages; /* points into rq_pages */ | |
215 | int rq_resused; /* number of pages used for result */ | |
1da177e4 | 216 | |
3cc03b16 N |
217 | struct kvec rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */ |
218 | ||
d8ed029d | 219 | __be32 rq_xid; /* transmission id */ |
1da177e4 LT |
220 | u32 rq_prog; /* program number */ |
221 | u32 rq_vers; /* program version */ | |
222 | u32 rq_proc; /* procedure number */ | |
223 | u32 rq_prot; /* IP protocol */ | |
224 | unsigned short | |
225 | rq_secure : 1; /* secure port */ | |
226 | ||
227 | ||
d8ed029d | 228 | __be32 rq_daddr; /* dest addr of request - reply from here */ |
1da177e4 LT |
229 | |
230 | void * rq_argp; /* decoded arguments */ | |
231 | void * rq_resp; /* xdr'd results */ | |
232 | void * rq_auth_data; /* flavor-specific data */ | |
233 | ||
234 | int rq_reserved; /* space on socket outq | |
235 | * reserved for this request | |
236 | */ | |
237 | ||
238 | struct cache_req rq_chandle; /* handle passed to caches for | |
239 | * request delaying | |
240 | */ | |
241 | /* Catering to nfsd */ | |
242 | struct auth_domain * rq_client; /* RPC peer info */ | |
243 | struct svc_cacherep * rq_cacherep; /* cache info */ | |
244 | struct knfsd_fh * rq_reffh; /* Referrence filehandle, used to | |
245 | * determine what device number | |
246 | * to report (real or virtual) | |
247 | */ | |
5c04c46a BF |
248 | int rq_sendfile_ok; /* turned off in gss privacy |
249 | * to prevent encrypting page | |
250 | * cache pages */ | |
1da177e4 | 251 | wait_queue_head_t rq_wait; /* synchronization */ |
a7455442 | 252 | struct task_struct *rq_task; /* service thread */ |
1da177e4 LT |
253 | }; |
254 | ||
255 | /* | |
256 | * Check buffer bounds after decoding arguments | |
257 | */ | |
258 | static inline int | |
d8ed029d | 259 | xdr_argsize_check(struct svc_rqst *rqstp, __be32 *p) |
1da177e4 LT |
260 | { |
261 | char *cp = (char *)p; | |
262 | struct kvec *vec = &rqstp->rq_arg.head[0]; | |
0ba7536d N |
263 | return cp >= (char*)vec->iov_base |
264 | && cp <= (char*)vec->iov_base + vec->iov_len; | |
1da177e4 LT |
265 | } |
266 | ||
267 | static inline int | |
d8ed029d | 268 | xdr_ressize_check(struct svc_rqst *rqstp, __be32 *p) |
1da177e4 LT |
269 | { |
270 | struct kvec *vec = &rqstp->rq_res.head[0]; | |
271 | char *cp = (char*)p; | |
272 | ||
273 | vec->iov_len = cp - (char*)vec->iov_base; | |
274 | ||
275 | return vec->iov_len <= PAGE_SIZE; | |
276 | } | |
277 | ||
44524359 | 278 | static inline void svc_free_res_pages(struct svc_rqst *rqstp) |
a257cdd0 | 279 | { |
44524359 N |
280 | while (rqstp->rq_resused) { |
281 | struct page **pp = (rqstp->rq_respages + | |
282 | --rqstp->rq_resused); | |
283 | if (*pp) { | |
284 | put_page(*pp); | |
285 | *pp = NULL; | |
1da177e4 LT |
286 | } |
287 | } | |
288 | } | |
289 | ||
1da177e4 LT |
290 | struct svc_deferred_req { |
291 | u32 prot; /* protocol (UDP or TCP) */ | |
292 | struct sockaddr_in addr; | |
293 | struct svc_sock *svsk; /* where reply must go */ | |
d8ed029d | 294 | __be32 daddr; /* where reply must come from */ |
1da177e4 LT |
295 | struct cache_deferred_req handle; |
296 | int argslen; | |
d8ed029d | 297 | __be32 args[0]; |
1da177e4 LT |
298 | }; |
299 | ||
300 | /* | |
9ba02638 | 301 | * List of RPC programs on the same transport endpoint |
1da177e4 LT |
302 | */ |
303 | struct svc_program { | |
9ba02638 | 304 | struct svc_program * pg_next; /* other programs (same xprt) */ |
1da177e4 LT |
305 | u32 pg_prog; /* program number */ |
306 | unsigned int pg_lovers; /* lowest version */ | |
307 | unsigned int pg_hivers; /* lowest version */ | |
308 | unsigned int pg_nvers; /* number of versions */ | |
309 | struct svc_version ** pg_vers; /* version array */ | |
310 | char * pg_name; /* service name */ | |
311 | char * pg_class; /* class name: services sharing authentication */ | |
312 | struct svc_stat * pg_stats; /* rpc statistics */ | |
313 | int (*pg_authenticate)(struct svc_rqst *); | |
314 | }; | |
315 | ||
316 | /* | |
317 | * RPC program version | |
318 | */ | |
319 | struct svc_version { | |
320 | u32 vs_vers; /* version number */ | |
321 | u32 vs_nproc; /* number of procedures */ | |
322 | struct svc_procedure * vs_proc; /* per-procedure info */ | |
323 | u32 vs_xdrsize; /* xdrsize needed for this version */ | |
324 | ||
bc5fea42 OK |
325 | unsigned int vs_hidden : 1; /* Don't register with portmapper. |
326 | * Only used for nfsacl so far. */ | |
327 | ||
1da177e4 LT |
328 | /* Override dispatch function (e.g. when caching replies). |
329 | * A return value of 0 means drop the request. | |
330 | * vs_dispatch == NULL means use default dispatcher. | |
331 | */ | |
d8ed029d | 332 | int (*vs_dispatch)(struct svc_rqst *, __be32 *); |
1da177e4 LT |
333 | }; |
334 | ||
335 | /* | |
336 | * RPC procedure info | |
337 | */ | |
7111c66e | 338 | typedef __be32 (*svc_procfunc)(struct svc_rqst *, void *argp, void *resp); |
1da177e4 LT |
339 | struct svc_procedure { |
340 | svc_procfunc pc_func; /* process the request */ | |
341 | kxdrproc_t pc_decode; /* XDR decode args */ | |
342 | kxdrproc_t pc_encode; /* XDR encode result */ | |
343 | kxdrproc_t pc_release; /* XDR free result */ | |
344 | unsigned int pc_argsize; /* argument struct size */ | |
345 | unsigned int pc_ressize; /* result struct size */ | |
346 | unsigned int pc_count; /* call count */ | |
347 | unsigned int pc_cachetype; /* cache info (NFS) */ | |
348 | unsigned int pc_xdrressize; /* maximum size of XDR reply */ | |
349 | }; | |
350 | ||
1da177e4 LT |
351 | /* |
352 | * Function prototypes. | |
353 | */ | |
bc591ccf N |
354 | struct svc_serv * svc_create(struct svc_program *, unsigned int, |
355 | void (*shutdown)(struct svc_serv*)); | |
1da177e4 LT |
356 | int svc_create_thread(svc_thread_fn, struct svc_serv *); |
357 | void svc_exit_thread(struct svc_rqst *); | |
a7455442 GB |
358 | struct svc_serv * svc_create_pooled(struct svc_program *, unsigned int, |
359 | void (*shutdown)(struct svc_serv*), | |
360 | svc_thread_fn, int sig, struct module *); | |
361 | int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int); | |
1da177e4 | 362 | void svc_destroy(struct svc_serv *); |
6fb2b47f | 363 | int svc_process(struct svc_rqst *); |
1da177e4 LT |
364 | int svc_register(struct svc_serv *, int, unsigned short); |
365 | void svc_wake_up(struct svc_serv *); | |
366 | void svc_reserve(struct svc_rqst *rqstp, int space); | |
bfd24160 | 367 | struct svc_pool * svc_pool_for_cpu(struct svc_serv *serv, int cpu); |
1da177e4 LT |
368 | |
369 | #endif /* SUNRPC_SVC_H */ |