xref: /linux-6.15/include/linux/sunrpc/svc.h (revision 8564c315)
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 <[email protected]>
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_sockets_queued;
43 	struct percpu_counter	sp_threads_woken;
44 	struct percpu_counter	sp_threads_timedout;
45 
46 #define	SP_TASK_PENDING		(0)		/* still work to do even if no
47 						 * xprt is queued. */
48 #define SP_CONGESTED		(1)
49 	unsigned long		sp_flags;
50 } ____cacheline_aligned_in_smp;
51 
52 /*
53  * RPC service.
54  *
55  * An RPC service is a ``daemon,'' possibly multithreaded, which
56  * receives and processes incoming RPC messages.
57  * It has one or more transport sockets associated with it, and maintains
58  * a list of idle threads waiting for input.
59  *
60  * We currently do not support more than one RPC program per daemon.
61  */
62 struct svc_serv {
63 	struct svc_program *	sv_program;	/* RPC program */
64 	struct svc_stat *	sv_stats;	/* RPC statistics */
65 	spinlock_t		sv_lock;
66 	struct kref		sv_refcnt;
67 	unsigned int		sv_nrthreads;	/* # of server threads */
68 	unsigned int		sv_maxconn;	/* max connections allowed or
69 						 * '0' causing max to be based
70 						 * on number of threads. */
71 
72 	unsigned int		sv_max_payload;	/* datagram payload size */
73 	unsigned int		sv_max_mesg;	/* max_payload + 1 page for overheads */
74 	unsigned int		sv_xdrsize;	/* XDR buffer size */
75 	struct list_head	sv_permsocks;	/* all permanent sockets */
76 	struct list_head	sv_tempsocks;	/* all temporary sockets */
77 	int			sv_tmpcnt;	/* count of temporary sockets */
78 	struct timer_list	sv_temptimer;	/* timer for aging temporary sockets */
79 
80 	char *			sv_name;	/* service name */
81 
82 	unsigned int		sv_nrpools;	/* number of thread pools */
83 	struct svc_pool *	sv_pools;	/* array of thread pools */
84 	int			(*sv_threadfn)(void *data);
85 
86 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
87 	struct list_head	sv_cb_list;	/* queue for callback requests
88 						 * that arrive over the same
89 						 * connection */
90 	spinlock_t		sv_cb_lock;	/* protects the svc_cb_list */
91 	wait_queue_head_t	sv_cb_waitq;	/* sleep here if there are no
92 						 * entries in the svc_cb_list */
93 	bool			sv_bc_enabled;	/* service uses backchannel */
94 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
95 };
96 
97 /**
98  * svc_get() - increment reference count on a SUNRPC serv
99  * @serv:  the svc_serv to have count incremented
100  *
101  * Returns: the svc_serv that was passed in.
102  */
103 static inline struct svc_serv *svc_get(struct svc_serv *serv)
104 {
105 	kref_get(&serv->sv_refcnt);
106 	return serv;
107 }
108 
109 void svc_destroy(struct kref *);
110 
111 /**
112  * svc_put - decrement reference count on a SUNRPC serv
113  * @serv:  the svc_serv to have count decremented
114  *
115  * When the reference count reaches zero, svc_destroy()
116  * is called to clean up and free the serv.
117  */
118 static inline void svc_put(struct svc_serv *serv)
119 {
120 	kref_put(&serv->sv_refcnt, svc_destroy);
121 }
122 
123 /**
124  * svc_put_not_last - decrement non-final reference count on SUNRPC serv
125  * @serv:  the svc_serv to have count decremented
126  *
127  * Returns: %true is refcount was decremented.
128  *
129  * If the refcount is 1, it is not decremented and instead failure is reported.
130  */
131 static inline bool svc_put_not_last(struct svc_serv *serv)
132 {
133 	return refcount_dec_not_one(&serv->sv_refcnt.refcount);
134 }
135 
136 /*
137  * Maximum payload size supported by a kernel RPC server.
138  * This is use to determine the max number of pages nfsd is
139  * willing to return in a single READ operation.
140  *
141  * These happen to all be powers of 2, which is not strictly
142  * necessary but helps enforce the real limitation, which is
143  * that they should be multiples of PAGE_SIZE.
144  *
145  * For UDP transports, a block plus NFS,RPC, and UDP headers
146  * has to fit into the IP datagram limit of 64K.  The largest
147  * feasible number for all known page sizes is probably 48K,
148  * but we choose 32K here.  This is the same as the historical
149  * Linux limit; someone who cares more about NFS/UDP performance
150  * can test a larger number.
151  *
152  * For TCP transports we have more freedom.  A size of 1MB is
153  * chosen to match the client limit.  Other OSes are known to
154  * have larger limits, but those numbers are probably beyond
155  * the point of diminishing returns.
156  */
157 #define RPCSVC_MAXPAYLOAD	(1*1024*1024u)
158 #define RPCSVC_MAXPAYLOAD_TCP	RPCSVC_MAXPAYLOAD
159 #define RPCSVC_MAXPAYLOAD_UDP	(32*1024u)
160 
161 extern u32 svc_max_payload(const struct svc_rqst *rqstp);
162 
163 /*
164  * RPC Requests and replies are stored in one or more pages.
165  * We maintain an array of pages for each server thread.
166  * Requests are copied into these pages as they arrive.  Remaining
167  * pages are available to write the reply into.
168  *
169  * Pages are sent using ->sendmsg with MSG_SPLICE_PAGES so each server thread
170  * needs to allocate more to replace those used in sending.  To help keep track
171  * of these pages we have a receive list where all pages initialy live, and a
172  * send list where pages are moved to when there are to be part of a reply.
173  *
174  * We use xdr_buf for holding responses as it fits well with NFS
175  * read responses (that have a header, and some data pages, and possibly
176  * a tail) and means we can share some client side routines.
177  *
178  * The xdr_buf.head kvec always points to the first page in the rq_*pages
179  * list.  The xdr_buf.pages pointer points to the second page on that
180  * list.  xdr_buf.tail points to the end of the first page.
181  * This assumes that the non-page part of an rpc reply will fit
182  * in a page - NFSd ensures this.  lockd also has no trouble.
183  *
184  * Each request/reply pair can have at most one "payload", plus two pages,
185  * one for the request, and one for the reply.
186  * We using ->sendfile to return read data, we might need one extra page
187  * if the request is not page-aligned.  So add another '1'.
188  */
189 #define RPCSVC_MAXPAGES		((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
190 				+ 2 + 1)
191 
192 /*
193  * The context of a single thread, including the request currently being
194  * processed.
195  */
196 struct svc_rqst {
197 	struct list_head	rq_all;		/* all threads list */
198 	struct rcu_head		rq_rcu_head;	/* for RCU deferred kfree */
199 	struct svc_xprt *	rq_xprt;	/* transport ptr */
200 
201 	struct sockaddr_storage	rq_addr;	/* peer address */
202 	size_t			rq_addrlen;
203 	struct sockaddr_storage	rq_daddr;	/* dest addr of request
204 						 *  - reply from here */
205 	size_t			rq_daddrlen;
206 
207 	struct svc_serv *	rq_server;	/* RPC service definition */
208 	struct svc_pool *	rq_pool;	/* thread pool */
209 	const struct svc_procedure *rq_procinfo;/* procedure info */
210 	struct auth_ops *	rq_authop;	/* authentication flavour */
211 	struct svc_cred		rq_cred;	/* auth info */
212 	void *			rq_xprt_ctxt;	/* transport specific context ptr */
213 	struct svc_deferred_req*rq_deferred;	/* deferred request we are replaying */
214 
215 	struct xdr_buf		rq_arg;
216 	struct xdr_stream	rq_arg_stream;
217 	struct xdr_stream	rq_res_stream;
218 	struct page		*rq_scratch_page;
219 	struct xdr_buf		rq_res;
220 	struct page		*rq_pages[RPCSVC_MAXPAGES + 1];
221 	struct page *		*rq_respages;	/* points into rq_pages */
222 	struct page *		*rq_next_page; /* next reply page to use */
223 	struct page *		*rq_page_end;  /* one past the last page */
224 
225 	struct folio_batch	rq_fbatch;
226 	struct kvec		rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
227 	struct bio_vec		rq_bvec[RPCSVC_MAXPAGES];
228 
229 	__be32			rq_xid;		/* transmission id */
230 	u32			rq_prog;	/* program number */
231 	u32			rq_vers;	/* program version */
232 	u32			rq_proc;	/* procedure number */
233 	u32			rq_prot;	/* IP protocol */
234 	int			rq_cachetype;	/* catering to nfsd */
235 #define	RQ_SECURE	(0)			/* secure port */
236 #define	RQ_LOCAL	(1)			/* local request */
237 #define	RQ_USEDEFERRAL	(2)			/* use deferral */
238 #define	RQ_DROPME	(3)			/* drop current reply */
239 #define	RQ_SPLICE_OK	(4)			/* turned off in gss privacy
240 						 * to prevent encrypting page
241 						 * cache pages */
242 #define	RQ_VICTIM	(5)			/* about to be shut down */
243 #define	RQ_BUSY		(6)			/* request is busy */
244 #define	RQ_DATA		(7)			/* request has data */
245 	unsigned long		rq_flags;	/* flags field */
246 	ktime_t			rq_qtime;	/* enqueue time */
247 
248 	void *			rq_argp;	/* decoded arguments */
249 	void *			rq_resp;	/* xdr'd results */
250 	__be32			*rq_accept_statp;
251 	void *			rq_auth_data;	/* flavor-specific data */
252 	__be32			rq_auth_stat;	/* authentication status */
253 	int			rq_auth_slack;	/* extra space xdr code
254 						 * should leave in head
255 						 * for krb5i, krb5p.
256 						 */
257 	int			rq_reserved;	/* space on socket outq
258 						 * reserved for this request
259 						 */
260 	ktime_t			rq_stime;	/* start time */
261 
262 	struct cache_req	rq_chandle;	/* handle passed to caches for
263 						 * request delaying
264 						 */
265 	/* Catering to nfsd */
266 	struct auth_domain *	rq_client;	/* RPC peer info */
267 	struct auth_domain *	rq_gssclient;	/* "gss/"-style peer info */
268 	struct svc_cacherep *	rq_cacherep;	/* cache info */
269 	struct task_struct	*rq_task;	/* service thread */
270 	struct net		*rq_bc_net;	/* pointer to backchannel's
271 						 * net namespace
272 						 */
273 	void **			rq_lease_breaker; /* The v4 client breaking a lease */
274 };
275 
276 #define SVC_NET(rqst) (rqst->rq_xprt ? rqst->rq_xprt->xpt_net : rqst->rq_bc_net)
277 
278 /*
279  * Rigorous type checking on sockaddr type conversions
280  */
281 static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst)
282 {
283 	return (struct sockaddr_in *) &rqst->rq_addr;
284 }
285 
286 static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst)
287 {
288 	return (struct sockaddr_in6 *) &rqst->rq_addr;
289 }
290 
291 static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst)
292 {
293 	return (struct sockaddr *) &rqst->rq_addr;
294 }
295 
296 static inline struct sockaddr_in *svc_daddr_in(const struct svc_rqst *rqst)
297 {
298 	return (struct sockaddr_in *) &rqst->rq_daddr;
299 }
300 
301 static inline struct sockaddr_in6 *svc_daddr_in6(const struct svc_rqst *rqst)
302 {
303 	return (struct sockaddr_in6 *) &rqst->rq_daddr;
304 }
305 
306 static inline struct sockaddr *svc_daddr(const struct svc_rqst *rqst)
307 {
308 	return (struct sockaddr *) &rqst->rq_daddr;
309 }
310 
311 struct svc_deferred_req {
312 	u32			prot;	/* protocol (UDP or TCP) */
313 	struct svc_xprt		*xprt;
314 	struct sockaddr_storage	addr;	/* where reply must go */
315 	size_t			addrlen;
316 	struct sockaddr_storage	daddr;	/* where reply must come from */
317 	size_t			daddrlen;
318 	void			*xprt_ctxt;
319 	struct cache_deferred_req handle;
320 	int			argslen;
321 	__be32			args[];
322 };
323 
324 struct svc_process_info {
325 	union {
326 		int  (*dispatch)(struct svc_rqst *rqstp);
327 		struct {
328 			unsigned int lovers;
329 			unsigned int hivers;
330 		} mismatch;
331 	};
332 };
333 
334 /*
335  * List of RPC programs on the same transport endpoint
336  */
337 struct svc_program {
338 	struct svc_program *	pg_next;	/* other programs (same xprt) */
339 	u32			pg_prog;	/* program number */
340 	unsigned int		pg_lovers;	/* lowest version */
341 	unsigned int		pg_hivers;	/* highest version */
342 	unsigned int		pg_nvers;	/* number of versions */
343 	const struct svc_version **pg_vers;	/* version array */
344 	char *			pg_name;	/* service name */
345 	char *			pg_class;	/* class name: services sharing authentication */
346 	struct svc_stat *	pg_stats;	/* rpc statistics */
347 	int			(*pg_authenticate)(struct svc_rqst *);
348 	__be32			(*pg_init_request)(struct svc_rqst *,
349 						   const struct svc_program *,
350 						   struct svc_process_info *);
351 	int			(*pg_rpcbind_set)(struct net *net,
352 						  const struct svc_program *,
353 						  u32 version, int family,
354 						  unsigned short proto,
355 						  unsigned short port);
356 };
357 
358 /*
359  * RPC program version
360  */
361 struct svc_version {
362 	u32			vs_vers;	/* version number */
363 	u32			vs_nproc;	/* number of procedures */
364 	const struct svc_procedure *vs_proc;	/* per-procedure info */
365 	unsigned long __percpu	*vs_count;	/* call counts */
366 	u32			vs_xdrsize;	/* xdrsize needed for this version */
367 
368 	/* Don't register with rpcbind */
369 	bool			vs_hidden;
370 
371 	/* Don't care if the rpcbind registration fails */
372 	bool			vs_rpcb_optnl;
373 
374 	/* Need xprt with congestion control */
375 	bool			vs_need_cong_ctrl;
376 
377 	/* Dispatch function */
378 	int			(*vs_dispatch)(struct svc_rqst *rqstp);
379 };
380 
381 /*
382  * RPC procedure info
383  */
384 struct svc_procedure {
385 	/* process the request: */
386 	__be32			(*pc_func)(struct svc_rqst *);
387 	/* XDR decode args: */
388 	bool			(*pc_decode)(struct svc_rqst *rqstp,
389 					     struct xdr_stream *xdr);
390 	/* XDR encode result: */
391 	bool			(*pc_encode)(struct svc_rqst *rqstp,
392 					     struct xdr_stream *xdr);
393 	/* XDR free result: */
394 	void			(*pc_release)(struct svc_rqst *);
395 	unsigned int		pc_argsize;	/* argument struct size */
396 	unsigned int		pc_argzero;	/* how much of argument to clear */
397 	unsigned int		pc_ressize;	/* result struct size */
398 	unsigned int		pc_cachetype;	/* cache info (NFS) */
399 	unsigned int		pc_xdrressize;	/* maximum size of XDR reply */
400 	const char *		pc_name;	/* for display */
401 };
402 
403 /*
404  * Function prototypes.
405  */
406 int svc_rpcb_setup(struct svc_serv *serv, struct net *net);
407 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net);
408 int svc_bind(struct svc_serv *serv, struct net *net);
409 struct svc_serv *svc_create(struct svc_program *, unsigned int,
410 			    int (*threadfn)(void *data));
411 struct svc_rqst *svc_rqst_alloc(struct svc_serv *serv,
412 					struct svc_pool *pool, int node);
413 bool		   svc_rqst_replace_page(struct svc_rqst *rqstp,
414 					 struct page *page);
415 void		   svc_rqst_release_pages(struct svc_rqst *rqstp);
416 void		   svc_rqst_free(struct svc_rqst *);
417 void		   svc_exit_thread(struct svc_rqst *);
418 struct svc_serv *  svc_create_pooled(struct svc_program *, unsigned int,
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_serv *serv, struct file *file);
422 void		   svc_process(struct svc_rqst *rqstp);
423 int		   bc_svc_process(struct svc_serv *, struct rpc_rqst *,
424 			struct svc_rqst *);
425 int		   svc_register(const struct svc_serv *, struct net *, const int,
426 				const unsigned short, const unsigned short);
427 
428 void		   svc_wake_up(struct svc_serv *);
429 void		   svc_reserve(struct svc_rqst *rqstp, int space);
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  */
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  */
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  */
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  */
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  */
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  */
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