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/lwq.h>
21 #include <linux/wait.h>
22 #include <linux/mm.h>
23 #include <linux/pagevec.h>
24 #include <linux/kthread.h>
25
26 /*
27 *
28 * RPC service thread pool.
29 *
30 * Pool of threads and temporary sockets. Generally there is only
31 * a single one of these per RPC service, but on NUMA machines those
32 * services that can benefit from it (i.e. nfs but not lockd) will
33 * have one pool per NUMA node. This optimisation reduces cross-
34 * node traffic on multi-node NUMA NFS servers.
35 */
36 struct svc_pool {
37 unsigned int sp_id; /* pool id; also node id on NUMA */
38 struct lwq sp_xprts; /* pending transports */
39 unsigned int sp_nrthreads; /* # of threads in pool */
40 struct list_head sp_all_threads; /* all server threads */
41 struct llist_head sp_idle_threads; /* idle server threads */
42
43 /* statistics on pool operation */
44 struct percpu_counter sp_messages_arrived;
45 struct percpu_counter sp_sockets_queued;
46 struct percpu_counter sp_threads_woken;
47
48 unsigned long sp_flags;
49 } ____cacheline_aligned_in_smp;
50
51 /* bits for sp_flags */
52 enum {
53 SP_TASK_PENDING, /* still work to do even if no xprt is queued */
54 SP_NEED_VICTIM, /* One thread needs to agree to exit */
55 SP_VICTIM_REMAINS, /* One thread needs to actually exit */
56 };
57
58
59 /*
60 * RPC service.
61 *
62 * An RPC service is a ``daemon,'' possibly multithreaded, which
63 * receives and processes incoming RPC messages.
64 * It has one or more transport sockets associated with it, and maintains
65 * a list of idle threads waiting for input.
66 *
67 * We currently do not support more than one RPC program per daemon.
68 */
69 struct svc_serv {
70 struct svc_program * sv_programs; /* RPC programs */
71 struct svc_stat * sv_stats; /* RPC statistics */
72 spinlock_t sv_lock;
73 unsigned int sv_nprogs; /* Number of sv_programs */
74 unsigned int sv_nrthreads; /* # of server threads */
75 unsigned int sv_max_payload; /* datagram payload size */
76 unsigned int sv_max_mesg; /* max_payload + 1 page for overheads */
77 unsigned int sv_xdrsize; /* XDR buffer size */
78 struct list_head sv_permsocks; /* all permanent sockets */
79 struct list_head sv_tempsocks; /* all temporary sockets */
80 int sv_tmpcnt; /* count of temporary "valid" sockets */
81 struct timer_list sv_temptimer; /* timer for aging temporary sockets */
82
83 char * sv_name; /* service name */
84
85 unsigned int sv_nrpools; /* number of thread pools */
86 bool sv_is_pooled; /* is this a pooled service? */
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 lwq sv_cb_list; /* queue for callback requests
92 * that arrive over the same
93 * connection */
94 bool sv_bc_enabled; /* service uses backchannel */
95 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
96 };
97
98 /* This is used by pool_stats to find and lock an svc */
99 struct svc_info {
100 struct svc_serv *serv;
101 struct mutex *mutex;
102 };
103
104 void svc_destroy(struct svc_serv **svcp);
105
106 /*
107 * Maximum payload size supported by a kernel RPC server.
108 * This is use to determine the max number of pages nfsd is
109 * willing to return in a single READ operation.
110 *
111 * These happen to all be powers of 2, which is not strictly
112 * necessary but helps enforce the real limitation, which is
113 * that they should be multiples of PAGE_SIZE.
114 *
115 * For UDP transports, a block plus NFS,RPC, and UDP headers
116 * has to fit into the IP datagram limit of 64K. The largest
117 * feasible number for all known page sizes is probably 48K,
118 * but we choose 32K here. This is the same as the historical
119 * Linux limit; someone who cares more about NFS/UDP performance
120 * can test a larger number.
121 *
122 * For non-UDP transports we have more freedom. A size of 4MB is
123 * chosen to accommodate clients that support larger I/O sizes.
124 */
125 enum {
126 RPCSVC_MAXPAYLOAD = 4 * 1024 * 1024,
127 RPCSVC_MAXPAYLOAD_TCP = RPCSVC_MAXPAYLOAD,
128 RPCSVC_MAXPAYLOAD_UDP = 32 * 1024,
129 };
130
131 extern u32 svc_max_payload(const struct svc_rqst *rqstp);
132
133 /*
134 * RPC Requests and replies are stored in one or more pages.
135 * We maintain an array of pages for each server thread.
136 * Requests are copied into these pages as they arrive. Remaining
137 * pages are available to write the reply into.
138 *
139 * Pages are sent using ->sendmsg with MSG_SPLICE_PAGES so each server thread
140 * needs to allocate more to replace those used in sending. To help keep track
141 * of these pages we have a receive list where all pages initialy live, and a
142 * send list where pages are moved to when there are to be part of a reply.
143 *
144 * We use xdr_buf for holding responses as it fits well with NFS
145 * read responses (that have a header, and some data pages, and possibly
146 * a tail) and means we can share some client side routines.
147 *
148 * The xdr_buf.head kvec always points to the first page in the rq_*pages
149 * list. The xdr_buf.pages pointer points to the second page on that
150 * list. xdr_buf.tail points to the end of the first page.
151 * This assumes that the non-page part of an rpc reply will fit
152 * in a page - NFSd ensures this. lockd also has no trouble.
153 */
154
155 /**
156 * svc_serv_maxpages - maximum count of pages needed for one RPC message
157 * @serv: RPC service context
158 *
159 * Returns a count of pages or vectors that can hold the maximum
160 * size RPC message for @serv.
161 *
162 * Each request/reply pair can have at most one "payload", plus two
163 * pages, one for the request, and one for the reply.
164 * nfsd_splice_actor() might need an extra page when a READ payload
165 * is not page-aligned.
166 */
svc_serv_maxpages(const struct svc_serv * serv)167 static inline unsigned long svc_serv_maxpages(const struct svc_serv *serv)
168 {
169 return DIV_ROUND_UP(serv->sv_max_mesg, PAGE_SIZE) + 2 + 1;
170 }
171
172 /*
173 * The context of a single thread, including the request currently being
174 * processed.
175 */
176 struct svc_rqst {
177 struct list_head rq_all; /* all threads list */
178 struct llist_node rq_idle; /* On the idle list */
179 struct rcu_head rq_rcu_head; /* for RCU deferred kfree */
180 struct svc_xprt * rq_xprt; /* transport ptr */
181
182 struct sockaddr_storage rq_addr; /* peer address */
183 size_t rq_addrlen;
184 struct sockaddr_storage rq_daddr; /* dest addr of request
185 * - reply from here */
186 size_t rq_daddrlen;
187
188 struct svc_serv * rq_server; /* RPC service definition */
189 struct svc_pool * rq_pool; /* thread pool */
190 const struct svc_procedure *rq_procinfo;/* procedure info */
191 struct auth_ops * rq_authop; /* authentication flavour */
192 struct svc_cred rq_cred; /* auth info */
193 void * rq_xprt_ctxt; /* transport specific context ptr */
194 struct svc_deferred_req*rq_deferred; /* deferred request we are replaying */
195
196 struct xdr_buf rq_arg;
197 struct xdr_stream rq_arg_stream;
198 struct xdr_stream rq_res_stream;
199 struct page *rq_scratch_page;
200 struct xdr_buf rq_res;
201 unsigned long rq_maxpages; /* num of entries in rq_pages */
202 struct page * *rq_pages;
203 struct page * *rq_respages; /* points into rq_pages */
204 struct page * *rq_next_page; /* next reply page to use */
205 struct page * *rq_page_end; /* one past the last page */
206
207 struct folio_batch rq_fbatch;
208 struct bio_vec *rq_bvec;
209
210 __be32 rq_xid; /* transmission id */
211 u32 rq_prog; /* program number */
212 u32 rq_vers; /* program version */
213 u32 rq_proc; /* procedure number */
214 u32 rq_prot; /* IP protocol */
215 int rq_cachetype; /* catering to nfsd */
216 unsigned long rq_flags; /* flags field */
217 ktime_t rq_qtime; /* enqueue time */
218
219 void * rq_argp; /* decoded arguments */
220 void * rq_resp; /* xdr'd results */
221 __be32 *rq_accept_statp;
222 void * rq_auth_data; /* flavor-specific data */
223 __be32 rq_auth_stat; /* authentication status */
224 int rq_auth_slack; /* extra space xdr code
225 * should leave in head
226 * for krb5i, krb5p.
227 */
228 int rq_reserved; /* space on socket outq
229 * reserved for this request
230 */
231 ktime_t rq_stime; /* start time */
232
233 struct cache_req rq_chandle; /* handle passed to caches for
234 * request delaying
235 */
236 /* Catering to nfsd */
237 struct auth_domain * rq_client; /* RPC peer info */
238 struct auth_domain * rq_gssclient; /* "gss/"-style peer info */
239 struct task_struct *rq_task; /* service thread */
240 struct net *rq_bc_net; /* pointer to backchannel's
241 * net namespace
242 */
243
244 int rq_err; /* Thread sets this to inidicate
245 * initialisation success.
246 */
247
248 unsigned long bc_to_initval;
249 unsigned int bc_to_retries;
250 unsigned int rq_status_counter; /* RPC processing counter */
251 void **rq_lease_breaker; /* The v4 client breaking a lease */
252 };
253
254 /* bits for rq_flags */
255 enum {
256 RQ_SECURE, /* secure port */
257 RQ_LOCAL, /* local request */
258 RQ_USEDEFERRAL, /* use deferral */
259 RQ_DROPME, /* drop current reply */
260 RQ_VICTIM, /* Have agreed to shut down */
261 RQ_DATA, /* request has data */
262 };
263
264 #define SVC_NET(rqst) (rqst->rq_xprt ? rqst->rq_xprt->xpt_net : rqst->rq_bc_net)
265
266 /*
267 * Rigorous type checking on sockaddr type conversions
268 */
svc_addr_in(const struct svc_rqst * rqst)269 static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst)
270 {
271 return (struct sockaddr_in *) &rqst->rq_addr;
272 }
273
svc_addr_in6(const struct svc_rqst * rqst)274 static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst)
275 {
276 return (struct sockaddr_in6 *) &rqst->rq_addr;
277 }
278
svc_addr(const struct svc_rqst * rqst)279 static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst)
280 {
281 return (struct sockaddr *) &rqst->rq_addr;
282 }
283
svc_daddr_in(const struct svc_rqst * rqst)284 static inline struct sockaddr_in *svc_daddr_in(const struct svc_rqst *rqst)
285 {
286 return (struct sockaddr_in *) &rqst->rq_daddr;
287 }
288
svc_daddr_in6(const struct svc_rqst * rqst)289 static inline struct sockaddr_in6 *svc_daddr_in6(const struct svc_rqst *rqst)
290 {
291 return (struct sockaddr_in6 *) &rqst->rq_daddr;
292 }
293
svc_daddr(const struct svc_rqst * rqst)294 static inline struct sockaddr *svc_daddr(const struct svc_rqst *rqst)
295 {
296 return (struct sockaddr *) &rqst->rq_daddr;
297 }
298
299 /**
300 * svc_thread_should_stop - check if this thread should stop
301 * @rqstp: the thread that might need to stop
302 *
303 * To stop an svc thread, the pool flags SP_NEED_VICTIM and SP_VICTIM_REMAINS
304 * are set. The first thread which sees SP_NEED_VICTIM clears it, becoming
305 * the victim using this function. It should then promptly call
306 * svc_exit_thread() to complete the process, clearing SP_VICTIM_REMAINS
307 * so the task waiting for a thread to exit can wake and continue.
308 *
309 * Return values:
310 * %true: caller should invoke svc_exit_thread()
311 * %false: caller should do nothing
312 */
svc_thread_should_stop(struct svc_rqst * rqstp)313 static inline bool svc_thread_should_stop(struct svc_rqst *rqstp)
314 {
315 if (test_and_clear_bit(SP_NEED_VICTIM, &rqstp->rq_pool->sp_flags))
316 set_bit(RQ_VICTIM, &rqstp->rq_flags);
317
318 return test_bit(RQ_VICTIM, &rqstp->rq_flags);
319 }
320
321 /**
322 * svc_thread_init_status - report whether thread has initialised successfully
323 * @rqstp: the thread in question
324 * @err: errno code
325 *
326 * After performing any initialisation that could fail, and before starting
327 * normal work, each sunrpc svc_thread must call svc_thread_init_status()
328 * with an appropriate error, or zero.
329 *
330 * If zero is passed, the thread is ready and must continue until
331 * svc_thread_should_stop() returns true. If a non-zero error is passed
332 * the call will not return - the thread will exit.
333 */
svc_thread_init_status(struct svc_rqst * rqstp,int err)334 static inline void svc_thread_init_status(struct svc_rqst *rqstp, int err)
335 {
336 store_release_wake_up(&rqstp->rq_err, err);
337 if (err)
338 kthread_exit(1);
339 }
340
341 struct svc_deferred_req {
342 u32 prot; /* protocol (UDP or TCP) */
343 struct svc_xprt *xprt;
344 struct sockaddr_storage addr; /* where reply must go */
345 size_t addrlen;
346 struct sockaddr_storage daddr; /* where reply must come from */
347 size_t daddrlen;
348 void *xprt_ctxt;
349 struct cache_deferred_req handle;
350 int argslen;
351 __be32 args[];
352 };
353
354 struct svc_process_info {
355 union {
356 int (*dispatch)(struct svc_rqst *rqstp);
357 struct {
358 unsigned int lovers;
359 unsigned int hivers;
360 } mismatch;
361 };
362 };
363
364 /*
365 * RPC program - an array of these can use the same transport endpoint
366 */
367 struct svc_program {
368 u32 pg_prog; /* program number */
369 unsigned int pg_lovers; /* lowest version */
370 unsigned int pg_hivers; /* highest version */
371 unsigned int pg_nvers; /* number of versions */
372 const struct svc_version **pg_vers; /* version array */
373 char * pg_name; /* service name */
374 char * pg_class; /* class name: services sharing authentication */
375 enum svc_auth_status (*pg_authenticate)(struct svc_rqst *rqstp);
376 __be32 (*pg_init_request)(struct svc_rqst *,
377 const struct svc_program *,
378 struct svc_process_info *);
379 int (*pg_rpcbind_set)(struct net *net,
380 const struct svc_program *,
381 u32 version, int family,
382 unsigned short proto,
383 unsigned short port);
384 };
385
386 /*
387 * RPC program version
388 */
389 struct svc_version {
390 u32 vs_vers; /* version number */
391 u32 vs_nproc; /* number of procedures */
392 const struct svc_procedure *vs_proc; /* per-procedure info */
393 unsigned long __percpu *vs_count; /* call counts */
394 u32 vs_xdrsize; /* xdrsize needed for this version */
395
396 /* Don't register with rpcbind */
397 bool vs_hidden;
398
399 /* Don't care if the rpcbind registration fails */
400 bool vs_rpcb_optnl;
401
402 /* Need xprt with congestion control */
403 bool vs_need_cong_ctrl;
404
405 /* Dispatch function */
406 int (*vs_dispatch)(struct svc_rqst *rqstp);
407 };
408
409 /*
410 * RPC procedure info
411 */
412 struct svc_procedure {
413 /* process the request: */
414 __be32 (*pc_func)(struct svc_rqst *);
415 /* XDR decode args: */
416 bool (*pc_decode)(struct svc_rqst *rqstp,
417 struct xdr_stream *xdr);
418 /* XDR encode result: */
419 bool (*pc_encode)(struct svc_rqst *rqstp,
420 struct xdr_stream *xdr);
421 /* XDR free result: */
422 void (*pc_release)(struct svc_rqst *);
423 unsigned int pc_argsize; /* argument struct size */
424 unsigned int pc_argzero; /* how much of argument to clear */
425 unsigned int pc_ressize; /* result struct size */
426 unsigned int pc_cachetype; /* cache info (NFS) */
427 unsigned int pc_xdrressize; /* maximum size of XDR reply */
428 const char * pc_name; /* for display */
429 };
430
431 /*
432 * Function prototypes.
433 */
434 int sunrpc_set_pool_mode(const char *val);
435 int sunrpc_get_pool_mode(char *val, size_t size);
436 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net);
437 int svc_bind(struct svc_serv *serv, struct net *net);
438 struct svc_serv *svc_create(struct svc_program *, unsigned int,
439 int (*threadfn)(void *data));
440 bool svc_rqst_replace_page(struct svc_rqst *rqstp,
441 struct page *page);
442 void svc_rqst_release_pages(struct svc_rqst *rqstp);
443 void svc_exit_thread(struct svc_rqst *);
444 struct svc_serv * svc_create_pooled(struct svc_program *prog,
445 unsigned int nprog,
446 struct svc_stat *stats,
447 unsigned int bufsize,
448 int (*threadfn)(void *data));
449 int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
450 int svc_pool_stats_open(struct svc_info *si, struct file *file);
451 void svc_process(struct svc_rqst *rqstp);
452 void svc_process_bc(struct rpc_rqst *req, struct svc_rqst *rqstp);
453 int svc_register(const struct svc_serv *, struct net *, const int,
454 const unsigned short, const unsigned short);
455
456 void svc_wake_up(struct svc_serv *);
457 void svc_reserve(struct svc_rqst *rqstp, int space);
458 void svc_pool_wake_idle_thread(struct svc_pool *pool);
459 struct svc_pool *svc_pool_for_cpu(struct svc_serv *serv);
460 char * svc_print_addr(struct svc_rqst *, char *, size_t);
461 const char * svc_proc_name(const struct svc_rqst *rqstp);
462 int svc_encode_result_payload(struct svc_rqst *rqstp,
463 unsigned int offset,
464 unsigned int length);
465 char *svc_fill_symlink_pathname(struct svc_rqst *rqstp,
466 struct kvec *first, void *p,
467 size_t total);
468 __be32 svc_generic_init_request(struct svc_rqst *rqstp,
469 const struct svc_program *progp,
470 struct svc_process_info *procinfo);
471 int svc_generic_rpcbind_set(struct net *net,
472 const struct svc_program *progp,
473 u32 version, int family,
474 unsigned short proto,
475 unsigned short port);
476
477 #define RPC_MAX_ADDRBUFLEN (63U)
478
479 /*
480 * When we want to reduce the size of the reserved space in the response
481 * buffer, we need to take into account the size of any checksum data that
482 * may be at the end of the packet. This is difficult to determine exactly
483 * for all cases without actually generating the checksum, so we just use a
484 * static value.
485 */
svc_reserve_auth(struct svc_rqst * rqstp,int space)486 static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space)
487 {
488 svc_reserve(rqstp, space + rqstp->rq_auth_slack);
489 }
490
491 /**
492 * svcxdr_init_decode - Prepare an xdr_stream for Call decoding
493 * @rqstp: controlling server RPC transaction context
494 *
495 */
svcxdr_init_decode(struct svc_rqst * rqstp)496 static inline void svcxdr_init_decode(struct svc_rqst *rqstp)
497 {
498 struct xdr_stream *xdr = &rqstp->rq_arg_stream;
499 struct xdr_buf *buf = &rqstp->rq_arg;
500 struct kvec *argv = buf->head;
501
502 WARN_ON(buf->len != buf->head->iov_len + buf->page_len + buf->tail->iov_len);
503 buf->len = buf->head->iov_len + buf->page_len + buf->tail->iov_len;
504
505 xdr_init_decode(xdr, buf, argv->iov_base, NULL);
506 xdr_set_scratch_page(xdr, rqstp->rq_scratch_page);
507 }
508
509 /**
510 * svcxdr_init_encode - Prepare an xdr_stream for svc Reply encoding
511 * @rqstp: controlling server RPC transaction context
512 *
513 */
svcxdr_init_encode(struct svc_rqst * rqstp)514 static inline void svcxdr_init_encode(struct svc_rqst *rqstp)
515 {
516 struct xdr_stream *xdr = &rqstp->rq_res_stream;
517 struct xdr_buf *buf = &rqstp->rq_res;
518 struct kvec *resv = buf->head;
519
520 xdr_reset_scratch_buffer(xdr);
521
522 xdr->buf = buf;
523 xdr->iov = resv;
524 xdr->p = resv->iov_base + resv->iov_len;
525 xdr->end = resv->iov_base + PAGE_SIZE;
526 buf->len = resv->iov_len;
527 xdr->page_ptr = buf->pages - 1;
528 buf->buflen = PAGE_SIZE * (rqstp->rq_page_end - buf->pages);
529 xdr->rqst = NULL;
530 }
531
532 /**
533 * svcxdr_encode_opaque_pages - Insert pages into an xdr_stream
534 * @xdr: xdr_stream to be updated
535 * @pages: array of pages to insert
536 * @base: starting offset of first data byte in @pages
537 * @len: number of data bytes in @pages to insert
538 *
539 * After the @pages are added, the tail iovec is instantiated pointing
540 * to end of the head buffer, and the stream is set up to encode
541 * subsequent items into the tail.
542 */
svcxdr_encode_opaque_pages(struct svc_rqst * rqstp,struct xdr_stream * xdr,struct page ** pages,unsigned int base,unsigned int len)543 static inline void svcxdr_encode_opaque_pages(struct svc_rqst *rqstp,
544 struct xdr_stream *xdr,
545 struct page **pages,
546 unsigned int base,
547 unsigned int len)
548 {
549 xdr_write_pages(xdr, pages, base, len);
550 xdr->page_ptr = rqstp->rq_next_page - 1;
551 }
552
553 /**
554 * svcxdr_set_auth_slack -
555 * @rqstp: RPC transaction
556 * @slack: buffer space to reserve for the transaction's security flavor
557 *
558 * Set the request's slack space requirement, and set aside that much
559 * space in the rqstp's rq_res.head for use when the auth wraps the Reply.
560 */
svcxdr_set_auth_slack(struct svc_rqst * rqstp,int slack)561 static inline void svcxdr_set_auth_slack(struct svc_rqst *rqstp, int slack)
562 {
563 struct xdr_stream *xdr = &rqstp->rq_res_stream;
564 struct xdr_buf *buf = &rqstp->rq_res;
565 struct kvec *resv = buf->head;
566
567 rqstp->rq_auth_slack = slack;
568
569 xdr->end -= XDR_QUADLEN(slack);
570 buf->buflen -= rqstp->rq_auth_slack;
571
572 WARN_ON(xdr->iov != resv);
573 WARN_ON(xdr->p > xdr->end);
574 }
575
576 /**
577 * svcxdr_set_accept_stat - Reserve space for the accept_stat field
578 * @rqstp: RPC transaction context
579 *
580 * Return values:
581 * %true: Success
582 * %false: No response buffer space was available
583 */
svcxdr_set_accept_stat(struct svc_rqst * rqstp)584 static inline bool svcxdr_set_accept_stat(struct svc_rqst *rqstp)
585 {
586 struct xdr_stream *xdr = &rqstp->rq_res_stream;
587
588 rqstp->rq_accept_statp = xdr_reserve_space(xdr, XDR_UNIT);
589 if (unlikely(!rqstp->rq_accept_statp))
590 return false;
591 *rqstp->rq_accept_statp = rpc_success;
592 return true;
593 }
594
595 #endif /* SUNRPC_SVC_H */
596