drivers/hv/channel.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2009, Microsoft Corporation.
   4  *
   5  * Authors:
   6  *   Haiyang Zhang <haiyangz@microsoft.com>
   7  *   Hank Janssen  <hjanssen@microsoft.com>
   8  */
   9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  10
  11 #include <linux/kernel.h>
  12 #include <linux/sched.h>
  13 #include <linux/wait.h>
  14 #include <linux/mm.h>
  15 #include <linux/slab.h>
  16 #include <linux/module.h>
  17 #include <linux/hyperv.h>
  18 #include <linux/uio.h>
  19 #include <linux/interrupt.h>
  20 #include <asm/page.h>
  21 #include <asm/mshyperv.h>
  22
  23 #include "hyperv_vmbus.h"
  24
  25 #define NUM_PAGES_SPANNED(addr, len) \
  26 ((PAGE_ALIGN(addr + len) >> PAGE_SHIFT) - (addr >> PAGE_SHIFT))
  27
  28 static unsigned long virt_to_hvpfn(void *addr)
  29 {
  30         phys_addr_t paddr;
  31
  32         if (is_vmalloc_addr(addr))
  33                 paddr = page_to_phys(vmalloc_to_page(addr)) +
  34                                          offset_in_page(addr);
  35         else
  36                 paddr = __pa(addr);
  37
  38         return  paddr >> PAGE_SHIFT;
  39 }
  40
  41 /*
  42  * vmbus_setevent- Trigger an event notification on the specified
  43  * channel.
  44  */
  45 void vmbus_setevent(struct vmbus_channel *channel)
  46 {
  47         struct hv_monitor_page *monitorpage;
  48
  49         trace_vmbus_setevent(channel);
  50
  51         /*
  52          * For channels marked as in "low latency" mode
  53          * bypass the monitor page mechanism.
  54          */
  55         if (channel->offermsg.monitor_allocated && !channel->low_latency) {
  56                 vmbus_send_interrupt(channel->offermsg.child_relid);
  57
  58                 /* Get the child to parent monitor page */
  59                 monitorpage = vmbus_connection.monitor_pages[1];
  60
  61                 sync_set_bit(channel->monitor_bit,
  62                         (unsigned long *)&monitorpage->trigger_group
  63                                         [channel->monitor_grp].pending);
  64
  65         } else {
  66                 vmbus_set_event(channel);
  67         }
  68 }
  69 EXPORT_SYMBOL_GPL(vmbus_setevent);
  70
  71 /* vmbus_free_ring - drop mapping of ring buffer */
  72 void vmbus_free_ring(struct vmbus_channel *channel)
  73 {
  74         hv_ringbuffer_cleanup(&channel->outbound);
  75         hv_ringbuffer_cleanup(&channel->inbound);
  76
  77         if (channel->ringbuffer_page) {
  78                 __free_pages(channel->ringbuffer_page,
  79                              get_order(channel->ringbuffer_pagecount
  80                                        << PAGE_SHIFT));
  81                 channel->ringbuffer_page = NULL;
  82         }
  83 }
  84 EXPORT_SYMBOL_GPL(vmbus_free_ring);
  85
  86 /* vmbus_alloc_ring - allocate and map pages for ring buffer */
  87 int vmbus_alloc_ring(struct vmbus_channel *newchannel,
  88                      u32 send_size, u32 recv_size)
  89 {
  90         struct page *page;
  91         int order;
  92
  93         if (send_size % PAGE_SIZE || recv_size % PAGE_SIZE)
  94                 return -EINVAL;
  95
  96         /* Allocate the ring buffer */
  97         order = get_order(send_size + recv_size);
  98         page = alloc_pages_node(cpu_to_node(newchannel->target_cpu),
  99                                 GFP_KERNEL|__GFP_ZERO, order);
 100
 101         if (!page)
 102                 page = alloc_pages(GFP_KERNEL|__GFP_ZERO, order);
 103
 104         if (!page)
 105                 return -ENOMEM;
 106
 107         newchannel->ringbuffer_page = page;
 108         newchannel->ringbuffer_pagecount = (send_size + recv_size) >> PAGE_SHIFT;
 109         newchannel->ringbuffer_send_offset = send_size >> PAGE_SHIFT;
 110
 111         return 0;
 112 }
 113 EXPORT_SYMBOL_GPL(vmbus_alloc_ring);
 114
 115 static int __vmbus_open(struct vmbus_channel *newchannel,
 116                        void *userdata, u32 userdatalen,
 117                        void (*onchannelcallback)(void *context), void *context)
 118 {
 119         struct vmbus_channel_open_channel *open_msg;
 120         struct vmbus_channel_msginfo *open_info = NULL;
 121         struct page *page = newchannel->ringbuffer_page;
 122         u32 send_pages, recv_pages;
 123         unsigned long flags;
 124         int err;
 125
 126         if (userdatalen > MAX_USER_DEFINED_BYTES)
 127                 return -EINVAL;
 128
 129         send_pages = newchannel->ringbuffer_send_offset;
 130         recv_pages = newchannel->ringbuffer_pagecount - send_pages;
 131
 132         if (newchannel->state != CHANNEL_OPEN_STATE)
 133                 return -EINVAL;
 134
 135         newchannel->state = CHANNEL_OPENING_STATE;
 136         newchannel->onchannel_callback = onchannelcallback;
 137         newchannel->channel_callback_context = context;
 138
 139         err = hv_ringbuffer_init(&newchannel->outbound, page, send_pages);
 140         if (err)
 141                 goto error_clean_ring;
 142
 143         err = hv_ringbuffer_init(&newchannel->inbound,
 144                                  &page[send_pages], recv_pages);
 145         if (err)
 146                 goto error_clean_ring;
 147
 148         /* Establish the gpadl for the ring buffer */
 149         newchannel->ringbuffer_gpadlhandle = 0;
 150
 151         err = vmbus_establish_gpadl(newchannel,
 152                                     page_address(newchannel->ringbuffer_page),
 153                                     (send_pages + recv_pages) << PAGE_SHIFT,
 154                                     &newchannel->ringbuffer_gpadlhandle);
 155         if (err)
 156                 goto error_clean_ring;
 157
 158         /* Create and init the channel open message */
 159         open_info = kmalloc(sizeof(*open_info) +
 160                            sizeof(struct vmbus_channel_open_channel),
 161                            GFP_KERNEL);
 162         if (!open_info) {
 163                 err = -ENOMEM;
 164                 goto error_free_gpadl;
 165         }
 166
 167         init_completion(&open_info->waitevent);
 168         open_info->waiting_channel = newchannel;
 169
 170         open_msg = (struct vmbus_channel_open_channel *)open_info->msg;
 171         open_msg->header.msgtype = CHANNELMSG_OPENCHANNEL;
 172         open_msg->openid = newchannel->offermsg.child_relid;
 173         open_msg->child_relid = newchannel->offermsg.child_relid;
 174         open_msg->ringbuffer_gpadlhandle = newchannel->ringbuffer_gpadlhandle;
 175         open_msg->downstream_ringbuffer_pageoffset = newchannel->ringbuffer_send_offset;
 176         open_msg->target_vp = hv_cpu_number_to_vp_number(newchannel->target_cpu);
 177
 178         if (userdatalen)
 179                 memcpy(open_msg->userdata, userdata, userdatalen);
 180
 181         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 182         list_add_tail(&open_info->msglistentry,
 183                       &vmbus_connection.chn_msg_list);
 184         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 185
 186         if (newchannel->rescind) {
 187                 err = -ENODEV;
 188                 goto error_free_info;
 189         }
 190
 191         err = vmbus_post_msg(open_msg,
 192                              sizeof(struct vmbus_channel_open_channel), true);
 193
 194         trace_vmbus_open(open_msg, err);
 195
 196         if (err != 0)
 197                 goto error_clean_msglist;
 198
 199         wait_for_completion(&open_info->waitevent);
 200
 201         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 202         list_del(&open_info->msglistentry);
 203         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 204
 205         if (newchannel->rescind) {
 206                 err = -ENODEV;
 207                 goto error_free_info;
 208         }
 209
 210         if (open_info->response.open_result.status) {
 211                 err = -EAGAIN;
 212                 goto error_free_info;
 213         }
 214
 215         newchannel->state = CHANNEL_OPENED_STATE;
 216         kfree(open_info);
 217         return 0;
 218
 219 error_clean_msglist:
 220         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 221         list_del(&open_info->msglistentry);
 222         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 223 error_free_info:
 224         kfree(open_info);
 225 error_free_gpadl:
 226         vmbus_teardown_gpadl(newchannel, newchannel->ringbuffer_gpadlhandle);
 227         newchannel->ringbuffer_gpadlhandle = 0;
 228 error_clean_ring:
 229         hv_ringbuffer_cleanup(&newchannel->outbound);
 230         hv_ringbuffer_cleanup(&newchannel->inbound);
 231         newchannel->state = CHANNEL_OPEN_STATE;
 232         return err;
 233 }
 234
 235 /*
 236  * vmbus_connect_ring - Open the channel but reuse ring buffer
 237  */
 238 int vmbus_connect_ring(struct vmbus_channel *newchannel,
 239                        void (*onchannelcallback)(void *context), void *context)
 240 {
 241         return  __vmbus_open(newchannel, NULL, 0, onchannelcallback, context);
 242 }
 243 EXPORT_SYMBOL_GPL(vmbus_connect_ring);
 244
 245 /*
 246  * vmbus_open - Open the specified channel.
 247  */
 248 int vmbus_open(struct vmbus_channel *newchannel,
 249                u32 send_ringbuffer_size, u32 recv_ringbuffer_size,
 250                void *userdata, u32 userdatalen,
 251                void (*onchannelcallback)(void *context), void *context)
 252 {
 253         int err;
 254
 255         err = vmbus_alloc_ring(newchannel, send_ringbuffer_size,
 256                                recv_ringbuffer_size);
 257         if (err)
 258                 return err;
 259
 260         err = __vmbus_open(newchannel, userdata, userdatalen,
 261                            onchannelcallback, context);
 262         if (err)
 263                 vmbus_free_ring(newchannel);
 264
 265         return err;
 266 }
 267 EXPORT_SYMBOL_GPL(vmbus_open);
 268
 269 /* Used for Hyper-V Socket: a guest client's connect() to the host */
 270 int vmbus_send_tl_connect_request(const guid_t *shv_guest_servie_id,
 271                                   const guid_t *shv_host_servie_id)
 272 {
 273         struct vmbus_channel_tl_connect_request conn_msg;
 274         int ret;
 275
 276         memset(&conn_msg, 0, sizeof(conn_msg));
 277         conn_msg.header.msgtype = CHANNELMSG_TL_CONNECT_REQUEST;
 278         conn_msg.guest_endpoint_id = *shv_guest_servie_id;
 279         conn_msg.host_service_id = *shv_host_servie_id;
 280
 281         ret = vmbus_post_msg(&conn_msg, sizeof(conn_msg), true);
 282
 283         trace_vmbus_send_tl_connect_request(&conn_msg, ret);
 284
 285         return ret;
 286 }
 287 EXPORT_SYMBOL_GPL(vmbus_send_tl_connect_request);
 288
 289 /*
 290  * Set/change the vCPU (@target_vp) the channel (@child_relid) will interrupt.
 291  *
 292  * CHANNELMSG_MODIFYCHANNEL messages are aynchronous.  Also, Hyper-V does not
 293  * ACK such messages.  IOW we can't know when the host will stop interrupting
 294  * the "old" vCPU and start interrupting the "new" vCPU for the given channel.
 295  *
 296  * The CHANNELMSG_MODIFYCHANNEL message type is supported since VMBus version
 297  * VERSION_WIN10_V4_1.
 298  */
 299 int vmbus_send_modifychannel(u32 child_relid, u32 target_vp)
 300 {
 301         struct vmbus_channel_modifychannel conn_msg;
 302         int ret;
 303
 304         memset(&conn_msg, 0, sizeof(conn_msg));
 305         conn_msg.header.msgtype = CHANNELMSG_MODIFYCHANNEL;
 306         conn_msg.child_relid = child_relid;
 307         conn_msg.target_vp = target_vp;
 308
 309         ret = vmbus_post_msg(&conn_msg, sizeof(conn_msg), true);
 310
 311         trace_vmbus_send_modifychannel(&conn_msg, ret);
 312
 313         return ret;
 314 }
 315 EXPORT_SYMBOL_GPL(vmbus_send_modifychannel);
 316
 317 /*
 318  * create_gpadl_header - Creates a gpadl for the specified buffer
 319  */
 320 static int create_gpadl_header(void *kbuffer, u32 size,
 321                                struct vmbus_channel_msginfo **msginfo)
 322 {
 323         int i;
 324         int pagecount;
 325         struct vmbus_channel_gpadl_header *gpadl_header;
 326         struct vmbus_channel_gpadl_body *gpadl_body;
 327         struct vmbus_channel_msginfo *msgheader;
 328         struct vmbus_channel_msginfo *msgbody = NULL;
 329         u32 msgsize;
 330
 331         int pfnsum, pfncount, pfnleft, pfncurr, pfnsize;
 332
 333         pagecount = size >> PAGE_SHIFT;
 334
 335         /* do we need a gpadl body msg */
 336         pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
 337                   sizeof(struct vmbus_channel_gpadl_header) -
 338                   sizeof(struct gpa_range);
 339         pfncount = pfnsize / sizeof(u64);
 340
 341         if (pagecount > pfncount) {
 342                 /* we need a gpadl body */
 343                 /* fill in the header */
 344                 msgsize = sizeof(struct vmbus_channel_msginfo) +
 345                           sizeof(struct vmbus_channel_gpadl_header) +
 346                           sizeof(struct gpa_range) + pfncount * sizeof(u64);
 347                 msgheader =  kzalloc(msgsize, GFP_KERNEL);
 348                 if (!msgheader)
 349                         goto nomem;
 350
 351                 INIT_LIST_HEAD(&msgheader->submsglist);
 352                 msgheader->msgsize = msgsize;
 353
 354                 gpadl_header = (struct vmbus_channel_gpadl_header *)
 355                         msgheader->msg;
 356                 gpadl_header->rangecount = 1;
 357                 gpadl_header->range_buflen = sizeof(struct gpa_range) +
 358                                          pagecount * sizeof(u64);
 359                 gpadl_header->range[0].byte_offset = 0;
 360                 gpadl_header->range[0].byte_count = size;
 361                 for (i = 0; i < pfncount; i++)
 362                         gpadl_header->range[0].pfn_array[i] = virt_to_hvpfn(
 363                                 kbuffer + PAGE_SIZE * i);
 364                 *msginfo = msgheader;
 365
 366                 pfnsum = pfncount;
 367                 pfnleft = pagecount - pfncount;
 368
 369                 /* how many pfns can we fit */
 370                 pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
 371                           sizeof(struct vmbus_channel_gpadl_body);
 372                 pfncount = pfnsize / sizeof(u64);
 373
 374                 /* fill in the body */
 375                 while (pfnleft) {
 376                         if (pfnleft > pfncount)
 377                                 pfncurr = pfncount;
 378                         else
 379                                 pfncurr = pfnleft;
 380
 381                         msgsize = sizeof(struct vmbus_channel_msginfo) +
 382                                   sizeof(struct vmbus_channel_gpadl_body) +
 383                                   pfncurr * sizeof(u64);
 384                         msgbody = kzalloc(msgsize, GFP_KERNEL);
 385
 386                         if (!msgbody) {
 387                                 struct vmbus_channel_msginfo *pos = NULL;
 388                                 struct vmbus_channel_msginfo *tmp = NULL;
 389                                 /*
 390                                  * Free up all the allocated messages.
 391                                  */
 392                                 list_for_each_entry_safe(pos, tmp,
 393                                         &msgheader->submsglist,
 394                                         msglistentry) {
 395
 396                                         list_del(&pos->msglistentry);
 397                                         kfree(pos);
 398                                 }
 399
 400                                 goto nomem;
 401                         }
 402
 403                         msgbody->msgsize = msgsize;
 404                         gpadl_body =
 405                                 (struct vmbus_channel_gpadl_body *)msgbody->msg;
 406
 407                         /*
 408                          * Gpadl is u32 and we are using a pointer which could
 409                          * be 64-bit
 410                          * This is governed by the guest/host protocol and
 411                          * so the hypervisor guarantees that this is ok.
 412                          */
 413                         for (i = 0; i < pfncurr; i++)
 414                                 gpadl_body->pfn[i] = virt_to_hvpfn(
 415                                         kbuffer + PAGE_SIZE * (pfnsum + i));
 416
 417                         /* add to msg header */
 418                         list_add_tail(&msgbody->msglistentry,
 419                                       &msgheader->submsglist);
 420                         pfnsum += pfncurr;
 421                         pfnleft -= pfncurr;
 422                 }
 423         } else {
 424                 /* everything fits in a header */
 425                 msgsize = sizeof(struct vmbus_channel_msginfo) +
 426                           sizeof(struct vmbus_channel_gpadl_header) +
 427                           sizeof(struct gpa_range) + pagecount * sizeof(u64);
 428                 msgheader = kzalloc(msgsize, GFP_KERNEL);
 429                 if (msgheader == NULL)
 430                         goto nomem;
 431
 432                 INIT_LIST_HEAD(&msgheader->submsglist);
 433                 msgheader->msgsize = msgsize;
 434
 435                 gpadl_header = (struct vmbus_channel_gpadl_header *)
 436                         msgheader->msg;
 437                 gpadl_header->rangecount = 1;
 438                 gpadl_header->range_buflen = sizeof(struct gpa_range) +
 439                                          pagecount * sizeof(u64);
 440                 gpadl_header->range[0].byte_offset = 0;
 441                 gpadl_header->range[0].byte_count = size;
 442                 for (i = 0; i < pagecount; i++)
 443                         gpadl_header->range[0].pfn_array[i] = virt_to_hvpfn(
 444                                 kbuffer + PAGE_SIZE * i);
 445
 446                 *msginfo = msgheader;
 447         }
 448
 449         return 0;
 450 nomem:
 451         kfree(msgheader);
 452         kfree(msgbody);
 453         return -ENOMEM;
 454 }
 455
 456 /*
 457  * vmbus_establish_gpadl - Establish a GPADL for the specified buffer
 458  *
 459  * @channel: a channel
 460  * @kbuffer: from kmalloc or vmalloc
 461  * @size: page-size multiple
 462  * @gpadl_handle: some funky thing
 463  */
 464 int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
 465                                u32 size, u32 *gpadl_handle)
 466 {
 467         struct vmbus_channel_gpadl_header *gpadlmsg;
 468         struct vmbus_channel_gpadl_body *gpadl_body;
 469         struct vmbus_channel_msginfo *msginfo = NULL;
 470         struct vmbus_channel_msginfo *submsginfo, *tmp;
 471         struct list_head *curr;
 472         u32 next_gpadl_handle;
 473         unsigned long flags;
 474         int ret = 0;
 475
 476         next_gpadl_handle =
 477                 (atomic_inc_return(&vmbus_connection.next_gpadl_handle) - 1);
 478
 479         ret = create_gpadl_header(kbuffer, size, &msginfo);
 480         if (ret)
 481                 return ret;
 482
 483         init_completion(&msginfo->waitevent);
 484         msginfo->waiting_channel = channel;
 485
 486         gpadlmsg = (struct vmbus_channel_gpadl_header *)msginfo->msg;
 487         gpadlmsg->header.msgtype = CHANNELMSG_GPADL_HEADER;
 488         gpadlmsg->child_relid = channel->offermsg.child_relid;
 489         gpadlmsg->gpadl = next_gpadl_handle;
 490
 491
 492         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 493         list_add_tail(&msginfo->msglistentry,
 494                       &vmbus_connection.chn_msg_list);
 495
 496         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 497
 498         if (channel->rescind) {
 499                 ret = -ENODEV;
 500                 goto cleanup;
 501         }
 502
 503         ret = vmbus_post_msg(gpadlmsg, msginfo->msgsize -
 504                              sizeof(*msginfo), true);
 505
 506         trace_vmbus_establish_gpadl_header(gpadlmsg, ret);
 507
 508         if (ret != 0)
 509                 goto cleanup;
 510
 511         list_for_each(curr, &msginfo->submsglist) {
 512                 submsginfo = (struct vmbus_channel_msginfo *)curr;
 513                 gpadl_body =
 514                         (struct vmbus_channel_gpadl_body *)submsginfo->msg;
 515
 516                 gpadl_body->header.msgtype =
 517                         CHANNELMSG_GPADL_BODY;
 518                 gpadl_body->gpadl = next_gpadl_handle;
 519
 520                 ret = vmbus_post_msg(gpadl_body,
 521                                      submsginfo->msgsize - sizeof(*submsginfo),
 522                                      true);
 523
 524                 trace_vmbus_establish_gpadl_body(gpadl_body, ret);
 525
 526                 if (ret != 0)
 527                         goto cleanup;
 528
 529         }
 530         wait_for_completion(&msginfo->waitevent);
 531
 532         if (msginfo->response.gpadl_created.creation_status != 0) {
 533                 pr_err("Failed to establish GPADL: err = 0x%x\n",
 534                        msginfo->response.gpadl_created.creation_status);
 535
 536                 ret = -EDQUOT;
 537                 goto cleanup;
 538         }
 539
 540         if (channel->rescind) {
 541                 ret = -ENODEV;
 542                 goto cleanup;
 543         }
 544
 545         /* At this point, we received the gpadl created msg */
 546         *gpadl_handle = gpadlmsg->gpadl;
 547
 548 cleanup:
 549         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 550         list_del(&msginfo->msglistentry);
 551         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 552         list_for_each_entry_safe(submsginfo, tmp, &msginfo->submsglist,
 553                                  msglistentry) {
 554                 kfree(submsginfo);
 555         }
 556
 557         kfree(msginfo);
 558         return ret;
 559 }
 560 EXPORT_SYMBOL_GPL(vmbus_establish_gpadl);
 561
 562 /*
 563  * vmbus_teardown_gpadl -Teardown the specified GPADL handle
 564  */
 565 int vmbus_teardown_gpadl(struct vmbus_channel *channel, u32 gpadl_handle)
 566 {
 567         struct vmbus_channel_gpadl_teardown *msg;
 568         struct vmbus_channel_msginfo *info;
 569         unsigned long flags;
 570         int ret;
 571
 572         info = kmalloc(sizeof(*info) +
 573                        sizeof(struct vmbus_channel_gpadl_teardown), GFP_KERNEL);
 574         if (!info)
 575                 return -ENOMEM;
 576
 577         init_completion(&info->waitevent);
 578         info->waiting_channel = channel;
 579
 580         msg = (struct vmbus_channel_gpadl_teardown *)info->msg;
 581
 582         msg->header.msgtype = CHANNELMSG_GPADL_TEARDOWN;
 583         msg->child_relid = channel->offermsg.child_relid;
 584         msg->gpadl = gpadl_handle;
 585
 586         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 587         list_add_tail(&info->msglistentry,
 588                       &vmbus_connection.chn_msg_list);
 589         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 590
 591         if (channel->rescind)
 592                 goto post_msg_err;
 593
 594         ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_gpadl_teardown),
 595                              true);
 596
 597         trace_vmbus_teardown_gpadl(msg, ret);
 598
 599         if (ret)
 600                 goto post_msg_err;
 601
 602         wait_for_completion(&info->waitevent);
 603
 604 post_msg_err:
 605         /*
 606          * If the channel has been rescinded;
 607          * we will be awakened by the rescind
 608          * handler; set the error code to zero so we don't leak memory.
 609          */
 610         if (channel->rescind)
 611                 ret = 0;
 612
 613         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 614         list_del(&info->msglistentry);
 615         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 616
 617         kfree(info);
 618         return ret;
 619 }
 620 EXPORT_SYMBOL_GPL(vmbus_teardown_gpadl);
 621
 622 void vmbus_reset_channel_cb(struct vmbus_channel *channel)
 623 {
 624         unsigned long flags;
 625
 626         /*
 627          * vmbus_on_event(), running in the per-channel tasklet, can race
 628          * with vmbus_close_internal() in the case of SMP guest, e.g., when
 629          * the former is accessing channel->inbound.ring_buffer, the latter
 630          * could be freeing the ring_buffer pages, so here we must stop it
 631          * first.
 632          *
 633          * vmbus_chan_sched() might call the netvsc driver callback function
 634          * that ends up scheduling NAPI work that accesses the ring buffer.
 635          * At this point, we have to ensure that any such work is completed
 636          * and that the channel ring buffer is no longer being accessed, cf.
 637          * the calls to napi_disable() in netvsc_device_remove().
 638          */
 639         tasklet_disable(&channel->callback_event);
 640
 641         /* See the inline comments in vmbus_chan_sched(). */
 642         spin_lock_irqsave(&channel->sched_lock, flags);
 643         channel->onchannel_callback = NULL;
 644         spin_unlock_irqrestore(&channel->sched_lock, flags);
 645
 646         channel->sc_creation_callback = NULL;
 647
 648         /* Re-enable tasklet for use on re-open */
 649         tasklet_enable(&channel->callback_event);
 650 }
 651
 652 static int vmbus_close_internal(struct vmbus_channel *channel)
 653 {
 654         struct vmbus_channel_close_channel *msg;
 655         int ret;
 656
 657         vmbus_reset_channel_cb(channel);
 658
 659         /*
 660          * In case a device driver's probe() fails (e.g.,
 661          * util_probe() -> vmbus_open() returns -ENOMEM) and the device is
 662          * rescinded later (e.g., we dynamically disable an Integrated Service
 663          * in Hyper-V Manager), the driver's remove() invokes vmbus_close():
 664          * here we should skip most of the below cleanup work.
 665          */
 666         if (channel->state != CHANNEL_OPENED_STATE)
 667                 return -EINVAL;
 668
 669         channel->state = CHANNEL_OPEN_STATE;
 670
 671         /* Send a closing message */
 672
 673         msg = &channel->close_msg.msg;
 674
 675         msg->header.msgtype = CHANNELMSG_CLOSECHANNEL;
 676         msg->child_relid = channel->offermsg.child_relid;
 677
 678         ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_close_channel),
 679                              true);
 680
 681         trace_vmbus_close_internal(msg, ret);
 682
 683         if (ret) {
 684                 pr_err("Close failed: close post msg return is %d\n", ret);
 685                 /*
 686                  * If we failed to post the close msg,
 687                  * it is perhaps better to leak memory.
 688                  */
 689         }
 690
 691         /* Tear down the gpadl for the channel's ring buffer */
 692         else if (channel->ringbuffer_gpadlhandle) {
 693                 ret = vmbus_teardown_gpadl(channel,
 694                                            channel->ringbuffer_gpadlhandle);
 695                 if (ret) {
 696                         pr_err("Close failed: teardown gpadl return %d\n", ret);
 697                         /*
 698                          * If we failed to teardown gpadl,
 699                          * it is perhaps better to leak memory.
 700                          */
 701                 }
 702
 703                 channel->ringbuffer_gpadlhandle = 0;
 704         }
 705
 706         return ret;
 707 }
 708
 709 /* disconnect ring - close all channels */
 710 int vmbus_disconnect_ring(struct vmbus_channel *channel)
 711 {
 712         struct vmbus_channel *cur_channel, *tmp;
 713         int ret;
 714
 715         if (channel->primary_channel != NULL)
 716                 return -EINVAL;
 717
 718         list_for_each_entry_safe(cur_channel, tmp, &channel->sc_list, sc_list) {
 719                 if (cur_channel->rescind)
 720                         wait_for_completion(&cur_channel->rescind_event);
 721
 722                 mutex_lock(&vmbus_connection.channel_mutex);
 723                 if (vmbus_close_internal(cur_channel) == 0) {
 724                         vmbus_free_ring(cur_channel);
 725
 726                         if (cur_channel->rescind)
 727                                 hv_process_channel_removal(cur_channel);
 728                 }
 729                 mutex_unlock(&vmbus_connection.channel_mutex);
 730         }
 731
 732         /*
 733          * Now close the primary.
 734          */
 735         mutex_lock(&vmbus_connection.channel_mutex);
 736         ret = vmbus_close_internal(channel);
 737         mutex_unlock(&vmbus_connection.channel_mutex);
 738
 739         return ret;
 740 }
 741 EXPORT_SYMBOL_GPL(vmbus_disconnect_ring);
 742
 743 /*
 744  * vmbus_close - Close the specified channel
 745  */
 746 void vmbus_close(struct vmbus_channel *channel)
 747 {
 748         if (vmbus_disconnect_ring(channel) == 0)
 749                 vmbus_free_ring(channel);
 750 }
 751 EXPORT_SYMBOL_GPL(vmbus_close);
 752
 753 /**
 754  * vmbus_sendpacket() - Send the specified buffer on the given channel
 755  * @channel: Pointer to vmbus_channel structure
 756  * @buffer: Pointer to the buffer you want to send the data from.
 757  * @bufferlen: Maximum size of what the buffer holds.
 758  * @requestid: Identifier of the request
 759  * @type: Type of packet that is being sent e.g. negotiate, time
 760  *        packet etc.
 761  * @flags: 0 or VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
 762  *
 763  * Sends data in @buffer directly to Hyper-V via the vmbus.
 764  * This will send the data unparsed to Hyper-V.
 765  *
 766  * Mainly used by Hyper-V drivers.
 767  */
 768 int vmbus_sendpacket(struct vmbus_channel *channel, void *buffer,
 769                            u32 bufferlen, u64 requestid,
 770                            enum vmbus_packet_type type, u32 flags)
 771 {
 772         struct vmpacket_descriptor desc;
 773         u32 packetlen = sizeof(struct vmpacket_descriptor) + bufferlen;
 774         u32 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
 775         struct kvec bufferlist[3];
 776         u64 aligned_data = 0;
 777         int num_vecs = ((bufferlen != 0) ? 3 : 1);
 778
 779
 780         /* Setup the descriptor */
 781         desc.type = type; /* VmbusPacketTypeDataInBand; */
 782         desc.flags = flags; /* VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED; */
 783         /* in 8-bytes granularity */
 784         desc.offset8 = sizeof(struct vmpacket_descriptor) >> 3;
 785         desc.len8 = (u16)(packetlen_aligned >> 3);
 786         desc.trans_id = requestid;
 787
 788         bufferlist[0].iov_base = &desc;
 789         bufferlist[0].iov_len = sizeof(struct vmpacket_descriptor);
 790         bufferlist[1].iov_base = buffer;
 791         bufferlist[1].iov_len = bufferlen;
 792         bufferlist[2].iov_base = &aligned_data;
 793         bufferlist[2].iov_len = (packetlen_aligned - packetlen);
 794
 795         return hv_ringbuffer_write(channel, bufferlist, num_vecs);
 796 }
 797 EXPORT_SYMBOL(vmbus_sendpacket);
 798
 799 /*
 800  * vmbus_sendpacket_pagebuffer - Send a range of single-page buffer
 801  * packets using a GPADL Direct packet type. This interface allows you
 802  * to control notifying the host. This will be useful for sending
 803  * batched data. Also the sender can control the send flags
 804  * explicitly.
 805  */
 806 int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
 807                                 struct hv_page_buffer pagebuffers[],
 808                                 u32 pagecount, void *buffer, u32 bufferlen,
 809                                 u64 requestid)
 810 {
 811         int i;
 812         struct vmbus_channel_packet_page_buffer desc;
 813         u32 descsize;
 814         u32 packetlen;
 815         u32 packetlen_aligned;
 816         struct kvec bufferlist[3];
 817         u64 aligned_data = 0;
 818
 819         if (pagecount > MAX_PAGE_BUFFER_COUNT)
 820                 return -EINVAL;
 821
 822         /*
 823          * Adjust the size down since vmbus_channel_packet_page_buffer is the
 824          * largest size we support
 825          */
 826         descsize = sizeof(struct vmbus_channel_packet_page_buffer) -
 827                           ((MAX_PAGE_BUFFER_COUNT - pagecount) *
 828                           sizeof(struct hv_page_buffer));
 829         packetlen = descsize + bufferlen;
 830         packetlen_aligned = ALIGN(packetlen, sizeof(u64));
 831
 832         /* Setup the descriptor */
 833         desc.type = VM_PKT_DATA_USING_GPA_DIRECT;
 834         desc.flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
 835         desc.dataoffset8 = descsize >> 3; /* in 8-bytes granularity */
 836         desc.length8 = (u16)(packetlen_aligned >> 3);
 837         desc.transactionid = requestid;
 838         desc.reserved = 0;
 839         desc.rangecount = pagecount;
 840
 841         for (i = 0; i < pagecount; i++) {
 842                 desc.range[i].len = pagebuffers[i].len;
 843                 desc.range[i].offset = pagebuffers[i].offset;
 844                 desc.range[i].pfn        = pagebuffers[i].pfn;
 845         }
 846
 847         bufferlist[0].iov_base = &desc;
 848         bufferlist[0].iov_len = descsize;
 849         bufferlist[1].iov_base = buffer;
 850         bufferlist[1].iov_len = bufferlen;
 851         bufferlist[2].iov_base = &aligned_data;
 852         bufferlist[2].iov_len = (packetlen_aligned - packetlen);
 853
 854         return hv_ringbuffer_write(channel, bufferlist, 3);
 855 }
 856 EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer);
 857
 858 /*
 859  * vmbus_sendpacket_multipagebuffer - Send a multi-page buffer packet
 860  * using a GPADL Direct packet type.
 861  * The buffer includes the vmbus descriptor.
 862  */
 863 int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
 864                               struct vmbus_packet_mpb_array *desc,
 865                               u32 desc_size,
 866                               void *buffer, u32 bufferlen, u64 requestid)
 867 {
 868         u32 packetlen;
 869         u32 packetlen_aligned;
 870         struct kvec bufferlist[3];
 871         u64 aligned_data = 0;
 872
 873         packetlen = desc_size + bufferlen;
 874         packetlen_aligned = ALIGN(packetlen, sizeof(u64));
 875
 876         /* Setup the descriptor */
 877         desc->type = VM_PKT_DATA_USING_GPA_DIRECT;
 878         desc->flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
 879         desc->dataoffset8 = desc_size >> 3; /* in 8-bytes granularity */
 880         desc->length8 = (u16)(packetlen_aligned >> 3);
 881         desc->transactionid = requestid;
 882         desc->reserved = 0;
 883         desc->rangecount = 1;
 884
 885         bufferlist[0].iov_base = desc;
 886         bufferlist[0].iov_len = desc_size;
 887         bufferlist[1].iov_base = buffer;
 888         bufferlist[1].iov_len = bufferlen;
 889         bufferlist[2].iov_base = &aligned_data;
 890         bufferlist[2].iov_len = (packetlen_aligned - packetlen);
 891
 892         return hv_ringbuffer_write(channel, bufferlist, 3);
 893 }
 894 EXPORT_SYMBOL_GPL(vmbus_sendpacket_mpb_desc);
 895
 896 /**
 897  * __vmbus_recvpacket() - Retrieve the user packet on the specified channel
 898  * @channel: Pointer to vmbus_channel structure
 899  * @buffer: Pointer to the buffer you want to receive the data into.
 900  * @bufferlen: Maximum size of what the buffer can hold.
 901  * @buffer_actual_len: The actual size of the data after it was received.
 902  * @requestid: Identifier of the request
 903  * @raw: true means keep the vmpacket_descriptor header in the received data.
 904  *
 905  * Receives directly from the hyper-v vmbus and puts the data it received
 906  * into Buffer. This will receive the data unparsed from hyper-v.
 907  *
 908  * Mainly used by Hyper-V drivers.
 909  */
 910 static inline int
 911 __vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
 912                    u32 bufferlen, u32 *buffer_actual_len, u64 *requestid,
 913                    bool raw)
 914 {
 915         return hv_ringbuffer_read(channel, buffer, bufferlen,
 916                                   buffer_actual_len, requestid, raw);
 917
 918 }
 919
 920 int vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
 921                      u32 bufferlen, u32 *buffer_actual_len,
 922                      u64 *requestid)
 923 {
 924         return __vmbus_recvpacket(channel, buffer, bufferlen,
 925                                   buffer_actual_len, requestid, false);
 926 }
 927 EXPORT_SYMBOL(vmbus_recvpacket);
 928
 929 /*
 930  * vmbus_recvpacket_raw - Retrieve the raw packet on the specified channel
 931  */
 932 int vmbus_recvpacket_raw(struct vmbus_channel *channel, void *buffer,
 933                               u32 bufferlen, u32 *buffer_actual_len,
 934                               u64 *requestid)
 935 {
 936         return __vmbus_recvpacket(channel, buffer, bufferlen,
 937                                   buffer_actual_len, requestid, true);
 938 }
 939 EXPORT_SYMBOL_GPL(vmbus_recvpacket_raw);