forked from ~ljy/RK356X_SDK_RELEASE
blame | history | patch | commit | commitdiff
hc
2024-02-20 ea08eeccae9297f7aabd2ef7f0c2517ac4549acc 
 kernel/net/vmw_vsock/hyperv_transport.c
....@@ -1,3 +1,4 @@
1
+// SPDX-License-Identifier: GPL-2.0-only
12 /*
23  * Hyper-V transport for vsock
34  *
....@@ -6,31 +7,22 @@
67  * support in the VM by introducing the new vsock transport.
78  *
89  * Copyright (c) 2017, Microsoft Corporation.
9
- *
10
- * This program is free software; you can redistribute it and/or modify it
11
- * under the terms and conditions of the GNU General Public License,
12
- * version 2, as published by the Free Software Foundation.
13
- *
14
- * This program is distributed in the hope it will be useful, but WITHOUT
15
- * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
16
- * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
17
- * more details.
18
- *
1910  */
2011 #include <linux/module.h>
2112 #include <linux/vmalloc.h>
2213 #include <linux/hyperv.h>
2314 #include <net/sock.h>
2415 #include <net/af_vsock.h>
16
+#include <asm/hyperv-tlfs.h>
2517 
26
-/* The host side's design of the feature requires 6 exact 4KB pages for
27
- * recv/send rings respectively -- this is suboptimal considering memory
28
- * consumption, however unluckily we have to live with it, before the
29
- * host comes up with a better design in the future.
18
+/* Older (VMBUS version 'VERSION_WIN10' or before) Windows hosts have some
19
+ * stricter requirements on the hv_sock ring buffer size of six 4K pages.
20
+ * hyperv-tlfs defines HV_HYP_PAGE_SIZE as 4K. Newer hosts don't have this
21
+ * limitation; but, keep the defaults the same for compat.
3022  */
31
-#define PAGE_SIZE_4K		4096
32
-#define RINGBUFFER_HVS_RCV_SIZE (PAGE_SIZE_4K * 6)
33
-#define RINGBUFFER_HVS_SND_SIZE (PAGE_SIZE_4K * 6)
23
+#define RINGBUFFER_HVS_RCV_SIZE (HV_HYP_PAGE_SIZE * 6)
24
+#define RINGBUFFER_HVS_SND_SIZE (HV_HYP_PAGE_SIZE * 6)
25
+#define RINGBUFFER_HVS_MAX_SIZE (HV_HYP_PAGE_SIZE * 64)
3426 
3527 /* The MTU is 16KB per the host side's design */
3628 #define HVS_MTU_SIZE		(1024 * 16)
....@@ -55,14 +47,16 @@
5547 };
5648 
5749 /* We can send up to HVS_MTU_SIZE bytes of payload to the host, but let's use
58
- * a small size, i.e. HVS_SEND_BUF_SIZE, to minimize the dynamically-allocated
59
- * buffer, because tests show there is no significant performance difference.
50
+ * a smaller size, i.e. HVS_SEND_BUF_SIZE, to maximize concurrency between the
51
+ * guest and the host processing as one VMBUS packet is the smallest processing
52
+ * unit.
6053  *
6154  * Note: the buffer can be eliminated in the future when we add new VMBus
6255  * ringbuffer APIs that allow us to directly copy data from userspace buffer
6356  * to VMBus ringbuffer.
6457  */
65
-#define HVS_SEND_BUF_SIZE (PAGE_SIZE_4K - sizeof(struct vmpipe_proto_header))
58
+#define HVS_SEND_BUF_SIZE \
59
+		(HV_HYP_PAGE_SIZE - sizeof(struct vmpipe_proto_header))
6660 
6761 struct hvs_send_buf {
6862 	/* The header before the payload data */
....@@ -85,11 +79,11 @@
8579 					 VMBUS_PKT_TRAILER_SIZE)
8680 
8781 union hvs_service_id {
88
-	uuid_le	srv_id;
82
+	guid_t	srv_id;
8983 
9084 	struct {
9185 		unsigned int svm_port;
92
-		unsigned char b[sizeof(uuid_le) - sizeof(unsigned int)];
86
+		unsigned char b[sizeof(guid_t) - sizeof(unsigned int)];
9387 	};
9488 };
9589 
....@@ -97,8 +91,8 @@
9791 struct hvsock {
9892 	struct vsock_sock *vsk;
9993 
100
-	uuid_le vm_srv_id;
101
-	uuid_le host_srv_id;
94
+	guid_t vm_srv_id;
95
+	guid_t host_srv_id;
10296 
10397 	struct vmbus_channel *chan;
10498 	struct vmpacket_descriptor *recv_desc;
....@@ -154,21 +148,23 @@
154148  */
155149 
156150 /* 00000000-facb-11e6-bd58-64006a7986d3 */
157
-static const uuid_le srv_id_template =
158
-	UUID_LE(0x00000000, 0xfacb, 0x11e6, 0xbd, 0x58,
159
-		0x64, 0x00, 0x6a, 0x79, 0x86, 0xd3);
151
+static const guid_t srv_id_template =
152
+	GUID_INIT(0x00000000, 0xfacb, 0x11e6, 0xbd, 0x58,
153
+		  0x64, 0x00, 0x6a, 0x79, 0x86, 0xd3);
160154 
161
-static bool is_valid_srv_id(const uuid_le *id)
155
+static bool hvs_check_transport(struct vsock_sock *vsk);
156
+
157
+static bool is_valid_srv_id(const guid_t *id)
162158 {
163
-	return !memcmp(&id->b[4], &srv_id_template.b[4], sizeof(uuid_le) - 4);
159
+	return !memcmp(&id->b[4], &srv_id_template.b[4], sizeof(guid_t) - 4);
164160 }
165161 
166
-static unsigned int get_port_by_srv_id(const uuid_le *svr_id)
162
+static unsigned int get_port_by_srv_id(const guid_t *svr_id)
167163 {
168164 	return *((unsigned int *)svr_id);
169165 }
170166 
171
-static void hvs_addr_init(struct sockaddr_vm *addr, const uuid_le *svr_id)
167
+static void hvs_addr_init(struct sockaddr_vm *addr, const guid_t *svr_id)
172168 {
173169 	unsigned int port = get_port_by_srv_id(svr_id);
174170 
....@@ -289,7 +285,7 @@
289285 
290286 static void hvs_open_connection(struct vmbus_channel *chan)
291287 {
292
-	uuid_le *if_instance, *if_type;
288
+	guid_t *if_instance, *if_type;
293289 	unsigned char conn_from_host;
294290 
295291 	struct sockaddr_vm addr;
....@@ -297,7 +293,9 @@
297293 	struct vsock_sock *vnew = NULL;
298294 	struct hvsock *hvs = NULL;
299295 	struct hvsock *hvs_new = NULL;
296
+	int rcvbuf;
300297 	int ret;
298
+	int sndbuf;
301299 
302300 	if_type = &chan->offermsg.offer.if_type;
303301 	if_instance = &chan->offermsg.offer.if_instance;
....@@ -319,8 +317,7 @@
319317 		if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog)
320318 			goto out;
321319 
322
-		new = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
323
-				     sk->sk_type, 0);
320
+		new = vsock_create_connected(sk);
324321 		if (!new)
325322 			goto out;
326323 
....@@ -333,6 +330,14 @@
333330 		vsock_addr_init(&vnew->remote_addr,
334331 				VMADDR_CID_HOST, VMADDR_PORT_ANY);
335332 		vnew->remote_addr.svm_port = get_port_by_srv_id(if_instance);
333
+		ret = vsock_assign_transport(vnew, vsock_sk(sk));
334
+		/* Transport assigned (looking at remote_addr) must be the
335
+		 * same where we received the request.
336
+		 */
337
+		if (ret || !hvs_check_transport(vnew)) {
338
+			sock_put(new);
339
+			goto out;
340
+		}
336341 		hvs_new = vnew->trans;
337342 		hvs_new->chan = chan;
338343 	} else {
....@@ -341,9 +346,34 @@
341346 	}
342347 
343348 	set_channel_read_mode(chan, HV_CALL_DIRECT);
344
-	ret = vmbus_open(chan, RINGBUFFER_HVS_SND_SIZE,
345
-			 RINGBUFFER_HVS_RCV_SIZE, NULL, 0,
346
-			 hvs_channel_cb, conn_from_host ? new : sk);
349
+
350
+	/* Use the socket buffer sizes as hints for the VMBUS ring size. For
351
+	 * server side sockets, 'sk' is the parent socket and thus, this will
352
+	 * allow the child sockets to inherit the size from the parent. Keep
353
+	 * the mins to the default value and align to page size as per VMBUS
354
+	 * requirements.
355
+	 * For the max, the socket core library will limit the socket buffer
356
+	 * size that can be set by the user, but, since currently, the hv_sock
357
+	 * VMBUS ring buffer is physically contiguous allocation, restrict it
358
+	 * further.
359
+	 * Older versions of hv_sock host side code cannot handle bigger VMBUS
360
+	 * ring buffer size. Use the version number to limit the change to newer
361
+	 * versions.
362
+	 */
363
+	if (vmbus_proto_version < VERSION_WIN10_V5) {
364
+		sndbuf = RINGBUFFER_HVS_SND_SIZE;
365
+		rcvbuf = RINGBUFFER_HVS_RCV_SIZE;
366
+	} else {
367
+		sndbuf = max_t(int, sk->sk_sndbuf, RINGBUFFER_HVS_SND_SIZE);
368
+		sndbuf = min_t(int, sndbuf, RINGBUFFER_HVS_MAX_SIZE);
369
+		sndbuf = ALIGN(sndbuf, HV_HYP_PAGE_SIZE);
370
+		rcvbuf = max_t(int, sk->sk_rcvbuf, RINGBUFFER_HVS_RCV_SIZE);
371
+		rcvbuf = min_t(int, rcvbuf, RINGBUFFER_HVS_MAX_SIZE);
372
+		rcvbuf = ALIGN(rcvbuf, HV_HYP_PAGE_SIZE);
373
+	}
374
+
375
+	ret = vmbus_open(chan, sndbuf, rcvbuf, NULL, 0, hvs_channel_cb,
376
+			 conn_from_host ? new : sk);
347377 	if (ret != 0) {
348378 		if (conn_from_host) {
349379 			hvs_new->chan = NULL;
....@@ -369,9 +399,8 @@
369399 
370400 	if (conn_from_host) {
371401 		new->sk_state = TCP_ESTABLISHED;
372
-		sk->sk_ack_backlog++;
402
+		sk_acceptq_added(sk);
373403 
374
-		hvs_addr_init(&vnew->local_addr, if_type);
375404 		hvs_new->vm_srv_id = *if_type;
376405 		hvs_new->host_srv_id = *if_instance;
377406 
....@@ -402,6 +431,7 @@
402431 static int hvs_sock_init(struct vsock_sock *vsk, struct vsock_sock *psk)
403432 {
404433 	struct hvsock *hvs;
434
+	struct sock *sk = sk_vsock(vsk);
405435 
406436 	hvs = kzalloc(sizeof(*hvs), GFP_KERNEL);
407437 	if (!hvs)
....@@ -409,7 +439,8 @@
409439 
410440 	vsk->trans = hvs;
411441 	hvs->vsk = vsk;
412
-
442
+	sk->sk_sndbuf = RINGBUFFER_HVS_SND_SIZE;
443
+	sk->sk_rcvbuf = RINGBUFFER_HVS_RCV_SIZE;
413444 	return 0;
414445 }
415446 
....@@ -491,12 +522,9 @@
491522 
492523 static void hvs_release(struct vsock_sock *vsk)
493524 {
494
-	struct sock *sk = sk_vsock(vsk);
495525 	bool remove_sock;
496526 
497
-	lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
498527 	remove_sock = hvs_close_lock_held(vsk);
499
-	release_sock(sk);
500528 	if (remove_sock)
501529 		vsock_remove_sock(vsk);
502530 }
....@@ -601,28 +629,44 @@
601629 	struct hvsock *hvs = vsk->trans;
602630 	struct vmbus_channel *chan = hvs->chan;
603631 	struct hvs_send_buf *send_buf;
604
-	ssize_t to_write, max_writable, ret;
632
+	ssize_t to_write, max_writable;
633
+	ssize_t ret = 0;
634
+	ssize_t bytes_written = 0;
605635 
606
-	BUILD_BUG_ON(sizeof(*send_buf) != PAGE_SIZE_4K);
636
+	BUILD_BUG_ON(sizeof(*send_buf) != HV_HYP_PAGE_SIZE);
607637 
608638 	send_buf = kmalloc(sizeof(*send_buf), GFP_KERNEL);
609639 	if (!send_buf)
610640 		return -ENOMEM;
611641 
612
-	max_writable = hvs_channel_writable_bytes(chan);
613
-	to_write = min_t(ssize_t, len, max_writable);
614
-	to_write = min_t(ssize_t, to_write, HVS_SEND_BUF_SIZE);
642
+	/* Reader(s) could be draining data from the channel as we write.
643
+	 * Maximize bandwidth, by iterating until the channel is found to be
644
+	 * full.
645
+	 */
646
+	while (len) {
647
+		max_writable = hvs_channel_writable_bytes(chan);
648
+		if (!max_writable)
649
+			break;
650
+		to_write = min_t(ssize_t, len, max_writable);
651
+		to_write = min_t(ssize_t, to_write, HVS_SEND_BUF_SIZE);
652
+		/* memcpy_from_msg is safe for loop as it advances the offsets
653
+		 * within the message iterator.
654
+		 */
655
+		ret = memcpy_from_msg(send_buf->data, msg, to_write);
656
+		if (ret < 0)
657
+			goto out;
615658 
616
-	ret = memcpy_from_msg(send_buf->data, msg, to_write);
617
-	if (ret < 0)
618
-		goto out;
659
+		ret = hvs_send_data(hvs->chan, send_buf, to_write);
660
+		if (ret < 0)
661
+			goto out;
619662 
620
-	ret = hvs_send_data(hvs->chan, send_buf, to_write);
621
-	if (ret < 0)
622
-		goto out;
623
-
624
-	ret = to_write;
663
+		bytes_written += to_write;
664
+		len -= to_write;
665
+	}
625666 out:
667
+	/* If any data has been sent, return that */
668
+	if (bytes_written)
669
+		ret = bytes_written;
626670 	kfree(send_buf);
627671 	return ret;
628672 }
....@@ -752,37 +796,9 @@
752796 	return 0;
753797 }
754798 
755
-static void hvs_set_buffer_size(struct vsock_sock *vsk, u64 val)
756
-{
757
-	/* Ignored. */
758
-}
759
-
760
-static void hvs_set_min_buffer_size(struct vsock_sock *vsk, u64 val)
761
-{
762
-	/* Ignored. */
763
-}
764
-
765
-static void hvs_set_max_buffer_size(struct vsock_sock *vsk, u64 val)
766
-{
767
-	/* Ignored. */
768
-}
769
-
770
-static u64 hvs_get_buffer_size(struct vsock_sock *vsk)
771
-{
772
-	return -ENOPROTOOPT;
773
-}
774
-
775
-static u64 hvs_get_min_buffer_size(struct vsock_sock *vsk)
776
-{
777
-	return -ENOPROTOOPT;
778
-}
779
-
780
-static u64 hvs_get_max_buffer_size(struct vsock_sock *vsk)
781
-{
782
-	return -ENOPROTOOPT;
783
-}
784
-
785799 static struct vsock_transport hvs_transport = {
800
+	.module                   = THIS_MODULE,
801
+
786802 	.get_local_cid            = hvs_get_local_cid,
787803 
788804 	.init                     = hvs_sock_init,
....@@ -815,13 +831,12 @@
815831 	.notify_send_pre_enqueue  = hvs_notify_send_pre_enqueue,
816832 	.notify_send_post_enqueue = hvs_notify_send_post_enqueue,
817833 
818
-	.set_buffer_size          = hvs_set_buffer_size,
819
-	.set_min_buffer_size      = hvs_set_min_buffer_size,
820
-	.set_max_buffer_size      = hvs_set_max_buffer_size,
821
-	.get_buffer_size          = hvs_get_buffer_size,
822
-	.get_min_buffer_size      = hvs_get_min_buffer_size,
823
-	.get_max_buffer_size      = hvs_get_max_buffer_size,
824834 };
835
+
836
+static bool hvs_check_transport(struct vsock_sock *vsk)
837
+{
838
+	return vsk->transport == &hvs_transport;
839
+}
825840 
826841 static int hvs_probe(struct hv_device *hdev,
827842 		     const struct hv_vmbus_device_id *dev_id)
....@@ -847,6 +862,24 @@
847862 	return 0;
848863 }
849864 
865
+/* hv_sock connections can not persist across hibernation, and all the hv_sock
866
+ * channels are forced to be rescinded before hibernation: see
867
+ * vmbus_bus_suspend(). Here the dummy hvs_suspend() and hvs_resume()
868
+ * are only needed because hibernation requires that every vmbus device's
869
+ * driver should have a .suspend and .resume callback: see vmbus_suspend().
870
+ */
871
+static int hvs_suspend(struct hv_device *hv_dev)
872
+{
873
+	/* Dummy */
874
+	return 0;
875
+}
876
+
877
+static int hvs_resume(struct hv_device *dev)
878
+{
879
+	/* Dummy */
880
+	return 0;
881
+}
882
+
850883 /* This isn't really used. See vmbus_match() and vmbus_probe() */
851884 static const struct hv_vmbus_device_id id_table[] = {
852885 	{},
....@@ -858,6 +891,8 @@
858891 	.id_table	= id_table,
859892 	.probe		= hvs_probe,
860893 	.remove		= hvs_remove,
894
+	.suspend	= hvs_suspend,
895
+	.resume		= hvs_resume,
861896 };
862897 
863898 static int __init hvs_init(void)
....@@ -871,7 +906,7 @@
871906 	if (ret != 0)
872907 		return ret;
873908 
874
-	ret = vsock_core_init(&hvs_transport);
909
+	ret = vsock_core_register(&hvs_transport, VSOCK_TRANSPORT_F_G2H);
875910 	if (ret) {
876911 		vmbus_driver_unregister(&hvs_drv);
877912 		return ret;
....@@ -882,7 +917,7 @@
882917 
883918 static void __exit hvs_exit(void)
884919 {
885
-	vsock_core_exit();
920
+	vsock_core_unregister(&hvs_transport);
886921 	vmbus_driver_unregister(&hvs_drv);
887922 }
888923