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
25 /*
26 *
27 * RPC service thread pool.
28 *
29 * Pool of threads and temporary sockets. Generally there is only
30 * a single one of these per RPC service, but on NUMA machines those
31 * services that can benefit from it (i.e. nfs but not lockd) will
32 * have one pool per NUMA node. This optimisation reduces cross-
33 * node traffic on multi-node NUMA NFS servers.
34 */
35 struct svc_pool {
36 unsigned int sp_id; /* pool id; also node id on NUMA */
37 struct lwq sp_xprts; /* pending transports */
38 atomic_t sp_nrthreads; /* # of threads in pool */
39 struct list_head sp_all_threads; /* all server threads */
40 struct llist_head sp_idle_threads; /* idle server threads */
41
42 /* statistics on pool operation */
43 struct percpu_counter sp_messages_arrived;
44 struct percpu_counter sp_sockets_queued;
45 struct percpu_counter sp_threads_woken;
46
47 unsigned long sp_flags;
48 } ____cacheline_aligned_in_smp;
49
50 /* bits for sp_flags */
51 enum {
52 SP_TASK_PENDING, /* still work to do even if no xprt is queued */
53 SP_NEED_VICTIM, /* One thread needs to agree to exit */
54 SP_VICTIM_REMAINS, /* One thread needs to actually exit */
55 };
56
57
58 /*
59 * RPC service.
60 *
61 * An RPC service is a ``daemon,'' possibly multithreaded, which
62 * receives and processes incoming RPC messages.
63 * It has one or more transport sockets associated with it, and maintains
64 * a list of idle threads waiting for input.
65 *
66 * We currently do not support more than one RPC program per daemon.
67 */
68 struct svc_serv {
69 struct svc_program * sv_program; /* RPC program */
70 struct svc_stat * sv_stats; /* RPC statistics */
71 spinlock_t sv_lock;
72 unsigned int sv_nrthreads; /* # of server threads */
73 unsigned int sv_maxconn; /* max connections allowed or
74 * '0' causing max to be based
75 * on number of threads. */
76
77 unsigned int sv_max_payload; /* datagram payload size */
78 unsigned int sv_max_mesg; /* max_payload + 1 page for overheads */
79 unsigned int sv_xdrsize; /* XDR buffer size */
80 struct list_head sv_permsocks; /* all permanent sockets */
81 struct list_head sv_tempsocks; /* all temporary sockets */
82 int sv_tmpcnt; /* count of temporary sockets */
83 struct timer_list sv_temptimer; /* timer for aging temporary sockets */
84
85 char * sv_name; /* service name */
86
87 unsigned int sv_nrpools; /* number of thread pools */
88 bool sv_is_pooled; /* is this a pooled service? */
89 struct svc_pool * sv_pools; /* array of thread pools */
90 int (*sv_threadfn)(void *data);
91
92 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
93 struct lwq sv_cb_list; /* queue for callback requests
94 * that arrive over the same
95 * connection */
96 bool sv_bc_enabled; /* service uses backchannel */
97 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
98 };
99
100 /* This is used by pool_stats to find and lock an svc */
101 struct svc_info {
102 struct svc_serv *serv;
103 struct mutex *mutex;
104 };
105
106 void svc_destroy(struct svc_serv **svcp);
107
108 /*
109 * Maximum payload size supported by a kernel RPC server.
110 * This is use to determine the max number of pages nfsd is
111 * willing to return in a single READ operation.
112 *
113 * These happen to all be powers of 2, which is not strictly
114 * necessary but helps enforce the real limitation, which is
115 * that they should be multiples of PAGE_SIZE.
116 *
117 * For UDP transports, a block plus NFS,RPC, and UDP headers
118 * has to fit into the IP datagram limit of 64K. The largest
119 * feasible number for all known page sizes is probably 48K,
120 * but we choose 32K here. This is the same as the historical
121 * Linux limit; someone who cares more about NFS/UDP performance
122 * can test a larger number.
123 *
124 * For TCP transports we have more freedom. A size of 1MB is
125 * chosen to match the client limit. Other OSes are known to
126 * have larger limits, but those numbers are probably beyond
127 * the point of diminishing returns.
128 */
129 #define RPCSVC_MAXPAYLOAD (1*1024*1024u)
130 #define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD
131 #define RPCSVC_MAXPAYLOAD_UDP (32*1024u)
132
133 extern u32 svc_max_payload(const struct svc_rqst *rqstp);
134
135 /*
136 * RPC Requests and replies are stored in one or more pages.
137 * We maintain an array of pages for each server thread.
138 * Requests are copied into these pages as they arrive. Remaining
139 * pages are available to write the reply into.
140 *
141 * Pages are sent using ->sendmsg with MSG_SPLICE_PAGES so each server thread
142 * needs to allocate more to replace those used in sending. To help keep track
143 * of these pages we have a receive list where all pages initialy live, and a
144 * send list where pages are moved to when there are to be part of a reply.
145 *
146 * We use xdr_buf for holding responses as it fits well with NFS
147 * read responses (that have a header, and some data pages, and possibly
148 * a tail) and means we can share some client side routines.
149 *
150 * The xdr_buf.head kvec always points to the first page in the rq_*pages
151 * list. The xdr_buf.pages pointer points to the second page on that
152 * list. xdr_buf.tail points to the end of the first page.
153 * This assumes that the non-page part of an rpc reply will fit
154 * in a page - NFSd ensures this. lockd also has no trouble.
155 *
156 * Each request/reply pair can have at most one "payload", plus two pages,
157 * one for the request, and one for the reply.
158 * We using ->sendfile to return read data, we might need one extra page
159 * if the request is not page-aligned. So add another '1'.
160 */
161 #define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
162 + 2 + 1)
163
164 /*
165 * The context of a single thread, including the request currently being
166 * processed.
167 */
168 struct svc_rqst {
169 struct list_head rq_all; /* all threads list */
170 struct llist_node rq_idle; /* On the idle list */
171 struct rcu_head rq_rcu_head; /* for RCU deferred kfree */
172 struct svc_xprt * rq_xprt; /* transport ptr */
173
174 struct sockaddr_storage rq_addr; /* peer address */
175 size_t rq_addrlen;
176 struct sockaddr_storage rq_daddr; /* dest addr of request
177 * - reply from here */
178 size_t rq_daddrlen;
179
180 struct svc_serv * rq_server; /* RPC service definition */
181 struct svc_pool * rq_pool; /* thread pool */
182 const struct svc_procedure *rq_procinfo;/* procedure info */
183 struct auth_ops * rq_authop; /* authentication flavour */
184 struct svc_cred rq_cred; /* auth info */
185 void * rq_xprt_ctxt; /* transport specific context ptr */
186 struct svc_deferred_req*rq_deferred; /* deferred request we are replaying */
187
188 struct xdr_buf rq_arg;
189 struct xdr_stream rq_arg_stream;
190 struct xdr_stream rq_res_stream;
191 struct page *rq_scratch_page;
192 struct xdr_buf rq_res;
193 struct page *rq_pages[RPCSVC_MAXPAGES + 1];
194 struct page * *rq_respages; /* points into rq_pages */
195 struct page * *rq_next_page; /* next reply page to use */
196 struct page * *rq_page_end; /* one past the last page */
197
198 struct folio_batch rq_fbatch;
199 struct kvec rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
200 struct bio_vec rq_bvec[RPCSVC_MAXPAGES];
201
202 __be32 rq_xid; /* transmission id */
203 u32 rq_prog; /* program number */
204 u32 rq_vers; /* program version */
205 u32 rq_proc; /* procedure number */
206 u32 rq_prot; /* IP protocol */
207 int rq_cachetype; /* catering to nfsd */
208 unsigned long rq_flags; /* flags field */
209 ktime_t rq_qtime; /* enqueue time */
210
211 void * rq_argp; /* decoded arguments */
212 void * rq_resp; /* xdr'd results */
213 __be32 *rq_accept_statp;
214 void * rq_auth_data; /* flavor-specific data */
215 __be32 rq_auth_stat; /* authentication status */
216 int rq_auth_slack; /* extra space xdr code
217 * should leave in head
218 * for krb5i, krb5p.
219 */
220 int rq_reserved; /* space on socket outq
221 * reserved for this request
222 */
223 ktime_t rq_stime; /* start time */
224
225 struct cache_req rq_chandle; /* handle passed to caches for
226 * request delaying
227 */
228 /* Catering to nfsd */
229 struct auth_domain * rq_client; /* RPC peer info */
230 struct auth_domain * rq_gssclient; /* "gss/"-style peer info */
231 struct task_struct *rq_task; /* service thread */
232 struct net *rq_bc_net; /* pointer to backchannel's
233 * net namespace
234 */
235 unsigned long bc_to_initval;
236 unsigned int bc_to_retries;
237 void ** rq_lease_breaker; /* The v4 client breaking a lease */
238 unsigned int rq_status_counter; /* RPC processing counter */
239 };
240
241 /* bits for rq_flags */
242 enum {
243 RQ_SECURE, /* secure port */
244 RQ_LOCAL, /* local request */
245 RQ_USEDEFERRAL, /* use deferral */
246 RQ_DROPME, /* drop current reply */
247 RQ_VICTIM, /* Have agreed to shut down */
248 RQ_DATA, /* request has data */
249 };
250
251 #define SVC_NET(rqst) (rqst->rq_xprt ? rqst->rq_xprt->xpt_net : rqst->rq_bc_net)
252
253 /*
254 * Rigorous type checking on sockaddr type conversions
255 */
svc_addr_in(const struct svc_rqst * rqst)256 static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst)
257 {
258 return (struct sockaddr_in *) &rqst->rq_addr;
259 }
260
svc_addr_in6(const struct svc_rqst * rqst)261 static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst)
262 {
263 return (struct sockaddr_in6 *) &rqst->rq_addr;
264 }
265
svc_addr(const struct svc_rqst * rqst)266 static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst)
267 {
268 return (struct sockaddr *) &rqst->rq_addr;
269 }
270
svc_daddr_in(const struct svc_rqst * rqst)271 static inline struct sockaddr_in *svc_daddr_in(const struct svc_rqst *rqst)
272 {
273 return (struct sockaddr_in *) &rqst->rq_daddr;
274 }
275
svc_daddr_in6(const struct svc_rqst * rqst)276 static inline struct sockaddr_in6 *svc_daddr_in6(const struct svc_rqst *rqst)
277 {
278 return (struct sockaddr_in6 *) &rqst->rq_daddr;
279 }
280
svc_daddr(const struct svc_rqst * rqst)281 static inline struct sockaddr *svc_daddr(const struct svc_rqst *rqst)
282 {
283 return (struct sockaddr *) &rqst->rq_daddr;
284 }
285
286 /**
287 * svc_thread_should_stop - check if this thread should stop
288 * @rqstp: the thread that might need to stop
289 *
290 * To stop an svc thread, the pool flags SP_NEED_VICTIM and SP_VICTIM_REMAINS
291 * are set. The first thread which sees SP_NEED_VICTIM clears it, becoming
292 * the victim using this function. It should then promptly call
293 * svc_exit_thread() to complete the process, clearing SP_VICTIM_REMAINS
294 * so the task waiting for a thread to exit can wake and continue.
295 *
296 * Return values:
297 * %true: caller should invoke svc_exit_thread()
298 * %false: caller should do nothing
299 */
svc_thread_should_stop(struct svc_rqst * rqstp)300 static inline bool svc_thread_should_stop(struct svc_rqst *rqstp)
301 {
302 if (test_and_clear_bit(SP_NEED_VICTIM, &rqstp->rq_pool->sp_flags))
303 set_bit(RQ_VICTIM, &rqstp->rq_flags);
304
305 return test_bit(RQ_VICTIM, &rqstp->rq_flags);
306 }
307
308 struct svc_deferred_req {
309 u32 prot; /* protocol (UDP or TCP) */
310 struct svc_xprt *xprt;
311 struct sockaddr_storage addr; /* where reply must go */
312 size_t addrlen;
313 struct sockaddr_storage daddr; /* where reply must come from */
314 size_t daddrlen;
315 void *xprt_ctxt;
316 struct cache_deferred_req handle;
317 int argslen;
318 __be32 args[];
319 };
320
321 struct svc_process_info {
322 union {
323 int (*dispatch)(struct svc_rqst *rqstp);
324 struct {
325 unsigned int lovers;
326 unsigned int hivers;
327 } mismatch;
328 };
329 };
330
331 /*
332 * List of RPC programs on the same transport endpoint
333 */
334 struct svc_program {
335 struct svc_program * pg_next; /* other programs (same xprt) */
336 u32 pg_prog; /* program number */
337 unsigned int pg_lovers; /* lowest version */
338 unsigned int pg_hivers; /* highest version */
339 unsigned int pg_nvers; /* number of versions */
340 const struct svc_version **pg_vers; /* version array */
341 char * pg_name; /* service name */
342 char * pg_class; /* class name: services sharing authentication */
343 enum svc_auth_status (*pg_authenticate)(struct svc_rqst *rqstp);
344 __be32 (*pg_init_request)(struct svc_rqst *,
345 const struct svc_program *,
346 struct svc_process_info *);
347 int (*pg_rpcbind_set)(struct net *net,
348 const struct svc_program *,
349 u32 version, int family,
350 unsigned short proto,
351 unsigned short port);
352 };
353
354 /*
355 * RPC program version
356 */
357 struct svc_version {
358 u32 vs_vers; /* version number */
359 u32 vs_nproc; /* number of procedures */
360 const struct svc_procedure *vs_proc; /* per-procedure info */
361 unsigned long __percpu *vs_count; /* call counts */
362 u32 vs_xdrsize; /* xdrsize needed for this version */
363
364 /* Don't register with rpcbind */
365 bool vs_hidden;
366
367 /* Don't care if the rpcbind registration fails */
368 bool vs_rpcb_optnl;
369
370 /* Need xprt with congestion control */
371 bool vs_need_cong_ctrl;
372
373 /* Dispatch function */
374 int (*vs_dispatch)(struct svc_rqst *rqstp);
375 };
376
377 /*
378 * RPC procedure info
379 */
380 struct svc_procedure {
381 /* process the request: */
382 __be32 (*pc_func)(struct svc_rqst *);
383 /* XDR decode args: */
384 bool (*pc_decode)(struct svc_rqst *rqstp,
385 struct xdr_stream *xdr);
386 /* XDR encode result: */
387 bool (*pc_encode)(struct svc_rqst *rqstp,
388 struct xdr_stream *xdr);
389 /* XDR free result: */
390 void (*pc_release)(struct svc_rqst *);
391 unsigned int pc_argsize; /* argument struct size */
392 unsigned int pc_argzero; /* how much of argument to clear */
393 unsigned int pc_ressize; /* result struct size */
394 unsigned int pc_cachetype; /* cache info (NFS) */
395 unsigned int pc_xdrressize; /* maximum size of XDR reply */
396 const char * pc_name; /* for display */
397 };
398
399 /*
400 * Function prototypes.
401 */
402 int sunrpc_set_pool_mode(const char *val);
403 int sunrpc_get_pool_mode(char *val, size_t size);
404 int svc_rpcb_setup(struct svc_serv *serv, struct net *net);
405 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net);
406 int svc_bind(struct svc_serv *serv, struct net *net);
407 struct svc_serv *svc_create(struct svc_program *, unsigned int,
408 int (*threadfn)(void *data));
409 struct svc_rqst *svc_rqst_alloc(struct svc_serv *serv,
410 struct svc_pool *pool, int node);
411 bool svc_rqst_replace_page(struct svc_rqst *rqstp,
412 struct page *page);
413 void svc_rqst_release_pages(struct svc_rqst *rqstp);
414 void svc_rqst_free(struct svc_rqst *);
415 void svc_exit_thread(struct svc_rqst *);
416 struct svc_serv * svc_create_pooled(struct svc_program *prog,
417 struct svc_stat *stats,
418 unsigned int bufsize,
419 int (*threadfn)(void *data));
420 int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
421 int svc_pool_stats_open(struct svc_info *si, struct file *file);
422 void svc_process(struct svc_rqst *rqstp);
423 void svc_process_bc(struct rpc_rqst *req, struct svc_rqst *rqstp);
424 int svc_register(const struct svc_serv *, struct net *, const int,
425 const unsigned short, const unsigned short);
426
427 void svc_wake_up(struct svc_serv *);
428 void svc_reserve(struct svc_rqst *rqstp, int space);
429 void svc_pool_wake_idle_thread(struct svc_pool *pool);
430 struct svc_pool *svc_pool_for_cpu(struct svc_serv *serv);
431 char * svc_print_addr(struct svc_rqst *, char *, size_t);
432 const char * svc_proc_name(const struct svc_rqst *rqstp);
433 int svc_encode_result_payload(struct svc_rqst *rqstp,
434 unsigned int offset,
435 unsigned int length);
436 unsigned int svc_fill_write_vector(struct svc_rqst *rqstp,
437 struct xdr_buf *payload);
438 char *svc_fill_symlink_pathname(struct svc_rqst *rqstp,
439 struct kvec *first, void *p,
440 size_t total);
441 __be32 svc_generic_init_request(struct svc_rqst *rqstp,
442 const struct svc_program *progp,
443 struct svc_process_info *procinfo);
444 int svc_generic_rpcbind_set(struct net *net,
445 const struct svc_program *progp,
446 u32 version, int family,
447 unsigned short proto,
448 unsigned short port);
449 int svc_rpcbind_set_version(struct net *net,
450 const struct svc_program *progp,
451 u32 version, int family,
452 unsigned short proto,
453 unsigned short port);
454
455 #define RPC_MAX_ADDRBUFLEN (63U)
456
457 /*
458 * When we want to reduce the size of the reserved space in the response
459 * buffer, we need to take into account the size of any checksum data that
460 * may be at the end of the packet. This is difficult to determine exactly
461 * for all cases without actually generating the checksum, so we just use a
462 * static value.
463 */
svc_reserve_auth(struct svc_rqst * rqstp,int space)464 static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space)
465 {
466 svc_reserve(rqstp, space + rqstp->rq_auth_slack);
467 }
468
469 /**
470 * svcxdr_init_decode - Prepare an xdr_stream for Call decoding
471 * @rqstp: controlling server RPC transaction context
472 *
473 */
svcxdr_init_decode(struct svc_rqst * rqstp)474 static inline void svcxdr_init_decode(struct svc_rqst *rqstp)
475 {
476 struct xdr_stream *xdr = &rqstp->rq_arg_stream;
477 struct xdr_buf *buf = &rqstp->rq_arg;
478 struct kvec *argv = buf->head;
479
480 WARN_ON(buf->len != buf->head->iov_len + buf->page_len + buf->tail->iov_len);
481 buf->len = buf->head->iov_len + buf->page_len + buf->tail->iov_len;
482
483 xdr_init_decode(xdr, buf, argv->iov_base, NULL);
484 xdr_set_scratch_page(xdr, rqstp->rq_scratch_page);
485 }
486
487 /**
488 * svcxdr_init_encode - Prepare an xdr_stream for svc Reply encoding
489 * @rqstp: controlling server RPC transaction context
490 *
491 */
svcxdr_init_encode(struct svc_rqst * rqstp)492 static inline void svcxdr_init_encode(struct svc_rqst *rqstp)
493 {
494 struct xdr_stream *xdr = &rqstp->rq_res_stream;
495 struct xdr_buf *buf = &rqstp->rq_res;
496 struct kvec *resv = buf->head;
497
498 xdr_reset_scratch_buffer(xdr);
499
500 xdr->buf = buf;
501 xdr->iov = resv;
502 xdr->p = resv->iov_base + resv->iov_len;
503 xdr->end = resv->iov_base + PAGE_SIZE;
504 buf->len = resv->iov_len;
505 xdr->page_ptr = buf->pages - 1;
506 buf->buflen = PAGE_SIZE * (rqstp->rq_page_end - buf->pages);
507 xdr->rqst = NULL;
508 }
509
510 /**
511 * svcxdr_encode_opaque_pages - Insert pages into an xdr_stream
512 * @xdr: xdr_stream to be updated
513 * @pages: array of pages to insert
514 * @base: starting offset of first data byte in @pages
515 * @len: number of data bytes in @pages to insert
516 *
517 * After the @pages are added, the tail iovec is instantiated pointing
518 * to end of the head buffer, and the stream is set up to encode
519 * subsequent items into the tail.
520 */
svcxdr_encode_opaque_pages(struct svc_rqst * rqstp,struct xdr_stream * xdr,struct page ** pages,unsigned int base,unsigned int len)521 static inline void svcxdr_encode_opaque_pages(struct svc_rqst *rqstp,
522 struct xdr_stream *xdr,
523 struct page **pages,
524 unsigned int base,
525 unsigned int len)
526 {
527 xdr_write_pages(xdr, pages, base, len);
528 xdr->page_ptr = rqstp->rq_next_page - 1;
529 }
530
531 /**
532 * svcxdr_set_auth_slack -
533 * @rqstp: RPC transaction
534 * @slack: buffer space to reserve for the transaction's security flavor
535 *
536 * Set the request's slack space requirement, and set aside that much
537 * space in the rqstp's rq_res.head for use when the auth wraps the Reply.
538 */
svcxdr_set_auth_slack(struct svc_rqst * rqstp,int slack)539 static inline void svcxdr_set_auth_slack(struct svc_rqst *rqstp, int slack)
540 {
541 struct xdr_stream *xdr = &rqstp->rq_res_stream;
542 struct xdr_buf *buf = &rqstp->rq_res;
543 struct kvec *resv = buf->head;
544
545 rqstp->rq_auth_slack = slack;
546
547 xdr->end -= XDR_QUADLEN(slack);
548 buf->buflen -= rqstp->rq_auth_slack;
549
550 WARN_ON(xdr->iov != resv);
551 WARN_ON(xdr->p > xdr->end);
552 }
553
554 /**
555 * svcxdr_set_accept_stat - Reserve space for the accept_stat field
556 * @rqstp: RPC transaction context
557 *
558 * Return values:
559 * %true: Success
560 * %false: No response buffer space was available
561 */
svcxdr_set_accept_stat(struct svc_rqst * rqstp)562 static inline bool svcxdr_set_accept_stat(struct svc_rqst *rqstp)
563 {
564 struct xdr_stream *xdr = &rqstp->rq_res_stream;
565
566 rqstp->rq_accept_statp = xdr_reserve_space(xdr, XDR_UNIT);
567 if (unlikely(!rqstp->rq_accept_statp))
568 return false;
569 *rqstp->rq_accept_statp = rpc_success;
570 return true;
571 }
572
573 #endif /* SUNRPC_SVC_H */
574