?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 5.8-rc4
[linux/fpc-iii.git] / fs / cifs / smbdirect.c
blobb029ed31ef916d979bde6d3523973c1b14e3276d
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
250 default:
251 break;
252 }
253 }
254
255 static inline void *smbd_request_payload(struct smbd_request *request)
256 {
257 return (void *)request->packet;
258 }
259
260 static inline void *smbd_response_payload(struct smbd_response *response)
261 {
262 return (void *)response->packet;
263 }
264
265 /* Called when a RDMA send is done */
266 static void send_done(struct ib_cq *cq, struct ib_wc *wc)
267 {
268 int i;
269 struct smbd_request *request =
270 container_of(wc->wr_cqe, struct smbd_request, cqe);
271
272 log_rdma_send(INFO, "smbd_request %p completed wc->status=%d\n",
273 request, wc->status);
274
275 if (wc->status != IB_WC_SUCCESS || wc->opcode != IB_WC_SEND) {
276 log_rdma_send(ERR, "wc->status=%d wc->opcode=%d\n",
277 wc->status, wc->opcode);
278 smbd_disconnect_rdma_connection(request->info);
279 }
280
281 for (i = 0; i < request->num_sge; i++)
282 ib_dma_unmap_single(request->info->id->device,
283 request->sge[i].addr,
284 request->sge[i].length,
285 DMA_TO_DEVICE);
286
287 if (atomic_dec_and_test(&request->info->send_pending))
288 wake_up(&request->info->wait_send_pending);
289
290 wake_up(&request->info->wait_post_send);
291
292 mempool_free(request, request->info->request_mempool);
293 }
294
295 static void dump_smbd_negotiate_resp(struct smbd_negotiate_resp *resp)
296 {
297 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",
298 resp->min_version, resp->max_version,
299 resp->negotiated_version, resp->credits_requested,
300 resp->credits_granted, resp->status,
301 resp->max_readwrite_size, resp->preferred_send_size,
302 resp->max_receive_size, resp->max_fragmented_size);
303 }
304
305 /*
306 * Process a negotiation response message, according to [MS-SMBD]3.1.5.7
307 * response, packet_length: the negotiation response message
308 * return value: true if negotiation is a success, false if failed
309 */
310 static bool process_negotiation_response(
311 struct smbd_response *response, int packet_length)
312 {
313 struct smbd_connection *info = response->info;
314 struct smbd_negotiate_resp *packet = smbd_response_payload(response);
315
316 if (packet_length < sizeof(struct smbd_negotiate_resp)) {
317 log_rdma_event(ERR,
318 "error: packet_length=%d\n", packet_length);
319 return false;
320 }
321
322 if (le16_to_cpu(packet->negotiated_version) != SMBD_V1) {
323 log_rdma_event(ERR, "error: negotiated_version=%x\n",
324 le16_to_cpu(packet->negotiated_version));
325 return false;
326 }
327 info->protocol = le16_to_cpu(packet->negotiated_version);
328
329 if (packet->credits_requested == 0) {
330 log_rdma_event(ERR, "error: credits_requested==0\n");
331 return false;
332 }
333 info->receive_credit_target = le16_to_cpu(packet->credits_requested);
334
335 if (packet->credits_granted == 0) {
336 log_rdma_event(ERR, "error: credits_granted==0\n");
337 return false;
338 }
339 atomic_set(&info->send_credits, le16_to_cpu(packet->credits_granted));
340
341 atomic_set(&info->receive_credits, 0);
342
343 if (le32_to_cpu(packet->preferred_send_size) > info->max_receive_size) {
344 log_rdma_event(ERR, "error: preferred_send_size=%d\n",
345 le32_to_cpu(packet->preferred_send_size));
346 return false;
347 }
348 info->max_receive_size = le32_to_cpu(packet->preferred_send_size);
349
350 if (le32_to_cpu(packet->max_receive_size) < SMBD_MIN_RECEIVE_SIZE) {
351 log_rdma_event(ERR, "error: max_receive_size=%d\n",
352 le32_to_cpu(packet->max_receive_size));
353 return false;
354 }
355 info->max_send_size = min_t(int, info->max_send_size,
356 le32_to_cpu(packet->max_receive_size));
357
358 if (le32_to_cpu(packet->max_fragmented_size) <
359 SMBD_MIN_FRAGMENTED_SIZE) {
360 log_rdma_event(ERR, "error: max_fragmented_size=%d\n",
361 le32_to_cpu(packet->max_fragmented_size));
362 return false;
363 }
364 info->max_fragmented_send_size =
365 le32_to_cpu(packet->max_fragmented_size);
366 info->rdma_readwrite_threshold =
367 rdma_readwrite_threshold > info->max_fragmented_send_size ?
368 info->max_fragmented_send_size :
369 rdma_readwrite_threshold;
370
371
372 info->max_readwrite_size = min_t(u32,
373 le32_to_cpu(packet->max_readwrite_size),
374 info->max_frmr_depth * PAGE_SIZE);
375 info->max_frmr_depth = info->max_readwrite_size / PAGE_SIZE;
376
377 return true;
378 }
379
380 static void smbd_post_send_credits(struct work_struct *work)
381 {
382 int ret = 0;
383 int use_receive_queue = 1;
384 int rc;
385 struct smbd_response *response;
386 struct smbd_connection *info =
387 container_of(work, struct smbd_connection,
388 post_send_credits_work);
389
390 if (info->transport_status != SMBD_CONNECTED) {
391 wake_up(&info->wait_receive_queues);
392 return;
393 }
394
395 if (info->receive_credit_target >
396 atomic_read(&info->receive_credits)) {
397 while (true) {
398 if (use_receive_queue)
399 response = get_receive_buffer(info);
400 else
401 response = get_empty_queue_buffer(info);
402 if (!response) {
403 /* now switch to emtpy packet queue */
404 if (use_receive_queue) {
405 use_receive_queue = 0;
406 continue;
407 } else
408 break;
409 }
410
411 response->type = SMBD_TRANSFER_DATA;
412 response->first_segment = false;
413 rc = smbd_post_recv(info, response);
414 if (rc) {
415 log_rdma_recv(ERR,
416 "post_recv failed rc=%d\n", rc);
417 put_receive_buffer(info, response);
418 break;
419 }
420
421 ret++;
422 }
423 }
424
425 spin_lock(&info->lock_new_credits_offered);
426 info->new_credits_offered += ret;
427 spin_unlock(&info->lock_new_credits_offered);
428
429 /* Promptly send an immediate packet as defined in [MS-SMBD] 3.1.1.1 */
430 info->send_immediate = true;
431 if (atomic_read(&info->receive_credits) <
432 info->receive_credit_target - 1) {
433 if (info->keep_alive_requested == KEEP_ALIVE_PENDING ||
434 info->send_immediate) {
435 log_keep_alive(INFO, "send an empty message\n");
436 smbd_post_send_empty(info);
437 }
438 }
439 }
440
441 /* Called from softirq, when recv is done */
442 static void recv_done(struct ib_cq *cq, struct ib_wc *wc)
443 {
444 struct smbd_data_transfer *data_transfer;
445 struct smbd_response *response =
446 container_of(wc->wr_cqe, struct smbd_response, cqe);
447 struct smbd_connection *info = response->info;
448 int data_length = 0;
449
450 log_rdma_recv(INFO, "response=%p type=%d wc status=%d wc opcode %d byte_len=%d pkey_index=%x\n",
451 response, response->type, wc->status, wc->opcode,
452 wc->byte_len, wc->pkey_index);
453
454 if (wc->status != IB_WC_SUCCESS || wc->opcode != IB_WC_RECV) {
455 log_rdma_recv(INFO, "wc->status=%d opcode=%d\n",
456 wc->status, wc->opcode);
457 smbd_disconnect_rdma_connection(info);
458 goto error;
459 }
460
461 ib_dma_sync_single_for_cpu(
462 wc->qp->device,
463 response->sge.addr,
464 response->sge.length,
465 DMA_FROM_DEVICE);
466
467 switch (response->type) {
468 /* SMBD negotiation response */
469 case SMBD_NEGOTIATE_RESP:
470 dump_smbd_negotiate_resp(smbd_response_payload(response));
471 info->full_packet_received = true;
472 info->negotiate_done =
473 process_negotiation_response(response, wc->byte_len);
474 complete(&info->negotiate_completion);
475 break;
476
477 /* SMBD data transfer packet */
478 case SMBD_TRANSFER_DATA:
479 data_transfer = smbd_response_payload(response);
480 data_length = le32_to_cpu(data_transfer->data_length);
481
482 /*
483 * If this is a packet with data playload place the data in
484 * reassembly queue and wake up the reading thread
485 */
486 if (data_length) {
487 if (info->full_packet_received)
488 response->first_segment = true;
489
490 if (le32_to_cpu(data_transfer->remaining_data_length))
491 info->full_packet_received = false;
492 else
493 info->full_packet_received = true;
494
495 enqueue_reassembly(
496 info,
497 response,
498 data_length);
499 } else
500 put_empty_packet(info, response);
501
502 if (data_length)
503 wake_up_interruptible(&info->wait_reassembly_queue);
504
505 atomic_dec(&info->receive_credits);
506 info->receive_credit_target =
507 le16_to_cpu(data_transfer->credits_requested);
508 if (le16_to_cpu(data_transfer->credits_granted)) {
509 atomic_add(le16_to_cpu(data_transfer->credits_granted),
510 &info->send_credits);
511 /*
512 * We have new send credits granted from remote peer
513 * If any sender is waiting for credits, unblock it
514 */
515 wake_up_interruptible(&info->wait_send_queue);
516 }
517
518 log_incoming(INFO, "data flags %d data_offset %d data_length %d remaining_data_length %d\n",
519 le16_to_cpu(data_transfer->flags),
520 le32_to_cpu(data_transfer->data_offset),
521 le32_to_cpu(data_transfer->data_length),
522 le32_to_cpu(data_transfer->remaining_data_length));
523
524 /* Send a KEEP_ALIVE response right away if requested */
525 info->keep_alive_requested = KEEP_ALIVE_NONE;
526 if (le16_to_cpu(data_transfer->flags) &
527 SMB_DIRECT_RESPONSE_REQUESTED) {
528 info->keep_alive_requested = KEEP_ALIVE_PENDING;
529 }
530
531 return;
532
533 default:
534 log_rdma_recv(ERR,
535 "unexpected response type=%d\n", response->type);
536 }
537
538 error:
539 put_receive_buffer(info, response);
540 }
541
542 static struct rdma_cm_id *smbd_create_id(
543 struct smbd_connection *info,
544 struct sockaddr *dstaddr, int port)
545 {
546 struct rdma_cm_id *id;
547 int rc;
548 __be16 *sport;
549
550 id = rdma_create_id(&init_net, smbd_conn_upcall, info,
551 RDMA_PS_TCP, IB_QPT_RC);
552 if (IS_ERR(id)) {
553 rc = PTR_ERR(id);
554 log_rdma_event(ERR, "rdma_create_id() failed %i\n", rc);
555 return id;
556 }
557
558 if (dstaddr->sa_family == AF_INET6)
559 sport = &((struct sockaddr_in6 *)dstaddr)->sin6_port;
560 else
561 sport = &((struct sockaddr_in *)dstaddr)->sin_port;
562
563 *sport = htons(port);
564
565 init_completion(&info->ri_done);
566 info->ri_rc = -ETIMEDOUT;
567
568 rc = rdma_resolve_addr(id, NULL, (struct sockaddr *)dstaddr,
569 RDMA_RESOLVE_TIMEOUT);
570 if (rc) {
571 log_rdma_event(ERR, "rdma_resolve_addr() failed %i\n", rc);
572 goto out;
573 }
574 wait_for_completion_interruptible_timeout(
575 &info->ri_done, msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT));
576 rc = info->ri_rc;
577 if (rc) {
578 log_rdma_event(ERR, "rdma_resolve_addr() completed %i\n", rc);
579 goto out;
580 }
581
582 info->ri_rc = -ETIMEDOUT;
583 rc = rdma_resolve_route(id, RDMA_RESOLVE_TIMEOUT);
584 if (rc) {
585 log_rdma_event(ERR, "rdma_resolve_route() failed %i\n", rc);
586 goto out;
587 }
588 wait_for_completion_interruptible_timeout(
589 &info->ri_done, msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT));
590 rc = info->ri_rc;
591 if (rc) {
592 log_rdma_event(ERR, "rdma_resolve_route() completed %i\n", rc);
593 goto out;
594 }
595
596 return id;
597
598 out:
599 rdma_destroy_id(id);
600 return ERR_PTR(rc);
601 }
602
603 /*
604 * Test if FRWR (Fast Registration Work Requests) is supported on the device
605 * This implementation requries FRWR on RDMA read/write
606 * return value: true if it is supported
607 */
608 static bool frwr_is_supported(struct ib_device_attr *attrs)
609 {
610 if (!(attrs->device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS))
611 return false;
612 if (attrs->max_fast_reg_page_list_len == 0)
613 return false;
614 return true;
615 }
616
617 static int smbd_ia_open(
618 struct smbd_connection *info,
619 struct sockaddr *dstaddr, int port)
620 {
621 int rc;
622
623 info->id = smbd_create_id(info, dstaddr, port);
624 if (IS_ERR(info->id)) {
625 rc = PTR_ERR(info->id);
626 goto out1;
627 }
628
629 if (!frwr_is_supported(&info->id->device->attrs)) {
630 log_rdma_event(ERR, "Fast Registration Work Requests (FRWR) is not supported\n");
631 log_rdma_event(ERR, "Device capability flags = %llx max_fast_reg_page_list_len = %u\n",
632 info->id->device->attrs.device_cap_flags,
633 info->id->device->attrs.max_fast_reg_page_list_len);
634 rc = -EPROTONOSUPPORT;
635 goto out2;
636 }
637 info->max_frmr_depth = min_t(int,
638 smbd_max_frmr_depth,
639 info->id->device->attrs.max_fast_reg_page_list_len);
640 info->mr_type = IB_MR_TYPE_MEM_REG;
641 if (info->id->device->attrs.device_cap_flags & IB_DEVICE_SG_GAPS_REG)
642 info->mr_type = IB_MR_TYPE_SG_GAPS;
643
644 info->pd = ib_alloc_pd(info->id->device, 0);
645 if (IS_ERR(info->pd)) {
646 rc = PTR_ERR(info->pd);
647 log_rdma_event(ERR, "ib_alloc_pd() returned %d\n", rc);
648 goto out2;
649 }
650
651 return 0;
652
653 out2:
654 rdma_destroy_id(info->id);
655 info->id = NULL;
656
657 out1:
658 return rc;
659 }
660
661 /*
662 * Send a negotiation request message to the peer
663 * The negotiation procedure is in [MS-SMBD] 3.1.5.2 and 3.1.5.3
664 * After negotiation, the transport is connected and ready for
665 * carrying upper layer SMB payload
666 */
667 static int smbd_post_send_negotiate_req(struct smbd_connection *info)
668 {
669 struct ib_send_wr send_wr;
670 int rc = -ENOMEM;
671 struct smbd_request *request;
672 struct smbd_negotiate_req *packet;
673
674 request = mempool_alloc(info->request_mempool, GFP_KERNEL);
675 if (!request)
676 return rc;
677
678 request->info = info;
679
680 packet = smbd_request_payload(request);
681 packet->min_version = cpu_to_le16(SMBD_V1);
682 packet->max_version = cpu_to_le16(SMBD_V1);
683 packet->reserved = 0;
684 packet->credits_requested = cpu_to_le16(info->send_credit_target);
685 packet->preferred_send_size = cpu_to_le32(info->max_send_size);
686 packet->max_receive_size = cpu_to_le32(info->max_receive_size);
687 packet->max_fragmented_size =
688 cpu_to_le32(info->max_fragmented_recv_size);
689
690 request->num_sge = 1;
691 request->sge[0].addr = ib_dma_map_single(
692 info->id->device, (void *)packet,
693 sizeof(*packet), DMA_TO_DEVICE);
694 if (ib_dma_mapping_error(info->id->device, request->sge[0].addr)) {
695 rc = -EIO;
696 goto dma_mapping_failed;
697 }
698
699 request->sge[0].length = sizeof(*packet);
700 request->sge[0].lkey = info->pd->local_dma_lkey;
701
702 ib_dma_sync_single_for_device(
703 info->id->device, request->sge[0].addr,
704 request->sge[0].length, DMA_TO_DEVICE);
705
706 request->cqe.done = send_done;
707
708 send_wr.next = NULL;
709 send_wr.wr_cqe = &request->cqe;
710 send_wr.sg_list = request->sge;
711 send_wr.num_sge = request->num_sge;
712 send_wr.opcode = IB_WR_SEND;
713 send_wr.send_flags = IB_SEND_SIGNALED;
714
715 log_rdma_send(INFO, "sge addr=%llx length=%x lkey=%x\n",
716 request->sge[0].addr,
717 request->sge[0].length, request->sge[0].lkey);
718
719 atomic_inc(&info->send_pending);
720 rc = ib_post_send(info->id->qp, &send_wr, NULL);
721 if (!rc)
722 return 0;
723
724 /* if we reach here, post send failed */
725 log_rdma_send(ERR, "ib_post_send failed rc=%d\n", rc);
726 atomic_dec(&info->send_pending);
727 ib_dma_unmap_single(info->id->device, request->sge[0].addr,
728 request->sge[0].length, DMA_TO_DEVICE);
729
730 smbd_disconnect_rdma_connection(info);
731
732 dma_mapping_failed:
733 mempool_free(request, info->request_mempool);
734 return rc;
735 }
736
737 /*
738 * Extend the credits to remote peer
739 * This implements [MS-SMBD] 3.1.5.9
740 * The idea is that we should extend credits to remote peer as quickly as
741 * it's allowed, to maintain data flow. We allocate as much receive
742 * buffer as possible, and extend the receive credits to remote peer
743 * return value: the new credtis being granted.
744 */
745 static int manage_credits_prior_sending(struct smbd_connection *info)
746 {
747 int new_credits;
748
749 spin_lock(&info->lock_new_credits_offered);
750 new_credits = info->new_credits_offered;
751 info->new_credits_offered = 0;
752 spin_unlock(&info->lock_new_credits_offered);
753
754 return new_credits;
755 }
756
757 /*
758 * Check if we need to send a KEEP_ALIVE message
759 * The idle connection timer triggers a KEEP_ALIVE message when expires
760 * SMB_DIRECT_RESPONSE_REQUESTED is set in the message flag to have peer send
761 * back a response.
762 * return value:
763 * 1 if SMB_DIRECT_RESPONSE_REQUESTED needs to be set
764 * 0: otherwise
765 */
766 static int manage_keep_alive_before_sending(struct smbd_connection *info)
767 {
768 if (info->keep_alive_requested == KEEP_ALIVE_PENDING) {
769 info->keep_alive_requested = KEEP_ALIVE_SENT;
770 return 1;
771 }
772 return 0;
773 }
774
775 /* Post the send request */
776 static int smbd_post_send(struct smbd_connection *info,
777 struct smbd_request *request)
778 {
779 struct ib_send_wr send_wr;
780 int rc, i;
781
782 for (i = 0; i < request->num_sge; i++) {
783 log_rdma_send(INFO,
784 "rdma_request sge[%d] addr=%llu length=%u\n",
785 i, request->sge[i].addr, request->sge[i].length);
786 ib_dma_sync_single_for_device(
787 info->id->device,
788 request->sge[i].addr,
789 request->sge[i].length,
790 DMA_TO_DEVICE);
791 }
792
793 request->cqe.done = send_done;
794
795 send_wr.next = NULL;
796 send_wr.wr_cqe = &request->cqe;
797 send_wr.sg_list = request->sge;
798 send_wr.num_sge = request->num_sge;
799 send_wr.opcode = IB_WR_SEND;
800 send_wr.send_flags = IB_SEND_SIGNALED;
801
802 rc = ib_post_send(info->id->qp, &send_wr, NULL);
803 if (rc) {
804 log_rdma_send(ERR, "ib_post_send failed rc=%d\n", rc);
805 smbd_disconnect_rdma_connection(info);
806 rc = -EAGAIN;
807 } else
808 /* Reset timer for idle connection after packet is sent */
809 mod_delayed_work(info->workqueue, &info->idle_timer_work,
810 info->keep_alive_interval*HZ);
811
812 return rc;
813 }
814
815 static int smbd_post_send_sgl(struct smbd_connection *info,
816 struct scatterlist *sgl, int data_length, int remaining_data_length)
817 {
818 int num_sgs;
819 int i, rc;
820 int header_length;
821 struct smbd_request *request;
822 struct smbd_data_transfer *packet;
823 int new_credits;
824 struct scatterlist *sg;
825
826 wait_credit:
827 /* Wait for send credits. A SMBD packet needs one credit */
828 rc = wait_event_interruptible(info->wait_send_queue,
829 atomic_read(&info->send_credits) > 0 ||
830 info->transport_status != SMBD_CONNECTED);
831 if (rc)
832 goto err_wait_credit;
833
834 if (info->transport_status != SMBD_CONNECTED) {
835 log_outgoing(ERR, "disconnected not sending on wait_credit\n");
836 rc = -EAGAIN;
837 goto err_wait_credit;
838 }
839 if (unlikely(atomic_dec_return(&info->send_credits) < 0)) {
840 atomic_inc(&info->send_credits);
841 goto wait_credit;
842 }
843
844 wait_send_queue:
845 wait_event(info->wait_post_send,
846 atomic_read(&info->send_pending) < info->send_credit_target ||
847 info->transport_status != SMBD_CONNECTED);
848
849 if (info->transport_status != SMBD_CONNECTED) {
850 log_outgoing(ERR, "disconnected not sending on wait_send_queue\n");
851 rc = -EAGAIN;
852 goto err_wait_send_queue;
853 }
854
855 if (unlikely(atomic_inc_return(&info->send_pending) >
856 info->send_credit_target)) {
857 atomic_dec(&info->send_pending);
858 goto wait_send_queue;
859 }
860
861 request = mempool_alloc(info->request_mempool, GFP_KERNEL);
862 if (!request) {
863 rc = -ENOMEM;
864 goto err_alloc;
865 }
866
867 request->info = info;
868
869 /* Fill in the packet header */
870 packet = smbd_request_payload(request);
871 packet->credits_requested = cpu_to_le16(info->send_credit_target);
872
873 new_credits = manage_credits_prior_sending(info);
874 atomic_add(new_credits, &info->receive_credits);
875 packet->credits_granted = cpu_to_le16(new_credits);
876
877 info->send_immediate = false;
878
879 packet->flags = 0;
880 if (manage_keep_alive_before_sending(info))
881 packet->flags |= cpu_to_le16(SMB_DIRECT_RESPONSE_REQUESTED);
882
883 packet->reserved = 0;
884 if (!data_length)
885 packet->data_offset = 0;
886 else
887 packet->data_offset = cpu_to_le32(24);
888 packet->data_length = cpu_to_le32(data_length);
889 packet->remaining_data_length = cpu_to_le32(remaining_data_length);
890 packet->padding = 0;
891
892 log_outgoing(INFO, "credits_requested=%d credits_granted=%d data_offset=%d data_length=%d remaining_data_length=%d\n",
893 le16_to_cpu(packet->credits_requested),
894 le16_to_cpu(packet->credits_granted),
895 le32_to_cpu(packet->data_offset),
896 le32_to_cpu(packet->data_length),
897 le32_to_cpu(packet->remaining_data_length));
898
899 /* Map the packet to DMA */
900 header_length = sizeof(struct smbd_data_transfer);
901 /* If this is a packet without payload, don't send padding */
902 if (!data_length)
903 header_length = offsetof(struct smbd_data_transfer, padding);
904
905 request->num_sge = 1;
906 request->sge[0].addr = ib_dma_map_single(info->id->device,
907 (void *)packet,
908 header_length,
909 DMA_TO_DEVICE);
910 if (ib_dma_mapping_error(info->id->device, request->sge[0].addr)) {
911 rc = -EIO;
912 request->sge[0].addr = 0;
913 goto err_dma;
914 }
915
916 request->sge[0].length = header_length;
917 request->sge[0].lkey = info->pd->local_dma_lkey;
918
919 /* Fill in the packet data payload */
920 num_sgs = sgl ? sg_nents(sgl) : 0;
921 for_each_sg(sgl, sg, num_sgs, i) {
922 request->sge[i+1].addr =
923 ib_dma_map_page(info->id->device, sg_page(sg),
924 sg->offset, sg->length, DMA_TO_DEVICE);
925 if (ib_dma_mapping_error(
926 info->id->device, request->sge[i+1].addr)) {
927 rc = -EIO;
928 request->sge[i+1].addr = 0;
929 goto err_dma;
930 }
931 request->sge[i+1].length = sg->length;
932 request->sge[i+1].lkey = info->pd->local_dma_lkey;
933 request->num_sge++;
934 }
935
936 rc = smbd_post_send(info, request);
937 if (!rc)
938 return 0;
939
940 err_dma:
941 for (i = 0; i < request->num_sge; i++)
942 if (request->sge[i].addr)
943 ib_dma_unmap_single(info->id->device,
944 request->sge[i].addr,
945 request->sge[i].length,
946 DMA_TO_DEVICE);
947 mempool_free(request, info->request_mempool);
948
949 /* roll back receive credits and credits to be offered */
950 spin_lock(&info->lock_new_credits_offered);
951 info->new_credits_offered += new_credits;
952 spin_unlock(&info->lock_new_credits_offered);
953 atomic_sub(new_credits, &info->receive_credits);
954
955 err_alloc:
956 if (atomic_dec_and_test(&info->send_pending))
957 wake_up(&info->wait_send_pending);
958
959 err_wait_send_queue:
960 /* roll back send credits and pending */
961 atomic_inc(&info->send_credits);
962
963 err_wait_credit:
964 return rc;
965 }
966
967 /*
968 * Send a page
969 * page: the page to send
970 * offset: offset in the page to send
971 * size: length in the page to send
972 * remaining_data_length: remaining data to send in this payload
973 */
974 static int smbd_post_send_page(struct smbd_connection *info, struct page *page,
975 unsigned long offset, size_t size, int remaining_data_length)
976 {
977 struct scatterlist sgl;
978
979 sg_init_table(&sgl, 1);
980 sg_set_page(&sgl, page, size, offset);
981
982 return smbd_post_send_sgl(info, &sgl, size, remaining_data_length);
983 }
984
985 /*
986 * Send an empty message
987 * Empty message is used to extend credits to peer to for keep live
988 * while there is no upper layer payload to send at the time
989 */
990 static int smbd_post_send_empty(struct smbd_connection *info)
991 {
992 info->count_send_empty++;
993 return smbd_post_send_sgl(info, NULL, 0, 0);
994 }
995
996 /*
997 * Send a data buffer
998 * iov: the iov array describing the data buffers
999 * n_vec: number of iov array
1000 * remaining_data_length: remaining data to send following this packet
1001 * in segmented SMBD packet
1002 */
1003 static int smbd_post_send_data(
1004 struct smbd_connection *info, struct kvec *iov, int n_vec,
1005 int remaining_data_length)
1006 {
1007 int i;
1008 u32 data_length = 0;
1009 struct scatterlist sgl[SMBDIRECT_MAX_SGE];
1010
1011 if (n_vec > SMBDIRECT_MAX_SGE) {
1012 cifs_dbg(VFS, "Can't fit data to SGL, n_vec=%d\n", n_vec);
1013 return -EINVAL;
1014 }
1015
1016 sg_init_table(sgl, n_vec);
1017 for (i = 0; i < n_vec; i++) {
1018 data_length += iov[i].iov_len;
1019 sg_set_buf(&sgl[i], iov[i].iov_base, iov[i].iov_len);
1020 }
1021
1022 return smbd_post_send_sgl(info, sgl, data_length, remaining_data_length);
1023 }
1024
1025 /*
1026 * Post a receive request to the transport
1027 * The remote peer can only send data when a receive request is posted
1028 * The interaction is controlled by send/receive credit system
1029 */
1030 static int smbd_post_recv(
1031 struct smbd_connection *info, struct smbd_response *response)
1032 {
1033 struct ib_recv_wr recv_wr;
1034 int rc = -EIO;
1035
1036 response->sge.addr = ib_dma_map_single(
1037 info->id->device, response->packet,
1038 info->max_receive_size, DMA_FROM_DEVICE);
1039 if (ib_dma_mapping_error(info->id->device, response->sge.addr))
1040 return rc;
1041
1042 response->sge.length = info->max_receive_size;
1043 response->sge.lkey = info->pd->local_dma_lkey;
1044
1045 response->cqe.done = recv_done;
1046
1047 recv_wr.wr_cqe = &response->cqe;
1048 recv_wr.next = NULL;
1049 recv_wr.sg_list = &response->sge;
1050 recv_wr.num_sge = 1;
1051
1052 rc = ib_post_recv(info->id->qp, &recv_wr, NULL);
1053 if (rc) {
1054 ib_dma_unmap_single(info->id->device, response->sge.addr,
1055 response->sge.length, DMA_FROM_DEVICE);
1056 smbd_disconnect_rdma_connection(info);
1057 log_rdma_recv(ERR, "ib_post_recv failed rc=%d\n", rc);
1058 }
1059
1060 return rc;
1061 }
1062
1063 /* Perform SMBD negotiate according to [MS-SMBD] 3.1.5.2 */
1064 static int smbd_negotiate(struct smbd_connection *info)
1065 {
1066 int rc;
1067 struct smbd_response *response = get_receive_buffer(info);
1068
1069 response->type = SMBD_NEGOTIATE_RESP;
1070 rc = smbd_post_recv(info, response);
1071 log_rdma_event(INFO, "smbd_post_recv rc=%d iov.addr=%llx iov.length=%x iov.lkey=%x\n",
1072 rc, response->sge.addr,
1073 response->sge.length, response->sge.lkey);
1074 if (rc)
1075 return rc;
1076
1077 init_completion(&info->negotiate_completion);
1078 info->negotiate_done = false;
1079 rc = smbd_post_send_negotiate_req(info);
1080 if (rc)
1081 return rc;
1082
1083 rc = wait_for_completion_interruptible_timeout(
1084 &info->negotiate_completion, SMBD_NEGOTIATE_TIMEOUT * HZ);
1085 log_rdma_event(INFO, "wait_for_completion_timeout rc=%d\n", rc);
1086
1087 if (info->negotiate_done)
1088 return 0;
1089
1090 if (rc == 0)
1091 rc = -ETIMEDOUT;
1092 else if (rc == -ERESTARTSYS)
1093 rc = -EINTR;
1094 else
1095 rc = -ENOTCONN;
1096
1097 return rc;
1098 }
1099
1100 static void put_empty_packet(
1101 struct smbd_connection *info, struct smbd_response *response)
1102 {
1103 spin_lock(&info->empty_packet_queue_lock);
1104 list_add_tail(&response->list, &info->empty_packet_queue);
1105 info->count_empty_packet_queue++;
1106 spin_unlock(&info->empty_packet_queue_lock);
1107
1108 queue_work(info->workqueue, &info->post_send_credits_work);
1109 }
1110
1111 /*
1112 * Implement Connection.FragmentReassemblyBuffer defined in [MS-SMBD] 3.1.1.1
1113 * This is a queue for reassembling upper layer payload and present to upper
1114 * layer. All the inncoming payload go to the reassembly queue, regardless of
1115 * if reassembly is required. The uuper layer code reads from the queue for all
1116 * incoming payloads.
1117 * Put a received packet to the reassembly queue
1118 * response: the packet received
1119 * data_length: the size of payload in this packet
1120 */
1121 static void enqueue_reassembly(
1122 struct smbd_connection *info,
1123 struct smbd_response *response,
1124 int data_length)
1125 {
1126 spin_lock(&info->reassembly_queue_lock);
1127 list_add_tail(&response->list, &info->reassembly_queue);
1128 info->reassembly_queue_length++;
1129 /*
1130 * Make sure reassembly_data_length is updated after list and
1131 * reassembly_queue_length are updated. On the dequeue side
1132 * reassembly_data_length is checked without a lock to determine
1133 * if reassembly_queue_length and list is up to date
1134 */
1135 virt_wmb();
1136 info->reassembly_data_length += data_length;
1137 spin_unlock(&info->reassembly_queue_lock);
1138 info->count_reassembly_queue++;
1139 info->count_enqueue_reassembly_queue++;
1140 }
1141
1142 /*
1143 * Get the first entry at the front of reassembly queue
1144 * Caller is responsible for locking
1145 * return value: the first entry if any, NULL if queue is empty
1146 */
1147 static struct smbd_response *_get_first_reassembly(struct smbd_connection *info)
1148 {
1149 struct smbd_response *ret = NULL;
1150
1151 if (!list_empty(&info->reassembly_queue)) {
1152 ret = list_first_entry(
1153 &info->reassembly_queue,
1154 struct smbd_response, list);
1155 }
1156 return ret;
1157 }
1158
1159 static struct smbd_response *get_empty_queue_buffer(
1160 struct smbd_connection *info)
1161 {
1162 struct smbd_response *ret = NULL;
1163 unsigned long flags;
1164
1165 spin_lock_irqsave(&info->empty_packet_queue_lock, flags);
1166 if (!list_empty(&info->empty_packet_queue)) {
1167 ret = list_first_entry(
1168 &info->empty_packet_queue,
1169 struct smbd_response, list);
1170 list_del(&ret->list);
1171 info->count_empty_packet_queue--;
1172 }
1173 spin_unlock_irqrestore(&info->empty_packet_queue_lock, flags);
1174
1175 return ret;
1176 }
1177
1178 /*
1179 * Get a receive buffer
1180 * For each remote send, we need to post a receive. The receive buffers are
1181 * pre-allocated in advance.
1182 * return value: the receive buffer, NULL if none is available
1183 */
1184 static struct smbd_response *get_receive_buffer(struct smbd_connection *info)
1185 {
1186 struct smbd_response *ret = NULL;
1187 unsigned long flags;
1188
1189 spin_lock_irqsave(&info->receive_queue_lock, flags);
1190 if (!list_empty(&info->receive_queue)) {
1191 ret = list_first_entry(
1192 &info->receive_queue,
1193 struct smbd_response, list);
1194 list_del(&ret->list);
1195 info->count_receive_queue--;
1196 info->count_get_receive_buffer++;
1197 }
1198 spin_unlock_irqrestore(&info->receive_queue_lock, flags);
1199
1200 return ret;
1201 }
1202
1203 /*
1204 * Return a receive buffer
1205 * Upon returning of a receive buffer, we can post new receive and extend
1206 * more receive credits to remote peer. This is done immediately after a
1207 * receive buffer is returned.
1208 */
1209 static void put_receive_buffer(
1210 struct smbd_connection *info, struct smbd_response *response)
1211 {
1212 unsigned long flags;
1213
1214 ib_dma_unmap_single(info->id->device, response->sge.addr,
1215 response->sge.length, DMA_FROM_DEVICE);
1216
1217 spin_lock_irqsave(&info->receive_queue_lock, flags);
1218 list_add_tail(&response->list, &info->receive_queue);
1219 info->count_receive_queue++;
1220 info->count_put_receive_buffer++;
1221 spin_unlock_irqrestore(&info->receive_queue_lock, flags);
1222
1223 queue_work(info->workqueue, &info->post_send_credits_work);
1224 }
1225
1226 /* Preallocate all receive buffer on transport establishment */
1227 static int allocate_receive_buffers(struct smbd_connection *info, int num_buf)
1228 {
1229 int i;
1230 struct smbd_response *response;
1231
1232 INIT_LIST_HEAD(&info->reassembly_queue);
1233 spin_lock_init(&info->reassembly_queue_lock);
1234 info->reassembly_data_length = 0;
1235 info->reassembly_queue_length = 0;
1236
1237 INIT_LIST_HEAD(&info->receive_queue);
1238 spin_lock_init(&info->receive_queue_lock);
1239 info->count_receive_queue = 0;
1240
1241 INIT_LIST_HEAD(&info->empty_packet_queue);
1242 spin_lock_init(&info->empty_packet_queue_lock);
1243 info->count_empty_packet_queue = 0;
1244
1245 init_waitqueue_head(&info->wait_receive_queues);
1246
1247 for (i = 0; i < num_buf; i++) {
1248 response = mempool_alloc(info->response_mempool, GFP_KERNEL);
1249 if (!response)
1250 goto allocate_failed;
1251
1252 response->info = info;
1253 list_add_tail(&response->list, &info->receive_queue);
1254 info->count_receive_queue++;
1255 }
1256
1257 return 0;
1258
1259 allocate_failed:
1260 while (!list_empty(&info->receive_queue)) {
1261 response = list_first_entry(
1262 &info->receive_queue,
1263 struct smbd_response, list);
1264 list_del(&response->list);
1265 info->count_receive_queue--;
1266
1267 mempool_free(response, info->response_mempool);
1268 }
1269 return -ENOMEM;
1270 }
1271
1272 static void destroy_receive_buffers(struct smbd_connection *info)
1273 {
1274 struct smbd_response *response;
1275
1276 while ((response = get_receive_buffer(info)))
1277 mempool_free(response, info->response_mempool);
1278
1279 while ((response = get_empty_queue_buffer(info)))
1280 mempool_free(response, info->response_mempool);
1281 }
1282
1283 /* Implement idle connection timer [MS-SMBD] 3.1.6.2 */
1284 static void idle_connection_timer(struct work_struct *work)
1285 {
1286 struct smbd_connection *info = container_of(
1287 work, struct smbd_connection,
1288 idle_timer_work.work);
1289
1290 if (info->keep_alive_requested != KEEP_ALIVE_NONE) {
1291 log_keep_alive(ERR,
1292 "error status info->keep_alive_requested=%d\n",
1293 info->keep_alive_requested);
1294 smbd_disconnect_rdma_connection(info);
1295 return;
1296 }
1297
1298 log_keep_alive(INFO, "about to send an empty idle message\n");
1299 smbd_post_send_empty(info);
1300
1301 /* Setup the next idle timeout work */
1302 queue_delayed_work(info->workqueue, &info->idle_timer_work,
1303 info->keep_alive_interval*HZ);
1304 }
1305
1306 /*
1307 * Destroy the transport and related RDMA and memory resources
1308 * Need to go through all the pending counters and make sure on one is using
1309 * the transport while it is destroyed
1310 */
1311 void smbd_destroy(struct TCP_Server_Info *server)
1312 {
1313 struct smbd_connection *info = server->smbd_conn;
1314 struct smbd_response *response;
1315 unsigned long flags;
1316
1317 if (!info) {
1318 log_rdma_event(INFO, "rdma session already destroyed\n");
1319 return;
1320 }
1321
1322 log_rdma_event(INFO, "destroying rdma session\n");
1323 if (info->transport_status != SMBD_DISCONNECTED) {
1324 rdma_disconnect(server->smbd_conn->id);
1325 log_rdma_event(INFO, "wait for transport being disconnected\n");
1326 wait_event_interruptible(
1327 info->disconn_wait,
1328 info->transport_status == SMBD_DISCONNECTED);
1329 }
1330
1331 log_rdma_event(INFO, "destroying qp\n");
1332 ib_drain_qp(info->id->qp);
1333 rdma_destroy_qp(info->id);
1334
1335 log_rdma_event(INFO, "cancelling idle timer\n");
1336 cancel_delayed_work_sync(&info->idle_timer_work);
1337
1338 log_rdma_event(INFO, "wait for all send posted to IB to finish\n");
1339 wait_event(info->wait_send_pending,
1340 atomic_read(&info->send_pending) == 0);
1341
1342 /* It's not posssible for upper layer to get to reassembly */
1343 log_rdma_event(INFO, "drain the reassembly queue\n");
1344 do {
1345 spin_lock_irqsave(&info->reassembly_queue_lock, flags);
1346 response = _get_first_reassembly(info);
1347 if (response) {
1348 list_del(&response->list);
1349 spin_unlock_irqrestore(
1350 &info->reassembly_queue_lock, flags);
1351 put_receive_buffer(info, response);
1352 } else
1353 spin_unlock_irqrestore(
1354 &info->reassembly_queue_lock, flags);
1355 } while (response);
1356 info->reassembly_data_length = 0;
1357
1358 log_rdma_event(INFO, "free receive buffers\n");
1359 wait_event(info->wait_receive_queues,
1360 info->count_receive_queue + info->count_empty_packet_queue
1361 == info->receive_credit_max);
1362 destroy_receive_buffers(info);
1363
1364 /*
1365 * For performance reasons, memory registration and deregistration
1366 * are not locked by srv_mutex. It is possible some processes are
1367 * blocked on transport srv_mutex while holding memory registration.
1368 * Release the transport srv_mutex to allow them to hit the failure
1369 * path when sending data, and then release memory registartions.
1370 */
1371 log_rdma_event(INFO, "freeing mr list\n");
1372 wake_up_interruptible_all(&info->wait_mr);
1373 while (atomic_read(&info->mr_used_count)) {
1374 mutex_unlock(&server->srv_mutex);
1375 msleep(1000);
1376 mutex_lock(&server->srv_mutex);
1377 }
1378 destroy_mr_list(info);
1379
1380 ib_free_cq(info->send_cq);
1381 ib_free_cq(info->recv_cq);
1382 ib_dealloc_pd(info->pd);
1383 rdma_destroy_id(info->id);
1384
1385 /* free mempools */
1386 mempool_destroy(info->request_mempool);
1387 kmem_cache_destroy(info->request_cache);
1388
1389 mempool_destroy(info->response_mempool);
1390 kmem_cache_destroy(info->response_cache);
1391
1392 info->transport_status = SMBD_DESTROYED;
1393
1394 destroy_workqueue(info->workqueue);
1395 log_rdma_event(INFO, "rdma session destroyed\n");
1396 kfree(info);
1397 }
1398
1399 /*
1400 * Reconnect this SMBD connection, called from upper layer
1401 * return value: 0 on success, or actual error code
1402 */
1403 int smbd_reconnect(struct TCP_Server_Info *server)
1404 {
1405 log_rdma_event(INFO, "reconnecting rdma session\n");
1406
1407 if (!server->smbd_conn) {
1408 log_rdma_event(INFO, "rdma session already destroyed\n");
1409 goto create_conn;
1410 }
1411
1412 /*
1413 * This is possible if transport is disconnected and we haven't received
1414 * notification from RDMA, but upper layer has detected timeout
1415 */
1416 if (server->smbd_conn->transport_status == SMBD_CONNECTED) {
1417 log_rdma_event(INFO, "disconnecting transport\n");
1418 smbd_destroy(server);
1419 }
1420
1421 create_conn:
1422 log_rdma_event(INFO, "creating rdma session\n");
1423 server->smbd_conn = smbd_get_connection(
1424 server, (struct sockaddr *) &server->dstaddr);
1425
1426 if (server->smbd_conn)
1427 cifs_dbg(VFS, "RDMA transport re-established\n");
1428
1429 return server->smbd_conn ? 0 : -ENOENT;
1430 }
1431
1432 static void destroy_caches_and_workqueue(struct smbd_connection *info)
1433 {
1434 destroy_receive_buffers(info);
1435 destroy_workqueue(info->workqueue);
1436 mempool_destroy(info->response_mempool);
1437 kmem_cache_destroy(info->response_cache);
1438 mempool_destroy(info->request_mempool);
1439 kmem_cache_destroy(info->request_cache);
1440 }
1441
1442 #define MAX_NAME_LEN 80
1443 static int allocate_caches_and_workqueue(struct smbd_connection *info)
1444 {
1445 char name[MAX_NAME_LEN];
1446 int rc;
1447
1448 scnprintf(name, MAX_NAME_LEN, "smbd_request_%p", info);
1449 info->request_cache =
1450 kmem_cache_create(
1451 name,
1452 sizeof(struct smbd_request) +
1453 sizeof(struct smbd_data_transfer),
1454 0, SLAB_HWCACHE_ALIGN, NULL);
1455 if (!info->request_cache)
1456 return -ENOMEM;
1457
1458 info->request_mempool =
1459 mempool_create(info->send_credit_target, mempool_alloc_slab,
1460 mempool_free_slab, info->request_cache);
1461 if (!info->request_mempool)
1462 goto out1;
1463
1464 scnprintf(name, MAX_NAME_LEN, "smbd_response_%p", info);
1465 info->response_cache =
1466 kmem_cache_create(
1467 name,
1468 sizeof(struct smbd_response) +
1469 info->max_receive_size,
1470 0, SLAB_HWCACHE_ALIGN, NULL);
1471 if (!info->response_cache)
1472 goto out2;
1473
1474 info->response_mempool =
1475 mempool_create(info->receive_credit_max, mempool_alloc_slab,
1476 mempool_free_slab, info->response_cache);
1477 if (!info->response_mempool)
1478 goto out3;
1479
1480 scnprintf(name, MAX_NAME_LEN, "smbd_%p", info);
1481 info->workqueue = create_workqueue(name);
1482 if (!info->workqueue)
1483 goto out4;
1484
1485 rc = allocate_receive_buffers(info, info->receive_credit_max);
1486 if (rc) {
1487 log_rdma_event(ERR, "failed to allocate receive buffers\n");
1488 goto out5;
1489 }
1490
1491 return 0;
1492
1493 out5:
1494 destroy_workqueue(info->workqueue);
1495 out4:
1496 mempool_destroy(info->response_mempool);
1497 out3:
1498 kmem_cache_destroy(info->response_cache);
1499 out2:
1500 mempool_destroy(info->request_mempool);
1501 out1:
1502 kmem_cache_destroy(info->request_cache);
1503 return -ENOMEM;
1504 }
1505
1506 /* Create a SMBD connection, called by upper layer */
1507 static struct smbd_connection *_smbd_get_connection(
1508 struct TCP_Server_Info *server, struct sockaddr *dstaddr, int port)
1509 {
1510 int rc;
1511 struct smbd_connection *info;
1512 struct rdma_conn_param conn_param;
1513 struct ib_qp_init_attr qp_attr;
1514 struct sockaddr_in *addr_in = (struct sockaddr_in *) dstaddr;
1515 struct ib_port_immutable port_immutable;
1516 u32 ird_ord_hdr[2];
1517
1518 info = kzalloc(sizeof(struct smbd_connection), GFP_KERNEL);
1519 if (!info)
1520 return NULL;
1521
1522 info->transport_status = SMBD_CONNECTING;
1523 rc = smbd_ia_open(info, dstaddr, port);
1524 if (rc) {
1525 log_rdma_event(INFO, "smbd_ia_open rc=%d\n", rc);
1526 goto create_id_failed;
1527 }
1528
1529 if (smbd_send_credit_target > info->id->device->attrs.max_cqe ||
1530 smbd_send_credit_target > info->id->device->attrs.max_qp_wr) {
1531 log_rdma_event(ERR, "consider lowering send_credit_target = %d. Possible CQE overrun, device reporting max_cpe %d max_qp_wr %d\n",
1532 smbd_send_credit_target,
1533 info->id->device->attrs.max_cqe,
1534 info->id->device->attrs.max_qp_wr);
1535 goto config_failed;
1536 }
1537
1538 if (smbd_receive_credit_max > info->id->device->attrs.max_cqe ||
1539 smbd_receive_credit_max > info->id->device->attrs.max_qp_wr) {
1540 log_rdma_event(ERR, "consider lowering receive_credit_max = %d. Possible CQE overrun, device reporting max_cpe %d max_qp_wr %d\n",
1541 smbd_receive_credit_max,
1542 info->id->device->attrs.max_cqe,
1543 info->id->device->attrs.max_qp_wr);
1544 goto config_failed;
1545 }
1546
1547 info->receive_credit_max = smbd_receive_credit_max;
1548 info->send_credit_target = smbd_send_credit_target;
1549 info->max_send_size = smbd_max_send_size;
1550 info->max_fragmented_recv_size = smbd_max_fragmented_recv_size;
1551 info->max_receive_size = smbd_max_receive_size;
1552 info->keep_alive_interval = smbd_keep_alive_interval;
1553
1554 if (info->id->device->attrs.max_send_sge < SMBDIRECT_MAX_SGE) {
1555 log_rdma_event(ERR,
1556 "warning: device max_send_sge = %d too small\n",
1557 info->id->device->attrs.max_send_sge);
1558 log_rdma_event(ERR, "Queue Pair creation may fail\n");
1559 }
1560 if (info->id->device->attrs.max_recv_sge < SMBDIRECT_MAX_SGE) {
1561 log_rdma_event(ERR,
1562 "warning: device max_recv_sge = %d too small\n",
1563 info->id->device->attrs.max_recv_sge);
1564 log_rdma_event(ERR, "Queue Pair creation may fail\n");
1565 }
1566
1567 info->send_cq = NULL;
1568 info->recv_cq = NULL;
1569 info->send_cq =
1570 ib_alloc_cq_any(info->id->device, info,
1571 info->send_credit_target, IB_POLL_SOFTIRQ);
1572 if (IS_ERR(info->send_cq)) {
1573 info->send_cq = NULL;
1574 goto alloc_cq_failed;
1575 }
1576
1577 info->recv_cq =
1578 ib_alloc_cq_any(info->id->device, info,
1579 info->receive_credit_max, IB_POLL_SOFTIRQ);
1580 if (IS_ERR(info->recv_cq)) {
1581 info->recv_cq = NULL;
1582 goto alloc_cq_failed;
1583 }
1584
1585 memset(&qp_attr, 0, sizeof(qp_attr));
1586 qp_attr.event_handler = smbd_qp_async_error_upcall;
1587 qp_attr.qp_context = info;
1588 qp_attr.cap.max_send_wr = info->send_credit_target;
1589 qp_attr.cap.max_recv_wr = info->receive_credit_max;
1590 qp_attr.cap.max_send_sge = SMBDIRECT_MAX_SGE;
1591 qp_attr.cap.max_recv_sge = SMBDIRECT_MAX_SGE;
1592 qp_attr.cap.max_inline_data = 0;
1593 qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
1594 qp_attr.qp_type = IB_QPT_RC;
1595 qp_attr.send_cq = info->send_cq;
1596 qp_attr.recv_cq = info->recv_cq;
1597 qp_attr.port_num = ~0;
1598
1599 rc = rdma_create_qp(info->id, info->pd, &qp_attr);
1600 if (rc) {
1601 log_rdma_event(ERR, "rdma_create_qp failed %i\n", rc);
1602 goto create_qp_failed;
1603 }
1604
1605 memset(&conn_param, 0, sizeof(conn_param));
1606 conn_param.initiator_depth = 0;
1607
1608 conn_param.responder_resources =
1609 info->id->device->attrs.max_qp_rd_atom
1610 < SMBD_CM_RESPONDER_RESOURCES ?
1611 info->id->device->attrs.max_qp_rd_atom :
1612 SMBD_CM_RESPONDER_RESOURCES;
1613 info->responder_resources = conn_param.responder_resources;
1614 log_rdma_mr(INFO, "responder_resources=%d\n",
1615 info->responder_resources);
1616
1617 /* Need to send IRD/ORD in private data for iWARP */
1618 info->id->device->ops.get_port_immutable(
1619 info->id->device, info->id->port_num, &port_immutable);
1620 if (port_immutable.core_cap_flags & RDMA_CORE_PORT_IWARP) {
1621 ird_ord_hdr[0] = info->responder_resources;
1622 ird_ord_hdr[1] = 1;
1623 conn_param.private_data = ird_ord_hdr;
1624 conn_param.private_data_len = sizeof(ird_ord_hdr);
1625 } else {
1626 conn_param.private_data = NULL;
1627 conn_param.private_data_len = 0;
1628 }
1629
1630 conn_param.retry_count = SMBD_CM_RETRY;
1631 conn_param.rnr_retry_count = SMBD_CM_RNR_RETRY;
1632 conn_param.flow_control = 0;
1633
1634 log_rdma_event(INFO, "connecting to IP %pI4 port %d\n",
1635 &addr_in->sin_addr, port);
1636
1637 init_waitqueue_head(&info->conn_wait);
1638 init_waitqueue_head(&info->disconn_wait);
1639 init_waitqueue_head(&info->wait_reassembly_queue);
1640 rc = rdma_connect(info->id, &conn_param);
1641 if (rc) {
1642 log_rdma_event(ERR, "rdma_connect() failed with %i\n", rc);
1643 goto rdma_connect_failed;
1644 }
1645
1646 wait_event_interruptible(
1647 info->conn_wait, info->transport_status != SMBD_CONNECTING);
1648
1649 if (info->transport_status != SMBD_CONNECTED) {
1650 log_rdma_event(ERR, "rdma_connect failed port=%d\n", port);
1651 goto rdma_connect_failed;
1652 }
1653
1654 log_rdma_event(INFO, "rdma_connect connected\n");
1655
1656 rc = allocate_caches_and_workqueue(info);
1657 if (rc) {
1658 log_rdma_event(ERR, "cache allocation failed\n");
1659 goto allocate_cache_failed;
1660 }
1661
1662 init_waitqueue_head(&info->wait_send_queue);
1663 INIT_DELAYED_WORK(&info->idle_timer_work, idle_connection_timer);
1664 queue_delayed_work(info->workqueue, &info->idle_timer_work,
1665 info->keep_alive_interval*HZ);
1666
1667 init_waitqueue_head(&info->wait_send_pending);
1668 atomic_set(&info->send_pending, 0);
1669
1670 init_waitqueue_head(&info->wait_post_send);
1671
1672 INIT_WORK(&info->disconnect_work, smbd_disconnect_rdma_work);
1673 INIT_WORK(&info->post_send_credits_work, smbd_post_send_credits);
1674 info->new_credits_offered = 0;
1675 spin_lock_init(&info->lock_new_credits_offered);
1676
1677 rc = smbd_negotiate(info);
1678 if (rc) {
1679 log_rdma_event(ERR, "smbd_negotiate rc=%d\n", rc);
1680 goto negotiation_failed;
1681 }
1682
1683 rc = allocate_mr_list(info);
1684 if (rc) {
1685 log_rdma_mr(ERR, "memory registration allocation failed\n");
1686 goto allocate_mr_failed;
1687 }
1688
1689 return info;
1690
1691 allocate_mr_failed:
1692 /* At this point, need to a full transport shutdown */
1693 smbd_destroy(server);
1694 return NULL;
1695
1696 negotiation_failed:
1697 cancel_delayed_work_sync(&info->idle_timer_work);
1698 destroy_caches_and_workqueue(info);
1699 info->transport_status = SMBD_NEGOTIATE_FAILED;
1700 init_waitqueue_head(&info->conn_wait);
1701 rdma_disconnect(info->id);
1702 wait_event(info->conn_wait,
1703 info->transport_status == SMBD_DISCONNECTED);
1704
1705 allocate_cache_failed:
1706 rdma_connect_failed:
1707 rdma_destroy_qp(info->id);
1708
1709 create_qp_failed:
1710 alloc_cq_failed:
1711 if (info->send_cq)
1712 ib_free_cq(info->send_cq);
1713 if (info->recv_cq)
1714 ib_free_cq(info->recv_cq);
1715
1716 config_failed:
1717 ib_dealloc_pd(info->pd);
1718 rdma_destroy_id(info->id);
1719
1720 create_id_failed:
1721 kfree(info);
1722 return NULL;
1723 }
1724
1725 struct smbd_connection *smbd_get_connection(
1726 struct TCP_Server_Info *server, struct sockaddr *dstaddr)
1727 {
1728 struct smbd_connection *ret;
1729 int port = SMBD_PORT;
1730
1731 try_again:
1732 ret = _smbd_get_connection(server, dstaddr, port);
1733
1734 /* Try SMB_PORT if SMBD_PORT doesn't work */
1735 if (!ret && port == SMBD_PORT) {
1736 port = SMB_PORT;
1737 goto try_again;
1738 }
1739 return ret;
1740 }
1741
1742 /*
1743 * Receive data from receive reassembly queue
1744 * All the incoming data packets are placed in reassembly queue
1745 * buf: the buffer to read data into
1746 * size: the length of data to read
1747 * return value: actual data read
1748 * Note: this implementation copies the data from reassebmly queue to receive
1749 * buffers used by upper layer. This is not the optimal code path. A better way
1750 * to do it is to not have upper layer allocate its receive buffers but rather
1751 * borrow the buffer from reassembly queue, and return it after data is
1752 * consumed. But this will require more changes to upper layer code, and also
1753 * need to consider packet boundaries while they still being reassembled.
1754 */
1755 static int smbd_recv_buf(struct smbd_connection *info, char *buf,
1756 unsigned int size)
1757 {
1758 struct smbd_response *response;
1759 struct smbd_data_transfer *data_transfer;
1760 int to_copy, to_read, data_read, offset;
1761 u32 data_length, remaining_data_length, data_offset;
1762 int rc;
1763
1764 again:
1765 /*
1766 * No need to hold the reassembly queue lock all the time as we are
1767 * the only one reading from the front of the queue. The transport
1768 * may add more entries to the back of the queue at the same time
1769 */
1770 log_read(INFO, "size=%d info->reassembly_data_length=%d\n", size,
1771 info->reassembly_data_length);
1772 if (info->reassembly_data_length >= size) {
1773 int queue_length;
1774 int queue_removed = 0;
1775
1776 /*
1777 * Need to make sure reassembly_data_length is read before
1778 * reading reassembly_queue_length and calling
1779 * _get_first_reassembly. This call is lock free
1780 * as we never read at the end of the queue which are being
1781 * updated in SOFTIRQ as more data is received
1782 */
1783 virt_rmb();
1784 queue_length = info->reassembly_queue_length;
1785 data_read = 0;
1786 to_read = size;
1787 offset = info->first_entry_offset;
1788 while (data_read < size) {
1789 response = _get_first_reassembly(info);
1790 data_transfer = smbd_response_payload(response);
1791 data_length = le32_to_cpu(data_transfer->data_length);
1792 remaining_data_length =
1793 le32_to_cpu(
1794 data_transfer->remaining_data_length);
1795 data_offset = le32_to_cpu(data_transfer->data_offset);
1796
1797 /*
1798 * The upper layer expects RFC1002 length at the
1799 * beginning of the payload. Return it to indicate
1800 * the total length of the packet. This minimize the
1801 * change to upper layer packet processing logic. This
1802 * will be eventually remove when an intermediate
1803 * transport layer is added
1804 */
1805 if (response->first_segment && size == 4) {
1806 unsigned int rfc1002_len =
1807 data_length + remaining_data_length;
1808 *((__be32 *)buf) = cpu_to_be32(rfc1002_len);
1809 data_read = 4;
1810 response->first_segment = false;
1811 log_read(INFO, "returning rfc1002 length %d\n",
1812 rfc1002_len);
1813 goto read_rfc1002_done;
1814 }
1815
1816 to_copy = min_t(int, data_length - offset, to_read);
1817 memcpy(
1818 buf + data_read,
1819 (char *)data_transfer + data_offset + offset,
1820 to_copy);
1821
1822 /* move on to the next buffer? */
1823 if (to_copy == data_length - offset) {
1824 queue_length--;
1825 /*
1826 * No need to lock if we are not at the
1827 * end of the queue
1828 */
1829 if (queue_length)
1830 list_del(&response->list);
1831 else {
1832 spin_lock_irq(
1833 &info->reassembly_queue_lock);
1834 list_del(&response->list);
1835 spin_unlock_irq(
1836 &info->reassembly_queue_lock);
1837 }
1838 queue_removed++;
1839 info->count_reassembly_queue--;
1840 info->count_dequeue_reassembly_queue++;
1841 put_receive_buffer(info, response);
1842 offset = 0;
1843 log_read(INFO, "put_receive_buffer offset=0\n");
1844 } else
1845 offset += to_copy;
1846
1847 to_read -= to_copy;
1848 data_read += to_copy;
1849
1850 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",
1851 to_copy, data_length - offset,
1852 to_read, data_read, offset);
1853 }
1854
1855 spin_lock_irq(&info->reassembly_queue_lock);
1856 info->reassembly_data_length -= data_read;
1857 info->reassembly_queue_length -= queue_removed;
1858 spin_unlock_irq(&info->reassembly_queue_lock);
1859
1860 info->first_entry_offset = offset;
1861 log_read(INFO, "returning to thread data_read=%d reassembly_data_length=%d first_entry_offset=%d\n",
1862 data_read, info->reassembly_data_length,
1863 info->first_entry_offset);
1864 read_rfc1002_done:
1865 return data_read;
1866 }
1867
1868 log_read(INFO, "wait_event on more data\n");
1869 rc = wait_event_interruptible(
1870 info->wait_reassembly_queue,
1871 info->reassembly_data_length >= size ||
1872 info->transport_status != SMBD_CONNECTED);
1873 /* Don't return any data if interrupted */
1874 if (rc)
1875 return rc;
1876
1877 if (info->transport_status != SMBD_CONNECTED) {
1878 log_read(ERR, "disconnected\n");
1879 return -ECONNABORTED;
1880 }
1881
1882 goto again;
1883 }
1884
1885 /*
1886 * Receive a page from receive reassembly queue
1887 * page: the page to read data into
1888 * to_read: the length of data to read
1889 * return value: actual data read
1890 */
1891 static int smbd_recv_page(struct smbd_connection *info,
1892 struct page *page, unsigned int page_offset,
1893 unsigned int to_read)
1894 {
1895 int ret;
1896 char *to_address;
1897 void *page_address;
1898
1899 /* make sure we have the page ready for read */
1900 ret = wait_event_interruptible(
1901 info->wait_reassembly_queue,
1902 info->reassembly_data_length >= to_read ||
1903 info->transport_status != SMBD_CONNECTED);
1904 if (ret)
1905 return ret;
1906
1907 /* now we can read from reassembly queue and not sleep */
1908 page_address = kmap_atomic(page);
1909 to_address = (char *) page_address + page_offset;
1910
1911 log_read(INFO, "reading from page=%p address=%p to_read=%d\n",
1912 page, to_address, to_read);
1913
1914 ret = smbd_recv_buf(info, to_address, to_read);
1915 kunmap_atomic(page_address);
1916
1917 return ret;
1918 }
1919
1920 /*
1921 * Receive data from transport
1922 * msg: a msghdr point to the buffer, can be ITER_KVEC or ITER_BVEC
1923 * return: total bytes read, or 0. SMB Direct will not do partial read.
1924 */
1925 int smbd_recv(struct smbd_connection *info, struct msghdr *msg)
1926 {
1927 char *buf;
1928 struct page *page;
1929 unsigned int to_read, page_offset;
1930 int rc;
1931
1932 if (iov_iter_rw(&msg->msg_iter) == WRITE) {
1933 /* It's a bug in upper layer to get there */
1934 cifs_dbg(VFS, "Invalid msg iter dir %u\n",
1935 iov_iter_rw(&msg->msg_iter));
1936 rc = -EINVAL;
1937 goto out;
1938 }
1939
1940 switch (iov_iter_type(&msg->msg_iter)) {
1941 case ITER_KVEC:
1942 buf = msg->msg_iter.kvec->iov_base;
1943 to_read = msg->msg_iter.kvec->iov_len;
1944 rc = smbd_recv_buf(info, buf, to_read);
1945 break;
1946
1947 case ITER_BVEC:
1948 page = msg->msg_iter.bvec->bv_page;
1949 page_offset = msg->msg_iter.bvec->bv_offset;
1950 to_read = msg->msg_iter.bvec->bv_len;
1951 rc = smbd_recv_page(info, page, page_offset, to_read);
1952 break;
1953
1954 default:
1955 /* It's a bug in upper layer to get there */
1956 cifs_dbg(VFS, "Invalid msg type %d\n",
1957 iov_iter_type(&msg->msg_iter));
1958 rc = -EINVAL;
1959 }
1960
1961 out:
1962 /* SMBDirect will read it all or nothing */
1963 if (rc > 0)
1964 msg->msg_iter.count = 0;
1965 return rc;
1966 }
1967
1968 /*
1969 * Send data to transport
1970 * Each rqst is transported as a SMBDirect payload
1971 * rqst: the data to write
1972 * return value: 0 if successfully write, otherwise error code
1973 */
1974 int smbd_send(struct TCP_Server_Info *server,
1975 int num_rqst, struct smb_rqst *rqst_array)
1976 {
1977 struct smbd_connection *info = server->smbd_conn;
1978 struct kvec vec;
1979 int nvecs;
1980 int size;
1981 unsigned int buflen, remaining_data_length;
1982 int start, i, j;
1983 int max_iov_size =
1984 info->max_send_size - sizeof(struct smbd_data_transfer);
1985 struct kvec *iov;
1986 int rc;
1987 struct smb_rqst *rqst;
1988 int rqst_idx;
1989
1990 if (info->transport_status != SMBD_CONNECTED) {
1991 rc = -EAGAIN;
1992 goto done;
1993 }
1994
1995 /*
1996 * Add in the page array if there is one. The caller needs to set
1997 * rq_tailsz to PAGE_SIZE when the buffer has multiple pages and
1998 * ends at page boundary
1999 */
2000 remaining_data_length = 0;
2001 for (i = 0; i < num_rqst; i++)
2002 remaining_data_length += smb_rqst_len(server, &rqst_array[i]);
2003
2004 if (remaining_data_length > info->max_fragmented_send_size) {
2005 log_write(ERR, "payload size %d > max size %d\n",
2006 remaining_data_length, info->max_fragmented_send_size);
2007 rc = -EINVAL;
2008 goto done;
2009 }
2010
2011 log_write(INFO, "num_rqst=%d total length=%u\n",
2012 num_rqst, remaining_data_length);
2013
2014 rqst_idx = 0;
2015 next_rqst:
2016 rqst = &rqst_array[rqst_idx];
2017 iov = rqst->rq_iov;
2018
2019 cifs_dbg(FYI, "Sending smb (RDMA): idx=%d smb_len=%lu\n",
2020 rqst_idx, smb_rqst_len(server, rqst));
2021 for (i = 0; i < rqst->rq_nvec; i++)
2022 dump_smb(iov[i].iov_base, iov[i].iov_len);
2023
2024
2025 log_write(INFO, "rqst_idx=%d nvec=%d rqst->rq_npages=%d rq_pagesz=%d rq_tailsz=%d buflen=%lu\n",
2026 rqst_idx, rqst->rq_nvec, rqst->rq_npages, rqst->rq_pagesz,
2027 rqst->rq_tailsz, smb_rqst_len(server, rqst));
2028
2029 start = i = 0;
2030 buflen = 0;
2031 while (true) {
2032 buflen += iov[i].iov_len;
2033 if (buflen > max_iov_size) {
2034 if (i > start) {
2035 remaining_data_length -=
2036 (buflen-iov[i].iov_len);
2037 log_write(INFO, "sending iov[] from start=%d i=%d nvecs=%d remaining_data_length=%d\n",
2038 start, i, i - start,
2039 remaining_data_length);
2040 rc = smbd_post_send_data(
2041 info, &iov[start], i-start,
2042 remaining_data_length);
2043 if (rc)
2044 goto done;
2045 } else {
2046 /* iov[start] is too big, break it */
2047 nvecs = (buflen+max_iov_size-1)/max_iov_size;
2048 log_write(INFO, "iov[%d] iov_base=%p buflen=%d break to %d vectors\n",
2049 start, iov[start].iov_base,
2050 buflen, nvecs);
2051 for (j = 0; j < nvecs; j++) {
2052 vec.iov_base =
2053 (char *)iov[start].iov_base +
2054 j*max_iov_size;
2055 vec.iov_len = max_iov_size;
2056 if (j == nvecs-1)
2057 vec.iov_len =
2058 buflen -
2059 max_iov_size*(nvecs-1);
2060 remaining_data_length -= vec.iov_len;
2061 log_write(INFO,
2062 "sending vec j=%d iov_base=%p iov_len=%zu remaining_data_length=%d\n",
2063 j, vec.iov_base, vec.iov_len,
2064 remaining_data_length);
2065 rc = smbd_post_send_data(
2066 info, &vec, 1,
2067 remaining_data_length);
2068 if (rc)
2069 goto done;
2070 }
2071 i++;
2072 if (i == rqst->rq_nvec)
2073 break;
2074 }
2075 start = i;
2076 buflen = 0;
2077 } else {
2078 i++;
2079 if (i == rqst->rq_nvec) {
2080 /* send out all remaining vecs */
2081 remaining_data_length -= buflen;
2082 log_write(INFO, "sending iov[] from start=%d i=%d nvecs=%d remaining_data_length=%d\n",
2083 start, i, i - start,
2084 remaining_data_length);
2085 rc = smbd_post_send_data(info, &iov[start],
2086 i-start, remaining_data_length);
2087 if (rc)
2088 goto done;
2089 break;
2090 }
2091 }
2092 log_write(INFO, "looping i=%d buflen=%d\n", i, buflen);
2093 }
2094
2095 /* now sending pages if there are any */
2096 for (i = 0; i < rqst->rq_npages; i++) {
2097 unsigned int offset;
2098
2099 rqst_page_get_length(rqst, i, &buflen, &offset);
2100 nvecs = (buflen + max_iov_size - 1) / max_iov_size;
2101 log_write(INFO, "sending pages buflen=%d nvecs=%d\n",
2102 buflen, nvecs);
2103 for (j = 0; j < nvecs; j++) {
2104 size = max_iov_size;
2105 if (j == nvecs-1)
2106 size = buflen - j*max_iov_size;
2107 remaining_data_length -= size;
2108 log_write(INFO, "sending pages i=%d offset=%d size=%d remaining_data_length=%d\n",
2109 i, j * max_iov_size + offset, size,
2110 remaining_data_length);
2111 rc = smbd_post_send_page(
2112 info, rqst->rq_pages[i],
2113 j*max_iov_size + offset,
2114 size, remaining_data_length);
2115 if (rc)
2116 goto done;
2117 }
2118 }
2119
2120 rqst_idx++;
2121 if (rqst_idx < num_rqst)
2122 goto next_rqst;
2123
2124 done:
2125 /*
2126 * As an optimization, we don't wait for individual I/O to finish
2127 * before sending the next one.
2128 * Send them all and wait for pending send count to get to 0
2129 * that means all the I/Os have been out and we are good to return
2130 */
2131
2132 wait_event(info->wait_send_pending,
2133 atomic_read(&info->send_pending) == 0);
2134
2135 return rc;
2136 }
2137
2138 static void register_mr_done(struct ib_cq *cq, struct ib_wc *wc)
2139 {
2140 struct smbd_mr *mr;
2141 struct ib_cqe *cqe;
2142
2143 if (wc->status) {
2144 log_rdma_mr(ERR, "status=%d\n", wc->status);
2145 cqe = wc->wr_cqe;
2146 mr = container_of(cqe, struct smbd_mr, cqe);
2147 smbd_disconnect_rdma_connection(mr->conn);
2148 }
2149 }
2150
2151 /*
2152 * The work queue function that recovers MRs
2153 * We need to call ib_dereg_mr() and ib_alloc_mr() before this MR can be used
2154 * again. Both calls are slow, so finish them in a workqueue. This will not
2155 * block I/O path.
2156 * There is one workqueue that recovers MRs, there is no need to lock as the
2157 * I/O requests calling smbd_register_mr will never update the links in the
2158 * mr_list.
2159 */
2160 static void smbd_mr_recovery_work(struct work_struct *work)
2161 {
2162 struct smbd_connection *info =
2163 container_of(work, struct smbd_connection, mr_recovery_work);
2164 struct smbd_mr *smbdirect_mr;
2165 int rc;
2166
2167 list_for_each_entry(smbdirect_mr, &info->mr_list, list) {
2168 if (smbdirect_mr->state == MR_ERROR) {
2169
2170 /* recover this MR entry */
2171 rc = ib_dereg_mr(smbdirect_mr->mr);
2172 if (rc) {
2173 log_rdma_mr(ERR,
2174 "ib_dereg_mr failed rc=%x\n",
2175 rc);
2176 smbd_disconnect_rdma_connection(info);
2177 continue;
2178 }
2179
2180 smbdirect_mr->mr = ib_alloc_mr(
2181 info->pd, info->mr_type,
2182 info->max_frmr_depth);
2183 if (IS_ERR(smbdirect_mr->mr)) {
2184 log_rdma_mr(ERR, "ib_alloc_mr failed mr_type=%x max_frmr_depth=%x\n",
2185 info->mr_type,
2186 info->max_frmr_depth);
2187 smbd_disconnect_rdma_connection(info);
2188 continue;
2189 }
2190 } else
2191 /* This MR is being used, don't recover it */
2192 continue;
2193
2194 smbdirect_mr->state = MR_READY;
2195
2196 /* smbdirect_mr->state is updated by this function
2197 * and is read and updated by I/O issuing CPUs trying
2198 * to get a MR, the call to atomic_inc_return
2199 * implicates a memory barrier and guarantees this
2200 * value is updated before waking up any calls to
2201 * get_mr() from the I/O issuing CPUs
2202 */
2203 if (atomic_inc_return(&info->mr_ready_count) == 1)
2204 wake_up_interruptible(&info->wait_mr);
2205 }
2206 }
2207
2208 static void destroy_mr_list(struct smbd_connection *info)
2209 {
2210 struct smbd_mr *mr, *tmp;
2211
2212 cancel_work_sync(&info->mr_recovery_work);
2213 list_for_each_entry_safe(mr, tmp, &info->mr_list, list) {
2214 if (mr->state == MR_INVALIDATED)
2215 ib_dma_unmap_sg(info->id->device, mr->sgl,
2216 mr->sgl_count, mr->dir);
2217 ib_dereg_mr(mr->mr);
2218 kfree(mr->sgl);
2219 kfree(mr);
2220 }
2221 }
2222
2223 /*
2224 * Allocate MRs used for RDMA read/write
2225 * The number of MRs will not exceed hardware capability in responder_resources
2226 * All MRs are kept in mr_list. The MR can be recovered after it's used
2227 * Recovery is done in smbd_mr_recovery_work. The content of list entry changes
2228 * as MRs are used and recovered for I/O, but the list links will not change
2229 */
2230 static int allocate_mr_list(struct smbd_connection *info)
2231 {
2232 int i;
2233 struct smbd_mr *smbdirect_mr, *tmp;
2234
2235 INIT_LIST_HEAD(&info->mr_list);
2236 init_waitqueue_head(&info->wait_mr);
2237 spin_lock_init(&info->mr_list_lock);
2238 atomic_set(&info->mr_ready_count, 0);
2239 atomic_set(&info->mr_used_count, 0);
2240 init_waitqueue_head(&info->wait_for_mr_cleanup);
2241 /* Allocate more MRs (2x) than hardware responder_resources */
2242 for (i = 0; i < info->responder_resources * 2; i++) {
2243 smbdirect_mr = kzalloc(sizeof(*smbdirect_mr), GFP_KERNEL);
2244 if (!smbdirect_mr)
2245 goto out;
2246 smbdirect_mr->mr = ib_alloc_mr(info->pd, info->mr_type,
2247 info->max_frmr_depth);
2248 if (IS_ERR(smbdirect_mr->mr)) {
2249 log_rdma_mr(ERR, "ib_alloc_mr failed mr_type=%x max_frmr_depth=%x\n",
2250 info->mr_type, info->max_frmr_depth);
2251 goto out;
2252 }
2253 smbdirect_mr->sgl = kcalloc(
2254 info->max_frmr_depth,
2255 sizeof(struct scatterlist),
2256 GFP_KERNEL);
2257 if (!smbdirect_mr->sgl) {
2258 log_rdma_mr(ERR, "failed to allocate sgl\n");
2259 ib_dereg_mr(smbdirect_mr->mr);
2260 goto out;
2261 }
2262 smbdirect_mr->state = MR_READY;
2263 smbdirect_mr->conn = info;
2264
2265 list_add_tail(&smbdirect_mr->list, &info->mr_list);
2266 atomic_inc(&info->mr_ready_count);
2267 }
2268 INIT_WORK(&info->mr_recovery_work, smbd_mr_recovery_work);
2269 return 0;
2270
2271 out:
2272 kfree(smbdirect_mr);
2273
2274 list_for_each_entry_safe(smbdirect_mr, tmp, &info->mr_list, list) {
2275 ib_dereg_mr(smbdirect_mr->mr);
2276 kfree(smbdirect_mr->sgl);
2277 kfree(smbdirect_mr);
2278 }
2279 return -ENOMEM;
2280 }
2281
2282 /*
2283 * Get a MR from mr_list. This function waits until there is at least one
2284 * MR available in the list. It may access the list while the
2285 * smbd_mr_recovery_work is recovering the MR list. This doesn't need a lock
2286 * as they never modify the same places. However, there may be several CPUs
2287 * issueing I/O trying to get MR at the same time, mr_list_lock is used to
2288 * protect this situation.
2289 */
2290 static struct smbd_mr *get_mr(struct smbd_connection *info)
2291 {
2292 struct smbd_mr *ret;
2293 int rc;
2294 again:
2295 rc = wait_event_interruptible(info->wait_mr,
2296 atomic_read(&info->mr_ready_count) ||
2297 info->transport_status != SMBD_CONNECTED);
2298 if (rc) {
2299 log_rdma_mr(ERR, "wait_event_interruptible rc=%x\n", rc);
2300 return NULL;
2301 }
2302
2303 if (info->transport_status != SMBD_CONNECTED) {
2304 log_rdma_mr(ERR, "info->transport_status=%x\n",
2305 info->transport_status);
2306 return NULL;
2307 }
2308
2309 spin_lock(&info->mr_list_lock);
2310 list_for_each_entry(ret, &info->mr_list, list) {
2311 if (ret->state == MR_READY) {
2312 ret->state = MR_REGISTERED;
2313 spin_unlock(&info->mr_list_lock);
2314 atomic_dec(&info->mr_ready_count);
2315 atomic_inc(&info->mr_used_count);
2316 return ret;
2317 }
2318 }
2319
2320 spin_unlock(&info->mr_list_lock);
2321 /*
2322 * It is possible that we could fail to get MR because other processes may
2323 * try to acquire a MR at the same time. If this is the case, retry it.
2324 */
2325 goto again;
2326 }
2327
2328 /*
2329 * Register memory for RDMA read/write
2330 * pages[]: the list of pages to register memory with
2331 * num_pages: the number of pages to register
2332 * tailsz: if non-zero, the bytes to register in the last page
2333 * writing: true if this is a RDMA write (SMB read), false for RDMA read
2334 * need_invalidate: true if this MR needs to be locally invalidated after I/O
2335 * return value: the MR registered, NULL if failed.
2336 */
2337 struct smbd_mr *smbd_register_mr(
2338 struct smbd_connection *info, struct page *pages[], int num_pages,
2339 int offset, int tailsz, bool writing, bool need_invalidate)
2340 {
2341 struct smbd_mr *smbdirect_mr;
2342 int rc, i;
2343 enum dma_data_direction dir;
2344 struct ib_reg_wr *reg_wr;
2345
2346 if (num_pages > info->max_frmr_depth) {
2347 log_rdma_mr(ERR, "num_pages=%d max_frmr_depth=%d\n",
2348 num_pages, info->max_frmr_depth);
2349 return NULL;
2350 }
2351
2352 smbdirect_mr = get_mr(info);
2353 if (!smbdirect_mr) {
2354 log_rdma_mr(ERR, "get_mr returning NULL\n");
2355 return NULL;
2356 }
2357 smbdirect_mr->need_invalidate = need_invalidate;
2358 smbdirect_mr->sgl_count = num_pages;
2359 sg_init_table(smbdirect_mr->sgl, num_pages);
2360
2361 log_rdma_mr(INFO, "num_pages=0x%x offset=0x%x tailsz=0x%x\n",
2362 num_pages, offset, tailsz);
2363
2364 if (num_pages == 1) {
2365 sg_set_page(&smbdirect_mr->sgl[0], pages[0], tailsz, offset);
2366 goto skip_multiple_pages;
2367 }
2368
2369 /* We have at least two pages to register */
2370 sg_set_page(
2371 &smbdirect_mr->sgl[0], pages[0], PAGE_SIZE - offset, offset);
2372 i = 1;
2373 while (i < num_pages - 1) {
2374 sg_set_page(&smbdirect_mr->sgl[i], pages[i], PAGE_SIZE, 0);
2375 i++;
2376 }
2377 sg_set_page(&smbdirect_mr->sgl[i], pages[i],
2378 tailsz ? tailsz : PAGE_SIZE, 0);
2379
2380 skip_multiple_pages:
2381 dir = writing ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
2382 smbdirect_mr->dir = dir;
2383 rc = ib_dma_map_sg(info->id->device, smbdirect_mr->sgl, num_pages, dir);
2384 if (!rc) {
2385 log_rdma_mr(ERR, "ib_dma_map_sg num_pages=%x dir=%x rc=%x\n",
2386 num_pages, dir, rc);
2387 goto dma_map_error;
2388 }
2389
2390 rc = ib_map_mr_sg(smbdirect_mr->mr, smbdirect_mr->sgl, num_pages,
2391 NULL, PAGE_SIZE);
2392 if (rc != num_pages) {
2393 log_rdma_mr(ERR,
2394 "ib_map_mr_sg failed rc = %d num_pages = %x\n",
2395 rc, num_pages);
2396 goto map_mr_error;
2397 }
2398
2399 ib_update_fast_reg_key(smbdirect_mr->mr,
2400 ib_inc_rkey(smbdirect_mr->mr->rkey));
2401 reg_wr = &smbdirect_mr->wr;
2402 reg_wr->wr.opcode = IB_WR_REG_MR;
2403 smbdirect_mr->cqe.done = register_mr_done;
2404 reg_wr->wr.wr_cqe = &smbdirect_mr->cqe;
2405 reg_wr->wr.num_sge = 0;
2406 reg_wr->wr.send_flags = IB_SEND_SIGNALED;
2407 reg_wr->mr = smbdirect_mr->mr;
2408 reg_wr->key = smbdirect_mr->mr->rkey;
2409 reg_wr->access = writing ?
2410 IB_ACCESS_REMOTE_WRITE | IB_ACCESS_LOCAL_WRITE :
2411 IB_ACCESS_REMOTE_READ;
2412
2413 /*
2414 * There is no need for waiting for complemtion on ib_post_send
2415 * on IB_WR_REG_MR. Hardware enforces a barrier and order of execution
2416 * on the next ib_post_send when we actaully send I/O to remote peer
2417 */
2418 rc = ib_post_send(info->id->qp, &reg_wr->wr, NULL);
2419 if (!rc)
2420 return smbdirect_mr;
2421
2422 log_rdma_mr(ERR, "ib_post_send failed rc=%x reg_wr->key=%x\n",
2423 rc, reg_wr->key);
2424
2425 /* If all failed, attempt to recover this MR by setting it MR_ERROR*/
2426 map_mr_error:
2427 ib_dma_unmap_sg(info->id->device, smbdirect_mr->sgl,
2428 smbdirect_mr->sgl_count, smbdirect_mr->dir);
2429
2430 dma_map_error:
2431 smbdirect_mr->state = MR_ERROR;
2432 if (atomic_dec_and_test(&info->mr_used_count))
2433 wake_up(&info->wait_for_mr_cleanup);
2434
2435 smbd_disconnect_rdma_connection(info);
2436
2437 return NULL;
2438 }
2439
2440 static void local_inv_done(struct ib_cq *cq, struct ib_wc *wc)
2441 {
2442 struct smbd_mr *smbdirect_mr;
2443 struct ib_cqe *cqe;
2444
2445 cqe = wc->wr_cqe;
2446 smbdirect_mr = container_of(cqe, struct smbd_mr, cqe);
2447 smbdirect_mr->state = MR_INVALIDATED;
2448 if (wc->status != IB_WC_SUCCESS) {
2449 log_rdma_mr(ERR, "invalidate failed status=%x\n", wc->status);
2450 smbdirect_mr->state = MR_ERROR;
2451 }
2452 complete(&smbdirect_mr->invalidate_done);
2453 }
2454
2455 /*
2456 * Deregister a MR after I/O is done
2457 * This function may wait if remote invalidation is not used
2458 * and we have to locally invalidate the buffer to prevent data is being
2459 * modified by remote peer after upper layer consumes it
2460 */
2461 int smbd_deregister_mr(struct smbd_mr *smbdirect_mr)
2462 {
2463 struct ib_send_wr *wr;
2464 struct smbd_connection *info = smbdirect_mr->conn;
2465 int rc = 0;
2466
2467 if (smbdirect_mr->need_invalidate) {
2468 /* Need to finish local invalidation before returning */
2469 wr = &smbdirect_mr->inv_wr;
2470 wr->opcode = IB_WR_LOCAL_INV;
2471 smbdirect_mr->cqe.done = local_inv_done;
2472 wr->wr_cqe = &smbdirect_mr->cqe;
2473 wr->num_sge = 0;
2474 wr->ex.invalidate_rkey = smbdirect_mr->mr->rkey;
2475 wr->send_flags = IB_SEND_SIGNALED;
2476
2477 init_completion(&smbdirect_mr->invalidate_done);
2478 rc = ib_post_send(info->id->qp, wr, NULL);
2479 if (rc) {
2480 log_rdma_mr(ERR, "ib_post_send failed rc=%x\n", rc);
2481 smbd_disconnect_rdma_connection(info);
2482 goto done;
2483 }
2484 wait_for_completion(&smbdirect_mr->invalidate_done);
2485 smbdirect_mr->need_invalidate = false;
2486 } else
2487 /*
2488 * For remote invalidation, just set it to MR_INVALIDATED
2489 * and defer to mr_recovery_work to recover the MR for next use
2490 */
2491 smbdirect_mr->state = MR_INVALIDATED;
2492
2493 if (smbdirect_mr->state == MR_INVALIDATED) {
2494 ib_dma_unmap_sg(
2495 info->id->device, smbdirect_mr->sgl,
2496 smbdirect_mr->sgl_count,
2497 smbdirect_mr->dir);
2498 smbdirect_mr->state = MR_READY;
2499 if (atomic_inc_return(&info->mr_ready_count) == 1)
2500 wake_up_interruptible(&info->wait_mr);
2501 } else
2502 /*
2503 * Schedule the work to do MR recovery for future I/Os MR
2504 * recovery is slow and don't want it to block current I/O
2505 */
2506 queue_work(info->workqueue, &info->mr_recovery_work);
2507
2508 done:
2509 if (atomic_dec_and_test(&info->mr_used_count))
2510 wake_up(&info->wait_for_mr_cleanup);
2511
2512 return rc;
2513 }
Linux with added FPC-III support