]>
Commit | Line | Data |
---|---|---|
a2268cfb | 1 | // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause |
f58851e6 | 2 | /* |
62b56a67 | 3 | * Copyright (c) 2014-2017 Oracle. All rights reserved. |
f58851e6 TT |
4 | * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved. |
5 | * | |
6 | * This software is available to you under a choice of one of two | |
7 | * licenses. You may choose to be licensed under the terms of the GNU | |
8 | * General Public License (GPL) Version 2, available from the file | |
9 | * COPYING in the main directory of this source tree, or the BSD-type | |
10 | * license below: | |
11 | * | |
12 | * Redistribution and use in source and binary forms, with or without | |
13 | * modification, are permitted provided that the following conditions | |
14 | * are met: | |
15 | * | |
16 | * Redistributions of source code must retain the above copyright | |
17 | * notice, this list of conditions and the following disclaimer. | |
18 | * | |
19 | * Redistributions in binary form must reproduce the above | |
20 | * copyright notice, this list of conditions and the following | |
21 | * disclaimer in the documentation and/or other materials provided | |
22 | * with the distribution. | |
23 | * | |
24 | * Neither the name of the Network Appliance, Inc. nor the names of | |
25 | * its contributors may be used to endorse or promote products | |
26 | * derived from this software without specific prior written | |
27 | * permission. | |
28 | * | |
29 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS | |
30 | * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT | |
31 | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR | |
32 | * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT | |
33 | * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, | |
34 | * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT | |
35 | * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, | |
36 | * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY | |
37 | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT | |
38 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE | |
39 | * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |
40 | */ | |
41 | ||
42 | /* | |
43 | * transport.c | |
44 | * | |
45 | * This file contains the top-level implementation of an RPC RDMA | |
46 | * transport. | |
47 | * | |
48 | * Naming convention: functions beginning with xprt_ are part of the | |
49 | * transport switch. All others are RPC RDMA internal. | |
50 | */ | |
51 | ||
52 | #include <linux/module.h> | |
5a0e3ad6 | 53 | #include <linux/slab.h> |
f58851e6 | 54 | #include <linux/seq_file.h> |
bd2abef3 CL |
55 | #include <linux/smp.h> |
56 | ||
5976687a | 57 | #include <linux/sunrpc/addr.h> |
bd2abef3 | 58 | #include <linux/sunrpc/svc_rdma.h> |
f58851e6 TT |
59 | |
60 | #include "xprt_rdma.h" | |
b6e717cb | 61 | #include <trace/events/rpcrdma.h> |
f58851e6 | 62 | |
f895b252 | 63 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) |
f58851e6 TT |
64 | # define RPCDBG_FACILITY RPCDBG_TRANS |
65 | #endif | |
66 | ||
f58851e6 TT |
67 | /* |
68 | * tunables | |
69 | */ | |
70 | ||
5bffb006 | 71 | static unsigned int xprt_rdma_slot_table_entries = RPCRDMA_DEF_SLOT_TABLE; |
5d252f90 | 72 | unsigned int xprt_rdma_max_inline_read = RPCRDMA_DEF_INLINE; |
94087e97 | 73 | unsigned int xprt_rdma_max_inline_write = RPCRDMA_DEF_INLINE; |
ce5b3717 | 74 | unsigned int xprt_rdma_memreg_strategy = RPCRDMA_FRWR; |
fff09594 | 75 | int xprt_rdma_pad_optimize; |
d3abc739 | 76 | static struct xprt_class xprt_rdma; |
f58851e6 | 77 | |
f895b252 | 78 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) |
f58851e6 TT |
79 | |
80 | static unsigned int min_slot_table_size = RPCRDMA_MIN_SLOT_TABLE; | |
81 | static unsigned int max_slot_table_size = RPCRDMA_MAX_SLOT_TABLE; | |
29c55422 CL |
82 | static unsigned int min_inline_size = RPCRDMA_MIN_INLINE; |
83 | static unsigned int max_inline_size = RPCRDMA_MAX_INLINE; | |
f58851e6 TT |
84 | static unsigned int max_padding = PAGE_SIZE; |
85 | static unsigned int min_memreg = RPCRDMA_BOUNCEBUFFERS; | |
86 | static unsigned int max_memreg = RPCRDMA_LAST - 1; | |
10492704 | 87 | static unsigned int dummy; |
f58851e6 TT |
88 | |
89 | static struct ctl_table_header *sunrpc_table_header; | |
90 | ||
fe2c6338 | 91 | static struct ctl_table xr_tunables_table[] = { |
f58851e6 | 92 | { |
f58851e6 TT |
93 | .procname = "rdma_slot_table_entries", |
94 | .data = &xprt_rdma_slot_table_entries, | |
95 | .maxlen = sizeof(unsigned int), | |
96 | .mode = 0644, | |
6d456111 | 97 | .proc_handler = proc_dointvec_minmax, |
f58851e6 TT |
98 | .extra1 = &min_slot_table_size, |
99 | .extra2 = &max_slot_table_size | |
100 | }, | |
101 | { | |
f58851e6 TT |
102 | .procname = "rdma_max_inline_read", |
103 | .data = &xprt_rdma_max_inline_read, | |
104 | .maxlen = sizeof(unsigned int), | |
105 | .mode = 0644, | |
44829d02 | 106 | .proc_handler = proc_dointvec_minmax, |
29c55422 CL |
107 | .extra1 = &min_inline_size, |
108 | .extra2 = &max_inline_size, | |
f58851e6 TT |
109 | }, |
110 | { | |
f58851e6 TT |
111 | .procname = "rdma_max_inline_write", |
112 | .data = &xprt_rdma_max_inline_write, | |
113 | .maxlen = sizeof(unsigned int), | |
114 | .mode = 0644, | |
44829d02 | 115 | .proc_handler = proc_dointvec_minmax, |
29c55422 CL |
116 | .extra1 = &min_inline_size, |
117 | .extra2 = &max_inline_size, | |
f58851e6 TT |
118 | }, |
119 | { | |
f58851e6 | 120 | .procname = "rdma_inline_write_padding", |
10492704 | 121 | .data = &dummy, |
f58851e6 TT |
122 | .maxlen = sizeof(unsigned int), |
123 | .mode = 0644, | |
6d456111 | 124 | .proc_handler = proc_dointvec_minmax, |
eec4844f | 125 | .extra1 = SYSCTL_ZERO, |
f58851e6 TT |
126 | .extra2 = &max_padding, |
127 | }, | |
128 | { | |
f58851e6 TT |
129 | .procname = "rdma_memreg_strategy", |
130 | .data = &xprt_rdma_memreg_strategy, | |
131 | .maxlen = sizeof(unsigned int), | |
132 | .mode = 0644, | |
6d456111 | 133 | .proc_handler = proc_dointvec_minmax, |
f58851e6 TT |
134 | .extra1 = &min_memreg, |
135 | .extra2 = &max_memreg, | |
136 | }, | |
9191ca3b | 137 | { |
9191ca3b TT |
138 | .procname = "rdma_pad_optimize", |
139 | .data = &xprt_rdma_pad_optimize, | |
140 | .maxlen = sizeof(unsigned int), | |
141 | .mode = 0644, | |
6d456111 | 142 | .proc_handler = proc_dointvec, |
9191ca3b | 143 | }, |
f8572d8f | 144 | { }, |
f58851e6 TT |
145 | }; |
146 | ||
fe2c6338 | 147 | static struct ctl_table sunrpc_table[] = { |
f58851e6 | 148 | { |
f58851e6 TT |
149 | .procname = "sunrpc", |
150 | .mode = 0555, | |
151 | .child = xr_tunables_table | |
152 | }, | |
f8572d8f | 153 | { }, |
f58851e6 TT |
154 | }; |
155 | ||
156 | #endif | |
157 | ||
d31ae254 | 158 | static const struct rpc_xprt_ops xprt_rdma_procs; |
f58851e6 | 159 | |
0dd39cae CL |
160 | static void |
161 | xprt_rdma_format_addresses4(struct rpc_xprt *xprt, struct sockaddr *sap) | |
162 | { | |
163 | struct sockaddr_in *sin = (struct sockaddr_in *)sap; | |
164 | char buf[20]; | |
165 | ||
166 | snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr)); | |
167 | xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL); | |
168 | ||
169 | xprt->address_strings[RPC_DISPLAY_NETID] = RPCBIND_NETID_RDMA; | |
170 | } | |
171 | ||
172 | static void | |
173 | xprt_rdma_format_addresses6(struct rpc_xprt *xprt, struct sockaddr *sap) | |
174 | { | |
175 | struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap; | |
176 | char buf[40]; | |
177 | ||
178 | snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr); | |
179 | xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL); | |
180 | ||
181 | xprt->address_strings[RPC_DISPLAY_NETID] = RPCBIND_NETID_RDMA6; | |
182 | } | |
183 | ||
5d252f90 | 184 | void |
5231eb97 | 185 | xprt_rdma_format_addresses(struct rpc_xprt *xprt, struct sockaddr *sap) |
f58851e6 | 186 | { |
0dd39cae CL |
187 | char buf[128]; |
188 | ||
189 | switch (sap->sa_family) { | |
190 | case AF_INET: | |
191 | xprt_rdma_format_addresses4(xprt, sap); | |
192 | break; | |
193 | case AF_INET6: | |
194 | xprt_rdma_format_addresses6(xprt, sap); | |
195 | break; | |
196 | default: | |
197 | pr_err("rpcrdma: Unrecognized address family\n"); | |
198 | return; | |
199 | } | |
f58851e6 | 200 | |
c877b849 CL |
201 | (void)rpc_ntop(sap, buf, sizeof(buf)); |
202 | xprt->address_strings[RPC_DISPLAY_ADDR] = kstrdup(buf, GFP_KERNEL); | |
f58851e6 | 203 | |
81160e66 | 204 | snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap)); |
c877b849 | 205 | xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL); |
f58851e6 | 206 | |
81160e66 | 207 | snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap)); |
c877b849 | 208 | xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL); |
f58851e6 | 209 | |
0dd39cae | 210 | xprt->address_strings[RPC_DISPLAY_PROTO] = "rdma"; |
f58851e6 TT |
211 | } |
212 | ||
5d252f90 | 213 | void |
f58851e6 TT |
214 | xprt_rdma_free_addresses(struct rpc_xprt *xprt) |
215 | { | |
33e01dc7 CL |
216 | unsigned int i; |
217 | ||
218 | for (i = 0; i < RPC_DISPLAY_MAX; i++) | |
219 | switch (i) { | |
220 | case RPC_DISPLAY_PROTO: | |
221 | case RPC_DISPLAY_NETID: | |
222 | continue; | |
223 | default: | |
224 | kfree(xprt->address_strings[i]); | |
225 | } | |
f58851e6 TT |
226 | } |
227 | ||
31e62d25 CL |
228 | /** |
229 | * xprt_rdma_connect_worker - establish connection in the background | |
230 | * @work: worker thread context | |
231 | * | |
232 | * Requester holds the xprt's send lock to prevent activity on this | |
233 | * transport while a fresh connection is being established. RPC tasks | |
234 | * sleep on the xprt's pending queue waiting for connect to complete. | |
235 | */ | |
f58851e6 TT |
236 | static void |
237 | xprt_rdma_connect_worker(struct work_struct *work) | |
238 | { | |
5abefb86 CL |
239 | struct rpcrdma_xprt *r_xprt = container_of(work, struct rpcrdma_xprt, |
240 | rx_connect_worker.work); | |
241 | struct rpc_xprt *xprt = &r_xprt->rx_xprt; | |
31e62d25 | 242 | int rc; |
d19751e7 | 243 | |
9144a803 | 244 | rc = rpcrdma_xprt_connect(r_xprt); |
f58851e6 | 245 | xprt_clear_connecting(xprt); |
2d97f463 | 246 | if (!rc) { |
e28ce900 | 247 | xprt->connect_cookie++; |
6cb28687 CL |
248 | xprt->stat.connect_count++; |
249 | xprt->stat.connect_time += (long)jiffies - | |
250 | xprt->stat.connect_start; | |
251 | xprt_set_connected(xprt); | |
252 | rc = -EAGAIN; | |
f99fa508 TM |
253 | } else |
254 | rpcrdma_xprt_disconnect(r_xprt); | |
255 | xprt_unlock_connect(xprt, r_xprt); | |
6cb28687 | 256 | xprt_wake_pending_tasks(xprt, rc); |
f58851e6 TT |
257 | } |
258 | ||
f26c32fa CL |
259 | /** |
260 | * xprt_rdma_inject_disconnect - inject a connection fault | |
261 | * @xprt: transport context | |
262 | * | |
7638e0bf CL |
263 | * If @xprt is connected, disconnect it to simulate spurious |
264 | * connection loss. Caller must hold @xprt's send lock to | |
265 | * ensure that data structures and hardware resources are | |
266 | * stable during the rdma_disconnect() call. | |
f26c32fa | 267 | */ |
4a068258 CL |
268 | static void |
269 | xprt_rdma_inject_disconnect(struct rpc_xprt *xprt) | |
270 | { | |
ad091180 | 271 | struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); |
4a068258 | 272 | |
395069fc | 273 | trace_xprtrdma_op_inject_dsc(r_xprt); |
e28ce900 | 274 | rdma_disconnect(r_xprt->rx_ep->re_id); |
4a068258 CL |
275 | } |
276 | ||
f26c32fa CL |
277 | /** |
278 | * xprt_rdma_destroy - Full tear down of transport | |
279 | * @xprt: doomed transport context | |
f58851e6 | 280 | * |
f26c32fa CL |
281 | * Caller guarantees there will be no more calls to us with |
282 | * this @xprt. | |
f58851e6 TT |
283 | */ |
284 | static void | |
285 | xprt_rdma_destroy(struct rpc_xprt *xprt) | |
286 | { | |
287 | struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); | |
f58851e6 | 288 | |
5abefb86 | 289 | cancel_delayed_work_sync(&r_xprt->rx_connect_worker); |
f58851e6 | 290 | |
9144a803 | 291 | rpcrdma_xprt_disconnect(r_xprt); |
72c02173 | 292 | rpcrdma_buffer_destroy(&r_xprt->rx_buf); |
f58851e6 TT |
293 | |
294 | xprt_rdma_free_addresses(xprt); | |
e204e621 | 295 | xprt_free(xprt); |
f58851e6 | 296 | |
f58851e6 TT |
297 | module_put(THIS_MODULE); |
298 | } | |
299 | ||
675dd90a | 300 | /* 60 second timeout, no retries */ |
2881ae74 TM |
301 | static const struct rpc_timeout xprt_rdma_default_timeout = { |
302 | .to_initval = 60 * HZ, | |
303 | .to_maxval = 60 * HZ, | |
304 | }; | |
305 | ||
f58851e6 TT |
306 | /** |
307 | * xprt_setup_rdma - Set up transport to use RDMA | |
308 | * | |
309 | * @args: rpc transport arguments | |
310 | */ | |
311 | static struct rpc_xprt * | |
312 | xprt_setup_rdma(struct xprt_create *args) | |
313 | { | |
f58851e6 TT |
314 | struct rpc_xprt *xprt; |
315 | struct rpcrdma_xprt *new_xprt; | |
5231eb97 | 316 | struct sockaddr *sap; |
f58851e6 TT |
317 | int rc; |
318 | ||
ddbb347f | 319 | if (args->addrlen > sizeof(xprt->addr)) |
f58851e6 | 320 | return ERR_PTR(-EBADF); |
f58851e6 | 321 | |
e28ce900 CL |
322 | if (!try_module_get(THIS_MODULE)) |
323 | return ERR_PTR(-EIO); | |
324 | ||
7581d901 CL |
325 | xprt = xprt_alloc(args->net, sizeof(struct rpcrdma_xprt), 0, |
326 | xprt_rdma_slot_table_entries); | |
e28ce900 CL |
327 | if (!xprt) { |
328 | module_put(THIS_MODULE); | |
f58851e6 | 329 | return ERR_PTR(-ENOMEM); |
e28ce900 | 330 | } |
f58851e6 | 331 | |
ba7392bb | 332 | xprt->timeout = &xprt_rdma_default_timeout; |
675dd90a CL |
333 | xprt->connect_timeout = xprt->timeout->to_initval; |
334 | xprt->max_reconnect_timeout = xprt->timeout->to_maxval; | |
bfaee096 CL |
335 | xprt->bind_timeout = RPCRDMA_BIND_TO; |
336 | xprt->reestablish_timeout = RPCRDMA_INIT_REEST_TO; | |
337 | xprt->idle_timeout = RPCRDMA_IDLE_DISC_TO; | |
f58851e6 TT |
338 | |
339 | xprt->resvport = 0; /* privileged port not needed */ | |
f58851e6 TT |
340 | xprt->ops = &xprt_rdma_procs; |
341 | ||
342 | /* | |
343 | * Set up RDMA-specific connect data. | |
344 | */ | |
dd229cee | 345 | sap = args->dstaddr; |
f58851e6 TT |
346 | |
347 | /* Ensure xprt->addr holds valid server TCP (not RDMA) | |
348 | * address, for any side protocols which peek at it */ | |
349 | xprt->prot = IPPROTO_TCP; | |
d3abc739 | 350 | xprt->xprt_class = &xprt_rdma; |
f58851e6 | 351 | xprt->addrlen = args->addrlen; |
5231eb97 | 352 | memcpy(&xprt->addr, sap, xprt->addrlen); |
f58851e6 | 353 | |
5231eb97 | 354 | if (rpc_get_port(sap)) |
f58851e6 | 355 | xprt_set_bound(xprt); |
d461f1f2 | 356 | xprt_rdma_format_addresses(xprt, sap); |
f58851e6 | 357 | |
f58851e6 | 358 | new_xprt = rpcx_to_rdmax(xprt); |
ac920d04 | 359 | rc = rpcrdma_buffer_create(new_xprt); |
e28ce900 CL |
360 | if (rc) { |
361 | xprt_rdma_free_addresses(xprt); | |
362 | xprt_free(xprt); | |
363 | module_put(THIS_MODULE); | |
364 | return ERR_PTR(rc); | |
365 | } | |
f58851e6 | 366 | |
18d065a5 CL |
367 | INIT_DELAYED_WORK(&new_xprt->rx_connect_worker, |
368 | xprt_rdma_connect_worker); | |
93aa8e0a | 369 | |
18d065a5 CL |
370 | xprt->max_payload = RPCRDMA_MAX_DATA_SEGS << PAGE_SHIFT; |
371 | ||
f58851e6 | 372 | return xprt; |
f58851e6 TT |
373 | } |
374 | ||
bebd0318 | 375 | /** |
f26c32fa CL |
376 | * xprt_rdma_close - close a transport connection |
377 | * @xprt: transport context | |
bebd0318 | 378 | * |
395069fc CL |
379 | * Called during autoclose or device removal. |
380 | * | |
f26c32fa CL |
381 | * Caller holds @xprt's send lock to prevent activity on this |
382 | * transport while the connection is torn down. | |
f58851e6 | 383 | */ |
0c0829bc | 384 | void xprt_rdma_close(struct rpc_xprt *xprt) |
f58851e6 TT |
385 | { |
386 | struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); | |
6d2d0ee2 | 387 | |
9144a803 | 388 | rpcrdma_xprt_disconnect(r_xprt); |
ef739b21 | 389 | |
f9e1afe0 | 390 | xprt->reestablish_timeout = 0; |
0c0829bc CL |
391 | ++xprt->connect_cookie; |
392 | xprt_disconnect_done(xprt); | |
f58851e6 TT |
393 | } |
394 | ||
20035edf CL |
395 | /** |
396 | * xprt_rdma_set_port - update server port with rpcbind result | |
397 | * @xprt: controlling RPC transport | |
398 | * @port: new port value | |
399 | * | |
400 | * Transport connect status is unchanged. | |
401 | */ | |
f58851e6 TT |
402 | static void |
403 | xprt_rdma_set_port(struct rpc_xprt *xprt, u16 port) | |
404 | { | |
20035edf CL |
405 | struct sockaddr *sap = (struct sockaddr *)&xprt->addr; |
406 | char buf[8]; | |
407 | ||
20035edf | 408 | rpc_set_port(sap, port); |
f58851e6 | 409 | |
20035edf CL |
410 | kfree(xprt->address_strings[RPC_DISPLAY_PORT]); |
411 | snprintf(buf, sizeof(buf), "%u", port); | |
412 | xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL); | |
413 | ||
414 | kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]); | |
415 | snprintf(buf, sizeof(buf), "%4hx", port); | |
416 | xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL); | |
f58851e6 TT |
417 | } |
418 | ||
33849792 CL |
419 | /** |
420 | * xprt_rdma_timer - invoked when an RPC times out | |
421 | * @xprt: controlling RPC transport | |
422 | * @task: RPC task that timed out | |
423 | * | |
424 | * Invoked when the transport is still connected, but an RPC | |
425 | * retransmit timeout occurs. | |
426 | * | |
427 | * Since RDMA connections don't have a keep-alive, forcibly | |
428 | * disconnect and retry to connect. This drives full | |
429 | * detection of the network path, and retransmissions of | |
430 | * all pending RPCs. | |
431 | */ | |
432 | static void | |
433 | xprt_rdma_timer(struct rpc_xprt *xprt, struct rpc_task *task) | |
434 | { | |
33849792 CL |
435 | xprt_force_disconnect(xprt); |
436 | } | |
437 | ||
f26c32fa | 438 | /** |
675dd90a CL |
439 | * xprt_rdma_set_connect_timeout - set timeouts for establishing a connection |
440 | * @xprt: controlling transport instance | |
441 | * @connect_timeout: reconnect timeout after client disconnects | |
442 | * @reconnect_timeout: reconnect timeout after server disconnects | |
443 | * | |
444 | */ | |
2a7f77c7 CL |
445 | static void xprt_rdma_set_connect_timeout(struct rpc_xprt *xprt, |
446 | unsigned long connect_timeout, | |
447 | unsigned long reconnect_timeout) | |
675dd90a CL |
448 | { |
449 | struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); | |
450 | ||
451 | trace_xprtrdma_op_set_cto(r_xprt, connect_timeout, reconnect_timeout); | |
452 | ||
453 | spin_lock(&xprt->transport_lock); | |
454 | ||
455 | if (connect_timeout < xprt->connect_timeout) { | |
456 | struct rpc_timeout to; | |
457 | unsigned long initval; | |
458 | ||
459 | to = *xprt->timeout; | |
460 | initval = connect_timeout; | |
461 | if (initval < RPCRDMA_INIT_REEST_TO << 1) | |
462 | initval = RPCRDMA_INIT_REEST_TO << 1; | |
463 | to.to_initval = initval; | |
464 | to.to_maxval = initval; | |
465 | r_xprt->rx_timeout = to; | |
466 | xprt->timeout = &r_xprt->rx_timeout; | |
467 | xprt->connect_timeout = connect_timeout; | |
468 | } | |
469 | ||
470 | if (reconnect_timeout < xprt->max_reconnect_timeout) | |
471 | xprt->max_reconnect_timeout = reconnect_timeout; | |
472 | ||
473 | spin_unlock(&xprt->transport_lock); | |
474 | } | |
475 | ||
476 | /** | |
477 | * xprt_rdma_connect - schedule an attempt to reconnect | |
f26c32fa | 478 | * @xprt: transport state |
675dd90a | 479 | * @task: RPC scheduler context (unused) |
f26c32fa CL |
480 | * |
481 | */ | |
f58851e6 | 482 | static void |
1b092092 | 483 | xprt_rdma_connect(struct rpc_xprt *xprt, struct rpc_task *task) |
f58851e6 | 484 | { |
f58851e6 | 485 | struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); |
e28ce900 | 486 | struct rpcrdma_ep *ep = r_xprt->rx_ep; |
675dd90a | 487 | unsigned long delay; |
f58851e6 | 488 | |
f99fa508 TM |
489 | WARN_ON_ONCE(!xprt_lock_connect(xprt, task, r_xprt)); |
490 | ||
675dd90a | 491 | delay = 0; |
e28ce900 | 492 | if (ep && ep->re_connect_status != 0) { |
675dd90a CL |
493 | delay = xprt_reconnect_delay(xprt); |
494 | xprt_reconnect_backoff(xprt, RPCRDMA_INIT_REEST_TO); | |
f58851e6 | 495 | } |
7b020f17 | 496 | trace_xprtrdma_op_connect(r_xprt, delay); |
675dd90a CL |
497 | queue_delayed_work(xprtiod_workqueue, &r_xprt->rx_connect_worker, |
498 | delay); | |
f58851e6 TT |
499 | } |
500 | ||
48be539d CL |
501 | /** |
502 | * xprt_rdma_alloc_slot - allocate an rpc_rqst | |
503 | * @xprt: controlling RPC transport | |
504 | * @task: RPC task requesting a fresh rpc_rqst | |
505 | * | |
506 | * tk_status values: | |
507 | * %0 if task->tk_rqstp points to a fresh rpc_rqst | |
508 | * %-EAGAIN if no rpc_rqst is available; queued on backlog | |
509 | */ | |
510 | static void | |
511 | xprt_rdma_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task) | |
512 | { | |
edb41e61 CL |
513 | struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); |
514 | struct rpcrdma_req *req; | |
48be539d | 515 | |
edb41e61 CL |
516 | req = rpcrdma_buffer_get(&r_xprt->rx_buf); |
517 | if (!req) | |
48be539d | 518 | goto out_sleep; |
edb41e61 | 519 | task->tk_rqstp = &req->rl_slot; |
48be539d CL |
520 | task->tk_status = 0; |
521 | return; | |
522 | ||
523 | out_sleep: | |
184efd65 | 524 | task->tk_status = -ENOMEM; |
e86be3a0 | 525 | xprt_add_backlog(xprt, task); |
48be539d CL |
526 | } |
527 | ||
528 | /** | |
529 | * xprt_rdma_free_slot - release an rpc_rqst | |
530 | * @xprt: controlling RPC transport | |
531 | * @rqst: rpc_rqst to release | |
532 | * | |
533 | */ | |
534 | static void | |
535 | xprt_rdma_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *rqst) | |
536 | { | |
5828ceba CL |
537 | struct rpcrdma_xprt *r_xprt = |
538 | container_of(xprt, struct rpcrdma_xprt, rx_xprt); | |
539 | ||
e86be3a0 TM |
540 | rpcrdma_reply_put(&r_xprt->rx_buf, rpcr_to_rdmar(rqst)); |
541 | if (!xprt_wake_up_backlog(xprt, rqst)) { | |
542 | memset(rqst, 0, sizeof(*rqst)); | |
543 | rpcrdma_buffer_put(&r_xprt->rx_buf, rpcr_to_rdmar(rqst)); | |
544 | } | |
48be539d CL |
545 | } |
546 | ||
0f665ceb CL |
547 | static bool rpcrdma_check_regbuf(struct rpcrdma_xprt *r_xprt, |
548 | struct rpcrdma_regbuf *rb, size_t size, | |
549 | gfp_t flags) | |
9c40c49f | 550 | { |
0f665ceb CL |
551 | if (unlikely(rdmab_length(rb) < size)) { |
552 | if (!rpcrdma_regbuf_realloc(rb, size, flags)) | |
553 | return false; | |
554 | r_xprt->rx_stats.hardway_register_count += size; | |
555 | } | |
9c40c49f CL |
556 | return true; |
557 | } | |
558 | ||
5fe6eaa1 CL |
559 | /** |
560 | * xprt_rdma_allocate - allocate transport resources for an RPC | |
561 | * @task: RPC task | |
562 | * | |
563 | * Return values: | |
564 | * 0: Success; rq_buffer points to RPC buffer to use | |
565 | * ENOMEM: Out of memory, call again later | |
566 | * EIO: A permanent error occurred, do not retry | |
f58851e6 | 567 | */ |
5fe6eaa1 CL |
568 | static int |
569 | xprt_rdma_allocate(struct rpc_task *task) | |
f58851e6 | 570 | { |
5fe6eaa1 | 571 | struct rpc_rqst *rqst = task->tk_rqstp; |
5fe6eaa1 | 572 | struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(rqst->rq_xprt); |
edb41e61 | 573 | struct rpcrdma_req *req = rpcr_to_rdmar(rqst); |
a0a1d50c | 574 | gfp_t flags; |
f58851e6 | 575 | |
5d252f90 | 576 | flags = RPCRDMA_DEF_GFP; |
bfd1d8f2 N |
577 | if (RPC_IS_ASYNC(task)) |
578 | flags = GFP_NOWAIT | __GFP_NOWARN; | |
a0a1d50c | 579 | if (RPC_IS_SWAPPER(task)) |
bfd1d8f2 | 580 | flags |= __GFP_MEMALLOC; |
a0a1d50c | 581 | |
0f665ceb CL |
582 | if (!rpcrdma_check_regbuf(r_xprt, req->rl_sendbuf, rqst->rq_callsize, |
583 | flags)) | |
9c40c49f | 584 | goto out_fail; |
0f665ceb CL |
585 | if (!rpcrdma_check_regbuf(r_xprt, req->rl_recvbuf, rqst->rq_rcvsize, |
586 | flags)) | |
9c40c49f CL |
587 | goto out_fail; |
588 | ||
8cec3dba CL |
589 | rqst->rq_buffer = rdmab_data(req->rl_sendbuf); |
590 | rqst->rq_rbuffer = rdmab_data(req->rl_recvbuf); | |
5fe6eaa1 | 591 | return 0; |
0ca77dc3 | 592 | |
0ca77dc3 | 593 | out_fail: |
5fe6eaa1 | 594 | return -ENOMEM; |
f58851e6 TT |
595 | } |
596 | ||
3435c74a CL |
597 | /** |
598 | * xprt_rdma_free - release resources allocated by xprt_rdma_allocate | |
599 | * @task: RPC task | |
600 | * | |
601 | * Caller guarantees rqst->rq_buffer is non-NULL. | |
f58851e6 TT |
602 | */ |
603 | static void | |
3435c74a | 604 | xprt_rdma_free(struct rpc_task *task) |
f58851e6 | 605 | { |
3435c74a | 606 | struct rpc_rqst *rqst = task->tk_rqstp; |
3435c74a | 607 | struct rpcrdma_req *req = rpcr_to_rdmar(rqst); |
f58851e6 | 608 | |
8e24e191 CL |
609 | if (unlikely(!list_empty(&req->rl_registered))) { |
610 | trace_xprtrdma_mrs_zap(task); | |
611 | frwr_unmap_sync(rpcx_to_rdmax(rqst->rq_xprt), req); | |
612 | } | |
0ab11523 CL |
613 | |
614 | /* XXX: If the RPC is completing because of a signal and | |
615 | * not because a reply was received, we ought to ensure | |
616 | * that the Send completion has fired, so that memory | |
617 | * involved with the Send is not still visible to the NIC. | |
618 | */ | |
f58851e6 TT |
619 | } |
620 | ||
7a89f9c6 CL |
621 | /** |
622 | * xprt_rdma_send_request - marshal and send an RPC request | |
adfa7144 | 623 | * @rqst: RPC message in rq_snd_buf |
7a89f9c6 | 624 | * |
bebd0318 CL |
625 | * Caller holds the transport's write lock. |
626 | * | |
cf73daf5 CL |
627 | * Returns: |
628 | * %0 if the RPC message has been sent | |
629 | * %-ENOTCONN if the caller should reconnect and call again | |
9e679d5e CL |
630 | * %-EAGAIN if the caller should call again |
631 | * %-ENOBUFS if the caller should call again after a delay | |
6946f823 CL |
632 | * %-EMSGSIZE if encoding ran out of buffer space. The request |
633 | * was not sent. Do not try to send this message again. | |
634 | * %-EIO if an I/O error occurred. The request was not sent. | |
635 | * Do not try to send this message again. | |
f58851e6 | 636 | */ |
f58851e6 | 637 | static int |
adfa7144 | 638 | xprt_rdma_send_request(struct rpc_rqst *rqst) |
f58851e6 | 639 | { |
a4f0835c | 640 | struct rpc_xprt *xprt = rqst->rq_xprt; |
f58851e6 TT |
641 | struct rpcrdma_req *req = rpcr_to_rdmar(rqst); |
642 | struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); | |
6ab59945 | 643 | int rc = 0; |
f58851e6 | 644 | |
cf73daf5 CL |
645 | #if defined(CONFIG_SUNRPC_BACKCHANNEL) |
646 | if (unlikely(!rqst->rq_buffer)) | |
647 | return xprt_rdma_bc_send_reply(rqst); | |
648 | #endif /* CONFIG_SUNRPC_BACKCHANNEL */ | |
649 | ||
bebd0318 | 650 | if (!xprt_connected(xprt)) |
0c0829bc | 651 | return -ENOTCONN; |
bebd0318 | 652 | |
75891f50 TM |
653 | if (!xprt_request_get_cong(xprt, rqst)) |
654 | return -EBADSLT; | |
655 | ||
09e60641 | 656 | rc = rpcrdma_marshal_req(r_xprt, rqst); |
6ab59945 CL |
657 | if (rc < 0) |
658 | goto failed_marshal; | |
f58851e6 | 659 | |
575448bd | 660 | /* Must suppress retransmit to maintain credits */ |
8a14793e | 661 | if (rqst->rq_connect_cookie == xprt->connect_cookie) |
575448bd | 662 | goto drop_connection; |
78215759 | 663 | rqst->rq_xtime = ktime_get(); |
575448bd | 664 | |
8d863b1f | 665 | if (frwr_send(r_xprt, req)) |
575448bd | 666 | goto drop_connection; |
f58851e6 | 667 | |
d60dbb20 | 668 | rqst->rq_xmit_bytes_sent += rqst->rq_snd_buf.len; |
fb14ae88 CL |
669 | |
670 | /* An RPC with no reply will throw off credit accounting, | |
671 | * so drop the connection to reset the credit grant. | |
672 | */ | |
50f484e2 | 673 | if (!rpc_reply_expected(rqst->rq_task)) |
fb14ae88 | 674 | goto drop_connection; |
f58851e6 | 675 | return 0; |
575448bd | 676 | |
c93c6223 | 677 | failed_marshal: |
7a89f9c6 CL |
678 | if (rc != -ENOTCONN) |
679 | return rc; | |
575448bd | 680 | drop_connection: |
0c0829bc CL |
681 | xprt_rdma_close(xprt); |
682 | return -ENOTCONN; | |
f58851e6 TT |
683 | } |
684 | ||
5d252f90 | 685 | void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) |
f58851e6 TT |
686 | { |
687 | struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); | |
688 | long idle_time = 0; | |
689 | ||
690 | if (xprt_connected(xprt)) | |
691 | idle_time = (long)(jiffies - xprt->last_used) / HZ; | |
692 | ||
763f7e4e CL |
693 | seq_puts(seq, "\txprt:\trdma "); |
694 | seq_printf(seq, "%u %lu %lu %lu %ld %lu %lu %lu %llu %llu ", | |
695 | 0, /* need a local port? */ | |
696 | xprt->stat.bind_count, | |
697 | xprt->stat.connect_count, | |
8440a886 | 698 | xprt->stat.connect_time / HZ, |
763f7e4e CL |
699 | idle_time, |
700 | xprt->stat.sends, | |
701 | xprt->stat.recvs, | |
702 | xprt->stat.bad_xids, | |
703 | xprt->stat.req_u, | |
704 | xprt->stat.bklog_u); | |
505bbe64 | 705 | seq_printf(seq, "%lu %lu %lu %llu %llu %llu %llu %lu %lu %lu %lu ", |
763f7e4e CL |
706 | r_xprt->rx_stats.read_chunk_count, |
707 | r_xprt->rx_stats.write_chunk_count, | |
708 | r_xprt->rx_stats.reply_chunk_count, | |
709 | r_xprt->rx_stats.total_rdma_request, | |
710 | r_xprt->rx_stats.total_rdma_reply, | |
711 | r_xprt->rx_stats.pullup_copy_count, | |
712 | r_xprt->rx_stats.fixup_copy_count, | |
713 | r_xprt->rx_stats.hardway_register_count, | |
714 | r_xprt->rx_stats.failed_marshal_count, | |
860477d1 CL |
715 | r_xprt->rx_stats.bad_reply_count, |
716 | r_xprt->rx_stats.nomsg_call_count); | |
01bb35c8 | 717 | seq_printf(seq, "%lu %lu %lu %lu %lu %lu\n", |
61da886b | 718 | r_xprt->rx_stats.mrs_recycled, |
e2ac236c | 719 | r_xprt->rx_stats.mrs_orphaned, |
c8b920bb | 720 | r_xprt->rx_stats.mrs_allocated, |
ae72950a | 721 | r_xprt->rx_stats.local_inv_needed, |
01bb35c8 CL |
722 | r_xprt->rx_stats.empty_sendctx_q, |
723 | r_xprt->rx_stats.reply_waits_for_send); | |
f58851e6 TT |
724 | } |
725 | ||
d67fa4d8 JL |
726 | static int |
727 | xprt_rdma_enable_swap(struct rpc_xprt *xprt) | |
728 | { | |
a0451788 | 729 | return 0; |
d67fa4d8 JL |
730 | } |
731 | ||
732 | static void | |
733 | xprt_rdma_disable_swap(struct rpc_xprt *xprt) | |
734 | { | |
735 | } | |
736 | ||
f58851e6 TT |
737 | /* |
738 | * Plumbing for rpc transport switch and kernel module | |
739 | */ | |
740 | ||
d31ae254 | 741 | static const struct rpc_xprt_ops xprt_rdma_procs = { |
e7ce710a | 742 | .reserve_xprt = xprt_reserve_xprt_cong, |
f58851e6 | 743 | .release_xprt = xprt_release_xprt_cong, /* sunrpc/xprt.c */ |
48be539d CL |
744 | .alloc_slot = xprt_rdma_alloc_slot, |
745 | .free_slot = xprt_rdma_free_slot, | |
f58851e6 | 746 | .release_request = xprt_release_rqst_cong, /* ditto */ |
8ba6a92d | 747 | .wait_for_reply_request = xprt_wait_for_reply_request_def, /* ditto */ |
33849792 | 748 | .timer = xprt_rdma_timer, |
f58851e6 TT |
749 | .rpcbind = rpcb_getport_async, /* sunrpc/rpcb_clnt.c */ |
750 | .set_port = xprt_rdma_set_port, | |
751 | .connect = xprt_rdma_connect, | |
752 | .buf_alloc = xprt_rdma_allocate, | |
753 | .buf_free = xprt_rdma_free, | |
754 | .send_request = xprt_rdma_send_request, | |
755 | .close = xprt_rdma_close, | |
756 | .destroy = xprt_rdma_destroy, | |
2a7f77c7 | 757 | .set_connect_timeout = xprt_rdma_set_connect_timeout, |
d67fa4d8 JL |
758 | .print_stats = xprt_rdma_print_stats, |
759 | .enable_swap = xprt_rdma_enable_swap, | |
760 | .disable_swap = xprt_rdma_disable_swap, | |
f531a5db CL |
761 | .inject_disconnect = xprt_rdma_inject_disconnect, |
762 | #if defined(CONFIG_SUNRPC_BACKCHANNEL) | |
763 | .bc_setup = xprt_rdma_bc_setup, | |
6b26cc8c | 764 | .bc_maxpayload = xprt_rdma_bc_maxpayload, |
7402a4fe | 765 | .bc_num_slots = xprt_rdma_bc_max_slots, |
f531a5db CL |
766 | .bc_free_rqst = xprt_rdma_bc_free_rqst, |
767 | .bc_destroy = xprt_rdma_bc_destroy, | |
768 | #endif | |
f58851e6 TT |
769 | }; |
770 | ||
771 | static struct xprt_class xprt_rdma = { | |
772 | .list = LIST_HEAD_INIT(xprt_rdma.list), | |
773 | .name = "rdma", | |
774 | .owner = THIS_MODULE, | |
775 | .ident = XPRT_TRANSPORT_RDMA, | |
776 | .setup = xprt_setup_rdma, | |
d5aa6b22 | 777 | .netid = { "rdma", "rdma6", "" }, |
f58851e6 TT |
778 | }; |
779 | ||
ffe1f0df | 780 | void xprt_rdma_cleanup(void) |
f58851e6 | 781 | { |
f895b252 | 782 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) |
f58851e6 TT |
783 | if (sunrpc_table_header) { |
784 | unregister_sysctl_table(sunrpc_table_header); | |
785 | sunrpc_table_header = NULL; | |
786 | } | |
787 | #endif | |
5d252f90 | 788 | |
ddbb347f CL |
789 | xprt_unregister_transport(&xprt_rdma); |
790 | xprt_unregister_transport(&xprt_rdma_bc); | |
f58851e6 TT |
791 | } |
792 | ||
ffe1f0df | 793 | int xprt_rdma_init(void) |
f58851e6 TT |
794 | { |
795 | int rc; | |
796 | ||
951e721c | 797 | rc = xprt_register_transport(&xprt_rdma); |
6d2d0ee2 | 798 | if (rc) |
951e721c | 799 | return rc; |
951e721c | 800 | |
5d252f90 CL |
801 | rc = xprt_register_transport(&xprt_rdma_bc); |
802 | if (rc) { | |
803 | xprt_unregister_transport(&xprt_rdma); | |
5d252f90 CL |
804 | return rc; |
805 | } | |
806 | ||
f895b252 | 807 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) |
f58851e6 TT |
808 | if (!sunrpc_table_header) |
809 | sunrpc_table_header = register_sysctl_table(sunrpc_table); | |
810 | #endif | |
811 | return 0; | |
812 | } |