1 /*
2 * Copyright (C) 2017, Microsoft Corporation.
3 *
4 * Author(s): Long Li <longli@microsoft.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
14 * the GNU General Public License for more details.
15 */
16 #include <linux/module.h>
17 #include <linux/highmem.h>
18 #include "smbdirect.h"
19 #include "cifs_debug.h"
20 #include "cifsproto.h"
21 #include "smb2proto.h"
22
23 static struct smbd_response *get_empty_queue_buffer(
24 struct smbd_connection *info);
25 static struct smbd_response *get_receive_buffer(
26 struct smbd_connection *info);
27 static void put_receive_buffer(
28 struct smbd_connection *info,
29 struct smbd_response *response);
30 static int allocate_receive_buffers(struct smbd_connection *info, int num_buf);
31 static void destroy_receive_buffers(struct smbd_connection *info);
32
33 static void put_empty_packet(
34 struct smbd_connection *info, struct smbd_response *response);
35 static void enqueue_reassembly(
36 struct smbd_connection *info,
37 struct smbd_response *response, int data_length);
38 static struct smbd_response *_get_first_reassembly(
39 struct smbd_connection *info);
40
41 static int smbd_post_recv(
42 struct smbd_connection *info,
43 struct smbd_response *response);
44
45 static int smbd_post_send_empty(struct smbd_connection *info);
46 static int smbd_post_send_data(
47 struct smbd_connection *info,
48 struct kvec *iov, int n_vec, int remaining_data_length);
49 static int smbd_post_send_page(struct smbd_connection *info,
50 struct page *page, unsigned long offset,
51 size_t size, int remaining_data_length);
52
53 static void destroy_mr_list(struct smbd_connection *info);
54 static int allocate_mr_list(struct smbd_connection *info);
55
56 /* SMBD version number */
57 #define SMBD_V1 0x0100
58
59 /* Port numbers for SMBD transport */
60 #define SMB_PORT 445
61 #define SMBD_PORT 5445
62
63 /* Address lookup and resolve timeout in ms */
64 #define RDMA_RESOLVE_TIMEOUT 5000
65
66 /* SMBD negotiation timeout in seconds */
67 #define SMBD_NEGOTIATE_TIMEOUT 120
68
69 /* SMBD minimum receive size and fragmented sized defined in [MS-SMBD] */
70 #define SMBD_MIN_RECEIVE_SIZE 128
71 #define SMBD_MIN_FRAGMENTED_SIZE 131072
72
73 /*
74 * Default maximum number of RDMA read/write outstanding on this connection
75 * This value is possibly decreased during QP creation on hardware limit
76 */
77 #define SMBD_CM_RESPONDER_RESOURCES 32
78
79 /* Maximum number of retries on data transfer operations */
80 #define SMBD_CM_RETRY 6
81 /* No need to retry on Receiver Not Ready since SMBD manages credits */
82 #define SMBD_CM_RNR_RETRY 0
83
84 /*
85 * User configurable initial values per SMBD transport connection
86 * as defined in [MS-SMBD] 3.1.1.1
87 * Those may change after a SMBD negotiation
88 */
89 /* The local peer's maximum number of credits to grant to the peer */
90 int smbd_receive_credit_max = 255;
91
92 /* The remote peer's credit request of local peer */
93 int smbd_send_credit_target = 255;
94
95 /* The maximum single message size can be sent to remote peer */
96 int smbd_max_send_size = 1364;
97
98 /* The maximum fragmented upper-layer payload receive size supported */
99 int smbd_max_fragmented_recv_size = 1024 * 1024;
100
101 /* The maximum single-message size which can be received */
102 int smbd_max_receive_size = 8192;
103
104 /* The timeout to initiate send of a keepalive message on idle */
105 int smbd_keep_alive_interval = 120;
106
107 /*
108 * User configurable initial values for RDMA transport
109 * The actual values used may be lower and are limited to hardware capabilities
110 */
111 /* Default maximum number of SGEs in a RDMA write/read */
112 int smbd_max_frmr_depth = 2048;
113
114 /* If payload is less than this byte, use RDMA send/recv not read/write */
115 int rdma_readwrite_threshold = 4096;
116
117 /* Transport logging functions
118 * Logging are defined as classes. They can be OR'ed to define the actual
119 * logging level via module parameter smbd_logging_class
120 * e.g. cifs.smbd_logging_class=0xa0 will log all log_rdma_recv() and
121 * log_rdma_event()
122 */
123 #define LOG_OUTGOING 0x1
124 #define LOG_INCOMING 0x2
125 #define LOG_READ 0x4
126 #define LOG_WRITE 0x8
127 #define LOG_RDMA_SEND 0x10
128 #define LOG_RDMA_RECV 0x20
129 #define LOG_KEEP_ALIVE 0x40
130 #define LOG_RDMA_EVENT 0x80
131 #define LOG_RDMA_MR 0x100
132 static unsigned int smbd_logging_class;
133 module_param(smbd_logging_class, uint, 0644);
134 MODULE_PARM_DESC(smbd_logging_class,
135 "Logging class for SMBD transport 0x0 to 0x100");
136
137 #define ERR 0x0
138 #define INFO 0x1
139 static unsigned int smbd_logging_level = ERR;
140 module_param(smbd_logging_level, uint, 0644);
141 MODULE_PARM_DESC(smbd_logging_level,
142 "Logging level for SMBD transport, 0 (default): error, 1: info");
143
144 #define log_rdma(level, class, fmt, args...) \
145 do { \
146 if (level <= smbd_logging_level || class & smbd_logging_class) \
147 cifs_dbg(VFS, "%s:%d " fmt, __func__, __LINE__, ##args);\
148 } while (0)
149
150 #define log_outgoing(level, fmt, args...) \
151 log_rdma(level, LOG_OUTGOING, fmt, ##args)
152 #define log_incoming(level, fmt, args...) \
153 log_rdma(level, LOG_INCOMING, fmt, ##args)
154 #define log_read(level, fmt, args...) log_rdma(level, LOG_READ, fmt, ##args)
155 #define log_write(level, fmt, args...) log_rdma(level, LOG_WRITE, fmt, ##args)
156 #define log_rdma_send(level, fmt, args...) \
157 log_rdma(level, LOG_RDMA_SEND, fmt, ##args)
158 #define log_rdma_recv(level, fmt, args...) \
159 log_rdma(level, LOG_RDMA_RECV, fmt, ##args)
160 #define log_keep_alive(level, fmt, args...) \
161 log_rdma(level, LOG_KEEP_ALIVE, fmt, ##args)
162 #define log_rdma_event(level, fmt, args...) \
163 log_rdma(level, LOG_RDMA_EVENT, fmt, ##args)
164 #define log_rdma_mr(level, fmt, args...) \
165 log_rdma(level, LOG_RDMA_MR, fmt, ##args)
166
167 /*
168 * Destroy the transport and related RDMA and memory resources
169 * Need to go through all the pending counters and make sure on one is using
170 * the transport while it is destroyed
171 */
172 static void smbd_destroy_rdma_work(struct work_struct *work)
173 {
174 struct smbd_response *response;
175 struct smbd_connection *info =
176 container_of(work, struct smbd_connection, destroy_work);
177 unsigned long flags;
178
179 log_rdma_event(INFO, "destroying qp\n");
180 ib_drain_qp(info->id->qp);
181 rdma_destroy_qp(info->id);
182
183 /* Unblock all I/O waiting on the send queue */
184 wake_up_interruptible_all(&info->wait_send_queue);
185
186 log_rdma_event(INFO, "cancelling idle timer\n");
187 cancel_delayed_work_sync(&info->idle_timer_work);
188 log_rdma_event(INFO, "cancelling send immediate work\n");
189 cancel_delayed_work_sync(&info->send_immediate_work);
190
191 log_rdma_event(INFO, "wait for all send to finish\n");
192 wait_event(info->wait_smbd_send_pending,
193 info->smbd_send_pending == 0);
194
195 log_rdma_event(INFO, "wait for all recv to finish\n");
196 wake_up_interruptible(&info->wait_reassembly_queue);
197 wait_event(info->wait_smbd_recv_pending,
198 info->smbd_recv_pending == 0);
199
200 log_rdma_event(INFO, "wait for all send posted to IB to finish\n");
201 wait_event(info->wait_send_pending,
202 atomic_read(&info->send_pending) == 0);
203 wait_event(info->wait_send_payload_pending,
204 atomic_read(&info->send_payload_pending) == 0);
205
206 log_rdma_event(INFO, "freeing mr list\n");
207 wake_up_interruptible_all(&info->wait_mr);
208 wait_event(info->wait_for_mr_cleanup,
209 atomic_read(&info->mr_used_count) == 0);
210 destroy_mr_list(info);
211
212 /* It's not posssible for upper layer to get to reassembly */
213 log_rdma_event(INFO, "drain the reassembly queue\n");
214 do {
215 spin_lock_irqsave(&info->reassembly_queue_lock, flags);
216 response = _get_first_reassembly(info);
217 if (response) {
218 list_del(&response->list);
219 spin_unlock_irqrestore(
220 &info->reassembly_queue_lock, flags);
221 put_receive_buffer(info, response);
222 } else
223 spin_unlock_irqrestore(&info->reassembly_queue_lock, flags);
224 } while (response);
225
226 info->reassembly_data_length = 0;
227
228 log_rdma_event(INFO, "free receive buffers\n");
229 wait_event(info->wait_receive_queues,
230 info->count_receive_queue + info->count_empty_packet_queue
231 == info->receive_credit_max);
232 destroy_receive_buffers(info);
233
234 ib_free_cq(info->send_cq);
235 ib_free_cq(info->recv_cq);
236 ib_dealloc_pd(info->pd);
237 rdma_destroy_id(info->id);
238
239 /* free mempools */
240 mempool_destroy(info->request_mempool);
241 kmem_cache_destroy(info->request_cache);
242
243 mempool_destroy(info->response_mempool);
244 kmem_cache_destroy(info->response_cache);
245
246 info->transport_status = SMBD_DESTROYED;
247 wake_up_all(&info->wait_destroy);
248 }
249
250 static int smbd_process_disconnected(struct smbd_connection *info)
251 {
252 schedule_work(&info->destroy_work);
253 return 0;
254 }
255
256 static void smbd_disconnect_rdma_work(struct work_struct *work)
257 {
258 struct smbd_connection *info =
259 container_of(work, struct smbd_connection, disconnect_work);
260
261 if (info->transport_status == SMBD_CONNECTED) {
262 info->transport_status = SMBD_DISCONNECTING;
263 rdma_disconnect(info->id);
264 }
265 }
266
267 static void smbd_disconnect_rdma_connection(struct smbd_connection *info)
268 {
269 queue_work(info->workqueue, &info->disconnect_work);
270 }
271
272 /* Upcall from RDMA CM */
273 static int smbd_conn_upcall(
274 struct rdma_cm_id *id, struct rdma_cm_event *event)
275 {
276 struct smbd_connection *info = id->context;
277
278 log_rdma_event(INFO, "event=%d status=%d\n",
279 event->event, event->status);
280
281 switch (event->event) {
282 case RDMA_CM_EVENT_ADDR_RESOLVED:
283 case RDMA_CM_EVENT_ROUTE_RESOLVED:
284 info->ri_rc = 0;
285 complete(&info->ri_done);
286 break;
287
288 case RDMA_CM_EVENT_ADDR_ERROR:
289 info->ri_rc = -EHOSTUNREACH;
290 complete(&info->ri_done);
291 break;
292
293 case RDMA_CM_EVENT_ROUTE_ERROR:
294 info->ri_rc = -ENETUNREACH;
295 complete(&info->ri_done);
296 break;
297
298 case RDMA_CM_EVENT_ESTABLISHED:
299 log_rdma_event(INFO, "connected event=%d\n", event->event);
300 info->transport_status = SMBD_CONNECTED;
301 wake_up_interruptible(&info->conn_wait);
302 break;
303
304 case RDMA_CM_EVENT_CONNECT_ERROR:
305 case RDMA_CM_EVENT_UNREACHABLE:
306 case RDMA_CM_EVENT_REJECTED:
307 log_rdma_event(INFO, "connecting failed event=%d\n", event->event);
308 info->transport_status = SMBD_DISCONNECTED;
309 wake_up_interruptible(&info->conn_wait);
310 break;
311
312 case RDMA_CM_EVENT_DEVICE_REMOVAL:
313 case RDMA_CM_EVENT_DISCONNECTED:
314 /* This happenes when we fail the negotiation */
315 if (info->transport_status == SMBD_NEGOTIATE_FAILED) {
316 info->transport_status = SMBD_DISCONNECTED;
317 wake_up(&info->conn_wait);
318 break;
319 }
320
321 info->transport_status = SMBD_DISCONNECTED;
322 smbd_process_disconnected(info);
323 break;
324
325 default:
326 break;
327 }
328
329 return 0;
330 }
331
332 /* Upcall from RDMA QP */
333 static void
334 smbd_qp_async_error_upcall(struct ib_event *event, void *context)
335 {
336 struct smbd_connection *info = context;
337
338 log_rdma_event(ERR, "%s on device %s info %p\n",
339 ib_event_msg(event->event), event->device->name, info);
340
341 switch (event->event) {
342 case IB_EVENT_CQ_ERR:
343 case IB_EVENT_QP_FATAL:
344 smbd_disconnect_rdma_connection(info);
345
346 default:
347 break;
348 }
349 }
350
351 static inline void *smbd_request_payload(struct smbd_request *request)
352 {
353 return (void *)request->packet;
354 }
355
356 static inline void *smbd_response_payload(struct smbd_response *response)
357 {
358 return (void *)response->packet;
359 }
360
361 /* Called when a RDMA send is done */
362 static void send_done(struct ib_cq *cq, struct ib_wc *wc)
363 {
364 int i;
365 struct smbd_request *request =
366 container_of(wc->wr_cqe, struct smbd_request, cqe);
367
368 log_rdma_send(INFO, "smbd_request %p completed wc->status=%d\n",
369 request, wc->status);
370
371 if (wc->status != IB_WC_SUCCESS || wc->opcode != IB_WC_SEND) {
372 log_rdma_send(ERR, "wc->status=%d wc->opcode=%d\n",
373 wc->status, wc->opcode);
374 smbd_disconnect_rdma_connection(request->info);
375 }
376
377 for (i = 0; i < request->num_sge; i++)
378 ib_dma_unmap_single(request->info->id->device,
379 request->sge[i].addr,
380 request->sge[i].length,
381 DMA_TO_DEVICE);
382
383 if (request->has_payload) {
384 if (atomic_dec_and_test(&request->info->send_payload_pending))
385 wake_up(&request->info->wait_send_payload_pending);
386 } else {
387 if (atomic_dec_and_test(&request->info->send_pending))
388 wake_up(&request->info->wait_send_pending);
389 }
390
391 mempool_free(request, request->info->request_mempool);
392 }
393
394 static void dump_smbd_negotiate_resp(struct smbd_negotiate_resp *resp)
395 {
396 log_rdma_event(INFO, "resp message min_version %u max_version %u "
397 "negotiated_version %u credits_requested %u "
398 "credits_granted %u status %u max_readwrite_size %u "
399 "preferred_send_size %u max_receive_size %u "
400 "max_fragmented_size %u\n",
401 resp->min_version, resp->max_version, resp->negotiated_version,
402 resp->credits_requested, resp->credits_granted, resp->status,
403 resp->max_readwrite_size, resp->preferred_send_size,
404 resp->max_receive_size, resp->max_fragmented_size);
405 }
406
407 /*
408 * Process a negotiation response message, according to [MS-SMBD]3.1.5.7
409 * response, packet_length: the negotiation response message
410 * return value: true if negotiation is a success, false if failed
411 */
412 static bool process_negotiation_response(
413 struct smbd_response *response, int packet_length)
414 {
415 struct smbd_connection *info = response->info;
416 struct smbd_negotiate_resp *packet = smbd_response_payload(response);
417
418 if (packet_length < sizeof(struct smbd_negotiate_resp)) {
419 log_rdma_event(ERR,
420 "error: packet_length=%d\n", packet_length);
421 return false;
422 }
423
424 if (le16_to_cpu(packet->negotiated_version) != SMBD_V1) {
425 log_rdma_event(ERR, "error: negotiated_version=%x\n",
426 le16_to_cpu(packet->negotiated_version));
427 return false;
428 }
429 info->protocol = le16_to_cpu(packet->negotiated_version);
430
431 if (packet->credits_requested == 0) {
432 log_rdma_event(ERR, "error: credits_requested==0\n");
433 return false;
434 }
435 info->receive_credit_target = le16_to_cpu(packet->credits_requested);
436
437 if (packet->credits_granted == 0) {
438 log_rdma_event(ERR, "error: credits_granted==0\n");
439 return false;
440 }
441 atomic_set(&info->send_credits, le16_to_cpu(packet->credits_granted));
442
443 atomic_set(&info->receive_credits, 0);
444
445 if (le32_to_cpu(packet->preferred_send_size) > info->max_receive_size) {
446 log_rdma_event(ERR, "error: preferred_send_size=%d\n",
447 le32_to_cpu(packet->preferred_send_size));
448 return false;
449 }
450 info->max_receive_size = le32_to_cpu(packet->preferred_send_size);
451
452 if (le32_to_cpu(packet->max_receive_size) < SMBD_MIN_RECEIVE_SIZE) {
453 log_rdma_event(ERR, "error: max_receive_size=%d\n",
454 le32_to_cpu(packet->max_receive_size));
455 return false;
456 }
457 info->max_send_size = min_t(int, info->max_send_size,
458 le32_to_cpu(packet->max_receive_size));
459
460 if (le32_to_cpu(packet->max_fragmented_size) <
461 SMBD_MIN_FRAGMENTED_SIZE) {
462 log_rdma_event(ERR, "error: max_fragmented_size=%d\n",
463 le32_to_cpu(packet->max_fragmented_size));
464 return false;
465 }
466 info->max_fragmented_send_size =
467 le32_to_cpu(packet->max_fragmented_size);
468 info->rdma_readwrite_threshold =
469 rdma_readwrite_threshold > info->max_fragmented_send_size ?
470 info->max_fragmented_send_size :
471 rdma_readwrite_threshold;
472
473
474 info->max_readwrite_size = min_t(u32,
475 le32_to_cpu(packet->max_readwrite_size),
476 info->max_frmr_depth * PAGE_SIZE);
477 info->max_frmr_depth = info->max_readwrite_size / PAGE_SIZE;
478
479 return true;
480 }
481
482 /*
483 * Check and schedule to send an immediate packet
484 * This is used to extend credtis to remote peer to keep the transport busy
485 */
486 static void check_and_send_immediate(struct smbd_connection *info)
487 {
488 if (info->transport_status != SMBD_CONNECTED)
489 return;
490
491 info->send_immediate = true;
492
493 /*
494 * Promptly send a packet if our peer is running low on receive
495 * credits
496 */
497 if (atomic_read(&info->receive_credits) <
498 info->receive_credit_target - 1)
499 queue_delayed_work(
500 info->workqueue, &info->send_immediate_work, 0);
501 }
502
503 static void smbd_post_send_credits(struct work_struct *work)
504 {
505 int ret = 0;
506 int use_receive_queue = 1;
507 int rc;
508 struct smbd_response *response;
509 struct smbd_connection *info =
510 container_of(work, struct smbd_connection,
511 post_send_credits_work);
512
513 if (info->transport_status != SMBD_CONNECTED) {
514 wake_up(&info->wait_receive_queues);
515 return;
516 }
517
518 if (info->receive_credit_target >
519 atomic_read(&info->receive_credits)) {
520 while (true) {
521 if (use_receive_queue)
522 response = get_receive_buffer(info);
523 else
524 response = get_empty_queue_buffer(info);
525 if (!response) {
526 /* now switch to emtpy packet queue */
527 if (use_receive_queue) {
528 use_receive_queue = 0;
529 continue;
530 } else
531 break;
532 }
533
534 response->type = SMBD_TRANSFER_DATA;
535 response->first_segment = false;
536 rc = smbd_post_recv(info, response);
537 if (rc) {
538 log_rdma_recv(ERR,
539 "post_recv failed rc=%d\n", rc);
540 put_receive_buffer(info, response);
541 break;
542 }
543
544 ret++;
545 }
546 }
547
548 spin_lock(&info->lock_new_credits_offered);
549 info->new_credits_offered += ret;
550 spin_unlock(&info->lock_new_credits_offered);
551
552 atomic_add(ret, &info->receive_credits);
553
554 /* Check if we can post new receive and grant credits to peer */
555 check_and_send_immediate(info);
556 }
557
558 static void smbd_recv_done_work(struct work_struct *work)
559 {
560 struct smbd_connection *info =
561 container_of(work, struct smbd_connection, recv_done_work);
562
563 /*
564 * We may have new send credits granted from remote peer
565 * If any sender is blcoked on lack of credets, unblock it
566 */
567 if (atomic_read(&info->send_credits))
568 wake_up_interruptible(&info->wait_send_queue);
569
570 /*
571 * Check if we need to send something to remote peer to
572 * grant more credits or respond to KEEP_ALIVE packet
573 */
574 check_and_send_immediate(info);
575 }
576
577 /* Called from softirq, when recv is done */
578 static void recv_done(struct ib_cq *cq, struct ib_wc *wc)
579 {
580 struct smbd_data_transfer *data_transfer;
581 struct smbd_response *response =
582 container_of(wc->wr_cqe, struct smbd_response, cqe);
583 struct smbd_connection *info = response->info;
584 int data_length = 0;
585
586 log_rdma_recv(INFO, "response=%p type=%d wc status=%d wc opcode %d "
587 "byte_len=%d pkey_index=%x\n",
588 response, response->type, wc->status, wc->opcode,
589 wc->byte_len, wc->pkey_index);
590
591 if (wc->status != IB_WC_SUCCESS || wc->opcode != IB_WC_RECV) {
592 log_rdma_recv(INFO, "wc->status=%d opcode=%d\n",
593 wc->status, wc->opcode);
594 smbd_disconnect_rdma_connection(info);
595 goto error;
596 }
597
598 ib_dma_sync_single_for_cpu(
599 wc->qp->device,
600 response->sge.addr,
601 response->sge.length,
602 DMA_FROM_DEVICE);
603
604 switch (response->type) {
605 /* SMBD negotiation response */
606 case SMBD_NEGOTIATE_RESP:
607 dump_smbd_negotiate_resp(smbd_response_payload(response));
608 info->full_packet_received = true;
609 info->negotiate_done =
610 process_negotiation_response(response, wc->byte_len);
611 complete(&info->negotiate_completion);
612 break;
613
614 /* SMBD data transfer packet */
615 case SMBD_TRANSFER_DATA:
616 data_transfer = smbd_response_payload(response);
617 data_length = le32_to_cpu(data_transfer->data_length);
618
619 /*
620 * If this is a packet with data playload place the data in
621 * reassembly queue and wake up the reading thread
622 */
623 if (data_length) {
624 if (info->full_packet_received)
625 response->first_segment = true;
626
627 if (le32_to_cpu(data_transfer->remaining_data_length))
628 info->full_packet_received = false;
629 else
630 info->full_packet_received = true;
631
632 enqueue_reassembly(
633 info,
634 response,
635 data_length);
636 } else
637 put_empty_packet(info, response);
638
639 if (data_length)
640 wake_up_interruptible(&info->wait_reassembly_queue);
641
642 atomic_dec(&info->receive_credits);
643 info->receive_credit_target =
644 le16_to_cpu(data_transfer->credits_requested);
645 atomic_add(le16_to_cpu(data_transfer->credits_granted),
646 &info->send_credits);
647
648 log_incoming(INFO, "data flags %d data_offset %d "
649 "data_length %d remaining_data_length %d\n",
650 le16_to_cpu(data_transfer->flags),
651 le32_to_cpu(data_transfer->data_offset),
652 le32_to_cpu(data_transfer->data_length),
653 le32_to_cpu(data_transfer->remaining_data_length));
654
655 /* Send a KEEP_ALIVE response right away if requested */
656 info->keep_alive_requested = KEEP_ALIVE_NONE;
657 if (le16_to_cpu(data_transfer->flags) &
658 SMB_DIRECT_RESPONSE_REQUESTED) {
659 info->keep_alive_requested = KEEP_ALIVE_PENDING;
660 }
661
662 queue_work(info->workqueue, &info->recv_done_work);
663 return;
664
665 default:
666 log_rdma_recv(ERR,
667 "unexpected response type=%d\n", response->type);
668 }
669
670 error:
671 put_receive_buffer(info, response);
672 }
673
674 static struct rdma_cm_id *smbd_create_id(
675 struct smbd_connection *info,
676 struct sockaddr *dstaddr, int port)
677 {
678 struct rdma_cm_id *id;
679 int rc;
680 __be16 *sport;
681
682 id = rdma_create_id(&init_net, smbd_conn_upcall, info,
683 RDMA_PS_TCP, IB_QPT_RC);
684 if (IS_ERR(id)) {
685 rc = PTR_ERR(id);
686 log_rdma_event(ERR, "rdma_create_id() failed %i\n", rc);
687 return id;
688 }
689
690 if (dstaddr->sa_family == AF_INET6)
691 sport = &((struct sockaddr_in6 *)dstaddr)->sin6_port;
692 else
693 sport = &((struct sockaddr_in *)dstaddr)->sin_port;
694
695 *sport = htons(port);
696
697 init_completion(&info->ri_done);
698 info->ri_rc = -ETIMEDOUT;
699
700 rc = rdma_resolve_addr(id, NULL, (struct sockaddr *)dstaddr,
701 RDMA_RESOLVE_TIMEOUT);
702 if (rc) {
703 log_rdma_event(ERR, "rdma_resolve_addr() failed %i\n", rc);
704 goto out;
705 }
706 wait_for_completion_interruptible_timeout(
707 &info->ri_done, msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT));
708 rc = info->ri_rc;
709 if (rc) {
710 log_rdma_event(ERR, "rdma_resolve_addr() completed %i\n", rc);
711 goto out;
712 }
713
714 info->ri_rc = -ETIMEDOUT;
715 rc = rdma_resolve_route(id, RDMA_RESOLVE_TIMEOUT);
716 if (rc) {
717 log_rdma_event(ERR, "rdma_resolve_route() failed %i\n", rc);
718 goto out;
719 }
720 wait_for_completion_interruptible_timeout(
721 &info->ri_done, msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT));
722 rc = info->ri_rc;
723 if (rc) {
724 log_rdma_event(ERR, "rdma_resolve_route() completed %i\n", rc);
725 goto out;
726 }
727
728 return id;
729
730 out:
731 rdma_destroy_id(id);
732 return ERR_PTR(rc);
733 }
734
735 /*
736 * Test if FRWR (Fast Registration Work Requests) is supported on the device
737 * This implementation requries FRWR on RDMA read/write
738 * return value: true if it is supported
739 */
740 static bool frwr_is_supported(struct ib_device_attr *attrs)
741 {
742 if (!(attrs->device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS))
743 return false;
744 if (attrs->max_fast_reg_page_list_len == 0)
745 return false;
746 return true;
747 }
748
749 static int smbd_ia_open(
750 struct smbd_connection *info,
751 struct sockaddr *dstaddr, int port)
752 {
753 int rc;
754
755 info->id = smbd_create_id(info, dstaddr, port);
756 if (IS_ERR(info->id)) {
757 rc = PTR_ERR(info->id);
758 goto out1;
759 }
760
761 if (!frwr_is_supported(&info->id->device->attrs)) {
762 log_rdma_event(ERR,
763 "Fast Registration Work Requests "
764 "(FRWR) is not supported\n");
765 log_rdma_event(ERR,
766 "Device capability flags = %llx "
767 "max_fast_reg_page_list_len = %u\n",
768 info->id->device->attrs.device_cap_flags,
769 info->id->device->attrs.max_fast_reg_page_list_len);
770 rc = -EPROTONOSUPPORT;
771 goto out2;
772 }
773 info->max_frmr_depth = min_t(int,
774 smbd_max_frmr_depth,
775 info->id->device->attrs.max_fast_reg_page_list_len);
776 info->mr_type = IB_MR_TYPE_MEM_REG;
777 if (info->id->device->attrs.device_cap_flags & IB_DEVICE_SG_GAPS_REG)
778 info->mr_type = IB_MR_TYPE_SG_GAPS;
779
780 info->pd = ib_alloc_pd(info->id->device, 0);
781 if (IS_ERR(info->pd)) {
782 rc = PTR_ERR(info->pd);
783 log_rdma_event(ERR, "ib_alloc_pd() returned %d\n", rc);
784 goto out2;
785 }
786
787 return 0;
788
789 out2:
790 rdma_destroy_id(info->id);
791 info->id = NULL;
792
793 out1:
794 return rc;
795 }
796
797 /*
798 * Send a negotiation request message to the peer
799 * The negotiation procedure is in [MS-SMBD] 3.1.5.2 and 3.1.5.3
800 * After negotiation, the transport is connected and ready for
801 * carrying upper layer SMB payload
802 */
803 static int smbd_post_send_negotiate_req(struct smbd_connection *info)
804 {
805 struct ib_send_wr send_wr;
806 int rc = -ENOMEM;
807 struct smbd_request *request;
808 struct smbd_negotiate_req *packet;
809
810 request = mempool_alloc(info->request_mempool, GFP_KERNEL);
811 if (!request)
812 return rc;
813
814 request->info = info;
815
816 packet = smbd_request_payload(request);
817 packet->min_version = cpu_to_le16(SMBD_V1);
818 packet->max_version = cpu_to_le16(SMBD_V1);
819 packet->reserved = 0;
820 packet->credits_requested = cpu_to_le16(info->send_credit_target);
821 packet->preferred_send_size = cpu_to_le32(info->max_send_size);
822 packet->max_receive_size = cpu_to_le32(info->max_receive_size);
823 packet->max_fragmented_size =
824 cpu_to_le32(info->max_fragmented_recv_size);
825
826 request->num_sge = 1;
827 request->sge[0].addr = ib_dma_map_single(
828 info->id->device, (void *)packet,
829 sizeof(*packet), DMA_TO_DEVICE);
830 if (ib_dma_mapping_error(info->id->device, request->sge[0].addr)) {
831 rc = -EIO;
832 goto dma_mapping_failed;
833 }
834
835 request->sge[0].length = sizeof(*packet);
836 request->sge[0].lkey = info->pd->local_dma_lkey;
837
838 ib_dma_sync_single_for_device(
839 info->id->device, request->sge[0].addr,
840 request->sge[0].length, DMA_TO_DEVICE);
841
842 request->cqe.done = send_done;
843
844 send_wr.next = NULL;
845 send_wr.wr_cqe = &request->cqe;
846 send_wr.sg_list = request->sge;
847 send_wr.num_sge = request->num_sge;
848 send_wr.opcode = IB_WR_SEND;
849 send_wr.send_flags = IB_SEND_SIGNALED;
850
851 log_rdma_send(INFO, "sge addr=%llx length=%x lkey=%x\n",
852 request->sge[0].addr,
853 request->sge[0].length, request->sge[0].lkey);
854
855 request->has_payload = false;
856 atomic_inc(&info->send_pending);
857 rc = ib_post_send(info->id->qp, &send_wr, NULL);
858 if (!rc)
859 return 0;
860
861 /* if we reach here, post send failed */
862 log_rdma_send(ERR, "ib_post_send failed rc=%d\n", rc);
863 atomic_dec(&info->send_pending);
864 ib_dma_unmap_single(info->id->device, request->sge[0].addr,
865 request->sge[0].length, DMA_TO_DEVICE);
866
867 smbd_disconnect_rdma_connection(info);
868
869 dma_mapping_failed:
870 mempool_free(request, info->request_mempool);
871 return rc;
872 }
873
874 /*
875 * Extend the credits to remote peer
876 * This implements [MS-SMBD] 3.1.5.9
877 * The idea is that we should extend credits to remote peer as quickly as
878 * it's allowed, to maintain data flow. We allocate as much receive
879 * buffer as possible, and extend the receive credits to remote peer
880 * return value: the new credtis being granted.
881 */
882 static int manage_credits_prior_sending(struct smbd_connection *info)
883 {
884 int new_credits;
885
886 spin_lock(&info->lock_new_credits_offered);
887 new_credits = info->new_credits_offered;
888 info->new_credits_offered = 0;
889 spin_unlock(&info->lock_new_credits_offered);
890
891 return new_credits;
892 }
893
894 /*
895 * Check if we need to send a KEEP_ALIVE message
896 * The idle connection timer triggers a KEEP_ALIVE message when expires
897 * SMB_DIRECT_RESPONSE_REQUESTED is set in the message flag to have peer send
898 * back a response.
899 * return value:
900 * 1 if SMB_DIRECT_RESPONSE_REQUESTED needs to be set
901 * 0: otherwise
902 */
903 static int manage_keep_alive_before_sending(struct smbd_connection *info)
904 {
905 if (info->keep_alive_requested == KEEP_ALIVE_PENDING) {
906 info->keep_alive_requested = KEEP_ALIVE_SENT;
907 return 1;
908 }
909 return 0;
910 }
911
912 /*
913 * Build and prepare the SMBD packet header
914 * This function waits for avaialbe send credits and build a SMBD packet
915 * header. The caller then optional append payload to the packet after
916 * the header
917 * intput values
918 * size: the size of the payload
919 * remaining_data_length: remaining data to send if this is part of a
920 * fragmented packet
921 * output values
922 * request_out: the request allocated from this function
923 * return values: 0 on success, otherwise actual error code returned
924 */
925 static int smbd_create_header(struct smbd_connection *info,
926 int size, int remaining_data_length,
927 struct smbd_request **request_out)
928 {
929 struct smbd_request *request;
930 struct smbd_data_transfer *packet;
931 int header_length;
932 int rc;
933
934 /* Wait for send credits. A SMBD packet needs one credit */
935 rc = wait_event_interruptible(info->wait_send_queue,
936 atomic_read(&info->send_credits) > 0 ||
937 info->transport_status != SMBD_CONNECTED);
938 if (rc)
939 return rc;
940
941 if (info->transport_status != SMBD_CONNECTED) {
942 log_outgoing(ERR, "disconnected not sending\n");
943 return -ENOENT;
944 }
945 atomic_dec(&info->send_credits);
946
947 request = mempool_alloc(info->request_mempool, GFP_KERNEL);
948 if (!request) {
949 rc = -ENOMEM;
950 goto err;
951 }
952
953 request->info = info;
954
955 /* Fill in the packet header */
956 packet = smbd_request_payload(request);
957 packet->credits_requested = cpu_to_le16(info->send_credit_target);
958 packet->credits_granted =
959 cpu_to_le16(manage_credits_prior_sending(info));
960 info->send_immediate = false;
961
962 packet->flags = 0;
963 if (manage_keep_alive_before_sending(info))
964 packet->flags |= cpu_to_le16(SMB_DIRECT_RESPONSE_REQUESTED);
965
966 packet->reserved = 0;
967 if (!size)
968 packet->data_offset = 0;
969 else
970 packet->data_offset = cpu_to_le32(24);
971 packet->data_length = cpu_to_le32(size);
972 packet->remaining_data_length = cpu_to_le32(remaining_data_length);
973 packet->padding = 0;
974
975 log_outgoing(INFO, "credits_requested=%d credits_granted=%d "
976 "data_offset=%d data_length=%d remaining_data_length=%d\n",
977 le16_to_cpu(packet->credits_requested),
978 le16_to_cpu(packet->credits_granted),
979 le32_to_cpu(packet->data_offset),
980 le32_to_cpu(packet->data_length),
981 le32_to_cpu(packet->remaining_data_length));
982
983 /* Map the packet to DMA */
984 header_length = sizeof(struct smbd_data_transfer);
985 /* If this is a packet without payload, don't send padding */
986 if (!size)
987 header_length = offsetof(struct smbd_data_transfer, padding);
988
989 request->num_sge = 1;
990 request->sge[0].addr = ib_dma_map_single(info->id->device,
991 (void *)packet,
992 header_length,
993 DMA_BIDIRECTIONAL);
994 if (ib_dma_mapping_error(info->id->device, request->sge[0].addr)) {
995 mempool_free(request, info->request_mempool);
996 rc = -EIO;
997 goto err;
998 }
999
1000 request->sge[0].length = header_length;
1001 request->sge[0].lkey = info->pd->local_dma_lkey;
1002
1003 *request_out = request;
1004 return 0;
1005
1006 err:
1007 atomic_inc(&info->send_credits);
1008 return rc;
1009 }
1010
1011 static void smbd_destroy_header(struct smbd_connection *info,
1012 struct smbd_request *request)
1013 {
1014
1015 ib_dma_unmap_single(info->id->device,
1016 request->sge[0].addr,
1017 request->sge[0].length,
1018 DMA_TO_DEVICE);
1019 mempool_free(request, info->request_mempool);
1020 atomic_inc(&info->send_credits);
1021 }
1022
1023 /* Post the send request */
1024 static int smbd_post_send(struct smbd_connection *info,
1025 struct smbd_request *request, bool has_payload)
1026 {
1027 struct ib_send_wr send_wr;
1028 int rc, i;
1029
1030 for (i = 0; i < request->num_sge; i++) {
1031 log_rdma_send(INFO,
1032 "rdma_request sge[%d] addr=%llu length=%u\n",
1033 i, request->sge[i].addr, request->sge[i].length);
1034 ib_dma_sync_single_for_device(
1035 info->id->device,
1036 request->sge[i].addr,
1037 request->sge[i].length,
1038 DMA_TO_DEVICE);
1039 }
1040
1041 request->cqe.done = send_done;
1042
1043 send_wr.next = NULL;
1044 send_wr.wr_cqe = &request->cqe;
1045 send_wr.sg_list = request->sge;
1046 send_wr.num_sge = request->num_sge;
1047 send_wr.opcode = IB_WR_SEND;
1048 send_wr.send_flags = IB_SEND_SIGNALED;
1049
1050 if (has_payload) {
1051 request->has_payload = true;
1052 atomic_inc(&info->send_payload_pending);
1053 } else {
1054 request->has_payload = false;
1055 atomic_inc(&info->send_pending);
1056 }
1057
1058 rc = ib_post_send(info->id->qp, &send_wr, NULL);
1059 if (rc) {
1060 log_rdma_send(ERR, "ib_post_send failed rc=%d\n", rc);
1061 if (has_payload) {
1062 if (atomic_dec_and_test(&info->send_payload_pending))
1063 wake_up(&info->wait_send_payload_pending);
1064 } else {
1065 if (atomic_dec_and_test(&info->send_pending))
1066 wake_up(&info->wait_send_pending);
1067 }
1068 smbd_disconnect_rdma_connection(info);
1069 } else
1070 /* Reset timer for idle connection after packet is sent */
1071 mod_delayed_work(info->workqueue, &info->idle_timer_work,
1072 info->keep_alive_interval*HZ);
1073
1074 return rc;
1075 }
1076
1077 static int smbd_post_send_sgl(struct smbd_connection *info,
1078 struct scatterlist *sgl, int data_length, int remaining_data_length)
1079 {
1080 int num_sgs;
1081 int i, rc;
1082 struct smbd_request *request;
1083 struct scatterlist *sg;
1084
1085 rc = smbd_create_header(
1086 info, data_length, remaining_data_length, &request);
1087 if (rc)
1088 return rc;
1089
1090 num_sgs = sgl ? sg_nents(sgl) : 0;
1091 for_each_sg(sgl, sg, num_sgs, i) {
1092 request->sge[i+1].addr =
1093 ib_dma_map_page(info->id->device, sg_page(sg),
1094 sg->offset, sg->length, DMA_BIDIRECTIONAL);
1095 if (ib_dma_mapping_error(
1096 info->id->device, request->sge[i+1].addr)) {
1097 rc = -EIO;
1098 request->sge[i+1].addr = 0;
1099 goto dma_mapping_failure;
1100 }
1101 request->sge[i+1].length = sg->length;
1102 request->sge[i+1].lkey = info->pd->local_dma_lkey;
1103 request->num_sge++;
1104 }
1105
1106 rc = smbd_post_send(info, request, data_length);
1107 if (!rc)
1108 return 0;
1109
1110 dma_mapping_failure:
1111 for (i = 1; i < request->num_sge; i++)
1112 if (request->sge[i].addr)
1113 ib_dma_unmap_single(info->id->device,
1114 request->sge[i].addr,
1115 request->sge[i].length,
1116 DMA_TO_DEVICE);
1117 smbd_destroy_header(info, request);
1118 return rc;
1119 }
1120
1121 /*
1122 * Send a page
1123 * page: the page to send
1124 * offset: offset in the page to send
1125 * size: length in the page to send
1126 * remaining_data_length: remaining data to send in this payload
1127 */
1128 static int smbd_post_send_page(struct smbd_connection *info, struct page *page,
1129 unsigned long offset, size_t size, int remaining_data_length)
1130 {
1131 struct scatterlist sgl;
1132
1133 sg_init_table(&sgl, 1);
1134 sg_set_page(&sgl, page, size, offset);
1135
1136 return smbd_post_send_sgl(info, &sgl, size, remaining_data_length);
1137 }
1138
1139 /*
1140 * Send an empty message
1141 * Empty message is used to extend credits to peer to for keep live
1142 * while there is no upper layer payload to send at the time
1143 */
1144 static int smbd_post_send_empty(struct smbd_connection *info)
1145 {
1146 info->count_send_empty++;
1147 return smbd_post_send_sgl(info, NULL, 0, 0);
1148 }
1149
1150 /*
1151 * Send a data buffer
1152 * iov: the iov array describing the data buffers
1153 * n_vec: number of iov array
1154 * remaining_data_length: remaining data to send following this packet
1155 * in segmented SMBD packet
1156 */
1157 static int smbd_post_send_data(
1158 struct smbd_connection *info, struct kvec *iov, int n_vec,
1159 int remaining_data_length)
1160 {
1161 int i;
1162 u32 data_length = 0;
1163 struct scatterlist sgl[SMBDIRECT_MAX_SGE];
1164
1165 if (n_vec > SMBDIRECT_MAX_SGE) {
1166 cifs_dbg(VFS, "Can't fit data to SGL, n_vec=%d\n", n_vec);
1167 return -ENOMEM;
1168 }
1169
1170 sg_init_table(sgl, n_vec);
1171 for (i = 0; i < n_vec; i++) {
1172 data_length += iov[i].iov_len;
1173 sg_set_buf(&sgl[i], iov[i].iov_base, iov[i].iov_len);
1174 }
1175
1176 return smbd_post_send_sgl(info, sgl, data_length, remaining_data_length);
1177 }
1178
1179 /*
1180 * Post a receive request to the transport
1181 * The remote peer can only send data when a receive request is posted
1182 * The interaction is controlled by send/receive credit system
1183 */
1184 static int smbd_post_recv(
1185 struct smbd_connection *info, struct smbd_response *response)
1186 {
1187 struct ib_recv_wr recv_wr;
1188 int rc = -EIO;
1189
1190 response->sge.addr = ib_dma_map_single(
1191 info->id->device, response->packet,
1192 info->max_receive_size, DMA_FROM_DEVICE);
1193 if (ib_dma_mapping_error(info->id->device, response->sge.addr))
1194 return rc;
1195
1196 response->sge.length = info->max_receive_size;
1197 response->sge.lkey = info->pd->local_dma_lkey;
1198
1199 response->cqe.done = recv_done;
1200
1201 recv_wr.wr_cqe = &response->cqe;
1202 recv_wr.next = NULL;
1203 recv_wr.sg_list = &response->sge;
1204 recv_wr.num_sge = 1;
1205
1206 rc = ib_post_recv(info->id->qp, &recv_wr, NULL);
1207 if (rc) {
1208 ib_dma_unmap_single(info->id->device, response->sge.addr,
1209 response->sge.length, DMA_FROM_DEVICE);
1210 smbd_disconnect_rdma_connection(info);
1211 log_rdma_recv(ERR, "ib_post_recv failed rc=%d\n", rc);
1212 }
1213
1214 return rc;
1215 }
1216
1217 /* Perform SMBD negotiate according to [MS-SMBD] 3.1.5.2 */
1218 static int smbd_negotiate(struct smbd_connection *info)
1219 {
1220 int rc;
1221 struct smbd_response *response = get_receive_buffer(info);
1222
1223 response->type = SMBD_NEGOTIATE_RESP;
1224 rc = smbd_post_recv(info, response);
1225 log_rdma_event(INFO,
1226 "smbd_post_recv rc=%d iov.addr=%llx iov.length=%x "
1227 "iov.lkey=%x\n",
1228 rc, response->sge.addr,
1229 response->sge.length, response->sge.lkey);
1230 if (rc)
1231 return rc;
1232
1233 init_completion(&info->negotiate_completion);
1234 info->negotiate_done = false;
1235 rc = smbd_post_send_negotiate_req(info);
1236 if (rc)
1237 return rc;
1238
1239 rc = wait_for_completion_interruptible_timeout(
1240 &info->negotiate_completion, SMBD_NEGOTIATE_TIMEOUT * HZ);
1241 log_rdma_event(INFO, "wait_for_completion_timeout rc=%d\n", rc);
1242
1243 if (info->negotiate_done)
1244 return 0;
1245
1246 if (rc == 0)
1247 rc = -ETIMEDOUT;
1248 else if (rc == -ERESTARTSYS)
1249 rc = -EINTR;
1250 else
1251 rc = -ENOTCONN;
1252
1253 return rc;
1254 }
1255
1256 static void put_empty_packet(
1257 struct smbd_connection *info, struct smbd_response *response)
1258 {
1259 spin_lock(&info->empty_packet_queue_lock);
1260 list_add_tail(&response->list, &info->empty_packet_queue);
1261 info->count_empty_packet_queue++;
1262 spin_unlock(&info->empty_packet_queue_lock);
1263
1264 queue_work(info->workqueue, &info->post_send_credits_work);
1265 }
1266
1267 /*
1268 * Implement Connection.FragmentReassemblyBuffer defined in [MS-SMBD] 3.1.1.1
1269 * This is a queue for reassembling upper layer payload and present to upper
1270 * layer. All the inncoming payload go to the reassembly queue, regardless of
1271 * if reassembly is required. The uuper layer code reads from the queue for all
1272 * incoming payloads.
1273 * Put a received packet to the reassembly queue
1274 * response: the packet received
1275 * data_length: the size of payload in this packet
1276 */
1277 static void enqueue_reassembly(
1278 struct smbd_connection *info,
1279 struct smbd_response *response,
1280 int data_length)
1281 {
1282 spin_lock(&info->reassembly_queue_lock);
1283 list_add_tail(&response->list, &info->reassembly_queue);
1284 info->reassembly_queue_length++;
1285 /*
1286 * Make sure reassembly_data_length is updated after list and
1287 * reassembly_queue_length are updated. On the dequeue side
1288 * reassembly_data_length is checked without a lock to determine
1289 * if reassembly_queue_length and list is up to date
1290 */
1291 virt_wmb();
1292 info->reassembly_data_length += data_length;
1293 spin_unlock(&info->reassembly_queue_lock);
1294 info->count_reassembly_queue++;
1295 info->count_enqueue_reassembly_queue++;
1296 }
1297
1298 /*
1299 * Get the first entry at the front of reassembly queue
1300 * Caller is responsible for locking
1301 * return value: the first entry if any, NULL if queue is empty
1302 */
1303 static struct smbd_response *_get_first_reassembly(struct smbd_connection *info)
1304 {
1305 struct smbd_response *ret = NULL;
1306
1307 if (!list_empty(&info->reassembly_queue)) {
1308 ret = list_first_entry(
1309 &info->reassembly_queue,
1310 struct smbd_response, list);
1311 }
1312 return ret;
1313 }
1314
1315 static struct smbd_response *get_empty_queue_buffer(
1316 struct smbd_connection *info)
1317 {
1318 struct smbd_response *ret = NULL;
1319 unsigned long flags;
1320
1321 spin_lock_irqsave(&info->empty_packet_queue_lock, flags);
1322 if (!list_empty(&info->empty_packet_queue)) {
1323 ret = list_first_entry(
1324 &info->empty_packet_queue,
1325 struct smbd_response, list);
1326 list_del(&ret->list);
1327 info->count_empty_packet_queue--;
1328 }
1329 spin_unlock_irqrestore(&info->empty_packet_queue_lock, flags);
1330
1331 return ret;
1332 }
1333
1334 /*
1335 * Get a receive buffer
1336 * For each remote send, we need to post a receive. The receive buffers are
1337 * pre-allocated in advance.
1338 * return value: the receive buffer, NULL if none is available
1339 */
1340 static struct smbd_response *get_receive_buffer(struct smbd_connection *info)
1341 {
1342 struct smbd_response *ret = NULL;
1343 unsigned long flags;
1344
1345 spin_lock_irqsave(&info->receive_queue_lock, flags);
1346 if (!list_empty(&info->receive_queue)) {
1347 ret = list_first_entry(
1348 &info->receive_queue,
1349 struct smbd_response, list);
1350 list_del(&ret->list);
1351 info->count_receive_queue--;
1352 info->count_get_receive_buffer++;
1353 }
1354 spin_unlock_irqrestore(&info->receive_queue_lock, flags);
1355
1356 return ret;
1357 }
1358
1359 /*
1360 * Return a receive buffer
1361 * Upon returning of a receive buffer, we can post new receive and extend
1362 * more receive credits to remote peer. This is done immediately after a
1363 * receive buffer is returned.
1364 */
1365 static void put_receive_buffer(
1366 struct smbd_connection *info, struct smbd_response *response)
1367 {
1368 unsigned long flags;
1369
1370 ib_dma_unmap_single(info->id->device, response->sge.addr,
1371 response->sge.length, DMA_FROM_DEVICE);
1372
1373 spin_lock_irqsave(&info->receive_queue_lock, flags);
1374 list_add_tail(&response->list, &info->receive_queue);
1375 info->count_receive_queue++;
1376 info->count_put_receive_buffer++;
1377 spin_unlock_irqrestore(&info->receive_queue_lock, flags);
1378
1379 queue_work(info->workqueue, &info->post_send_credits_work);
1380 }
1381
1382 /* Preallocate all receive buffer on transport establishment */
1383 static int allocate_receive_buffers(struct smbd_connection *info, int num_buf)
1384 {
1385 int i;
1386 struct smbd_response *response;
1387
1388 INIT_LIST_HEAD(&info->reassembly_queue);
1389 spin_lock_init(&info->reassembly_queue_lock);
1390 info->reassembly_data_length = 0;
1391 info->reassembly_queue_length = 0;
1392
1393 INIT_LIST_HEAD(&info->receive_queue);
1394 spin_lock_init(&info->receive_queue_lock);
1395 info->count_receive_queue = 0;
1396
1397 INIT_LIST_HEAD(&info->empty_packet_queue);
1398 spin_lock_init(&info->empty_packet_queue_lock);
1399 info->count_empty_packet_queue = 0;
1400
1401 init_waitqueue_head(&info->wait_receive_queues);
1402
1403 for (i = 0; i < num_buf; i++) {
1404 response = mempool_alloc(info->response_mempool, GFP_KERNEL);
1405 if (!response)
1406 goto allocate_failed;
1407
1408 response->info = info;
1409 list_add_tail(&response->list, &info->receive_queue);
1410 info->count_receive_queue++;
1411 }
1412
1413 return 0;
1414
1415 allocate_failed:
1416 while (!list_empty(&info->receive_queue)) {
1417 response = list_first_entry(
1418 &info->receive_queue,
1419 struct smbd_response, list);
1420 list_del(&response->list);
1421 info->count_receive_queue--;
1422
1423 mempool_free(response, info->response_mempool);
1424 }
1425 return -ENOMEM;
1426 }
1427
1428 static void destroy_receive_buffers(struct smbd_connection *info)
1429 {
1430 struct smbd_response *response;
1431
1432 while ((response = get_receive_buffer(info)))
1433 mempool_free(response, info->response_mempool);
1434
1435 while ((response = get_empty_queue_buffer(info)))
1436 mempool_free(response, info->response_mempool);
1437 }
1438
1439 /*
1440 * Check and send an immediate or keep alive packet
1441 * The condition to send those packets are defined in [MS-SMBD] 3.1.1.1
1442 * Connection.KeepaliveRequested and Connection.SendImmediate
1443 * The idea is to extend credits to server as soon as it becomes available
1444 */
1445 static void send_immediate_work(struct work_struct *work)
1446 {
1447 struct smbd_connection *info = container_of(
1448 work, struct smbd_connection,
1449 send_immediate_work.work);
1450
1451 if (info->keep_alive_requested == KEEP_ALIVE_PENDING ||
1452 info->send_immediate) {
1453 log_keep_alive(INFO, "send an empty message\n");
1454 smbd_post_send_empty(info);
1455 }
1456 }
1457
1458 /* Implement idle connection timer [MS-SMBD] 3.1.6.2 */
1459 static void idle_connection_timer(struct work_struct *work)
1460 {
1461 struct smbd_connection *info = container_of(
1462 work, struct smbd_connection,
1463 idle_timer_work.work);
1464
1465 if (info->keep_alive_requested != KEEP_ALIVE_NONE) {
1466 log_keep_alive(ERR,
1467 "error status info->keep_alive_requested=%d\n",
1468 info->keep_alive_requested);
1469 smbd_disconnect_rdma_connection(info);
1470 return;
1471 }
1472
1473 log_keep_alive(INFO, "about to send an empty idle message\n");
1474 smbd_post_send_empty(info);
1475
1476 /* Setup the next idle timeout work */
1477 queue_delayed_work(info->workqueue, &info->idle_timer_work,
1478 info->keep_alive_interval*HZ);
1479 }
1480
1481 /* Destroy this SMBD connection, called from upper layer */
1482 void smbd_destroy(struct smbd_connection *info)
1483 {
1484 log_rdma_event(INFO, "destroying rdma session\n");
1485
1486 /* Kick off the disconnection process */
1487 smbd_disconnect_rdma_connection(info);
1488
1489 log_rdma_event(INFO, "wait for transport being destroyed\n");
1490 wait_event(info->wait_destroy,
1491 info->transport_status == SMBD_DESTROYED);
1492
1493 destroy_workqueue(info->workqueue);
1494 kfree(info);
1495 }
1496
1497 /*
1498 * Reconnect this SMBD connection, called from upper layer
1499 * return value: 0 on success, or actual error code
1500 */
1501 int smbd_reconnect(struct TCP_Server_Info *server)
1502 {
1503 log_rdma_event(INFO, "reconnecting rdma session\n");
1504
1505 if (!server->smbd_conn) {
1506 log_rdma_event(INFO, "rdma session already destroyed\n");
1507 goto create_conn;
1508 }
1509
1510 /*
1511 * This is possible if transport is disconnected and we haven't received
1512 * notification from RDMA, but upper layer has detected timeout
1513 */
1514 if (server->smbd_conn->transport_status == SMBD_CONNECTED) {
1515 log_rdma_event(INFO, "disconnecting transport\n");
1516 smbd_disconnect_rdma_connection(server->smbd_conn);
1517 }
1518
1519 /* wait until the transport is destroyed */
1520 if (!wait_event_timeout(server->smbd_conn->wait_destroy,
1521 server->smbd_conn->transport_status == SMBD_DESTROYED, 5*HZ))
1522 return -EAGAIN;
1523
1524 destroy_workqueue(server->smbd_conn->workqueue);
1525 kfree(server->smbd_conn);
1526
1527 create_conn:
1528 log_rdma_event(INFO, "creating rdma session\n");
1529 server->smbd_conn = smbd_get_connection(
1530 server, (struct sockaddr *) &server->dstaddr);
1531 log_rdma_event(INFO, "created rdma session info=%p\n",
1532 server->smbd_conn);
1533
1534 return server->smbd_conn ? 0 : -ENOENT;
1535 }
1536
1537 static void destroy_caches_and_workqueue(struct smbd_connection *info)
1538 {
1539 destroy_receive_buffers(info);
1540 destroy_workqueue(info->workqueue);
1541 mempool_destroy(info->response_mempool);
1542 kmem_cache_destroy(info->response_cache);
1543 mempool_destroy(info->request_mempool);
1544 kmem_cache_destroy(info->request_cache);
1545 }
1546
1547 #define MAX_NAME_LEN 80
1548 static int allocate_caches_and_workqueue(struct smbd_connection *info)
1549 {
1550 char name[MAX_NAME_LEN];
1551 int rc;
1552
1553 snprintf(name, MAX_NAME_LEN, "smbd_request_%p", info);
1554 info->request_cache =
1555 kmem_cache_create(
1556 name,
1557 sizeof(struct smbd_request) +
1558 sizeof(struct smbd_data_transfer),
1559 0, SLAB_HWCACHE_ALIGN, NULL);
1560 if (!info->request_cache)
1561 return -ENOMEM;
1562
1563 info->request_mempool =
1564 mempool_create(info->send_credit_target, mempool_alloc_slab,
1565 mempool_free_slab, info->request_cache);
1566 if (!info->request_mempool)
1567 goto out1;
1568
1569 snprintf(name, MAX_NAME_LEN, "smbd_response_%p", info);
1570 info->response_cache =
1571 kmem_cache_create(
1572 name,
1573 sizeof(struct smbd_response) +
1574 info->max_receive_size,
1575 0, SLAB_HWCACHE_ALIGN, NULL);
1576 if (!info->response_cache)
1577 goto out2;
1578
1579 info->response_mempool =
1580 mempool_create(info->receive_credit_max, mempool_alloc_slab,
1581 mempool_free_slab, info->response_cache);
1582 if (!info->response_mempool)
1583 goto out3;
1584
1585 snprintf(name, MAX_NAME_LEN, "smbd_%p", info);
1586 info->workqueue = create_workqueue(name);
1587 if (!info->workqueue)
1588 goto out4;
1589
1590 rc = allocate_receive_buffers(info, info->receive_credit_max);
1591 if (rc) {
1592 log_rdma_event(ERR, "failed to allocate receive buffers\n");
1593 goto out5;
1594 }
1595
1596 return 0;
1597
1598 out5:
1599 destroy_workqueue(info->workqueue);
1600 out4:
1601 mempool_destroy(info->response_mempool);
1602 out3:
1603 kmem_cache_destroy(info->response_cache);
1604 out2:
1605 mempool_destroy(info->request_mempool);
1606 out1:
1607 kmem_cache_destroy(info->request_cache);
1608 return -ENOMEM;
1609 }
1610
1611 /* Create a SMBD connection, called by upper layer */
1612 static struct smbd_connection *_smbd_get_connection(
1613 struct TCP_Server_Info *server, struct sockaddr *dstaddr, int port)
1614 {
1615 int rc;
1616 struct smbd_connection *info;
1617 struct rdma_conn_param conn_param;
1618 struct ib_qp_init_attr qp_attr;
1619 struct sockaddr_in *addr_in = (struct sockaddr_in *) dstaddr;
1620 struct ib_port_immutable port_immutable;
1621 u32 ird_ord_hdr[2];
1622
1623 info = kzalloc(sizeof(struct smbd_connection), GFP_KERNEL);
1624 if (!info)
1625 return NULL;
1626
1627 info->transport_status = SMBD_CONNECTING;
1628 rc = smbd_ia_open(info, dstaddr, port);
1629 if (rc) {
1630 log_rdma_event(INFO, "smbd_ia_open rc=%d\n", rc);
1631 goto create_id_failed;
1632 }
1633
1634 if (smbd_send_credit_target > info->id->device->attrs.max_cqe ||
1635 smbd_send_credit_target > info->id->device->attrs.max_qp_wr) {
1636 log_rdma_event(ERR,
1637 "consider lowering send_credit_target = %d. "
1638 "Possible CQE overrun, device "
1639 "reporting max_cpe %d max_qp_wr %d\n",
1640 smbd_send_credit_target,
1641 info->id->device->attrs.max_cqe,
1642 info->id->device->attrs.max_qp_wr);
1643 goto config_failed;
1644 }
1645
1646 if (smbd_receive_credit_max > info->id->device->attrs.max_cqe ||
1647 smbd_receive_credit_max > info->id->device->attrs.max_qp_wr) {
1648 log_rdma_event(ERR,
1649 "consider lowering receive_credit_max = %d. "
1650 "Possible CQE overrun, device "
1651 "reporting max_cpe %d max_qp_wr %d\n",
1652 smbd_receive_credit_max,
1653 info->id->device->attrs.max_cqe,
1654 info->id->device->attrs.max_qp_wr);
1655 goto config_failed;
1656 }
1657
1658 info->receive_credit_max = smbd_receive_credit_max;
1659 info->send_credit_target = smbd_send_credit_target;
1660 info->max_send_size = smbd_max_send_size;
1661 info->max_fragmented_recv_size = smbd_max_fragmented_recv_size;
1662 info->max_receive_size = smbd_max_receive_size;
1663 info->keep_alive_interval = smbd_keep_alive_interval;
1664
1665 if (info->id->device->attrs.max_send_sge < SMBDIRECT_MAX_SGE) {
1666 log_rdma_event(ERR,
1667 "warning: device max_send_sge = %d too small\n",
1668 info->id->device->attrs.max_send_sge);
1669 log_rdma_event(ERR, "Queue Pair creation may fail\n");
1670 }
1671 if (info->id->device->attrs.max_recv_sge < SMBDIRECT_MAX_SGE) {
1672 log_rdma_event(ERR,
1673 "warning: device max_recv_sge = %d too small\n",
1674 info->id->device->attrs.max_recv_sge);
1675 log_rdma_event(ERR, "Queue Pair creation may fail\n");
1676 }
1677
1678 info->send_cq = NULL;
1679 info->recv_cq = NULL;
1680 info->send_cq = ib_alloc_cq(info->id->device, info,
1681 info->send_credit_target, 0, IB_POLL_SOFTIRQ);
1682 if (IS_ERR(info->send_cq)) {
1683 info->send_cq = NULL;
1684 goto alloc_cq_failed;
1685 }
1686
1687 info->recv_cq = ib_alloc_cq(info->id->device, info,
1688 info->receive_credit_max, 0, IB_POLL_SOFTIRQ);
1689 if (IS_ERR(info->recv_cq)) {
1690 info->recv_cq = NULL;
1691 goto alloc_cq_failed;
1692 }
1693
1694 memset(&qp_attr, 0, sizeof(qp_attr));
1695 qp_attr.event_handler = smbd_qp_async_error_upcall;
1696 qp_attr.qp_context = info;
1697 qp_attr.cap.max_send_wr = info->send_credit_target;
1698 qp_attr.cap.max_recv_wr = info->receive_credit_max;
1699 qp_attr.cap.max_send_sge = SMBDIRECT_MAX_SGE;
1700 qp_attr.cap.max_recv_sge = SMBDIRECT_MAX_SGE;
1701 qp_attr.cap.max_inline_data = 0;
1702 qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
1703 qp_attr.qp_type = IB_QPT_RC;
1704 qp_attr.send_cq = info->send_cq;
1705 qp_attr.recv_cq = info->recv_cq;
1706 qp_attr.port_num = ~0;
1707
1708 rc = rdma_create_qp(info->id, info->pd, &qp_attr);
1709 if (rc) {
1710 log_rdma_event(ERR, "rdma_create_qp failed %i\n", rc);
1711 goto create_qp_failed;
1712 }
1713
1714 memset(&conn_param, 0, sizeof(conn_param));
1715 conn_param.initiator_depth = 0;
1716
1717 conn_param.responder_resources =
1718 info->id->device->attrs.max_qp_rd_atom
1719 < SMBD_CM_RESPONDER_RESOURCES ?
1720 info->id->device->attrs.max_qp_rd_atom :
1721 SMBD_CM_RESPONDER_RESOURCES;
1722 info->responder_resources = conn_param.responder_resources;
1723 log_rdma_mr(INFO, "responder_resources=%d\n",
1724 info->responder_resources);
1725
1726 /* Need to send IRD/ORD in private data for iWARP */
1727 info->id->device->get_port_immutable(
1728 info->id->device, info->id->port_num, &port_immutable);
1729 if (port_immutable.core_cap_flags & RDMA_CORE_PORT_IWARP) {
1730 ird_ord_hdr[0] = info->responder_resources;
1731 ird_ord_hdr[1] = 1;
1732 conn_param.private_data = ird_ord_hdr;
1733 conn_param.private_data_len = sizeof(ird_ord_hdr);
1734 } else {
1735 conn_param.private_data = NULL;
1736 conn_param.private_data_len = 0;
1737 }
1738
1739 conn_param.retry_count = SMBD_CM_RETRY;
1740 conn_param.rnr_retry_count = SMBD_CM_RNR_RETRY;
1741 conn_param.flow_control = 0;
1742 init_waitqueue_head(&info->wait_destroy);
1743
1744 log_rdma_event(INFO, "connecting to IP %pI4 port %d\n",
1745 &addr_in->sin_addr, port);
1746
1747 init_waitqueue_head(&info->conn_wait);
1748 rc = rdma_connect(info->id, &conn_param);
1749 if (rc) {
1750 log_rdma_event(ERR, "rdma_connect() failed with %i\n", rc);
1751 goto rdma_connect_failed;
1752 }
1753
1754 wait_event_interruptible(
1755 info->conn_wait, info->transport_status != SMBD_CONNECTING);
1756
1757 if (info->transport_status != SMBD_CONNECTED) {
1758 log_rdma_event(ERR, "rdma_connect failed port=%d\n", port);
1759 goto rdma_connect_failed;
1760 }
1761
1762 log_rdma_event(INFO, "rdma_connect connected\n");
1763
1764 rc = allocate_caches_and_workqueue(info);
1765 if (rc) {
1766 log_rdma_event(ERR, "cache allocation failed\n");
1767 goto allocate_cache_failed;
1768 }
1769
1770 init_waitqueue_head(&info->wait_send_queue);
1771 init_waitqueue_head(&info->wait_reassembly_queue);
1772
1773 INIT_DELAYED_WORK(&info->idle_timer_work, idle_connection_timer);
1774 INIT_DELAYED_WORK(&info->send_immediate_work, send_immediate_work);
1775 queue_delayed_work(info->workqueue, &info->idle_timer_work,
1776 info->keep_alive_interval*HZ);
1777
1778 init_waitqueue_head(&info->wait_smbd_send_pending);
1779 info->smbd_send_pending = 0;
1780
1781 init_waitqueue_head(&info->wait_smbd_recv_pending);
1782 info->smbd_recv_pending = 0;
1783
1784 init_waitqueue_head(&info->wait_send_pending);
1785 atomic_set(&info->send_pending, 0);
1786
1787 init_waitqueue_head(&info->wait_send_payload_pending);
1788 atomic_set(&info->send_payload_pending, 0);
1789
1790 INIT_WORK(&info->disconnect_work, smbd_disconnect_rdma_work);
1791 INIT_WORK(&info->destroy_work, smbd_destroy_rdma_work);
1792 INIT_WORK(&info->recv_done_work, smbd_recv_done_work);
1793 INIT_WORK(&info->post_send_credits_work, smbd_post_send_credits);
1794 info->new_credits_offered = 0;
1795 spin_lock_init(&info->lock_new_credits_offered);
1796
1797 rc = smbd_negotiate(info);
1798 if (rc) {
1799 log_rdma_event(ERR, "smbd_negotiate rc=%d\n", rc);
1800 goto negotiation_failed;
1801 }
1802
1803 rc = allocate_mr_list(info);
1804 if (rc) {
1805 log_rdma_mr(ERR, "memory registration allocation failed\n");
1806 goto allocate_mr_failed;
1807 }
1808
1809 return info;
1810
1811 allocate_mr_failed:
1812 /* At this point, need to a full transport shutdown */
1813 smbd_destroy(info);
1814 return NULL;
1815
1816 negotiation_failed:
1817 cancel_delayed_work_sync(&info->idle_timer_work);
1818 destroy_caches_and_workqueue(info);
1819 info->transport_status = SMBD_NEGOTIATE_FAILED;
1820 init_waitqueue_head(&info->conn_wait);
1821 rdma_disconnect(info->id);
1822 wait_event(info->conn_wait,
1823 info->transport_status == SMBD_DISCONNECTED);
1824
1825 allocate_cache_failed:
1826 rdma_connect_failed:
1827 rdma_destroy_qp(info->id);
1828
1829 create_qp_failed:
1830 alloc_cq_failed:
1831 if (info->send_cq)
1832 ib_free_cq(info->send_cq);
1833 if (info->recv_cq)
1834 ib_free_cq(info->recv_cq);
1835
1836 config_failed:
1837 ib_dealloc_pd(info->pd);
1838 rdma_destroy_id(info->id);
1839
1840 create_id_failed:
1841 kfree(info);
1842 return NULL;
1843 }
1844
1845 struct smbd_connection *smbd_get_connection(
1846 struct TCP_Server_Info *server, struct sockaddr *dstaddr)
1847 {
1848 struct smbd_connection *ret;
1849 int port = SMBD_PORT;
1850
1851 try_again:
1852 ret = _smbd_get_connection(server, dstaddr, port);
1853
1854 /* Try SMB_PORT if SMBD_PORT doesn't work */
1855 if (!ret && port == SMBD_PORT) {
1856 port = SMB_PORT;
1857 goto try_again;
1858 }
1859 return ret;
1860 }
1861
1862 /*
1863 * Receive data from receive reassembly queue
1864 * All the incoming data packets are placed in reassembly queue
1865 * buf: the buffer to read data into
1866 * size: the length of data to read
1867 * return value: actual data read
1868 * Note: this implementation copies the data from reassebmly queue to receive
1869 * buffers used by upper layer. This is not the optimal code path. A better way
1870 * to do it is to not have upper layer allocate its receive buffers but rather
1871 * borrow the buffer from reassembly queue, and return it after data is
1872 * consumed. But this will require more changes to upper layer code, and also
1873 * need to consider packet boundaries while they still being reassembled.
1874 */
1875 static int smbd_recv_buf(struct smbd_connection *info, char *buf,
1876 unsigned int size)
1877 {
1878 struct smbd_response *response;
1879 struct smbd_data_transfer *data_transfer;
1880 int to_copy, to_read, data_read, offset;
1881 u32 data_length, remaining_data_length, data_offset;
1882 int rc;
1883
1884 again:
1885 if (info->transport_status != SMBD_CONNECTED) {
1886 log_read(ERR, "disconnected\n");
1887 return -ENODEV;
1888 }
1889
1890 /*
1891 * No need to hold the reassembly queue lock all the time as we are
1892 * the only one reading from the front of the queue. The transport
1893 * may add more entries to the back of the queue at the same time
1894 */
1895 log_read(INFO, "size=%d info->reassembly_data_length=%d\n", size,
1896 info->reassembly_data_length);
1897 if (info->reassembly_data_length >= size) {
1898 int queue_length;
1899 int queue_removed = 0;
1900
1901 /*
1902 * Need to make sure reassembly_data_length is read before
1903 * reading reassembly_queue_length and calling
1904 * _get_first_reassembly. This call is lock free
1905 * as we never read at the end of the queue which are being
1906 * updated in SOFTIRQ as more data is received
1907 */
1908 virt_rmb();
1909 queue_length = info->reassembly_queue_length;
1910 data_read = 0;
1911 to_read = size;
1912 offset = info->first_entry_offset;
1913 while (data_read < size) {
1914 response = _get_first_reassembly(info);
1915 data_transfer = smbd_response_payload(response);
1916 data_length = le32_to_cpu(data_transfer->data_length);
1917 remaining_data_length =
1918 le32_to_cpu(
1919 data_transfer->remaining_data_length);
1920 data_offset = le32_to_cpu(data_transfer->data_offset);
1921
1922 /*
1923 * The upper layer expects RFC1002 length at the
1924 * beginning of the payload. Return it to indicate
1925 * the total length of the packet. This minimize the
1926 * change to upper layer packet processing logic. This
1927 * will be eventually remove when an intermediate
1928 * transport layer is added
1929 */
1930 if (response->first_segment && size == 4) {
1931 unsigned int rfc1002_len =
1932 data_length + remaining_data_length;
1933 *((__be32 *)buf) = cpu_to_be32(rfc1002_len);
1934 data_read = 4;
1935 response->first_segment = false;
1936 log_read(INFO, "returning rfc1002 length %d\n",
1937 rfc1002_len);
1938 goto read_rfc1002_done;
1939 }
1940
1941 to_copy = min_t(int, data_length - offset, to_read);
1942 memcpy(
1943 buf + data_read,
1944 (char *)data_transfer + data_offset + offset,
1945 to_copy);
1946
1947 /* move on to the next buffer? */
1948 if (to_copy == data_length - offset) {
1949 queue_length--;
1950 /*
1951 * No need to lock if we are not at the
1952 * end of the queue
1953 */
1954 if (queue_length)
1955 list_del(&response->list);
1956 else {
1957 spin_lock_irq(
1958 &info->reassembly_queue_lock);
1959 list_del(&response->list);
1960 spin_unlock_irq(
1961 &info->reassembly_queue_lock);
1962 }
1963 queue_removed++;
1964 info->count_reassembly_queue--;
1965 info->count_dequeue_reassembly_queue++;
1966 put_receive_buffer(info, response);
1967 offset = 0;
1968 log_read(INFO, "put_receive_buffer offset=0\n");
1969 } else
1970 offset += to_copy;
1971
1972 to_read -= to_copy;
1973 data_read += to_copy;
1974
1975 log_read(INFO, "_get_first_reassembly memcpy %d bytes "
1976 "data_transfer_length-offset=%d after that "
1977 "to_read=%d data_read=%d offset=%d\n",
1978 to_copy, data_length - offset,
1979 to_read, data_read, offset);
1980 }
1981
1982 spin_lock_irq(&info->reassembly_queue_lock);
1983 info->reassembly_data_length -= data_read;
1984 info->reassembly_queue_length -= queue_removed;
1985 spin_unlock_irq(&info->reassembly_queue_lock);
1986
1987 info->first_entry_offset = offset;
1988 log_read(INFO, "returning to thread data_read=%d "
1989 "reassembly_data_length=%d first_entry_offset=%d\n",
1990 data_read, info->reassembly_data_length,
1991 info->first_entry_offset);
1992 read_rfc1002_done:
1993 return data_read;
1994 }
1995
1996 log_read(INFO, "wait_event on more data\n");
1997 rc = wait_event_interruptible(
1998 info->wait_reassembly_queue,
1999 info->reassembly_data_length >= size ||
2000 info->transport_status != SMBD_CONNECTED);
2001 /* Don't return any data if interrupted */
2002 if (rc)
2003 return -ENODEV;
2004
2005 goto again;
2006 }
2007
2008 /*
2009 * Receive a page from receive reassembly queue
2010 * page: the page to read data into
2011 * to_read: the length of data to read
2012 * return value: actual data read
2013 */
2014 static int smbd_recv_page(struct smbd_connection *info,
2015 struct page *page, unsigned int page_offset,
2016 unsigned int to_read)
2017 {
2018 int ret;
2019 char *to_address;
2020 void *page_address;
2021
2022 /* make sure we have the page ready for read */
2023 ret = wait_event_interruptible(
2024 info->wait_reassembly_queue,
2025 info->reassembly_data_length >= to_read ||
2026 info->transport_status != SMBD_CONNECTED);
2027 if (ret)
2028 return ret;
2029
2030 /* now we can read from reassembly queue and not sleep */
2031 page_address = kmap_atomic(page);
2032 to_address = (char *) page_address + page_offset;
2033
2034 log_read(INFO, "reading from page=%p address=%p to_read=%d\n",
2035 page, to_address, to_read);
2036
2037 ret = smbd_recv_buf(info, to_address, to_read);
2038 kunmap_atomic(page_address);
2039
2040 return ret;
2041 }
2042
2043 /*
2044 * Receive data from transport
2045 * msg: a msghdr point to the buffer, can be ITER_KVEC or ITER_BVEC
2046 * return: total bytes read, or 0. SMB Direct will not do partial read.
2047 */
2048 int smbd_recv(struct smbd_connection *info, struct msghdr *msg)
2049 {
2050 char *buf;
2051 struct page *page;
2052 unsigned int to_read, page_offset;
2053 int rc;
2054
2055 info->smbd_recv_pending++;
2056
2057 if (iov_iter_rw(&msg->msg_iter) == WRITE) {
2058 /* It's a bug in upper layer to get there */
2059 cifs_dbg(VFS, "CIFS: invalid msg iter dir %u\n",
2060 iov_iter_rw(&msg->msg_iter));
2061 rc = -EINVAL;
2062 goto out;
2063 }
2064
2065 switch (iov_iter_type(&msg->msg_iter)) {
2066 case ITER_KVEC:
2067 buf = msg->msg_iter.kvec->iov_base;
2068 to_read = msg->msg_iter.kvec->iov_len;
2069 rc = smbd_recv_buf(info, buf, to_read);
2070 break;
2071
2072 case ITER_BVEC:
2073 page = msg->msg_iter.bvec->bv_page;
2074 page_offset = msg->msg_iter.bvec->bv_offset;
2075 to_read = msg->msg_iter.bvec->bv_len;
2076 rc = smbd_recv_page(info, page, page_offset, to_read);
2077 break;
2078
2079 default:
2080 /* It's a bug in upper layer to get there */
2081 cifs_dbg(VFS, "CIFS: invalid msg type %d\n",
2082 iov_iter_type(&msg->msg_iter));
2083 rc = -EINVAL;
2084 }
2085
2086 out:
2087 info->smbd_recv_pending--;
2088 wake_up(&info->wait_smbd_recv_pending);
2089
2090 /* SMBDirect will read it all or nothing */
2091 if (rc > 0)
2092 msg->msg_iter.count = 0;
2093 return rc;
2094 }
2095
2096 /*
2097 * Send data to transport
2098 * Each rqst is transported as a SMBDirect payload
2099 * rqst: the data to write
2100 * return value: 0 if successfully write, otherwise error code
2101 */
2102 int smbd_send(struct TCP_Server_Info *server, struct smb_rqst *rqst)
2103 {
2104 struct smbd_connection *info = server->smbd_conn;
2105 struct kvec vec;
2106 int nvecs;
2107 int size;
2108 unsigned int buflen, remaining_data_length;
2109 int start, i, j;
2110 int max_iov_size =
2111 info->max_send_size - sizeof(struct smbd_data_transfer);
2112 struct kvec *iov;
2113 int rc;
2114
2115 info->smbd_send_pending++;
2116 if (info->transport_status != SMBD_CONNECTED) {
2117 rc = -ENODEV;
2118 goto done;
2119 }
2120
2121 /*
2122 * Skip the RFC1002 length defined in MS-SMB2 section 2.1
2123 * It is used only for TCP transport in the iov[0]
2124 * In future we may want to add a transport layer under protocol
2125 * layer so this will only be issued to TCP transport
2126 */
2127
2128 if (rqst->rq_iov[0].iov_len != 4) {
2129 log_write(ERR, "expected the pdu length in 1st iov, but got %zu\n", rqst->rq_iov[0].iov_len);
2130 return -EINVAL;
2131 }
2132
2133 /*
2134 * Add in the page array if there is one. The caller needs to set
2135 * rq_tailsz to PAGE_SIZE when the buffer has multiple pages and
2136 * ends at page boundary
2137 */
2138 buflen = smb_rqst_len(server, rqst);
2139
2140 if (buflen + sizeof(struct smbd_data_transfer) >
2141 info->max_fragmented_send_size) {
2142 log_write(ERR, "payload size %d > max size %d\n",
2143 buflen, info->max_fragmented_send_size);
2144 rc = -EINVAL;
2145 goto done;
2146 }
2147
2148 iov = &rqst->rq_iov[1];
2149
2150 cifs_dbg(FYI, "Sending smb (RDMA): smb_len=%u\n", buflen);
2151 for (i = 0; i < rqst->rq_nvec-1; i++)
2152 dump_smb(iov[i].iov_base, iov[i].iov_len);
2153
2154 remaining_data_length = buflen;
2155
2156 log_write(INFO, "rqst->rq_nvec=%d rqst->rq_npages=%d rq_pagesz=%d "
2157 "rq_tailsz=%d buflen=%d\n",
2158 rqst->rq_nvec, rqst->rq_npages, rqst->rq_pagesz,
2159 rqst->rq_tailsz, buflen);
2160
2161 start = i = iov[0].iov_len ? 0 : 1;
2162 buflen = 0;
2163 while (true) {
2164 buflen += iov[i].iov_len;
2165 if (buflen > max_iov_size) {
2166 if (i > start) {
2167 remaining_data_length -=
2168 (buflen-iov[i].iov_len);
2169 log_write(INFO, "sending iov[] from start=%d "
2170 "i=%d nvecs=%d "
2171 "remaining_data_length=%d\n",
2172 start, i, i-start,
2173 remaining_data_length);
2174 rc = smbd_post_send_data(
2175 info, &iov[start], i-start,
2176 remaining_data_length);
2177 if (rc)
2178 goto done;
2179 } else {
2180 /* iov[start] is too big, break it */
2181 nvecs = (buflen+max_iov_size-1)/max_iov_size;
2182 log_write(INFO, "iov[%d] iov_base=%p buflen=%d"
2183 " break to %d vectors\n",
2184 start, iov[start].iov_base,
2185 buflen, nvecs);
2186 for (j = 0; j < nvecs; j++) {
2187 vec.iov_base =
2188 (char *)iov[start].iov_base +
2189 j*max_iov_size;
2190 vec.iov_len = max_iov_size;
2191 if (j == nvecs-1)
2192 vec.iov_len =
2193 buflen -
2194 max_iov_size*(nvecs-1);
2195 remaining_data_length -= vec.iov_len;
2196 log_write(INFO,
2197 "sending vec j=%d iov_base=%p"
2198 " iov_len=%zu "
2199 "remaining_data_length=%d\n",
2200 j, vec.iov_base, vec.iov_len,
2201 remaining_data_length);
2202 rc = smbd_post_send_data(
2203 info, &vec, 1,
2204 remaining_data_length);
2205 if (rc)
2206 goto done;
2207 }
2208 i++;
2209 if (i == rqst->rq_nvec-1)
2210 break;
2211 }
2212 start = i;
2213 buflen = 0;
2214 } else {
2215 i++;
2216 if (i == rqst->rq_nvec-1) {
2217 /* send out all remaining vecs */
2218 remaining_data_length -= buflen;
2219 log_write(INFO,
2220 "sending iov[] from start=%d i=%d "
2221 "nvecs=%d remaining_data_length=%d\n",
2222 start, i, i-start,
2223 remaining_data_length);
2224 rc = smbd_post_send_data(info, &iov[start],
2225 i-start, remaining_data_length);
2226 if (rc)
2227 goto done;
2228 break;
2229 }
2230 }
2231 log_write(INFO, "looping i=%d buflen=%d\n", i, buflen);
2232 }
2233
2234 /* now sending pages if there are any */
2235 for (i = 0; i < rqst->rq_npages; i++) {
2236 unsigned int offset;
2237
2238 rqst_page_get_length(rqst, i, &buflen, &offset);
2239 nvecs = (buflen + max_iov_size - 1) / max_iov_size;
2240 log_write(INFO, "sending pages buflen=%d nvecs=%d\n",
2241 buflen, nvecs);
2242 for (j = 0; j < nvecs; j++) {
2243 size = max_iov_size;
2244 if (j == nvecs-1)
2245 size = buflen - j*max_iov_size;
2246 remaining_data_length -= size;
2247 log_write(INFO, "sending pages i=%d offset=%d size=%d"
2248 " remaining_data_length=%d\n",
2249 i, j*max_iov_size+offset, size,
2250 remaining_data_length);
2251 rc = smbd_post_send_page(
2252 info, rqst->rq_pages[i],
2253 j*max_iov_size + offset,
2254 size, remaining_data_length);
2255 if (rc)
2256 goto done;
2257 }
2258 }
2259
2260 done:
2261 /*
2262 * As an optimization, we don't wait for individual I/O to finish
2263 * before sending the next one.
2264 * Send them all and wait for pending send count to get to 0
2265 * that means all the I/Os have been out and we are good to return
2266 */
2267
2268 wait_event(info->wait_send_payload_pending,
2269 atomic_read(&info->send_payload_pending) == 0);
2270
2271 info->smbd_send_pending--;
2272 wake_up(&info->wait_smbd_send_pending);
2273
2274 return rc;
2275 }
2276
2277 static void register_mr_done(struct ib_cq *cq, struct ib_wc *wc)
2278 {
2279 struct smbd_mr *mr;
2280 struct ib_cqe *cqe;
2281
2282 if (wc->status) {
2283 log_rdma_mr(ERR, "status=%d\n", wc->status);
2284 cqe = wc->wr_cqe;
2285 mr = container_of(cqe, struct smbd_mr, cqe);
2286 smbd_disconnect_rdma_connection(mr->conn);
2287 }
2288 }
2289
2290 /*
2291 * The work queue function that recovers MRs
2292 * We need to call ib_dereg_mr() and ib_alloc_mr() before this MR can be used
2293 * again. Both calls are slow, so finish them in a workqueue. This will not
2294 * block I/O path.
2295 * There is one workqueue that recovers MRs, there is no need to lock as the
2296 * I/O requests calling smbd_register_mr will never update the links in the
2297 * mr_list.
2298 */
2299 static void smbd_mr_recovery_work(struct work_struct *work)
2300 {
2301 struct smbd_connection *info =
2302 container_of(work, struct smbd_connection, mr_recovery_work);
2303 struct smbd_mr *smbdirect_mr;
2304 int rc;
2305
2306 list_for_each_entry(smbdirect_mr, &info->mr_list, list) {
2307 if (smbdirect_mr->state == MR_INVALIDATED)
2308 ib_dma_unmap_sg(
2309 info->id->device, smbdirect_mr->sgl,
2310 smbdirect_mr->sgl_count,
2311 smbdirect_mr->dir);
2312 else if (smbdirect_mr->state == MR_ERROR) {
2313
2314 /* recover this MR entry */
2315 rc = ib_dereg_mr(smbdirect_mr->mr);
2316 if (rc) {
2317 log_rdma_mr(ERR,
2318 "ib_dereg_mr failed rc=%x\n",
2319 rc);
2320 smbd_disconnect_rdma_connection(info);
2321 continue;
2322 }
2323
2324 smbdirect_mr->mr = ib_alloc_mr(
2325 info->pd, info->mr_type,
2326 info->max_frmr_depth);
2327 if (IS_ERR(smbdirect_mr->mr)) {
2328 log_rdma_mr(ERR,
2329 "ib_alloc_mr failed mr_type=%x "
2330 "max_frmr_depth=%x\n",
2331 info->mr_type,
2332 info->max_frmr_depth);
2333 smbd_disconnect_rdma_connection(info);
2334 continue;
2335 }
2336 } else
2337 /* This MR is being used, don't recover it */
2338 continue;
2339
2340 smbdirect_mr->state = MR_READY;
2341
2342 /* smbdirect_mr->state is updated by this function
2343 * and is read and updated by I/O issuing CPUs trying
2344 * to get a MR, the call to atomic_inc_return
2345 * implicates a memory barrier and guarantees this
2346 * value is updated before waking up any calls to
2347 * get_mr() from the I/O issuing CPUs
2348 */
2349 if (atomic_inc_return(&info->mr_ready_count) == 1)
2350 wake_up_interruptible(&info->wait_mr);
2351 }
2352 }
2353
2354 static void destroy_mr_list(struct smbd_connection *info)
2355 {
2356 struct smbd_mr *mr, *tmp;
2357
2358 cancel_work_sync(&info->mr_recovery_work);
2359 list_for_each_entry_safe(mr, tmp, &info->mr_list, list) {
2360 if (mr->state == MR_INVALIDATED)
2361 ib_dma_unmap_sg(info->id->device, mr->sgl,
2362 mr->sgl_count, mr->dir);
2363 ib_dereg_mr(mr->mr);
2364 kfree(mr->sgl);
2365 kfree(mr);
2366 }
2367 }
2368
2369 /*
2370 * Allocate MRs used for RDMA read/write
2371 * The number of MRs will not exceed hardware capability in responder_resources
2372 * All MRs are kept in mr_list. The MR can be recovered after it's used
2373 * Recovery is done in smbd_mr_recovery_work. The content of list entry changes
2374 * as MRs are used and recovered for I/O, but the list links will not change
2375 */
2376 static int allocate_mr_list(struct smbd_connection *info)
2377 {
2378 int i;
2379 struct smbd_mr *smbdirect_mr, *tmp;
2380
2381 INIT_LIST_HEAD(&info->mr_list);
2382 init_waitqueue_head(&info->wait_mr);
2383 spin_lock_init(&info->mr_list_lock);
2384 atomic_set(&info->mr_ready_count, 0);
2385 atomic_set(&info->mr_used_count, 0);
2386 init_waitqueue_head(&info->wait_for_mr_cleanup);
2387 /* Allocate more MRs (2x) than hardware responder_resources */
2388 for (i = 0; i < info->responder_resources * 2; i++) {
2389 smbdirect_mr = kzalloc(sizeof(*smbdirect_mr), GFP_KERNEL);
2390 if (!smbdirect_mr)
2391 goto out;
2392 smbdirect_mr->mr = ib_alloc_mr(info->pd, info->mr_type,
2393 info->max_frmr_depth);
2394 if (IS_ERR(smbdirect_mr->mr)) {
2395 log_rdma_mr(ERR, "ib_alloc_mr failed mr_type=%x "
2396 "max_frmr_depth=%x\n",
2397 info->mr_type, info->max_frmr_depth);
2398 goto out;
2399 }
2400 smbdirect_mr->sgl = kcalloc(
2401 info->max_frmr_depth,
2402 sizeof(struct scatterlist),
2403 GFP_KERNEL);
2404 if (!smbdirect_mr->sgl) {
2405 log_rdma_mr(ERR, "failed to allocate sgl\n");
2406 ib_dereg_mr(smbdirect_mr->mr);
2407 goto out;
2408 }
2409 smbdirect_mr->state = MR_READY;
2410 smbdirect_mr->conn = info;
2411
2412 list_add_tail(&smbdirect_mr->list, &info->mr_list);
2413 atomic_inc(&info->mr_ready_count);
2414 }
2415 INIT_WORK(&info->mr_recovery_work, smbd_mr_recovery_work);
2416 return 0;
2417
2418 out:
2419 kfree(smbdirect_mr);
2420
2421 list_for_each_entry_safe(smbdirect_mr, tmp, &info->mr_list, list) {
2422 ib_dereg_mr(smbdirect_mr->mr);
2423 kfree(smbdirect_mr->sgl);
2424 kfree(smbdirect_mr);
2425 }
2426 return -ENOMEM;
2427 }
2428
2429 /*
2430 * Get a MR from mr_list. This function waits until there is at least one
2431 * MR available in the list. It may access the list while the
2432 * smbd_mr_recovery_work is recovering the MR list. This doesn't need a lock
2433 * as they never modify the same places. However, there may be several CPUs
2434 * issueing I/O trying to get MR at the same time, mr_list_lock is used to
2435 * protect this situation.
2436 */
2437 static struct smbd_mr *get_mr(struct smbd_connection *info)
2438 {
2439 struct smbd_mr *ret;
2440 int rc;
2441 again:
2442 rc = wait_event_interruptible(info->wait_mr,
2443 atomic_read(&info->mr_ready_count) ||
2444 info->transport_status != SMBD_CONNECTED);
2445 if (rc) {
2446 log_rdma_mr(ERR, "wait_event_interruptible rc=%x\n", rc);
2447 return NULL;
2448 }
2449
2450 if (info->transport_status != SMBD_CONNECTED) {
2451 log_rdma_mr(ERR, "info->transport_status=%x\n",
2452 info->transport_status);
2453 return NULL;
2454 }
2455
2456 spin_lock(&info->mr_list_lock);
2457 list_for_each_entry(ret, &info->mr_list, list) {
2458 if (ret->state == MR_READY) {
2459 ret->state = MR_REGISTERED;
2460 spin_unlock(&info->mr_list_lock);
2461 atomic_dec(&info->mr_ready_count);
2462 atomic_inc(&info->mr_used_count);
2463 return ret;
2464 }
2465 }
2466
2467 spin_unlock(&info->mr_list_lock);
2468 /*
2469 * It is possible that we could fail to get MR because other processes may
2470 * try to acquire a MR at the same time. If this is the case, retry it.
2471 */
2472 goto again;
2473 }
2474
2475 /*
2476 * Register memory for RDMA read/write
2477 * pages[]: the list of pages to register memory with
2478 * num_pages: the number of pages to register
2479 * tailsz: if non-zero, the bytes to register in the last page
2480 * writing: true if this is a RDMA write (SMB read), false for RDMA read
2481 * need_invalidate: true if this MR needs to be locally invalidated after I/O
2482 * return value: the MR registered, NULL if failed.
2483 */
2484 struct smbd_mr *smbd_register_mr(
2485 struct smbd_connection *info, struct page *pages[], int num_pages,
2486 int offset, int tailsz, bool writing, bool need_invalidate)
2487 {
2488 struct smbd_mr *smbdirect_mr;
2489 int rc, i;
2490 enum dma_data_direction dir;
2491 struct ib_reg_wr *reg_wr;
2492
2493 if (num_pages > info->max_frmr_depth) {
2494 log_rdma_mr(ERR, "num_pages=%d max_frmr_depth=%d\n",
2495 num_pages, info->max_frmr_depth);
2496 return NULL;
2497 }
2498
2499 smbdirect_mr = get_mr(info);
2500 if (!smbdirect_mr) {
2501 log_rdma_mr(ERR, "get_mr returning NULL\n");
2502 return NULL;
2503 }
2504 smbdirect_mr->need_invalidate = need_invalidate;
2505 smbdirect_mr->sgl_count = num_pages;
2506 sg_init_table(smbdirect_mr->sgl, num_pages);
2507
2508 log_rdma_mr(INFO, "num_pages=0x%x offset=0x%x tailsz=0x%x\n",
2509 num_pages, offset, tailsz);
2510
2511 if (num_pages == 1) {
2512 sg_set_page(&smbdirect_mr->sgl[0], pages[0], tailsz, offset);
2513 goto skip_multiple_pages;
2514 }
2515
2516 /* We have at least two pages to register */
2517 sg_set_page(
2518 &smbdirect_mr->sgl[0], pages[0], PAGE_SIZE - offset, offset);
2519 i = 1;
2520 while (i < num_pages - 1) {
2521 sg_set_page(&smbdirect_mr->sgl[i], pages[i], PAGE_SIZE, 0);
2522 i++;
2523 }
2524 sg_set_page(&smbdirect_mr->sgl[i], pages[i],
2525 tailsz ? tailsz : PAGE_SIZE, 0);
2526
2527 skip_multiple_pages:
2528 dir = writing ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
2529 smbdirect_mr->dir = dir;
2530 rc = ib_dma_map_sg(info->id->device, smbdirect_mr->sgl, num_pages, dir);
2531 if (!rc) {
2532 log_rdma_mr(ERR, "ib_dma_map_sg num_pages=%x dir=%x rc=%x\n",
2533 num_pages, dir, rc);
2534 goto dma_map_error;
2535 }
2536
2537 rc = ib_map_mr_sg(smbdirect_mr->mr, smbdirect_mr->sgl, num_pages,
2538 NULL, PAGE_SIZE);
2539 if (rc != num_pages) {
2540 log_rdma_mr(ERR,
2541 "ib_map_mr_sg failed rc = %d num_pages = %x\n",
2542 rc, num_pages);
2543 goto map_mr_error;
2544 }
2545
2546 ib_update_fast_reg_key(smbdirect_mr->mr,
2547 ib_inc_rkey(smbdirect_mr->mr->rkey));
2548 reg_wr = &smbdirect_mr->wr;
2549 reg_wr->wr.opcode = IB_WR_REG_MR;
2550 smbdirect_mr->cqe.done = register_mr_done;
2551 reg_wr->wr.wr_cqe = &smbdirect_mr->cqe;
2552 reg_wr->wr.num_sge = 0;
2553 reg_wr->wr.send_flags = IB_SEND_SIGNALED;
2554 reg_wr->mr = smbdirect_mr->mr;
2555 reg_wr->key = smbdirect_mr->mr->rkey;
2556 reg_wr->access = writing ?
2557 IB_ACCESS_REMOTE_WRITE | IB_ACCESS_LOCAL_WRITE :
2558 IB_ACCESS_REMOTE_READ;
2559
2560 /*
2561 * There is no need for waiting for complemtion on ib_post_send
2562 * on IB_WR_REG_MR. Hardware enforces a barrier and order of execution
2563 * on the next ib_post_send when we actaully send I/O to remote peer
2564 */
2565 rc = ib_post_send(info->id->qp, ®_wr->wr, NULL);
2566 if (!rc)
2567 return smbdirect_mr;
2568
2569 log_rdma_mr(ERR, "ib_post_send failed rc=%x reg_wr->key=%x\n",
2570 rc, reg_wr->key);
2571
2572 /* If all failed, attempt to recover this MR by setting it MR_ERROR*/
2573 map_mr_error:
2574 ib_dma_unmap_sg(info->id->device, smbdirect_mr->sgl,
2575 smbdirect_mr->sgl_count, smbdirect_mr->dir);
2576
2577 dma_map_error:
2578 smbdirect_mr->state = MR_ERROR;
2579 if (atomic_dec_and_test(&info->mr_used_count))
2580 wake_up(&info->wait_for_mr_cleanup);
2581
2582 smbd_disconnect_rdma_connection(info);
2583
2584 return NULL;
2585 }
2586
2587 static void local_inv_done(struct ib_cq *cq, struct ib_wc *wc)
2588 {
2589 struct smbd_mr *smbdirect_mr;
2590 struct ib_cqe *cqe;
2591
2592 cqe = wc->wr_cqe;
2593 smbdirect_mr = container_of(cqe, struct smbd_mr, cqe);
2594 smbdirect_mr->state = MR_INVALIDATED;
2595 if (wc->status != IB_WC_SUCCESS) {
2596 log_rdma_mr(ERR, "invalidate failed status=%x\n", wc->status);
2597 smbdirect_mr->state = MR_ERROR;
2598 }
2599 complete(&smbdirect_mr->invalidate_done);
2600 }
2601
2602 /*
2603 * Deregister a MR after I/O is done
2604 * This function may wait if remote invalidation is not used
2605 * and we have to locally invalidate the buffer to prevent data is being
2606 * modified by remote peer after upper layer consumes it
2607 */
2608 int smbd_deregister_mr(struct smbd_mr *smbdirect_mr)
2609 {
2610 struct ib_send_wr *wr;
2611 struct smbd_connection *info = smbdirect_mr->conn;
2612 int rc = 0;
2613
2614 if (smbdirect_mr->need_invalidate) {
2615 /* Need to finish local invalidation before returning */
2616 wr = &smbdirect_mr->inv_wr;
2617 wr->opcode = IB_WR_LOCAL_INV;
2618 smbdirect_mr->cqe.done = local_inv_done;
2619 wr->wr_cqe = &smbdirect_mr->cqe;
2620 wr->num_sge = 0;
2621 wr->ex.invalidate_rkey = smbdirect_mr->mr->rkey;
2622 wr->send_flags = IB_SEND_SIGNALED;
2623
2624 init_completion(&smbdirect_mr->invalidate_done);
2625 rc = ib_post_send(info->id->qp, wr, NULL);
2626 if (rc) {
2627 log_rdma_mr(ERR, "ib_post_send failed rc=%x\n", rc);
2628 smbd_disconnect_rdma_connection(info);
2629 goto done;
2630 }
2631 wait_for_completion(&smbdirect_mr->invalidate_done);
2632 smbdirect_mr->need_invalidate = false;
2633 } else
2634 /*
2635 * For remote invalidation, just set it to MR_INVALIDATED
2636 * and defer to mr_recovery_work to recover the MR for next use
2637 */
2638 smbdirect_mr->state = MR_INVALIDATED;
2639
2640 /*
2641 * Schedule the work to do MR recovery for future I/Os
2642 * MR recovery is slow and we don't want it to block the current I/O
2643 */
2644 queue_work(info->workqueue, &info->mr_recovery_work);
2645
2646 done:
2647 if (atomic_dec_and_test(&info->mr_used_count))
2648 wake_up(&info->wait_for_mr_cleanup);
2649
2650 return rc;
2651 }
2652
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.