1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3 * linux/include/linux/sunrpc/svc.h
4 *
5 * RPC server declarations.
6 *
7 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
8 */
9
10
11 #ifndef SUNRPC_SVC_H
12 #define SUNRPC_SVC_H
13
14 #include <linux/in.h>
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>
21 #include <linux/mm.h>
22 #include <linux/pagevec.h>
23
24 /*
25 *
26 * RPC service thread pool.
27 *
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.
33 */
34 struct svc_pool {
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 */
40
41 /* statistics on pool operation */
42 struct percpu_counter sp_messages_arrived;
43 struct percpu_counter sp_sockets_queued;
44 struct percpu_counter sp_threads_woken;
45
46 unsigned long sp_flags;
47 } ____cacheline_aligned_in_smp;
48
49 /* bits for sp_flags */
50 enum {
51 SP_TASK_PENDING, /* still work to do even if no xprt is queued */
52 SP_CONGESTED, /* all threads are busy, none idle */
53 };
54
55
56 /*
57 * RPC service.
58 *
59 * An RPC service is a ``daemon,'' possibly multithreaded, which
60 * receives and processes incoming RPC messages.
61 * It has one or more transport sockets associated with it, and maintains
62 * a list of idle threads waiting for input.
63 *
64 * We currently do not support more than one RPC program per daemon.
65 */
66 struct svc_serv {
67 struct svc_program * sv_program; /* RPC program */
68 struct svc_stat * sv_stats; /* RPC statistics */
69 spinlock_t sv_lock;
70 struct kref sv_refcnt;
71 unsigned int sv_nrthreads; /* # of server threads */
72 unsigned int sv_maxconn; /* max connections allowed or
73 * '0' causing max to be based
74 * on number of threads. */
75
76 unsigned int sv_max_payload; /* datagram payload size */
77 unsigned int sv_max_mesg; /* max_payload + 1 page for overheads */
78 unsigned int sv_xdrsize; /* XDR buffer size */
79 struct list_head sv_permsocks; /* all permanent sockets */
80 struct list_head sv_tempsocks; /* all temporary sockets */
81 int sv_tmpcnt; /* count of temporary sockets */
82 struct timer_list sv_temptimer; /* timer for aging temporary sockets */
83
84 char * sv_name; /* service name */
85
86 unsigned int sv_nrpools; /* number of thread pools */
87 struct svc_pool * sv_pools; /* array of thread pools */
88 int (*sv_threadfn)(void *data);
89
90 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
91 struct list_head sv_cb_list; /* queue for callback requests
92 * that arrive over the same
93 * connection */
94 spinlock_t sv_cb_lock; /* protects the svc_cb_list */
95 wait_queue_head_t sv_cb_waitq; /* sleep here if there are no
96 * entries in the svc_cb_list */
97 bool sv_bc_enabled; /* service uses backchannel */
98 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
99 };
100
101 /**
102 * svc_get() - increment reference count on a SUNRPC serv
103 * @serv: the svc_serv to have count incremented
104 *
105 * Returns: the svc_serv that was passed in.
106 */
svc_get(struct svc_serv * serv)107 static inline struct svc_serv *svc_get(struct svc_serv *serv)
108 {
109 kref_get(&serv->sv_refcnt);
110 return serv;
111 }
112
113 void svc_destroy(struct kref *);
114
115 /**
116 * svc_put - decrement reference count on a SUNRPC serv
117 * @serv: the svc_serv to have count decremented
118 *
119 * When the reference count reaches zero, svc_destroy()
120 * is called to clean up and free the serv.
121 */
svc_put(struct svc_serv * serv)122 static inline void svc_put(struct svc_serv *serv)
123 {
124 kref_put(&serv->sv_refcnt, svc_destroy);
125 }
126
127 /*
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.
131 *
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.
135 *
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.
142 *
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.
147 */
148 #define RPCSVC_MAXPAYLOAD (1*1024*1024u)
149 #define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD
150 #define RPCSVC_MAXPAYLOAD_UDP (32*1024u)
151
152 extern u32 svc_max_payload(const struct svc_rqst *rqstp);
153
154 /*
155 * RPC Requests 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.
159 *
160 * Pages are sent using ->sendmsg with MSG_SPLICE_PAGES so each server thread
161 * needs to 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, and a
163 * send list where pages are moved to when there are to be part of a reply.
164 *
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.
168 *
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.
174 *
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'.
179 */
180 #define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
181 + 2 + 1)
182
183 /*
184 * The context of a single thread, including the request currently being
185 * processed.
186 */
187 struct svc_rqst {
188 struct list_head rq_all; /* all threads list */
189 struct rcu_head rq_rcu_head; /* for RCU deferred kfree */
190 struct svc_xprt * rq_xprt; /* transport ptr */
191
192 struct sockaddr_storage rq_addr; /* peer address */
193 size_t rq_addrlen;
194 struct sockaddr_storage rq_daddr; /* dest addr of request
195 * - reply from here */
196 size_t rq_daddrlen;
197
198 struct svc_serv * rq_server; /* RPC service definition */
199 struct svc_pool * rq_pool; /* thread pool */
200 const struct svc_procedure *rq_procinfo;/* procedure info */
201 struct auth_ops * rq_authop; /* authentication flavour */
202 struct svc_cred rq_cred; /* auth info */
203 void * rq_xprt_ctxt; /* transport specific context ptr */
204 struct svc_deferred_req*rq_deferred; /* deferred request we are replaying */
205
206 struct xdr_buf rq_arg;
207 struct xdr_stream rq_arg_stream;
208 struct xdr_stream rq_res_stream;
209 struct page *rq_scratch_page;
210 struct xdr_buf rq_res;
211 struct page *rq_pages[RPCSVC_MAXPAGES + 1];
212 struct page * *rq_respages; /* points into rq_pages */
213 struct page * *rq_next_page; /* next reply page to use */
214 struct page * *rq_page_end; /* one past the last page */
215
216 struct folio_batch rq_fbatch;
217 struct kvec rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
218 struct bio_vec rq_bvec[RPCSVC_MAXPAGES];
219
220 __be32 rq_xid; /* transmission id */
221 u32 rq_prog; /* program number */
222 u32 rq_vers; /* program version */
223 u32 rq_proc; /* procedure number */
224 u32 rq_prot; /* IP protocol */
225 int rq_cachetype; /* catering to nfsd */
226 unsigned long rq_flags; /* flags field */
227 ktime_t rq_qtime; /* enqueue time */
228
229 void * rq_argp; /* decoded arguments */
230 void * rq_resp; /* xdr'd results */
231 __be32 *rq_accept_statp;
232 void * rq_auth_data; /* flavor-specific data */
233 __be32 rq_auth_stat; /* authentication status */
234 int rq_auth_slack; /* extra space xdr code
235 * should leave in head
236 * for krb5i, krb5p.
237 */
238 int rq_reserved; /* space on socket outq
239 * reserved for this request
240 */
241 ktime_t rq_stime; /* start time */
242
243 struct cache_req rq_chandle; /* handle passed to caches for
244 * request delaying
245 */
246 /* Catering to nfsd */
247 struct auth_domain * rq_client; /* RPC peer info */
248 struct auth_domain * rq_gssclient; /* "gss/"-style peer info */
249 struct task_struct *rq_task; /* service thread */
250 struct net *rq_bc_net; /* pointer to backchannel's
251 * net namespace
252 */
253 void ** rq_lease_breaker; /* The v4 client breaking a lease */
254 };
255
256 /* bits for rq_flags */
257 enum {
258 RQ_SECURE, /* secure port */
259 RQ_LOCAL, /* local request */
260 RQ_USEDEFERRAL, /* use deferral */
261 RQ_DROPME, /* drop current reply */
262 RQ_SPLICE_OK, /* turned off in gss privacy to prevent
263 * encrypting page cache pages */
264 RQ_VICTIM, /* about to be shut down */
265 RQ_BUSY, /* request is busy */
266 RQ_DATA, /* request has data */
267 };
268
269 #define SVC_NET(rqst) (rqst->rq_xprt ? rqst->rq_xprt->xpt_net : rqst->rq_bc_net)
270
271 /*
272 * Rigorous type checking on sockaddr type conversions
273 */
svc_addr_in(const struct svc_rqst * rqst)274 static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst)
275 {
276 return (struct sockaddr_in *) &rqst->rq_addr;
277 }
278
svc_addr_in6(const struct svc_rqst * rqst)279 static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst)
280 {
281 return (struct sockaddr_in6 *) &rqst->rq_addr;
282 }
283
svc_addr(const struct svc_rqst * rqst)284 static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst)
285 {
286 return (struct sockaddr *) &rqst->rq_addr;
287 }
288
svc_daddr_in(const struct svc_rqst * rqst)289 static inline struct sockaddr_in *svc_daddr_in(const struct svc_rqst *rqst)
290 {
291 return (struct sockaddr_in *) &rqst->rq_daddr;
292 }
293
svc_daddr_in6(const struct svc_rqst * rqst)294 static inline struct sockaddr_in6 *svc_daddr_in6(const struct svc_rqst *rqst)
295 {
296 return (struct sockaddr_in6 *) &rqst->rq_daddr;
297 }
298
svc_daddr(const struct svc_rqst * rqst)299 static inline struct sockaddr *svc_daddr(const struct svc_rqst *rqst)
300 {
301 return (struct sockaddr *) &rqst->rq_daddr;
302 }
303
304 struct svc_deferred_req {
305 u32 prot; /* protocol (UDP or TCP) */
306 struct svc_xprt *xprt;
307 struct sockaddr_storage addr; /* where reply must go */
308 size_t addrlen;
309 struct sockaddr_storage daddr; /* where reply must come from */
310 size_t daddrlen;
311 void *xprt_ctxt;
312 struct cache_deferred_req handle;
313 int argslen;
314 __be32 args[];
315 };
316
317 struct svc_process_info {
318 union {
319 int (*dispatch)(struct svc_rqst *rqstp);
320 struct {
321 unsigned int lovers;
322 unsigned int hivers;
323 } mismatch;
324 };
325 };
326
327 /*
328 * List of RPC programs on the same transport endpoint
329 */
330 struct svc_program {
331 struct svc_program * pg_next; /* other programs (same xprt) */
332 u32 pg_prog; /* program number */
333 unsigned int pg_lovers; /* lowest version */
334 unsigned int pg_hivers; /* highest version */
335 unsigned int pg_nvers; /* number of versions */
336 const struct svc_version **pg_vers; /* version array */
337 char * pg_name; /* service name */
338 char * pg_class; /* class name: services sharing authentication */
339 struct svc_stat * pg_stats; /* rpc statistics */
340 enum svc_auth_status (*pg_authenticate)(struct svc_rqst *rqstp);
341 __be32 (*pg_init_request)(struct svc_rqst *,
342 const struct svc_program *,
343 struct svc_process_info *);
344 int (*pg_rpcbind_set)(struct net *net,
345 const struct svc_program *,
346 u32 version, int family,
347 unsigned short proto,
348 unsigned short port);
349 };
350
351 /*
352 * RPC program version
353 */
354 struct svc_version {
355 u32 vs_vers; /* version number */
356 u32 vs_nproc; /* number of procedures */
357 const struct svc_procedure *vs_proc; /* per-procedure info */
358 unsigned long __percpu *vs_count; /* call counts */
359 u32 vs_xdrsize; /* xdrsize needed for this version */
360
361 /* Don't register with rpcbind */
362 bool vs_hidden;
363
364 /* Don't care if the rpcbind registration fails */
365 bool vs_rpcb_optnl;
366
367 /* Need xprt with congestion control */
368 bool vs_need_cong_ctrl;
369
370 /* Dispatch function */
371 int (*vs_dispatch)(struct svc_rqst *rqstp);
372 };
373
374 /*
375 * RPC procedure info
376 */
377 struct svc_procedure {
378 /* process the request: */
379 __be32 (*pc_func)(struct svc_rqst *);
380 /* XDR decode args: */
381 bool (*pc_decode)(struct svc_rqst *rqstp,
382 struct xdr_stream *xdr);
383 /* XDR encode result: */
384 bool (*pc_encode)(struct svc_rqst *rqstp,
385 struct xdr_stream *xdr);
386 /* XDR free result: */
387 void (*pc_release)(struct svc_rqst *);
388 unsigned int pc_argsize; /* argument struct size */
389 unsigned int pc_argzero; /* how much of argument to clear */
390 unsigned int pc_ressize; /* result struct size */
391 unsigned int pc_cachetype; /* cache info (NFS) */
392 unsigned int pc_xdrressize; /* maximum size of XDR reply */
393 const char * pc_name; /* for display */
394 };
395
396 /*
397 * Function prototypes.
398 */
399 int svc_rpcb_setup(struct svc_serv *serv, struct net *net);
400 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net);
401 int svc_bind(struct svc_serv *serv, struct net *net);
402 struct svc_serv *svc_create(struct svc_program *, unsigned int,
403 int (*threadfn)(void *data));
404 struct svc_rqst *svc_rqst_alloc(struct svc_serv *serv,
405 struct svc_pool *pool, int node);
406 bool svc_rqst_replace_page(struct svc_rqst *rqstp,
407 struct page *page);
408 void svc_rqst_release_pages(struct svc_rqst *rqstp);
409 void svc_rqst_free(struct svc_rqst *);
410 void svc_exit_thread(struct svc_rqst *);
411 struct svc_serv * svc_create_pooled(struct svc_program *, unsigned int,
412 int (*threadfn)(void *data));
413 int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
414 int svc_pool_stats_open(struct svc_serv *serv, struct file *file);
415 void svc_process(struct svc_rqst *rqstp);
416 int bc_svc_process(struct svc_serv *, struct rpc_rqst *,
417 struct svc_rqst *);
418 int svc_register(const struct svc_serv *, struct net *, const int,
419 const unsigned short, const unsigned short);
420
421 void svc_wake_up(struct svc_serv *);
422 void svc_reserve(struct svc_rqst *rqstp, int space);
423 void svc_pool_wake_idle_thread(struct svc_pool *pool);
424 struct svc_pool *svc_pool_for_cpu(struct svc_serv *serv);
425 char * svc_print_addr(struct svc_rqst *, char *, size_t);
426 const char * svc_proc_name(const struct svc_rqst *rqstp);
427 int svc_encode_result_payload(struct svc_rqst *rqstp,
428 unsigned int offset,
429 unsigned int length);
430 unsigned int svc_fill_write_vector(struct svc_rqst *rqstp,
431 struct xdr_buf *payload);
432 char *svc_fill_symlink_pathname(struct svc_rqst *rqstp,
433 struct kvec *first, void *p,
434 size_t total);
435 __be32 svc_generic_init_request(struct svc_rqst *rqstp,
436 const struct svc_program *progp,
437 struct svc_process_info *procinfo);
438 int svc_generic_rpcbind_set(struct net *net,
439 const struct svc_program *progp,
440 u32 version, int family,
441 unsigned short proto,
442 unsigned short port);
443 int svc_rpcbind_set_version(struct net *net,
444 const struct svc_program *progp,
445 u32 version, int family,
446 unsigned short proto,
447 unsigned short port);
448
449 #define RPC_MAX_ADDRBUFLEN (63U)
450
451 /*
452 * When we want to reduce the size of the reserved space in the response
453 * buffer, we need to take into account the size of any checksum data that
454 * may be at the end of the packet. This is difficult to determine exactly
455 * for all cases without actually generating the checksum, so we just use a
456 * static value.
457 */
svc_reserve_auth(struct svc_rqst * rqstp,int space)458 static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space)
459 {
460 svc_reserve(rqstp, space + rqstp->rq_auth_slack);
461 }
462
463 /**
464 * svcxdr_init_decode - Prepare an xdr_stream for Call decoding
465 * @rqstp: controlling server RPC transaction context
466 *
467 */
svcxdr_init_decode(struct svc_rqst * rqstp)468 static inline void svcxdr_init_decode(struct svc_rqst *rqstp)
469 {
470 struct xdr_stream *xdr = &rqstp->rq_arg_stream;
471 struct xdr_buf *buf = &rqstp->rq_arg;
472 struct kvec *argv = buf->head;
473
474 WARN_ON(buf->len != buf->head->iov_len + buf->page_len + buf->tail->iov_len);
475 buf->len = buf->head->iov_len + buf->page_len + buf->tail->iov_len;
476
477 xdr_init_decode(xdr, buf, argv->iov_base, NULL);
478 xdr_set_scratch_page(xdr, rqstp->rq_scratch_page);
479 }
480
481 /**
482 * svcxdr_init_encode - Prepare an xdr_stream for svc Reply encoding
483 * @rqstp: controlling server RPC transaction context
484 *
485 */
svcxdr_init_encode(struct svc_rqst * rqstp)486 static inline void svcxdr_init_encode(struct svc_rqst *rqstp)
487 {
488 struct xdr_stream *xdr = &rqstp->rq_res_stream;
489 struct xdr_buf *buf = &rqstp->rq_res;
490 struct kvec *resv = buf->head;
491
492 xdr_reset_scratch_buffer(xdr);
493
494 xdr->buf = buf;
495 xdr->iov = resv;
496 xdr->p = resv->iov_base + resv->iov_len;
497 xdr->end = resv->iov_base + PAGE_SIZE;
498 buf->len = resv->iov_len;
499 xdr->page_ptr = buf->pages - 1;
500 buf->buflen = PAGE_SIZE * (rqstp->rq_page_end - buf->pages);
501 xdr->rqst = NULL;
502 }
503
504 /**
505 * svcxdr_encode_opaque_pages - Insert pages into an xdr_stream
506 * @xdr: xdr_stream to be updated
507 * @pages: array of pages to insert
508 * @base: starting offset of first data byte in @pages
509 * @len: number of data bytes in @pages to insert
510 *
511 * After the @pages are added, the tail iovec is instantiated pointing
512 * to end of the head buffer, and the stream is set up to encode
513 * subsequent items into the tail.
514 */
svcxdr_encode_opaque_pages(struct svc_rqst * rqstp,struct xdr_stream * xdr,struct page ** pages,unsigned int base,unsigned int len)515 static inline void svcxdr_encode_opaque_pages(struct svc_rqst *rqstp,
516 struct xdr_stream *xdr,
517 struct page **pages,
518 unsigned int base,
519 unsigned int len)
520 {
521 xdr_write_pages(xdr, pages, base, len);
522 xdr->page_ptr = rqstp->rq_next_page - 1;
523 }
524
525 /**
526 * svcxdr_set_auth_slack -
527 * @rqstp: RPC transaction
528 * @slack: buffer space to reserve for the transaction's security flavor
529 *
530 * Set the request's slack space requirement, and set aside that much
531 * space in the rqstp's rq_res.head for use when the auth wraps the Reply.
532 */
svcxdr_set_auth_slack(struct svc_rqst * rqstp,int slack)533 static inline void svcxdr_set_auth_slack(struct svc_rqst *rqstp, int slack)
534 {
535 struct xdr_stream *xdr = &rqstp->rq_res_stream;
536 struct xdr_buf *buf = &rqstp->rq_res;
537 struct kvec *resv = buf->head;
538
539 rqstp->rq_auth_slack = slack;
540
541 xdr->end -= XDR_QUADLEN(slack);
542 buf->buflen -= rqstp->rq_auth_slack;
543
544 WARN_ON(xdr->iov != resv);
545 WARN_ON(xdr->p > xdr->end);
546 }
547
548 /**
549 * svcxdr_set_accept_stat - Reserve space for the accept_stat field
550 * @rqstp: RPC transaction context
551 *
552 * Return values:
553 * %true: Success
554 * %false: No response buffer space was available
555 */
svcxdr_set_accept_stat(struct svc_rqst * rqstp)556 static inline bool svcxdr_set_accept_stat(struct svc_rqst *rqstp)
557 {
558 struct xdr_stream *xdr = &rqstp->rq_res_stream;
559
560 rqstp->rq_accept_statp = xdr_reserve_space(xdr, XDR_UNIT);
561 if (unlikely(!rqstp->rq_accept_statp))
562 return false;
563 *rqstp->rq_accept_statp = rpc_success;
564 return true;
565 }
566
567 #endif /* SUNRPC_SVC_H */
568