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>
22 /* statistics for svc_pool structures */
23 struct svc_pool_stats {
24 atomic_long_t packets;
25 unsigned long sockets_queued;
26 atomic_long_t threads_woken;
27 atomic_long_t threads_timedout;
32 * RPC service thread pool.
34 * Pool of threads and temporary sockets. Generally there is only
35 * a single one of these per RPC service, but on NUMA machines those
36 * services that can benefit from it (i.e. nfs but not lockd) will
37 * have one pool per NUMA node. This optimisation reduces cross-
38 * node traffic on multi-node NUMA NFS servers.
41 unsigned int sp_id; /* pool id; also node id on NUMA */
42 spinlock_t sp_lock; /* protects all fields */
43 struct list_head sp_sockets; /* pending sockets */
44 unsigned int sp_nrthreads; /* # of threads in pool */
45 struct list_head sp_all_threads; /* all server threads */
46 struct svc_pool_stats sp_stats; /* statistics on pool operation */
47 #define SP_TASK_PENDING (0) /* still work to do even if no
49 #define SP_CONGESTED (1)
50 unsigned long sp_flags;
51 } ____cacheline_aligned_in_smp;
56 /* Callback to use when last thread exits. */
57 void (*svo_shutdown)(struct svc_serv *, struct net *);
59 /* function for service threads to run */
60 int (*svo_function)(void *);
62 /* queue up a transport for servicing */
63 void (*svo_enqueue_xprt)(struct svc_xprt *);
65 /* set up thread (or whatever) execution context */
66 int (*svo_setup)(struct svc_serv *, struct svc_pool *, int);
68 /* optional module to count when adding threads (pooled svcs only) */
69 struct module *svo_module;
75 * An RPC service is a ``daemon,'' possibly multithreaded, which
76 * receives and processes incoming RPC messages.
77 * It has one or more transport sockets associated with it, and maintains
78 * a list of idle threads waiting for input.
80 * We currently do not support more than one RPC program per daemon.
83 struct svc_program * sv_program; /* RPC program */
84 struct svc_stat * sv_stats; /* RPC statistics */
86 unsigned int sv_nrthreads; /* # of server threads */
87 unsigned int sv_maxconn; /* max connections allowed or
88 * '0' causing max to be based
89 * on number of threads. */
91 unsigned int sv_max_payload; /* datagram payload size */
92 unsigned int sv_max_mesg; /* max_payload + 1 page for overheads */
93 unsigned int sv_xdrsize; /* XDR buffer size */
94 struct list_head sv_permsocks; /* all permanent sockets */
95 struct list_head sv_tempsocks; /* all temporary sockets */
96 int sv_tmpcnt; /* count of temporary sockets */
97 struct timer_list sv_temptimer; /* timer for aging temporary sockets */
99 char * sv_name; /* service name */
101 unsigned int sv_nrpools; /* number of thread pools */
102 struct svc_pool * sv_pools; /* array of thread pools */
103 const struct svc_serv_ops *sv_ops; /* server operations */
104 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
105 struct list_head sv_cb_list; /* queue for callback requests
106 * that arrive over the same
108 spinlock_t sv_cb_lock; /* protects the svc_cb_list */
109 wait_queue_head_t sv_cb_waitq; /* sleep here if there are no
110 * entries in the svc_cb_list */
111 struct svc_xprt *sv_bc_xprt; /* callback on fore channel */
112 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
116 * We use sv_nrthreads as a reference count. svc_destroy() drops
117 * this refcount, so we need to bump it up around operations that
118 * change the number of threads. Horrible, but there it is.
119 * Should be called with the "service mutex" held.
121 static inline void svc_get(struct svc_serv *serv)
123 serv->sv_nrthreads++;
127 * Maximum payload size supported by a kernel RPC server.
128 * This is use to determine the max number of pages nfsd is
129 * willing to return in a single READ operation.
131 * These happen to all be powers of 2, which is not strictly
132 * necessary but helps enforce the real limitation, which is
133 * that they should be multiples of PAGE_SIZE.
135 * For UDP transports, a block plus NFS,RPC, and UDP headers
136 * has to fit into the IP datagram limit of 64K. The largest
137 * feasible number for all known page sizes is probably 48K,
138 * but we choose 32K here. This is the same as the historical
139 * Linux limit; someone who cares more about NFS/UDP performance
140 * can test a larger number.
142 * For TCP transports we have more freedom. A size of 1MB is
143 * chosen to match the client limit. Other OSes are known to
144 * have larger limits, but those numbers are probably beyond
145 * the point of diminishing returns.
147 #define RPCSVC_MAXPAYLOAD (1*1024*1024u)
148 #define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD
149 #define RPCSVC_MAXPAYLOAD_UDP (32*1024u)
151 extern u32 svc_max_payload(const struct svc_rqst *rqstp);
154 * RPC Requsts and replies are stored in one or more pages.
155 * We maintain an array of pages for each server thread.
156 * Requests are copied into these pages as they arrive. Remaining
157 * pages are available to write the reply into.
159 * Pages are sent using ->sendpage so each server thread needs to
160 * allocate more to replace those used in sending. To help keep track
161 * of these pages we have a receive list where all pages initialy live,
162 * and a send list where pages are moved to when there are to be part
165 * We use xdr_buf for holding responses as it fits well with NFS
166 * read responses (that have a header, and some data pages, and possibly
167 * a tail) and means we can share some client side routines.
169 * The xdr_buf.head kvec always points to the first page in the rq_*pages
170 * list. The xdr_buf.pages pointer points to the second page on that
171 * list. xdr_buf.tail points to the end of the first page.
172 * This assumes that the non-page part of an rpc reply will fit
173 * in a page - NFSd ensures this. lockd also has no trouble.
175 * Each request/reply pair can have at most one "payload", plus two pages,
176 * one for the request, and one for the reply.
177 * We using ->sendfile to return read data, we might need one extra page
178 * if the request is not page-aligned. So add another '1'.
180 #define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
183 static inline u32 svc_getnl(struct kvec *iov)
188 iov->iov_base = (void*)vp;
189 iov->iov_len -= sizeof(__be32);
193 static inline void svc_putnl(struct kvec *iov, u32 val)
195 __be32 *vp = iov->iov_base + iov->iov_len;
197 iov->iov_len += sizeof(__be32);
200 static inline __be32 svc_getu32(struct kvec *iov)
205 iov->iov_base = (void*)vp;
206 iov->iov_len -= sizeof(__be32);
210 static inline void svc_ungetu32(struct kvec *iov)
212 __be32 *vp = (__be32 *)iov->iov_base;
213 iov->iov_base = (void *)(vp - 1);
214 iov->iov_len += sizeof(*vp);
217 static inline void svc_putu32(struct kvec *iov, __be32 val)
219 __be32 *vp = iov->iov_base + iov->iov_len;
221 iov->iov_len += sizeof(__be32);
225 * The context of a single thread, including the request currently being
229 struct list_head rq_all; /* all threads list */
230 struct rcu_head rq_rcu_head; /* for RCU deferred kfree */
231 struct svc_xprt * rq_xprt; /* transport ptr */
233 struct sockaddr_storage rq_addr; /* peer address */
235 struct sockaddr_storage rq_daddr; /* dest addr of request
236 * - reply from here */
239 struct svc_serv * rq_server; /* RPC service definition */
240 struct svc_pool * rq_pool; /* thread pool */
241 const struct svc_procedure *rq_procinfo;/* procedure info */
242 struct auth_ops * rq_authop; /* authentication flavour */
243 struct svc_cred rq_cred; /* auth info */
244 void * rq_xprt_ctxt; /* transport specific context ptr */
245 struct svc_deferred_req*rq_deferred; /* deferred request we are replaying */
247 size_t rq_xprt_hlen; /* xprt header len */
248 struct xdr_buf rq_arg;
249 struct xdr_buf rq_res;
250 struct page *rq_pages[RPCSVC_MAXPAGES + 1];
251 struct page * *rq_respages; /* points into rq_pages */
252 struct page * *rq_next_page; /* next reply page to use */
253 struct page * *rq_page_end; /* one past the last page */
255 struct kvec rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
257 __be32 rq_xid; /* transmission id */
258 u32 rq_prog; /* program number */
259 u32 rq_vers; /* program version */
260 u32 rq_proc; /* procedure number */
261 u32 rq_prot; /* IP protocol */
262 int rq_cachetype; /* catering to nfsd */
263 #define RQ_SECURE (0) /* secure port */
264 #define RQ_LOCAL (1) /* local request */
265 #define RQ_USEDEFERRAL (2) /* use deferral */
266 #define RQ_DROPME (3) /* drop current reply */
267 #define RQ_SPLICE_OK (4) /* turned off in gss privacy
268 * to prevent encrypting page
270 #define RQ_VICTIM (5) /* about to be shut down */
271 #define RQ_BUSY (6) /* request is busy */
272 #define RQ_DATA (7) /* request has data */
273 unsigned long rq_flags; /* flags field */
275 void * rq_argp; /* decoded arguments */
276 void * rq_resp; /* xdr'd results */
277 void * rq_auth_data; /* flavor-specific data */
278 int rq_auth_slack; /* extra space xdr code
279 * should leave in head
282 int rq_reserved; /* space on socket outq
283 * reserved for this request
286 struct cache_req rq_chandle; /* handle passed to caches for
289 /* Catering to nfsd */
290 struct auth_domain * rq_client; /* RPC peer info */
291 struct auth_domain * rq_gssclient; /* "gss/"-style peer info */
292 struct svc_cacherep * rq_cacherep; /* cache info */
293 struct task_struct *rq_task; /* service thread */
294 spinlock_t rq_lock; /* per-request lock */
297 #define SVC_NET(svc_rqst) (svc_rqst->rq_xprt->xpt_net)
300 * Rigorous type checking on sockaddr type conversions
302 static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst)
304 return (struct sockaddr_in *) &rqst->rq_addr;
307 static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst)
309 return (struct sockaddr_in6 *) &rqst->rq_addr;
312 static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst)
314 return (struct sockaddr *) &rqst->rq_addr;
317 static inline struct sockaddr_in *svc_daddr_in(const struct svc_rqst *rqst)
319 return (struct sockaddr_in *) &rqst->rq_daddr;
322 static inline struct sockaddr_in6 *svc_daddr_in6(const struct svc_rqst *rqst)
324 return (struct sockaddr_in6 *) &rqst->rq_daddr;
327 static inline struct sockaddr *svc_daddr(const struct svc_rqst *rqst)
329 return (struct sockaddr *) &rqst->rq_daddr;
333 * Check buffer bounds after decoding arguments
336 xdr_argsize_check(struct svc_rqst *rqstp, __be32 *p)
338 char *cp = (char *)p;
339 struct kvec *vec = &rqstp->rq_arg.head[0];
340 return cp >= (char*)vec->iov_base
341 && cp <= (char*)vec->iov_base + vec->iov_len;
345 xdr_ressize_check(struct svc_rqst *rqstp, __be32 *p)
347 struct kvec *vec = &rqstp->rq_res.head[0];
350 vec->iov_len = cp - (char*)vec->iov_base;
352 return vec->iov_len <= PAGE_SIZE;
355 static inline void svc_free_res_pages(struct svc_rqst *rqstp)
357 while (rqstp->rq_next_page != rqstp->rq_respages) {
358 struct page **pp = --rqstp->rq_next_page;
366 struct svc_deferred_req {
367 u32 prot; /* protocol (UDP or TCP) */
368 struct svc_xprt *xprt;
369 struct sockaddr_storage addr; /* where reply must go */
371 struct sockaddr_storage daddr; /* where reply must come from */
373 struct cache_deferred_req handle;
380 * List of RPC programs on the same transport endpoint
383 struct svc_program * pg_next; /* other programs (same xprt) */
384 u32 pg_prog; /* program number */
385 unsigned int pg_lovers; /* lowest version */
386 unsigned int pg_hivers; /* highest version */
387 unsigned int pg_nvers; /* number of versions */
388 const struct svc_version **pg_vers; /* version array */
389 char * pg_name; /* service name */
390 char * pg_class; /* class name: services sharing authentication */
391 struct svc_stat * pg_stats; /* rpc statistics */
392 int (*pg_authenticate)(struct svc_rqst *);
396 * RPC program version
399 u32 vs_vers; /* version number */
400 u32 vs_nproc; /* number of procedures */
401 const struct svc_procedure *vs_proc; /* per-procedure info */
402 unsigned int *vs_count; /* call counts */
403 u32 vs_xdrsize; /* xdrsize needed for this version */
405 /* Don't register with rpcbind */
408 /* Don't care if the rpcbind registration fails */
411 /* Need xprt with congestion control */
412 bool vs_need_cong_ctrl;
414 /* Override dispatch function (e.g. when caching replies).
415 * A return value of 0 means drop the request.
416 * vs_dispatch == NULL means use default dispatcher.
418 int (*vs_dispatch)(struct svc_rqst *, __be32 *);
424 struct svc_procedure {
425 /* process the request: */
426 __be32 (*pc_func)(struct svc_rqst *);
427 /* XDR decode args: */
428 int (*pc_decode)(struct svc_rqst *, __be32 *data);
429 /* XDR encode result: */
430 int (*pc_encode)(struct svc_rqst *, __be32 *data);
431 /* XDR free result: */
432 void (*pc_release)(struct svc_rqst *);
433 unsigned int pc_argsize; /* argument struct size */
434 unsigned int pc_ressize; /* result struct size */
435 unsigned int pc_cachetype; /* cache info (NFS) */
436 unsigned int pc_xdrressize; /* maximum size of XDR reply */
440 * Mode for mapping cpus to pools.
443 SVC_POOL_AUTO = -1, /* choose one of the others */
444 SVC_POOL_GLOBAL, /* no mapping, just a single global pool
445 * (legacy & UP mode) */
446 SVC_POOL_PERCPU, /* one pool per cpu */
447 SVC_POOL_PERNODE /* one pool per numa node */
450 struct svc_pool_map {
451 int count; /* How many svc_servs use us */
452 int mode; /* Note: int not enum to avoid
453 * warnings about "enumeration value
454 * not handled in switch" */
456 unsigned int *pool_to; /* maps pool id to cpu or node */
457 unsigned int *to_pool; /* maps cpu or node to pool id */
460 extern struct svc_pool_map svc_pool_map;
463 * Function prototypes.
465 int svc_rpcb_setup(struct svc_serv *serv, struct net *net);
466 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net);
467 int svc_bind(struct svc_serv *serv, struct net *net);
468 struct svc_serv *svc_create(struct svc_program *, unsigned int,
469 const struct svc_serv_ops *);
470 struct svc_rqst *svc_rqst_alloc(struct svc_serv *serv,
471 struct svc_pool *pool, int node);
472 struct svc_rqst *svc_prepare_thread(struct svc_serv *serv,
473 struct svc_pool *pool, int node);
474 void svc_rqst_free(struct svc_rqst *);
475 void svc_exit_thread(struct svc_rqst *);
476 unsigned int svc_pool_map_get(void);
477 void svc_pool_map_put(void);
478 struct svc_serv * svc_create_pooled(struct svc_program *, unsigned int,
479 const struct svc_serv_ops *);
480 int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
481 int svc_set_num_threads_sync(struct svc_serv *, struct svc_pool *, int);
482 int svc_pool_stats_open(struct svc_serv *serv, struct file *file);
483 void svc_destroy(struct svc_serv *);
484 void svc_shutdown_net(struct svc_serv *, struct net *);
485 int svc_process(struct svc_rqst *);
486 int bc_svc_process(struct svc_serv *, struct rpc_rqst *,
488 int svc_register(const struct svc_serv *, struct net *, const int,
489 const unsigned short, const unsigned short);
491 void svc_wake_up(struct svc_serv *);
492 void svc_reserve(struct svc_rqst *rqstp, int space);
493 struct svc_pool * svc_pool_for_cpu(struct svc_serv *serv, int cpu);
494 char * svc_print_addr(struct svc_rqst *, char *, size_t);
496 #define RPC_MAX_ADDRBUFLEN (63U)
499 * When we want to reduce the size of the reserved space in the response
500 * buffer, we need to take into account the size of any checksum data that
501 * may be at the end of the packet. This is difficult to determine exactly
502 * for all cases without actually generating the checksum, so we just use a
505 static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space)
507 svc_reserve(rqstp, space + rqstp->rq_auth_slack);
510 #endif /* SUNRPC_SVC_H */