• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 <linux/set_memory.h>
21 #include <asm/page.h>
22 #include <asm/mem_encrypt.h>
23 #include <asm/mshyperv.h>
24 
25 #include "hyperv_vmbus.h"
26 
27 /*
28  * hv_gpadl_size - Return the real size of a gpadl, the size that Hyper-V uses
29  *
30  * For BUFFER gpadl, Hyper-V uses the exact same size as the guest does.
31  *
32  * For RING gpadl, in each ring, the guest uses one PAGE_SIZE as the header
33  * (because of the alignment requirement), however, the hypervisor only
34  * uses the first HV_HYP_PAGE_SIZE as the header, therefore leaving a
35  * (PAGE_SIZE - HV_HYP_PAGE_SIZE) gap. And since there are two rings in a
36  * ringbuffer, the total size for a RING gpadl that Hyper-V uses is the
37  * total size that the guest uses minus twice of the gap size.
38  */
hv_gpadl_size(enum hv_gpadl_type type,u32 size)39 static inline u32 hv_gpadl_size(enum hv_gpadl_type type, u32 size)
40 {
41 	switch (type) {
42 	case HV_GPADL_BUFFER:
43 		return size;
44 	case HV_GPADL_RING:
45 		/* The size of a ringbuffer must be page-aligned */
46 		BUG_ON(size % PAGE_SIZE);
47 		/*
48 		 * Two things to notice here:
49 		 * 1) We're processing two ring buffers as a unit
50 		 * 2) We're skipping any space larger than HV_HYP_PAGE_SIZE in
51 		 * the first guest-size page of each of the two ring buffers.
52 		 * So we effectively subtract out two guest-size pages, and add
53 		 * back two Hyper-V size pages.
54 		 */
55 		return size - 2 * (PAGE_SIZE - HV_HYP_PAGE_SIZE);
56 	}
57 	BUG();
58 	return 0;
59 }
60 
61 /*
62  * hv_ring_gpadl_send_hvpgoffset - Calculate the send offset (in unit of
63  *                                 HV_HYP_PAGE) in a ring gpadl based on the
64  *                                 offset in the guest
65  *
66  * @offset: the offset (in bytes) where the send ringbuffer starts in the
67  *               virtual address space of the guest
68  */
hv_ring_gpadl_send_hvpgoffset(u32 offset)69 static inline u32 hv_ring_gpadl_send_hvpgoffset(u32 offset)
70 {
71 
72 	/*
73 	 * For RING gpadl, in each ring, the guest uses one PAGE_SIZE as the
74 	 * header (because of the alignment requirement), however, the
75 	 * hypervisor only uses the first HV_HYP_PAGE_SIZE as the header,
76 	 * therefore leaving a (PAGE_SIZE - HV_HYP_PAGE_SIZE) gap.
77 	 *
78 	 * And to calculate the effective send offset in gpadl, we need to
79 	 * substract this gap.
80 	 */
81 	return (offset - (PAGE_SIZE - HV_HYP_PAGE_SIZE)) >> HV_HYP_PAGE_SHIFT;
82 }
83 
84 /*
85  * hv_gpadl_hvpfn - Return the Hyper-V page PFN of the @i th Hyper-V page in
86  *                  the gpadl
87  *
88  * @type: the type of the gpadl
89  * @kbuffer: the pointer to the gpadl in the guest
90  * @size: the total size (in bytes) of the gpadl
91  * @send_offset: the offset (in bytes) where the send ringbuffer starts in the
92  *               virtual address space of the guest
93  * @i: the index
94  */
hv_gpadl_hvpfn(enum hv_gpadl_type type,void * kbuffer,u32 size,u32 send_offset,int i)95 static inline u64 hv_gpadl_hvpfn(enum hv_gpadl_type type, void *kbuffer,
96 				 u32 size, u32 send_offset, int i)
97 {
98 	int send_idx = hv_ring_gpadl_send_hvpgoffset(send_offset);
99 	unsigned long delta = 0UL;
100 
101 	switch (type) {
102 	case HV_GPADL_BUFFER:
103 		break;
104 	case HV_GPADL_RING:
105 		if (i == 0)
106 			delta = 0;
107 		else if (i <= send_idx)
108 			delta = PAGE_SIZE - HV_HYP_PAGE_SIZE;
109 		else
110 			delta = 2 * (PAGE_SIZE - HV_HYP_PAGE_SIZE);
111 		break;
112 	default:
113 		BUG();
114 		break;
115 	}
116 
117 	return virt_to_hvpfn(kbuffer + delta + (HV_HYP_PAGE_SIZE * i));
118 }
119 
120 /*
121  * vmbus_setevent- Trigger an event notification on the specified
122  * channel.
123  */
vmbus_setevent(struct vmbus_channel * channel)124 void vmbus_setevent(struct vmbus_channel *channel)
125 {
126 	struct hv_monitor_page *monitorpage;
127 
128 	trace_vmbus_setevent(channel);
129 
130 	/*
131 	 * For channels marked as in "low latency" mode
132 	 * bypass the monitor page mechanism.
133 	 */
134 	if (channel->offermsg.monitor_allocated && !channel->low_latency) {
135 		vmbus_send_interrupt(channel->offermsg.child_relid);
136 
137 		/* Get the child to parent monitor page */
138 		monitorpage = vmbus_connection.monitor_pages[1];
139 
140 		sync_set_bit(channel->monitor_bit,
141 			(unsigned long *)&monitorpage->trigger_group
142 					[channel->monitor_grp].pending);
143 
144 	} else {
145 		vmbus_set_event(channel);
146 	}
147 }
148 EXPORT_SYMBOL_GPL(vmbus_setevent);
149 
150 /* vmbus_free_ring - drop mapping of ring buffer */
vmbus_free_ring(struct vmbus_channel * channel)151 void vmbus_free_ring(struct vmbus_channel *channel)
152 {
153 	hv_ringbuffer_cleanup(&channel->outbound);
154 	hv_ringbuffer_cleanup(&channel->inbound);
155 
156 	if (channel->ringbuffer_page) {
157 		/* In a CoCo VM leak the memory if it didn't get re-encrypted */
158 		if (!channel->ringbuffer_gpadlhandle.decrypted)
159 			__free_pages(channel->ringbuffer_page,
160 			     get_order(channel->ringbuffer_pagecount
161 				       << PAGE_SHIFT));
162 		channel->ringbuffer_page = NULL;
163 	}
164 }
165 EXPORT_SYMBOL_GPL(vmbus_free_ring);
166 
167 /* vmbus_alloc_ring - allocate and map pages for ring buffer */
vmbus_alloc_ring(struct vmbus_channel * newchannel,u32 send_size,u32 recv_size)168 int vmbus_alloc_ring(struct vmbus_channel *newchannel,
169 		     u32 send_size, u32 recv_size)
170 {
171 	struct page *page;
172 	int order;
173 
174 	if (send_size % PAGE_SIZE || recv_size % PAGE_SIZE)
175 		return -EINVAL;
176 
177 	/* Allocate the ring buffer */
178 	order = get_order(send_size + recv_size);
179 	page = alloc_pages_node(cpu_to_node(newchannel->target_cpu),
180 				GFP_KERNEL|__GFP_ZERO, order);
181 
182 	if (!page)
183 		page = alloc_pages(GFP_KERNEL|__GFP_ZERO, order);
184 
185 	if (!page)
186 		return -ENOMEM;
187 
188 	newchannel->ringbuffer_page = page;
189 	newchannel->ringbuffer_pagecount = (send_size + recv_size) >> PAGE_SHIFT;
190 	newchannel->ringbuffer_send_offset = send_size >> PAGE_SHIFT;
191 
192 	return 0;
193 }
194 EXPORT_SYMBOL_GPL(vmbus_alloc_ring);
195 
196 /* Used for Hyper-V Socket: a guest client's connect() to the host */
vmbus_send_tl_connect_request(const guid_t * shv_guest_servie_id,const guid_t * shv_host_servie_id)197 int vmbus_send_tl_connect_request(const guid_t *shv_guest_servie_id,
198 				  const guid_t *shv_host_servie_id)
199 {
200 	struct vmbus_channel_tl_connect_request conn_msg;
201 	int ret;
202 
203 	memset(&conn_msg, 0, sizeof(conn_msg));
204 	conn_msg.header.msgtype = CHANNELMSG_TL_CONNECT_REQUEST;
205 	conn_msg.guest_endpoint_id = *shv_guest_servie_id;
206 	conn_msg.host_service_id = *shv_host_servie_id;
207 
208 	ret = vmbus_post_msg(&conn_msg, sizeof(conn_msg), true);
209 
210 	trace_vmbus_send_tl_connect_request(&conn_msg, ret);
211 
212 	return ret;
213 }
214 EXPORT_SYMBOL_GPL(vmbus_send_tl_connect_request);
215 
send_modifychannel_without_ack(struct vmbus_channel * channel,u32 target_vp)216 static int send_modifychannel_without_ack(struct vmbus_channel *channel, u32 target_vp)
217 {
218 	struct vmbus_channel_modifychannel msg;
219 	int ret;
220 
221 	memset(&msg, 0, sizeof(msg));
222 	msg.header.msgtype = CHANNELMSG_MODIFYCHANNEL;
223 	msg.child_relid = channel->offermsg.child_relid;
224 	msg.target_vp = target_vp;
225 
226 	ret = vmbus_post_msg(&msg, sizeof(msg), true);
227 	trace_vmbus_send_modifychannel(&msg, ret);
228 
229 	return ret;
230 }
231 
send_modifychannel_with_ack(struct vmbus_channel * channel,u32 target_vp)232 static int send_modifychannel_with_ack(struct vmbus_channel *channel, u32 target_vp)
233 {
234 	struct vmbus_channel_modifychannel *msg;
235 	struct vmbus_channel_msginfo *info;
236 	unsigned long flags;
237 	int ret;
238 
239 	info = kzalloc(sizeof(struct vmbus_channel_msginfo) +
240 				sizeof(struct vmbus_channel_modifychannel),
241 		       GFP_KERNEL);
242 	if (!info)
243 		return -ENOMEM;
244 
245 	init_completion(&info->waitevent);
246 	info->waiting_channel = channel;
247 
248 	msg = (struct vmbus_channel_modifychannel *)info->msg;
249 	msg->header.msgtype = CHANNELMSG_MODIFYCHANNEL;
250 	msg->child_relid = channel->offermsg.child_relid;
251 	msg->target_vp = target_vp;
252 
253 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
254 	list_add_tail(&info->msglistentry, &vmbus_connection.chn_msg_list);
255 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
256 
257 	ret = vmbus_post_msg(msg, sizeof(*msg), true);
258 	trace_vmbus_send_modifychannel(msg, ret);
259 	if (ret != 0) {
260 		spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
261 		list_del(&info->msglistentry);
262 		spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
263 		goto free_info;
264 	}
265 
266 	/*
267 	 * Release channel_mutex; otherwise, vmbus_onoffer_rescind() could block on
268 	 * the mutex and be unable to signal the completion.
269 	 *
270 	 * See the caller target_cpu_store() for information about the usage of the
271 	 * mutex.
272 	 */
273 	mutex_unlock(&vmbus_connection.channel_mutex);
274 	wait_for_completion(&info->waitevent);
275 	mutex_lock(&vmbus_connection.channel_mutex);
276 
277 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
278 	list_del(&info->msglistentry);
279 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
280 
281 	if (info->response.modify_response.status)
282 		ret = -EAGAIN;
283 
284 free_info:
285 	kfree(info);
286 	return ret;
287 }
288 
289 /*
290  * Set/change the vCPU (@target_vp) the channel (@child_relid) will interrupt.
291  *
292  * CHANNELMSG_MODIFYCHANNEL messages are aynchronous.  When VMbus version 5.3
293  * or later is negotiated, Hyper-V always sends an ACK in response to such a
294  * message.  For VMbus version 5.2 and earlier, it never sends an ACK.  With-
295  * out an ACK, we can not know when the host will stop interrupting the "old"
296  * vCPU and start interrupting the "new" vCPU for the given channel.
297  *
298  * The CHANNELMSG_MODIFYCHANNEL message type is supported since VMBus version
299  * VERSION_WIN10_V4_1.
300  */
vmbus_send_modifychannel(struct vmbus_channel * channel,u32 target_vp)301 int vmbus_send_modifychannel(struct vmbus_channel *channel, u32 target_vp)
302 {
303 	if (vmbus_proto_version >= VERSION_WIN10_V5_3)
304 		return send_modifychannel_with_ack(channel, target_vp);
305 	return send_modifychannel_without_ack(channel, target_vp);
306 }
307 EXPORT_SYMBOL_GPL(vmbus_send_modifychannel);
308 
309 /*
310  * create_gpadl_header - Creates a gpadl for the specified buffer
311  */
create_gpadl_header(enum hv_gpadl_type type,void * kbuffer,u32 size,u32 send_offset,struct vmbus_channel_msginfo ** msginfo)312 static int create_gpadl_header(enum hv_gpadl_type type, void *kbuffer,
313 			       u32 size, u32 send_offset,
314 			       struct vmbus_channel_msginfo **msginfo)
315 {
316 	int i;
317 	int pagecount;
318 	struct vmbus_channel_gpadl_header *gpadl_header;
319 	struct vmbus_channel_gpadl_body *gpadl_body;
320 	struct vmbus_channel_msginfo *msgheader;
321 	struct vmbus_channel_msginfo *msgbody = NULL;
322 	u32 msgsize;
323 
324 	int pfnsum, pfncount, pfnleft, pfncurr, pfnsize;
325 
326 	pagecount = hv_gpadl_size(type, size) >> HV_HYP_PAGE_SHIFT;
327 
328 	/* do we need a gpadl body msg */
329 	pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
330 		  sizeof(struct vmbus_channel_gpadl_header) -
331 		  sizeof(struct gpa_range);
332 	pfncount = pfnsize / sizeof(u64);
333 
334 	if (pagecount > pfncount) {
335 		/* we need a gpadl body */
336 		/* fill in the header */
337 		msgsize = sizeof(struct vmbus_channel_msginfo) +
338 			  sizeof(struct vmbus_channel_gpadl_header) +
339 			  sizeof(struct gpa_range) + pfncount * sizeof(u64);
340 		msgheader =  kzalloc(msgsize, GFP_KERNEL);
341 		if (!msgheader)
342 			goto nomem;
343 
344 		INIT_LIST_HEAD(&msgheader->submsglist);
345 		msgheader->msgsize = msgsize;
346 
347 		gpadl_header = (struct vmbus_channel_gpadl_header *)
348 			msgheader->msg;
349 		gpadl_header->rangecount = 1;
350 		gpadl_header->range_buflen = sizeof(struct gpa_range) +
351 					 pagecount * sizeof(u64);
352 		gpadl_header->range[0].byte_offset = 0;
353 		gpadl_header->range[0].byte_count = hv_gpadl_size(type, size);
354 		for (i = 0; i < pfncount; i++)
355 			gpadl_header->range[0].pfn_array[i] = hv_gpadl_hvpfn(
356 				type, kbuffer, size, send_offset, i);
357 		*msginfo = msgheader;
358 
359 		pfnsum = pfncount;
360 		pfnleft = pagecount - pfncount;
361 
362 		/* how many pfns can we fit */
363 		pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
364 			  sizeof(struct vmbus_channel_gpadl_body);
365 		pfncount = pfnsize / sizeof(u64);
366 
367 		/* fill in the body */
368 		while (pfnleft) {
369 			if (pfnleft > pfncount)
370 				pfncurr = pfncount;
371 			else
372 				pfncurr = pfnleft;
373 
374 			msgsize = sizeof(struct vmbus_channel_msginfo) +
375 				  sizeof(struct vmbus_channel_gpadl_body) +
376 				  pfncurr * sizeof(u64);
377 			msgbody = kzalloc(msgsize, GFP_KERNEL);
378 
379 			if (!msgbody) {
380 				struct vmbus_channel_msginfo *pos = NULL;
381 				struct vmbus_channel_msginfo *tmp = NULL;
382 				/*
383 				 * Free up all the allocated messages.
384 				 */
385 				list_for_each_entry_safe(pos, tmp,
386 					&msgheader->submsglist,
387 					msglistentry) {
388 
389 					list_del(&pos->msglistentry);
390 					kfree(pos);
391 				}
392 
393 				goto nomem;
394 			}
395 
396 			msgbody->msgsize = msgsize;
397 			gpadl_body =
398 				(struct vmbus_channel_gpadl_body *)msgbody->msg;
399 
400 			/*
401 			 * Gpadl is u32 and we are using a pointer which could
402 			 * be 64-bit
403 			 * This is governed by the guest/host protocol and
404 			 * so the hypervisor guarantees that this is ok.
405 			 */
406 			for (i = 0; i < pfncurr; i++)
407 				gpadl_body->pfn[i] = hv_gpadl_hvpfn(type,
408 					kbuffer, size, send_offset, pfnsum + i);
409 
410 			/* add to msg header */
411 			list_add_tail(&msgbody->msglistentry,
412 				      &msgheader->submsglist);
413 			pfnsum += pfncurr;
414 			pfnleft -= pfncurr;
415 		}
416 	} else {
417 		/* everything fits in a header */
418 		msgsize = sizeof(struct vmbus_channel_msginfo) +
419 			  sizeof(struct vmbus_channel_gpadl_header) +
420 			  sizeof(struct gpa_range) + pagecount * sizeof(u64);
421 		msgheader = kzalloc(msgsize, GFP_KERNEL);
422 		if (msgheader == NULL)
423 			goto nomem;
424 
425 		INIT_LIST_HEAD(&msgheader->submsglist);
426 		msgheader->msgsize = msgsize;
427 
428 		gpadl_header = (struct vmbus_channel_gpadl_header *)
429 			msgheader->msg;
430 		gpadl_header->rangecount = 1;
431 		gpadl_header->range_buflen = sizeof(struct gpa_range) +
432 					 pagecount * sizeof(u64);
433 		gpadl_header->range[0].byte_offset = 0;
434 		gpadl_header->range[0].byte_count = hv_gpadl_size(type, size);
435 		for (i = 0; i < pagecount; i++)
436 			gpadl_header->range[0].pfn_array[i] = hv_gpadl_hvpfn(
437 				type, kbuffer, size, send_offset, i);
438 
439 		*msginfo = msgheader;
440 	}
441 
442 	return 0;
443 nomem:
444 	kfree(msgheader);
445 	kfree(msgbody);
446 	return -ENOMEM;
447 }
448 
449 /*
450  * __vmbus_establish_gpadl - Establish a GPADL for a buffer or ringbuffer
451  *
452  * @channel: a channel
453  * @type: the type of the corresponding GPADL, only meaningful for the guest.
454  * @kbuffer: from kmalloc or vmalloc
455  * @size: page-size multiple
456  * @send_offset: the offset (in bytes) where the send ring buffer starts,
457  *              should be 0 for BUFFER type gpadl
458  * @gpadl_handle: some funky thing
459  */
__vmbus_establish_gpadl(struct vmbus_channel * channel,enum hv_gpadl_type type,void * kbuffer,u32 size,u32 send_offset,struct vmbus_gpadl * gpadl)460 static int __vmbus_establish_gpadl(struct vmbus_channel *channel,
461 				   enum hv_gpadl_type type, void *kbuffer,
462 				   u32 size, u32 send_offset,
463 				   struct vmbus_gpadl *gpadl)
464 {
465 	struct vmbus_channel_gpadl_header *gpadlmsg;
466 	struct vmbus_channel_gpadl_body *gpadl_body;
467 	struct vmbus_channel_msginfo *msginfo = NULL;
468 	struct vmbus_channel_msginfo *submsginfo, *tmp;
469 	struct list_head *curr;
470 	u32 next_gpadl_handle;
471 	unsigned long flags;
472 	int ret = 0;
473 
474 	next_gpadl_handle =
475 		(atomic_inc_return(&vmbus_connection.next_gpadl_handle) - 1);
476 
477 	ret = create_gpadl_header(type, kbuffer, size, send_offset, &msginfo);
478 	if (ret) {
479 		gpadl->decrypted = false;
480 		return ret;
481 	}
482 
483 	/*
484 	 * Set the "decrypted" flag to true for the set_memory_decrypted()
485 	 * success case. In the failure case, the encryption state of the
486 	 * memory is unknown. Leave "decrypted" as true to ensure the
487 	 * memory will be leaked instead of going back on the free list.
488 	 */
489 	gpadl->decrypted = true;
490 	ret = set_memory_decrypted((unsigned long)kbuffer,
491 				   PFN_UP(size));
492 	if (ret) {
493 		dev_warn(&channel->device_obj->device,
494 			 "Failed to set host visibility for new GPADL %d.\n",
495 			 ret);
496 		return ret;
497 	}
498 
499 	init_completion(&msginfo->waitevent);
500 	msginfo->waiting_channel = channel;
501 
502 	gpadlmsg = (struct vmbus_channel_gpadl_header *)msginfo->msg;
503 	gpadlmsg->header.msgtype = CHANNELMSG_GPADL_HEADER;
504 	gpadlmsg->child_relid = channel->offermsg.child_relid;
505 	gpadlmsg->gpadl = next_gpadl_handle;
506 
507 
508 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
509 	list_add_tail(&msginfo->msglistentry,
510 		      &vmbus_connection.chn_msg_list);
511 
512 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
513 
514 	if (channel->rescind) {
515 		ret = -ENODEV;
516 		goto cleanup;
517 	}
518 
519 	ret = vmbus_post_msg(gpadlmsg, msginfo->msgsize -
520 			     sizeof(*msginfo), true);
521 
522 	trace_vmbus_establish_gpadl_header(gpadlmsg, ret);
523 
524 	if (ret != 0)
525 		goto cleanup;
526 
527 	list_for_each(curr, &msginfo->submsglist) {
528 		submsginfo = (struct vmbus_channel_msginfo *)curr;
529 		gpadl_body =
530 			(struct vmbus_channel_gpadl_body *)submsginfo->msg;
531 
532 		gpadl_body->header.msgtype =
533 			CHANNELMSG_GPADL_BODY;
534 		gpadl_body->gpadl = next_gpadl_handle;
535 
536 		ret = vmbus_post_msg(gpadl_body,
537 				     submsginfo->msgsize - sizeof(*submsginfo),
538 				     true);
539 
540 		trace_vmbus_establish_gpadl_body(gpadl_body, ret);
541 
542 		if (ret != 0)
543 			goto cleanup;
544 
545 	}
546 	wait_for_completion(&msginfo->waitevent);
547 
548 	if (msginfo->response.gpadl_created.creation_status != 0) {
549 		pr_err("Failed to establish GPADL: err = 0x%x\n",
550 		       msginfo->response.gpadl_created.creation_status);
551 
552 		ret = -EDQUOT;
553 		goto cleanup;
554 	}
555 
556 	if (channel->rescind) {
557 		ret = -ENODEV;
558 		goto cleanup;
559 	}
560 
561 	/* At this point, we received the gpadl created msg */
562 	gpadl->gpadl_handle = gpadlmsg->gpadl;
563 	gpadl->buffer = kbuffer;
564 	gpadl->size = size;
565 
566 
567 cleanup:
568 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
569 	list_del(&msginfo->msglistentry);
570 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
571 	list_for_each_entry_safe(submsginfo, tmp, &msginfo->submsglist,
572 				 msglistentry) {
573 		kfree(submsginfo);
574 	}
575 
576 	kfree(msginfo);
577 
578 	if (ret) {
579 		/*
580 		 * If set_memory_encrypted() fails, the decrypted flag is
581 		 * left as true so the memory is leaked instead of being
582 		 * put back on the free list.
583 		 */
584 		if (!set_memory_encrypted((unsigned long)kbuffer, PFN_UP(size)))
585 			gpadl->decrypted = false;
586 	}
587 
588 	return ret;
589 }
590 
591 /*
592  * vmbus_establish_gpadl - Establish a GPADL for the specified buffer
593  *
594  * @channel: a channel
595  * @kbuffer: from kmalloc or vmalloc
596  * @size: page-size multiple
597  * @gpadl_handle: some funky thing
598  */
vmbus_establish_gpadl(struct vmbus_channel * channel,void * kbuffer,u32 size,struct vmbus_gpadl * gpadl)599 int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
600 			  u32 size, struct vmbus_gpadl *gpadl)
601 {
602 	return __vmbus_establish_gpadl(channel, HV_GPADL_BUFFER, kbuffer, size,
603 				       0U, gpadl);
604 }
605 EXPORT_SYMBOL_GPL(vmbus_establish_gpadl);
606 
607 /**
608  * request_arr_init - Allocates memory for the requestor array. Each slot
609  * keeps track of the next available slot in the array. Initially, each
610  * slot points to the next one (as in a Linked List). The last slot
611  * does not point to anything, so its value is U64_MAX by default.
612  * @size The size of the array
613  */
request_arr_init(u32 size)614 static u64 *request_arr_init(u32 size)
615 {
616 	int i;
617 	u64 *req_arr;
618 
619 	req_arr = kcalloc(size, sizeof(u64), GFP_KERNEL);
620 	if (!req_arr)
621 		return NULL;
622 
623 	for (i = 0; i < size - 1; i++)
624 		req_arr[i] = i + 1;
625 
626 	/* Last slot (no more available slots) */
627 	req_arr[i] = U64_MAX;
628 
629 	return req_arr;
630 }
631 
632 /*
633  * vmbus_alloc_requestor - Initializes @rqstor's fields.
634  * Index 0 is the first free slot
635  * @size: Size of the requestor array
636  */
vmbus_alloc_requestor(struct vmbus_requestor * rqstor,u32 size)637 static int vmbus_alloc_requestor(struct vmbus_requestor *rqstor, u32 size)
638 {
639 	u64 *rqst_arr;
640 	unsigned long *bitmap;
641 
642 	rqst_arr = request_arr_init(size);
643 	if (!rqst_arr)
644 		return -ENOMEM;
645 
646 	bitmap = bitmap_zalloc(size, GFP_KERNEL);
647 	if (!bitmap) {
648 		kfree(rqst_arr);
649 		return -ENOMEM;
650 	}
651 
652 	rqstor->req_arr = rqst_arr;
653 	rqstor->req_bitmap = bitmap;
654 	rqstor->size = size;
655 	rqstor->next_request_id = 0;
656 	spin_lock_init(&rqstor->req_lock);
657 
658 	return 0;
659 }
660 
661 /*
662  * vmbus_free_requestor - Frees memory allocated for @rqstor
663  * @rqstor: Pointer to the requestor struct
664  */
vmbus_free_requestor(struct vmbus_requestor * rqstor)665 static void vmbus_free_requestor(struct vmbus_requestor *rqstor)
666 {
667 	kfree(rqstor->req_arr);
668 	bitmap_free(rqstor->req_bitmap);
669 }
670 
__vmbus_open(struct vmbus_channel * newchannel,void * userdata,u32 userdatalen,void (* onchannelcallback)(void * context),void * context)671 static int __vmbus_open(struct vmbus_channel *newchannel,
672 		       void *userdata, u32 userdatalen,
673 		       void (*onchannelcallback)(void *context), void *context)
674 {
675 	struct vmbus_channel_open_channel *open_msg;
676 	struct vmbus_channel_msginfo *open_info = NULL;
677 	struct page *page = newchannel->ringbuffer_page;
678 	u32 send_pages, recv_pages;
679 	unsigned long flags;
680 	int err;
681 
682 	if (userdatalen > MAX_USER_DEFINED_BYTES)
683 		return -EINVAL;
684 
685 	send_pages = newchannel->ringbuffer_send_offset;
686 	recv_pages = newchannel->ringbuffer_pagecount - send_pages;
687 
688 	if (newchannel->state != CHANNEL_OPEN_STATE)
689 		return -EINVAL;
690 
691 	/* Create and init requestor */
692 	if (newchannel->rqstor_size) {
693 		if (vmbus_alloc_requestor(&newchannel->requestor, newchannel->rqstor_size))
694 			return -ENOMEM;
695 	}
696 
697 	newchannel->state = CHANNEL_OPENING_STATE;
698 	newchannel->onchannel_callback = onchannelcallback;
699 	newchannel->channel_callback_context = context;
700 
701 	if (!newchannel->max_pkt_size)
702 		newchannel->max_pkt_size = VMBUS_DEFAULT_MAX_PKT_SIZE;
703 
704 	/* Establish the gpadl for the ring buffer */
705 	newchannel->ringbuffer_gpadlhandle.gpadl_handle = 0;
706 
707 	err = __vmbus_establish_gpadl(newchannel, HV_GPADL_RING,
708 				      page_address(newchannel->ringbuffer_page),
709 				      (send_pages + recv_pages) << PAGE_SHIFT,
710 				      newchannel->ringbuffer_send_offset << PAGE_SHIFT,
711 				      &newchannel->ringbuffer_gpadlhandle);
712 	if (err)
713 		goto error_clean_ring;
714 
715 	err = hv_ringbuffer_init(&newchannel->outbound,
716 				 page, send_pages, 0);
717 	if (err)
718 		goto error_free_gpadl;
719 
720 	err = hv_ringbuffer_init(&newchannel->inbound, &page[send_pages],
721 				 recv_pages, newchannel->max_pkt_size);
722 	if (err)
723 		goto error_free_gpadl;
724 
725 	/* Create and init the channel open message */
726 	open_info = kzalloc(sizeof(*open_info) +
727 			   sizeof(struct vmbus_channel_open_channel),
728 			   GFP_KERNEL);
729 	if (!open_info) {
730 		err = -ENOMEM;
731 		goto error_free_gpadl;
732 	}
733 
734 	init_completion(&open_info->waitevent);
735 	open_info->waiting_channel = newchannel;
736 
737 	open_msg = (struct vmbus_channel_open_channel *)open_info->msg;
738 	open_msg->header.msgtype = CHANNELMSG_OPENCHANNEL;
739 	open_msg->openid = newchannel->offermsg.child_relid;
740 	open_msg->child_relid = newchannel->offermsg.child_relid;
741 	open_msg->ringbuffer_gpadlhandle
742 		= newchannel->ringbuffer_gpadlhandle.gpadl_handle;
743 	/*
744 	 * The unit of ->downstream_ringbuffer_pageoffset is HV_HYP_PAGE and
745 	 * the unit of ->ringbuffer_send_offset (i.e. send_pages) is PAGE, so
746 	 * here we calculate it into HV_HYP_PAGE.
747 	 */
748 	open_msg->downstream_ringbuffer_pageoffset =
749 		hv_ring_gpadl_send_hvpgoffset(send_pages << PAGE_SHIFT);
750 	open_msg->target_vp = hv_cpu_number_to_vp_number(newchannel->target_cpu);
751 
752 	if (userdatalen)
753 		memcpy(open_msg->userdata, userdata, userdatalen);
754 
755 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
756 	list_add_tail(&open_info->msglistentry,
757 		      &vmbus_connection.chn_msg_list);
758 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
759 
760 	if (newchannel->rescind) {
761 		err = -ENODEV;
762 		goto error_clean_msglist;
763 	}
764 
765 	err = vmbus_post_msg(open_msg,
766 			     sizeof(struct vmbus_channel_open_channel), true);
767 
768 	trace_vmbus_open(open_msg, err);
769 
770 	if (err != 0)
771 		goto error_clean_msglist;
772 
773 	wait_for_completion(&open_info->waitevent);
774 
775 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
776 	list_del(&open_info->msglistentry);
777 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
778 
779 	if (newchannel->rescind) {
780 		err = -ENODEV;
781 		goto error_free_info;
782 	}
783 
784 	if (open_info->response.open_result.status) {
785 		err = -EAGAIN;
786 		goto error_free_info;
787 	}
788 
789 	newchannel->state = CHANNEL_OPENED_STATE;
790 	kfree(open_info);
791 	return 0;
792 
793 error_clean_msglist:
794 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
795 	list_del(&open_info->msglistentry);
796 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
797 error_free_info:
798 	kfree(open_info);
799 error_free_gpadl:
800 	vmbus_teardown_gpadl(newchannel, &newchannel->ringbuffer_gpadlhandle);
801 error_clean_ring:
802 	hv_ringbuffer_cleanup(&newchannel->outbound);
803 	hv_ringbuffer_cleanup(&newchannel->inbound);
804 	vmbus_free_requestor(&newchannel->requestor);
805 	newchannel->state = CHANNEL_OPEN_STATE;
806 	return err;
807 }
808 
809 /*
810  * vmbus_connect_ring - Open the channel but reuse ring buffer
811  */
vmbus_connect_ring(struct vmbus_channel * newchannel,void (* onchannelcallback)(void * context),void * context)812 int vmbus_connect_ring(struct vmbus_channel *newchannel,
813 		       void (*onchannelcallback)(void *context), void *context)
814 {
815 	return  __vmbus_open(newchannel, NULL, 0, onchannelcallback, context);
816 }
817 EXPORT_SYMBOL_GPL(vmbus_connect_ring);
818 
819 /*
820  * vmbus_open - Open the specified channel.
821  */
vmbus_open(struct vmbus_channel * newchannel,u32 send_ringbuffer_size,u32 recv_ringbuffer_size,void * userdata,u32 userdatalen,void (* onchannelcallback)(void * context),void * context)822 int vmbus_open(struct vmbus_channel *newchannel,
823 	       u32 send_ringbuffer_size, u32 recv_ringbuffer_size,
824 	       void *userdata, u32 userdatalen,
825 	       void (*onchannelcallback)(void *context), void *context)
826 {
827 	int err;
828 
829 	err = vmbus_alloc_ring(newchannel, send_ringbuffer_size,
830 			       recv_ringbuffer_size);
831 	if (err)
832 		return err;
833 
834 	err = __vmbus_open(newchannel, userdata, userdatalen,
835 			   onchannelcallback, context);
836 	if (err)
837 		vmbus_free_ring(newchannel);
838 
839 	return err;
840 }
841 EXPORT_SYMBOL_GPL(vmbus_open);
842 
843 /*
844  * vmbus_teardown_gpadl -Teardown the specified GPADL handle
845  */
vmbus_teardown_gpadl(struct vmbus_channel * channel,struct vmbus_gpadl * gpadl)846 int vmbus_teardown_gpadl(struct vmbus_channel *channel, struct vmbus_gpadl *gpadl)
847 {
848 	struct vmbus_channel_gpadl_teardown *msg;
849 	struct vmbus_channel_msginfo *info;
850 	unsigned long flags;
851 	int ret;
852 
853 	info = kzalloc(sizeof(*info) +
854 		       sizeof(struct vmbus_channel_gpadl_teardown), GFP_KERNEL);
855 	if (!info)
856 		return -ENOMEM;
857 
858 	init_completion(&info->waitevent);
859 	info->waiting_channel = channel;
860 
861 	msg = (struct vmbus_channel_gpadl_teardown *)info->msg;
862 
863 	msg->header.msgtype = CHANNELMSG_GPADL_TEARDOWN;
864 	msg->child_relid = channel->offermsg.child_relid;
865 	msg->gpadl = gpadl->gpadl_handle;
866 
867 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
868 	list_add_tail(&info->msglistentry,
869 		      &vmbus_connection.chn_msg_list);
870 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
871 
872 	if (channel->rescind)
873 		goto post_msg_err;
874 
875 	ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_gpadl_teardown),
876 			     true);
877 
878 	trace_vmbus_teardown_gpadl(msg, ret);
879 
880 	if (ret)
881 		goto post_msg_err;
882 
883 	wait_for_completion(&info->waitevent);
884 
885 	gpadl->gpadl_handle = 0;
886 
887 post_msg_err:
888 	/*
889 	 * If the channel has been rescinded;
890 	 * we will be awakened by the rescind
891 	 * handler; set the error code to zero so we don't leak memory.
892 	 */
893 	if (channel->rescind)
894 		ret = 0;
895 
896 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
897 	list_del(&info->msglistentry);
898 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
899 
900 	kfree(info);
901 
902 	ret = set_memory_encrypted((unsigned long)gpadl->buffer,
903 				   PFN_UP(gpadl->size));
904 	if (ret)
905 		pr_warn("Fail to set mem host visibility in GPADL teardown %d.\n", ret);
906 
907 	gpadl->decrypted = ret;
908 
909 	return ret;
910 }
911 EXPORT_SYMBOL_GPL(vmbus_teardown_gpadl);
912 
vmbus_reset_channel_cb(struct vmbus_channel * channel)913 void vmbus_reset_channel_cb(struct vmbus_channel *channel)
914 {
915 	unsigned long flags;
916 
917 	/*
918 	 * vmbus_on_event(), running in the per-channel tasklet, can race
919 	 * with vmbus_close_internal() in the case of SMP guest, e.g., when
920 	 * the former is accessing channel->inbound.ring_buffer, the latter
921 	 * could be freeing the ring_buffer pages, so here we must stop it
922 	 * first.
923 	 *
924 	 * vmbus_chan_sched() might call the netvsc driver callback function
925 	 * that ends up scheduling NAPI work that accesses the ring buffer.
926 	 * At this point, we have to ensure that any such work is completed
927 	 * and that the channel ring buffer is no longer being accessed, cf.
928 	 * the calls to napi_disable() in netvsc_device_remove().
929 	 */
930 	tasklet_disable(&channel->callback_event);
931 
932 	/* See the inline comments in vmbus_chan_sched(). */
933 	spin_lock_irqsave(&channel->sched_lock, flags);
934 	channel->onchannel_callback = NULL;
935 	spin_unlock_irqrestore(&channel->sched_lock, flags);
936 
937 	channel->sc_creation_callback = NULL;
938 
939 	/* Re-enable tasklet for use on re-open */
940 	tasklet_enable(&channel->callback_event);
941 }
942 
vmbus_close_internal(struct vmbus_channel * channel)943 static int vmbus_close_internal(struct vmbus_channel *channel)
944 {
945 	struct vmbus_channel_close_channel *msg;
946 	int ret;
947 
948 	vmbus_reset_channel_cb(channel);
949 
950 	/*
951 	 * In case a device driver's probe() fails (e.g.,
952 	 * util_probe() -> vmbus_open() returns -ENOMEM) and the device is
953 	 * rescinded later (e.g., we dynamically disable an Integrated Service
954 	 * in Hyper-V Manager), the driver's remove() invokes vmbus_close():
955 	 * here we should skip most of the below cleanup work.
956 	 */
957 	if (channel->state != CHANNEL_OPENED_STATE)
958 		return -EINVAL;
959 
960 	channel->state = CHANNEL_OPEN_STATE;
961 
962 	/* Send a closing message */
963 
964 	msg = &channel->close_msg.msg;
965 
966 	msg->header.msgtype = CHANNELMSG_CLOSECHANNEL;
967 	msg->child_relid = channel->offermsg.child_relid;
968 
969 	ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_close_channel),
970 			     true);
971 
972 	trace_vmbus_close_internal(msg, ret);
973 
974 	if (ret) {
975 		pr_err("Close failed: close post msg return is %d\n", ret);
976 		/*
977 		 * If we failed to post the close msg,
978 		 * it is perhaps better to leak memory.
979 		 */
980 	}
981 
982 	/* Tear down the gpadl for the channel's ring buffer */
983 	else if (channel->ringbuffer_gpadlhandle.gpadl_handle) {
984 		ret = vmbus_teardown_gpadl(channel, &channel->ringbuffer_gpadlhandle);
985 		if (ret) {
986 			pr_err("Close failed: teardown gpadl return %d\n", ret);
987 			/*
988 			 * If we failed to teardown gpadl,
989 			 * it is perhaps better to leak memory.
990 			 */
991 		}
992 	}
993 
994 	if (!ret)
995 		vmbus_free_requestor(&channel->requestor);
996 
997 	return ret;
998 }
999 
1000 /* disconnect ring - close all channels */
vmbus_disconnect_ring(struct vmbus_channel * channel)1001 int vmbus_disconnect_ring(struct vmbus_channel *channel)
1002 {
1003 	struct vmbus_channel *cur_channel, *tmp;
1004 	int ret;
1005 
1006 	if (channel->primary_channel != NULL)
1007 		return -EINVAL;
1008 
1009 	list_for_each_entry_safe(cur_channel, tmp, &channel->sc_list, sc_list) {
1010 		if (cur_channel->rescind)
1011 			wait_for_completion(&cur_channel->rescind_event);
1012 
1013 		mutex_lock(&vmbus_connection.channel_mutex);
1014 		if (vmbus_close_internal(cur_channel) == 0) {
1015 			vmbus_free_ring(cur_channel);
1016 
1017 			if (cur_channel->rescind)
1018 				hv_process_channel_removal(cur_channel);
1019 		}
1020 		mutex_unlock(&vmbus_connection.channel_mutex);
1021 	}
1022 
1023 	/*
1024 	 * Now close the primary.
1025 	 */
1026 	mutex_lock(&vmbus_connection.channel_mutex);
1027 	ret = vmbus_close_internal(channel);
1028 	mutex_unlock(&vmbus_connection.channel_mutex);
1029 
1030 	return ret;
1031 }
1032 EXPORT_SYMBOL_GPL(vmbus_disconnect_ring);
1033 
1034 /*
1035  * vmbus_close - Close the specified channel
1036  */
vmbus_close(struct vmbus_channel * channel)1037 void vmbus_close(struct vmbus_channel *channel)
1038 {
1039 	if (vmbus_disconnect_ring(channel) == 0)
1040 		vmbus_free_ring(channel);
1041 }
1042 EXPORT_SYMBOL_GPL(vmbus_close);
1043 
1044 /**
1045  * vmbus_sendpacket_getid() - Send the specified buffer on the given channel
1046  * @channel: Pointer to vmbus_channel structure
1047  * @buffer: Pointer to the buffer you want to send the data from.
1048  * @bufferlen: Maximum size of what the buffer holds.
1049  * @requestid: Identifier of the request
1050  * @trans_id: Identifier of the transaction associated to this request, if
1051  *            the send is successful; undefined, otherwise.
1052  * @type: Type of packet that is being sent e.g. negotiate, time
1053  *	  packet etc.
1054  * @flags: 0 or VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
1055  *
1056  * Sends data in @buffer directly to Hyper-V via the vmbus.
1057  * This will send the data unparsed to Hyper-V.
1058  *
1059  * Mainly used by Hyper-V drivers.
1060  */
vmbus_sendpacket_getid(struct vmbus_channel * channel,void * buffer,u32 bufferlen,u64 requestid,u64 * trans_id,enum vmbus_packet_type type,u32 flags)1061 int vmbus_sendpacket_getid(struct vmbus_channel *channel, void *buffer,
1062 			   u32 bufferlen, u64 requestid, u64 *trans_id,
1063 			   enum vmbus_packet_type type, u32 flags)
1064 {
1065 	struct vmpacket_descriptor desc;
1066 	u32 packetlen = sizeof(struct vmpacket_descriptor) + bufferlen;
1067 	u32 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
1068 	struct kvec bufferlist[3];
1069 	u64 aligned_data = 0;
1070 	int num_vecs = ((bufferlen != 0) ? 3 : 1);
1071 
1072 
1073 	/* Setup the descriptor */
1074 	desc.type = type; /* VmbusPacketTypeDataInBand; */
1075 	desc.flags = flags; /* VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED; */
1076 	/* in 8-bytes granularity */
1077 	desc.offset8 = sizeof(struct vmpacket_descriptor) >> 3;
1078 	desc.len8 = (u16)(packetlen_aligned >> 3);
1079 	desc.trans_id = VMBUS_RQST_ERROR; /* will be updated in hv_ringbuffer_write() */
1080 
1081 	bufferlist[0].iov_base = &desc;
1082 	bufferlist[0].iov_len = sizeof(struct vmpacket_descriptor);
1083 	bufferlist[1].iov_base = buffer;
1084 	bufferlist[1].iov_len = bufferlen;
1085 	bufferlist[2].iov_base = &aligned_data;
1086 	bufferlist[2].iov_len = (packetlen_aligned - packetlen);
1087 
1088 	return hv_ringbuffer_write(channel, bufferlist, num_vecs, requestid, trans_id);
1089 }
1090 EXPORT_SYMBOL(vmbus_sendpacket_getid);
1091 
1092 /**
1093  * vmbus_sendpacket() - Send the specified buffer on the given channel
1094  * @channel: Pointer to vmbus_channel structure
1095  * @buffer: Pointer to the buffer you want to send the data from.
1096  * @bufferlen: Maximum size of what the buffer holds.
1097  * @requestid: Identifier of the request
1098  * @type: Type of packet that is being sent e.g. negotiate, time
1099  *	  packet etc.
1100  * @flags: 0 or VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
1101  *
1102  * Sends data in @buffer directly to Hyper-V via the vmbus.
1103  * This will send the data unparsed to Hyper-V.
1104  *
1105  * Mainly used by Hyper-V drivers.
1106  */
vmbus_sendpacket(struct vmbus_channel * channel,void * buffer,u32 bufferlen,u64 requestid,enum vmbus_packet_type type,u32 flags)1107 int vmbus_sendpacket(struct vmbus_channel *channel, void *buffer,
1108 		     u32 bufferlen, u64 requestid,
1109 		     enum vmbus_packet_type type, u32 flags)
1110 {
1111 	return vmbus_sendpacket_getid(channel, buffer, bufferlen,
1112 				      requestid, NULL, type, flags);
1113 }
1114 EXPORT_SYMBOL(vmbus_sendpacket);
1115 
1116 /*
1117  * vmbus_sendpacket_pagebuffer - Send a range of single-page buffer
1118  * packets using a GPADL Direct packet type. This interface allows you
1119  * to control notifying the host. This will be useful for sending
1120  * batched data. Also the sender can control the send flags
1121  * explicitly.
1122  */
vmbus_sendpacket_pagebuffer(struct vmbus_channel * channel,struct hv_page_buffer pagebuffers[],u32 pagecount,void * buffer,u32 bufferlen,u64 requestid)1123 int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
1124 				struct hv_page_buffer pagebuffers[],
1125 				u32 pagecount, void *buffer, u32 bufferlen,
1126 				u64 requestid)
1127 {
1128 	int i;
1129 	struct vmbus_channel_packet_page_buffer desc;
1130 	u32 descsize;
1131 	u32 packetlen;
1132 	u32 packetlen_aligned;
1133 	struct kvec bufferlist[3];
1134 	u64 aligned_data = 0;
1135 
1136 	if (pagecount > MAX_PAGE_BUFFER_COUNT)
1137 		return -EINVAL;
1138 
1139 	/*
1140 	 * Adjust the size down since vmbus_channel_packet_page_buffer is the
1141 	 * largest size we support
1142 	 */
1143 	descsize = sizeof(struct vmbus_channel_packet_page_buffer) -
1144 			  ((MAX_PAGE_BUFFER_COUNT - pagecount) *
1145 			  sizeof(struct hv_page_buffer));
1146 	packetlen = descsize + bufferlen;
1147 	packetlen_aligned = ALIGN(packetlen, sizeof(u64));
1148 
1149 	/* Setup the descriptor */
1150 	desc.type = VM_PKT_DATA_USING_GPA_DIRECT;
1151 	desc.flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
1152 	desc.dataoffset8 = descsize >> 3; /* in 8-bytes granularity */
1153 	desc.length8 = (u16)(packetlen_aligned >> 3);
1154 	desc.transactionid = VMBUS_RQST_ERROR; /* will be updated in hv_ringbuffer_write() */
1155 	desc.reserved = 0;
1156 	desc.rangecount = pagecount;
1157 
1158 	for (i = 0; i < pagecount; i++) {
1159 		desc.range[i].len = pagebuffers[i].len;
1160 		desc.range[i].offset = pagebuffers[i].offset;
1161 		desc.range[i].pfn	 = pagebuffers[i].pfn;
1162 	}
1163 
1164 	bufferlist[0].iov_base = &desc;
1165 	bufferlist[0].iov_len = descsize;
1166 	bufferlist[1].iov_base = buffer;
1167 	bufferlist[1].iov_len = bufferlen;
1168 	bufferlist[2].iov_base = &aligned_data;
1169 	bufferlist[2].iov_len = (packetlen_aligned - packetlen);
1170 
1171 	return hv_ringbuffer_write(channel, bufferlist, 3, requestid, NULL);
1172 }
1173 EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer);
1174 
1175 /*
1176  * vmbus_sendpacket_multipagebuffer - Send a multi-page buffer packet
1177  * using a GPADL Direct packet type.
1178  * The buffer includes the vmbus descriptor.
1179  */
vmbus_sendpacket_mpb_desc(struct vmbus_channel * channel,struct vmbus_packet_mpb_array * desc,u32 desc_size,void * buffer,u32 bufferlen,u64 requestid)1180 int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
1181 			      struct vmbus_packet_mpb_array *desc,
1182 			      u32 desc_size,
1183 			      void *buffer, u32 bufferlen, u64 requestid)
1184 {
1185 	u32 packetlen;
1186 	u32 packetlen_aligned;
1187 	struct kvec bufferlist[3];
1188 	u64 aligned_data = 0;
1189 
1190 	packetlen = desc_size + bufferlen;
1191 	packetlen_aligned = ALIGN(packetlen, sizeof(u64));
1192 
1193 	/* Setup the descriptor */
1194 	desc->type = VM_PKT_DATA_USING_GPA_DIRECT;
1195 	desc->flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
1196 	desc->dataoffset8 = desc_size >> 3; /* in 8-bytes granularity */
1197 	desc->length8 = (u16)(packetlen_aligned >> 3);
1198 	desc->transactionid = VMBUS_RQST_ERROR; /* will be updated in hv_ringbuffer_write() */
1199 	desc->reserved = 0;
1200 	desc->rangecount = 1;
1201 
1202 	bufferlist[0].iov_base = desc;
1203 	bufferlist[0].iov_len = desc_size;
1204 	bufferlist[1].iov_base = buffer;
1205 	bufferlist[1].iov_len = bufferlen;
1206 	bufferlist[2].iov_base = &aligned_data;
1207 	bufferlist[2].iov_len = (packetlen_aligned - packetlen);
1208 
1209 	return hv_ringbuffer_write(channel, bufferlist, 3, requestid, NULL);
1210 }
1211 EXPORT_SYMBOL_GPL(vmbus_sendpacket_mpb_desc);
1212 
1213 /**
1214  * __vmbus_recvpacket() - Retrieve the user packet on the specified channel
1215  * @channel: Pointer to vmbus_channel structure
1216  * @buffer: Pointer to the buffer you want to receive the data into.
1217  * @bufferlen: Maximum size of what the buffer can hold.
1218  * @buffer_actual_len: The actual size of the data after it was received.
1219  * @requestid: Identifier of the request
1220  * @raw: true means keep the vmpacket_descriptor header in the received data.
1221  *
1222  * Receives directly from the hyper-v vmbus and puts the data it received
1223  * into Buffer. This will receive the data unparsed from hyper-v.
1224  *
1225  * Mainly used by Hyper-V drivers.
1226  */
1227 static inline int
__vmbus_recvpacket(struct vmbus_channel * channel,void * buffer,u32 bufferlen,u32 * buffer_actual_len,u64 * requestid,bool raw)1228 __vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
1229 		   u32 bufferlen, u32 *buffer_actual_len, u64 *requestid,
1230 		   bool raw)
1231 {
1232 	return hv_ringbuffer_read(channel, buffer, bufferlen,
1233 				  buffer_actual_len, requestid, raw);
1234 
1235 }
1236 
vmbus_recvpacket(struct vmbus_channel * channel,void * buffer,u32 bufferlen,u32 * buffer_actual_len,u64 * requestid)1237 int vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
1238 		     u32 bufferlen, u32 *buffer_actual_len,
1239 		     u64 *requestid)
1240 {
1241 	return __vmbus_recvpacket(channel, buffer, bufferlen,
1242 				  buffer_actual_len, requestid, false);
1243 }
1244 EXPORT_SYMBOL(vmbus_recvpacket);
1245 
1246 /*
1247  * vmbus_recvpacket_raw - Retrieve the raw packet on the specified channel
1248  */
vmbus_recvpacket_raw(struct vmbus_channel * channel,void * buffer,u32 bufferlen,u32 * buffer_actual_len,u64 * requestid)1249 int vmbus_recvpacket_raw(struct vmbus_channel *channel, void *buffer,
1250 			      u32 bufferlen, u32 *buffer_actual_len,
1251 			      u64 *requestid)
1252 {
1253 	return __vmbus_recvpacket(channel, buffer, bufferlen,
1254 				  buffer_actual_len, requestid, true);
1255 }
1256 EXPORT_SYMBOL_GPL(vmbus_recvpacket_raw);
1257 
1258 /*
1259  * vmbus_next_request_id - Returns a new request id. It is also
1260  * the index at which the guest memory address is stored.
1261  * Uses a spin lock to avoid race conditions.
1262  * @channel: Pointer to the VMbus channel struct
1263  * @rqst_add: Guest memory address to be stored in the array
1264  */
vmbus_next_request_id(struct vmbus_channel * channel,u64 rqst_addr)1265 u64 vmbus_next_request_id(struct vmbus_channel *channel, u64 rqst_addr)
1266 {
1267 	struct vmbus_requestor *rqstor = &channel->requestor;
1268 	unsigned long flags;
1269 	u64 current_id;
1270 
1271 	/* Check rqstor has been initialized */
1272 	if (!channel->rqstor_size)
1273 		return VMBUS_NO_RQSTOR;
1274 
1275 	lock_requestor(channel, flags);
1276 	current_id = rqstor->next_request_id;
1277 
1278 	/* Requestor array is full */
1279 	if (current_id >= rqstor->size) {
1280 		unlock_requestor(channel, flags);
1281 		return VMBUS_RQST_ERROR;
1282 	}
1283 
1284 	rqstor->next_request_id = rqstor->req_arr[current_id];
1285 	rqstor->req_arr[current_id] = rqst_addr;
1286 
1287 	/* The already held spin lock provides atomicity */
1288 	bitmap_set(rqstor->req_bitmap, current_id, 1);
1289 
1290 	unlock_requestor(channel, flags);
1291 
1292 	/*
1293 	 * Cannot return an ID of 0, which is reserved for an unsolicited
1294 	 * message from Hyper-V; Hyper-V does not acknowledge (respond to)
1295 	 * VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED requests with ID of
1296 	 * 0 sent by the guest.
1297 	 */
1298 	return current_id + 1;
1299 }
1300 EXPORT_SYMBOL_GPL(vmbus_next_request_id);
1301 
1302 /* As in vmbus_request_addr_match() but without the requestor lock */
__vmbus_request_addr_match(struct vmbus_channel * channel,u64 trans_id,u64 rqst_addr)1303 u64 __vmbus_request_addr_match(struct vmbus_channel *channel, u64 trans_id,
1304 			       u64 rqst_addr)
1305 {
1306 	struct vmbus_requestor *rqstor = &channel->requestor;
1307 	u64 req_addr;
1308 
1309 	/* Check rqstor has been initialized */
1310 	if (!channel->rqstor_size)
1311 		return VMBUS_NO_RQSTOR;
1312 
1313 	/* Hyper-V can send an unsolicited message with ID of 0 */
1314 	if (!trans_id)
1315 		return VMBUS_RQST_ERROR;
1316 
1317 	/* Data corresponding to trans_id is stored at trans_id - 1 */
1318 	trans_id--;
1319 
1320 	/* Invalid trans_id */
1321 	if (trans_id >= rqstor->size || !test_bit(trans_id, rqstor->req_bitmap))
1322 		return VMBUS_RQST_ERROR;
1323 
1324 	req_addr = rqstor->req_arr[trans_id];
1325 	if (rqst_addr == VMBUS_RQST_ADDR_ANY || req_addr == rqst_addr) {
1326 		rqstor->req_arr[trans_id] = rqstor->next_request_id;
1327 		rqstor->next_request_id = trans_id;
1328 
1329 		/* The already held spin lock provides atomicity */
1330 		bitmap_clear(rqstor->req_bitmap, trans_id, 1);
1331 	}
1332 
1333 	return req_addr;
1334 }
1335 EXPORT_SYMBOL_GPL(__vmbus_request_addr_match);
1336 
1337 /*
1338  * vmbus_request_addr_match - Clears/removes @trans_id from the @channel's
1339  * requestor, provided the memory address stored at @trans_id equals @rqst_addr
1340  * (or provided @rqst_addr matches the sentinel value VMBUS_RQST_ADDR_ANY).
1341  *
1342  * Returns the memory address stored at @trans_id, or VMBUS_RQST_ERROR if
1343  * @trans_id is not contained in the requestor.
1344  *
1345  * Acquires and releases the requestor spin lock.
1346  */
vmbus_request_addr_match(struct vmbus_channel * channel,u64 trans_id,u64 rqst_addr)1347 u64 vmbus_request_addr_match(struct vmbus_channel *channel, u64 trans_id,
1348 			     u64 rqst_addr)
1349 {
1350 	unsigned long flags;
1351 	u64 req_addr;
1352 
1353 	lock_requestor(channel, flags);
1354 	req_addr = __vmbus_request_addr_match(channel, trans_id, rqst_addr);
1355 	unlock_requestor(channel, flags);
1356 
1357 	return req_addr;
1358 }
1359 EXPORT_SYMBOL_GPL(vmbus_request_addr_match);
1360 
1361 /*
1362  * vmbus_request_addr - Returns the memory address stored at @trans_id
1363  * in @rqstor. Uses a spin lock to avoid race conditions.
1364  * @channel: Pointer to the VMbus channel struct
1365  * @trans_id: Request id sent back from Hyper-V. Becomes the requestor's
1366  * next request id.
1367  */
vmbus_request_addr(struct vmbus_channel * channel,u64 trans_id)1368 u64 vmbus_request_addr(struct vmbus_channel *channel, u64 trans_id)
1369 {
1370 	return vmbus_request_addr_match(channel, trans_id, VMBUS_RQST_ADDR_ANY);
1371 }
1372 EXPORT_SYMBOL_GPL(vmbus_request_addr);
1373