xprt_rdma.h 20.5 KB
Newer Older
1
/* SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause */
2
/*
C
Chuck Lever 已提交
3
 * Copyright (c) 2014-2017 Oracle.  All rights reserved.
4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46
 * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the BSD-type
 * license below:
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 *      Redistributions of source code must retain the above copyright
 *      notice, this list of conditions and the following disclaimer.
 *
 *      Redistributions in binary form must reproduce the above
 *      copyright notice, this list of conditions and the following
 *      disclaimer in the documentation and/or other materials provided
 *      with the distribution.
 *
 *      Neither the name of the Network Appliance, Inc. nor the names of
 *      its contributors may be used to endorse or promote products
 *      derived from this software without specific prior written
 *      permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

#ifndef _LINUX_SUNRPC_XPRT_RDMA_H
#define _LINUX_SUNRPC_XPRT_RDMA_H

#include <linux/wait.h> 		/* wait_queue_head_t, etc */
#include <linux/spinlock.h> 		/* spinlock_t, etc */
A
Arun Sharma 已提交
47
#include <linux/atomic.h>			/* atomic_t, etc */
48
#include <linux/workqueue.h>		/* struct work_struct */
49 50 51 52 53 54 55 56

#include <rdma/rdma_cm.h>		/* RDMA connection api */
#include <rdma/ib_verbs.h>		/* RDMA verbs api */

#include <linux/sunrpc/clnt.h> 		/* rpc_xprt */
#include <linux/sunrpc/rpc_rdma.h> 	/* RPC/RDMA protocol */
#include <linux/sunrpc/xprtrdma.h> 	/* xprt parameters */

57 58 59
#define RDMA_RESOLVE_TIMEOUT	(5000)	/* 5 seconds */
#define RDMA_CONNECT_RETRY_MAX	(2)	/* retries if no listener backlog */

60 61 62 63 64
#define RPCRDMA_BIND_TO		(60U * HZ)
#define RPCRDMA_INIT_REEST_TO	(5U * HZ)
#define RPCRDMA_MAX_REEST_TO	(30U * HZ)
#define RPCRDMA_IDLE_DISC_TO	(5U * 60 * HZ)

65 66 67 68
/*
 * Interface Adapter -- one per transport instance
 */
struct rpcrdma_ia {
69
	const struct rpcrdma_memreg_ops	*ri_ops;
70
	struct ib_device	*ri_device;
71 72 73
	struct rdma_cm_id 	*ri_id;
	struct ib_pd		*ri_pd;
	struct completion	ri_done;
74
	struct completion	ri_remove_done;
75
	int			ri_async_rc;
76
	unsigned int		ri_max_segs;
77
	unsigned int		ri_max_frwr_depth;
C
Chuck Lever 已提交
78 79
	unsigned int		ri_max_inline_write;
	unsigned int		ri_max_inline_read;
80
	unsigned int		ri_max_send_sges;
81
	bool			ri_implicit_roundup;
C
Chuck Lever 已提交
82
	enum ib_mr_type		ri_mrtype;
83
	unsigned long		ri_flags;
84 85
	struct ib_qp_attr	ri_qp_attr;
	struct ib_qp_init_attr	ri_qp_init_attr;
86 87
};

88 89 90 91
enum {
	RPCRDMA_IAF_REMOVING = 0,
};

92 93 94 95 96
/*
 * RDMA Endpoint -- one per transport instance
 */

struct rpcrdma_ep {
97 98
	unsigned int		rep_send_count;
	unsigned int		rep_send_batch;
99 100 101
	int			rep_connected;
	struct ib_qp_init_attr	rep_attr;
	wait_queue_head_t 	rep_connect_wait;
102
	struct rpcrdma_connect_private	rep_cm_private;
103
	struct rdma_conn_param	rep_remote_cma;
104
	struct delayed_work	rep_connect_worker;
105 106
};

107 108 109 110 111 112 113 114 115 116
/* Pre-allocate extra Work Requests for handling backward receives
 * and sends. This is a fixed value because the Work Queues are
 * allocated when the forward channel is set up.
 */
#if defined(CONFIG_SUNRPC_BACKCHANNEL)
#define RPCRDMA_BACKWARD_WRS		(8)
#else
#define RPCRDMA_BACKWARD_WRS		(0)
#endif

117 118 119 120 121 122 123 124
/* Registered buffer -- registered kmalloc'd memory for RDMA SEND/RECV
 *
 * The below structure appears at the front of a large region of kmalloc'd
 * memory, which always starts on a good alignment boundary.
 */

struct rpcrdma_regbuf {
	struct ib_sge		rg_iov;
125
	struct ib_device	*rg_device;
126
	enum dma_data_direction	rg_direction;
127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147
	__be32			rg_base[0] __attribute__ ((aligned(256)));
};

static inline u64
rdmab_addr(struct rpcrdma_regbuf *rb)
{
	return rb->rg_iov.addr;
}

static inline u32
rdmab_length(struct rpcrdma_regbuf *rb)
{
	return rb->rg_iov.length;
}

static inline u32
rdmab_lkey(struct rpcrdma_regbuf *rb)
{
	return rb->rg_iov.lkey;
}

148 149 150 151 152 153
static inline struct ib_device *
rdmab_device(struct rpcrdma_regbuf *rb)
{
	return rb->rg_device;
}

154 155
#define RPCRDMA_DEF_GFP		(GFP_NOIO | __GFP_NOWARN)

156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173
/* To ensure a transport can always make forward progress,
 * the number of RDMA segments allowed in header chunk lists
 * is capped at 8. This prevents less-capable devices and
 * memory registrations from overrunning the Send buffer
 * while building chunk lists.
 *
 * Elements of the Read list take up more room than the
 * Write list or Reply chunk. 8 read segments means the Read
 * list (or Write list or Reply chunk) cannot consume more
 * than
 *
 * ((8 + 2) * read segment size) + 1 XDR words, or 244 bytes.
 *
 * And the fixed part of the header is another 24 bytes.
 *
 * The smallest inline threshold is 1024 bytes, ensuring that
 * at least 750 bytes are available for RPC messages.
 */
174 175 176 177
enum {
	RPCRDMA_MAX_HDR_SEGS = 8,
	RPCRDMA_HDRBUF_SIZE = 256,
};
178

179
/*
180 181 182
 * struct rpcrdma_rep -- this structure encapsulates state required
 * to receive and complete an RPC Reply, asychronously. It needs
 * several pieces of state:
183
 *
184 185 186
 *   o receive buffer and ib_sge (donated to provider)
 *   o status of receive (success or not, length, inv rkey)
 *   o bookkeeping state to get run by reply handler (XDR stream)
187
 *
188 189 190
 * These structures are allocated during transport initialization.
 * N of these are associated with a transport instance, managed by
 * struct rpcrdma_buffer. N is the max number of outstanding RPCs.
191 192 193
 */

struct rpcrdma_rep {
194
	struct ib_cqe		rr_cqe;
195 196 197
	__be32			rr_xid;
	__be32			rr_vers;
	__be32			rr_proc;
198 199
	int			rr_wc_flags;
	u32			rr_inv_rkey;
200
	struct rpcrdma_regbuf	*rr_rdmabuf;
201
	struct rpcrdma_xprt	*rr_rxprt;
202
	struct work_struct	rr_work;
203 204
	struct xdr_buf		rr_hdrbuf;
	struct xdr_stream	rr_stream;
205
	struct rpc_rqst		*rr_rqst;
206
	struct list_head	rr_list;
207
	struct ib_recv_wr	rr_recv_wr;
208 209
};

210 211
/* struct rpcrdma_sendctx - DMA mapped SGEs to unmap after Send completes
 */
212
struct rpcrdma_req;
213 214 215 216 217
struct rpcrdma_xprt;
struct rpcrdma_sendctx {
	struct ib_send_wr	sc_wr;
	struct ib_cqe		sc_cqe;
	struct rpcrdma_xprt	*sc_xprt;
218
	struct rpcrdma_req	*sc_req;
219 220 221 222 223 224 225 226 227 228 229 230 231 232
	unsigned int		sc_unmap_count;
	struct ib_sge		sc_sges[];
};

/* Limit the number of SGEs that can be unmapped during one
 * Send completion. This caps the amount of work a single
 * completion can do before returning to the provider.
 *
 * Setting this to zero disables Send completion batching.
 */
enum {
	RPCRDMA_MAX_SEND_BATCH = 7,
};

233
/*
C
Chuck Lever 已提交
234
 * struct rpcrdma_mr - external memory region metadata
235 236 237 238
 *
 * An external memory region is any buffer or page that is registered
 * on the fly (ie, not pre-registered).
 *
C
Chuck Lever 已提交
239
 * Each rpcrdma_buffer has a list of free MWs anchored in rb_mrs. During
240 241 242 243 244 245
 * call_allocate, rpcrdma_buffer_get() assigns one to each segment in
 * an rpcrdma_req. Then rpcrdma_register_external() grabs these to keep
 * track of registration metadata while each RPC is pending.
 * rpcrdma_deregister_external() uses this metadata to unmap and
 * release these resources when an RPC is complete.
 */
246 247 248 249 250
enum rpcrdma_frwr_state {
	FRWR_IS_INVALID,	/* ready to be used */
	FRWR_IS_VALID,		/* in use */
	FRWR_FLUSHED_FR,	/* flushed FASTREG WR */
	FRWR_FLUSHED_LI,	/* flushed LOCALINV WR */
251 252
};

253
struct rpcrdma_frwr {
254
	struct ib_mr			*fr_mr;
255
	struct ib_cqe			fr_cqe;
256
	enum rpcrdma_frwr_state		fr_state;
257
	struct completion		fr_linv_done;
258 259 260 261
	union {
		struct ib_reg_wr	fr_regwr;
		struct ib_send_wr	fr_invwr;
	};
262 263
};

264
struct rpcrdma_fmr {
265 266
	struct ib_fmr		*fm_mr;
	u64			*fm_physaddrs;
267 268
};

C
Chuck Lever 已提交
269 270 271 272 273
struct rpcrdma_mr {
	struct list_head	mr_list;
	struct scatterlist	*mr_sg;
	int			mr_nents;
	enum dma_data_direction	mr_dir;
274
	union {
275
		struct rpcrdma_fmr	fmr;
276
		struct rpcrdma_frwr	frwr;
277
	};
C
Chuck Lever 已提交
278 279 280 281 282
	struct rpcrdma_xprt	*mr_xprt;
	u32			mr_handle;
	u32			mr_length;
	u64			mr_offset;
	struct list_head	mr_all;
283 284
};

285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302
/*
 * struct rpcrdma_req -- structure central to the request/reply sequence.
 *
 * N of these are associated with a transport instance, and stored in
 * struct rpcrdma_buffer. N is the max number of outstanding requests.
 *
 * It includes pre-registered buffer memory for send AND recv.
 * The recv buffer, however, is not owned by this structure, and
 * is "donated" to the hardware when a recv is posted. When a
 * reply is handled, the recv buffer used is given back to the
 * struct rpcrdma_req associated with the request.
 *
 * In addition to the basic memory, this structure includes an array
 * of iovs for send operations. The reason is that the iovs passed to
 * ib_post_{send,recv} must not be modified until the work request
 * completes.
 */

303 304 305 306 307 308 309 310 311 312
/* Maximum number of page-sized "segments" per chunk list to be
 * registered or invalidated. Must handle a Reply chunk:
 */
enum {
	RPCRDMA_MAX_IOV_SEGS	= 3,
	RPCRDMA_MAX_DATA_SEGS	= ((1 * 1024 * 1024) / PAGE_SIZE) + 1,
	RPCRDMA_MAX_SEGS	= RPCRDMA_MAX_DATA_SEGS +
				  RPCRDMA_MAX_IOV_SEGS,
};

313 314 315 316 317 318
struct rpcrdma_mr_seg {		/* chunk descriptors */
	u32		mr_len;		/* length of chunk or segment */
	struct page	*mr_page;	/* owning page, if any */
	char		*mr_offset;	/* kva if no page, else offset */
};

C
Chuck Lever 已提交
319 320
/* The Send SGE array is provisioned to send a maximum size
 * inline request:
321 322
 * - RPC-over-RDMA header
 * - xdr_buf head iovec
C
Chuck Lever 已提交
323
 * - RPCRDMA_MAX_INLINE bytes, in pages
324
 * - xdr_buf tail iovec
C
Chuck Lever 已提交
325 326 327
 *
 * The actual number of array elements consumed by each RPC
 * depends on the device's max_sge limit.
328 329
 */
enum {
330
	RPCRDMA_MIN_SEND_SGES = 3,
C
Chuck Lever 已提交
331
	RPCRDMA_MAX_PAGE_SGES = RPCRDMA_MAX_INLINE >> PAGE_SHIFT,
332 333
	RPCRDMA_MAX_SEND_SGES = 1 + 1 + RPCRDMA_MAX_PAGE_SGES + 1,
};
334

335
struct rpcrdma_buffer;
336
struct rpcrdma_req {
337
	struct list_head	rl_list;
338
	struct rpcrdma_buffer	*rl_buffer;
339
	struct rpcrdma_rep	*rl_reply;
340 341
	struct xdr_stream	rl_stream;
	struct xdr_buf		rl_hdrbuf;
342
	struct rpcrdma_sendctx	*rl_sendctx;
343 344 345
	struct rpcrdma_regbuf	*rl_rdmabuf;	/* xprt header */
	struct rpcrdma_regbuf	*rl_sendbuf;	/* rq_snd_buf */
	struct rpcrdma_regbuf	*rl_recvbuf;	/* rq_rcv_buf */
346 347

	struct list_head	rl_all;
348
	unsigned long		rl_flags;
349 350 351

	struct list_head	rl_registered;	/* registered segments */
	struct rpcrdma_mr_seg	rl_segments[RPCRDMA_MAX_SEGS];
352
};
353

354 355
/* rl_flags */
enum {
356
	RPCRDMA_REQ_F_PENDING = 0,
357
	RPCRDMA_REQ_F_TX_RESOURCES,
358 359
};

360 361 362 363 364 365
static inline void
rpcrdma_set_xprtdata(struct rpc_rqst *rqst, struct rpcrdma_req *req)
{
	rqst->rq_xprtdata = req;
}

366
static inline struct rpcrdma_req *
367
rpcr_to_rdmar(const struct rpc_rqst *rqst)
368
{
369
	return rqst->rq_xprtdata;
370
}
371

372
static inline void
C
Chuck Lever 已提交
373
rpcrdma_mr_push(struct rpcrdma_mr *mr, struct list_head *list)
374
{
C
Chuck Lever 已提交
375
	list_add_tail(&mr->mr_list, list);
376 377
}

C
Chuck Lever 已提交
378 379
static inline struct rpcrdma_mr *
rpcrdma_mr_pop(struct list_head *list)
380
{
C
Chuck Lever 已提交
381
	struct rpcrdma_mr *mr;
382

C
Chuck Lever 已提交
383 384 385
	mr = list_first_entry(list, struct rpcrdma_mr, mr_list);
	list_del(&mr->mr_list);
	return mr;
386 387
}

388 389 390 391 392 393 394
/*
 * struct rpcrdma_buffer -- holds list/queue of pre-registered memory for
 * inline requests/replies, and client/server credits.
 *
 * One of these is associated with a transport instance
 */
struct rpcrdma_buffer {
C
Chuck Lever 已提交
395 396
	spinlock_t		rb_mrlock;	/* protect rb_mrs list */
	struct list_head	rb_mrs;
C
Chuck Lever 已提交
397 398
	struct list_head	rb_all;

399 400 401 402 403
	unsigned long		rb_sc_head;
	unsigned long		rb_sc_tail;
	unsigned long		rb_sc_last;
	struct rpcrdma_sendctx	**rb_sc_ctxs;

404
	spinlock_t		rb_lock;	/* protect buf lists */
405
	int			rb_send_count, rb_recv_count;
406 407
	struct list_head	rb_send_bufs;
	struct list_head	rb_recv_bufs;
C
Chuck Lever 已提交
408
	u32			rb_max_requests;
409
	u32			rb_credits;	/* most recent credit grant */
410 411 412 413

	u32			rb_bc_srv_max_requests;
	spinlock_t		rb_reqslock;	/* protect rb_allreqs */
	struct list_head	rb_allreqs;
414 415

	u32			rb_bc_max_requests;
416 417 418 419

	spinlock_t		rb_recovery_lock; /* protect rb_stale_mrs */
	struct list_head	rb_stale_mrs;
	struct delayed_work	rb_recovery_worker;
C
Chuck Lever 已提交
420
	struct delayed_work	rb_refresh_worker;
421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441
};
#define rdmab_to_ia(b) (&container_of((b), struct rpcrdma_xprt, rx_buf)->rx_ia)

/*
 * Internal structure for transport instance creation. This
 * exists primarily for modularity.
 *
 * This data should be set with mount options
 */
struct rpcrdma_create_data_internal {
	unsigned int	max_requests;	/* max requests (slots) in flight */
	unsigned int	rsize;		/* mount rsize - max read hdr+data */
	unsigned int	wsize;		/* mount wsize - max write hdr+data */
	unsigned int	inline_rsize;	/* max non-rdma read data payload */
	unsigned int	inline_wsize;	/* max non-rdma write data payload */
};

/*
 * Statistics for RPCRDMA
 */
struct rpcrdma_stats {
442
	/* accessed when sending a call */
443 444 445 446 447
	unsigned long		read_chunk_count;
	unsigned long		write_chunk_count;
	unsigned long		reply_chunk_count;
	unsigned long long	total_rdma_request;

448
	/* rarely accessed error counters */
449 450 451 452
	unsigned long long	pullup_copy_count;
	unsigned long		hardway_register_count;
	unsigned long		failed_marshal_count;
	unsigned long		bad_reply_count;
453 454
	unsigned long		mrs_recovered;
	unsigned long		mrs_orphaned;
C
Chuck Lever 已提交
455
	unsigned long		mrs_allocated;
456
	unsigned long		empty_sendctx_q;
457 458 459 460

	/* accessed when receiving a reply */
	unsigned long long	total_rdma_reply;
	unsigned long long	fixup_copy_count;
461
	unsigned long		reply_waits_for_send;
462
	unsigned long		local_inv_needed;
463 464
	unsigned long		nomsg_call_count;
	unsigned long		bcall_count;
465 466
};

467 468 469
/*
 * Per-registration mode operations
 */
470
struct rpcrdma_xprt;
471
struct rpcrdma_memreg_ops {
472 473
	struct rpcrdma_mr_seg *
			(*ro_map)(struct rpcrdma_xprt *,
474
				  struct rpcrdma_mr_seg *, int, bool,
C
Chuck Lever 已提交
475
				  struct rpcrdma_mr **);
476 477
	int		(*ro_send)(struct rpcrdma_ia *ia,
				   struct rpcrdma_req *req);
478
	void		(*ro_reminv)(struct rpcrdma_rep *rep,
C
Chuck Lever 已提交
479
				     struct list_head *mrs);
480
	void		(*ro_unmap_sync)(struct rpcrdma_xprt *,
481
					 struct list_head *);
C
Chuck Lever 已提交
482
	void		(*ro_recover_mr)(struct rpcrdma_mr *mr);
C
Chuck Lever 已提交
483 484 485
	int		(*ro_open)(struct rpcrdma_ia *,
				   struct rpcrdma_ep *,
				   struct rpcrdma_create_data_internal *);
486
	size_t		(*ro_maxpages)(struct rpcrdma_xprt *);
C
Chuck Lever 已提交
487
	int		(*ro_init_mr)(struct rpcrdma_ia *,
C
Chuck Lever 已提交
488 489
				      struct rpcrdma_mr *);
	void		(*ro_release_mr)(struct rpcrdma_mr *mr);
490
	const char	*ro_displayname;
491
	const int	ro_send_w_inv_ok;
492 493 494 495 496
};

extern const struct rpcrdma_memreg_ops rpcrdma_fmr_memreg_ops;
extern const struct rpcrdma_memreg_ops rpcrdma_frwr_memreg_ops;

497 498 499 500 501 502 503 504 505 506 507
/*
 * RPCRDMA transport -- encapsulates the structures above for
 * integration with RPC.
 *
 * The contained structures are embedded, not pointers,
 * for convenience. This structure need not be visible externally.
 *
 * It is allocated and initialized during mount, and released
 * during unmount.
 */
struct rpcrdma_xprt {
508
	struct rpc_xprt		rx_xprt;
509 510 511 512
	struct rpcrdma_ia	rx_ia;
	struct rpcrdma_ep	rx_ep;
	struct rpcrdma_buffer	rx_buf;
	struct rpcrdma_create_data_internal rx_data;
513
	struct delayed_work	rx_connect_worker;
514 515 516
	struct rpcrdma_stats	rx_stats;
};

517
#define rpcx_to_rdmax(x) container_of(x, struct rpcrdma_xprt, rx_xprt)
518 519
#define rpcx_to_rdmad(x) (rpcx_to_rdmax(x)->rx_data)

520 521 522 523 524 525 526 527 528 529 530 531
static inline const char *
rpcrdma_addrstr(const struct rpcrdma_xprt *r_xprt)
{
	return r_xprt->rx_xprt.address_strings[RPC_DISPLAY_ADDR];
}

static inline const char *
rpcrdma_portstr(const struct rpcrdma_xprt *r_xprt)
{
	return r_xprt->rx_xprt.address_strings[RPC_DISPLAY_PORT];
}

532 533 534 535 536
/* Setting this to 0 ensures interoperability with early servers.
 * Setting this to 1 enhances certain unaligned read/write performance.
 * Default is 0, see sysctl entry and rpc_rdma.c rpcrdma_convert_iovs() */
extern int xprt_rdma_pad_optimize;

537 538 539 540 541
/* This setting controls the hunt for a supported memory
 * registration strategy.
 */
extern unsigned int xprt_rdma_memreg_strategy;

542 543 544
/*
 * Interface Adapter calls - xprtrdma/verbs.c
 */
545
int rpcrdma_ia_open(struct rpcrdma_xprt *xprt);
546
void rpcrdma_ia_remove(struct rpcrdma_ia *ia);
547
void rpcrdma_ia_close(struct rpcrdma_ia *);
548 549
bool frwr_is_supported(struct rpcrdma_ia *);
bool fmr_is_supported(struct rpcrdma_ia *);
550

551 552
extern struct workqueue_struct *rpcrdma_receive_wq;

553 554 555 556 557
/*
 * Endpoint calls - xprtrdma/verbs.c
 */
int rpcrdma_ep_create(struct rpcrdma_ep *, struct rpcrdma_ia *,
				struct rpcrdma_create_data_internal *);
558
void rpcrdma_ep_destroy(struct rpcrdma_ep *, struct rpcrdma_ia *);
559
int rpcrdma_ep_connect(struct rpcrdma_ep *, struct rpcrdma_ia *);
560
void rpcrdma_conn_func(struct rpcrdma_ep *ep);
561
void rpcrdma_ep_disconnect(struct rpcrdma_ep *, struct rpcrdma_ia *);
562 563 564

int rpcrdma_ep_post(struct rpcrdma_ia *, struct rpcrdma_ep *,
				struct rpcrdma_req *);
565
int rpcrdma_ep_post_recv(struct rpcrdma_ia *, struct rpcrdma_rep *);
566 567 568 569

/*
 * Buffer calls - xprtrdma/verbs.c
 */
570
struct rpcrdma_req *rpcrdma_create_req(struct rpcrdma_xprt *);
571
void rpcrdma_destroy_req(struct rpcrdma_req *);
572
int rpcrdma_create_rep(struct rpcrdma_xprt *r_xprt);
573
int rpcrdma_buffer_create(struct rpcrdma_xprt *);
574
void rpcrdma_buffer_destroy(struct rpcrdma_buffer *);
575 576
struct rpcrdma_sendctx *rpcrdma_sendctx_get_locked(struct rpcrdma_buffer *buf);
void rpcrdma_sendctx_put_locked(struct rpcrdma_sendctx *sc);
577

C
Chuck Lever 已提交
578 579
struct rpcrdma_mr *rpcrdma_mr_get(struct rpcrdma_xprt *r_xprt);
void rpcrdma_mr_put(struct rpcrdma_mr *mr);
580
void rpcrdma_mr_unmap_and_put(struct rpcrdma_mr *mr);
C
Chuck Lever 已提交
581 582
void rpcrdma_mr_defer_recovery(struct rpcrdma_mr *mr);

583 584 585 586 587
struct rpcrdma_req *rpcrdma_buffer_get(struct rpcrdma_buffer *);
void rpcrdma_buffer_put(struct rpcrdma_req *);
void rpcrdma_recv_buffer_get(struct rpcrdma_req *);
void rpcrdma_recv_buffer_put(struct rpcrdma_rep *);

588
struct rpcrdma_regbuf *rpcrdma_alloc_regbuf(size_t, enum dma_data_direction,
589
					    gfp_t);
590
bool __rpcrdma_dma_map_regbuf(struct rpcrdma_ia *, struct rpcrdma_regbuf *);
591
void rpcrdma_free_regbuf(struct rpcrdma_regbuf *);
592

593 594 595 596 597 598 599 600 601 602 603 604 605 606
static inline bool
rpcrdma_regbuf_is_mapped(struct rpcrdma_regbuf *rb)
{
	return rb->rg_device != NULL;
}

static inline bool
rpcrdma_dma_map_regbuf(struct rpcrdma_ia *ia, struct rpcrdma_regbuf *rb)
{
	if (likely(rpcrdma_regbuf_is_mapped(rb)))
		return true;
	return __rpcrdma_dma_map_regbuf(ia, rb);
}

607
int rpcrdma_ep_post_extra_recv(struct rpcrdma_xprt *, unsigned int);
608

609 610 611
int rpcrdma_alloc_wq(void);
void rpcrdma_destroy_wq(void);

612 613 614 615 616 617 618 619 620 621
/*
 * Wrappers for chunk registration, shared by read/write chunk code.
 */

static inline enum dma_data_direction
rpcrdma_data_dir(bool writing)
{
	return writing ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
}

622 623 624
/*
 * RPC/RDMA protocol calls - xprtrdma/rpc_rdma.c
 */
625 626 627 628 629 630 631 632 633

enum rpcrdma_chunktype {
	rpcrdma_noch = 0,
	rpcrdma_readch,
	rpcrdma_areadch,
	rpcrdma_writech,
	rpcrdma_replych
};

634 635 636 637
int rpcrdma_prepare_send_sges(struct rpcrdma_xprt *r_xprt,
			      struct rpcrdma_req *req, u32 hdrlen,
			      struct xdr_buf *xdr,
			      enum rpcrdma_chunktype rtype);
638
void rpcrdma_unmap_sendctx(struct rpcrdma_sendctx *sc);
639
int rpcrdma_marshal_req(struct rpcrdma_xprt *r_xprt, struct rpc_rqst *rqst);
640
void rpcrdma_set_max_header_sizes(struct rpcrdma_xprt *);
641
void rpcrdma_complete_rqst(struct rpcrdma_rep *rep);
642
void rpcrdma_reply_handler(struct rpcrdma_rep *rep);
643 644
void rpcrdma_release_rqst(struct rpcrdma_xprt *r_xprt,
			  struct rpcrdma_req *req);
645
void rpcrdma_deferred_completion(struct work_struct *work);
646

647 648 649 650 651 652
static inline void rpcrdma_set_xdrlen(struct xdr_buf *xdr, size_t len)
{
	xdr->head[0].iov_len = len;
	xdr->len = len;
}

653 654
/* RPC/RDMA module init - xprtrdma/transport.c
 */
655 656 657
extern unsigned int xprt_rdma_max_inline_read;
void xprt_rdma_format_addresses(struct rpc_xprt *xprt, struct sockaddr *sap);
void xprt_rdma_free_addresses(struct rpc_xprt *xprt);
658
void rpcrdma_connect_worker(struct work_struct *work);
659
void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq);
660 661 662
int xprt_rdma_init(void);
void xprt_rdma_cleanup(void);

663 664 665 666
/* Backchannel calls - xprtrdma/backchannel.c
 */
#if defined(CONFIG_SUNRPC_BACKCHANNEL)
int xprt_rdma_bc_setup(struct rpc_xprt *, unsigned int);
667
int xprt_rdma_bc_up(struct svc_serv *, struct net *);
668
size_t xprt_rdma_bc_maxpayload(struct rpc_xprt *);
669
int rpcrdma_bc_post_recv(struct rpcrdma_xprt *, unsigned int);
670
void rpcrdma_bc_receive_call(struct rpcrdma_xprt *, struct rpcrdma_rep *);
671
int xprt_rdma_bc_send_reply(struct rpc_rqst *rqst);
672 673 674 675
void xprt_rdma_bc_free_rqst(struct rpc_rqst *);
void xprt_rdma_bc_destroy(struct rpc_xprt *, unsigned int);
#endif	/* CONFIG_SUNRPC_BACKCHANNEL */

676
extern struct xprt_class xprt_rdma_bc;
677

678
#endif				/* _LINUX_SUNRPC_XPRT_RDMA_H */
679 680

#include <trace/events/rpcrdma.h>