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 */
35struct 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 */
51enum {
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 */
68struct svc_serv {
69 struct svc_program * sv_program; /* RPC program */
70 struct svc_stat * sv_stats; /* RPC statistics */
71 spinlock_t sv_lock;
72 struct kref sv_refcnt;
73 unsigned int sv_nrthreads; /* # of server threads */
74 unsigned int sv_maxconn; /* max connections allowed or
75 * '0' causing max to be based
76 * on number of threads. */
77
78 unsigned int sv_max_payload; /* datagram payload size */
79 unsigned int sv_max_mesg; /* max_payload + 1 page for overheads */
80 unsigned int sv_xdrsize; /* XDR buffer size */
81 struct list_head sv_permsocks; /* all permanent sockets */
82 struct list_head sv_tempsocks; /* all temporary sockets */
83 int sv_tmpcnt; /* count of temporary sockets */
84 struct timer_list sv_temptimer; /* timer for aging temporary sockets */
85
86 char * sv_name; /* service name */
87
88 unsigned int sv_nrpools; /* number of thread pools */
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/**
101 * svc_get() - increment reference count on a SUNRPC serv
102 * @serv: the svc_serv to have count incremented
103 *
104 * Returns: the svc_serv that was passed in.
105 */
106static inline struct svc_serv *svc_get(struct svc_serv *serv)
107{
108 kref_get(kref: &serv->sv_refcnt);
109 return serv;
110}
111
112void svc_destroy(struct kref *);
113
114/**
115 * svc_put - decrement reference count on a SUNRPC serv
116 * @serv: the svc_serv to have count decremented
117 *
118 * When the reference count reaches zero, svc_destroy()
119 * is called to clean up and free the serv.
120 */
121static inline void svc_put(struct svc_serv *serv)
122{
123 kref_put(kref: &serv->sv_refcnt, release: svc_destroy);
124}
125
126/*
127 * Maximum payload size supported by a kernel RPC server.
128 * This is use to determine the max number of pages nfsd is
129 * willing to return in a single READ operation.
130 *
131 * These happen to all be powers of 2, which is not strictly
132 * necessary but helps enforce the real limitation, which is
133 * that they should be multiples of PAGE_SIZE.
134 *
135 * For UDP transports, a block plus NFS,RPC, and UDP headers
136 * has to fit into the IP datagram limit of 64K. The largest
137 * feasible number for all known page sizes is probably 48K,
138 * but we choose 32K here. This is the same as the historical
139 * Linux limit; someone who cares more about NFS/UDP performance
140 * can test a larger number.
141 *
142 * For TCP transports we have more freedom. A size of 1MB is
143 * chosen to match the client limit. Other OSes are known to
144 * have larger limits, but those numbers are probably beyond
145 * the point of diminishing returns.
146 */
147#define RPCSVC_MAXPAYLOAD (1*1024*1024u)
148#define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD
149#define RPCSVC_MAXPAYLOAD_UDP (32*1024u)
150
151extern u32 svc_max_payload(const struct svc_rqst *rqstp);
152
153/*
154 * RPC Requests and replies are stored in one or more pages.
155 * We maintain an array of pages for each server thread.
156 * Requests are copied into these pages as they arrive. Remaining
157 * pages are available to write the reply into.
158 *
159 * Pages are sent using ->sendmsg with MSG_SPLICE_PAGES so each server thread
160 * needs to allocate more to replace those used in sending. To help keep track
161 * of these pages we have a receive list where all pages initialy live, and a
162 * send list where pages are moved to when there are to be part of a reply.
163 *
164 * We use xdr_buf for holding responses as it fits well with NFS
165 * read responses (that have a header, and some data pages, and possibly
166 * a tail) and means we can share some client side routines.
167 *
168 * The xdr_buf.head kvec always points to the first page in the rq_*pages
169 * list. The xdr_buf.pages pointer points to the second page on that
170 * list. xdr_buf.tail points to the end of the first page.
171 * This assumes that the non-page part of an rpc reply will fit
172 * in a page - NFSd ensures this. lockd also has no trouble.
173 *
174 * Each request/reply pair can have at most one "payload", plus two pages,
175 * one for the request, and one for the reply.
176 * We using ->sendfile to return read data, we might need one extra page
177 * if the request is not page-aligned. So add another '1'.
178 */
179#define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
180 + 2 + 1)
181
182/*
183 * The context of a single thread, including the request currently being
184 * processed.
185 */
186struct svc_rqst {
187 struct list_head rq_all; /* all threads list */
188 struct llist_node rq_idle; /* On the idle 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 unsigned int rq_status_counter; /* RPC processing counter */
255};
256
257/* bits for rq_flags */
258enum {
259 RQ_SECURE, /* secure port */
260 RQ_LOCAL, /* local request */
261 RQ_USEDEFERRAL, /* use deferral */
262 RQ_DROPME, /* drop current reply */
263 RQ_SPLICE_OK, /* turned off in gss privacy to prevent
264 * encrypting page cache pages */
265 RQ_VICTIM, /* Have agreed to shut down */
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 */
274static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst)
275{
276 return (struct sockaddr_in *) &rqst->rq_addr;
277}
278
279static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst)
280{
281 return (struct sockaddr_in6 *) &rqst->rq_addr;
282}
283
284static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst)
285{
286 return (struct sockaddr *) &rqst->rq_addr;
287}
288
289static inline struct sockaddr_in *svc_daddr_in(const struct svc_rqst *rqst)
290{
291 return (struct sockaddr_in *) &rqst->rq_daddr;
292}
293
294static inline struct sockaddr_in6 *svc_daddr_in6(const struct svc_rqst *rqst)
295{
296 return (struct sockaddr_in6 *) &rqst->rq_daddr;
297}
298
299static inline struct sockaddr *svc_daddr(const struct svc_rqst *rqst)
300{
301 return (struct sockaddr *) &rqst->rq_daddr;
302}
303
304/**
305 * svc_thread_should_stop - check if this thread should stop
306 * @rqstp: the thread that might need to stop
307 *
308 * To stop an svc thread, the pool flags SP_NEED_VICTIM and SP_VICTIM_REMAINS
309 * are set. The first thread which sees SP_NEED_VICTIM clears it, becoming
310 * the victim using this function. It should then promptly call
311 * svc_exit_thread() to complete the process, clearing SP_VICTIM_REMAINS
312 * so the task waiting for a thread to exit can wake and continue.
313 *
314 * Return values:
315 * %true: caller should invoke svc_exit_thread()
316 * %false: caller should do nothing
317 */
318static inline bool svc_thread_should_stop(struct svc_rqst *rqstp)
319{
320 if (test_and_clear_bit(nr: SP_NEED_VICTIM, addr: &rqstp->rq_pool->sp_flags))
321 set_bit(nr: RQ_VICTIM, addr: &rqstp->rq_flags);
322
323 return test_bit(RQ_VICTIM, &rqstp->rq_flags);
324}
325
326struct svc_deferred_req {
327 u32 prot; /* protocol (UDP or TCP) */
328 struct svc_xprt *xprt;
329 struct sockaddr_storage addr; /* where reply must go */
330 size_t addrlen;
331 struct sockaddr_storage daddr; /* where reply must come from */
332 size_t daddrlen;
333 void *xprt_ctxt;
334 struct cache_deferred_req handle;
335 int argslen;
336 __be32 args[];
337};
338
339struct svc_process_info {
340 union {
341 int (*dispatch)(struct svc_rqst *rqstp);
342 struct {
343 unsigned int lovers;
344 unsigned int hivers;
345 } mismatch;
346 };
347};
348
349/*
350 * List of RPC programs on the same transport endpoint
351 */
352struct svc_program {
353 struct svc_program * pg_next; /* other programs (same xprt) */
354 u32 pg_prog; /* program number */
355 unsigned int pg_lovers; /* lowest version */
356 unsigned int pg_hivers; /* highest version */
357 unsigned int pg_nvers; /* number of versions */
358 const struct svc_version **pg_vers; /* version array */
359 char * pg_name; /* service name */
360 char * pg_class; /* class name: services sharing authentication */
361 struct svc_stat * pg_stats; /* rpc statistics */
362 enum svc_auth_status (*pg_authenticate)(struct svc_rqst *rqstp);
363 __be32 (*pg_init_request)(struct svc_rqst *,
364 const struct svc_program *,
365 struct svc_process_info *);
366 int (*pg_rpcbind_set)(struct net *net,
367 const struct svc_program *,
368 u32 version, int family,
369 unsigned short proto,
370 unsigned short port);
371};
372
373/*
374 * RPC program version
375 */
376struct svc_version {
377 u32 vs_vers; /* version number */
378 u32 vs_nproc; /* number of procedures */
379 const struct svc_procedure *vs_proc; /* per-procedure info */
380 unsigned long __percpu *vs_count; /* call counts */
381 u32 vs_xdrsize; /* xdrsize needed for this version */
382
383 /* Don't register with rpcbind */
384 bool vs_hidden;
385
386 /* Don't care if the rpcbind registration fails */
387 bool vs_rpcb_optnl;
388
389 /* Need xprt with congestion control */
390 bool vs_need_cong_ctrl;
391
392 /* Dispatch function */
393 int (*vs_dispatch)(struct svc_rqst *rqstp);
394};
395
396/*
397 * RPC procedure info
398 */
399struct svc_procedure {
400 /* process the request: */
401 __be32 (*pc_func)(struct svc_rqst *);
402 /* XDR decode args: */
403 bool (*pc_decode)(struct svc_rqst *rqstp,
404 struct xdr_stream *xdr);
405 /* XDR encode result: */
406 bool (*pc_encode)(struct svc_rqst *rqstp,
407 struct xdr_stream *xdr);
408 /* XDR free result: */
409 void (*pc_release)(struct svc_rqst *);
410 unsigned int pc_argsize; /* argument struct size */
411 unsigned int pc_argzero; /* how much of argument to clear */
412 unsigned int pc_ressize; /* result struct size */
413 unsigned int pc_cachetype; /* cache info (NFS) */
414 unsigned int pc_xdrressize; /* maximum size of XDR reply */
415 const char * pc_name; /* for display */
416};
417
418/*
419 * Function prototypes.
420 */
421int svc_rpcb_setup(struct svc_serv *serv, struct net *net);
422void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net);
423int svc_bind(struct svc_serv *serv, struct net *net);
424struct svc_serv *svc_create(struct svc_program *, unsigned int,
425 int (*threadfn)(void *data));
426struct svc_rqst *svc_rqst_alloc(struct svc_serv *serv,
427 struct svc_pool *pool, int node);
428bool svc_rqst_replace_page(struct svc_rqst *rqstp,
429 struct page *page);
430void svc_rqst_release_pages(struct svc_rqst *rqstp);
431void svc_rqst_free(struct svc_rqst *);
432void svc_exit_thread(struct svc_rqst *);
433struct svc_serv * svc_create_pooled(struct svc_program *, unsigned int,
434 int (*threadfn)(void *data));
435int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
436int svc_pool_stats_open(struct svc_serv *serv, struct file *file);
437void svc_process(struct svc_rqst *rqstp);
438void svc_process_bc(struct rpc_rqst *req, struct svc_rqst *rqstp);
439int svc_register(const struct svc_serv *, struct net *, const int,
440 const unsigned short, const unsigned short);
441
442void svc_wake_up(struct svc_serv *);
443void svc_reserve(struct svc_rqst *rqstp, int space);
444void svc_pool_wake_idle_thread(struct svc_pool *pool);
445struct svc_pool *svc_pool_for_cpu(struct svc_serv *serv);
446char * svc_print_addr(struct svc_rqst *, char *, size_t);
447const char * svc_proc_name(const struct svc_rqst *rqstp);
448int svc_encode_result_payload(struct svc_rqst *rqstp,
449 unsigned int offset,
450 unsigned int length);
451unsigned int svc_fill_write_vector(struct svc_rqst *rqstp,
452 struct xdr_buf *payload);
453char *svc_fill_symlink_pathname(struct svc_rqst *rqstp,
454 struct kvec *first, void *p,
455 size_t total);
456__be32 svc_generic_init_request(struct svc_rqst *rqstp,
457 const struct svc_program *progp,
458 struct svc_process_info *procinfo);
459int svc_generic_rpcbind_set(struct net *net,
460 const struct svc_program *progp,
461 u32 version, int family,
462 unsigned short proto,
463 unsigned short port);
464int svc_rpcbind_set_version(struct net *net,
465 const struct svc_program *progp,
466 u32 version, int family,
467 unsigned short proto,
468 unsigned short port);
469
470#define RPC_MAX_ADDRBUFLEN (63U)
471
472/*
473 * When we want to reduce the size of the reserved space in the response
474 * buffer, we need to take into account the size of any checksum data that
475 * may be at the end of the packet. This is difficult to determine exactly
476 * for all cases without actually generating the checksum, so we just use a
477 * static value.
478 */
479static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space)
480{
481 svc_reserve(rqstp, space: space + rqstp->rq_auth_slack);
482}
483
484/**
485 * svcxdr_init_decode - Prepare an xdr_stream for Call decoding
486 * @rqstp: controlling server RPC transaction context
487 *
488 */
489static inline void svcxdr_init_decode(struct svc_rqst *rqstp)
490{
491 struct xdr_stream *xdr = &rqstp->rq_arg_stream;
492 struct xdr_buf *buf = &rqstp->rq_arg;
493 struct kvec *argv = buf->head;
494
495 WARN_ON(buf->len != buf->head->iov_len + buf->page_len + buf->tail->iov_len);
496 buf->len = buf->head->iov_len + buf->page_len + buf->tail->iov_len;
497
498 xdr_init_decode(xdr, buf, p: argv->iov_base, NULL);
499 xdr_set_scratch_page(xdr, page: rqstp->rq_scratch_page);
500}
501
502/**
503 * svcxdr_init_encode - Prepare an xdr_stream for svc Reply encoding
504 * @rqstp: controlling server RPC transaction context
505 *
506 */
507static inline void svcxdr_init_encode(struct svc_rqst *rqstp)
508{
509 struct xdr_stream *xdr = &rqstp->rq_res_stream;
510 struct xdr_buf *buf = &rqstp->rq_res;
511 struct kvec *resv = buf->head;
512
513 xdr_reset_scratch_buffer(xdr);
514
515 xdr->buf = buf;
516 xdr->iov = resv;
517 xdr->p = resv->iov_base + resv->iov_len;
518 xdr->end = resv->iov_base + PAGE_SIZE;
519 buf->len = resv->iov_len;
520 xdr->page_ptr = buf->pages - 1;
521 buf->buflen = PAGE_SIZE * (rqstp->rq_page_end - buf->pages);
522 xdr->rqst = NULL;
523}
524
525/**
526 * svcxdr_encode_opaque_pages - Insert pages into an xdr_stream
527 * @xdr: xdr_stream to be updated
528 * @pages: array of pages to insert
529 * @base: starting offset of first data byte in @pages
530 * @len: number of data bytes in @pages to insert
531 *
532 * After the @pages are added, the tail iovec is instantiated pointing
533 * to end of the head buffer, and the stream is set up to encode
534 * subsequent items into the tail.
535 */
536static inline void svcxdr_encode_opaque_pages(struct svc_rqst *rqstp,
537 struct xdr_stream *xdr,
538 struct page **pages,
539 unsigned int base,
540 unsigned int len)
541{
542 xdr_write_pages(xdr, pages, base, len);
543 xdr->page_ptr = rqstp->rq_next_page - 1;
544}
545
546/**
547 * svcxdr_set_auth_slack -
548 * @rqstp: RPC transaction
549 * @slack: buffer space to reserve for the transaction's security flavor
550 *
551 * Set the request's slack space requirement, and set aside that much
552 * space in the rqstp's rq_res.head for use when the auth wraps the Reply.
553 */
554static inline void svcxdr_set_auth_slack(struct svc_rqst *rqstp, int slack)
555{
556 struct xdr_stream *xdr = &rqstp->rq_res_stream;
557 struct xdr_buf *buf = &rqstp->rq_res;
558 struct kvec *resv = buf->head;
559
560 rqstp->rq_auth_slack = slack;
561
562 xdr->end -= XDR_QUADLEN(slack);
563 buf->buflen -= rqstp->rq_auth_slack;
564
565 WARN_ON(xdr->iov != resv);
566 WARN_ON(xdr->p > xdr->end);
567}
568
569/**
570 * svcxdr_set_accept_stat - Reserve space for the accept_stat field
571 * @rqstp: RPC transaction context
572 *
573 * Return values:
574 * %true: Success
575 * %false: No response buffer space was available
576 */
577static inline bool svcxdr_set_accept_stat(struct svc_rqst *rqstp)
578{
579 struct xdr_stream *xdr = &rqstp->rq_res_stream;
580
581 rqstp->rq_accept_statp = xdr_reserve_space(xdr, XDR_UNIT);
582 if (unlikely(!rqstp->rq_accept_statp))
583 return false;
584 *rqstp->rq_accept_statp = rpc_success;
585 return true;
586}
587
588#endif /* SUNRPC_SVC_H */
589

source code of linux/include/linux/sunrpc/svc.h