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