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>
22 #include <linux/pagevec.h>
24 /* statistics for svc_pool structures */
25 struct svc_pool_stats
{
26 atomic_long_t packets
;
27 unsigned long sockets_queued
;
28 atomic_long_t threads_woken
;
29 atomic_long_t threads_timedout
;
34 * RPC service thread pool.
36 * Pool of threads and temporary sockets. Generally there is only
37 * a single one of these per RPC service, but on NUMA machines those
38 * services that can benefit from it (i.e. nfs but not lockd) will
39 * have one pool per NUMA node. This optimisation reduces cross-
40 * node traffic on multi-node NUMA NFS servers.
43 unsigned int sp_id
; /* pool id; also node id on NUMA */
44 spinlock_t sp_lock
; /* protects all fields */
45 struct list_head sp_sockets
; /* pending sockets */
46 unsigned int sp_nrthreads
; /* # of threads in pool */
47 struct list_head sp_all_threads
; /* all server threads */
48 struct svc_pool_stats sp_stats
; /* statistics on pool operation */
49 #define SP_TASK_PENDING (0) /* still work to do even if no
51 #define SP_CONGESTED (1)
52 unsigned long sp_flags
;
53 } ____cacheline_aligned_in_smp
;
58 /* Callback to use when last thread exits. */
59 void (*svo_shutdown
)(struct svc_serv
*, struct net
*);
61 /* function for service threads to run */
62 int (*svo_function
)(void *);
64 /* queue up a transport for servicing */
65 void (*svo_enqueue_xprt
)(struct svc_xprt
*);
67 /* set up thread (or whatever) execution context */
68 int (*svo_setup
)(struct svc_serv
*, struct svc_pool
*, int);
70 /* optional module to count when adding threads (pooled svcs only) */
71 struct module
*svo_module
;
77 * An RPC service is a ``daemon,'' possibly multithreaded, which
78 * receives and processes incoming RPC messages.
79 * It has one or more transport sockets associated with it, and maintains
80 * a list of idle threads waiting for input.
82 * We currently do not support more than one RPC program per daemon.
85 struct svc_program
* sv_program
; /* RPC program */
86 struct svc_stat
* sv_stats
; /* RPC statistics */
88 unsigned int sv_nrthreads
; /* # of server threads */
89 unsigned int sv_maxconn
; /* max connections allowed or
90 * '0' causing max to be based
91 * on number of threads. */
93 unsigned int sv_max_payload
; /* datagram payload size */
94 unsigned int sv_max_mesg
; /* max_payload + 1 page for overheads */
95 unsigned int sv_xdrsize
; /* XDR buffer size */
96 struct list_head sv_permsocks
; /* all permanent sockets */
97 struct list_head sv_tempsocks
; /* all temporary sockets */
98 int sv_tmpcnt
; /* count of temporary sockets */
99 struct timer_list sv_temptimer
; /* timer for aging temporary sockets */
101 char * sv_name
; /* service name */
103 unsigned int sv_nrpools
; /* number of thread pools */
104 struct svc_pool
* sv_pools
; /* array of thread pools */
105 const struct svc_serv_ops
*sv_ops
; /* server operations */
106 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
107 struct list_head sv_cb_list
; /* queue for callback requests
108 * that arrive over the same
110 spinlock_t sv_cb_lock
; /* protects the svc_cb_list */
111 wait_queue_head_t sv_cb_waitq
; /* sleep here if there are no
112 * entries in the svc_cb_list */
113 bool sv_bc_enabled
; /* service uses backchannel */
114 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
118 * We use sv_nrthreads as a reference count. svc_destroy() drops
119 * this refcount, so we need to bump it up around operations that
120 * change the number of threads. Horrible, but there it is.
121 * Should be called with the "service mutex" held.
123 static inline void svc_get(struct svc_serv
*serv
)
125 serv
->sv_nrthreads
++;
129 * Maximum payload size supported by a kernel RPC server.
130 * This is use to determine the max number of pages nfsd is
131 * willing to return in a single READ operation.
133 * These happen to all be powers of 2, which is not strictly
134 * necessary but helps enforce the real limitation, which is
135 * that they should be multiples of PAGE_SIZE.
137 * For UDP transports, a block plus NFS,RPC, and UDP headers
138 * has to fit into the IP datagram limit of 64K. The largest
139 * feasible number for all known page sizes is probably 48K,
140 * but we choose 32K here. This is the same as the historical
141 * Linux limit; someone who cares more about NFS/UDP performance
142 * can test a larger number.
144 * For TCP transports we have more freedom. A size of 1MB is
145 * chosen to match the client limit. Other OSes are known to
146 * have larger limits, but those numbers are probably beyond
147 * the point of diminishing returns.
149 #define RPCSVC_MAXPAYLOAD (1*1024*1024u)
150 #define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD
151 #define RPCSVC_MAXPAYLOAD_UDP (32*1024u)
153 extern u32
svc_max_payload(const struct svc_rqst
*rqstp
);
156 * RPC Requsts and replies are stored in one or more pages.
157 * We maintain an array of pages for each server thread.
158 * Requests are copied into these pages as they arrive. Remaining
159 * pages are available to write the reply into.
161 * Pages are sent using ->sendpage so each server thread needs to
162 * allocate more to replace those used in sending. To help keep track
163 * of these pages we have a receive list where all pages initialy live,
164 * and a send list where pages are moved to when there are to be part
167 * We use xdr_buf for holding responses as it fits well with NFS
168 * read responses (that have a header, and some data pages, and possibly
169 * a tail) and means we can share some client side routines.
171 * The xdr_buf.head kvec always points to the first page in the rq_*pages
172 * list. The xdr_buf.pages pointer points to the second page on that
173 * list. xdr_buf.tail points to the end of the first page.
174 * This assumes that the non-page part of an rpc reply will fit
175 * in a page - NFSd ensures this. lockd also has no trouble.
177 * Each request/reply pair can have at most one "payload", plus two pages,
178 * one for the request, and one for the reply.
179 * We using ->sendfile to return read data, we might need one extra page
180 * if the request is not page-aligned. So add another '1'.
182 #define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
185 static inline u32
svc_getnl(struct kvec
*iov
)
190 iov
->iov_base
= (void*)vp
;
191 iov
->iov_len
-= sizeof(__be32
);
195 static inline void svc_putnl(struct kvec
*iov
, u32 val
)
197 __be32
*vp
= iov
->iov_base
+ iov
->iov_len
;
199 iov
->iov_len
+= sizeof(__be32
);
202 static inline __be32
svc_getu32(struct kvec
*iov
)
207 iov
->iov_base
= (void*)vp
;
208 iov
->iov_len
-= sizeof(__be32
);
212 static inline void svc_ungetu32(struct kvec
*iov
)
214 __be32
*vp
= (__be32
*)iov
->iov_base
;
215 iov
->iov_base
= (void *)(vp
- 1);
216 iov
->iov_len
+= sizeof(*vp
);
219 static inline void svc_putu32(struct kvec
*iov
, __be32 val
)
221 __be32
*vp
= iov
->iov_base
+ iov
->iov_len
;
223 iov
->iov_len
+= sizeof(__be32
);
227 * The context of a single thread, including the request currently being
231 struct list_head rq_all
; /* all threads list */
232 struct rcu_head rq_rcu_head
; /* for RCU deferred kfree */
233 struct svc_xprt
* rq_xprt
; /* transport ptr */
235 struct sockaddr_storage rq_addr
; /* peer address */
237 struct sockaddr_storage rq_daddr
; /* dest addr of request
238 * - reply from here */
241 struct svc_serv
* rq_server
; /* RPC service definition */
242 struct svc_pool
* rq_pool
; /* thread pool */
243 const struct svc_procedure
*rq_procinfo
;/* procedure info */
244 struct auth_ops
* rq_authop
; /* authentication flavour */
245 struct svc_cred rq_cred
; /* auth info */
246 void * rq_xprt_ctxt
; /* transport specific context ptr */
247 struct svc_deferred_req
*rq_deferred
; /* deferred request we are replaying */
249 size_t rq_xprt_hlen
; /* xprt header len */
250 struct xdr_buf rq_arg
;
251 struct xdr_stream rq_arg_stream
;
252 struct xdr_stream rq_res_stream
;
253 struct page
*rq_scratch_page
;
254 struct xdr_buf rq_res
;
255 struct page
*rq_pages
[RPCSVC_MAXPAGES
+ 1];
256 struct page
* *rq_respages
; /* points into rq_pages */
257 struct page
* *rq_next_page
; /* next reply page to use */
258 struct page
* *rq_page_end
; /* one past the last page */
260 struct pagevec rq_pvec
;
261 struct kvec rq_vec
[RPCSVC_MAXPAGES
]; /* generally useful.. */
262 struct bio_vec rq_bvec
[RPCSVC_MAXPAGES
];
264 __be32 rq_xid
; /* transmission id */
265 u32 rq_prog
; /* program number */
266 u32 rq_vers
; /* program version */
267 u32 rq_proc
; /* procedure number */
268 u32 rq_prot
; /* IP protocol */
269 int rq_cachetype
; /* catering to nfsd */
270 #define RQ_SECURE (0) /* secure port */
271 #define RQ_LOCAL (1) /* local request */
272 #define RQ_USEDEFERRAL (2) /* use deferral */
273 #define RQ_DROPME (3) /* drop current reply */
274 #define RQ_SPLICE_OK (4) /* turned off in gss privacy
275 * to prevent encrypting page
277 #define RQ_VICTIM (5) /* about to be shut down */
278 #define RQ_BUSY (6) /* request is busy */
279 #define RQ_DATA (7) /* request has data */
280 unsigned long rq_flags
; /* flags field */
281 ktime_t rq_qtime
; /* enqueue time */
283 void * rq_argp
; /* decoded arguments */
284 void * rq_resp
; /* xdr'd results */
285 void * rq_auth_data
; /* flavor-specific data */
286 __be32 rq_auth_stat
; /* authentication status */
287 int rq_auth_slack
; /* extra space xdr code
288 * should leave in head
291 int rq_reserved
; /* space on socket outq
292 * reserved for this request
294 ktime_t rq_stime
; /* start time */
296 struct cache_req rq_chandle
; /* handle passed to caches for
299 /* Catering to nfsd */
300 struct auth_domain
* rq_client
; /* RPC peer info */
301 struct auth_domain
* rq_gssclient
; /* "gss/"-style peer info */
302 struct svc_cacherep
* rq_cacherep
; /* cache info */
303 struct task_struct
*rq_task
; /* service thread */
304 spinlock_t rq_lock
; /* per-request lock */
305 struct net
*rq_bc_net
; /* pointer to backchannel's
308 void ** rq_lease_breaker
; /* The v4 client breaking a lease */
311 #define SVC_NET(rqst) (rqst->rq_xprt ? rqst->rq_xprt->xpt_net : rqst->rq_bc_net)
314 * Rigorous type checking on sockaddr type conversions
316 static inline struct sockaddr_in
*svc_addr_in(const struct svc_rqst
*rqst
)
318 return (struct sockaddr_in
*) &rqst
->rq_addr
;
321 static inline struct sockaddr_in6
*svc_addr_in6(const struct svc_rqst
*rqst
)
323 return (struct sockaddr_in6
*) &rqst
->rq_addr
;
326 static inline struct sockaddr
*svc_addr(const struct svc_rqst
*rqst
)
328 return (struct sockaddr
*) &rqst
->rq_addr
;
331 static inline struct sockaddr_in
*svc_daddr_in(const struct svc_rqst
*rqst
)
333 return (struct sockaddr_in
*) &rqst
->rq_daddr
;
336 static inline struct sockaddr_in6
*svc_daddr_in6(const struct svc_rqst
*rqst
)
338 return (struct sockaddr_in6
*) &rqst
->rq_daddr
;
341 static inline struct sockaddr
*svc_daddr(const struct svc_rqst
*rqst
)
343 return (struct sockaddr
*) &rqst
->rq_daddr
;
347 * Check buffer bounds after decoding arguments
350 xdr_argsize_check(struct svc_rqst
*rqstp
, __be32
*p
)
352 char *cp
= (char *)p
;
353 struct kvec
*vec
= &rqstp
->rq_arg
.head
[0];
354 return cp
>= (char*)vec
->iov_base
355 && cp
<= (char*)vec
->iov_base
+ vec
->iov_len
;
359 xdr_ressize_check(struct svc_rqst
*rqstp
, __be32
*p
)
361 struct kvec
*vec
= &rqstp
->rq_res
.head
[0];
364 vec
->iov_len
= cp
- (char*)vec
->iov_base
;
366 return vec
->iov_len
<= PAGE_SIZE
;
369 static inline void svc_free_res_pages(struct svc_rqst
*rqstp
)
371 while (rqstp
->rq_next_page
!= rqstp
->rq_respages
) {
372 struct page
**pp
= --rqstp
->rq_next_page
;
380 struct svc_deferred_req
{
381 u32 prot
; /* protocol (UDP or TCP) */
382 struct svc_xprt
*xprt
;
383 struct sockaddr_storage addr
; /* where reply must go */
385 struct sockaddr_storage daddr
; /* where reply must come from */
387 struct cache_deferred_req handle
;
393 struct svc_process_info
{
395 int (*dispatch
)(struct svc_rqst
*, __be32
*);
404 * List of RPC programs on the same transport endpoint
407 struct svc_program
* pg_next
; /* other programs (same xprt) */
408 u32 pg_prog
; /* program number */
409 unsigned int pg_lovers
; /* lowest version */
410 unsigned int pg_hivers
; /* highest version */
411 unsigned int pg_nvers
; /* number of versions */
412 const struct svc_version
**pg_vers
; /* version array */
413 char * pg_name
; /* service name */
414 char * pg_class
; /* class name: services sharing authentication */
415 struct svc_stat
* pg_stats
; /* rpc statistics */
416 int (*pg_authenticate
)(struct svc_rqst
*);
417 __be32 (*pg_init_request
)(struct svc_rqst
*,
418 const struct svc_program
*,
419 struct svc_process_info
*);
420 int (*pg_rpcbind_set
)(struct net
*net
,
421 const struct svc_program
*,
422 u32 version
, int family
,
423 unsigned short proto
,
424 unsigned short port
);
428 * RPC program version
431 u32 vs_vers
; /* version number */
432 u32 vs_nproc
; /* number of procedures */
433 const struct svc_procedure
*vs_proc
; /* per-procedure info */
434 unsigned int *vs_count
; /* call counts */
435 u32 vs_xdrsize
; /* xdrsize needed for this version */
437 /* Don't register with rpcbind */
440 /* Don't care if the rpcbind registration fails */
443 /* Need xprt with congestion control */
444 bool vs_need_cong_ctrl
;
446 /* Override dispatch function (e.g. when caching replies).
447 * A return value of 0 means drop the request.
448 * vs_dispatch == NULL means use default dispatcher.
450 int (*vs_dispatch
)(struct svc_rqst
*, __be32
*);
456 struct svc_procedure
{
457 /* process the request: */
458 __be32 (*pc_func
)(struct svc_rqst
*);
459 /* XDR decode args: */
460 int (*pc_decode
)(struct svc_rqst
*, __be32
*data
);
461 /* XDR encode result: */
462 int (*pc_encode
)(struct svc_rqst
*, __be32
*data
);
463 /* XDR free result: */
464 void (*pc_release
)(struct svc_rqst
*);
465 unsigned int pc_argsize
; /* argument struct size */
466 unsigned int pc_ressize
; /* result struct size */
467 unsigned int pc_cachetype
; /* cache info (NFS) */
468 unsigned int pc_xdrressize
; /* maximum size of XDR reply */
469 const char * pc_name
; /* for display */
473 * Mode for mapping cpus to pools.
476 SVC_POOL_AUTO
= -1, /* choose one of the others */
477 SVC_POOL_GLOBAL
, /* no mapping, just a single global pool
478 * (legacy & UP mode) */
479 SVC_POOL_PERCPU
, /* one pool per cpu */
480 SVC_POOL_PERNODE
/* one pool per numa node */
483 struct svc_pool_map
{
484 int count
; /* How many svc_servs use us */
485 int mode
; /* Note: int not enum to avoid
486 * warnings about "enumeration value
487 * not handled in switch" */
489 unsigned int *pool_to
; /* maps pool id to cpu or node */
490 unsigned int *to_pool
; /* maps cpu or node to pool id */
493 extern struct svc_pool_map svc_pool_map
;
496 * Function prototypes.
498 int svc_rpcb_setup(struct svc_serv
*serv
, struct net
*net
);
499 void svc_rpcb_cleanup(struct svc_serv
*serv
, struct net
*net
);
500 int svc_bind(struct svc_serv
*serv
, struct net
*net
);
501 struct svc_serv
*svc_create(struct svc_program
*, unsigned int,
502 const struct svc_serv_ops
*);
503 struct svc_rqst
*svc_rqst_alloc(struct svc_serv
*serv
,
504 struct svc_pool
*pool
, int node
);
505 struct svc_rqst
*svc_prepare_thread(struct svc_serv
*serv
,
506 struct svc_pool
*pool
, int node
);
507 void svc_rqst_replace_page(struct svc_rqst
*rqstp
,
509 void svc_rqst_free(struct svc_rqst
*);
510 void svc_exit_thread(struct svc_rqst
*);
511 unsigned int svc_pool_map_get(void);
512 void svc_pool_map_put(void);
513 struct svc_serv
* svc_create_pooled(struct svc_program
*, unsigned int,
514 const struct svc_serv_ops
*);
515 int svc_set_num_threads(struct svc_serv
*, struct svc_pool
*, int);
516 int svc_set_num_threads_sync(struct svc_serv
*, struct svc_pool
*, int);
517 int svc_pool_stats_open(struct svc_serv
*serv
, struct file
*file
);
518 void svc_destroy(struct svc_serv
*);
519 void svc_shutdown_net(struct svc_serv
*, struct net
*);
520 int svc_process(struct svc_rqst
*);
521 int bc_svc_process(struct svc_serv
*, struct rpc_rqst
*,
523 int svc_register(const struct svc_serv
*, struct net
*, const int,
524 const unsigned short, const unsigned short);
526 void svc_wake_up(struct svc_serv
*);
527 void svc_reserve(struct svc_rqst
*rqstp
, int space
);
528 struct svc_pool
* svc_pool_for_cpu(struct svc_serv
*serv
, int cpu
);
529 char * svc_print_addr(struct svc_rqst
*, char *, size_t);
530 const char * svc_proc_name(const struct svc_rqst
*rqstp
);
531 int svc_encode_result_payload(struct svc_rqst
*rqstp
,
533 unsigned int length
);
534 unsigned int svc_fill_write_vector(struct svc_rqst
*rqstp
,
536 struct kvec
*first
, size_t total
);
537 char *svc_fill_symlink_pathname(struct svc_rqst
*rqstp
,
538 struct kvec
*first
, void *p
,
540 __be32
svc_generic_init_request(struct svc_rqst
*rqstp
,
541 const struct svc_program
*progp
,
542 struct svc_process_info
*procinfo
);
543 int svc_generic_rpcbind_set(struct net
*net
,
544 const struct svc_program
*progp
,
545 u32 version
, int family
,
546 unsigned short proto
,
547 unsigned short port
);
548 int svc_rpcbind_set_version(struct net
*net
,
549 const struct svc_program
*progp
,
550 u32 version
, int family
,
551 unsigned short proto
,
552 unsigned short port
);
554 #define RPC_MAX_ADDRBUFLEN (63U)
557 * When we want to reduce the size of the reserved space in the response
558 * buffer, we need to take into account the size of any checksum data that
559 * may be at the end of the packet. This is difficult to determine exactly
560 * for all cases without actually generating the checksum, so we just use a
563 static inline void svc_reserve_auth(struct svc_rqst
*rqstp
, int space
)
565 svc_reserve(rqstp
, space
+ rqstp
->rq_auth_slack
);
569 * svcxdr_init_decode - Prepare an xdr_stream for svc Call decoding
570 * @rqstp: controlling server RPC transaction context
573 static inline void svcxdr_init_decode(struct svc_rqst
*rqstp
)
575 struct xdr_stream
*xdr
= &rqstp
->rq_arg_stream
;
576 struct kvec
*argv
= rqstp
->rq_arg
.head
;
578 xdr_init_decode(xdr
, &rqstp
->rq_arg
, argv
->iov_base
, NULL
);
579 xdr_set_scratch_page(xdr
, rqstp
->rq_scratch_page
);
583 * svcxdr_init_encode - Prepare an xdr_stream for svc Reply encoding
584 * @rqstp: controlling server RPC transaction context
587 static inline void svcxdr_init_encode(struct svc_rqst
*rqstp
)
589 struct xdr_stream
*xdr
= &rqstp
->rq_res_stream
;
590 struct xdr_buf
*buf
= &rqstp
->rq_res
;
591 struct kvec
*resv
= buf
->head
;
593 xdr_reset_scratch_buffer(xdr
);
597 xdr
->p
= resv
->iov_base
+ resv
->iov_len
;
598 xdr
->end
= resv
->iov_base
+ PAGE_SIZE
- rqstp
->rq_auth_slack
;
599 buf
->len
= resv
->iov_len
;
600 xdr
->page_ptr
= buf
->pages
- 1;
601 buf
->buflen
= PAGE_SIZE
* (1 + rqstp
->rq_page_end
- buf
->pages
);
602 buf
->buflen
-= rqstp
->rq_auth_slack
;
606 #endif /* SUNRPC_SVC_H */