xref: /linux-6.15/include/linux/sunrpc/xprt.h (revision bb6d3fb3)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  *  linux/include/linux/sunrpc/xprt.h
4  *
5  *  Declarations for the RPC transport interface.
6  *
7  *  Copyright (C) 1995, 1996 Olaf Kirch <[email protected]>
8  */
9 
10 #ifndef _LINUX_SUNRPC_XPRT_H
11 #define _LINUX_SUNRPC_XPRT_H
12 
13 #include <linux/uio.h>
14 #include <linux/socket.h>
15 #include <linux/in.h>
16 #include <linux/ktime.h>
17 #include <linux/kref.h>
18 #include <linux/sunrpc/sched.h>
19 #include <linux/sunrpc/xdr.h>
20 #include <linux/sunrpc/msg_prot.h>
21 
22 #define RPC_MIN_SLOT_TABLE	(2U)
23 #define RPC_DEF_SLOT_TABLE	(16U)
24 #define RPC_MAX_SLOT_TABLE_LIMIT	(65536U)
25 #define RPC_MAX_SLOT_TABLE	RPC_MAX_SLOT_TABLE_LIMIT
26 
27 #define RPC_CWNDSHIFT		(8U)
28 #define RPC_CWNDSCALE		(1U << RPC_CWNDSHIFT)
29 #define RPC_INITCWND		RPC_CWNDSCALE
30 #define RPC_MAXCWND(xprt)	((xprt)->max_reqs << RPC_CWNDSHIFT)
31 #define RPCXPRT_CONGESTED(xprt) ((xprt)->cong >= (xprt)->cwnd)
32 
33 /*
34  * This describes a timeout strategy
35  */
36 struct rpc_timeout {
37 	unsigned long		to_initval,		/* initial timeout */
38 				to_maxval,		/* max timeout */
39 				to_increment;		/* if !exponential */
40 	unsigned int		to_retries;		/* max # of retries */
41 	unsigned char		to_exponential;
42 };
43 
44 enum rpc_display_format_t {
45 	RPC_DISPLAY_ADDR = 0,
46 	RPC_DISPLAY_PORT,
47 	RPC_DISPLAY_PROTO,
48 	RPC_DISPLAY_HEX_ADDR,
49 	RPC_DISPLAY_HEX_PORT,
50 	RPC_DISPLAY_NETID,
51 	RPC_DISPLAY_MAX,
52 };
53 
54 struct rpc_task;
55 struct rpc_xprt;
56 struct seq_file;
57 struct svc_serv;
58 struct net;
59 
60 /*
61  * This describes a complete RPC request
62  */
63 struct rpc_rqst {
64 	/*
65 	 * This is the user-visible part
66 	 */
67 	struct rpc_xprt *	rq_xprt;		/* RPC client */
68 	struct xdr_buf		rq_snd_buf;		/* send buffer */
69 	struct xdr_buf		rq_rcv_buf;		/* recv buffer */
70 
71 	/*
72 	 * This is the private part
73 	 */
74 	struct rpc_task *	rq_task;	/* RPC task data */
75 	struct rpc_cred *	rq_cred;	/* Bound cred */
76 	__be32			rq_xid;		/* request XID */
77 	int			rq_cong;	/* has incremented xprt->cong */
78 	u32			rq_seqno;	/* gss seq no. used on req. */
79 	int			rq_enc_pages_num;
80 	struct page		**rq_enc_pages;	/* scratch pages for use by
81 						   gss privacy code */
82 	void (*rq_release_snd_buf)(struct rpc_rqst *); /* release rq_enc_pages */
83 
84 	union {
85 		struct list_head	rq_list;	/* Slot allocation list */
86 		struct rb_node		rq_recv;	/* Receive queue */
87 	};
88 
89 	struct list_head	rq_xmit;	/* Send queue */
90 	struct list_head	rq_xmit2;	/* Send queue */
91 
92 	void			*rq_buffer;	/* Call XDR encode buffer */
93 	size_t			rq_callsize;
94 	void			*rq_rbuffer;	/* Reply XDR decode buffer */
95 	size_t			rq_rcvsize;
96 	size_t			rq_xmit_bytes_sent;	/* total bytes sent */
97 	size_t			rq_reply_bytes_recvd;	/* total reply bytes */
98 							/* received */
99 
100 	struct xdr_buf		rq_private_buf;		/* The receive buffer
101 							 * used in the softirq.
102 							 */
103 	unsigned long		rq_majortimeo;	/* major timeout alarm */
104 	unsigned long		rq_timeout;	/* Current timeout value */
105 	ktime_t			rq_rtt;		/* round-trip time */
106 	unsigned int		rq_retries;	/* # of retries */
107 	unsigned int		rq_connect_cookie;
108 						/* A cookie used to track the
109 						   state of the transport
110 						   connection */
111 	atomic_t		rq_pin;
112 
113 	/*
114 	 * Partial send handling
115 	 */
116 	u32			rq_bytes_sent;	/* Bytes we have sent */
117 
118 	ktime_t			rq_xtime;	/* transmit time stamp */
119 	int			rq_ntrans;
120 
121 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
122 	struct list_head	rq_bc_list;	/* Callback service list */
123 	unsigned long		rq_bc_pa_state;	/* Backchannel prealloc state */
124 	struct list_head	rq_bc_pa_list;	/* Backchannel prealloc list */
125 #endif /* CONFIG_SUNRPC_BACKCHANEL */
126 };
127 #define rq_svec			rq_snd_buf.head
128 #define rq_slen			rq_snd_buf.len
129 
130 struct rpc_xprt_ops {
131 	void		(*set_buffer_size)(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize);
132 	int		(*reserve_xprt)(struct rpc_xprt *xprt, struct rpc_task *task);
133 	void		(*release_xprt)(struct rpc_xprt *xprt, struct rpc_task *task);
134 	void		(*alloc_slot)(struct rpc_xprt *xprt, struct rpc_task *task);
135 	void		(*free_slot)(struct rpc_xprt *xprt,
136 				     struct rpc_rqst *req);
137 	void		(*rpcbind)(struct rpc_task *task);
138 	void		(*set_port)(struct rpc_xprt *xprt, unsigned short port);
139 	void		(*connect)(struct rpc_xprt *xprt, struct rpc_task *task);
140 	int		(*buf_alloc)(struct rpc_task *task);
141 	void		(*buf_free)(struct rpc_task *task);
142 	void		(*prepare_request)(struct rpc_rqst *req);
143 	int		(*send_request)(struct rpc_rqst *req);
144 	void		(*wait_for_reply_request)(struct rpc_task *task);
145 	void		(*timer)(struct rpc_xprt *xprt, struct rpc_task *task);
146 	void		(*release_request)(struct rpc_task *task);
147 	void		(*close)(struct rpc_xprt *xprt);
148 	void		(*destroy)(struct rpc_xprt *xprt);
149 	void		(*set_connect_timeout)(struct rpc_xprt *xprt,
150 					unsigned long connect_timeout,
151 					unsigned long reconnect_timeout);
152 	void		(*print_stats)(struct rpc_xprt *xprt, struct seq_file *seq);
153 	int		(*enable_swap)(struct rpc_xprt *xprt);
154 	void		(*disable_swap)(struct rpc_xprt *xprt);
155 	void		(*inject_disconnect)(struct rpc_xprt *xprt);
156 	int		(*bc_setup)(struct rpc_xprt *xprt,
157 				    unsigned int min_reqs);
158 	size_t		(*bc_maxpayload)(struct rpc_xprt *xprt);
159 	unsigned int	(*bc_num_slots)(struct rpc_xprt *xprt);
160 	void		(*bc_free_rqst)(struct rpc_rqst *rqst);
161 	void		(*bc_destroy)(struct rpc_xprt *xprt,
162 				      unsigned int max_reqs);
163 };
164 
165 /*
166  * RPC transport identifiers
167  *
168  * To preserve compatibility with the historical use of raw IP protocol
169  * id's for transport selection, UDP and TCP identifiers are specified
170  * with the previous values. No such restriction exists for new transports,
171  * except that they may not collide with these values (17 and 6,
172  * respectively).
173  */
174 #define XPRT_TRANSPORT_BC       (1 << 31)
175 enum xprt_transports {
176 	XPRT_TRANSPORT_UDP	= IPPROTO_UDP,
177 	XPRT_TRANSPORT_TCP	= IPPROTO_TCP,
178 	XPRT_TRANSPORT_BC_TCP	= IPPROTO_TCP | XPRT_TRANSPORT_BC,
179 	XPRT_TRANSPORT_RDMA	= 256,
180 	XPRT_TRANSPORT_BC_RDMA	= XPRT_TRANSPORT_RDMA | XPRT_TRANSPORT_BC,
181 	XPRT_TRANSPORT_LOCAL	= 257,
182 };
183 
184 struct rpc_xprt {
185 	struct kref		kref;		/* Reference count */
186 	const struct rpc_xprt_ops *ops;		/* transport methods */
187 
188 	const struct rpc_timeout *timeout;	/* timeout parms */
189 	struct sockaddr_storage	addr;		/* server address */
190 	size_t			addrlen;	/* size of server address */
191 	int			prot;		/* IP protocol */
192 
193 	unsigned long		cong;		/* current congestion */
194 	unsigned long		cwnd;		/* congestion window */
195 
196 	size_t			max_payload;	/* largest RPC payload size,
197 						   in bytes */
198 
199 	struct rpc_wait_queue	binding;	/* requests waiting on rpcbind */
200 	struct rpc_wait_queue	sending;	/* requests waiting to send */
201 	struct rpc_wait_queue	pending;	/* requests in flight */
202 	struct rpc_wait_queue	backlog;	/* waiting for slot */
203 	struct list_head	free;		/* free slots */
204 	unsigned int		max_reqs;	/* max number of slots */
205 	unsigned int		min_reqs;	/* min number of slots */
206 	unsigned int		num_reqs;	/* total slots */
207 	unsigned long		state;		/* transport state */
208 	unsigned char		resvport   : 1,	/* use a reserved port */
209 				reuseport  : 1; /* reuse port on reconnect */
210 	atomic_t		swapper;	/* we're swapping over this
211 						   transport */
212 	unsigned int		bind_index;	/* bind function index */
213 
214 	/*
215 	 * Multipath
216 	 */
217 	struct list_head	xprt_switch;
218 
219 	/*
220 	 * Connection of transports
221 	 */
222 	unsigned long		bind_timeout,
223 				reestablish_timeout;
224 	unsigned int		connect_cookie;	/* A cookie that gets bumped
225 						   every time the transport
226 						   is reconnected */
227 
228 	/*
229 	 * Disconnection of idle transports
230 	 */
231 	struct work_struct	task_cleanup;
232 	struct timer_list	timer;
233 	unsigned long		last_used,
234 				idle_timeout,
235 				connect_timeout,
236 				max_reconnect_timeout;
237 
238 	/*
239 	 * Send stuff
240 	 */
241 	atomic_long_t		queuelen;
242 	spinlock_t		transport_lock;	/* lock transport info */
243 	spinlock_t		reserve_lock;	/* lock slot table */
244 	spinlock_t		queue_lock;	/* send/receive queue lock */
245 	u32			xid;		/* Next XID value to use */
246 	struct rpc_task *	snd_task;	/* Task blocked in send */
247 
248 	struct list_head	xmit_queue;	/* Send queue */
249 
250 	struct svc_xprt		*bc_xprt;	/* NFSv4.1 backchannel */
251 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
252 	struct svc_serv		*bc_serv;       /* The RPC service which will */
253 						/* process the callback */
254 	unsigned int		bc_alloc_max;
255 	unsigned int		bc_alloc_count;	/* Total number of preallocs */
256 	atomic_t		bc_slot_count;	/* Number of allocated slots */
257 	spinlock_t		bc_pa_lock;	/* Protects the preallocated
258 						 * items */
259 	struct list_head	bc_pa_list;	/* List of preallocated
260 						 * backchannel rpc_rqst's */
261 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
262 
263 	struct rb_root		recv_queue;	/* Receive queue */
264 
265 	struct {
266 		unsigned long		bind_count,	/* total number of binds */
267 					connect_count,	/* total number of connects */
268 					connect_start,	/* connect start timestamp */
269 					connect_time,	/* jiffies waiting for connect */
270 					sends,		/* how many complete requests */
271 					recvs,		/* how many complete requests */
272 					bad_xids,	/* lookup_rqst didn't find XID */
273 					max_slots;	/* max rpc_slots used */
274 
275 		unsigned long long	req_u,		/* average requests on the wire */
276 					bklog_u,	/* backlog queue utilization */
277 					sending_u,	/* send q utilization */
278 					pending_u;	/* pend q utilization */
279 	} stat;
280 
281 	struct net		*xprt_net;
282 	const char		*servername;
283 	const char		*address_strings[RPC_DISPLAY_MAX];
284 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
285 	struct dentry		*debugfs;		/* debugfs directory */
286 	atomic_t		inject_disconnect;
287 #endif
288 	struct rcu_head		rcu;
289 };
290 
291 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
292 /*
293  * Backchannel flags
294  */
295 #define	RPC_BC_PA_IN_USE	0x0001		/* Preallocated backchannel */
296 						/* buffer in use */
297 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
298 
299 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
300 static inline int bc_prealloc(struct rpc_rqst *req)
301 {
302 	return test_bit(RPC_BC_PA_IN_USE, &req->rq_bc_pa_state);
303 }
304 #else
305 static inline int bc_prealloc(struct rpc_rqst *req)
306 {
307 	return 0;
308 }
309 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
310 
311 #define XPRT_CREATE_INFINITE_SLOTS	(1U)
312 #define XPRT_CREATE_NO_IDLE_TIMEOUT	(1U << 1)
313 
314 struct xprt_create {
315 	int			ident;		/* XPRT_TRANSPORT identifier */
316 	struct net *		net;
317 	struct sockaddr *	srcaddr;	/* optional local address */
318 	struct sockaddr *	dstaddr;	/* remote peer address */
319 	size_t			addrlen;
320 	const char		*servername;
321 	struct svc_xprt		*bc_xprt;	/* NFSv4.1 backchannel */
322 	struct rpc_xprt_switch	*bc_xps;
323 	unsigned int		flags;
324 };
325 
326 struct xprt_class {
327 	struct list_head	list;
328 	int			ident;		/* XPRT_TRANSPORT identifier */
329 	struct rpc_xprt *	(*setup)(struct xprt_create *);
330 	struct module		*owner;
331 	char			name[32];
332 };
333 
334 /*
335  * Generic internal transport functions
336  */
337 struct rpc_xprt		*xprt_create_transport(struct xprt_create *args);
338 void			xprt_connect(struct rpc_task *task);
339 unsigned long		xprt_reconnect_delay(const struct rpc_xprt *xprt);
340 void			xprt_reconnect_backoff(struct rpc_xprt *xprt,
341 					       unsigned long init_to);
342 void			xprt_reserve(struct rpc_task *task);
343 void			xprt_retry_reserve(struct rpc_task *task);
344 int			xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task);
345 int			xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task);
346 void			xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task);
347 void			xprt_free_slot(struct rpc_xprt *xprt,
348 				       struct rpc_rqst *req);
349 void			xprt_request_prepare(struct rpc_rqst *req);
350 bool			xprt_prepare_transmit(struct rpc_task *task);
351 void			xprt_request_enqueue_transmit(struct rpc_task *task);
352 void			xprt_request_enqueue_receive(struct rpc_task *task);
353 void			xprt_request_wait_receive(struct rpc_task *task);
354 void			xprt_request_dequeue_xprt(struct rpc_task *task);
355 bool			xprt_request_need_retransmit(struct rpc_task *task);
356 void			xprt_transmit(struct rpc_task *task);
357 void			xprt_end_transmit(struct rpc_task *task);
358 int			xprt_adjust_timeout(struct rpc_rqst *req);
359 void			xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task);
360 void			xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task);
361 void			xprt_release(struct rpc_task *task);
362 struct rpc_xprt *	xprt_get(struct rpc_xprt *xprt);
363 void			xprt_put(struct rpc_xprt *xprt);
364 struct rpc_xprt *	xprt_alloc(struct net *net, size_t size,
365 				unsigned int num_prealloc,
366 				unsigned int max_req);
367 void			xprt_free(struct rpc_xprt *);
368 
369 static inline int
370 xprt_enable_swap(struct rpc_xprt *xprt)
371 {
372 	return xprt->ops->enable_swap(xprt);
373 }
374 
375 static inline void
376 xprt_disable_swap(struct rpc_xprt *xprt)
377 {
378 	xprt->ops->disable_swap(xprt);
379 }
380 
381 /*
382  * Transport switch helper functions
383  */
384 int			xprt_register_transport(struct xprt_class *type);
385 int			xprt_unregister_transport(struct xprt_class *type);
386 int			xprt_load_transport(const char *);
387 void			xprt_wait_for_reply_request_def(struct rpc_task *task);
388 void			xprt_wait_for_reply_request_rtt(struct rpc_task *task);
389 void			xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status);
390 void			xprt_wait_for_buffer_space(struct rpc_xprt *xprt);
391 bool			xprt_write_space(struct rpc_xprt *xprt);
392 void			xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result);
393 struct rpc_rqst *	xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid);
394 void			xprt_update_rtt(struct rpc_task *task);
395 void			xprt_complete_rqst(struct rpc_task *task, int copied);
396 void			xprt_pin_rqst(struct rpc_rqst *req);
397 void			xprt_unpin_rqst(struct rpc_rqst *req);
398 void			xprt_release_rqst_cong(struct rpc_task *task);
399 bool			xprt_request_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req);
400 void			xprt_disconnect_done(struct rpc_xprt *xprt);
401 void			xprt_force_disconnect(struct rpc_xprt *xprt);
402 void			xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie);
403 
404 bool			xprt_lock_connect(struct rpc_xprt *, struct rpc_task *, void *);
405 void			xprt_unlock_connect(struct rpc_xprt *, void *);
406 
407 /*
408  * Reserved bit positions in xprt->state
409  */
410 #define XPRT_LOCKED		(0)
411 #define XPRT_CONNECTED		(1)
412 #define XPRT_CONNECTING		(2)
413 #define XPRT_CLOSE_WAIT		(3)
414 #define XPRT_BOUND		(4)
415 #define XPRT_BINDING		(5)
416 #define XPRT_CLOSING		(6)
417 #define XPRT_CONGESTED		(9)
418 #define XPRT_CWND_WAIT		(10)
419 #define XPRT_WRITE_SPACE	(11)
420 
421 static inline void xprt_set_connected(struct rpc_xprt *xprt)
422 {
423 	set_bit(XPRT_CONNECTED, &xprt->state);
424 }
425 
426 static inline void xprt_clear_connected(struct rpc_xprt *xprt)
427 {
428 	clear_bit(XPRT_CONNECTED, &xprt->state);
429 }
430 
431 static inline int xprt_connected(struct rpc_xprt *xprt)
432 {
433 	return test_bit(XPRT_CONNECTED, &xprt->state);
434 }
435 
436 static inline int xprt_test_and_set_connected(struct rpc_xprt *xprt)
437 {
438 	return test_and_set_bit(XPRT_CONNECTED, &xprt->state);
439 }
440 
441 static inline int xprt_test_and_clear_connected(struct rpc_xprt *xprt)
442 {
443 	return test_and_clear_bit(XPRT_CONNECTED, &xprt->state);
444 }
445 
446 static inline void xprt_clear_connecting(struct rpc_xprt *xprt)
447 {
448 	smp_mb__before_atomic();
449 	clear_bit(XPRT_CONNECTING, &xprt->state);
450 	smp_mb__after_atomic();
451 }
452 
453 static inline int xprt_connecting(struct rpc_xprt *xprt)
454 {
455 	return test_bit(XPRT_CONNECTING, &xprt->state);
456 }
457 
458 static inline int xprt_test_and_set_connecting(struct rpc_xprt *xprt)
459 {
460 	return test_and_set_bit(XPRT_CONNECTING, &xprt->state);
461 }
462 
463 static inline void xprt_set_bound(struct rpc_xprt *xprt)
464 {
465 	test_and_set_bit(XPRT_BOUND, &xprt->state);
466 }
467 
468 static inline int xprt_bound(struct rpc_xprt *xprt)
469 {
470 	return test_bit(XPRT_BOUND, &xprt->state);
471 }
472 
473 static inline void xprt_clear_bound(struct rpc_xprt *xprt)
474 {
475 	clear_bit(XPRT_BOUND, &xprt->state);
476 }
477 
478 static inline void xprt_clear_binding(struct rpc_xprt *xprt)
479 {
480 	smp_mb__before_atomic();
481 	clear_bit(XPRT_BINDING, &xprt->state);
482 	smp_mb__after_atomic();
483 }
484 
485 static inline int xprt_test_and_set_binding(struct rpc_xprt *xprt)
486 {
487 	return test_and_set_bit(XPRT_BINDING, &xprt->state);
488 }
489 
490 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
491 extern unsigned int rpc_inject_disconnect;
492 static inline void xprt_inject_disconnect(struct rpc_xprt *xprt)
493 {
494 	if (!rpc_inject_disconnect)
495 		return;
496 	if (atomic_dec_return(&xprt->inject_disconnect))
497 		return;
498 	atomic_set(&xprt->inject_disconnect, rpc_inject_disconnect);
499 	xprt->ops->inject_disconnect(xprt);
500 }
501 #else
502 static inline void xprt_inject_disconnect(struct rpc_xprt *xprt)
503 {
504 }
505 #endif
506 
507 #endif /* _LINUX_SUNRPC_XPRT_H */
508