1 /* SPDX-License-Identifier: GPL-2.0 */
3 * linux/include/linux/sunrpc/svc.h
5 * RPC server declarations.
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>
22 #include <linux/pagevec.h>
26 * RPC service thread pool.
28 * Pool of threads and temporary sockets. Generally there is only
29 * a single one of these per RPC service, but on NUMA machines those
30 * services that can benefit from it (i.e. nfs but not lockd) will
31 * have one pool per NUMA node. This optimisation reduces cross-
32 * node traffic on multi-node NUMA NFS servers.
35 unsigned int sp_id; /* pool id; also node id on NUMA */
36 spinlock_t sp_lock; /* protects all fields */
37 struct list_head sp_sockets; /* pending sockets */
38 unsigned int sp_nrthreads; /* # of threads in pool */
39 struct list_head sp_all_threads; /* all server threads */
41 /* statistics on pool operation */
42 struct percpu_counter sp_sockets_queued;
43 struct percpu_counter sp_threads_woken;
44 struct percpu_counter sp_threads_timedout;
46 #define SP_TASK_PENDING (0) /* still work to do even if no
48 #define SP_CONGESTED (1)
49 unsigned long sp_flags;
50 } ____cacheline_aligned_in_smp;
55 * An RPC service is a ``daemon,'' possibly multithreaded, which
56 * receives and processes incoming RPC messages.
57 * It has one or more transport sockets associated with it, and maintains
58 * a list of idle threads waiting for input.
60 * We currently do not support more than one RPC program per daemon.
63 struct svc_program * sv_program; /* RPC program */
64 struct svc_stat * sv_stats; /* RPC statistics */
66 struct kref sv_refcnt;
67 unsigned int sv_nrthreads; /* # of server threads */
68 unsigned int sv_maxconn; /* max connections allowed or
69 * '0' causing max to be based
70 * on number of threads. */
72 unsigned int sv_max_payload; /* datagram payload size */
73 unsigned int sv_max_mesg; /* max_payload + 1 page for overheads */
74 unsigned int sv_xdrsize; /* XDR buffer size */
75 struct list_head sv_permsocks; /* all permanent sockets */
76 struct list_head sv_tempsocks; /* all temporary sockets */
77 int sv_tmpcnt; /* count of temporary sockets */
78 struct timer_list sv_temptimer; /* timer for aging temporary sockets */
80 char * sv_name; /* service name */
82 unsigned int sv_nrpools; /* number of thread pools */
83 struct svc_pool * sv_pools; /* array of thread pools */
84 int (*sv_threadfn)(void *data);
86 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
87 struct list_head sv_cb_list; /* queue for callback requests
88 * that arrive over the same
90 spinlock_t sv_cb_lock; /* protects the svc_cb_list */
91 wait_queue_head_t sv_cb_waitq; /* sleep here if there are no
92 * entries in the svc_cb_list */
93 bool sv_bc_enabled; /* service uses backchannel */
94 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
98 * svc_get() - increment reference count on a SUNRPC serv
99 * @serv: the svc_serv to have count incremented
101 * Returns: the svc_serv that was passed in.
103 static inline struct svc_serv *svc_get(struct svc_serv *serv)
105 kref_get(&serv->sv_refcnt);
109 void svc_destroy(struct kref *);
112 * svc_put - decrement reference count on a SUNRPC serv
113 * @serv: the svc_serv to have count decremented
115 * When the reference count reaches zero, svc_destroy()
116 * is called to clean up and free the serv.
118 static inline void svc_put(struct svc_serv *serv)
120 kref_put(&serv->sv_refcnt, svc_destroy);
124 * svc_put_not_last - decrement non-final reference count on SUNRPC serv
125 * @serv: the svc_serv to have count decremented
127 * Returns: %true is refcount was decremented.
129 * If the refcount is 1, it is not decremented and instead failure is reported.
131 static inline bool svc_put_not_last(struct svc_serv *serv)
133 return refcount_dec_not_one(&serv->sv_refcnt.refcount);
137 * Maximum payload size supported by a kernel RPC server.
138 * This is use to determine the max number of pages nfsd is
139 * willing to return in a single READ operation.
141 * These happen to all be powers of 2, which is not strictly
142 * necessary but helps enforce the real limitation, which is
143 * that they should be multiples of PAGE_SIZE.
145 * For UDP transports, a block plus NFS,RPC, and UDP headers
146 * has to fit into the IP datagram limit of 64K. The largest
147 * feasible number for all known page sizes is probably 48K,
148 * but we choose 32K here. This is the same as the historical
149 * Linux limit; someone who cares more about NFS/UDP performance
150 * can test a larger number.
152 * For TCP transports we have more freedom. A size of 1MB is
153 * chosen to match the client limit. Other OSes are known to
154 * have larger limits, but those numbers are probably beyond
155 * the point of diminishing returns.
157 #define RPCSVC_MAXPAYLOAD (1*1024*1024u)
158 #define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD
159 #define RPCSVC_MAXPAYLOAD_UDP (32*1024u)
161 extern u32 svc_max_payload(const struct svc_rqst *rqstp);
164 * RPC Requsts and replies are stored in one or more pages.
165 * We maintain an array of pages for each server thread.
166 * Requests are copied into these pages as they arrive. Remaining
167 * pages are available to write the reply into.
169 * Pages are sent using ->sendpage so each server thread needs to
170 * allocate more to replace those used in sending. To help keep track
171 * of these pages we have a receive list where all pages initialy live,
172 * and a send list where pages are moved to when there are to be part
175 * We use xdr_buf for holding responses as it fits well with NFS
176 * read responses (that have a header, and some data pages, and possibly
177 * a tail) and means we can share some client side routines.
179 * The xdr_buf.head kvec always points to the first page in the rq_*pages
180 * list. The xdr_buf.pages pointer points to the second page on that
181 * list. xdr_buf.tail points to the end of the first page.
182 * This assumes that the non-page part of an rpc reply will fit
183 * in a page - NFSd ensures this. lockd also has no trouble.
185 * Each request/reply pair can have at most one "payload", plus two pages,
186 * one for the request, and one for the reply.
187 * We using ->sendfile to return read data, we might need one extra page
188 * if the request is not page-aligned. So add another '1'.
190 #define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
194 * The context of a single thread, including the request currently being
198 struct list_head rq_all; /* all threads list */
199 struct rcu_head rq_rcu_head; /* for RCU deferred kfree */
200 struct svc_xprt * rq_xprt; /* transport ptr */
202 struct sockaddr_storage rq_addr; /* peer address */
204 struct sockaddr_storage rq_daddr; /* dest addr of request
205 * - reply from here */
208 struct svc_serv * rq_server; /* RPC service definition */
209 struct svc_pool * rq_pool; /* thread pool */
210 const struct svc_procedure *rq_procinfo;/* procedure info */
211 struct auth_ops * rq_authop; /* authentication flavour */
212 struct svc_cred rq_cred; /* auth info */
213 void * rq_xprt_ctxt; /* transport specific context ptr */
214 struct svc_deferred_req*rq_deferred; /* deferred request we are replaying */
216 struct xdr_buf rq_arg;
217 struct xdr_stream rq_arg_stream;
218 struct xdr_stream rq_res_stream;
219 struct page *rq_scratch_page;
220 struct xdr_buf rq_res;
221 struct page *rq_pages[RPCSVC_MAXPAGES + 1];
222 struct page * *rq_respages; /* points into rq_pages */
223 struct page * *rq_next_page; /* next reply page to use */
224 struct page * *rq_page_end; /* one past the last page */
226 struct pagevec rq_pvec;
227 struct kvec rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
228 struct bio_vec rq_bvec[RPCSVC_MAXPAGES];
230 __be32 rq_xid; /* transmission id */
231 u32 rq_prog; /* program number */
232 u32 rq_vers; /* program version */
233 u32 rq_proc; /* procedure number */
234 u32 rq_prot; /* IP protocol */
235 int rq_cachetype; /* catering to nfsd */
236 #define RQ_SECURE (0) /* secure port */
237 #define RQ_LOCAL (1) /* local request */
238 #define RQ_USEDEFERRAL (2) /* use deferral */
239 #define RQ_DROPME (3) /* drop current reply */
240 #define RQ_SPLICE_OK (4) /* turned off in gss privacy
241 * to prevent encrypting page
243 #define RQ_VICTIM (5) /* about to be shut down */
244 #define RQ_BUSY (6) /* request is busy */
245 #define RQ_DATA (7) /* request has data */
246 unsigned long rq_flags; /* flags field */
247 ktime_t rq_qtime; /* enqueue time */
249 void * rq_argp; /* decoded arguments */
250 void * rq_resp; /* xdr'd results */
251 __be32 *rq_accept_statp;
252 void * rq_auth_data; /* flavor-specific data */
253 __be32 rq_auth_stat; /* authentication status */
254 int rq_auth_slack; /* extra space xdr code
255 * should leave in head
258 int rq_reserved; /* space on socket outq
259 * reserved for this request
261 ktime_t rq_stime; /* start time */
263 struct cache_req rq_chandle; /* handle passed to caches for
266 /* Catering to nfsd */
267 struct auth_domain * rq_client; /* RPC peer info */
268 struct auth_domain * rq_gssclient; /* "gss/"-style peer info */
269 struct svc_cacherep * rq_cacherep; /* cache info */
270 struct task_struct *rq_task; /* service thread */
271 struct net *rq_bc_net; /* pointer to backchannel's
274 void ** rq_lease_breaker; /* The v4 client breaking a lease */
277 #define SVC_NET(rqst) (rqst->rq_xprt ? rqst->rq_xprt->xpt_net : rqst->rq_bc_net)
280 * Rigorous type checking on sockaddr type conversions
282 static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst)
284 return (struct sockaddr_in *) &rqst->rq_addr;
287 static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst)
289 return (struct sockaddr_in6 *) &rqst->rq_addr;
292 static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst)
294 return (struct sockaddr *) &rqst->rq_addr;
297 static inline struct sockaddr_in *svc_daddr_in(const struct svc_rqst *rqst)
299 return (struct sockaddr_in *) &rqst->rq_daddr;
302 static inline struct sockaddr_in6 *svc_daddr_in6(const struct svc_rqst *rqst)
304 return (struct sockaddr_in6 *) &rqst->rq_daddr;
307 static inline struct sockaddr *svc_daddr(const struct svc_rqst *rqst)
309 return (struct sockaddr *) &rqst->rq_daddr;
312 struct svc_deferred_req {
313 u32 prot; /* protocol (UDP or TCP) */
314 struct svc_xprt *xprt;
315 struct sockaddr_storage addr; /* where reply must go */
317 struct sockaddr_storage daddr; /* where reply must come from */
320 struct cache_deferred_req handle;
325 struct svc_process_info {
327 int (*dispatch)(struct svc_rqst *rqstp);
336 * List of RPC programs on the same transport endpoint
339 struct svc_program * pg_next; /* other programs (same xprt) */
340 u32 pg_prog; /* program number */
341 unsigned int pg_lovers; /* lowest version */
342 unsigned int pg_hivers; /* highest version */
343 unsigned int pg_nvers; /* number of versions */
344 const struct svc_version **pg_vers; /* version array */
345 char * pg_name; /* service name */
346 char * pg_class; /* class name: services sharing authentication */
347 struct svc_stat * pg_stats; /* rpc statistics */
348 int (*pg_authenticate)(struct svc_rqst *);
349 __be32 (*pg_init_request)(struct svc_rqst *,
350 const struct svc_program *,
351 struct svc_process_info *);
352 int (*pg_rpcbind_set)(struct net *net,
353 const struct svc_program *,
354 u32 version, int family,
355 unsigned short proto,
356 unsigned short port);
360 * RPC program version
363 u32 vs_vers; /* version number */
364 u32 vs_nproc; /* number of procedures */
365 const struct svc_procedure *vs_proc; /* per-procedure info */
366 unsigned long __percpu *vs_count; /* call counts */
367 u32 vs_xdrsize; /* xdrsize needed for this version */
369 /* Don't register with rpcbind */
372 /* Don't care if the rpcbind registration fails */
375 /* Need xprt with congestion control */
376 bool vs_need_cong_ctrl;
378 /* Dispatch function */
379 int (*vs_dispatch)(struct svc_rqst *rqstp);
385 struct svc_procedure {
386 /* process the request: */
387 __be32 (*pc_func)(struct svc_rqst *);
388 /* XDR decode args: */
389 bool (*pc_decode)(struct svc_rqst *rqstp,
390 struct xdr_stream *xdr);
391 /* XDR encode result: */
392 bool (*pc_encode)(struct svc_rqst *rqstp,
393 struct xdr_stream *xdr);
394 /* XDR free result: */
395 void (*pc_release)(struct svc_rqst *);
396 unsigned int pc_argsize; /* argument struct size */
397 unsigned int pc_argzero; /* how much of argument to clear */
398 unsigned int pc_ressize; /* result struct size */
399 unsigned int pc_cachetype; /* cache info (NFS) */
400 unsigned int pc_xdrressize; /* maximum size of XDR reply */
401 const char * pc_name; /* for display */
405 * Function prototypes.
407 int svc_rpcb_setup(struct svc_serv *serv, struct net *net);
408 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net);
409 int svc_bind(struct svc_serv *serv, struct net *net);
410 struct svc_serv *svc_create(struct svc_program *, unsigned int,
411 int (*threadfn)(void *data));
412 struct svc_rqst *svc_rqst_alloc(struct svc_serv *serv,
413 struct svc_pool *pool, int node);
414 bool svc_rqst_replace_page(struct svc_rqst *rqstp,
416 void svc_rqst_release_pages(struct svc_rqst *rqstp);
417 void svc_rqst_free(struct svc_rqst *);
418 void svc_exit_thread(struct svc_rqst *);
419 struct svc_serv * svc_create_pooled(struct svc_program *, unsigned int,
420 int (*threadfn)(void *data));
421 int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
422 int svc_pool_stats_open(struct svc_serv *serv, struct file *file);
423 void svc_process(struct svc_rqst *rqstp);
424 int bc_svc_process(struct svc_serv *, struct rpc_rqst *,
426 int svc_register(const struct svc_serv *, struct net *, const int,
427 const unsigned short, const unsigned short);
429 void svc_wake_up(struct svc_serv *);
430 void svc_reserve(struct svc_rqst *rqstp, int space);
431 struct svc_pool *svc_pool_for_cpu(struct svc_serv *serv);
432 char * svc_print_addr(struct svc_rqst *, char *, size_t);
433 const char * svc_proc_name(const struct svc_rqst *rqstp);
434 int svc_encode_result_payload(struct svc_rqst *rqstp,
436 unsigned int length);
437 unsigned int svc_fill_write_vector(struct svc_rqst *rqstp,
438 struct xdr_buf *payload);
439 char *svc_fill_symlink_pathname(struct svc_rqst *rqstp,
440 struct kvec *first, void *p,
442 __be32 svc_generic_init_request(struct svc_rqst *rqstp,
443 const struct svc_program *progp,
444 struct svc_process_info *procinfo);
445 int svc_generic_rpcbind_set(struct net *net,
446 const struct svc_program *progp,
447 u32 version, int family,
448 unsigned short proto,
449 unsigned short port);
450 int svc_rpcbind_set_version(struct net *net,
451 const struct svc_program *progp,
452 u32 version, int family,
453 unsigned short proto,
454 unsigned short port);
456 #define RPC_MAX_ADDRBUFLEN (63U)
459 * When we want to reduce the size of the reserved space in the response
460 * buffer, we need to take into account the size of any checksum data that
461 * may be at the end of the packet. This is difficult to determine exactly
462 * for all cases without actually generating the checksum, so we just use a
465 static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space)
467 svc_reserve(rqstp, space + rqstp->rq_auth_slack);
471 * svcxdr_init_decode - Prepare an xdr_stream for Call decoding
472 * @rqstp: controlling server RPC transaction context
475 static inline void svcxdr_init_decode(struct svc_rqst *rqstp)
477 struct xdr_stream *xdr = &rqstp->rq_arg_stream;
478 struct xdr_buf *buf = &rqstp->rq_arg;
479 struct kvec *argv = buf->head;
481 WARN_ON(buf->len != buf->head->iov_len + buf->page_len + buf->tail->iov_len);
482 buf->len = buf->head->iov_len + buf->page_len + buf->tail->iov_len;
484 xdr_init_decode(xdr, buf, argv->iov_base, NULL);
485 xdr_set_scratch_page(xdr, rqstp->rq_scratch_page);
489 * svcxdr_init_encode - Prepare an xdr_stream for svc Reply encoding
490 * @rqstp: controlling server RPC transaction context
493 static inline void svcxdr_init_encode(struct svc_rqst *rqstp)
495 struct xdr_stream *xdr = &rqstp->rq_res_stream;
496 struct xdr_buf *buf = &rqstp->rq_res;
497 struct kvec *resv = buf->head;
499 xdr_reset_scratch_buffer(xdr);
503 xdr->p = resv->iov_base + resv->iov_len;
504 xdr->end = resv->iov_base + PAGE_SIZE;
505 buf->len = resv->iov_len;
506 xdr->page_ptr = buf->pages - 1;
507 buf->buflen = PAGE_SIZE * (rqstp->rq_page_end - buf->pages);
512 * svcxdr_set_auth_slack -
513 * @rqstp: RPC transaction
514 * @slack: buffer space to reserve for the transaction's security flavor
516 * Set the request's slack space requirement, and set aside that much
517 * space in the rqstp's rq_res.head for use when the auth wraps the Reply.
519 static inline void svcxdr_set_auth_slack(struct svc_rqst *rqstp, int slack)
521 struct xdr_stream *xdr = &rqstp->rq_res_stream;
522 struct xdr_buf *buf = &rqstp->rq_res;
523 struct kvec *resv = buf->head;
525 rqstp->rq_auth_slack = slack;
527 xdr->end -= XDR_QUADLEN(slack);
528 buf->buflen -= rqstp->rq_auth_slack;
530 WARN_ON(xdr->iov != resv);
531 WARN_ON(xdr->p > xdr->end);
535 * svcxdr_set_accept_stat - Reserve space for the accept_stat field
536 * @rqstp: RPC transaction context
540 * %false: No response buffer space was available
542 static inline bool svcxdr_set_accept_stat(struct svc_rqst *rqstp)
544 struct xdr_stream *xdr = &rqstp->rq_res_stream;
546 rqstp->rq_accept_statp = xdr_reserve_space(xdr, XDR_UNIT);
547 if (unlikely(!rqstp->rq_accept_statp))
549 *rqstp->rq_accept_statp = rpc_success;
553 #endif /* SUNRPC_SVC_H */