?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'regmap-fix-v5.11-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux/fpc-iii.git] / fs / cifs / smbdirect.c
blob10dfe50067920b96924ea8fc4892ac2b2efc7fda
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright (C) 2017, Microsoft Corporation.
4 *
5 * Author(s): Long Li <longli@microsoft.com>
6 */
7 #include <linux/module.h>
8 #include <linux/highmem.h>
9 #include "smbdirect.h"
10 #include "cifs_debug.h"
11 #include "cifsproto.h"
12 #include "smb2proto.h"
13
14 static struct smbd_response *get_empty_queue_buffer(
15 struct smbd_connection *info);
16 static struct smbd_response *get_receive_buffer(
17 struct smbd_connection *info);
18 static void put_receive_buffer(
19 struct smbd_connection *info,
20 struct smbd_response *response);
21 static int allocate_receive_buffers(struct smbd_connection *info, int num_buf);
22 static void destroy_receive_buffers(struct smbd_connection *info);
23
24 static void put_empty_packet(
25 struct smbd_connection *info, struct smbd_response *response);
26 static void enqueue_reassembly(
27 struct smbd_connection *info,
28 struct smbd_response *response, int data_length);
29 static struct smbd_response *_get_first_reassembly(
30 struct smbd_connection *info);
31
32 static int smbd_post_recv(
33 struct smbd_connection *info,
34 struct smbd_response *response);
35
36 static int smbd_post_send_empty(struct smbd_connection *info);
37 static int smbd_post_send_data(
38 struct smbd_connection *info,
39 struct kvec *iov, int n_vec, int remaining_data_length);
40 static int smbd_post_send_page(struct smbd_connection *info,
41 struct page *page, unsigned long offset,
42 size_t size, int remaining_data_length);
43
44 static void destroy_mr_list(struct smbd_connection *info);
45 static int allocate_mr_list(struct smbd_connection *info);
46
47 /* SMBD version number */
48 #define SMBD_V1 0x0100
49
50 /* Port numbers for SMBD transport */
51 #define SMB_PORT 445
52 #define SMBD_PORT 5445
53
54 /* Address lookup and resolve timeout in ms */
55 #define RDMA_RESOLVE_TIMEOUT 5000
56
57 /* SMBD negotiation timeout in seconds */
58 #define SMBD_NEGOTIATE_TIMEOUT 120
59
60 /* SMBD minimum receive size and fragmented sized defined in [MS-SMBD] */
61 #define SMBD_MIN_RECEIVE_SIZE 128
62 #define SMBD_MIN_FRAGMENTED_SIZE 131072
63
64 /*
65 * Default maximum number of RDMA read/write outstanding on this connection
66 * This value is possibly decreased during QP creation on hardware limit
67 */
68 #define SMBD_CM_RESPONDER_RESOURCES 32
69
70 /* Maximum number of retries on data transfer operations */
71 #define SMBD_CM_RETRY 6
72 /* No need to retry on Receiver Not Ready since SMBD manages credits */
73 #define SMBD_CM_RNR_RETRY 0
74
75 /*
76 * User configurable initial values per SMBD transport connection
77 * as defined in [MS-SMBD] 3.1.1.1
78 * Those may change after a SMBD negotiation
79 */
80 /* The local peer's maximum number of credits to grant to the peer */
81 int smbd_receive_credit_max = 255;
82
83 /* The remote peer's credit request of local peer */
84 int smbd_send_credit_target = 255;
85
86 /* The maximum single message size can be sent to remote peer */
87 int smbd_max_send_size = 1364;
88
89 /* The maximum fragmented upper-layer payload receive size supported */
90 int smbd_max_fragmented_recv_size = 1024 * 1024;
91
92 /* The maximum single-message size which can be received */
93 int smbd_max_receive_size = 8192;
94
95 /* The timeout to initiate send of a keepalive message on idle */
96 int smbd_keep_alive_interval = 120;
97
98 /*
99 * User configurable initial values for RDMA transport
100 * The actual values used may be lower and are limited to hardware capabilities
101 */
102 /* Default maximum number of SGEs in a RDMA write/read */
103 int smbd_max_frmr_depth = 2048;
104
105 /* If payload is less than this byte, use RDMA send/recv not read/write */
106 int rdma_readwrite_threshold = 4096;
107
108 /* Transport logging functions
109 * Logging are defined as classes. They can be OR'ed to define the actual
110 * logging level via module parameter smbd_logging_class
111 * e.g. cifs.smbd_logging_class=0xa0 will log all log_rdma_recv() and
112 * log_rdma_event()
113 */
114 #define LOG_OUTGOING 0x1
115 #define LOG_INCOMING 0x2
116 #define LOG_READ 0x4
117 #define LOG_WRITE 0x8
118 #define LOG_RDMA_SEND 0x10
119 #define LOG_RDMA_RECV 0x20
120 #define LOG_KEEP_ALIVE 0x40
121 #define LOG_RDMA_EVENT 0x80
122 #define LOG_RDMA_MR 0x100
123 static unsigned int smbd_logging_class;
124 module_param(smbd_logging_class, uint, 0644);
125 MODULE_PARM_DESC(smbd_logging_class,
126 "Logging class for SMBD transport 0x0 to 0x100");
127
128 #define ERR 0x0
129 #define INFO 0x1
130 static unsigned int smbd_logging_level = ERR;
131 module_param(smbd_logging_level, uint, 0644);
132 MODULE_PARM_DESC(smbd_logging_level,
133 "Logging level for SMBD transport, 0 (default): error, 1: info");
134
135 #define log_rdma(level, class, fmt, args...) \
136 do { \
137 if (level <= smbd_logging_level || class & smbd_logging_class) \
138 cifs_dbg(VFS, "%s:%d " fmt, __func__, __LINE__, ##args);\
139 } while (0)
140
141 #define log_outgoing(level, fmt, args...) \
142 log_rdma(level, LOG_OUTGOING, fmt, ##args)
143 #define log_incoming(level, fmt, args...) \
144 log_rdma(level, LOG_INCOMING, fmt, ##args)
145 #define log_read(level, fmt, args...) log_rdma(level, LOG_READ, fmt, ##args)
146 #define log_write(level, fmt, args...) log_rdma(level, LOG_WRITE, fmt, ##args)
147 #define log_rdma_send(level, fmt, args...) \
148 log_rdma(level, LOG_RDMA_SEND, fmt, ##args)
149 #define log_rdma_recv(level, fmt, args...) \
150 log_rdma(level, LOG_RDMA_RECV, fmt, ##args)
151 #define log_keep_alive(level, fmt, args...) \
152 log_rdma(level, LOG_KEEP_ALIVE, fmt, ##args)
153 #define log_rdma_event(level, fmt, args...) \
154 log_rdma(level, LOG_RDMA_EVENT, fmt, ##args)
155 #define log_rdma_mr(level, fmt, args...) \
156 log_rdma(level, LOG_RDMA_MR, fmt, ##args)
157
158 static void smbd_disconnect_rdma_work(struct work_struct *work)
159 {
160 struct smbd_connection *info =
161 container_of(work, struct smbd_connection, disconnect_work);
162
163 if (info->transport_status == SMBD_CONNECTED) {
164 info->transport_status = SMBD_DISCONNECTING;
165 rdma_disconnect(info->id);
166 }
167 }
168
169 static void smbd_disconnect_rdma_connection(struct smbd_connection *info)
170 {
171 queue_work(info->workqueue, &info->disconnect_work);
172 }
173
174 /* Upcall from RDMA CM */
175 static int smbd_conn_upcall(
176 struct rdma_cm_id *id, struct rdma_cm_event *event)
177 {
178 struct smbd_connection *info = id->context;
179
180 log_rdma_event(INFO, "event=%d status=%d\n",
181 event->event, event->status);
182
183 switch (event->event) {
184 case RDMA_CM_EVENT_ADDR_RESOLVED:
185 case RDMA_CM_EVENT_ROUTE_RESOLVED:
186 info->ri_rc = 0;
187 complete(&info->ri_done);
188 break;
189
190 case RDMA_CM_EVENT_ADDR_ERROR:
191 info->ri_rc = -EHOSTUNREACH;
192 complete(&info->ri_done);
193 break;
194
195 case RDMA_CM_EVENT_ROUTE_ERROR:
196 info->ri_rc = -ENETUNREACH;
197 complete(&info->ri_done);
198 break;
199
200 case RDMA_CM_EVENT_ESTABLISHED:
201 log_rdma_event(INFO, "connected event=%d\n", event->event);
202 info->transport_status = SMBD_CONNECTED;
203 wake_up_interruptible(&info->conn_wait);
204 break;
205
206 case RDMA_CM_EVENT_CONNECT_ERROR:
207 case RDMA_CM_EVENT_UNREACHABLE:
208 case RDMA_CM_EVENT_REJECTED:
209 log_rdma_event(INFO, "connecting failed event=%d\n", event->event);
210 info->transport_status = SMBD_DISCONNECTED;
211 wake_up_interruptible(&info->conn_wait);
212 break;
213
214 case RDMA_CM_EVENT_DEVICE_REMOVAL:
215 case RDMA_CM_EVENT_DISCONNECTED:
216 /* This happenes when we fail the negotiation */
217 if (info->transport_status == SMBD_NEGOTIATE_FAILED) {
218 info->transport_status = SMBD_DISCONNECTED;
219 wake_up(&info->conn_wait);
220 break;
221 }
222
223 info->transport_status = SMBD_DISCONNECTED;
224 wake_up_interruptible(&info->disconn_wait);
225 wake_up_interruptible(&info->wait_reassembly_queue);
226 wake_up_interruptible_all(&info->wait_send_queue);
227 break;
228
229 default:
230 break;
231 }
232
233 return 0;
234 }
235
236 /* Upcall from RDMA QP */
237 static void
238 smbd_qp_async_error_upcall(struct ib_event *event, void *context)
239 {
240 struct smbd_connection *info = context;
241
242 log_rdma_event(ERR, "%s on device %s info %p\n",
243 ib_event_msg(event->event), event->device->name, info);
244
245 switch (event->event) {
246 case IB_EVENT_CQ_ERR:
247 case IB_EVENT_QP_FATAL:
248 smbd_disconnect_rdma_connection(info);
249 break;
250
251 default:
252 break;
253 }
254 }
255
256 static inline void *smbd_request_payload(struct smbd_request *request)
257 {
258 return (void *)request->packet;
259 }
260
261 static inline void *smbd_response_payload(struct smbd_response *response)
262 {
263 return (void *)response->packet;
264 }
265
266 /* Called when a RDMA send is done */
267 static void send_done(struct ib_cq *cq, struct ib_wc *wc)
268 {
269 int i;
270 struct smbd_request *request =
271 container_of(wc->wr_cqe, struct smbd_request, cqe);
272
273 log_rdma_send(INFO, "smbd_request %p completed wc->status=%d\n",
274 request, wc->status);
275
276 if (wc->status != IB_WC_SUCCESS || wc->opcode != IB_WC_SEND) {
277 log_rdma_send(ERR, "wc->status=%d wc->opcode=%d\n",
278 wc->status, wc->opcode);
279 smbd_disconnect_rdma_connection(request->info);
280 }
281
282 for (i = 0; i < request->num_sge; i++)
283 ib_dma_unmap_single(request->info->id->device,
284 request->sge[i].addr,
285 request->sge[i].length,
286 DMA_TO_DEVICE);
287
288 if (atomic_dec_and_test(&request->info->send_pending))
289 wake_up(&request->info->wait_send_pending);
290
291 wake_up(&request->info->wait_post_send);
292
293 mempool_free(request, request->info->request_mempool);
294 }
295
296 static void dump_smbd_negotiate_resp(struct smbd_negotiate_resp *resp)
297 {
298 log_rdma_event(INFO, "resp message min_version %u max_version %u negotiated_version %u credits_requested %u credits_granted %u status %u max_readwrite_size %u preferred_send_size %u max_receive_size %u max_fragmented_size %u\n",
299 resp->min_version, resp->max_version,
300 resp->negotiated_version, resp->credits_requested,
301 resp->credits_granted, resp->status,
302 resp->max_readwrite_size, resp->preferred_send_size,
303 resp->max_receive_size, resp->max_fragmented_size);
304 }
305
306 /*
307 * Process a negotiation response message, according to [MS-SMBD]3.1.5.7
308 * response, packet_length: the negotiation response message
309 * return value: true if negotiation is a success, false if failed
310 */
311 static bool process_negotiation_response(
312 struct smbd_response *response, int packet_length)
313 {
314 struct smbd_connection *info = response->info;
315 struct smbd_negotiate_resp *packet = smbd_response_payload(response);
316
317 if (packet_length < sizeof(struct smbd_negotiate_resp)) {
318 log_rdma_event(ERR,
319 "error: packet_length=%d\n", packet_length);
320 return false;
321 }
322
323 if (le16_to_cpu(packet->negotiated_version) != SMBD_V1) {
324 log_rdma_event(ERR, "error: negotiated_version=%x\n",
325 le16_to_cpu(packet->negotiated_version));
326 return false;
327 }
328 info->protocol = le16_to_cpu(packet->negotiated_version);
329
330 if (packet->credits_requested == 0) {
331 log_rdma_event(ERR, "error: credits_requested==0\n");
332 return false;
333 }
334 info->receive_credit_target = le16_to_cpu(packet->credits_requested);
335
336 if (packet->credits_granted == 0) {
337 log_rdma_event(ERR, "error: credits_granted==0\n");
338 return false;
339 }
340 atomic_set(&info->send_credits, le16_to_cpu(packet->credits_granted));
341
342 atomic_set(&info->receive_credits, 0);
343
344 if (le32_to_cpu(packet->preferred_send_size) > info->max_receive_size) {
345 log_rdma_event(ERR, "error: preferred_send_size=%d\n",
346 le32_to_cpu(packet->preferred_send_size));
347 return false;
348 }
349 info->max_receive_size = le32_to_cpu(packet->preferred_send_size);
350
351 if (le32_to_cpu(packet->max_receive_size) < SMBD_MIN_RECEIVE_SIZE) {
352 log_rdma_event(ERR, "error: max_receive_size=%d\n",
353 le32_to_cpu(packet->max_receive_size));
354 return false;
355 }
356 info->max_send_size = min_t(int, info->max_send_size,
357 le32_to_cpu(packet->max_receive_size));
358
359 if (le32_to_cpu(packet->max_fragmented_size) <
360 SMBD_MIN_FRAGMENTED_SIZE) {
361 log_rdma_event(ERR, "error: max_fragmented_size=%d\n",
362 le32_to_cpu(packet->max_fragmented_size));
363 return false;
364 }
365 info->max_fragmented_send_size =
366 le32_to_cpu(packet->max_fragmented_size);
367 info->rdma_readwrite_threshold =
368 rdma_readwrite_threshold > info->max_fragmented_send_size ?
369 info->max_fragmented_send_size :
370 rdma_readwrite_threshold;
371
372
373 info->max_readwrite_size = min_t(u32,
374 le32_to_cpu(packet->max_readwrite_size),
375 info->max_frmr_depth * PAGE_SIZE);
376 info->max_frmr_depth = info->max_readwrite_size / PAGE_SIZE;
377
378 return true;
379 }
380
381 static void smbd_post_send_credits(struct work_struct *work)
382 {
383 int ret = 0;
384 int use_receive_queue = 1;
385 int rc;
386 struct smbd_response *response;
387 struct smbd_connection *info =
388 container_of(work, struct smbd_connection,
389 post_send_credits_work);
390
391 if (info->transport_status != SMBD_CONNECTED) {
392 wake_up(&info->wait_receive_queues);
393 return;
394 }
395
396 if (info->receive_credit_target >
397 atomic_read(&info->receive_credits)) {
398 while (true) {
399 if (use_receive_queue)
400 response = get_receive_buffer(info);
401 else
402 response = get_empty_queue_buffer(info);
403 if (!response) {
404 /* now switch to emtpy packet queue */
405 if (use_receive_queue) {
406 use_receive_queue = 0;
407 continue;
408 } else
409 break;
410 }
411
412 response->type = SMBD_TRANSFER_DATA;
413 response->first_segment = false;
414 rc = smbd_post_recv(info, response);
415 if (rc) {
416 log_rdma_recv(ERR,
417 "post_recv failed rc=%d\n", rc);
418 put_receive_buffer(info, response);
419 break;
420 }
421
422 ret++;
423 }
424 }
425
426 spin_lock(&info->lock_new_credits_offered);
427 info->new_credits_offered += ret;
428 spin_unlock(&info->lock_new_credits_offered);
429
430 /* Promptly send an immediate packet as defined in [MS-SMBD] 3.1.1.1 */
431 info->send_immediate = true;
432 if (atomic_read(&info->receive_credits) <
433 info->receive_credit_target - 1) {
434 if (info->keep_alive_requested == KEEP_ALIVE_PENDING ||
435 info->send_immediate) {
436 log_keep_alive(INFO, "send an empty message\n");
437 smbd_post_send_empty(info);
438 }
439 }
440 }
441
442 /* Called from softirq, when recv is done */
443 static void recv_done(struct ib_cq *cq, struct ib_wc *wc)
444 {
445 struct smbd_data_transfer *data_transfer;
446 struct smbd_response *response =
447 container_of(wc->wr_cqe, struct smbd_response, cqe);
448 struct smbd_connection *info = response->info;
449 int data_length = 0;
450
451 log_rdma_recv(INFO, "response=%p type=%d wc status=%d wc opcode %d byte_len=%d pkey_index=%x\n",
452 response, response->type, wc->status, wc->opcode,
453 wc->byte_len, wc->pkey_index);
454
455 if (wc->status != IB_WC_SUCCESS || wc->opcode != IB_WC_RECV) {
456 log_rdma_recv(INFO, "wc->status=%d opcode=%d\n",
457 wc->status, wc->opcode);
458 smbd_disconnect_rdma_connection(info);
459 goto error;
460 }
461
462 ib_dma_sync_single_for_cpu(
463 wc->qp->device,
464 response->sge.addr,
465 response->sge.length,
466 DMA_FROM_DEVICE);
467
468 switch (response->type) {
469 /* SMBD negotiation response */
470 case SMBD_NEGOTIATE_RESP:
471 dump_smbd_negotiate_resp(smbd_response_payload(response));
472 info->full_packet_received = true;
473 info->negotiate_done =
474 process_negotiation_response(response, wc->byte_len);
475 complete(&info->negotiate_completion);
476 break;
477
478 /* SMBD data transfer packet */
479 case SMBD_TRANSFER_DATA:
480 data_transfer = smbd_response_payload(response);
481 data_length = le32_to_cpu(data_transfer->data_length);
482
483 /*
484 * If this is a packet with data playload place the data in
485 * reassembly queue and wake up the reading thread
486 */
487 if (data_length) {
488 if (info->full_packet_received)
489 response->first_segment = true;
490
491 if (le32_to_cpu(data_transfer->remaining_data_length))
492 info->full_packet_received = false;
493 else
494 info->full_packet_received = true;
495
496 enqueue_reassembly(
497 info,
498 response,
499 data_length);
500 } else
501 put_empty_packet(info, response);
502
503 if (data_length)
504 wake_up_interruptible(&info->wait_reassembly_queue);
505
506 atomic_dec(&info->receive_credits);
507 info->receive_credit_target =
508 le16_to_cpu(data_transfer->credits_requested);
509 if (le16_to_cpu(data_transfer->credits_granted)) {
510 atomic_add(le16_to_cpu(data_transfer->credits_granted),
511 &info->send_credits);
512 /*
513 * We have new send credits granted from remote peer
514 * If any sender is waiting for credits, unblock it
515 */
516 wake_up_interruptible(&info->wait_send_queue);
517 }
518
519 log_incoming(INFO, "data flags %d data_offset %d data_length %d remaining_data_length %d\n",
520 le16_to_cpu(data_transfer->flags),
521 le32_to_cpu(data_transfer->data_offset),
522 le32_to_cpu(data_transfer->data_length),
523 le32_to_cpu(data_transfer->remaining_data_length));
524
525 /* Send a KEEP_ALIVE response right away if requested */
526 info->keep_alive_requested = KEEP_ALIVE_NONE;
527 if (le16_to_cpu(data_transfer->flags) &
528 SMB_DIRECT_RESPONSE_REQUESTED) {
529 info->keep_alive_requested = KEEP_ALIVE_PENDING;
530 }
531
532 return;
533
534 default:
535 log_rdma_recv(ERR,
536 "unexpected response type=%d\n", response->type);
537 }
538
539 error:
540 put_receive_buffer(info, response);
541 }
542
543 static struct rdma_cm_id *smbd_create_id(
544 struct smbd_connection *info,
545 struct sockaddr *dstaddr, int port)
546 {
547 struct rdma_cm_id *id;
548 int rc;
549 __be16 *sport;
550
551 id = rdma_create_id(&init_net, smbd_conn_upcall, info,
552 RDMA_PS_TCP, IB_QPT_RC);
553 if (IS_ERR(id)) {
554 rc = PTR_ERR(id);
555 log_rdma_event(ERR, "rdma_create_id() failed %i\n", rc);
556 return id;
557 }
558
559 if (dstaddr->sa_family == AF_INET6)
560 sport = &((struct sockaddr_in6 *)dstaddr)->sin6_port;
561 else
562 sport = &((struct sockaddr_in *)dstaddr)->sin_port;
563
564 *sport = htons(port);
565
566 init_completion(&info->ri_done);
567 info->ri_rc = -ETIMEDOUT;
568
569 rc = rdma_resolve_addr(id, NULL, (struct sockaddr *)dstaddr,
570 RDMA_RESOLVE_TIMEOUT);
571 if (rc) {
572 log_rdma_event(ERR, "rdma_resolve_addr() failed %i\n", rc);
573 goto out;
574 }
575 wait_for_completion_interruptible_timeout(
576 &info->ri_done, msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT));
577 rc = info->ri_rc;
578 if (rc) {
579 log_rdma_event(ERR, "rdma_resolve_addr() completed %i\n", rc);
580 goto out;
581 }
582
583 info->ri_rc = -ETIMEDOUT;
584 rc = rdma_resolve_route(id, RDMA_RESOLVE_TIMEOUT);
585 if (rc) {
586 log_rdma_event(ERR, "rdma_resolve_route() failed %i\n", rc);
587 goto out;
588 }
589 wait_for_completion_interruptible_timeout(
590 &info->ri_done, msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT));
591 rc = info->ri_rc;
592 if (rc) {
593 log_rdma_event(ERR, "rdma_resolve_route() completed %i\n", rc);
594 goto out;
595 }
596
597 return id;
598
599 out:
600 rdma_destroy_id(id);
601 return ERR_PTR(rc);
602 }
603
604 /*
605 * Test if FRWR (Fast Registration Work Requests) is supported on the device
606 * This implementation requries FRWR on RDMA read/write
607 * return value: true if it is supported
608 */
609 static bool frwr_is_supported(struct ib_device_attr *attrs)
610 {
611 if (!(attrs->device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS))
612 return false;
613 if (attrs->max_fast_reg_page_list_len == 0)
614 return false;
615 return true;
616 }
617
618 static int smbd_ia_open(
619 struct smbd_connection *info,
620 struct sockaddr *dstaddr, int port)
621 {
622 int rc;
623
624 info->id = smbd_create_id(info, dstaddr, port);
625 if (IS_ERR(info->id)) {
626 rc = PTR_ERR(info->id);
627 goto out1;
628 }
629
630 if (!frwr_is_supported(&info->id->device->attrs)) {
631 log_rdma_event(ERR, "Fast Registration Work Requests (FRWR) is not supported\n");
632 log_rdma_event(ERR, "Device capability flags = %llx max_fast_reg_page_list_len = %u\n",
633 info->id->device->attrs.device_cap_flags,
634 info->id->device->attrs.max_fast_reg_page_list_len);
635 rc = -EPROTONOSUPPORT;
636 goto out2;
637 }
638 info->max_frmr_depth = min_t(int,
639 smbd_max_frmr_depth,
640 info->id->device->attrs.max_fast_reg_page_list_len);
641 info->mr_type = IB_MR_TYPE_MEM_REG;
642 if (info->id->device->attrs.device_cap_flags & IB_DEVICE_SG_GAPS_REG)
643 info->mr_type = IB_MR_TYPE_SG_GAPS;
644
645 info->pd = ib_alloc_pd(info->id->device, 0);
646 if (IS_ERR(info->pd)) {
647 rc = PTR_ERR(info->pd);
648 log_rdma_event(ERR, "ib_alloc_pd() returned %d\n", rc);
649 goto out2;
650 }
651
652 return 0;
653
654 out2:
655 rdma_destroy_id(info->id);
656 info->id = NULL;
657
658 out1:
659 return rc;
660 }
661
662 /*
663 * Send a negotiation request message to the peer
664 * The negotiation procedure is in [MS-SMBD] 3.1.5.2 and 3.1.5.3
665 * After negotiation, the transport is connected and ready for
666 * carrying upper layer SMB payload
667 */
668 static int smbd_post_send_negotiate_req(struct smbd_connection *info)
669 {
670 struct ib_send_wr send_wr;
671 int rc = -ENOMEM;
672 struct smbd_request *request;
673 struct smbd_negotiate_req *packet;
674
675 request = mempool_alloc(info->request_mempool, GFP_KERNEL);
676 if (!request)
677 return rc;
678
679 request->info = info;
680
681 packet = smbd_request_payload(request);
682 packet->min_version = cpu_to_le16(SMBD_V1);
683 packet->max_version = cpu_to_le16(SMBD_V1);
684 packet->reserved = 0;
685 packet->credits_requested = cpu_to_le16(info->send_credit_target);
686 packet->preferred_send_size = cpu_to_le32(info->max_send_size);
687 packet->max_receive_size = cpu_to_le32(info->max_receive_size);
688 packet->max_fragmented_size =
689 cpu_to_le32(info->max_fragmented_recv_size);
690
691 request->num_sge = 1;
692 request->sge[0].addr = ib_dma_map_single(
693 info->id->device, (void *)packet,
694 sizeof(*packet), DMA_TO_DEVICE);
695 if (ib_dma_mapping_error(info->id->device, request->sge[0].addr)) {
696 rc = -EIO;
697 goto dma_mapping_failed;
698 }
699
700 request->sge[0].length = sizeof(*packet);
701 request->sge[0].lkey = info->pd->local_dma_lkey;
702
703 ib_dma_sync_single_for_device(
704 info->id->device, request->sge[0].addr,
705 request->sge[0].length, DMA_TO_DEVICE);
706
707 request->cqe.done = send_done;
708
709 send_wr.next = NULL;
710 send_wr.wr_cqe = &request->cqe;
711 send_wr.sg_list = request->sge;
712 send_wr.num_sge = request->num_sge;
713 send_wr.opcode = IB_WR_SEND;
714 send_wr.send_flags = IB_SEND_SIGNALED;
715
716 log_rdma_send(INFO, "sge addr=%llx length=%x lkey=%x\n",
717 request->sge[0].addr,
718 request->sge[0].length, request->sge[0].lkey);
719
720 atomic_inc(&info->send_pending);
721 rc = ib_post_send(info->id->qp, &send_wr, NULL);
722 if (!rc)
723 return 0;
724
725 /* if we reach here, post send failed */
726 log_rdma_send(ERR, "ib_post_send failed rc=%d\n", rc);
727 atomic_dec(&info->send_pending);
728 ib_dma_unmap_single(info->id->device, request->sge[0].addr,
729 request->sge[0].length, DMA_TO_DEVICE);
730
731 smbd_disconnect_rdma_connection(info);
732
733 dma_mapping_failed:
734 mempool_free(request, info->request_mempool);
735 return rc;
736 }
737
738 /*
739 * Extend the credits to remote peer
740 * This implements [MS-SMBD] 3.1.5.9
741 * The idea is that we should extend credits to remote peer as quickly as
742 * it's allowed, to maintain data flow. We allocate as much receive
743 * buffer as possible, and extend the receive credits to remote peer
744 * return value: the new credtis being granted.
745 */
746 static int manage_credits_prior_sending(struct smbd_connection *info)
747 {
748 int new_credits;
749
750 spin_lock(&info->lock_new_credits_offered);
751 new_credits = info->new_credits_offered;
752 info->new_credits_offered = 0;
753 spin_unlock(&info->lock_new_credits_offered);
754
755 return new_credits;
756 }
757
758 /*
759 * Check if we need to send a KEEP_ALIVE message
760 * The idle connection timer triggers a KEEP_ALIVE message when expires
761 * SMB_DIRECT_RESPONSE_REQUESTED is set in the message flag to have peer send
762 * back a response.
763 * return value:
764 * 1 if SMB_DIRECT_RESPONSE_REQUESTED needs to be set
765 * 0: otherwise
766 */
767 static int manage_keep_alive_before_sending(struct smbd_connection *info)
768 {
769 if (info->keep_alive_requested == KEEP_ALIVE_PENDING) {
770 info->keep_alive_requested = KEEP_ALIVE_SENT;
771 return 1;
772 }
773 return 0;
774 }
775
776 /* Post the send request */
777 static int smbd_post_send(struct smbd_connection *info,
778 struct smbd_request *request)
779 {
780 struct ib_send_wr send_wr;
781 int rc, i;
782
783 for (i = 0; i < request->num_sge; i++) {
784 log_rdma_send(INFO,
785 "rdma_request sge[%d] addr=%llu length=%u\n",
786 i, request->sge[i].addr, request->sge[i].length);
787 ib_dma_sync_single_for_device(
788 info->id->device,
789 request->sge[i].addr,
790 request->sge[i].length,
791 DMA_TO_DEVICE);
792 }
793
794 request->cqe.done = send_done;
795
796 send_wr.next = NULL;
797 send_wr.wr_cqe = &request->cqe;
798 send_wr.sg_list = request->sge;
799 send_wr.num_sge = request->num_sge;
800 send_wr.opcode = IB_WR_SEND;
801 send_wr.send_flags = IB_SEND_SIGNALED;
802
803 rc = ib_post_send(info->id->qp, &send_wr, NULL);
804 if (rc) {
805 log_rdma_send(ERR, "ib_post_send failed rc=%d\n", rc);
806 smbd_disconnect_rdma_connection(info);
807 rc = -EAGAIN;
808 } else
809 /* Reset timer for idle connection after packet is sent */
810 mod_delayed_work(info->workqueue, &info->idle_timer_work,
811 info->keep_alive_interval*HZ);
812
813 return rc;
814 }
815
816 static int smbd_post_send_sgl(struct smbd_connection *info,
817 struct scatterlist *sgl, int data_length, int remaining_data_length)
818 {
819 int num_sgs;
820 int i, rc;
821 int header_length;
822 struct smbd_request *request;
823 struct smbd_data_transfer *packet;
824 int new_credits;
825 struct scatterlist *sg;
826
827 wait_credit:
828 /* Wait for send credits. A SMBD packet needs one credit */
829 rc = wait_event_interruptible(info->wait_send_queue,
830 atomic_read(&info->send_credits) > 0 ||
831 info->transport_status != SMBD_CONNECTED);
832 if (rc)
833 goto err_wait_credit;
834
835 if (info->transport_status != SMBD_CONNECTED) {
836 log_outgoing(ERR, "disconnected not sending on wait_credit\n");
837 rc = -EAGAIN;
838 goto err_wait_credit;
839 }
840 if (unlikely(atomic_dec_return(&info->send_credits) < 0)) {
841 atomic_inc(&info->send_credits);
842 goto wait_credit;
843 }
844
845 wait_send_queue:
846 wait_event(info->wait_post_send,
847 atomic_read(&info->send_pending) < info->send_credit_target ||
848 info->transport_status != SMBD_CONNECTED);
849
850 if (info->transport_status != SMBD_CONNECTED) {
851 log_outgoing(ERR, "disconnected not sending on wait_send_queue\n");
852 rc = -EAGAIN;
853 goto err_wait_send_queue;
854 }
855
856 if (unlikely(atomic_inc_return(&info->send_pending) >
857 info->send_credit_target)) {
858 atomic_dec(&info->send_pending);
859 goto wait_send_queue;
860 }
861
862 request = mempool_alloc(info->request_mempool, GFP_KERNEL);
863 if (!request) {
864 rc = -ENOMEM;
865 goto err_alloc;
866 }
867
868 request->info = info;
869
870 /* Fill in the packet header */
871 packet = smbd_request_payload(request);
872 packet->credits_requested = cpu_to_le16(info->send_credit_target);
873
874 new_credits = manage_credits_prior_sending(info);
875 atomic_add(new_credits, &info->receive_credits);
876 packet->credits_granted = cpu_to_le16(new_credits);
877
878 info->send_immediate = false;
879
880 packet->flags = 0;
881 if (manage_keep_alive_before_sending(info))
882 packet->flags |= cpu_to_le16(SMB_DIRECT_RESPONSE_REQUESTED);
883
884 packet->reserved = 0;
885 if (!data_length)
886 packet->data_offset = 0;
887 else
888 packet->data_offset = cpu_to_le32(24);
889 packet->data_length = cpu_to_le32(data_length);
890 packet->remaining_data_length = cpu_to_le32(remaining_data_length);
891 packet->padding = 0;
892
893 log_outgoing(INFO, "credits_requested=%d credits_granted=%d data_offset=%d data_length=%d remaining_data_length=%d\n",
894 le16_to_cpu(packet->credits_requested),
895 le16_to_cpu(packet->credits_granted),
896 le32_to_cpu(packet->data_offset),
897 le32_to_cpu(packet->data_length),
898 le32_to_cpu(packet->remaining_data_length));
899
900 /* Map the packet to DMA */
901 header_length = sizeof(struct smbd_data_transfer);
902 /* If this is a packet without payload, don't send padding */
903 if (!data_length)
904 header_length = offsetof(struct smbd_data_transfer, padding);
905
906 request->num_sge = 1;
907 request->sge[0].addr = ib_dma_map_single(info->id->device,
908 (void *)packet,
909 header_length,
910 DMA_TO_DEVICE);
911 if (ib_dma_mapping_error(info->id->device, request->sge[0].addr)) {
912 rc = -EIO;
913 request->sge[0].addr = 0;
914 goto err_dma;
915 }
916
917 request->sge[0].length = header_length;
918 request->sge[0].lkey = info->pd->local_dma_lkey;
919
920 /* Fill in the packet data payload */
921 num_sgs = sgl ? sg_nents(sgl) : 0;
922 for_each_sg(sgl, sg, num_sgs, i) {
923 request->sge[i+1].addr =
924 ib_dma_map_page(info->id->device, sg_page(sg),
925 sg->offset, sg->length, DMA_TO_DEVICE);
926 if (ib_dma_mapping_error(
927 info->id->device, request->sge[i+1].addr)) {
928 rc = -EIO;
929 request->sge[i+1].addr = 0;
930 goto err_dma;
931 }
932 request->sge[i+1].length = sg->length;
933 request->sge[i+1].lkey = info->pd->local_dma_lkey;
934 request->num_sge++;
935 }
936
937 rc = smbd_post_send(info, request);
938 if (!rc)
939 return 0;
940
941 err_dma:
942 for (i = 0; i < request->num_sge; i++)
943 if (request->sge[i].addr)
944 ib_dma_unmap_single(info->id->device,
945 request->sge[i].addr,
946 request->sge[i].length,
947 DMA_TO_DEVICE);
948 mempool_free(request, info->request_mempool);
949
950 /* roll back receive credits and credits to be offered */
951 spin_lock(&info->lock_new_credits_offered);
952 info->new_credits_offered += new_credits;
953 spin_unlock(&info->lock_new_credits_offered);
954 atomic_sub(new_credits, &info->receive_credits);
955
956 err_alloc:
957 if (atomic_dec_and_test(&info->send_pending))
958 wake_up(&info->wait_send_pending);
959
960 err_wait_send_queue:
961 /* roll back send credits and pending */
962 atomic_inc(&info->send_credits);
963
964 err_wait_credit:
965 return rc;
966 }
967
968 /*
969 * Send a page
970 * page: the page to send
971 * offset: offset in the page to send
972 * size: length in the page to send
973 * remaining_data_length: remaining data to send in this payload
974 */
975 static int smbd_post_send_page(struct smbd_connection *info, struct page *page,
976 unsigned long offset, size_t size, int remaining_data_length)
977 {
978 struct scatterlist sgl;
979
980 sg_init_table(&sgl, 1);
981 sg_set_page(&sgl, page, size, offset);
982
983 return smbd_post_send_sgl(info, &sgl, size, remaining_data_length);
984 }
985
986 /*
987 * Send an empty message
988 * Empty message is used to extend credits to peer to for keep live
989 * while there is no upper layer payload to send at the time
990 */
991 static int smbd_post_send_empty(struct smbd_connection *info)
992 {
993 info->count_send_empty++;
994 return smbd_post_send_sgl(info, NULL, 0, 0);
995 }
996
997 /*
998 * Send a data buffer
999 * iov: the iov array describing the data buffers
1000 * n_vec: number of iov array
1001 * remaining_data_length: remaining data to send following this packet
1002 * in segmented SMBD packet
1003 */
1004 static int smbd_post_send_data(
1005 struct smbd_connection *info, struct kvec *iov, int n_vec,
1006 int remaining_data_length)
1007 {
1008 int i;
1009 u32 data_length = 0;
1010 struct scatterlist sgl[SMBDIRECT_MAX_SGE];
1011
1012 if (n_vec > SMBDIRECT_MAX_SGE) {
1013 cifs_dbg(VFS, "Can't fit data to SGL, n_vec=%d\n", n_vec);
1014 return -EINVAL;
1015 }
1016
1017 sg_init_table(sgl, n_vec);
1018 for (i = 0; i < n_vec; i++) {
1019 data_length += iov[i].iov_len;
1020 sg_set_buf(&sgl[i], iov[i].iov_base, iov[i].iov_len);
1021 }
1022
1023 return smbd_post_send_sgl(info, sgl, data_length, remaining_data_length);
1024 }
1025
1026 /*
1027 * Post a receive request to the transport
1028 * The remote peer can only send data when a receive request is posted
1029 * The interaction is controlled by send/receive credit system
1030 */
1031 static int smbd_post_recv(
1032 struct smbd_connection *info, struct smbd_response *response)
1033 {
1034 struct ib_recv_wr recv_wr;
1035 int rc = -EIO;
1036
1037 response->sge.addr = ib_dma_map_single(
1038 info->id->device, response->packet,
1039 info->max_receive_size, DMA_FROM_DEVICE);
1040 if (ib_dma_mapping_error(info->id->device, response->sge.addr))
1041 return rc;
1042
1043 response->sge.length = info->max_receive_size;
1044 response->sge.lkey = info->pd->local_dma_lkey;
1045
1046 response->cqe.done = recv_done;
1047
1048 recv_wr.wr_cqe = &response->cqe;
1049 recv_wr.next = NULL;
1050 recv_wr.sg_list = &response->sge;
1051 recv_wr.num_sge = 1;
1052
1053 rc = ib_post_recv(info->id->qp, &recv_wr, NULL);
1054 if (rc) {
1055 ib_dma_unmap_single(info->id->device, response->sge.addr,
1056 response->sge.length, DMA_FROM_DEVICE);
1057 smbd_disconnect_rdma_connection(info);
1058 log_rdma_recv(ERR, "ib_post_recv failed rc=%d\n", rc);
1059 }
1060
1061 return rc;
1062 }
1063
1064 /* Perform SMBD negotiate according to [MS-SMBD] 3.1.5.2 */
1065 static int smbd_negotiate(struct smbd_connection *info)
1066 {
1067 int rc;
1068 struct smbd_response *response = get_receive_buffer(info);
1069
1070 response->type = SMBD_NEGOTIATE_RESP;
1071 rc = smbd_post_recv(info, response);
1072 log_rdma_event(INFO, "smbd_post_recv rc=%d iov.addr=%llx iov.length=%x iov.lkey=%x\n",
1073 rc, response->sge.addr,
1074 response->sge.length, response->sge.lkey);
1075 if (rc)
1076 return rc;
1077
1078 init_completion(&info->negotiate_completion);
1079 info->negotiate_done = false;
1080 rc = smbd_post_send_negotiate_req(info);
1081 if (rc)
1082 return rc;
1083
1084 rc = wait_for_completion_interruptible_timeout(
1085 &info->negotiate_completion, SMBD_NEGOTIATE_TIMEOUT * HZ);
1086 log_rdma_event(INFO, "wait_for_completion_timeout rc=%d\n", rc);
1087
1088 if (info->negotiate_done)
1089 return 0;
1090
1091 if (rc == 0)
1092 rc = -ETIMEDOUT;
1093 else if (rc == -ERESTARTSYS)
1094 rc = -EINTR;
1095 else
1096 rc = -ENOTCONN;
1097
1098 return rc;
1099 }
1100
1101 static void put_empty_packet(
1102 struct smbd_connection *info, struct smbd_response *response)
1103 {
1104 spin_lock(&info->empty_packet_queue_lock);
1105 list_add_tail(&response->list, &info->empty_packet_queue);
1106 info->count_empty_packet_queue++;
1107 spin_unlock(&info->empty_packet_queue_lock);
1108
1109 queue_work(info->workqueue, &info->post_send_credits_work);
1110 }
1111
1112 /*
1113 * Implement Connection.FragmentReassemblyBuffer defined in [MS-SMBD] 3.1.1.1
1114 * This is a queue for reassembling upper layer payload and present to upper
1115 * layer. All the inncoming payload go to the reassembly queue, regardless of
1116 * if reassembly is required. The uuper layer code reads from the queue for all
1117 * incoming payloads.
1118 * Put a received packet to the reassembly queue
1119 * response: the packet received
1120 * data_length: the size of payload in this packet
1121 */
1122 static void enqueue_reassembly(
1123 struct smbd_connection *info,
1124 struct smbd_response *response,
1125 int data_length)
1126 {
1127 spin_lock(&info->reassembly_queue_lock);
1128 list_add_tail(&response->list, &info->reassembly_queue);
1129 info->reassembly_queue_length++;
1130 /*
1131 * Make sure reassembly_data_length is updated after list and
1132 * reassembly_queue_length are updated. On the dequeue side
1133 * reassembly_data_length is checked without a lock to determine
1134 * if reassembly_queue_length and list is up to date
1135 */
1136 virt_wmb();
1137 info->reassembly_data_length += data_length;
1138 spin_unlock(&info->reassembly_queue_lock);
1139 info->count_reassembly_queue++;
1140 info->count_enqueue_reassembly_queue++;
1141 }
1142
1143 /*
1144 * Get the first entry at the front of reassembly queue
1145 * Caller is responsible for locking
1146 * return value: the first entry if any, NULL if queue is empty
1147 */
1148 static struct smbd_response *_get_first_reassembly(struct smbd_connection *info)
1149 {
1150 struct smbd_response *ret = NULL;
1151
1152 if (!list_empty(&info->reassembly_queue)) {
1153 ret = list_first_entry(
1154 &info->reassembly_queue,
1155 struct smbd_response, list);
1156 }
1157 return ret;
1158 }
1159
1160 static struct smbd_response *get_empty_queue_buffer(
1161 struct smbd_connection *info)
1162 {
1163 struct smbd_response *ret = NULL;
1164 unsigned long flags;
1165
1166 spin_lock_irqsave(&info->empty_packet_queue_lock, flags);
1167 if (!list_empty(&info->empty_packet_queue)) {
1168 ret = list_first_entry(
1169 &info->empty_packet_queue,
1170 struct smbd_response, list);
1171 list_del(&ret->list);
1172 info->count_empty_packet_queue--;
1173 }
1174 spin_unlock_irqrestore(&info->empty_packet_queue_lock, flags);
1175
1176 return ret;
1177 }
1178
1179 /*
1180 * Get a receive buffer
1181 * For each remote send, we need to post a receive. The receive buffers are
1182 * pre-allocated in advance.
1183 * return value: the receive buffer, NULL if none is available
1184 */
1185 static struct smbd_response *get_receive_buffer(struct smbd_connection *info)
1186 {
1187 struct smbd_response *ret = NULL;
1188 unsigned long flags;
1189
1190 spin_lock_irqsave(&info->receive_queue_lock, flags);
1191 if (!list_empty(&info->receive_queue)) {
1192 ret = list_first_entry(
1193 &info->receive_queue,
1194 struct smbd_response, list);
1195 list_del(&ret->list);
1196 info->count_receive_queue--;
1197 info->count_get_receive_buffer++;
1198 }
1199 spin_unlock_irqrestore(&info->receive_queue_lock, flags);
1200
1201 return ret;
1202 }
1203
1204 /*
1205 * Return a receive buffer
1206 * Upon returning of a receive buffer, we can post new receive and extend
1207 * more receive credits to remote peer. This is done immediately after a
1208 * receive buffer is returned.
1209 */
1210 static void put_receive_buffer(
1211 struct smbd_connection *info, struct smbd_response *response)
1212 {
1213 unsigned long flags;
1214
1215 ib_dma_unmap_single(info->id->device, response->sge.addr,
1216 response->sge.length, DMA_FROM_DEVICE);
1217
1218 spin_lock_irqsave(&info->receive_queue_lock, flags);
1219 list_add_tail(&response->list, &info->receive_queue);
1220 info->count_receive_queue++;
1221 info->count_put_receive_buffer++;
1222 spin_unlock_irqrestore(&info->receive_queue_lock, flags);
1223
1224 queue_work(info->workqueue, &info->post_send_credits_work);
1225 }
1226
1227 /* Preallocate all receive buffer on transport establishment */
1228 static int allocate_receive_buffers(struct smbd_connection *info, int num_buf)
1229 {
1230 int i;
1231 struct smbd_response *response;
1232
1233 INIT_LIST_HEAD(&info->reassembly_queue);
1234 spin_lock_init(&info->reassembly_queue_lock);
1235 info->reassembly_data_length = 0;
1236 info->reassembly_queue_length = 0;
1237
1238 INIT_LIST_HEAD(&info->receive_queue);
1239 spin_lock_init(&info->receive_queue_lock);
1240 info->count_receive_queue = 0;
1241
1242 INIT_LIST_HEAD(&info->empty_packet_queue);
1243 spin_lock_init(&info->empty_packet_queue_lock);
1244 info->count_empty_packet_queue = 0;
1245
1246 init_waitqueue_head(&info->wait_receive_queues);
1247
1248 for (i = 0; i < num_buf; i++) {
1249 response = mempool_alloc(info->response_mempool, GFP_KERNEL);
1250 if (!response)
1251 goto allocate_failed;
1252
1253 response->info = info;
1254 list_add_tail(&response->list, &info->receive_queue);
1255 info->count_receive_queue++;
1256 }
1257
1258 return 0;
1259
1260 allocate_failed:
1261 while (!list_empty(&info->receive_queue)) {
1262 response = list_first_entry(
1263 &info->receive_queue,
1264 struct smbd_response, list);
1265 list_del(&response->list);
1266 info->count_receive_queue--;
1267
1268 mempool_free(response, info->response_mempool);
1269 }
1270 return -ENOMEM;
1271 }
1272
1273 static void destroy_receive_buffers(struct smbd_connection *info)
1274 {
1275 struct smbd_response *response;
1276
1277 while ((response = get_receive_buffer(info)))
1278 mempool_free(response, info->response_mempool);
1279
1280 while ((response = get_empty_queue_buffer(info)))
1281 mempool_free(response, info->response_mempool);
1282 }
1283
1284 /* Implement idle connection timer [MS-SMBD] 3.1.6.2 */
1285 static void idle_connection_timer(struct work_struct *work)
1286 {
1287 struct smbd_connection *info = container_of(
1288 work, struct smbd_connection,
1289 idle_timer_work.work);
1290
1291 if (info->keep_alive_requested != KEEP_ALIVE_NONE) {
1292 log_keep_alive(ERR,
1293 "error status info->keep_alive_requested=%d\n",
1294 info->keep_alive_requested);
1295 smbd_disconnect_rdma_connection(info);
1296 return;
1297 }
1298
1299 log_keep_alive(INFO, "about to send an empty idle message\n");
1300 smbd_post_send_empty(info);
1301
1302 /* Setup the next idle timeout work */
1303 queue_delayed_work(info->workqueue, &info->idle_timer_work,
1304 info->keep_alive_interval*HZ);
1305 }
1306
1307 /*
1308 * Destroy the transport and related RDMA and memory resources
1309 * Need to go through all the pending counters and make sure on one is using
1310 * the transport while it is destroyed
1311 */
1312 void smbd_destroy(struct TCP_Server_Info *server)
1313 {
1314 struct smbd_connection *info = server->smbd_conn;
1315 struct smbd_response *response;
1316 unsigned long flags;
1317
1318 if (!info) {
1319 log_rdma_event(INFO, "rdma session already destroyed\n");
1320 return;
1321 }
1322
1323 log_rdma_event(INFO, "destroying rdma session\n");
1324 if (info->transport_status != SMBD_DISCONNECTED) {
1325 rdma_disconnect(server->smbd_conn->id);
1326 log_rdma_event(INFO, "wait for transport being disconnected\n");
1327 wait_event_interruptible(
1328 info->disconn_wait,
1329 info->transport_status == SMBD_DISCONNECTED);
1330 }
1331
1332 log_rdma_event(INFO, "destroying qp\n");
1333 ib_drain_qp(info->id->qp);
1334 rdma_destroy_qp(info->id);
1335
1336 log_rdma_event(INFO, "cancelling idle timer\n");
1337 cancel_delayed_work_sync(&info->idle_timer_work);
1338
1339 log_rdma_event(INFO, "wait for all send posted to IB to finish\n");
1340 wait_event(info->wait_send_pending,
1341 atomic_read(&info->send_pending) == 0);
1342
1343 /* It's not posssible for upper layer to get to reassembly */
1344 log_rdma_event(INFO, "drain the reassembly queue\n");
1345 do {
1346 spin_lock_irqsave(&info->reassembly_queue_lock, flags);
1347 response = _get_first_reassembly(info);
1348 if (response) {
1349 list_del(&response->list);
1350 spin_unlock_irqrestore(
1351 &info->reassembly_queue_lock, flags);
1352 put_receive_buffer(info, response);
1353 } else
1354 spin_unlock_irqrestore(
1355 &info->reassembly_queue_lock, flags);
1356 } while (response);
1357 info->reassembly_data_length = 0;
1358
1359 log_rdma_event(INFO, "free receive buffers\n");
1360 wait_event(info->wait_receive_queues,
1361 info->count_receive_queue + info->count_empty_packet_queue
1362 == info->receive_credit_max);
1363 destroy_receive_buffers(info);
1364
1365 /*
1366 * For performance reasons, memory registration and deregistration
1367 * are not locked by srv_mutex. It is possible some processes are
1368 * blocked on transport srv_mutex while holding memory registration.
1369 * Release the transport srv_mutex to allow them to hit the failure
1370 * path when sending data, and then release memory registartions.
1371 */
1372 log_rdma_event(INFO, "freeing mr list\n");
1373 wake_up_interruptible_all(&info->wait_mr);
1374 while (atomic_read(&info->mr_used_count)) {
1375 mutex_unlock(&server->srv_mutex);
1376 msleep(1000);
1377 mutex_lock(&server->srv_mutex);
1378 }
1379 destroy_mr_list(info);
1380
1381 ib_free_cq(info->send_cq);
1382 ib_free_cq(info->recv_cq);
1383 ib_dealloc_pd(info->pd);
1384 rdma_destroy_id(info->id);
1385
1386 /* free mempools */
1387 mempool_destroy(info->request_mempool);
1388 kmem_cache_destroy(info->request_cache);
1389
1390 mempool_destroy(info->response_mempool);
1391 kmem_cache_destroy(info->response_cache);
1392
1393 info->transport_status = SMBD_DESTROYED;
1394
1395 destroy_workqueue(info->workqueue);
1396 log_rdma_event(INFO, "rdma session destroyed\n");
1397 kfree(info);
1398 }
1399
1400 /*
1401 * Reconnect this SMBD connection, called from upper layer
1402 * return value: 0 on success, or actual error code
1403 */
1404 int smbd_reconnect(struct TCP_Server_Info *server)
1405 {
1406 log_rdma_event(INFO, "reconnecting rdma session\n");
1407
1408 if (!server->smbd_conn) {
1409 log_rdma_event(INFO, "rdma session already destroyed\n");
1410 goto create_conn;
1411 }
1412
1413 /*
1414 * This is possible if transport is disconnected and we haven't received
1415 * notification from RDMA, but upper layer has detected timeout
1416 */
1417 if (server->smbd_conn->transport_status == SMBD_CONNECTED) {
1418 log_rdma_event(INFO, "disconnecting transport\n");
1419 smbd_destroy(server);
1420 }
1421
1422 create_conn:
1423 log_rdma_event(INFO, "creating rdma session\n");
1424 server->smbd_conn = smbd_get_connection(
1425 server, (struct sockaddr *) &server->dstaddr);
1426
1427 if (server->smbd_conn)
1428 cifs_dbg(VFS, "RDMA transport re-established\n");
1429
1430 return server->smbd_conn ? 0 : -ENOENT;
1431 }
1432
1433 static void destroy_caches_and_workqueue(struct smbd_connection *info)
1434 {
1435 destroy_receive_buffers(info);
1436 destroy_workqueue(info->workqueue);
1437 mempool_destroy(info->response_mempool);
1438 kmem_cache_destroy(info->response_cache);
1439 mempool_destroy(info->request_mempool);
1440 kmem_cache_destroy(info->request_cache);
1441 }
1442
1443 #define MAX_NAME_LEN 80
1444 static int allocate_caches_and_workqueue(struct smbd_connection *info)
1445 {
1446 char name[MAX_NAME_LEN];
1447 int rc;
1448
1449 scnprintf(name, MAX_NAME_LEN, "smbd_request_%p", info);
1450 info->request_cache =
1451 kmem_cache_create(
1452 name,
1453 sizeof(struct smbd_request) +
1454 sizeof(struct smbd_data_transfer),
1455 0, SLAB_HWCACHE_ALIGN, NULL);
1456 if (!info->request_cache)
1457 return -ENOMEM;
1458
1459 info->request_mempool =
1460 mempool_create(info->send_credit_target, mempool_alloc_slab,
1461 mempool_free_slab, info->request_cache);
1462 if (!info->request_mempool)
1463 goto out1;
1464
1465 scnprintf(name, MAX_NAME_LEN, "smbd_response_%p", info);
1466 info->response_cache =
1467 kmem_cache_create(
1468 name,
1469 sizeof(struct smbd_response) +
1470 info->max_receive_size,
1471 0, SLAB_HWCACHE_ALIGN, NULL);
1472 if (!info->response_cache)
1473 goto out2;
1474
1475 info->response_mempool =
1476 mempool_create(info->receive_credit_max, mempool_alloc_slab,
1477 mempool_free_slab, info->response_cache);
1478 if (!info->response_mempool)
1479 goto out3;
1480
1481 scnprintf(name, MAX_NAME_LEN, "smbd_%p", info);
1482 info->workqueue = create_workqueue(name);
1483 if (!info->workqueue)
1484 goto out4;
1485
1486 rc = allocate_receive_buffers(info, info->receive_credit_max);
1487 if (rc) {
1488 log_rdma_event(ERR, "failed to allocate receive buffers\n");
1489 goto out5;
1490 }
1491
1492 return 0;
1493
1494 out5:
1495 destroy_workqueue(info->workqueue);
1496 out4:
1497 mempool_destroy(info->response_mempool);
1498 out3:
1499 kmem_cache_destroy(info->response_cache);
1500 out2:
1501 mempool_destroy(info->request_mempool);
1502 out1:
1503 kmem_cache_destroy(info->request_cache);
1504 return -ENOMEM;
1505 }
1506
1507 /* Create a SMBD connection, called by upper layer */
1508 static struct smbd_connection *_smbd_get_connection(
1509 struct TCP_Server_Info *server, struct sockaddr *dstaddr, int port)
1510 {
1511 int rc;
1512 struct smbd_connection *info;
1513 struct rdma_conn_param conn_param;
1514 struct ib_qp_init_attr qp_attr;
1515 struct sockaddr_in *addr_in = (struct sockaddr_in *) dstaddr;
1516 struct ib_port_immutable port_immutable;
1517 u32 ird_ord_hdr[2];
1518
1519 info = kzalloc(sizeof(struct smbd_connection), GFP_KERNEL);
1520 if (!info)
1521 return NULL;
1522
1523 info->transport_status = SMBD_CONNECTING;
1524 rc = smbd_ia_open(info, dstaddr, port);
1525 if (rc) {
1526 log_rdma_event(INFO, "smbd_ia_open rc=%d\n", rc);
1527 goto create_id_failed;
1528 }
1529
1530 if (smbd_send_credit_target > info->id->device->attrs.max_cqe ||
1531 smbd_send_credit_target > info->id->device->attrs.max_qp_wr) {
1532 log_rdma_event(ERR, "consider lowering send_credit_target = %d. Possible CQE overrun, device reporting max_cpe %d max_qp_wr %d\n",
1533 smbd_send_credit_target,
1534 info->id->device->attrs.max_cqe,
1535 info->id->device->attrs.max_qp_wr);
1536 goto config_failed;
1537 }
1538
1539 if (smbd_receive_credit_max > info->id->device->attrs.max_cqe ||
1540 smbd_receive_credit_max > info->id->device->attrs.max_qp_wr) {
1541 log_rdma_event(ERR, "consider lowering receive_credit_max = %d. Possible CQE overrun, device reporting max_cpe %d max_qp_wr %d\n",
1542 smbd_receive_credit_max,
1543 info->id->device->attrs.max_cqe,
1544 info->id->device->attrs.max_qp_wr);
1545 goto config_failed;
1546 }
1547
1548 info->receive_credit_max = smbd_receive_credit_max;
1549 info->send_credit_target = smbd_send_credit_target;
1550 info->max_send_size = smbd_max_send_size;
1551 info->max_fragmented_recv_size = smbd_max_fragmented_recv_size;
1552 info->max_receive_size = smbd_max_receive_size;
1553 info->keep_alive_interval = smbd_keep_alive_interval;
1554
1555 if (info->id->device->attrs.max_send_sge < SMBDIRECT_MAX_SGE) {
1556 log_rdma_event(ERR,
1557 "warning: device max_send_sge = %d too small\n",
1558 info->id->device->attrs.max_send_sge);
1559 log_rdma_event(ERR, "Queue Pair creation may fail\n");
1560 }
1561 if (info->id->device->attrs.max_recv_sge < SMBDIRECT_MAX_SGE) {
1562 log_rdma_event(ERR,
1563 "warning: device max_recv_sge = %d too small\n",
1564 info->id->device->attrs.max_recv_sge);
1565 log_rdma_event(ERR, "Queue Pair creation may fail\n");
1566 }
1567
1568 info->send_cq = NULL;
1569 info->recv_cq = NULL;
1570 info->send_cq =
1571 ib_alloc_cq_any(info->id->device, info,
1572 info->send_credit_target, IB_POLL_SOFTIRQ);
1573 if (IS_ERR(info->send_cq)) {
1574 info->send_cq = NULL;
1575 goto alloc_cq_failed;
1576 }
1577
1578 info->recv_cq =
1579 ib_alloc_cq_any(info->id->device, info,
1580 info->receive_credit_max, IB_POLL_SOFTIRQ);
1581 if (IS_ERR(info->recv_cq)) {
1582 info->recv_cq = NULL;
1583 goto alloc_cq_failed;
1584 }
1585
1586 memset(&qp_attr, 0, sizeof(qp_attr));
1587 qp_attr.event_handler = smbd_qp_async_error_upcall;
1588 qp_attr.qp_context = info;
1589 qp_attr.cap.max_send_wr = info->send_credit_target;
1590 qp_attr.cap.max_recv_wr = info->receive_credit_max;
1591 qp_attr.cap.max_send_sge = SMBDIRECT_MAX_SGE;
1592 qp_attr.cap.max_recv_sge = SMBDIRECT_MAX_SGE;
1593 qp_attr.cap.max_inline_data = 0;
1594 qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
1595 qp_attr.qp_type = IB_QPT_RC;
1596 qp_attr.send_cq = info->send_cq;
1597 qp_attr.recv_cq = info->recv_cq;
1598 qp_attr.port_num = ~0;
1599
1600 rc = rdma_create_qp(info->id, info->pd, &qp_attr);
1601 if (rc) {
1602 log_rdma_event(ERR, "rdma_create_qp failed %i\n", rc);
1603 goto create_qp_failed;
1604 }
1605
1606 memset(&conn_param, 0, sizeof(conn_param));
1607 conn_param.initiator_depth = 0;
1608
1609 conn_param.responder_resources =
1610 info->id->device->attrs.max_qp_rd_atom
1611 < SMBD_CM_RESPONDER_RESOURCES ?
1612 info->id->device->attrs.max_qp_rd_atom :
1613 SMBD_CM_RESPONDER_RESOURCES;
1614 info->responder_resources = conn_param.responder_resources;
1615 log_rdma_mr(INFO, "responder_resources=%d\n",
1616 info->responder_resources);
1617
1618 /* Need to send IRD/ORD in private data for iWARP */
1619 info->id->device->ops.get_port_immutable(
1620 info->id->device, info->id->port_num, &port_immutable);
1621 if (port_immutable.core_cap_flags & RDMA_CORE_PORT_IWARP) {
1622 ird_ord_hdr[0] = info->responder_resources;
1623 ird_ord_hdr[1] = 1;
1624 conn_param.private_data = ird_ord_hdr;
1625 conn_param.private_data_len = sizeof(ird_ord_hdr);
1626 } else {
1627 conn_param.private_data = NULL;
1628 conn_param.private_data_len = 0;
1629 }
1630
1631 conn_param.retry_count = SMBD_CM_RETRY;
1632 conn_param.rnr_retry_count = SMBD_CM_RNR_RETRY;
1633 conn_param.flow_control = 0;
1634
1635 log_rdma_event(INFO, "connecting to IP %pI4 port %d\n",
1636 &addr_in->sin_addr, port);
1637
1638 init_waitqueue_head(&info->conn_wait);
1639 init_waitqueue_head(&info->disconn_wait);
1640 init_waitqueue_head(&info->wait_reassembly_queue);
1641 rc = rdma_connect(info->id, &conn_param);
1642 if (rc) {
1643 log_rdma_event(ERR, "rdma_connect() failed with %i\n", rc);
1644 goto rdma_connect_failed;
1645 }
1646
1647 wait_event_interruptible(
1648 info->conn_wait, info->transport_status != SMBD_CONNECTING);
1649
1650 if (info->transport_status != SMBD_CONNECTED) {
1651 log_rdma_event(ERR, "rdma_connect failed port=%d\n", port);
1652 goto rdma_connect_failed;
1653 }
1654
1655 log_rdma_event(INFO, "rdma_connect connected\n");
1656
1657 rc = allocate_caches_and_workqueue(info);
1658 if (rc) {
1659 log_rdma_event(ERR, "cache allocation failed\n");
1660 goto allocate_cache_failed;
1661 }
1662
1663 init_waitqueue_head(&info->wait_send_queue);
1664 INIT_DELAYED_WORK(&info->idle_timer_work, idle_connection_timer);
1665 queue_delayed_work(info->workqueue, &info->idle_timer_work,
1666 info->keep_alive_interval*HZ);
1667
1668 init_waitqueue_head(&info->wait_send_pending);
1669 atomic_set(&info->send_pending, 0);
1670
1671 init_waitqueue_head(&info->wait_post_send);
1672
1673 INIT_WORK(&info->disconnect_work, smbd_disconnect_rdma_work);
1674 INIT_WORK(&info->post_send_credits_work, smbd_post_send_credits);
1675 info->new_credits_offered = 0;
1676 spin_lock_init(&info->lock_new_credits_offered);
1677
1678 rc = smbd_negotiate(info);
1679 if (rc) {
1680 log_rdma_event(ERR, "smbd_negotiate rc=%d\n", rc);
1681 goto negotiation_failed;
1682 }
1683
1684 rc = allocate_mr_list(info);
1685 if (rc) {
1686 log_rdma_mr(ERR, "memory registration allocation failed\n");
1687 goto allocate_mr_failed;
1688 }
1689
1690 return info;
1691
1692 allocate_mr_failed:
1693 /* At this point, need to a full transport shutdown */
1694 smbd_destroy(server);
1695 return NULL;
1696
1697 negotiation_failed:
1698 cancel_delayed_work_sync(&info->idle_timer_work);
1699 destroy_caches_and_workqueue(info);
1700 info->transport_status = SMBD_NEGOTIATE_FAILED;
1701 init_waitqueue_head(&info->conn_wait);
1702 rdma_disconnect(info->id);
1703 wait_event(info->conn_wait,
1704 info->transport_status == SMBD_DISCONNECTED);
1705
1706 allocate_cache_failed:
1707 rdma_connect_failed:
1708 rdma_destroy_qp(info->id);
1709
1710 create_qp_failed:
1711 alloc_cq_failed:
1712 if (info->send_cq)
1713 ib_free_cq(info->send_cq);
1714 if (info->recv_cq)
1715 ib_free_cq(info->recv_cq);
1716
1717 config_failed:
1718 ib_dealloc_pd(info->pd);
1719 rdma_destroy_id(info->id);
1720
1721 create_id_failed:
1722 kfree(info);
1723 return NULL;
1724 }
1725
1726 struct smbd_connection *smbd_get_connection(
1727 struct TCP_Server_Info *server, struct sockaddr *dstaddr)
1728 {
1729 struct smbd_connection *ret;
1730 int port = SMBD_PORT;
1731
1732 try_again:
1733 ret = _smbd_get_connection(server, dstaddr, port);
1734
1735 /* Try SMB_PORT if SMBD_PORT doesn't work */
1736 if (!ret && port == SMBD_PORT) {
1737 port = SMB_PORT;
1738 goto try_again;
1739 }
1740 return ret;
1741 }
1742
1743 /*
1744 * Receive data from receive reassembly queue
1745 * All the incoming data packets are placed in reassembly queue
1746 * buf: the buffer to read data into
1747 * size: the length of data to read
1748 * return value: actual data read
1749 * Note: this implementation copies the data from reassebmly queue to receive
1750 * buffers used by upper layer. This is not the optimal code path. A better way
1751 * to do it is to not have upper layer allocate its receive buffers but rather
1752 * borrow the buffer from reassembly queue, and return it after data is
1753 * consumed. But this will require more changes to upper layer code, and also
1754 * need to consider packet boundaries while they still being reassembled.
1755 */
1756 static int smbd_recv_buf(struct smbd_connection *info, char *buf,
1757 unsigned int size)
1758 {
1759 struct smbd_response *response;
1760 struct smbd_data_transfer *data_transfer;
1761 int to_copy, to_read, data_read, offset;
1762 u32 data_length, remaining_data_length, data_offset;
1763 int rc;
1764
1765 again:
1766 /*
1767 * No need to hold the reassembly queue lock all the time as we are
1768 * the only one reading from the front of the queue. The transport
1769 * may add more entries to the back of the queue at the same time
1770 */
1771 log_read(INFO, "size=%d info->reassembly_data_length=%d\n", size,
1772 info->reassembly_data_length);
1773 if (info->reassembly_data_length >= size) {
1774 int queue_length;
1775 int queue_removed = 0;
1776
1777 /*
1778 * Need to make sure reassembly_data_length is read before
1779 * reading reassembly_queue_length and calling
1780 * _get_first_reassembly. This call is lock free
1781 * as we never read at the end of the queue which are being
1782 * updated in SOFTIRQ as more data is received
1783 */
1784 virt_rmb();
1785 queue_length = info->reassembly_queue_length;
1786 data_read = 0;
1787 to_read = size;
1788 offset = info->first_entry_offset;
1789 while (data_read < size) {
1790 response = _get_first_reassembly(info);
1791 data_transfer = smbd_response_payload(response);
1792 data_length = le32_to_cpu(data_transfer->data_length);
1793 remaining_data_length =
1794 le32_to_cpu(
1795 data_transfer->remaining_data_length);
1796 data_offset = le32_to_cpu(data_transfer->data_offset);
1797
1798 /*
1799 * The upper layer expects RFC1002 length at the
1800 * beginning of the payload. Return it to indicate
1801 * the total length of the packet. This minimize the
1802 * change to upper layer packet processing logic. This
1803 * will be eventually remove when an intermediate
1804 * transport layer is added
1805 */
1806 if (response->first_segment && size == 4) {
1807 unsigned int rfc1002_len =
1808 data_length + remaining_data_length;
1809 *((__be32 *)buf) = cpu_to_be32(rfc1002_len);
1810 data_read = 4;
1811 response->first_segment = false;
1812 log_read(INFO, "returning rfc1002 length %d\n",
1813 rfc1002_len);
1814 goto read_rfc1002_done;
1815 }
1816
1817 to_copy = min_t(int, data_length - offset, to_read);
1818 memcpy(
1819 buf + data_read,
1820 (char *)data_transfer + data_offset + offset,
1821 to_copy);
1822
1823 /* move on to the next buffer? */
1824 if (to_copy == data_length - offset) {
1825 queue_length--;
1826 /*
1827 * No need to lock if we are not at the
1828 * end of the queue
1829 */
1830 if (queue_length)
1831 list_del(&response->list);
1832 else {
1833 spin_lock_irq(
1834 &info->reassembly_queue_lock);
1835 list_del(&response->list);
1836 spin_unlock_irq(
1837 &info->reassembly_queue_lock);
1838 }
1839 queue_removed++;
1840 info->count_reassembly_queue--;
1841 info->count_dequeue_reassembly_queue++;
1842 put_receive_buffer(info, response);
1843 offset = 0;
1844 log_read(INFO, "put_receive_buffer offset=0\n");
1845 } else
1846 offset += to_copy;
1847
1848 to_read -= to_copy;
1849 data_read += to_copy;
1850
1851 log_read(INFO, "_get_first_reassembly memcpy %d bytes data_transfer_length-offset=%d after that to_read=%d data_read=%d offset=%d\n",
1852 to_copy, data_length - offset,
1853 to_read, data_read, offset);
1854 }
1855
1856 spin_lock_irq(&info->reassembly_queue_lock);
1857 info->reassembly_data_length -= data_read;
1858 info->reassembly_queue_length -= queue_removed;
1859 spin_unlock_irq(&info->reassembly_queue_lock);
1860
1861 info->first_entry_offset = offset;
1862 log_read(INFO, "returning to thread data_read=%d reassembly_data_length=%d first_entry_offset=%d\n",
1863 data_read, info->reassembly_data_length,
1864 info->first_entry_offset);
1865 read_rfc1002_done:
1866 return data_read;
1867 }
1868
1869 log_read(INFO, "wait_event on more data\n");
1870 rc = wait_event_interruptible(
1871 info->wait_reassembly_queue,
1872 info->reassembly_data_length >= size ||
1873 info->transport_status != SMBD_CONNECTED);
1874 /* Don't return any data if interrupted */
1875 if (rc)
1876 return rc;
1877
1878 if (info->transport_status != SMBD_CONNECTED) {
1879 log_read(ERR, "disconnected\n");
1880 return -ECONNABORTED;
1881 }
1882
1883 goto again;
1884 }
1885
1886 /*
1887 * Receive a page from receive reassembly queue
1888 * page: the page to read data into
1889 * to_read: the length of data to read
1890 * return value: actual data read
1891 */
1892 static int smbd_recv_page(struct smbd_connection *info,
1893 struct page *page, unsigned int page_offset,
1894 unsigned int to_read)
1895 {
1896 int ret;
1897 char *to_address;
1898 void *page_address;
1899
1900 /* make sure we have the page ready for read */
1901 ret = wait_event_interruptible(
1902 info->wait_reassembly_queue,
1903 info->reassembly_data_length >= to_read ||
1904 info->transport_status != SMBD_CONNECTED);
1905 if (ret)
1906 return ret;
1907
1908 /* now we can read from reassembly queue and not sleep */
1909 page_address = kmap_atomic(page);
1910 to_address = (char *) page_address + page_offset;
1911
1912 log_read(INFO, "reading from page=%p address=%p to_read=%d\n",
1913 page, to_address, to_read);
1914
1915 ret = smbd_recv_buf(info, to_address, to_read);
1916 kunmap_atomic(page_address);
1917
1918 return ret;
1919 }
1920
1921 /*
1922 * Receive data from transport
1923 * msg: a msghdr point to the buffer, can be ITER_KVEC or ITER_BVEC
1924 * return: total bytes read, or 0. SMB Direct will not do partial read.
1925 */
1926 int smbd_recv(struct smbd_connection *info, struct msghdr *msg)
1927 {
1928 char *buf;
1929 struct page *page;
1930 unsigned int to_read, page_offset;
1931 int rc;
1932
1933 if (iov_iter_rw(&msg->msg_iter) == WRITE) {
1934 /* It's a bug in upper layer to get there */
1935 cifs_dbg(VFS, "Invalid msg iter dir %u\n",
1936 iov_iter_rw(&msg->msg_iter));
1937 rc = -EINVAL;
1938 goto out;
1939 }
1940
1941 switch (iov_iter_type(&msg->msg_iter)) {
1942 case ITER_KVEC:
1943 buf = msg->msg_iter.kvec->iov_base;
1944 to_read = msg->msg_iter.kvec->iov_len;
1945 rc = smbd_recv_buf(info, buf, to_read);
1946 break;
1947
1948 case ITER_BVEC:
1949 page = msg->msg_iter.bvec->bv_page;
1950 page_offset = msg->msg_iter.bvec->bv_offset;
1951 to_read = msg->msg_iter.bvec->bv_len;
1952 rc = smbd_recv_page(info, page, page_offset, to_read);
1953 break;
1954
1955 default:
1956 /* It's a bug in upper layer to get there */
1957 cifs_dbg(VFS, "Invalid msg type %d\n",
1958 iov_iter_type(&msg->msg_iter));
1959 rc = -EINVAL;
1960 }
1961
1962 out:
1963 /* SMBDirect will read it all or nothing */
1964 if (rc > 0)
1965 msg->msg_iter.count = 0;
1966 return rc;
1967 }
1968
1969 /*
1970 * Send data to transport
1971 * Each rqst is transported as a SMBDirect payload
1972 * rqst: the data to write
1973 * return value: 0 if successfully write, otherwise error code
1974 */
1975 int smbd_send(struct TCP_Server_Info *server,
1976 int num_rqst, struct smb_rqst *rqst_array)
1977 {
1978 struct smbd_connection *info = server->smbd_conn;
1979 struct kvec vec;
1980 int nvecs;
1981 int size;
1982 unsigned int buflen, remaining_data_length;
1983 int start, i, j;
1984 int max_iov_size =
1985 info->max_send_size - sizeof(struct smbd_data_transfer);
1986 struct kvec *iov;
1987 int rc;
1988 struct smb_rqst *rqst;
1989 int rqst_idx;
1990
1991 if (info->transport_status != SMBD_CONNECTED) {
1992 rc = -EAGAIN;
1993 goto done;
1994 }
1995
1996 /*
1997 * Add in the page array if there is one. The caller needs to set
1998 * rq_tailsz to PAGE_SIZE when the buffer has multiple pages and
1999 * ends at page boundary
2000 */
2001 remaining_data_length = 0;
2002 for (i = 0; i < num_rqst; i++)
2003 remaining_data_length += smb_rqst_len(server, &rqst_array[i]);
2004
2005 if (remaining_data_length > info->max_fragmented_send_size) {
2006 log_write(ERR, "payload size %d > max size %d\n",
2007 remaining_data_length, info->max_fragmented_send_size);
2008 rc = -EINVAL;
2009 goto done;
2010 }
2011
2012 log_write(INFO, "num_rqst=%d total length=%u\n",
2013 num_rqst, remaining_data_length);
2014
2015 rqst_idx = 0;
2016 next_rqst:
2017 rqst = &rqst_array[rqst_idx];
2018 iov = rqst->rq_iov;
2019
2020 cifs_dbg(FYI, "Sending smb (RDMA): idx=%d smb_len=%lu\n",
2021 rqst_idx, smb_rqst_len(server, rqst));
2022 for (i = 0; i < rqst->rq_nvec; i++)
2023 dump_smb(iov[i].iov_base, iov[i].iov_len);
2024
2025
2026 log_write(INFO, "rqst_idx=%d nvec=%d rqst->rq_npages=%d rq_pagesz=%d rq_tailsz=%d buflen=%lu\n",
2027 rqst_idx, rqst->rq_nvec, rqst->rq_npages, rqst->rq_pagesz,
2028 rqst->rq_tailsz, smb_rqst_len(server, rqst));
2029
2030 start = i = 0;
2031 buflen = 0;
2032 while (true) {
2033 buflen += iov[i].iov_len;
2034 if (buflen > max_iov_size) {
2035 if (i > start) {
2036 remaining_data_length -=
2037 (buflen-iov[i].iov_len);
2038 log_write(INFO, "sending iov[] from start=%d i=%d nvecs=%d remaining_data_length=%d\n",
2039 start, i, i - start,
2040 remaining_data_length);
2041 rc = smbd_post_send_data(
2042 info, &iov[start], i-start,
2043 remaining_data_length);
2044 if (rc)
2045 goto done;
2046 } else {
2047 /* iov[start] is too big, break it */
2048 nvecs = (buflen+max_iov_size-1)/max_iov_size;
2049 log_write(INFO, "iov[%d] iov_base=%p buflen=%d break to %d vectors\n",
2050 start, iov[start].iov_base,
2051 buflen, nvecs);
2052 for (j = 0; j < nvecs; j++) {
2053 vec.iov_base =
2054 (char *)iov[start].iov_base +
2055 j*max_iov_size;
2056 vec.iov_len = max_iov_size;
2057 if (j == nvecs-1)
2058 vec.iov_len =
2059 buflen -
2060 max_iov_size*(nvecs-1);
2061 remaining_data_length -= vec.iov_len;
2062 log_write(INFO,
2063 "sending vec j=%d iov_base=%p iov_len=%zu remaining_data_length=%d\n",
2064 j, vec.iov_base, vec.iov_len,
2065 remaining_data_length);
2066 rc = smbd_post_send_data(
2067 info, &vec, 1,
2068 remaining_data_length);
2069 if (rc)
2070 goto done;
2071 }
2072 i++;
2073 if (i == rqst->rq_nvec)
2074 break;
2075 }
2076 start = i;
2077 buflen = 0;
2078 } else {
2079 i++;
2080 if (i == rqst->rq_nvec) {
2081 /* send out all remaining vecs */
2082 remaining_data_length -= buflen;
2083 log_write(INFO, "sending iov[] from start=%d i=%d nvecs=%d remaining_data_length=%d\n",
2084 start, i, i - start,
2085 remaining_data_length);
2086 rc = smbd_post_send_data(info, &iov[start],
2087 i-start, remaining_data_length);
2088 if (rc)
2089 goto done;
2090 break;
2091 }
2092 }
2093 log_write(INFO, "looping i=%d buflen=%d\n", i, buflen);
2094 }
2095
2096 /* now sending pages if there are any */
2097 for (i = 0; i < rqst->rq_npages; i++) {
2098 unsigned int offset;
2099
2100 rqst_page_get_length(rqst, i, &buflen, &offset);
2101 nvecs = (buflen + max_iov_size - 1) / max_iov_size;
2102 log_write(INFO, "sending pages buflen=%d nvecs=%d\n",
2103 buflen, nvecs);
2104 for (j = 0; j < nvecs; j++) {
2105 size = max_iov_size;
2106 if (j == nvecs-1)
2107 size = buflen - j*max_iov_size;
2108 remaining_data_length -= size;
2109 log_write(INFO, "sending pages i=%d offset=%d size=%d remaining_data_length=%d\n",
2110 i, j * max_iov_size + offset, size,
2111 remaining_data_length);
2112 rc = smbd_post_send_page(
2113 info, rqst->rq_pages[i],
2114 j*max_iov_size + offset,
2115 size, remaining_data_length);
2116 if (rc)
2117 goto done;
2118 }
2119 }
2120
2121 rqst_idx++;
2122 if (rqst_idx < num_rqst)
2123 goto next_rqst;
2124
2125 done:
2126 /*
2127 * As an optimization, we don't wait for individual I/O to finish
2128 * before sending the next one.
2129 * Send them all and wait for pending send count to get to 0
2130 * that means all the I/Os have been out and we are good to return
2131 */
2132
2133 wait_event(info->wait_send_pending,
2134 atomic_read(&info->send_pending) == 0);
2135
2136 return rc;
2137 }
2138
2139 static void register_mr_done(struct ib_cq *cq, struct ib_wc *wc)
2140 {
2141 struct smbd_mr *mr;
2142 struct ib_cqe *cqe;
2143
2144 if (wc->status) {
2145 log_rdma_mr(ERR, "status=%d\n", wc->status);
2146 cqe = wc->wr_cqe;
2147 mr = container_of(cqe, struct smbd_mr, cqe);
2148 smbd_disconnect_rdma_connection(mr->conn);
2149 }
2150 }
2151
2152 /*
2153 * The work queue function that recovers MRs
2154 * We need to call ib_dereg_mr() and ib_alloc_mr() before this MR can be used
2155 * again. Both calls are slow, so finish them in a workqueue. This will not
2156 * block I/O path.
2157 * There is one workqueue that recovers MRs, there is no need to lock as the
2158 * I/O requests calling smbd_register_mr will never update the links in the
2159 * mr_list.
2160 */
2161 static void smbd_mr_recovery_work(struct work_struct *work)
2162 {
2163 struct smbd_connection *info =
2164 container_of(work, struct smbd_connection, mr_recovery_work);
2165 struct smbd_mr *smbdirect_mr;
2166 int rc;
2167
2168 list_for_each_entry(smbdirect_mr, &info->mr_list, list) {
2169 if (smbdirect_mr->state == MR_ERROR) {
2170
2171 /* recover this MR entry */
2172 rc = ib_dereg_mr(smbdirect_mr->mr);
2173 if (rc) {
2174 log_rdma_mr(ERR,
2175 "ib_dereg_mr failed rc=%x\n",
2176 rc);
2177 smbd_disconnect_rdma_connection(info);
2178 continue;
2179 }
2180
2181 smbdirect_mr->mr = ib_alloc_mr(
2182 info->pd, info->mr_type,
2183 info->max_frmr_depth);
2184 if (IS_ERR(smbdirect_mr->mr)) {
2185 log_rdma_mr(ERR, "ib_alloc_mr failed mr_type=%x max_frmr_depth=%x\n",
2186 info->mr_type,
2187 info->max_frmr_depth);
2188 smbd_disconnect_rdma_connection(info);
2189 continue;
2190 }
2191 } else
2192 /* This MR is being used, don't recover it */
2193 continue;
2194
2195 smbdirect_mr->state = MR_READY;
2196
2197 /* smbdirect_mr->state is updated by this function
2198 * and is read and updated by I/O issuing CPUs trying
2199 * to get a MR, the call to atomic_inc_return
2200 * implicates a memory barrier and guarantees this
2201 * value is updated before waking up any calls to
2202 * get_mr() from the I/O issuing CPUs
2203 */
2204 if (atomic_inc_return(&info->mr_ready_count) == 1)
2205 wake_up_interruptible(&info->wait_mr);
2206 }
2207 }
2208
2209 static void destroy_mr_list(struct smbd_connection *info)
2210 {
2211 struct smbd_mr *mr, *tmp;
2212
2213 cancel_work_sync(&info->mr_recovery_work);
2214 list_for_each_entry_safe(mr, tmp, &info->mr_list, list) {
2215 if (mr->state == MR_INVALIDATED)
2216 ib_dma_unmap_sg(info->id->device, mr->sgl,
2217 mr->sgl_count, mr->dir);
2218 ib_dereg_mr(mr->mr);
2219 kfree(mr->sgl);
2220 kfree(mr);
2221 }
2222 }
2223
2224 /*
2225 * Allocate MRs used for RDMA read/write
2226 * The number of MRs will not exceed hardware capability in responder_resources
2227 * All MRs are kept in mr_list. The MR can be recovered after it's used
2228 * Recovery is done in smbd_mr_recovery_work. The content of list entry changes
2229 * as MRs are used and recovered for I/O, but the list links will not change
2230 */
2231 static int allocate_mr_list(struct smbd_connection *info)
2232 {
2233 int i;
2234 struct smbd_mr *smbdirect_mr, *tmp;
2235
2236 INIT_LIST_HEAD(&info->mr_list);
2237 init_waitqueue_head(&info->wait_mr);
2238 spin_lock_init(&info->mr_list_lock);
2239 atomic_set(&info->mr_ready_count, 0);
2240 atomic_set(&info->mr_used_count, 0);
2241 init_waitqueue_head(&info->wait_for_mr_cleanup);
2242 /* Allocate more MRs (2x) than hardware responder_resources */
2243 for (i = 0; i < info->responder_resources * 2; i++) {
2244 smbdirect_mr = kzalloc(sizeof(*smbdirect_mr), GFP_KERNEL);
2245 if (!smbdirect_mr)
2246 goto out;
2247 smbdirect_mr->mr = ib_alloc_mr(info->pd, info->mr_type,
2248 info->max_frmr_depth);
2249 if (IS_ERR(smbdirect_mr->mr)) {
2250 log_rdma_mr(ERR, "ib_alloc_mr failed mr_type=%x max_frmr_depth=%x\n",
2251 info->mr_type, info->max_frmr_depth);
2252 goto out;
2253 }
2254 smbdirect_mr->sgl = kcalloc(
2255 info->max_frmr_depth,
2256 sizeof(struct scatterlist),
2257 GFP_KERNEL);
2258 if (!smbdirect_mr->sgl) {
2259 log_rdma_mr(ERR, "failed to allocate sgl\n");
2260 ib_dereg_mr(smbdirect_mr->mr);
2261 goto out;
2262 }
2263 smbdirect_mr->state = MR_READY;
2264 smbdirect_mr->conn = info;
2265
2266 list_add_tail(&smbdirect_mr->list, &info->mr_list);
2267 atomic_inc(&info->mr_ready_count);
2268 }
2269 INIT_WORK(&info->mr_recovery_work, smbd_mr_recovery_work);
2270 return 0;
2271
2272 out:
2273 kfree(smbdirect_mr);
2274
2275 list_for_each_entry_safe(smbdirect_mr, tmp, &info->mr_list, list) {
2276 ib_dereg_mr(smbdirect_mr->mr);
2277 kfree(smbdirect_mr->sgl);
2278 kfree(smbdirect_mr);
2279 }
2280 return -ENOMEM;
2281 }
2282
2283 /*
2284 * Get a MR from mr_list. This function waits until there is at least one
2285 * MR available in the list. It may access the list while the
2286 * smbd_mr_recovery_work is recovering the MR list. This doesn't need a lock
2287 * as they never modify the same places. However, there may be several CPUs
2288 * issueing I/O trying to get MR at the same time, mr_list_lock is used to
2289 * protect this situation.
2290 */
2291 static struct smbd_mr *get_mr(struct smbd_connection *info)
2292 {
2293 struct smbd_mr *ret;
2294 int rc;
2295 again:
2296 rc = wait_event_interruptible(info->wait_mr,
2297 atomic_read(&info->mr_ready_count) ||
2298 info->transport_status != SMBD_CONNECTED);
2299 if (rc) {
2300 log_rdma_mr(ERR, "wait_event_interruptible rc=%x\n", rc);
2301 return NULL;
2302 }
2303
2304 if (info->transport_status != SMBD_CONNECTED) {
2305 log_rdma_mr(ERR, "info->transport_status=%x\n",
2306 info->transport_status);
2307 return NULL;
2308 }
2309
2310 spin_lock(&info->mr_list_lock);
2311 list_for_each_entry(ret, &info->mr_list, list) {
2312 if (ret->state == MR_READY) {
2313 ret->state = MR_REGISTERED;
2314 spin_unlock(&info->mr_list_lock);
2315 atomic_dec(&info->mr_ready_count);
2316 atomic_inc(&info->mr_used_count);
2317 return ret;
2318 }
2319 }
2320
2321 spin_unlock(&info->mr_list_lock);
2322 /*
2323 * It is possible that we could fail to get MR because other processes may
2324 * try to acquire a MR at the same time. If this is the case, retry it.
2325 */
2326 goto again;
2327 }
2328
2329 /*
2330 * Register memory for RDMA read/write
2331 * pages[]: the list of pages to register memory with
2332 * num_pages: the number of pages to register
2333 * tailsz: if non-zero, the bytes to register in the last page
2334 * writing: true if this is a RDMA write (SMB read), false for RDMA read
2335 * need_invalidate: true if this MR needs to be locally invalidated after I/O
2336 * return value: the MR registered, NULL if failed.
2337 */
2338 struct smbd_mr *smbd_register_mr(
2339 struct smbd_connection *info, struct page *pages[], int num_pages,
2340 int offset, int tailsz, bool writing, bool need_invalidate)
2341 {
2342 struct smbd_mr *smbdirect_mr;
2343 int rc, i;
2344 enum dma_data_direction dir;
2345 struct ib_reg_wr *reg_wr;
2346
2347 if (num_pages > info->max_frmr_depth) {
2348 log_rdma_mr(ERR, "num_pages=%d max_frmr_depth=%d\n",
2349 num_pages, info->max_frmr_depth);
2350 return NULL;
2351 }
2352
2353 smbdirect_mr = get_mr(info);
2354 if (!smbdirect_mr) {
2355 log_rdma_mr(ERR, "get_mr returning NULL\n");
2356 return NULL;
2357 }
2358 smbdirect_mr->need_invalidate = need_invalidate;
2359 smbdirect_mr->sgl_count = num_pages;
2360 sg_init_table(smbdirect_mr->sgl, num_pages);
2361
2362 log_rdma_mr(INFO, "num_pages=0x%x offset=0x%x tailsz=0x%x\n",
2363 num_pages, offset, tailsz);
2364
2365 if (num_pages == 1) {
2366 sg_set_page(&smbdirect_mr->sgl[0], pages[0], tailsz, offset);
2367 goto skip_multiple_pages;
2368 }
2369
2370 /* We have at least two pages to register */
2371 sg_set_page(
2372 &smbdirect_mr->sgl[0], pages[0], PAGE_SIZE - offset, offset);
2373 i = 1;
2374 while (i < num_pages - 1) {
2375 sg_set_page(&smbdirect_mr->sgl[i], pages[i], PAGE_SIZE, 0);
2376 i++;
2377 }
2378 sg_set_page(&smbdirect_mr->sgl[i], pages[i],
2379 tailsz ? tailsz : PAGE_SIZE, 0);
2380
2381 skip_multiple_pages:
2382 dir = writing ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
2383 smbdirect_mr->dir = dir;
2384 rc = ib_dma_map_sg(info->id->device, smbdirect_mr->sgl, num_pages, dir);
2385 if (!rc) {
2386 log_rdma_mr(ERR, "ib_dma_map_sg num_pages=%x dir=%x rc=%x\n",
2387 num_pages, dir, rc);
2388 goto dma_map_error;
2389 }
2390
2391 rc = ib_map_mr_sg(smbdirect_mr->mr, smbdirect_mr->sgl, num_pages,
2392 NULL, PAGE_SIZE);
2393 if (rc != num_pages) {
2394 log_rdma_mr(ERR,
2395 "ib_map_mr_sg failed rc = %d num_pages = %x\n",
2396 rc, num_pages);
2397 goto map_mr_error;
2398 }
2399
2400 ib_update_fast_reg_key(smbdirect_mr->mr,
2401 ib_inc_rkey(smbdirect_mr->mr->rkey));
2402 reg_wr = &smbdirect_mr->wr;
2403 reg_wr->wr.opcode = IB_WR_REG_MR;
2404 smbdirect_mr->cqe.done = register_mr_done;
2405 reg_wr->wr.wr_cqe = &smbdirect_mr->cqe;
2406 reg_wr->wr.num_sge = 0;
2407 reg_wr->wr.send_flags = IB_SEND_SIGNALED;
2408 reg_wr->mr = smbdirect_mr->mr;
2409 reg_wr->key = smbdirect_mr->mr->rkey;
2410 reg_wr->access = writing ?
2411 IB_ACCESS_REMOTE_WRITE | IB_ACCESS_LOCAL_WRITE :
2412 IB_ACCESS_REMOTE_READ;
2413
2414 /*
2415 * There is no need for waiting for complemtion on ib_post_send
2416 * on IB_WR_REG_MR. Hardware enforces a barrier and order of execution
2417 * on the next ib_post_send when we actaully send I/O to remote peer
2418 */
2419 rc = ib_post_send(info->id->qp, &reg_wr->wr, NULL);
2420 if (!rc)
2421 return smbdirect_mr;
2422
2423 log_rdma_mr(ERR, "ib_post_send failed rc=%x reg_wr->key=%x\n",
2424 rc, reg_wr->key);
2425
2426 /* If all failed, attempt to recover this MR by setting it MR_ERROR*/
2427 map_mr_error:
2428 ib_dma_unmap_sg(info->id->device, smbdirect_mr->sgl,
2429 smbdirect_mr->sgl_count, smbdirect_mr->dir);
2430
2431 dma_map_error:
2432 smbdirect_mr->state = MR_ERROR;
2433 if (atomic_dec_and_test(&info->mr_used_count))
2434 wake_up(&info->wait_for_mr_cleanup);
2435
2436 smbd_disconnect_rdma_connection(info);
2437
2438 return NULL;
2439 }
2440
2441 static void local_inv_done(struct ib_cq *cq, struct ib_wc *wc)
2442 {
2443 struct smbd_mr *smbdirect_mr;
2444 struct ib_cqe *cqe;
2445
2446 cqe = wc->wr_cqe;
2447 smbdirect_mr = container_of(cqe, struct smbd_mr, cqe);
2448 smbdirect_mr->state = MR_INVALIDATED;
2449 if (wc->status != IB_WC_SUCCESS) {
2450 log_rdma_mr(ERR, "invalidate failed status=%x\n", wc->status);
2451 smbdirect_mr->state = MR_ERROR;
2452 }
2453 complete(&smbdirect_mr->invalidate_done);
2454 }
2455
2456 /*
2457 * Deregister a MR after I/O is done
2458 * This function may wait if remote invalidation is not used
2459 * and we have to locally invalidate the buffer to prevent data is being
2460 * modified by remote peer after upper layer consumes it
2461 */
2462 int smbd_deregister_mr(struct smbd_mr *smbdirect_mr)
2463 {
2464 struct ib_send_wr *wr;
2465 struct smbd_connection *info = smbdirect_mr->conn;
2466 int rc = 0;
2467
2468 if (smbdirect_mr->need_invalidate) {
2469 /* Need to finish local invalidation before returning */
2470 wr = &smbdirect_mr->inv_wr;
2471 wr->opcode = IB_WR_LOCAL_INV;
2472 smbdirect_mr->cqe.done = local_inv_done;
2473 wr->wr_cqe = &smbdirect_mr->cqe;
2474 wr->num_sge = 0;
2475 wr->ex.invalidate_rkey = smbdirect_mr->mr->rkey;
2476 wr->send_flags = IB_SEND_SIGNALED;
2477
2478 init_completion(&smbdirect_mr->invalidate_done);
2479 rc = ib_post_send(info->id->qp, wr, NULL);
2480 if (rc) {
2481 log_rdma_mr(ERR, "ib_post_send failed rc=%x\n", rc);
2482 smbd_disconnect_rdma_connection(info);
2483 goto done;
2484 }
2485 wait_for_completion(&smbdirect_mr->invalidate_done);
2486 smbdirect_mr->need_invalidate = false;
2487 } else
2488 /*
2489 * For remote invalidation, just set it to MR_INVALIDATED
2490 * and defer to mr_recovery_work to recover the MR for next use
2491 */
2492 smbdirect_mr->state = MR_INVALIDATED;
2493
2494 if (smbdirect_mr->state == MR_INVALIDATED) {
2495 ib_dma_unmap_sg(
2496 info->id->device, smbdirect_mr->sgl,
2497 smbdirect_mr->sgl_count,
2498 smbdirect_mr->dir);
2499 smbdirect_mr->state = MR_READY;
2500 if (atomic_inc_return(&info->mr_ready_count) == 1)
2501 wake_up_interruptible(&info->wait_mr);
2502 } else
2503 /*
2504 * Schedule the work to do MR recovery for future I/Os MR
2505 * recovery is slow and don't want it to block current I/O
2506 */
2507 queue_work(info->workqueue, &info->mr_recovery_work);
2508
2509 done:
2510 if (atomic_dec_and_test(&info->mr_used_count))
2511 wake_up(&info->wait_for_mr_cleanup);
2512
2513 return rc;
2514 }
Linux with added FPC-III support