1 /* SPDX-License-Identifier: GPL-2.0 */
3 * linux/include/linux/sunrpc/svc.h
5 * RPC server declarations.
7 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
15 #include <linux/in6.h>
16 #include <linux/sunrpc/types.h>
17 #include <linux/sunrpc/xdr.h>
18 #include <linux/sunrpc/auth.h>
19 #include <linux/sunrpc/svcauth.h>
20 #include <linux/wait.h>
23 /* statistics for svc_pool structures */
24 struct svc_pool_stats {
25 atomic_long_t packets;
26 unsigned long sockets_queued;
27 atomic_long_t threads_woken;
28 atomic_long_t threads_timedout;
33 * RPC service thread pool.
35 * Pool of threads and temporary sockets. Generally there is only
36 * a single one of these per RPC service, but on NUMA machines those
37 * services that can benefit from it (i.e. nfs but not lockd) will
38 * have one pool per NUMA node. This optimisation reduces cross-
39 * node traffic on multi-node NUMA NFS servers.
42 unsigned int sp_id; /* pool id; also node id on NUMA */
43 spinlock_t sp_lock; /* protects all fields */
44 struct list_head sp_sockets; /* pending sockets */
45 unsigned int sp_nrthreads; /* # of threads in pool */
46 struct list_head sp_all_threads; /* all server threads */
47 struct svc_pool_stats sp_stats; /* statistics on pool operation */
48 #define SP_TASK_PENDING (0) /* still work to do even if no
50 #define SP_CONGESTED (1)
51 unsigned long sp_flags;
52 } ____cacheline_aligned_in_smp;
57 /* Callback to use when last thread exits. */
58 void (*svo_shutdown)(struct svc_serv *, struct net *);
60 /* function for service threads to run */
61 int (*svo_function)(void *);
63 /* queue up a transport for servicing */
64 void (*svo_enqueue_xprt)(struct svc_xprt *);
66 /* set up thread (or whatever) execution context */
67 int (*svo_setup)(struct svc_serv *, struct svc_pool *, int);
69 /* optional module to count when adding threads (pooled svcs only) */
70 struct module *svo_module;
76 * An RPC service is a ``daemon,'' possibly multithreaded, which
77 * receives and processes incoming RPC messages.
78 * It has one or more transport sockets associated with it, and maintains
79 * a list of idle threads waiting for input.
81 * We currently do not support more than one RPC program per daemon.
84 struct svc_program * sv_program; /* RPC program */
85 struct svc_stat * sv_stats; /* RPC statistics */
87 unsigned int sv_nrthreads; /* # of server threads */
88 unsigned int sv_maxconn; /* max connections allowed or
89 * '0' causing max to be based
90 * on number of threads. */
92 unsigned int sv_max_payload; /* datagram payload size */
93 unsigned int sv_max_mesg; /* max_payload + 1 page for overheads */
94 unsigned int sv_xdrsize; /* XDR buffer size */
95 struct list_head sv_permsocks; /* all permanent sockets */
96 struct list_head sv_tempsocks; /* all temporary sockets */
97 int sv_tmpcnt; /* count of temporary sockets */
98 struct timer_list sv_temptimer; /* timer for aging temporary sockets */
100 char * sv_name; /* service name */
102 unsigned int sv_nrpools; /* number of thread pools */
103 struct svc_pool * sv_pools; /* array of thread pools */
104 const struct svc_serv_ops *sv_ops; /* server operations */
105 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
106 struct list_head sv_cb_list; /* queue for callback requests
107 * that arrive over the same
109 spinlock_t sv_cb_lock; /* protects the svc_cb_list */
110 wait_queue_head_t sv_cb_waitq; /* sleep here if there are no
111 * entries in the svc_cb_list */
112 bool sv_bc_enabled; /* service uses backchannel */
113 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
117 * We use sv_nrthreads as a reference count. svc_destroy() drops
118 * this refcount, so we need to bump it up around operations that
119 * change the number of threads. Horrible, but there it is.
120 * Should be called with the "service mutex" held.
122 static inline void svc_get(struct svc_serv *serv)
124 serv->sv_nrthreads++;
128 * Maximum payload size supported by a kernel RPC server.
129 * This is use to determine the max number of pages nfsd is
130 * willing to return in a single READ operation.
132 * These happen to all be powers of 2, which is not strictly
133 * necessary but helps enforce the real limitation, which is
134 * that they should be multiples of PAGE_SIZE.
136 * For UDP transports, a block plus NFS,RPC, and UDP headers
137 * has to fit into the IP datagram limit of 64K. The largest
138 * feasible number for all known page sizes is probably 48K,
139 * but we choose 32K here. This is the same as the historical
140 * Linux limit; someone who cares more about NFS/UDP performance
141 * can test a larger number.
143 * For TCP transports we have more freedom. A size of 1MB is
144 * chosen to match the client limit. Other OSes are known to
145 * have larger limits, but those numbers are probably beyond
146 * the point of diminishing returns.
148 #define RPCSVC_MAXPAYLOAD (1*1024*1024u)
149 #define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD
150 #define RPCSVC_MAXPAYLOAD_UDP (32*1024u)
152 extern u32 svc_max_payload(const struct svc_rqst *rqstp);
155 * RPC Requsts and replies are stored in one or more pages.
156 * We maintain an array of pages for each server thread.
157 * Requests are copied into these pages as they arrive. Remaining
158 * pages are available to write the reply into.
160 * Pages are sent using ->sendpage so each server thread needs to
161 * allocate more to replace those used in sending. To help keep track
162 * of these pages we have a receive list where all pages initialy live,
163 * and a send list where pages are moved to when there are to be part
166 * We use xdr_buf for holding responses as it fits well with NFS
167 * read responses (that have a header, and some data pages, and possibly
168 * a tail) and means we can share some client side routines.
170 * The xdr_buf.head kvec always points to the first page in the rq_*pages
171 * list. The xdr_buf.pages pointer points to the second page on that
172 * list. xdr_buf.tail points to the end of the first page.
173 * This assumes that the non-page part of an rpc reply will fit
174 * in a page - NFSd ensures this. lockd also has no trouble.
176 * Each request/reply pair can have at most one "payload", plus two pages,
177 * one for the request, and one for the reply.
178 * We using ->sendfile to return read data, we might need one extra page
179 * if the request is not page-aligned. So add another '1'.
181 #define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
184 static inline u32 svc_getnl(struct kvec *iov)
189 iov->iov_base = (void*)vp;
190 iov->iov_len -= sizeof(__be32);
194 static inline void svc_putnl(struct kvec *iov, u32 val)
196 __be32 *vp = iov->iov_base + iov->iov_len;
198 iov->iov_len += sizeof(__be32);
201 static inline __be32 svc_getu32(struct kvec *iov)
206 iov->iov_base = (void*)vp;
207 iov->iov_len -= sizeof(__be32);
211 static inline void svc_ungetu32(struct kvec *iov)
213 __be32 *vp = (__be32 *)iov->iov_base;
214 iov->iov_base = (void *)(vp - 1);
215 iov->iov_len += sizeof(*vp);
218 static inline void svc_putu32(struct kvec *iov, __be32 val)
220 __be32 *vp = iov->iov_base + iov->iov_len;
222 iov->iov_len += sizeof(__be32);
226 * The context of a single thread, including the request currently being
230 struct list_head rq_all; /* all threads list */
231 struct rcu_head rq_rcu_head; /* for RCU deferred kfree */
232 struct svc_xprt * rq_xprt; /* transport ptr */
234 struct sockaddr_storage rq_addr; /* peer address */
236 struct sockaddr_storage rq_daddr; /* dest addr of request
237 * - reply from here */
240 struct svc_serv * rq_server; /* RPC service definition */
241 struct svc_pool * rq_pool; /* thread pool */
242 const struct svc_procedure *rq_procinfo;/* procedure info */
243 struct auth_ops * rq_authop; /* authentication flavour */
244 struct svc_cred rq_cred; /* auth info */
245 void * rq_xprt_ctxt; /* transport specific context ptr */
246 struct svc_deferred_req*rq_deferred; /* deferred request we are replaying */
248 size_t rq_xprt_hlen; /* xprt header len */
249 struct xdr_buf rq_arg;
250 struct xdr_stream rq_arg_stream;
251 struct page *rq_scratch_page;
252 struct xdr_buf rq_res;
253 struct page *rq_pages[RPCSVC_MAXPAGES + 1];
254 struct page * *rq_respages; /* points into rq_pages */
255 struct page * *rq_next_page; /* next reply page to use */
256 struct page * *rq_page_end; /* one past the last page */
258 struct kvec rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
259 struct bio_vec rq_bvec[RPCSVC_MAXPAGES];
261 __be32 rq_xid; /* transmission id */
262 u32 rq_prog; /* program number */
263 u32 rq_vers; /* program version */
264 u32 rq_proc; /* procedure number */
265 u32 rq_prot; /* IP protocol */
266 int rq_cachetype; /* catering to nfsd */
267 #define RQ_SECURE (0) /* secure port */
268 #define RQ_LOCAL (1) /* local request */
269 #define RQ_USEDEFERRAL (2) /* use deferral */
270 #define RQ_DROPME (3) /* drop current reply */
271 #define RQ_SPLICE_OK (4) /* turned off in gss privacy
272 * to prevent encrypting page
274 #define RQ_VICTIM (5) /* about to be shut down */
275 #define RQ_BUSY (6) /* request is busy */
276 #define RQ_DATA (7) /* request has data */
277 #define RQ_AUTHERR (8) /* Request status is auth error */
278 unsigned long rq_flags; /* flags field */
279 ktime_t rq_qtime; /* enqueue time */
281 void * rq_argp; /* decoded arguments */
282 void * rq_resp; /* xdr'd results */
283 void * rq_auth_data; /* flavor-specific data */
284 int rq_auth_slack; /* extra space xdr code
285 * should leave in head
288 int rq_reserved; /* space on socket outq
289 * reserved for this request
291 ktime_t rq_stime; /* start time */
293 struct cache_req rq_chandle; /* handle passed to caches for
296 /* Catering to nfsd */
297 struct auth_domain * rq_client; /* RPC peer info */
298 struct auth_domain * rq_gssclient; /* "gss/"-style peer info */
299 struct svc_cacherep * rq_cacherep; /* cache info */
300 struct task_struct *rq_task; /* service thread */
301 spinlock_t rq_lock; /* per-request lock */
302 struct net *rq_bc_net; /* pointer to backchannel's
305 void ** rq_lease_breaker; /* The v4 client breaking a lease */
308 #define SVC_NET(rqst) (rqst->rq_xprt ? rqst->rq_xprt->xpt_net : rqst->rq_bc_net)
311 * Rigorous type checking on sockaddr type conversions
313 static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst)
315 return (struct sockaddr_in *) &rqst->rq_addr;
318 static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst)
320 return (struct sockaddr_in6 *) &rqst->rq_addr;
323 static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst)
325 return (struct sockaddr *) &rqst->rq_addr;
328 static inline struct sockaddr_in *svc_daddr_in(const struct svc_rqst *rqst)
330 return (struct sockaddr_in *) &rqst->rq_daddr;
333 static inline struct sockaddr_in6 *svc_daddr_in6(const struct svc_rqst *rqst)
335 return (struct sockaddr_in6 *) &rqst->rq_daddr;
338 static inline struct sockaddr *svc_daddr(const struct svc_rqst *rqst)
340 return (struct sockaddr *) &rqst->rq_daddr;
344 * Check buffer bounds after decoding arguments
347 xdr_argsize_check(struct svc_rqst *rqstp, __be32 *p)
349 char *cp = (char *)p;
350 struct kvec *vec = &rqstp->rq_arg.head[0];
351 return cp >= (char*)vec->iov_base
352 && cp <= (char*)vec->iov_base + vec->iov_len;
356 xdr_ressize_check(struct svc_rqst *rqstp, __be32 *p)
358 struct kvec *vec = &rqstp->rq_res.head[0];
361 vec->iov_len = cp - (char*)vec->iov_base;
363 return vec->iov_len <= PAGE_SIZE;
366 static inline void svc_free_res_pages(struct svc_rqst *rqstp)
368 while (rqstp->rq_next_page != rqstp->rq_respages) {
369 struct page **pp = --rqstp->rq_next_page;
377 struct svc_deferred_req {
378 u32 prot; /* protocol (UDP or TCP) */
379 struct svc_xprt *xprt;
380 struct sockaddr_storage addr; /* where reply must go */
382 struct sockaddr_storage daddr; /* where reply must come from */
384 struct cache_deferred_req handle;
390 struct svc_process_info {
392 int (*dispatch)(struct svc_rqst *, __be32 *);
401 * List of RPC programs on the same transport endpoint
404 struct svc_program * pg_next; /* other programs (same xprt) */
405 u32 pg_prog; /* program number */
406 unsigned int pg_lovers; /* lowest version */
407 unsigned int pg_hivers; /* highest version */
408 unsigned int pg_nvers; /* number of versions */
409 const struct svc_version **pg_vers; /* version array */
410 char * pg_name; /* service name */
411 char * pg_class; /* class name: services sharing authentication */
412 struct svc_stat * pg_stats; /* rpc statistics */
413 int (*pg_authenticate)(struct svc_rqst *);
414 __be32 (*pg_init_request)(struct svc_rqst *,
415 const struct svc_program *,
416 struct svc_process_info *);
417 int (*pg_rpcbind_set)(struct net *net,
418 const struct svc_program *,
419 u32 version, int family,
420 unsigned short proto,
421 unsigned short port);
425 * RPC program version
428 u32 vs_vers; /* version number */
429 u32 vs_nproc; /* number of procedures */
430 const struct svc_procedure *vs_proc; /* per-procedure info */
431 unsigned int *vs_count; /* call counts */
432 u32 vs_xdrsize; /* xdrsize needed for this version */
434 /* Don't register with rpcbind */
437 /* Don't care if the rpcbind registration fails */
440 /* Need xprt with congestion control */
441 bool vs_need_cong_ctrl;
443 /* Override dispatch function (e.g. when caching replies).
444 * A return value of 0 means drop the request.
445 * vs_dispatch == NULL means use default dispatcher.
447 int (*vs_dispatch)(struct svc_rqst *, __be32 *);
453 struct svc_procedure {
454 /* process the request: */
455 __be32 (*pc_func)(struct svc_rqst *);
456 /* XDR decode args: */
457 int (*pc_decode)(struct svc_rqst *, __be32 *data);
458 /* XDR encode result: */
459 int (*pc_encode)(struct svc_rqst *, __be32 *data);
460 /* XDR free result: */
461 void (*pc_release)(struct svc_rqst *);
462 unsigned int pc_argsize; /* argument struct size */
463 unsigned int pc_ressize; /* result struct size */
464 unsigned int pc_cachetype; /* cache info (NFS) */
465 unsigned int pc_xdrressize; /* maximum size of XDR reply */
469 * Mode for mapping cpus to pools.
472 SVC_POOL_AUTO = -1, /* choose one of the others */
473 SVC_POOL_GLOBAL, /* no mapping, just a single global pool
474 * (legacy & UP mode) */
475 SVC_POOL_PERCPU, /* one pool per cpu */
476 SVC_POOL_PERNODE /* one pool per numa node */
479 struct svc_pool_map {
480 int count; /* How many svc_servs use us */
481 int mode; /* Note: int not enum to avoid
482 * warnings about "enumeration value
483 * not handled in switch" */
485 unsigned int *pool_to; /* maps pool id to cpu or node */
486 unsigned int *to_pool; /* maps cpu or node to pool id */
489 extern struct svc_pool_map svc_pool_map;
492 * Function prototypes.
494 int svc_rpcb_setup(struct svc_serv *serv, struct net *net);
495 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net);
496 int svc_bind(struct svc_serv *serv, struct net *net);
497 struct svc_serv *svc_create(struct svc_program *, unsigned int,
498 const struct svc_serv_ops *);
499 struct svc_rqst *svc_rqst_alloc(struct svc_serv *serv,
500 struct svc_pool *pool, int node);
501 struct svc_rqst *svc_prepare_thread(struct svc_serv *serv,
502 struct svc_pool *pool, int node);
503 void svc_rqst_free(struct svc_rqst *);
504 void svc_exit_thread(struct svc_rqst *);
505 unsigned int svc_pool_map_get(void);
506 void svc_pool_map_put(void);
507 struct svc_serv * svc_create_pooled(struct svc_program *, unsigned int,
508 const struct svc_serv_ops *);
509 int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
510 int svc_set_num_threads_sync(struct svc_serv *, struct svc_pool *, int);
511 int svc_pool_stats_open(struct svc_serv *serv, struct file *file);
512 void svc_destroy(struct svc_serv *);
513 void svc_shutdown_net(struct svc_serv *, struct net *);
514 int svc_process(struct svc_rqst *);
515 int bc_svc_process(struct svc_serv *, struct rpc_rqst *,
517 int svc_register(const struct svc_serv *, struct net *, const int,
518 const unsigned short, const unsigned short);
520 void svc_wake_up(struct svc_serv *);
521 void svc_reserve(struct svc_rqst *rqstp, int space);
522 struct svc_pool * svc_pool_for_cpu(struct svc_serv *serv, int cpu);
523 char * svc_print_addr(struct svc_rqst *, char *, size_t);
524 int svc_encode_result_payload(struct svc_rqst *rqstp,
526 unsigned int length);
527 unsigned int svc_fill_write_vector(struct svc_rqst *rqstp,
529 struct kvec *first, size_t total);
530 char *svc_fill_symlink_pathname(struct svc_rqst *rqstp,
531 struct kvec *first, void *p,
533 __be32 svc_return_autherr(struct svc_rqst *rqstp, __be32 auth_err);
534 __be32 svc_generic_init_request(struct svc_rqst *rqstp,
535 const struct svc_program *progp,
536 struct svc_process_info *procinfo);
537 int svc_generic_rpcbind_set(struct net *net,
538 const struct svc_program *progp,
539 u32 version, int family,
540 unsigned short proto,
541 unsigned short port);
542 int svc_rpcbind_set_version(struct net *net,
543 const struct svc_program *progp,
544 u32 version, int family,
545 unsigned short proto,
546 unsigned short port);
548 #define RPC_MAX_ADDRBUFLEN (63U)
551 * When we want to reduce the size of the reserved space in the response
552 * buffer, we need to take into account the size of any checksum data that
553 * may be at the end of the packet. This is difficult to determine exactly
554 * for all cases without actually generating the checksum, so we just use a
557 static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space)
559 svc_reserve(rqstp, space + rqstp->rq_auth_slack);
563 * svcxdr_init_decode - Prepare an xdr_stream for svc Call decoding
564 * @rqstp: controlling server RPC transaction context
567 static inline void svcxdr_init_decode(struct svc_rqst *rqstp)
569 struct xdr_stream *xdr = &rqstp->rq_arg_stream;
570 struct kvec *argv = rqstp->rq_arg.head;
572 xdr_init_decode(xdr, &rqstp->rq_arg, argv->iov_base, NULL);
573 xdr_set_scratch_page(xdr, rqstp->rq_scratch_page);
576 #endif /* SUNRPC_SVC_H */