blob: 0b8ca5e5c706815f069fad67b6964a8f5d248a8a [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 {
96 atomic_t rep_cqcount;
97 int rep_cqinit;
98 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 Lever8d38de62016-11-29 10:52:16 -0500107static inline void
108rpcrdma_init_cqcount(struct rpcrdma_ep *ep, int count)
109{
110 atomic_set(&ep->rep_cqcount, ep->rep_cqinit - count);
111}
112
113/* To update send queue accounting, provider must take a
114 * send completion every now and then.
115 */
116static inline void
117rpcrdma_set_signaled(struct rpcrdma_ep *ep, struct ib_send_wr *send_wr)
118{
119 send_wr->send_flags = 0;
120 if (unlikely(atomic_sub_return(1, &ep->rep_cqcount) <= 0)) {
121 rpcrdma_init_cqcount(ep, 0);
122 send_wr->send_flags = IB_SEND_SIGNALED;
123 }
124}
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400125
Chuck Lever124fa172015-10-24 17:27:51 -0400126/* Pre-allocate extra Work Requests for handling backward receives
127 * and sends. This is a fixed value because the Work Queues are
128 * allocated when the forward channel is set up.
129 */
130#if defined(CONFIG_SUNRPC_BACKCHANNEL)
131#define RPCRDMA_BACKWARD_WRS (8)
132#else
133#define RPCRDMA_BACKWARD_WRS (0)
134#endif
135
Chuck Lever9128c3e2015-01-21 11:04:00 -0500136/* Registered buffer -- registered kmalloc'd memory for RDMA SEND/RECV
137 *
138 * The below structure appears at the front of a large region of kmalloc'd
139 * memory, which always starts on a good alignment boundary.
140 */
141
142struct rpcrdma_regbuf {
Chuck Lever9128c3e2015-01-21 11:04:00 -0500143 struct ib_sge rg_iov;
Chuck Lever54cbd6b2016-09-15 10:56:18 -0400144 struct ib_device *rg_device;
Chuck Lever99ef4db2016-09-15 10:56:10 -0400145 enum dma_data_direction rg_direction;
Chuck Lever9128c3e2015-01-21 11:04:00 -0500146 __be32 rg_base[0] __attribute__ ((aligned(256)));
147};
148
149static inline u64
150rdmab_addr(struct rpcrdma_regbuf *rb)
151{
152 return rb->rg_iov.addr;
153}
154
155static inline u32
156rdmab_length(struct rpcrdma_regbuf *rb)
157{
158 return rb->rg_iov.length;
159}
160
161static inline u32
162rdmab_lkey(struct rpcrdma_regbuf *rb)
163{
164 return rb->rg_iov.lkey;
165}
166
167static inline struct rpcrdma_msg *
168rdmab_to_msg(struct rpcrdma_regbuf *rb)
169{
170 return (struct rpcrdma_msg *)rb->rg_base;
171}
172
Chuck Lever91a10c52017-04-11 13:23:02 -0400173static inline struct ib_device *
174rdmab_device(struct rpcrdma_regbuf *rb)
175{
176 return rb->rg_device;
177}
178
Chuck Lever5d252f92016-01-07 14:50:10 -0500179#define RPCRDMA_DEF_GFP (GFP_NOIO | __GFP_NOWARN)
180
Chuck Lever94931742016-05-02 14:40:56 -0400181/* To ensure a transport can always make forward progress,
182 * the number of RDMA segments allowed in header chunk lists
183 * is capped at 8. This prevents less-capable devices and
184 * memory registrations from overrunning the Send buffer
185 * while building chunk lists.
186 *
187 * Elements of the Read list take up more room than the
188 * Write list or Reply chunk. 8 read segments means the Read
189 * list (or Write list or Reply chunk) cannot consume more
190 * than
191 *
192 * ((8 + 2) * read segment size) + 1 XDR words, or 244 bytes.
193 *
194 * And the fixed part of the header is another 24 bytes.
195 *
196 * The smallest inline threshold is 1024 bytes, ensuring that
197 * at least 750 bytes are available for RPC messages.
198 */
Chuck Lever08cf2ef2016-09-15 10:56:02 -0400199enum {
200 RPCRDMA_MAX_HDR_SEGS = 8,
201 RPCRDMA_HDRBUF_SIZE = 256,
202};
Chuck Lever94931742016-05-02 14:40:56 -0400203
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400204/*
Chuck Levere1352c92017-10-16 15:01:22 -0400205 * struct rpcrdma_rep -- this structure encapsulates state required
206 * to receive and complete an RPC Reply, asychronously. It needs
207 * several pieces of state:
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400208 *
Chuck Levere1352c92017-10-16 15:01:22 -0400209 * o receive buffer and ib_sge (donated to provider)
210 * o status of receive (success or not, length, inv rkey)
211 * o bookkeeping state to get run by reply handler (XDR stream)
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400212 *
Chuck Levere1352c92017-10-16 15:01:22 -0400213 * These structures are allocated during transport initialization.
214 * N of these are associated with a transport instance, managed by
215 * struct rpcrdma_buffer. N is the max number of outstanding RPCs.
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400216 */
217
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400218struct rpcrdma_rep {
Chuck Lever552bf222016-03-04 11:28:36 -0500219 struct ib_cqe rr_cqe;
Chuck Lever5381e0e2017-10-16 15:01:14 -0400220 __be32 rr_xid;
221 __be32 rr_vers;
222 __be32 rr_proc;
Chuck Leverc8b920b2016-09-15 10:57:16 -0400223 int rr_wc_flags;
224 u32 rr_inv_rkey;
Chuck Leverc1bcb682017-08-03 14:30:52 -0400225 struct rpcrdma_regbuf *rr_rdmabuf;
Chuck Leverfed171b2015-05-26 11:51:37 -0400226 struct rpcrdma_xprt *rr_rxprt;
Chuck Leverfe97b472015-10-24 17:27:10 -0400227 struct work_struct rr_work;
Chuck Lever96f87782017-08-03 14:30:03 -0400228 struct xdr_buf rr_hdrbuf;
229 struct xdr_stream rr_stream;
Chuck Levere1352c92017-10-16 15:01:22 -0400230 struct rpc_rqst *rr_rqst;
Chuck Lever6b1184c2015-01-21 11:04:25 -0500231 struct list_head rr_list;
Chuck Lever6ea8e712016-09-15 10:56:51 -0400232 struct ib_recv_wr rr_recv_wr;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400233};
234
235/*
Chuck Lever0dbb4102014-07-29 17:24:09 -0400236 * struct rpcrdma_mw - external memory region metadata
237 *
238 * An external memory region is any buffer or page that is registered
239 * on the fly (ie, not pre-registered).
240 *
Chuck Lever3111d722014-07-29 17:24:28 -0400241 * Each rpcrdma_buffer has a list of free MWs anchored in rb_mws. During
Chuck Lever0dbb4102014-07-29 17:24:09 -0400242 * call_allocate, rpcrdma_buffer_get() assigns one to each segment in
243 * an rpcrdma_req. Then rpcrdma_register_external() grabs these to keep
244 * track of registration metadata while each RPC is pending.
245 * rpcrdma_deregister_external() uses this metadata to unmap and
246 * release these resources when an RPC is complete.
247 */
248enum rpcrdma_frmr_state {
249 FRMR_IS_INVALID, /* ready to be used */
250 FRMR_IS_VALID, /* in use */
Chuck Lever62bdf942016-11-07 16:16:24 -0500251 FRMR_FLUSHED_FR, /* flushed FASTREG WR */
252 FRMR_FLUSHED_LI, /* flushed LOCALINV WR */
Chuck Lever0dbb4102014-07-29 17:24:09 -0400253};
254
255struct rpcrdma_frmr {
Chuck Lever0dbb4102014-07-29 17:24:09 -0400256 struct ib_mr *fr_mr;
Chuck Lever2fa8f882016-03-04 11:28:53 -0500257 struct ib_cqe fr_cqe;
Chuck Lever0dbb4102014-07-29 17:24:09 -0400258 enum rpcrdma_frmr_state fr_state;
Chuck Lever2fa8f882016-03-04 11:28:53 -0500259 struct completion fr_linv_done;
Chuck Lever3cf4e162015-12-16 17:22:31 -0500260 union {
261 struct ib_reg_wr fr_regwr;
262 struct ib_send_wr fr_invwr;
263 };
Chuck Lever0dbb4102014-07-29 17:24:09 -0400264};
265
Chuck Leveracb9da72015-05-26 11:53:23 -0400266struct rpcrdma_fmr {
Chuck Lever88975eb2016-06-29 13:52:37 -0400267 struct ib_fmr *fm_mr;
268 u64 *fm_physaddrs;
Chuck Lever0dbb4102014-07-29 17:24:09 -0400269};
270
271struct rpcrdma_mw {
Chuck Lever564471d2016-06-29 13:52:21 -0400272 struct list_head mw_list;
273 struct scatterlist *mw_sg;
274 int mw_nents;
275 enum dma_data_direction mw_dir;
Chuck Lever4b196dc62017-06-08 11:51:56 -0400276 unsigned long mw_flags;
Chuck Lever0dbb4102014-07-29 17:24:09 -0400277 union {
Chuck Leveracb9da72015-05-26 11:53:23 -0400278 struct rpcrdma_fmr fmr;
Chuck Lever0dbb4102014-07-29 17:24:09 -0400279 struct rpcrdma_frmr frmr;
Chuck Leverc882a652016-03-04 11:28:45 -0500280 };
Chuck Lever766656b2016-05-02 14:42:29 -0400281 struct rpcrdma_xprt *mw_xprt;
Chuck Lever9d6b0402016-06-29 13:54:16 -0400282 u32 mw_handle;
283 u32 mw_length;
284 u64 mw_offset;
Chuck Lever3111d722014-07-29 17:24:28 -0400285 struct list_head mw_all;
Chuck Lever0dbb4102014-07-29 17:24:09 -0400286};
287
Chuck Lever4b196dc62017-06-08 11:51:56 -0400288/* mw_flags */
289enum {
290 RPCRDMA_MW_F_RI = 1,
291};
292
Chuck Lever0dbb4102014-07-29 17:24:09 -0400293/*
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400294 * struct rpcrdma_req -- structure central to the request/reply sequence.
295 *
296 * N of these are associated with a transport instance, and stored in
297 * struct rpcrdma_buffer. N is the max number of outstanding requests.
298 *
299 * It includes pre-registered buffer memory for send AND recv.
300 * The recv buffer, however, is not owned by this structure, and
301 * is "donated" to the hardware when a recv is posted. When a
302 * reply is handled, the recv buffer used is given back to the
303 * struct rpcrdma_req associated with the request.
304 *
305 * In addition to the basic memory, this structure includes an array
306 * of iovs for send operations. The reason is that the iovs passed to
307 * ib_post_{send,recv} must not be modified until the work request
308 * completes.
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400309 */
310
Chuck Lever5ab81422016-06-29 13:54:25 -0400311/* Maximum number of page-sized "segments" per chunk list to be
312 * registered or invalidated. Must handle a Reply chunk:
313 */
314enum {
315 RPCRDMA_MAX_IOV_SEGS = 3,
316 RPCRDMA_MAX_DATA_SEGS = ((1 * 1024 * 1024) / PAGE_SIZE) + 1,
317 RPCRDMA_MAX_SEGS = RPCRDMA_MAX_DATA_SEGS +
318 RPCRDMA_MAX_IOV_SEGS,
319};
320
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400321struct rpcrdma_mr_seg { /* chunk descriptors */
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400322 u32 mr_len; /* length of chunk or segment */
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400323 struct page *mr_page; /* owning page, if any */
324 char *mr_offset; /* kva if no page, else offset */
325};
326
Chuck Leverc6f5b472017-02-08 17:00:18 -0500327/* The Send SGE array is provisioned to send a maximum size
328 * inline request:
Chuck Lever655fec62016-09-15 10:57:24 -0400329 * - RPC-over-RDMA header
330 * - xdr_buf head iovec
Chuck Leverc6f5b472017-02-08 17:00:18 -0500331 * - RPCRDMA_MAX_INLINE bytes, in pages
Chuck Lever655fec62016-09-15 10:57:24 -0400332 * - xdr_buf tail iovec
Chuck Leverc6f5b472017-02-08 17:00:18 -0500333 *
334 * The actual number of array elements consumed by each RPC
335 * depends on the device's max_sge limit.
Chuck Lever655fec62016-09-15 10:57:24 -0400336 */
337enum {
Chuck Lever16f906d2017-02-08 17:00:10 -0500338 RPCRDMA_MIN_SEND_SGES = 3,
Chuck Leverc6f5b472017-02-08 17:00:18 -0500339 RPCRDMA_MAX_PAGE_SGES = RPCRDMA_MAX_INLINE >> PAGE_SHIFT,
Chuck Lever655fec62016-09-15 10:57:24 -0400340 RPCRDMA_MAX_SEND_SGES = 1 + 1 + RPCRDMA_MAX_PAGE_SGES + 1,
341};
Chuck Leverb3221d62015-08-03 13:03:39 -0400342
Chuck Lever5ab81422016-06-29 13:54:25 -0400343struct rpcrdma_buffer;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400344struct rpcrdma_req {
Chuck Levera80d66c2017-06-08 11:52:12 -0400345 struct list_head rl_list;
Chuck Lever655fec62016-09-15 10:57:24 -0400346 unsigned int rl_mapped_sges;
Chuck Leverb3221d62015-08-03 13:03:39 -0400347 unsigned int rl_connect_cookie;
348 struct rpcrdma_buffer *rl_buffer;
Chuck Lever90aab602016-09-15 10:56:43 -0400349 struct rpcrdma_rep *rl_reply;
Chuck Lever7a80f3f2017-08-10 12:47:28 -0400350 struct xdr_stream rl_stream;
351 struct xdr_buf rl_hdrbuf;
Chuck Lever90aab602016-09-15 10:56:43 -0400352 struct ib_send_wr rl_send_wr;
Chuck Lever655fec62016-09-15 10:57:24 -0400353 struct ib_sge rl_send_sge[RPCRDMA_MAX_SEND_SGES];
Chuck Lever9c40c492016-09-15 10:55:53 -0400354 struct rpcrdma_regbuf *rl_rdmabuf; /* xprt header */
355 struct rpcrdma_regbuf *rl_sendbuf; /* rq_snd_buf */
356 struct rpcrdma_regbuf *rl_recvbuf; /* rq_rcv_buf */
Chuck Leverf531a5d2015-10-24 17:27:43 -0400357
Chuck Lever2fa8f882016-03-04 11:28:53 -0500358 struct ib_cqe rl_cqe;
Chuck Leverf531a5d2015-10-24 17:27:43 -0400359 struct list_head rl_all;
360 bool rl_backchannel;
Chuck Lever5ab81422016-06-29 13:54:25 -0400361
362 struct list_head rl_registered; /* registered segments */
363 struct rpcrdma_mr_seg rl_segments[RPCRDMA_MAX_SEGS];
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400364};
Chuck Lever0ca77dc2015-01-21 11:04:08 -0500365
Chuck Lever5a6d1db2016-09-15 10:55:45 -0400366static inline void
367rpcrdma_set_xprtdata(struct rpc_rqst *rqst, struct rpcrdma_req *req)
368{
369 rqst->rq_xprtdata = req;
370}
371
Chuck Lever0ca77dc2015-01-21 11:04:08 -0500372static inline struct rpcrdma_req *
373rpcr_to_rdmar(struct rpc_rqst *rqst)
374{
Chuck Lever5a6d1db2016-09-15 10:55:45 -0400375 return rqst->rq_xprtdata;
Chuck Lever0ca77dc2015-01-21 11:04:08 -0500376}
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400377
Chuck Lever9a5c63e2017-02-08 17:00:43 -0500378static inline void
379rpcrdma_push_mw(struct rpcrdma_mw *mw, struct list_head *list)
380{
381 list_add_tail(&mw->mw_list, list);
382}
383
384static inline struct rpcrdma_mw *
385rpcrdma_pop_mw(struct list_head *list)
386{
387 struct rpcrdma_mw *mw;
388
389 mw = list_first_entry(list, struct rpcrdma_mw, mw_list);
390 list_del(&mw->mw_list);
391 return mw;
392}
393
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400394/*
395 * struct rpcrdma_buffer -- holds list/queue of pre-registered memory for
396 * inline requests/replies, and client/server credits.
397 *
398 * One of these is associated with a transport instance
399 */
400struct rpcrdma_buffer {
Chuck Lever58d1dcf2015-05-26 11:53:13 -0400401 spinlock_t rb_mwlock; /* protect rb_mws list */
402 struct list_head rb_mws;
403 struct list_head rb_all;
Chuck Lever58d1dcf2015-05-26 11:53:13 -0400404
Chuck Lever1e465fd2015-10-24 17:27:02 -0400405 spinlock_t rb_lock; /* protect buf lists */
Chuck Lever05c97462016-09-06 11:22:58 -0400406 int rb_send_count, rb_recv_count;
Chuck Lever1e465fd2015-10-24 17:27:02 -0400407 struct list_head rb_send_bufs;
408 struct list_head rb_recv_bufs;
Chuck Lever58d1dcf2015-05-26 11:53:13 -0400409 u32 rb_max_requests;
Chuck Leverbe798f92017-10-16 15:01:39 -0400410 u32 rb_credits; /* most recent credit grant */
Chuck Leverf531a5d2015-10-24 17:27:43 -0400411
412 u32 rb_bc_srv_max_requests;
413 spinlock_t rb_reqslock; /* protect rb_allreqs */
414 struct list_head rb_allreqs;
Chuck Lever5d252f92016-01-07 14:50:10 -0500415
416 u32 rb_bc_max_requests;
Chuck Lever505bbe62016-06-29 13:52:54 -0400417
418 spinlock_t rb_recovery_lock; /* protect rb_stale_mrs */
419 struct list_head rb_stale_mrs;
420 struct delayed_work rb_recovery_worker;
Chuck Levere2ac2362016-06-29 13:54:00 -0400421 struct delayed_work rb_refresh_worker;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400422};
423#define rdmab_to_ia(b) (&container_of((b), struct rpcrdma_xprt, rx_buf)->rx_ia)
424
425/*
426 * Internal structure for transport instance creation. This
427 * exists primarily for modularity.
428 *
429 * This data should be set with mount options
430 */
431struct rpcrdma_create_data_internal {
432 struct sockaddr_storage addr; /* RDMA server address */
433 unsigned int max_requests; /* max requests (slots) in flight */
434 unsigned int rsize; /* mount rsize - max read hdr+data */
435 unsigned int wsize; /* mount wsize - max write hdr+data */
436 unsigned int inline_rsize; /* max non-rdma read data payload */
437 unsigned int inline_wsize; /* max non-rdma write data payload */
438 unsigned int padding; /* non-rdma write header padding */
439};
440
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400441/*
442 * Statistics for RPCRDMA
443 */
444struct rpcrdma_stats {
Chuck Lever67af6f62017-08-22 11:19:35 -0400445 /* accessed when sending a call */
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400446 unsigned long read_chunk_count;
447 unsigned long write_chunk_count;
448 unsigned long reply_chunk_count;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400449 unsigned long long total_rdma_request;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400450
Chuck Lever67af6f62017-08-22 11:19:35 -0400451 /* rarely accessed error counters */
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400452 unsigned long long pullup_copy_count;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400453 unsigned long hardway_register_count;
454 unsigned long failed_marshal_count;
455 unsigned long bad_reply_count;
Chuck Lever505bbe62016-06-29 13:52:54 -0400456 unsigned long mrs_recovered;
457 unsigned long mrs_orphaned;
Chuck Levere2ac2362016-06-29 13:54:00 -0400458 unsigned long mrs_allocated;
Chuck Lever67af6f62017-08-22 11:19:35 -0400459
460 /* accessed when receiving a reply */
461 unsigned long long total_rdma_reply;
462 unsigned long long fixup_copy_count;
Chuck Leverc8b920b2016-09-15 10:57:16 -0400463 unsigned long local_inv_needed;
Chuck Lever67af6f62017-08-22 11:19:35 -0400464 unsigned long nomsg_call_count;
465 unsigned long bcall_count;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400466};
467
468/*
Chuck Levera0ce85f2015-03-30 14:34:21 -0400469 * Per-registration mode operations
470 */
Chuck Lever1c9351e2015-03-30 14:34:30 -0400471struct rpcrdma_xprt;
Chuck Levera0ce85f2015-03-30 14:34:21 -0400472struct rpcrdma_memreg_ops {
Chuck Lever6748b0c2017-08-14 15:38:30 -0400473 struct rpcrdma_mr_seg *
474 (*ro_map)(struct rpcrdma_xprt *,
Chuck Lever9d6b0402016-06-29 13:54:16 -0400475 struct rpcrdma_mr_seg *, int, bool,
476 struct rpcrdma_mw **);
Chuck Lever32d0cee2015-12-16 17:22:39 -0500477 void (*ro_unmap_sync)(struct rpcrdma_xprt *,
Chuck Lever451d26e2017-06-08 11:52:04 -0400478 struct list_head *);
Chuck Lever505bbe62016-06-29 13:52:54 -0400479 void (*ro_recover_mr)(struct rpcrdma_mw *);
Chuck Lever3968cb52015-03-30 14:35:26 -0400480 int (*ro_open)(struct rpcrdma_ia *,
481 struct rpcrdma_ep *,
482 struct rpcrdma_create_data_internal *);
Chuck Lever1c9351e2015-03-30 14:34:30 -0400483 size_t (*ro_maxpages)(struct rpcrdma_xprt *);
Chuck Levere2ac2362016-06-29 13:54:00 -0400484 int (*ro_init_mr)(struct rpcrdma_ia *,
485 struct rpcrdma_mw *);
486 void (*ro_release_mr)(struct rpcrdma_mw *);
Chuck Levera0ce85f2015-03-30 14:34:21 -0400487 const char *ro_displayname;
Chuck Leverc8b920b2016-09-15 10:57:16 -0400488 const int ro_send_w_inv_ok;
Chuck Levera0ce85f2015-03-30 14:34:21 -0400489};
490
491extern const struct rpcrdma_memreg_ops rpcrdma_fmr_memreg_ops;
492extern const struct rpcrdma_memreg_ops rpcrdma_frwr_memreg_ops;
Chuck Levera0ce85f2015-03-30 14:34:21 -0400493
494/*
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400495 * RPCRDMA transport -- encapsulates the structures above for
496 * integration with RPC.
497 *
498 * The contained structures are embedded, not pointers,
499 * for convenience. This structure need not be visible externally.
500 *
501 * It is allocated and initialized during mount, and released
502 * during unmount.
503 */
504struct rpcrdma_xprt {
Chuck Lever5abefb82015-01-21 11:02:37 -0500505 struct rpc_xprt rx_xprt;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400506 struct rpcrdma_ia rx_ia;
507 struct rpcrdma_ep rx_ep;
508 struct rpcrdma_buffer rx_buf;
509 struct rpcrdma_create_data_internal rx_data;
Chuck Lever5abefb82015-01-21 11:02:37 -0500510 struct delayed_work rx_connect_worker;
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400511 struct rpcrdma_stats rx_stats;
512};
513
Chuck Lever5abefb82015-01-21 11:02:37 -0500514#define rpcx_to_rdmax(x) container_of(x, struct rpcrdma_xprt, rx_xprt)
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400515#define rpcx_to_rdmad(x) (rpcx_to_rdmax(x)->rx_data)
516
Tom Talpey9191ca32008-10-09 15:01:11 -0400517/* Setting this to 0 ensures interoperability with early servers.
518 * Setting this to 1 enhances certain unaligned read/write performance.
519 * Default is 0, see sysctl entry and rpc_rdma.c rpcrdma_convert_iovs() */
520extern int xprt_rdma_pad_optimize;
521
Chuck Leverfff09592017-04-11 13:22:54 -0400522/* This setting controls the hunt for a supported memory
523 * registration strategy.
524 */
525extern unsigned int xprt_rdma_memreg_strategy;
526
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400527/*
528 * Interface Adapter calls - xprtrdma/verbs.c
529 */
Chuck Leverfff09592017-04-11 13:22:54 -0400530int rpcrdma_ia_open(struct rpcrdma_xprt *xprt, struct sockaddr *addr);
Chuck Leverbebd0312017-04-11 13:23:10 -0400531void rpcrdma_ia_remove(struct rpcrdma_ia *ia);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400532void rpcrdma_ia_close(struct rpcrdma_ia *);
Chuck Leverb54054c2016-06-29 13:53:27 -0400533bool frwr_is_supported(struct rpcrdma_ia *);
534bool fmr_is_supported(struct rpcrdma_ia *);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400535
Chuck Leverd8f532d2017-10-16 15:01:30 -0400536extern struct workqueue_struct *rpcrdma_receive_wq;
537
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400538/*
539 * Endpoint calls - xprtrdma/verbs.c
540 */
541int rpcrdma_ep_create(struct rpcrdma_ep *, struct rpcrdma_ia *,
542 struct rpcrdma_create_data_internal *);
Chuck Lever7f1d5412014-05-28 10:33:16 -0400543void rpcrdma_ep_destroy(struct rpcrdma_ep *, struct rpcrdma_ia *);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400544int rpcrdma_ep_connect(struct rpcrdma_ep *, struct rpcrdma_ia *);
Chuck Lever3a72dc72016-11-29 10:53:37 -0500545void rpcrdma_conn_func(struct rpcrdma_ep *ep);
Chuck Lever282191c2014-07-29 17:25:55 -0400546void rpcrdma_ep_disconnect(struct rpcrdma_ep *, struct rpcrdma_ia *);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400547
548int rpcrdma_ep_post(struct rpcrdma_ia *, struct rpcrdma_ep *,
549 struct rpcrdma_req *);
Chuck Leverb1573802016-09-15 10:56:35 -0400550int rpcrdma_ep_post_recv(struct rpcrdma_ia *, struct rpcrdma_rep *);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400551
552/*
553 * Buffer calls - xprtrdma/verbs.c
554 */
Chuck Leverf531a5d2015-10-24 17:27:43 -0400555struct rpcrdma_req *rpcrdma_create_req(struct rpcrdma_xprt *);
556struct rpcrdma_rep *rpcrdma_create_rep(struct rpcrdma_xprt *);
Chuck Lever13650c22016-09-15 10:56:26 -0400557void rpcrdma_destroy_req(struct rpcrdma_req *);
Chuck Leverac920d02015-01-21 11:03:44 -0500558int rpcrdma_buffer_create(struct rpcrdma_xprt *);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400559void rpcrdma_buffer_destroy(struct rpcrdma_buffer *);
560
Chuck Lever346aa662015-05-26 11:52:06 -0400561struct rpcrdma_mw *rpcrdma_get_mw(struct rpcrdma_xprt *);
562void rpcrdma_put_mw(struct rpcrdma_xprt *, struct rpcrdma_mw *);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400563struct rpcrdma_req *rpcrdma_buffer_get(struct rpcrdma_buffer *);
564void rpcrdma_buffer_put(struct rpcrdma_req *);
565void rpcrdma_recv_buffer_get(struct rpcrdma_req *);
566void rpcrdma_recv_buffer_put(struct rpcrdma_rep *);
567
Chuck Lever505bbe62016-06-29 13:52:54 -0400568void rpcrdma_defer_mr_recovery(struct rpcrdma_mw *);
569
Chuck Lever13650c22016-09-15 10:56:26 -0400570struct rpcrdma_regbuf *rpcrdma_alloc_regbuf(size_t, enum dma_data_direction,
Chuck Lever99ef4db2016-09-15 10:56:10 -0400571 gfp_t);
Chuck Lever54cbd6b2016-09-15 10:56:18 -0400572bool __rpcrdma_dma_map_regbuf(struct rpcrdma_ia *, struct rpcrdma_regbuf *);
Chuck Lever13650c22016-09-15 10:56:26 -0400573void rpcrdma_free_regbuf(struct rpcrdma_regbuf *);
Chuck Lever9128c3e2015-01-21 11:04:00 -0500574
Chuck Lever54cbd6b2016-09-15 10:56:18 -0400575static inline bool
576rpcrdma_regbuf_is_mapped(struct rpcrdma_regbuf *rb)
577{
578 return rb->rg_device != NULL;
579}
580
581static inline bool
582rpcrdma_dma_map_regbuf(struct rpcrdma_ia *ia, struct rpcrdma_regbuf *rb)
583{
584 if (likely(rpcrdma_regbuf_is_mapped(rb)))
585 return true;
586 return __rpcrdma_dma_map_regbuf(ia, rb);
587}
588
Chuck Leverf531a5d2015-10-24 17:27:43 -0400589int rpcrdma_ep_post_extra_recv(struct rpcrdma_xprt *, unsigned int);
Chuck Leverd6547882015-03-30 14:35:44 -0400590
Chuck Leverfe97b472015-10-24 17:27:10 -0400591int rpcrdma_alloc_wq(void);
592void rpcrdma_destroy_wq(void);
593
Chuck Leverd6547882015-03-30 14:35:44 -0400594/*
595 * Wrappers for chunk registration, shared by read/write chunk code.
596 */
597
Chuck Leverd6547882015-03-30 14:35:44 -0400598static inline enum dma_data_direction
599rpcrdma_data_dir(bool writing)
600{
601 return writing ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
602}
603
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400604/*
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400605 * RPC/RDMA protocol calls - xprtrdma/rpc_rdma.c
606 */
Chuck Lever655fec62016-09-15 10:57:24 -0400607
608enum rpcrdma_chunktype {
609 rpcrdma_noch = 0,
610 rpcrdma_readch,
611 rpcrdma_areadch,
612 rpcrdma_writech,
613 rpcrdma_replych
614};
615
Chuck Lever857f9ac2017-10-20 10:47:55 -0400616int rpcrdma_prepare_send_sges(struct rpcrdma_xprt *r_xprt,
617 struct rpcrdma_req *req, u32 hdrlen,
618 struct xdr_buf *xdr,
619 enum rpcrdma_chunktype rtype);
Chuck Lever655fec62016-09-15 10:57:24 -0400620void rpcrdma_unmap_sges(struct rpcrdma_ia *, struct rpcrdma_req *);
Chuck Lever09e60642017-08-10 12:47:12 -0400621int rpcrdma_marshal_req(struct rpcrdma_xprt *r_xprt, struct rpc_rqst *rqst);
Chuck Lever87cfb9a2016-09-15 10:57:07 -0400622void rpcrdma_set_max_header_sizes(struct rpcrdma_xprt *);
Chuck Levere1352c92017-10-16 15:01:22 -0400623void rpcrdma_complete_rqst(struct rpcrdma_rep *rep);
Chuck Leverd8f532d2017-10-16 15:01:30 -0400624void rpcrdma_reply_handler(struct rpcrdma_rep *rep);
625void rpcrdma_deferred_completion(struct work_struct *work);
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400626
Chuck Lever96f87782017-08-03 14:30:03 -0400627static inline void rpcrdma_set_xdrlen(struct xdr_buf *xdr, size_t len)
628{
629 xdr->head[0].iov_len = len;
630 xdr->len = len;
631}
632
Chuck Leverffe1f0d2015-06-04 11:21:42 -0400633/* RPC/RDMA module init - xprtrdma/transport.c
634 */
Chuck Lever5d252f92016-01-07 14:50:10 -0500635extern unsigned int xprt_rdma_max_inline_read;
636void xprt_rdma_format_addresses(struct rpc_xprt *xprt, struct sockaddr *sap);
637void xprt_rdma_free_addresses(struct rpc_xprt *xprt);
Chuck Lever3a72dc72016-11-29 10:53:37 -0500638void rpcrdma_connect_worker(struct work_struct *work);
Chuck Lever5d252f92016-01-07 14:50:10 -0500639void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq);
Chuck Leverffe1f0d2015-06-04 11:21:42 -0400640int xprt_rdma_init(void);
641void xprt_rdma_cleanup(void);
642
Chuck Leverf531a5d2015-10-24 17:27:43 -0400643/* Backchannel calls - xprtrdma/backchannel.c
644 */
645#if defined(CONFIG_SUNRPC_BACKCHANNEL)
646int xprt_rdma_bc_setup(struct rpc_xprt *, unsigned int);
Chuck Lever76566772015-10-24 17:28:32 -0400647int xprt_rdma_bc_up(struct svc_serv *, struct net *);
Chuck Lever6b26cc82016-05-02 14:40:40 -0400648size_t xprt_rdma_bc_maxpayload(struct rpc_xprt *);
Chuck Leverf531a5d2015-10-24 17:27:43 -0400649int rpcrdma_bc_post_recv(struct rpcrdma_xprt *, unsigned int);
Chuck Lever63cae472015-10-24 17:28:08 -0400650void rpcrdma_bc_receive_call(struct rpcrdma_xprt *, struct rpcrdma_rep *);
Chuck Lever83128a62015-10-24 17:27:59 -0400651int rpcrdma_bc_marshal_reply(struct rpc_rqst *);
Chuck Leverf531a5d2015-10-24 17:27:43 -0400652void xprt_rdma_bc_free_rqst(struct rpc_rqst *);
653void xprt_rdma_bc_destroy(struct rpc_xprt *, unsigned int);
654#endif /* CONFIG_SUNRPC_BACKCHANNEL */
655
Chuck Lever5d252f92016-01-07 14:50:10 -0500656extern struct xprt_class xprt_rdma_bc;
Tom Tuckercec56c82012-02-15 11:30:00 -0600657
\"Talpey, Thomas\f58851e2007-09-10 13:50:12 -0400658#endif /* _LINUX_SUNRPC_XPRT_RDMA_H */