]>
Commit | Line | Data |
---|---|---|
f58851e6 TT |
1 | /* |
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 | /* | |
41 | * transport.c | |
42 | * | |
43 | * This file contains the top-level implementation of an RPC RDMA | |
44 | * transport. | |
45 | * | |
46 | * Naming convention: functions beginning with xprt_ are part of the | |
47 | * transport switch. All others are RPC RDMA internal. | |
48 | */ | |
49 | ||
50 | #include <linux/module.h> | |
51 | #include <linux/init.h> | |
5a0e3ad6 | 52 | #include <linux/slab.h> |
f58851e6 | 53 | #include <linux/seq_file.h> |
5976687a | 54 | #include <linux/sunrpc/addr.h> |
f58851e6 TT |
55 | |
56 | #include "xprt_rdma.h" | |
57 | ||
f895b252 | 58 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) |
f58851e6 TT |
59 | # define RPCDBG_FACILITY RPCDBG_TRANS |
60 | #endif | |
61 | ||
62 | MODULE_LICENSE("Dual BSD/GPL"); | |
63 | ||
64 | MODULE_DESCRIPTION("RPC/RDMA Transport for Linux kernel NFS"); | |
65 | MODULE_AUTHOR("Network Appliance, Inc."); | |
66 | ||
67 | /* | |
68 | * tunables | |
69 | */ | |
70 | ||
71 | static unsigned int xprt_rdma_slot_table_entries = RPCRDMA_DEF_SLOT_TABLE; | |
72 | static unsigned int xprt_rdma_max_inline_read = RPCRDMA_DEF_INLINE; | |
73 | static unsigned int xprt_rdma_max_inline_write = RPCRDMA_DEF_INLINE; | |
74 | static unsigned int xprt_rdma_inline_write_padding; | |
3197d309 | 75 | static unsigned int xprt_rdma_memreg_strategy = RPCRDMA_FRMR; |
d5440e27 | 76 | int xprt_rdma_pad_optimize = 1; |
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; | |
82 | static unsigned int zero; | |
83 | static unsigned int max_padding = PAGE_SIZE; | |
84 | static unsigned int min_memreg = RPCRDMA_BOUNCEBUFFERS; | |
85 | static unsigned int max_memreg = RPCRDMA_LAST - 1; | |
86 | ||
87 | static struct ctl_table_header *sunrpc_table_header; | |
88 | ||
fe2c6338 | 89 | static struct ctl_table xr_tunables_table[] = { |
f58851e6 | 90 | { |
f58851e6 TT |
91 | .procname = "rdma_slot_table_entries", |
92 | .data = &xprt_rdma_slot_table_entries, | |
93 | .maxlen = sizeof(unsigned int), | |
94 | .mode = 0644, | |
6d456111 | 95 | .proc_handler = proc_dointvec_minmax, |
f58851e6 TT |
96 | .extra1 = &min_slot_table_size, |
97 | .extra2 = &max_slot_table_size | |
98 | }, | |
99 | { | |
f58851e6 TT |
100 | .procname = "rdma_max_inline_read", |
101 | .data = &xprt_rdma_max_inline_read, | |
102 | .maxlen = sizeof(unsigned int), | |
103 | .mode = 0644, | |
6d456111 | 104 | .proc_handler = proc_dointvec, |
f58851e6 TT |
105 | }, |
106 | { | |
f58851e6 TT |
107 | .procname = "rdma_max_inline_write", |
108 | .data = &xprt_rdma_max_inline_write, | |
109 | .maxlen = sizeof(unsigned int), | |
110 | .mode = 0644, | |
6d456111 | 111 | .proc_handler = proc_dointvec, |
f58851e6 TT |
112 | }, |
113 | { | |
f58851e6 TT |
114 | .procname = "rdma_inline_write_padding", |
115 | .data = &xprt_rdma_inline_write_padding, | |
116 | .maxlen = sizeof(unsigned int), | |
117 | .mode = 0644, | |
6d456111 | 118 | .proc_handler = proc_dointvec_minmax, |
f58851e6 TT |
119 | .extra1 = &zero, |
120 | .extra2 = &max_padding, | |
121 | }, | |
122 | { | |
f58851e6 TT |
123 | .procname = "rdma_memreg_strategy", |
124 | .data = &xprt_rdma_memreg_strategy, | |
125 | .maxlen = sizeof(unsigned int), | |
126 | .mode = 0644, | |
6d456111 | 127 | .proc_handler = proc_dointvec_minmax, |
f58851e6 TT |
128 | .extra1 = &min_memreg, |
129 | .extra2 = &max_memreg, | |
130 | }, | |
9191ca3b | 131 | { |
9191ca3b TT |
132 | .procname = "rdma_pad_optimize", |
133 | .data = &xprt_rdma_pad_optimize, | |
134 | .maxlen = sizeof(unsigned int), | |
135 | .mode = 0644, | |
6d456111 | 136 | .proc_handler = proc_dointvec, |
9191ca3b | 137 | }, |
f8572d8f | 138 | { }, |
f58851e6 TT |
139 | }; |
140 | ||
fe2c6338 | 141 | static struct ctl_table sunrpc_table[] = { |
f58851e6 | 142 | { |
f58851e6 TT |
143 | .procname = "sunrpc", |
144 | .mode = 0555, | |
145 | .child = xr_tunables_table | |
146 | }, | |
f8572d8f | 147 | { }, |
f58851e6 TT |
148 | }; |
149 | ||
150 | #endif | |
151 | ||
bfaee096 CL |
152 | #define RPCRDMA_BIND_TO (60U * HZ) |
153 | #define RPCRDMA_INIT_REEST_TO (5U * HZ) | |
154 | #define RPCRDMA_MAX_REEST_TO (30U * HZ) | |
155 | #define RPCRDMA_IDLE_DISC_TO (5U * 60 * HZ) | |
156 | ||
f58851e6 TT |
157 | static struct rpc_xprt_ops xprt_rdma_procs; /* forward reference */ |
158 | ||
159 | static void | |
160 | xprt_rdma_format_addresses(struct rpc_xprt *xprt) | |
161 | { | |
c877b849 | 162 | struct sockaddr *sap = (struct sockaddr *) |
f58851e6 | 163 | &rpcx_to_rdmad(xprt).addr; |
c877b849 CL |
164 | struct sockaddr_in *sin = (struct sockaddr_in *)sap; |
165 | char buf[64]; | |
f58851e6 | 166 | |
c877b849 CL |
167 | (void)rpc_ntop(sap, buf, sizeof(buf)); |
168 | xprt->address_strings[RPC_DISPLAY_ADDR] = kstrdup(buf, GFP_KERNEL); | |
f58851e6 | 169 | |
81160e66 | 170 | snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap)); |
c877b849 | 171 | xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL); |
f58851e6 TT |
172 | |
173 | xprt->address_strings[RPC_DISPLAY_PROTO] = "rdma"; | |
174 | ||
81160e66 | 175 | snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr)); |
c877b849 CL |
176 | xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL); |
177 | ||
81160e66 | 178 | snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap)); |
c877b849 | 179 | xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL); |
f58851e6 | 180 | |
f58851e6 TT |
181 | /* netid */ |
182 | xprt->address_strings[RPC_DISPLAY_NETID] = "rdma"; | |
183 | } | |
184 | ||
185 | static void | |
186 | xprt_rdma_free_addresses(struct rpc_xprt *xprt) | |
187 | { | |
33e01dc7 CL |
188 | unsigned int i; |
189 | ||
190 | for (i = 0; i < RPC_DISPLAY_MAX; i++) | |
191 | switch (i) { | |
192 | case RPC_DISPLAY_PROTO: | |
193 | case RPC_DISPLAY_NETID: | |
194 | continue; | |
195 | default: | |
196 | kfree(xprt->address_strings[i]); | |
197 | } | |
f58851e6 TT |
198 | } |
199 | ||
200 | static void | |
201 | xprt_rdma_connect_worker(struct work_struct *work) | |
202 | { | |
5abefb86 CL |
203 | struct rpcrdma_xprt *r_xprt = container_of(work, struct rpcrdma_xprt, |
204 | rx_connect_worker.work); | |
205 | struct rpc_xprt *xprt = &r_xprt->rx_xprt; | |
f58851e6 TT |
206 | int rc = 0; |
207 | ||
d19751e7 TM |
208 | xprt_clear_connected(xprt); |
209 | ||
210 | dprintk("RPC: %s: %sconnect\n", __func__, | |
211 | r_xprt->rx_ep.rep_connected != 0 ? "re" : ""); | |
212 | rc = rpcrdma_ep_connect(&r_xprt->rx_ep, &r_xprt->rx_ia); | |
213 | if (rc) | |
214 | xprt_wake_pending_tasks(xprt, rc); | |
215 | ||
f58851e6 TT |
216 | dprintk("RPC: %s: exit\n", __func__); |
217 | xprt_clear_connecting(xprt); | |
218 | } | |
219 | ||
220 | /* | |
221 | * xprt_rdma_destroy | |
222 | * | |
223 | * Destroy the xprt. | |
224 | * Free all memory associated with the object, including its own. | |
225 | * NOTE: none of the *destroy methods free memory for their top-level | |
226 | * objects, even though they may have allocated it (they do free | |
227 | * private memory). It's up to the caller to handle it. In this | |
228 | * case (RDMA transport), all structure memory is inlined with the | |
229 | * struct rpcrdma_xprt. | |
230 | */ | |
231 | static void | |
232 | xprt_rdma_destroy(struct rpc_xprt *xprt) | |
233 | { | |
234 | struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); | |
f58851e6 TT |
235 | |
236 | dprintk("RPC: %s: called\n", __func__); | |
237 | ||
5abefb86 | 238 | cancel_delayed_work_sync(&r_xprt->rx_connect_worker); |
f58851e6 TT |
239 | |
240 | xprt_clear_connected(xprt); | |
241 | ||
242 | rpcrdma_buffer_destroy(&r_xprt->rx_buf); | |
7f1d5419 | 243 | rpcrdma_ep_destroy(&r_xprt->rx_ep, &r_xprt->rx_ia); |
f58851e6 TT |
244 | rpcrdma_ia_close(&r_xprt->rx_ia); |
245 | ||
246 | xprt_rdma_free_addresses(xprt); | |
247 | ||
e204e621 | 248 | xprt_free(xprt); |
f58851e6 TT |
249 | |
250 | dprintk("RPC: %s: returning\n", __func__); | |
251 | ||
252 | module_put(THIS_MODULE); | |
253 | } | |
254 | ||
2881ae74 TM |
255 | static const struct rpc_timeout xprt_rdma_default_timeout = { |
256 | .to_initval = 60 * HZ, | |
257 | .to_maxval = 60 * HZ, | |
258 | }; | |
259 | ||
f58851e6 TT |
260 | /** |
261 | * xprt_setup_rdma - Set up transport to use RDMA | |
262 | * | |
263 | * @args: rpc transport arguments | |
264 | */ | |
265 | static struct rpc_xprt * | |
266 | xprt_setup_rdma(struct xprt_create *args) | |
267 | { | |
268 | struct rpcrdma_create_data_internal cdata; | |
269 | struct rpc_xprt *xprt; | |
270 | struct rpcrdma_xprt *new_xprt; | |
271 | struct rpcrdma_ep *new_ep; | |
272 | struct sockaddr_in *sin; | |
273 | int rc; | |
274 | ||
275 | if (args->addrlen > sizeof(xprt->addr)) { | |
276 | dprintk("RPC: %s: address too large\n", __func__); | |
277 | return ERR_PTR(-EBADF); | |
278 | } | |
279 | ||
37aa2133 | 280 | xprt = xprt_alloc(args->net, sizeof(struct rpcrdma_xprt), |
d9ba131d | 281 | xprt_rdma_slot_table_entries, |
bd1722d4 | 282 | xprt_rdma_slot_table_entries); |
f58851e6 TT |
283 | if (xprt == NULL) { |
284 | dprintk("RPC: %s: couldn't allocate rpcrdma_xprt\n", | |
285 | __func__); | |
286 | return ERR_PTR(-ENOMEM); | |
287 | } | |
288 | ||
f58851e6 | 289 | /* 60 second timeout, no retries */ |
ba7392bb | 290 | xprt->timeout = &xprt_rdma_default_timeout; |
bfaee096 CL |
291 | xprt->bind_timeout = RPCRDMA_BIND_TO; |
292 | xprt->reestablish_timeout = RPCRDMA_INIT_REEST_TO; | |
293 | xprt->idle_timeout = RPCRDMA_IDLE_DISC_TO; | |
f58851e6 TT |
294 | |
295 | xprt->resvport = 0; /* privileged port not needed */ | |
296 | xprt->tsh_size = 0; /* RPC-RDMA handles framing */ | |
f58851e6 TT |
297 | xprt->ops = &xprt_rdma_procs; |
298 | ||
299 | /* | |
300 | * Set up RDMA-specific connect data. | |
301 | */ | |
302 | ||
303 | /* Put server RDMA address in local cdata */ | |
304 | memcpy(&cdata.addr, args->dstaddr, args->addrlen); | |
305 | ||
306 | /* Ensure xprt->addr holds valid server TCP (not RDMA) | |
307 | * address, for any side protocols which peek at it */ | |
308 | xprt->prot = IPPROTO_TCP; | |
309 | xprt->addrlen = args->addrlen; | |
310 | memcpy(&xprt->addr, &cdata.addr, xprt->addrlen); | |
311 | ||
312 | sin = (struct sockaddr_in *)&cdata.addr; | |
313 | if (ntohs(sin->sin_port) != 0) | |
314 | xprt_set_bound(xprt); | |
315 | ||
21454aaa HH |
316 | dprintk("RPC: %s: %pI4:%u\n", |
317 | __func__, &sin->sin_addr.s_addr, ntohs(sin->sin_port)); | |
f58851e6 TT |
318 | |
319 | /* Set max requests */ | |
320 | cdata.max_requests = xprt->max_reqs; | |
321 | ||
322 | /* Set some length limits */ | |
323 | cdata.rsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA write max */ | |
324 | cdata.wsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA read max */ | |
325 | ||
326 | cdata.inline_wsize = xprt_rdma_max_inline_write; | |
327 | if (cdata.inline_wsize > cdata.wsize) | |
328 | cdata.inline_wsize = cdata.wsize; | |
329 | ||
330 | cdata.inline_rsize = xprt_rdma_max_inline_read; | |
331 | if (cdata.inline_rsize > cdata.rsize) | |
332 | cdata.inline_rsize = cdata.rsize; | |
333 | ||
334 | cdata.padding = xprt_rdma_inline_write_padding; | |
335 | ||
336 | /* | |
337 | * Create new transport instance, which includes initialized | |
338 | * o ia | |
339 | * o endpoint | |
340 | * o buffers | |
341 | */ | |
342 | ||
343 | new_xprt = rpcx_to_rdmax(xprt); | |
344 | ||
345 | rc = rpcrdma_ia_open(new_xprt, (struct sockaddr *) &cdata.addr, | |
346 | xprt_rdma_memreg_strategy); | |
347 | if (rc) | |
348 | goto out1; | |
349 | ||
350 | /* | |
351 | * initialize and create ep | |
352 | */ | |
353 | new_xprt->rx_data = cdata; | |
354 | new_ep = &new_xprt->rx_ep; | |
355 | new_ep->rep_remote_addr = cdata.addr; | |
356 | ||
357 | rc = rpcrdma_ep_create(&new_xprt->rx_ep, | |
358 | &new_xprt->rx_ia, &new_xprt->rx_data); | |
359 | if (rc) | |
360 | goto out2; | |
361 | ||
362 | /* | |
363 | * Allocate pre-registered send and receive buffers for headers and | |
364 | * any inline data. Also specify any padding which will be provided | |
365 | * from a preregistered zero buffer. | |
366 | */ | |
ac920d04 | 367 | rc = rpcrdma_buffer_create(new_xprt); |
f58851e6 TT |
368 | if (rc) |
369 | goto out3; | |
370 | ||
371 | /* | |
372 | * Register a callback for connection events. This is necessary because | |
373 | * connection loss notification is async. We also catch connection loss | |
374 | * when reaping receives. | |
375 | */ | |
5abefb86 CL |
376 | INIT_DELAYED_WORK(&new_xprt->rx_connect_worker, |
377 | xprt_rdma_connect_worker); | |
f58851e6 TT |
378 | |
379 | xprt_rdma_format_addresses(xprt); | |
43e95988 CL |
380 | xprt->max_payload = rpcrdma_max_payload(new_xprt); |
381 | dprintk("RPC: %s: transport data payload maximum: %zu bytes\n", | |
382 | __func__, xprt->max_payload); | |
f58851e6 TT |
383 | |
384 | if (!try_module_get(THIS_MODULE)) | |
385 | goto out4; | |
386 | ||
387 | return xprt; | |
388 | ||
389 | out4: | |
390 | xprt_rdma_free_addresses(xprt); | |
391 | rc = -EINVAL; | |
392 | out3: | |
7f1d5419 | 393 | rpcrdma_ep_destroy(new_ep, &new_xprt->rx_ia); |
f58851e6 TT |
394 | out2: |
395 | rpcrdma_ia_close(&new_xprt->rx_ia); | |
396 | out1: | |
e204e621 | 397 | xprt_free(xprt); |
f58851e6 TT |
398 | return ERR_PTR(rc); |
399 | } | |
400 | ||
401 | /* | |
402 | * Close a connection, during shutdown or timeout/reconnect | |
403 | */ | |
404 | static void | |
405 | xprt_rdma_close(struct rpc_xprt *xprt) | |
406 | { | |
407 | struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); | |
408 | ||
409 | dprintk("RPC: %s: closing\n", __func__); | |
08ca0dce TT |
410 | if (r_xprt->rx_ep.rep_connected > 0) |
411 | xprt->reestablish_timeout = 0; | |
62da3b24 | 412 | xprt_disconnect_done(xprt); |
282191cb | 413 | rpcrdma_ep_disconnect(&r_xprt->rx_ep, &r_xprt->rx_ia); |
f58851e6 TT |
414 | } |
415 | ||
416 | static void | |
417 | xprt_rdma_set_port(struct rpc_xprt *xprt, u16 port) | |
418 | { | |
419 | struct sockaddr_in *sap; | |
420 | ||
421 | sap = (struct sockaddr_in *)&xprt->addr; | |
422 | sap->sin_port = htons(port); | |
423 | sap = (struct sockaddr_in *)&rpcx_to_rdmad(xprt).addr; | |
424 | sap->sin_port = htons(port); | |
425 | dprintk("RPC: %s: %u\n", __func__, port); | |
426 | } | |
427 | ||
428 | static void | |
1b092092 | 429 | xprt_rdma_connect(struct rpc_xprt *xprt, struct rpc_task *task) |
f58851e6 | 430 | { |
f58851e6 TT |
431 | struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); |
432 | ||
0b9e7943 TM |
433 | if (r_xprt->rx_ep.rep_connected != 0) { |
434 | /* Reconnect */ | |
5abefb86 CL |
435 | schedule_delayed_work(&r_xprt->rx_connect_worker, |
436 | xprt->reestablish_timeout); | |
0b9e7943 | 437 | xprt->reestablish_timeout <<= 1; |
bfaee096 CL |
438 | if (xprt->reestablish_timeout > RPCRDMA_MAX_REEST_TO) |
439 | xprt->reestablish_timeout = RPCRDMA_MAX_REEST_TO; | |
440 | else if (xprt->reestablish_timeout < RPCRDMA_INIT_REEST_TO) | |
441 | xprt->reestablish_timeout = RPCRDMA_INIT_REEST_TO; | |
0b9e7943 | 442 | } else { |
5abefb86 | 443 | schedule_delayed_work(&r_xprt->rx_connect_worker, 0); |
0b9e7943 | 444 | if (!RPC_IS_ASYNC(task)) |
5abefb86 | 445 | flush_delayed_work(&r_xprt->rx_connect_worker); |
f58851e6 TT |
446 | } |
447 | } | |
448 | ||
f58851e6 TT |
449 | /* |
450 | * The RDMA allocate/free functions need the task structure as a place | |
451 | * to hide the struct rpcrdma_req, which is necessary for the actual send/recv | |
0ca77dc3 CL |
452 | * sequence. |
453 | * | |
454 | * The RPC layer allocates both send and receive buffers in the same call | |
455 | * (rq_send_buf and rq_rcv_buf are both part of a single contiguous buffer). | |
456 | * We may register rq_rcv_buf when using reply chunks. | |
f58851e6 TT |
457 | */ |
458 | static void * | |
459 | xprt_rdma_allocate(struct rpc_task *task, size_t size) | |
460 | { | |
a4f0835c | 461 | struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt; |
0ca77dc3 CL |
462 | struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); |
463 | struct rpcrdma_regbuf *rb; | |
464 | struct rpcrdma_req *req; | |
465 | size_t min_size; | |
466 | gfp_t flags = task->tk_flags & RPC_TASK_SWAPPER ? | |
467 | GFP_ATOMIC : GFP_NOFS; | |
f58851e6 | 468 | |
0ca77dc3 | 469 | req = rpcrdma_buffer_get(&r_xprt->rx_buf); |
c977dea2 CL |
470 | if (req == NULL) |
471 | return NULL; | |
f58851e6 | 472 | |
85275c87 CL |
473 | if (req->rl_rdmabuf == NULL) |
474 | goto out_rdmabuf; | |
0ca77dc3 CL |
475 | if (req->rl_sendbuf == NULL) |
476 | goto out_sendbuf; | |
477 | if (size > req->rl_sendbuf->rg_size) | |
478 | goto out_sendbuf; | |
479 | ||
480 | out: | |
f58851e6 | 481 | dprintk("RPC: %s: size %zd, request 0x%p\n", __func__, size, req); |
575448bd | 482 | req->rl_connect_cookie = 0; /* our reserved value */ |
0ca77dc3 CL |
483 | return req->rl_sendbuf->rg_base; |
484 | ||
85275c87 CL |
485 | out_rdmabuf: |
486 | min_size = RPCRDMA_INLINE_WRITE_THRESHOLD(task->tk_rqstp); | |
487 | rb = rpcrdma_alloc_regbuf(&r_xprt->rx_ia, min_size, flags); | |
488 | if (IS_ERR(rb)) | |
489 | goto out_fail; | |
490 | req->rl_rdmabuf = rb; | |
491 | ||
0ca77dc3 CL |
492 | out_sendbuf: |
493 | /* XDR encoding and RPC/RDMA marshaling of this request has not | |
494 | * yet occurred. Thus a lower bound is needed to prevent buffer | |
495 | * overrun during marshaling. | |
496 | * | |
497 | * RPC/RDMA marshaling may choose to send payload bearing ops | |
498 | * inline, if the result is smaller than the inline threshold. | |
499 | * The value of the "size" argument accounts for header | |
500 | * requirements but not for the payload in these cases. | |
501 | * | |
502 | * Likewise, allocate enough space to receive a reply up to the | |
503 | * size of the inline threshold. | |
504 | * | |
505 | * It's unlikely that both the send header and the received | |
506 | * reply will be large, but slush is provided here to allow | |
507 | * flexibility when marshaling. | |
508 | */ | |
509 | min_size = RPCRDMA_INLINE_READ_THRESHOLD(task->tk_rqstp); | |
510 | min_size += RPCRDMA_INLINE_WRITE_THRESHOLD(task->tk_rqstp); | |
511 | if (size < min_size) | |
512 | size = min_size; | |
513 | ||
514 | rb = rpcrdma_alloc_regbuf(&r_xprt->rx_ia, size, flags); | |
515 | if (IS_ERR(rb)) | |
516 | goto out_fail; | |
517 | rb->rg_owner = req; | |
518 | ||
519 | r_xprt->rx_stats.hardway_register_count += size; | |
520 | rpcrdma_free_regbuf(&r_xprt->rx_ia, req->rl_sendbuf); | |
521 | req->rl_sendbuf = rb; | |
522 | goto out; | |
523 | ||
524 | out_fail: | |
f58851e6 | 525 | rpcrdma_buffer_put(req); |
0ca77dc3 | 526 | r_xprt->rx_stats.failed_marshal_count++; |
f58851e6 TT |
527 | return NULL; |
528 | } | |
529 | ||
530 | /* | |
531 | * This function returns all RDMA resources to the pool. | |
532 | */ | |
533 | static void | |
534 | xprt_rdma_free(void *buffer) | |
535 | { | |
536 | struct rpcrdma_req *req; | |
537 | struct rpcrdma_xprt *r_xprt; | |
0ca77dc3 | 538 | struct rpcrdma_regbuf *rb; |
f58851e6 TT |
539 | int i; |
540 | ||
541 | if (buffer == NULL) | |
542 | return; | |
543 | ||
0ca77dc3 CL |
544 | rb = container_of(buffer, struct rpcrdma_regbuf, rg_base[0]); |
545 | req = rb->rg_owner; | |
546 | r_xprt = container_of(req->rl_buffer, struct rpcrdma_xprt, rx_buf); | |
f58851e6 | 547 | |
0ca77dc3 | 548 | dprintk("RPC: %s: called on 0x%p\n", __func__, req->rl_reply); |
f58851e6 | 549 | |
f58851e6 TT |
550 | for (i = 0; req->rl_nchunks;) { |
551 | --req->rl_nchunks; | |
552 | i += rpcrdma_deregister_external( | |
13c9ff8f | 553 | &req->rl_segments[i], r_xprt); |
f58851e6 TT |
554 | } |
555 | ||
f58851e6 TT |
556 | rpcrdma_buffer_put(req); |
557 | } | |
558 | ||
559 | /* | |
560 | * send_request invokes the meat of RPC RDMA. It must do the following: | |
561 | * 1. Marshal the RPC request into an RPC RDMA request, which means | |
562 | * putting a header in front of data, and creating IOVs for RDMA | |
563 | * from those in the request. | |
564 | * 2. In marshaling, detect opportunities for RDMA, and use them. | |
565 | * 3. Post a recv message to set up asynch completion, then send | |
566 | * the request (rpcrdma_ep_post). | |
567 | * 4. No partial sends are possible in the RPC-RDMA protocol (as in UDP). | |
568 | */ | |
569 | ||
570 | static int | |
571 | xprt_rdma_send_request(struct rpc_task *task) | |
572 | { | |
573 | struct rpc_rqst *rqst = task->tk_rqstp; | |
a4f0835c | 574 | struct rpc_xprt *xprt = rqst->rq_xprt; |
f58851e6 TT |
575 | struct rpcrdma_req *req = rpcr_to_rdmar(rqst); |
576 | struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); | |
6ab59945 | 577 | int rc = 0; |
f58851e6 | 578 | |
6ab59945 | 579 | if (req->rl_niovs == 0) |
c93c6223 | 580 | rc = rpcrdma_marshal_req(rqst); |
467c9674 | 581 | else if (r_xprt->rx_ia.ri_memreg_strategy != RPCRDMA_ALLPHYSICAL) |
6ab59945 CL |
582 | rc = rpcrdma_marshal_chunks(rqst, 0); |
583 | if (rc < 0) | |
584 | goto failed_marshal; | |
f58851e6 TT |
585 | |
586 | if (req->rl_reply == NULL) /* e.g. reconnection */ | |
587 | rpcrdma_recv_buffer_get(req); | |
588 | ||
589 | if (req->rl_reply) { | |
590 | req->rl_reply->rr_func = rpcrdma_reply_handler; | |
591 | /* this need only be done once, but... */ | |
592 | req->rl_reply->rr_xprt = xprt; | |
593 | } | |
594 | ||
575448bd TT |
595 | /* Must suppress retransmit to maintain credits */ |
596 | if (req->rl_connect_cookie == xprt->connect_cookie) | |
597 | goto drop_connection; | |
598 | req->rl_connect_cookie = xprt->connect_cookie; | |
599 | ||
600 | if (rpcrdma_ep_post(&r_xprt->rx_ia, &r_xprt->rx_ep, req)) | |
601 | goto drop_connection; | |
f58851e6 | 602 | |
d60dbb20 | 603 | rqst->rq_xmit_bytes_sent += rqst->rq_snd_buf.len; |
f58851e6 TT |
604 | rqst->rq_bytes_sent = 0; |
605 | return 0; | |
575448bd | 606 | |
c93c6223 CL |
607 | failed_marshal: |
608 | r_xprt->rx_stats.failed_marshal_count++; | |
609 | dprintk("RPC: %s: rpcrdma_marshal_req failed, status %i\n", | |
610 | __func__, rc); | |
611 | if (rc == -EIO) | |
612 | return -EIO; | |
575448bd TT |
613 | drop_connection: |
614 | xprt_disconnect_done(xprt); | |
615 | return -ENOTCONN; /* implies disconnect */ | |
f58851e6 TT |
616 | } |
617 | ||
618 | static void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) | |
619 | { | |
620 | struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); | |
621 | long idle_time = 0; | |
622 | ||
623 | if (xprt_connected(xprt)) | |
624 | idle_time = (long)(jiffies - xprt->last_used) / HZ; | |
625 | ||
626 | seq_printf(seq, | |
627 | "\txprt:\trdma %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu " | |
628 | "%lu %lu %lu %Lu %Lu %Lu %Lu %lu %lu %lu\n", | |
629 | ||
630 | 0, /* need a local port? */ | |
631 | xprt->stat.bind_count, | |
632 | xprt->stat.connect_count, | |
633 | xprt->stat.connect_time, | |
634 | idle_time, | |
635 | xprt->stat.sends, | |
636 | xprt->stat.recvs, | |
637 | xprt->stat.bad_xids, | |
638 | xprt->stat.req_u, | |
639 | xprt->stat.bklog_u, | |
640 | ||
641 | r_xprt->rx_stats.read_chunk_count, | |
642 | r_xprt->rx_stats.write_chunk_count, | |
643 | r_xprt->rx_stats.reply_chunk_count, | |
644 | r_xprt->rx_stats.total_rdma_request, | |
645 | r_xprt->rx_stats.total_rdma_reply, | |
646 | r_xprt->rx_stats.pullup_copy_count, | |
647 | r_xprt->rx_stats.fixup_copy_count, | |
648 | r_xprt->rx_stats.hardway_register_count, | |
649 | r_xprt->rx_stats.failed_marshal_count, | |
650 | r_xprt->rx_stats.bad_reply_count); | |
651 | } | |
652 | ||
653 | /* | |
654 | * Plumbing for rpc transport switch and kernel module | |
655 | */ | |
656 | ||
657 | static struct rpc_xprt_ops xprt_rdma_procs = { | |
e7ce710a | 658 | .reserve_xprt = xprt_reserve_xprt_cong, |
f58851e6 | 659 | .release_xprt = xprt_release_xprt_cong, /* sunrpc/xprt.c */ |
f39c1bfb | 660 | .alloc_slot = xprt_alloc_slot, |
f58851e6 TT |
661 | .release_request = xprt_release_rqst_cong, /* ditto */ |
662 | .set_retrans_timeout = xprt_set_retrans_timeout_def, /* ditto */ | |
663 | .rpcbind = rpcb_getport_async, /* sunrpc/rpcb_clnt.c */ | |
664 | .set_port = xprt_rdma_set_port, | |
665 | .connect = xprt_rdma_connect, | |
666 | .buf_alloc = xprt_rdma_allocate, | |
667 | .buf_free = xprt_rdma_free, | |
668 | .send_request = xprt_rdma_send_request, | |
669 | .close = xprt_rdma_close, | |
670 | .destroy = xprt_rdma_destroy, | |
671 | .print_stats = xprt_rdma_print_stats | |
672 | }; | |
673 | ||
674 | static struct xprt_class xprt_rdma = { | |
675 | .list = LIST_HEAD_INIT(xprt_rdma.list), | |
676 | .name = "rdma", | |
677 | .owner = THIS_MODULE, | |
678 | .ident = XPRT_TRANSPORT_RDMA, | |
679 | .setup = xprt_setup_rdma, | |
680 | }; | |
681 | ||
682 | static void __exit xprt_rdma_cleanup(void) | |
683 | { | |
684 | int rc; | |
685 | ||
3a0799a9 | 686 | dprintk("RPCRDMA Module Removed, deregister RPC RDMA transport\n"); |
f895b252 | 687 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) |
f58851e6 TT |
688 | if (sunrpc_table_header) { |
689 | unregister_sysctl_table(sunrpc_table_header); | |
690 | sunrpc_table_header = NULL; | |
691 | } | |
692 | #endif | |
693 | rc = xprt_unregister_transport(&xprt_rdma); | |
694 | if (rc) | |
695 | dprintk("RPC: %s: xprt_unregister returned %i\n", | |
696 | __func__, rc); | |
697 | } | |
698 | ||
699 | static int __init xprt_rdma_init(void) | |
700 | { | |
701 | int rc; | |
702 | ||
703 | rc = xprt_register_transport(&xprt_rdma); | |
704 | ||
705 | if (rc) | |
706 | return rc; | |
707 | ||
3a0799a9 | 708 | dprintk("RPCRDMA Module Init, register RPC RDMA transport\n"); |
f58851e6 | 709 | |
3a0799a9 CL |
710 | dprintk("Defaults:\n"); |
711 | dprintk("\tSlots %d\n" | |
f58851e6 TT |
712 | "\tMaxInlineRead %d\n\tMaxInlineWrite %d\n", |
713 | xprt_rdma_slot_table_entries, | |
714 | xprt_rdma_max_inline_read, xprt_rdma_max_inline_write); | |
3a0799a9 | 715 | dprintk("\tPadding %d\n\tMemreg %d\n", |
f58851e6 TT |
716 | xprt_rdma_inline_write_padding, xprt_rdma_memreg_strategy); |
717 | ||
f895b252 | 718 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) |
f58851e6 TT |
719 | if (!sunrpc_table_header) |
720 | sunrpc_table_header = register_sysctl_table(sunrpc_table); | |
721 | #endif | |
722 | return 0; | |
723 | } | |
724 | ||
725 | module_init(xprt_rdma_init); | |
726 | module_exit(xprt_rdma_cleanup); |