fs/cifs/smbdirect.c

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