blob: 8b9954c41ee4dd36bbe0c777d945ac507003791a [file] [log] [blame]
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -04001/*
2 * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
3 *
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the BSD-type
8 * license below:
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 *
14 * Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 *
17 * Redistributions in binary form must reproduce the above
18 * copyright notice, this list of conditions and the following
19 * disclaimer in the documentation and/or other materials provided
20 * with the distribution.
21 *
22 * Neither the name of the Network Appliance, Inc. nor the names of
23 * its contributors may be used to endorse or promote products
24 * derived from this software without specific prior written
25 * permission.
26 *
27 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
28 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
29 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
30 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
31 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
32 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
33 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
34 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
35 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
36 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
37 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
38 */
39
40#ifndef _LINUX_SUNRPC_XPRT_RDMA_H
41#define _LINUX_SUNRPC_XPRT_RDMA_H
42
43#include <linux/wait.h> /* wait_queue_head_t, etc */
44#include <linux/spinlock.h> /* spinlock_t, etc */
Arun Sharma600634972011-07-26 16:09:06 -070045#include <linux/atomic.h> /* atomic_t, etc */
Chuck Lever254f91e2014-05-28 10:32:17 -040046#include <linux/workqueue.h> /* struct work_struct */
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -040047
48#include <rdma/rdma_cm.h> /* RDMA connection api */
49#include <rdma/ib_verbs.h> /* RDMA verbs api */
50
51#include <linux/sunrpc/clnt.h> /* rpc_xprt */
52#include <linux/sunrpc/rpc_rdma.h> /* RPC/RDMA protocol */
53#include <linux/sunrpc/xprtrdma.h> /* xprt parameters */
54
Tom Talpey5675add2008-10-09 15:01:41 -040055#define RDMA_RESOLVE_TIMEOUT (5000) /* 5 seconds */
56#define RDMA_CONNECT_RETRY_MAX (2) /* retries if no listener backlog */
57
Chuck Lever5d252f92016-01-07 14:50:10 -050058#define RPCRDMA_BIND_TO (60U * HZ)
59#define RPCRDMA_INIT_REEST_TO (5U * HZ)
60#define RPCRDMA_MAX_REEST_TO (30U * HZ)
61#define RPCRDMA_IDLE_DISC_TO (5U * 60 * HZ)
62
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -040063/*
64 * Interface Adapter -- one per transport instance
65 */
66struct rpcrdma_ia {
Chuck Levera0ce85f2015-03-30 14:34:21 -040067 const struct rpcrdma_memreg_ops *ri_ops;
Chuck Lever89e0d1122015-05-26 11:51:56 -040068 struct ib_device *ri_device;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -040069 struct rdma_cm_id *ri_id;
70 struct ib_pd *ri_pd;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -040071 struct completion ri_done;
Chuck Leverbebd0312017-04-11 13:23:10 -040072 struct completion ri_remove_done;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -040073 int ri_async_rc;
Chuck Lever87cfb9a2016-09-15 10:57:07 -040074 unsigned int ri_max_segs;
Steve Wise0fc6c4e2014-05-28 10:32:00 -040075 unsigned int ri_max_frmr_depth;
Chuck Lever302d3de2016-05-02 14:41:05 -040076 unsigned int ri_max_inline_write;
77 unsigned int ri_max_inline_read;
Chuck Lever16f906d2017-02-08 17:00:10 -050078 unsigned int ri_max_send_sges;
Chuck Leverc8b920b2016-09-15 10:57:16 -040079 bool ri_reminv_expected;
Chuck Leverb5f0afb2017-02-08 16:59:54 -050080 bool ri_implicit_roundup;
Chuck Lever5e9fc6a2016-11-29 10:52:24 -050081 enum ib_mr_type ri_mrtype;
Chuck Leverbebd0312017-04-11 13:23:10 -040082 unsigned long ri_flags;
Chuck Leverce1ab9a2015-01-21 11:03:35 -050083 struct ib_qp_attr ri_qp_attr;
84 struct ib_qp_init_attr ri_qp_init_attr;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -040085};
86
Chuck Leverbebd0312017-04-11 13:23:10 -040087enum {
88 RPCRDMA_IAF_REMOVING = 0,
89};
90
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -040091/*
92 * RDMA Endpoint -- one per transport instance
93 */
94
95struct rpcrdma_ep {
Chuck Leverae729502017-10-20 10:48:12 -040096 unsigned int rep_send_count;
97 unsigned int rep_send_batch;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -040098 int rep_connected;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -040099 struct ib_qp_init_attr rep_attr;
100 wait_queue_head_t rep_connect_wait;
Chuck Lever87cfb9a2016-09-15 10:57:07 -0400101 struct rpcrdma_connect_private rep_cm_private;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400102 struct rdma_conn_param rep_remote_cma;
103 struct sockaddr_storage rep_remote_addr;
Chuck Lever254f91e2014-05-28 10:32:17 -0400104 struct delayed_work rep_connect_worker;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400105};
106
Chuck Lever124fa172015-10-24 17:27:51 -0400107/* Pre-allocate extra Work Requests for handling backward receives
108 * and sends. This is a fixed value because the Work Queues are
109 * allocated when the forward channel is set up.
110 */
111#if defined(CONFIG_SUNRPC_BACKCHANNEL)
112#define RPCRDMA_BACKWARD_WRS (8)
113#else
114#define RPCRDMA_BACKWARD_WRS (0)
115#endif
116
Chuck Lever9128c3e2015-01-21 11:04:00 -0500117/* Registered buffer -- registered kmalloc'd memory for RDMA SEND/RECV
118 *
119 * The below structure appears at the front of a large region of kmalloc'd
120 * memory, which always starts on a good alignment boundary.
121 */
122
123struct rpcrdma_regbuf {
Chuck Lever9128c3e2015-01-21 11:04:00 -0500124 struct ib_sge rg_iov;
Chuck Lever54cbd6b2016-09-15 10:56:18 -0400125 struct ib_device *rg_device;
Chuck Lever99ef4db2016-09-15 10:56:10 -0400126 enum dma_data_direction rg_direction;
Chuck Lever9128c3e2015-01-21 11:04:00 -0500127 __be32 rg_base[0] __attribute__ ((aligned(256)));
128};
129
130static inline u64
131rdmab_addr(struct rpcrdma_regbuf *rb)
132{
133 return rb->rg_iov.addr;
134}
135
136static inline u32
137rdmab_length(struct rpcrdma_regbuf *rb)
138{
139 return rb->rg_iov.length;
140}
141
142static inline u32
143rdmab_lkey(struct rpcrdma_regbuf *rb)
144{
145 return rb->rg_iov.lkey;
146}
147
Chuck Lever91a10c52017-04-11 13:23:02 -0400148static inline struct ib_device *
149rdmab_device(struct rpcrdma_regbuf *rb)
150{
151 return rb->rg_device;
152}
153
Chuck Lever5d252f92016-01-07 14:50:10 -0500154#define RPCRDMA_DEF_GFP (GFP_NOIO | __GFP_NOWARN)
155
Chuck Lever94931742016-05-02 14:40:56 -0400156/* To ensure a transport can always make forward progress,
157 * the number of RDMA segments allowed in header chunk lists
158 * is capped at 8. This prevents less-capable devices and
159 * memory registrations from overrunning the Send buffer
160 * while building chunk lists.
161 *
162 * Elements of the Read list take up more room than the
163 * Write list or Reply chunk. 8 read segments means the Read
164 * list (or Write list or Reply chunk) cannot consume more
165 * than
166 *
167 * ((8 + 2) * read segment size) + 1 XDR words, or 244 bytes.
168 *
169 * And the fixed part of the header is another 24 bytes.
170 *
171 * The smallest inline threshold is 1024 bytes, ensuring that
172 * at least 750 bytes are available for RPC messages.
173 */
Chuck Lever08cf2ef2016-09-15 10:56:02 -0400174enum {
175 RPCRDMA_MAX_HDR_SEGS = 8,
176 RPCRDMA_HDRBUF_SIZE = 256,
177};
Chuck Lever94931742016-05-02 14:40:56 -0400178
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400179/*
Chuck Levere1352c92017-10-16 15:01:22 -0400180 * struct rpcrdma_rep -- this structure encapsulates state required
181 * to receive and complete an RPC Reply, asychronously. It needs
182 * several pieces of state:
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400183 *
Chuck Levere1352c92017-10-16 15:01:22 -0400184 * o receive buffer and ib_sge (donated to provider)
185 * o status of receive (success or not, length, inv rkey)
186 * o bookkeeping state to get run by reply handler (XDR stream)
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400187 *
Chuck Levere1352c92017-10-16 15:01:22 -0400188 * These structures are allocated during transport initialization.
189 * N of these are associated with a transport instance, managed by
190 * struct rpcrdma_buffer. N is the max number of outstanding RPCs.
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400191 */
192
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400193struct rpcrdma_rep {
Chuck Lever552bf222016-03-04 11:28:36 -0500194 struct ib_cqe rr_cqe;
Chuck Lever5381e0e2017-10-16 15:01:14 -0400195 __be32 rr_xid;
196 __be32 rr_vers;
197 __be32 rr_proc;
Chuck Leverc8b920b2016-09-15 10:57:16 -0400198 int rr_wc_flags;
199 u32 rr_inv_rkey;
Chuck Leverc1bcb682017-08-03 14:30:52 -0400200 struct rpcrdma_regbuf *rr_rdmabuf;
Chuck Leverfed171b2015-05-26 11:51:37 -0400201 struct rpcrdma_xprt *rr_rxprt;
Chuck Leverfe97b472015-10-24 17:27:10 -0400202 struct work_struct rr_work;
Chuck Lever96f87782017-08-03 14:30:03 -0400203 struct xdr_buf rr_hdrbuf;
204 struct xdr_stream rr_stream;
Chuck Levere1352c92017-10-16 15:01:22 -0400205 struct rpc_rqst *rr_rqst;
Chuck Lever6b1184c2015-01-21 11:04:25 -0500206 struct list_head rr_list;
Chuck Lever6ea8e712016-09-15 10:56:51 -0400207 struct ib_recv_wr rr_recv_wr;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400208};
209
Chuck Leverae729502017-10-20 10:48:12 -0400210/* struct rpcrdma_sendctx - DMA mapped SGEs to unmap after Send completes
211 */
Chuck Lever01bb35c2017-10-20 10:48:36 -0400212struct rpcrdma_req;
Chuck Leverae729502017-10-20 10:48:12 -0400213struct rpcrdma_xprt;
214struct rpcrdma_sendctx {
215 struct ib_send_wr sc_wr;
216 struct ib_cqe sc_cqe;
217 struct rpcrdma_xprt *sc_xprt;
Chuck Lever01bb35c2017-10-20 10:48:36 -0400218 struct rpcrdma_req *sc_req;
Chuck Leverae729502017-10-20 10:48:12 -0400219 unsigned int sc_unmap_count;
220 struct ib_sge sc_sges[];
221};
222
223/* Limit the number of SGEs that can be unmapped during one
224 * Send completion. This caps the amount of work a single
225 * completion can do before returning to the provider.
226 *
227 * Setting this to zero disables Send completion batching.
228 */
229enum {
230 RPCRDMA_MAX_SEND_BATCH = 7,
231};
232
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400233/*
Chuck Lever0dbb4102014-07-29 17:24:09 -0400234 * struct rpcrdma_mw - external memory region metadata
235 *
236 * An external memory region is any buffer or page that is registered
237 * on the fly (ie, not pre-registered).
238 *
Chuck Lever3111d722014-07-29 17:24:28 -0400239 * Each rpcrdma_buffer has a list of free MWs anchored in rb_mws. During
Chuck Lever0dbb4102014-07-29 17:24:09 -0400240 * call_allocate, rpcrdma_buffer_get() assigns one to each segment in
241 * an rpcrdma_req. Then rpcrdma_register_external() grabs these to keep
242 * track of registration metadata while each RPC is pending.
243 * rpcrdma_deregister_external() uses this metadata to unmap and
244 * release these resources when an RPC is complete.
245 */
246enum rpcrdma_frmr_state {
247 FRMR_IS_INVALID, /* ready to be used */
248 FRMR_IS_VALID, /* in use */
Chuck Lever62bdf942016-11-07 16:16:24 -0500249 FRMR_FLUSHED_FR, /* flushed FASTREG WR */
250 FRMR_FLUSHED_LI, /* flushed LOCALINV WR */
Chuck Lever0dbb4102014-07-29 17:24:09 -0400251};
252
253struct rpcrdma_frmr {
Chuck Lever0dbb4102014-07-29 17:24:09 -0400254 struct ib_mr *fr_mr;
Chuck Lever2fa8f882016-03-04 11:28:53 -0500255 struct ib_cqe fr_cqe;
Chuck Lever0dbb4102014-07-29 17:24:09 -0400256 enum rpcrdma_frmr_state fr_state;
Chuck Lever2fa8f882016-03-04 11:28:53 -0500257 struct completion fr_linv_done;
Chuck Lever3cf4e162015-12-16 17:22:31 -0500258 union {
259 struct ib_reg_wr fr_regwr;
260 struct ib_send_wr fr_invwr;
261 };
Chuck Lever0dbb4102014-07-29 17:24:09 -0400262};
263
Chuck Leveracb9da72015-05-26 11:53:23 -0400264struct rpcrdma_fmr {
Chuck Lever88975eb2016-06-29 13:52:37 -0400265 struct ib_fmr *fm_mr;
266 u64 *fm_physaddrs;
Chuck Lever0dbb4102014-07-29 17:24:09 -0400267};
268
269struct rpcrdma_mw {
Chuck Lever564471d2016-06-29 13:52:21 -0400270 struct list_head mw_list;
271 struct scatterlist *mw_sg;
272 int mw_nents;
273 enum dma_data_direction mw_dir;
Chuck Lever4b196dc62017-06-08 11:51:56 -0400274 unsigned long mw_flags;
Chuck Lever0dbb4102014-07-29 17:24:09 -0400275 union {
Chuck Leveracb9da72015-05-26 11:53:23 -0400276 struct rpcrdma_fmr fmr;
Chuck Lever0dbb4102014-07-29 17:24:09 -0400277 struct rpcrdma_frmr frmr;
Chuck Leverc882a652016-03-04 11:28:45 -0500278 };
Chuck Lever766656b2016-05-02 14:42:29 -0400279 struct rpcrdma_xprt *mw_xprt;
Chuck Lever9d6b0402016-06-29 13:54:16 -0400280 u32 mw_handle;
281 u32 mw_length;
282 u64 mw_offset;
Chuck Lever3111d722014-07-29 17:24:28 -0400283 struct list_head mw_all;
Chuck Lever0dbb4102014-07-29 17:24:09 -0400284};
285
Chuck Lever4b196dc62017-06-08 11:51:56 -0400286/* mw_flags */
287enum {
288 RPCRDMA_MW_F_RI = 1,
289};
290
Chuck Lever0dbb4102014-07-29 17:24:09 -0400291/*
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400292 * struct rpcrdma_req -- structure central to the request/reply sequence.
293 *
294 * N of these are associated with a transport instance, and stored in
295 * struct rpcrdma_buffer. N is the max number of outstanding requests.
296 *
297 * It includes pre-registered buffer memory for send AND recv.
298 * The recv buffer, however, is not owned by this structure, and
299 * is "donated" to the hardware when a recv is posted. When a
300 * reply is handled, the recv buffer used is given back to the
301 * struct rpcrdma_req associated with the request.
302 *
303 * In addition to the basic memory, this structure includes an array
304 * of iovs for send operations. The reason is that the iovs passed to
305 * ib_post_{send,recv} must not be modified until the work request
306 * completes.
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400307 */
308
Chuck Lever5ab81422016-06-29 13:54:25 -0400309/* Maximum number of page-sized "segments" per chunk list to be
310 * registered or invalidated. Must handle a Reply chunk:
311 */
312enum {
313 RPCRDMA_MAX_IOV_SEGS = 3,
314 RPCRDMA_MAX_DATA_SEGS = ((1 * 1024 * 1024) / PAGE_SIZE) + 1,
315 RPCRDMA_MAX_SEGS = RPCRDMA_MAX_DATA_SEGS +
316 RPCRDMA_MAX_IOV_SEGS,
317};
318
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400319struct rpcrdma_mr_seg { /* chunk descriptors */
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400320 u32 mr_len; /* length of chunk or segment */
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400321 struct page *mr_page; /* owning page, if any */
322 char *mr_offset; /* kva if no page, else offset */
323};
324
Chuck Leverc6f5b472017-02-08 17:00:18 -0500325/* The Send SGE array is provisioned to send a maximum size
326 * inline request:
Chuck Lever655fec62016-09-15 10:57:24 -0400327 * - RPC-over-RDMA header
328 * - xdr_buf head iovec
Chuck Leverc6f5b472017-02-08 17:00:18 -0500329 * - RPCRDMA_MAX_INLINE bytes, in pages
Chuck Lever655fec62016-09-15 10:57:24 -0400330 * - xdr_buf tail iovec
Chuck Leverc6f5b472017-02-08 17:00:18 -0500331 *
332 * The actual number of array elements consumed by each RPC
333 * depends on the device's max_sge limit.
Chuck Lever655fec62016-09-15 10:57:24 -0400334 */
335enum {
Chuck Lever16f906d2017-02-08 17:00:10 -0500336 RPCRDMA_MIN_SEND_SGES = 3,
Chuck Leverc6f5b472017-02-08 17:00:18 -0500337 RPCRDMA_MAX_PAGE_SGES = RPCRDMA_MAX_INLINE >> PAGE_SHIFT,
Chuck Lever655fec62016-09-15 10:57:24 -0400338 RPCRDMA_MAX_SEND_SGES = 1 + 1 + RPCRDMA_MAX_PAGE_SGES + 1,
339};
Chuck Leverb3221d62015-08-03 13:03:39 -0400340
Chuck Lever5ab81422016-06-29 13:54:25 -0400341struct rpcrdma_buffer;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400342struct rpcrdma_req {
Chuck Levera80d66c2017-06-08 11:52:12 -0400343 struct list_head rl_list;
Chuck Leverb3221d62015-08-03 13:03:39 -0400344 unsigned int rl_connect_cookie;
345 struct rpcrdma_buffer *rl_buffer;
Chuck Lever90aab602016-09-15 10:56:43 -0400346 struct rpcrdma_rep *rl_reply;
Chuck Lever7a80f3f2017-08-10 12:47:28 -0400347 struct xdr_stream rl_stream;
348 struct xdr_buf rl_hdrbuf;
Chuck Leverae729502017-10-20 10:48:12 -0400349 struct rpcrdma_sendctx *rl_sendctx;
Chuck Lever9c40c492016-09-15 10:55:53 -0400350 struct rpcrdma_regbuf *rl_rdmabuf; /* xprt header */
351 struct rpcrdma_regbuf *rl_sendbuf; /* rq_snd_buf */
352 struct rpcrdma_regbuf *rl_recvbuf; /* rq_rcv_buf */
Chuck Leverf531a5d2015-10-24 17:27:43 -0400353
354 struct list_head rl_all;
Chuck Lever531cca02017-10-20 10:48:20 -0400355 unsigned long rl_flags;
Chuck Lever5ab81422016-06-29 13:54:25 -0400356
357 struct list_head rl_registered; /* registered segments */
358 struct rpcrdma_mr_seg rl_segments[RPCRDMA_MAX_SEGS];
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400359};
Chuck Lever0ca77dc2015-01-21 11:04:08 -0500360
Chuck Lever531cca02017-10-20 10:48:20 -0400361/* rl_flags */
362enum {
363 RPCRDMA_REQ_F_BACKCHANNEL = 0,
Chuck Lever0ba6f372017-10-20 10:48:28 -0400364 RPCRDMA_REQ_F_PENDING,
Chuck Lever01bb35c2017-10-20 10:48:36 -0400365 RPCRDMA_REQ_F_TX_RESOURCES,
Chuck Lever531cca02017-10-20 10:48:20 -0400366};
367
Chuck Lever5a6d1db2016-09-15 10:55:45 -0400368static inline void
369rpcrdma_set_xprtdata(struct rpc_rqst *rqst, struct rpcrdma_req *req)
370{
371 rqst->rq_xprtdata = req;
372}
373
Chuck Lever0ca77dc2015-01-21 11:04:08 -0500374static inline struct rpcrdma_req *
375rpcr_to_rdmar(struct rpc_rqst *rqst)
376{
Chuck Lever5a6d1db2016-09-15 10:55:45 -0400377 return rqst->rq_xprtdata;
Chuck Lever0ca77dc2015-01-21 11:04:08 -0500378}
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400379
Chuck Lever9a5c63e2017-02-08 17:00:43 -0500380static inline void
381rpcrdma_push_mw(struct rpcrdma_mw *mw, struct list_head *list)
382{
383 list_add_tail(&mw->mw_list, list);
384}
385
386static inline struct rpcrdma_mw *
387rpcrdma_pop_mw(struct list_head *list)
388{
389 struct rpcrdma_mw *mw;
390
391 mw = list_first_entry(list, struct rpcrdma_mw, mw_list);
392 list_del(&mw->mw_list);
393 return mw;
394}
395
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400396/*
397 * struct rpcrdma_buffer -- holds list/queue of pre-registered memory for
398 * inline requests/replies, and client/server credits.
399 *
400 * One of these is associated with a transport instance
401 */
402struct rpcrdma_buffer {
Chuck Lever58d1dcf2015-05-26 11:53:13 -0400403 spinlock_t rb_mwlock; /* protect rb_mws list */
404 struct list_head rb_mws;
405 struct list_head rb_all;
Chuck Lever58d1dcf2015-05-26 11:53:13 -0400406
Chuck Leverae729502017-10-20 10:48:12 -0400407 unsigned long rb_sc_head;
408 unsigned long rb_sc_tail;
409 unsigned long rb_sc_last;
410 struct rpcrdma_sendctx **rb_sc_ctxs;
411
Chuck Lever1e465fd2015-10-24 17:27:02 -0400412 spinlock_t rb_lock; /* protect buf lists */
Chuck Lever05c97462016-09-06 11:22:58 -0400413 int rb_send_count, rb_recv_count;
Chuck Lever1e465fd2015-10-24 17:27:02 -0400414 struct list_head rb_send_bufs;
415 struct list_head rb_recv_bufs;
Chuck Lever58d1dcf2015-05-26 11:53:13 -0400416 u32 rb_max_requests;
Chuck Leverbe798f92017-10-16 15:01:39 -0400417 u32 rb_credits; /* most recent credit grant */
Chuck Leverf531a5d2015-10-24 17:27:43 -0400418
419 u32 rb_bc_srv_max_requests;
420 spinlock_t rb_reqslock; /* protect rb_allreqs */
421 struct list_head rb_allreqs;
Chuck Lever5d252f92016-01-07 14:50:10 -0500422
423 u32 rb_bc_max_requests;
Chuck Lever505bbe62016-06-29 13:52:54 -0400424
425 spinlock_t rb_recovery_lock; /* protect rb_stale_mrs */
426 struct list_head rb_stale_mrs;
427 struct delayed_work rb_recovery_worker;
Chuck Levere2ac2362016-06-29 13:54:00 -0400428 struct delayed_work rb_refresh_worker;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400429};
430#define rdmab_to_ia(b) (&container_of((b), struct rpcrdma_xprt, rx_buf)->rx_ia)
431
432/*
433 * Internal structure for transport instance creation. This
434 * exists primarily for modularity.
435 *
436 * This data should be set with mount options
437 */
438struct rpcrdma_create_data_internal {
439 struct sockaddr_storage addr; /* RDMA server address */
440 unsigned int max_requests; /* max requests (slots) in flight */
441 unsigned int rsize; /* mount rsize - max read hdr+data */
442 unsigned int wsize; /* mount wsize - max write hdr+data */
443 unsigned int inline_rsize; /* max non-rdma read data payload */
444 unsigned int inline_wsize; /* max non-rdma write data payload */
445 unsigned int padding; /* non-rdma write header padding */
446};
447
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400448/*
449 * Statistics for RPCRDMA
450 */
451struct rpcrdma_stats {
Chuck Lever67af6f62017-08-22 11:19:35 -0400452 /* accessed when sending a call */
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400453 unsigned long read_chunk_count;
454 unsigned long write_chunk_count;
455 unsigned long reply_chunk_count;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400456 unsigned long long total_rdma_request;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400457
Chuck Lever67af6f62017-08-22 11:19:35 -0400458 /* rarely accessed error counters */
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400459 unsigned long long pullup_copy_count;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400460 unsigned long hardway_register_count;
461 unsigned long failed_marshal_count;
462 unsigned long bad_reply_count;
Chuck Lever505bbe62016-06-29 13:52:54 -0400463 unsigned long mrs_recovered;
464 unsigned long mrs_orphaned;
Chuck Levere2ac2362016-06-29 13:54:00 -0400465 unsigned long mrs_allocated;
Chuck Leverae729502017-10-20 10:48:12 -0400466 unsigned long empty_sendctx_q;
Chuck Lever67af6f62017-08-22 11:19:35 -0400467
468 /* accessed when receiving a reply */
469 unsigned long long total_rdma_reply;
470 unsigned long long fixup_copy_count;
Chuck Lever01bb35c2017-10-20 10:48:36 -0400471 unsigned long reply_waits_for_send;
Chuck Leverc8b920b2016-09-15 10:57:16 -0400472 unsigned long local_inv_needed;
Chuck Lever67af6f62017-08-22 11:19:35 -0400473 unsigned long nomsg_call_count;
474 unsigned long bcall_count;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400475};
476
477/*
Chuck Levera0ce85f2015-03-30 14:34:21 -0400478 * Per-registration mode operations
479 */
Chuck Lever1c9351e2015-03-30 14:34:30 -0400480struct rpcrdma_xprt;
Chuck Levera0ce85f2015-03-30 14:34:21 -0400481struct rpcrdma_memreg_ops {
Chuck Lever6748b0c2017-08-14 15:38:30 -0400482 struct rpcrdma_mr_seg *
483 (*ro_map)(struct rpcrdma_xprt *,
Chuck Lever9d6b0402016-06-29 13:54:16 -0400484 struct rpcrdma_mr_seg *, int, bool,
485 struct rpcrdma_mw **);
Chuck Lever32d0cee2015-12-16 17:22:39 -0500486 void (*ro_unmap_sync)(struct rpcrdma_xprt *,
Chuck Lever451d26e2017-06-08 11:52:04 -0400487 struct list_head *);
Chuck Lever505bbe62016-06-29 13:52:54 -0400488 void (*ro_recover_mr)(struct rpcrdma_mw *);
Chuck Lever3968cb52015-03-30 14:35:26 -0400489 int (*ro_open)(struct rpcrdma_ia *,
490 struct rpcrdma_ep *,
491 struct rpcrdma_create_data_internal *);
Chuck Lever1c9351e2015-03-30 14:34:30 -0400492 size_t (*ro_maxpages)(struct rpcrdma_xprt *);
Chuck Levere2ac2362016-06-29 13:54:00 -0400493 int (*ro_init_mr)(struct rpcrdma_ia *,
494 struct rpcrdma_mw *);
495 void (*ro_release_mr)(struct rpcrdma_mw *);
Chuck Levera0ce85f2015-03-30 14:34:21 -0400496 const char *ro_displayname;
Chuck Leverc8b920b2016-09-15 10:57:16 -0400497 const int ro_send_w_inv_ok;
Chuck Levera0ce85f2015-03-30 14:34:21 -0400498};
499
500extern const struct rpcrdma_memreg_ops rpcrdma_fmr_memreg_ops;
501extern const struct rpcrdma_memreg_ops rpcrdma_frwr_memreg_ops;
Chuck Levera0ce85f2015-03-30 14:34:21 -0400502
503/*
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400504 * RPCRDMA transport -- encapsulates the structures above for
505 * integration with RPC.
506 *
507 * The contained structures are embedded, not pointers,
508 * for convenience. This structure need not be visible externally.
509 *
510 * It is allocated and initialized during mount, and released
511 * during unmount.
512 */
513struct rpcrdma_xprt {
Chuck Lever5abefb82015-01-21 11:02:37 -0500514 struct rpc_xprt rx_xprt;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400515 struct rpcrdma_ia rx_ia;
516 struct rpcrdma_ep rx_ep;
517 struct rpcrdma_buffer rx_buf;
518 struct rpcrdma_create_data_internal rx_data;
Chuck Lever5abefb82015-01-21 11:02:37 -0500519 struct delayed_work rx_connect_worker;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400520 struct rpcrdma_stats rx_stats;
521};
522
Chuck Lever5abefb82015-01-21 11:02:37 -0500523#define rpcx_to_rdmax(x) container_of(x, struct rpcrdma_xprt, rx_xprt)
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400524#define rpcx_to_rdmad(x) (rpcx_to_rdmax(x)->rx_data)
525
Tom Talpey9191ca32008-10-09 15:01:11 -0400526/* Setting this to 0 ensures interoperability with early servers.
527 * Setting this to 1 enhances certain unaligned read/write performance.
528 * Default is 0, see sysctl entry and rpc_rdma.c rpcrdma_convert_iovs() */
529extern int xprt_rdma_pad_optimize;
530
Chuck Leverfff09592017-04-11 13:22:54 -0400531/* This setting controls the hunt for a supported memory
532 * registration strategy.
533 */
534extern unsigned int xprt_rdma_memreg_strategy;
535
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400536/*
537 * Interface Adapter calls - xprtrdma/verbs.c
538 */
Chuck Leverfff09592017-04-11 13:22:54 -0400539int rpcrdma_ia_open(struct rpcrdma_xprt *xprt, struct sockaddr *addr);
Chuck Leverbebd0312017-04-11 13:23:10 -0400540void rpcrdma_ia_remove(struct rpcrdma_ia *ia);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400541void rpcrdma_ia_close(struct rpcrdma_ia *);
Chuck Leverb54054c2016-06-29 13:53:27 -0400542bool frwr_is_supported(struct rpcrdma_ia *);
543bool fmr_is_supported(struct rpcrdma_ia *);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400544
Chuck Leverd8f532d2017-10-16 15:01:30 -0400545extern struct workqueue_struct *rpcrdma_receive_wq;
546
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400547/*
548 * Endpoint calls - xprtrdma/verbs.c
549 */
550int rpcrdma_ep_create(struct rpcrdma_ep *, struct rpcrdma_ia *,
551 struct rpcrdma_create_data_internal *);
Chuck Lever7f1d5412014-05-28 10:33:16 -0400552void rpcrdma_ep_destroy(struct rpcrdma_ep *, struct rpcrdma_ia *);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400553int rpcrdma_ep_connect(struct rpcrdma_ep *, struct rpcrdma_ia *);
Chuck Lever3a72dc72016-11-29 10:53:37 -0500554void rpcrdma_conn_func(struct rpcrdma_ep *ep);
Chuck Lever282191c2014-07-29 17:25:55 -0400555void rpcrdma_ep_disconnect(struct rpcrdma_ep *, struct rpcrdma_ia *);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400556
557int rpcrdma_ep_post(struct rpcrdma_ia *, struct rpcrdma_ep *,
558 struct rpcrdma_req *);
Chuck Leverb1573802016-09-15 10:56:35 -0400559int rpcrdma_ep_post_recv(struct rpcrdma_ia *, struct rpcrdma_rep *);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400560
561/*
562 * Buffer calls - xprtrdma/verbs.c
563 */
Chuck Leverf531a5d2015-10-24 17:27:43 -0400564struct rpcrdma_req *rpcrdma_create_req(struct rpcrdma_xprt *);
565struct rpcrdma_rep *rpcrdma_create_rep(struct rpcrdma_xprt *);
Chuck Lever13650c22016-09-15 10:56:26 -0400566void rpcrdma_destroy_req(struct rpcrdma_req *);
Chuck Leverac920d02015-01-21 11:03:44 -0500567int rpcrdma_buffer_create(struct rpcrdma_xprt *);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400568void rpcrdma_buffer_destroy(struct rpcrdma_buffer *);
Chuck Leverae729502017-10-20 10:48:12 -0400569struct rpcrdma_sendctx *rpcrdma_sendctx_get_locked(struct rpcrdma_buffer *buf);
570void rpcrdma_sendctx_put_locked(struct rpcrdma_sendctx *sc);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400571
Chuck Lever346aa662015-05-26 11:52:06 -0400572struct rpcrdma_mw *rpcrdma_get_mw(struct rpcrdma_xprt *);
573void rpcrdma_put_mw(struct rpcrdma_xprt *, struct rpcrdma_mw *);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400574struct rpcrdma_req *rpcrdma_buffer_get(struct rpcrdma_buffer *);
575void rpcrdma_buffer_put(struct rpcrdma_req *);
576void rpcrdma_recv_buffer_get(struct rpcrdma_req *);
577void rpcrdma_recv_buffer_put(struct rpcrdma_rep *);
578
Chuck Lever505bbe62016-06-29 13:52:54 -0400579void rpcrdma_defer_mr_recovery(struct rpcrdma_mw *);
580
Chuck Lever13650c22016-09-15 10:56:26 -0400581struct rpcrdma_regbuf *rpcrdma_alloc_regbuf(size_t, enum dma_data_direction,
Chuck Lever99ef4db2016-09-15 10:56:10 -0400582 gfp_t);
Chuck Lever54cbd6b2016-09-15 10:56:18 -0400583bool __rpcrdma_dma_map_regbuf(struct rpcrdma_ia *, struct rpcrdma_regbuf *);
Chuck Lever13650c22016-09-15 10:56:26 -0400584void rpcrdma_free_regbuf(struct rpcrdma_regbuf *);
Chuck Lever9128c3e2015-01-21 11:04:00 -0500585
Chuck Lever54cbd6b2016-09-15 10:56:18 -0400586static inline bool
587rpcrdma_regbuf_is_mapped(struct rpcrdma_regbuf *rb)
588{
589 return rb->rg_device != NULL;
590}
591
592static inline bool
593rpcrdma_dma_map_regbuf(struct rpcrdma_ia *ia, struct rpcrdma_regbuf *rb)
594{
595 if (likely(rpcrdma_regbuf_is_mapped(rb)))
596 return true;
597 return __rpcrdma_dma_map_regbuf(ia, rb);
598}
599
Chuck Leverf531a5d2015-10-24 17:27:43 -0400600int rpcrdma_ep_post_extra_recv(struct rpcrdma_xprt *, unsigned int);
Chuck Leverd6547882015-03-30 14:35:44 -0400601
Chuck Leverfe97b472015-10-24 17:27:10 -0400602int rpcrdma_alloc_wq(void);
603void rpcrdma_destroy_wq(void);
604
Chuck Leverd6547882015-03-30 14:35:44 -0400605/*
606 * Wrappers for chunk registration, shared by read/write chunk code.
607 */
608
Chuck Leverd6547882015-03-30 14:35:44 -0400609static inline enum dma_data_direction
610rpcrdma_data_dir(bool writing)
611{
612 return writing ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
613}
614
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400615/*
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400616 * RPC/RDMA protocol calls - xprtrdma/rpc_rdma.c
617 */
Chuck Lever655fec62016-09-15 10:57:24 -0400618
619enum rpcrdma_chunktype {
620 rpcrdma_noch = 0,
621 rpcrdma_readch,
622 rpcrdma_areadch,
623 rpcrdma_writech,
624 rpcrdma_replych
625};
626
Chuck Lever857f9ac2017-10-20 10:47:55 -0400627int rpcrdma_prepare_send_sges(struct rpcrdma_xprt *r_xprt,
628 struct rpcrdma_req *req, u32 hdrlen,
629 struct xdr_buf *xdr,
630 enum rpcrdma_chunktype rtype);
Chuck Leverae729502017-10-20 10:48:12 -0400631void rpcrdma_unmap_sendctx(struct rpcrdma_sendctx *sc);
Chuck Lever09e60642017-08-10 12:47:12 -0400632int rpcrdma_marshal_req(struct rpcrdma_xprt *r_xprt, struct rpc_rqst *rqst);
Chuck Lever87cfb9a2016-09-15 10:57:07 -0400633void rpcrdma_set_max_header_sizes(struct rpcrdma_xprt *);
Chuck Levere1352c92017-10-16 15:01:22 -0400634void rpcrdma_complete_rqst(struct rpcrdma_rep *rep);
Chuck Leverd8f532d2017-10-16 15:01:30 -0400635void rpcrdma_reply_handler(struct rpcrdma_rep *rep);
Chuck Lever0ba6f372017-10-20 10:48:28 -0400636void rpcrdma_release_rqst(struct rpcrdma_xprt *r_xprt,
637 struct rpcrdma_req *req);
Chuck Leverd8f532d2017-10-16 15:01:30 -0400638void rpcrdma_deferred_completion(struct work_struct *work);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400639
Chuck Lever96f87782017-08-03 14:30:03 -0400640static inline void rpcrdma_set_xdrlen(struct xdr_buf *xdr, size_t len)
641{
642 xdr->head[0].iov_len = len;
643 xdr->len = len;
644}
645
Chuck Leverffe1f0d2015-06-04 11:21:42 -0400646/* RPC/RDMA module init - xprtrdma/transport.c
647 */
Chuck Lever5d252f92016-01-07 14:50:10 -0500648extern unsigned int xprt_rdma_max_inline_read;
649void xprt_rdma_format_addresses(struct rpc_xprt *xprt, struct sockaddr *sap);
650void xprt_rdma_free_addresses(struct rpc_xprt *xprt);
Chuck Lever3a72dc72016-11-29 10:53:37 -0500651void rpcrdma_connect_worker(struct work_struct *work);
Chuck Lever5d252f92016-01-07 14:50:10 -0500652void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq);
Chuck Leverffe1f0d2015-06-04 11:21:42 -0400653int xprt_rdma_init(void);
654void xprt_rdma_cleanup(void);
655
Chuck Leverf531a5d2015-10-24 17:27:43 -0400656/* Backchannel calls - xprtrdma/backchannel.c
657 */
658#if defined(CONFIG_SUNRPC_BACKCHANNEL)
659int xprt_rdma_bc_setup(struct rpc_xprt *, unsigned int);
Chuck Lever76566772015-10-24 17:28:32 -0400660int xprt_rdma_bc_up(struct svc_serv *, struct net *);
Chuck Lever6b26cc82016-05-02 14:40:40 -0400661size_t xprt_rdma_bc_maxpayload(struct rpc_xprt *);
Chuck Leverf531a5d2015-10-24 17:27:43 -0400662int rpcrdma_bc_post_recv(struct rpcrdma_xprt *, unsigned int);
Chuck Lever63cae472015-10-24 17:28:08 -0400663void rpcrdma_bc_receive_call(struct rpcrdma_xprt *, struct rpcrdma_rep *);
Chuck Lever83128a62015-10-24 17:27:59 -0400664int rpcrdma_bc_marshal_reply(struct rpc_rqst *);
Chuck Leverf531a5d2015-10-24 17:27:43 -0400665void xprt_rdma_bc_free_rqst(struct rpc_rqst *);
666void xprt_rdma_bc_destroy(struct rpc_xprt *, unsigned int);
667#endif /* CONFIG_SUNRPC_BACKCHANNEL */
668
Chuck Lever5d252f92016-01-07 14:50:10 -0500669extern struct xprt_class xprt_rdma_bc;
Tom Tuckercec56c82012-02-15 11:30:00 -0600670
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400671#endif /* _LINUX_SUNRPC_XPRT_RDMA_H */