hv_util.c revision 255414 
1/*-
2 * Copyright (c) 2009-2012 Microsoft Corp.
3 * Copyright (c) 2012 NetApp Inc.
4 * Copyright (c) 2012 Citrix Inc.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice unmodified, this list of conditions, and the following
12 *    disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 */
28
29/**
30 * A common driver for all hyper-V util services.
31 */
32
33#include <sys/param.h>
34#include <sys/kernel.h>
35#include <sys/bus.h>
36#include <sys/malloc.h>
37#include <sys/module.h>
38#include <sys/reboot.h>
39#include <sys/timetc.h>
40#include <sys/syscallsubr.h>
41
42#include <dev/hyperv/include/hyperv.h>
43#include "hv_kvp.h"
44
45/* Time Sync data */
46typedef struct {
47	uint64_t data;
48} time_sync_data;
49
50static void hv_shutdown_cb(void *context);
51static void hv_heartbeat_cb(void *context);
52static void hv_timesync_cb(void *context);
53
54static int hv_timesync_init(hv_vmbus_service *serv);
55
56/**
57 * Note: GUID codes below are predefined by the host hypervisor
58 * (Hyper-V and Azure)interface and required for correct operation.
59 */
60hv_vmbus_service service_table[] = {
61	/* Shutdown Service */
62	{ .guid.data = {0x31, 0x60, 0x0B, 0X0E, 0x13, 0x52, 0x34, 0x49,
63			0x81, 0x8B, 0x38, 0XD9, 0x0C, 0xED, 0x39, 0xDB},
64	  .name  = "Hyper-V Shutdown Service\n",
65	  .enabled = TRUE,
66	  .callback = hv_shutdown_cb,
67	},
68
69        /* Time Synch Service */
70        { .guid.data = {0x30, 0xe6, 0x27, 0x95, 0xae, 0xd0, 0x7b, 0x49,
71			0xad, 0xce, 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf},
72	  .name = "Hyper-V Time Synch Service\n",
73	  .enabled = TRUE,
74	  .init = hv_timesync_init,
75	  .callback = hv_timesync_cb,
76	},
77
78        /* Heartbeat Service */
79        { .guid.data = {0x39, 0x4f, 0x16, 0x57, 0x15, 0x91, 0x78, 0x4e,
80			0xab, 0x55, 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d},
81	  .name = "Hyper-V Heartbeat Service\n",
82	  .enabled = TRUE,
83          .callback = hv_heartbeat_cb,
84	},
85
86        /* KVP (Key Value Pair) Service */
87        { .guid.data = {0xe7, 0xf4, 0xa0, 0xa9, 0x45, 0x5a, 0x96, 0x4d,
88			0xb8, 0x27, 0x8a, 0x84, 0x1e, 0x8c, 0x3,  0xe6},
89	  .name = "Hyper-V KVP Service\n",
90	  .enabled = TRUE,
91	  .init = hv_kvp_init,
92	  .callback = hv_kvp_callback,
93	},
94};
95
96/*
97 * Receive buffer pointers. There is one buffer per utility service. The
98 * buffer is allocated during attach().
99 */
100uint8_t *receive_buffer[HV_MAX_UTIL_SERVICES];
101
102struct hv_ictimesync_data {
103	uint64_t    parenttime;
104	uint64_t    childtime;
105	uint64_t    roundtriptime;
106	uint8_t     flags;
107} __packed;
108
109static int
110hv_timesync_init(hv_vmbus_service *serv)
111{
112
113	serv->work_queue = hv_work_queue_create("Time Sync");
114	if (serv->work_queue == NULL)
115		return (ENOMEM);
116	return (0);
117}
118
119static void
120hv_negotiate_version(
121	struct hv_vmbus_icmsg_hdr*		icmsghdrp,
122	struct hv_vmbus_icmsg_negotiate*	negop,
123	uint8_t*				buf)
124{
125	icmsghdrp->icmsgsize = 0x10;
126
127	negop = (struct hv_vmbus_icmsg_negotiate *)&buf[
128		sizeof(struct hv_vmbus_pipe_hdr) +
129		sizeof(struct hv_vmbus_icmsg_hdr)];
130
131	if (negop->icframe_vercnt >= 2 &&
132	    negop->icversion_data[1].major == 3) {
133		negop->icversion_data[0].major = 3;
134		negop->icversion_data[0].minor = 0;
135		negop->icversion_data[1].major = 3;
136		negop->icversion_data[1].minor = 0;
137	} else {
138		negop->icversion_data[0].major = 1;
139		negop->icversion_data[0].minor = 0;
140		negop->icversion_data[1].major = 1;
141		negop->icversion_data[1].minor = 0;
142	}
143
144	negop->icframe_vercnt = 1;
145	negop->icmsg_vercnt = 1;
146}
147
148
149/**
150 * Set host time based on time sync message from host
151 */
152static void
153hv_set_host_time(void *context)
154{
155 	time_sync_data *time_msg = context;
156	uint64_t hosttime = time_msg->data;
157	struct timespec guest_ts, host_ts;
158	uint64_t host_tns;
159	int64_t diff;
160	int error;
161
162	host_tns = (hosttime - HV_WLTIMEDELTA) * 100;
163	host_ts.tv_sec = (time_t)(host_tns/HV_NANO_SEC_PER_SEC);
164	host_ts.tv_nsec = (long)(host_tns%HV_NANO_SEC_PER_SEC);
165
166	nanotime(&guest_ts);
167
168	diff = (int64_t)host_ts.tv_sec - (int64_t)guest_ts.tv_sec;
169
170	/*
171	 * If host differs by 5 seconds then make the guest catch up
172	 */
173	if (diff > 5 || diff < -5) {
174		error = kern_clock_settime(curthread, CLOCK_REALTIME,
175		    &host_ts);
176	}
177
178	/*
179	 * Free the hosttime that was allocated in hv_adj_guesttime()
180	 */
181	free(time_msg, M_DEVBUF);
182}
183
184/**
185 * @brief Synchronize time with host after reboot, restore, etc.
186 *
187 * ICTIMESYNCFLAG_SYNC flag bit indicates reboot, restore events of the VM.
188 * After reboot the flag ICTIMESYNCFLAG_SYNC is included in the first time
189 * message after the timesync channel is opened. Since the hv_utils module is
190 * loaded after hv_vmbus, the first message is usually missed. The other
191 * thing is, systime is automatically set to emulated hardware clock which may
192 * not be UTC time or in the same time zone. So, to override these effects, we
193 * use the first 50 time samples for initial system time setting.
194 */
195static inline
196void hv_adj_guesttime(uint64_t hosttime, uint8_t flags)
197{
198	time_sync_data* time_msg;
199
200	time_msg = malloc(sizeof(time_sync_data), M_DEVBUF, M_NOWAIT);
201
202	if (time_msg == NULL)
203		return;
204
205	time_msg->data = hosttime;
206
207	if ((flags & HV_ICTIMESYNCFLAG_SYNC) != 0) {
208		hv_queue_work_item(service_table[HV_TIME_SYNCH].work_queue,
209		    hv_set_host_time, time_msg);
210	} else if ((flags & HV_ICTIMESYNCFLAG_SAMPLE) != 0) {
211		hv_queue_work_item(service_table[HV_TIME_SYNCH].work_queue,
212		    hv_set_host_time, time_msg);
213	} else {
214		free(time_msg, M_DEVBUF);
215	}
216}
217
218/**
219 * Time Sync Channel message handler
220 */
221static void
222hv_timesync_cb(void *context)
223{
224	hv_vmbus_channel*	channel = context;
225	hv_vmbus_icmsg_hdr*	icmsghdrp;
226	uint32_t		recvlen;
227	uint64_t		requestId;
228	int			ret;
229	uint8_t*		time_buf;
230	struct hv_ictimesync_data* timedatap;
231
232	time_buf = receive_buffer[HV_TIME_SYNCH];
233
234	ret = hv_vmbus_channel_recv_packet(channel, time_buf,
235					    PAGE_SIZE, &recvlen, &requestId);
236
237	if ((ret == 0) && recvlen > 0) {
238	    icmsghdrp = (struct hv_vmbus_icmsg_hdr *) &time_buf[
239		sizeof(struct hv_vmbus_pipe_hdr)];
240
241	    if (icmsghdrp->icmsgtype == HV_ICMSGTYPE_NEGOTIATE) {
242		hv_negotiate_version(icmsghdrp, NULL, time_buf);
243	    } else {
244		timedatap = (struct hv_ictimesync_data *) &time_buf[
245		    sizeof(struct hv_vmbus_pipe_hdr) +
246			sizeof(struct hv_vmbus_icmsg_hdr)];
247		hv_adj_guesttime(timedatap->parenttime, timedatap->flags);
248	    }
249
250	    icmsghdrp->icflags = HV_ICMSGHDRFLAG_TRANSACTION
251		| HV_ICMSGHDRFLAG_RESPONSE;
252
253	    hv_vmbus_channel_send_packet(channel, time_buf,
254		recvlen, requestId,
255		HV_VMBUS_PACKET_TYPE_DATA_IN_BAND, 0);
256	}
257}
258
259/**
260 * Shutdown
261 */
262static void
263hv_shutdown_cb(void *context)
264{
265	uint8_t*			buf;
266	hv_vmbus_channel*		channel = context;
267	uint8_t				execute_shutdown = 0;
268	hv_vmbus_icmsg_hdr*		icmsghdrp;
269	uint32_t			recv_len;
270	uint64_t			request_id;
271	int				ret;
272	hv_vmbus_shutdown_msg_data*	shutdown_msg;
273
274	buf = receive_buffer[HV_SHUT_DOWN];
275
276	ret = hv_vmbus_channel_recv_packet(channel, buf, PAGE_SIZE,
277					    &recv_len, &request_id);
278
279	if ((ret == 0) && recv_len > 0) {
280
281	    icmsghdrp = (struct hv_vmbus_icmsg_hdr *)
282		&buf[sizeof(struct hv_vmbus_pipe_hdr)];
283
284	    if (icmsghdrp->icmsgtype == HV_ICMSGTYPE_NEGOTIATE) {
285		hv_negotiate_version(icmsghdrp, NULL, buf);
286
287	    } else {
288		shutdown_msg =
289		    (struct hv_vmbus_shutdown_msg_data *)
290		    &buf[sizeof(struct hv_vmbus_pipe_hdr) +
291			sizeof(struct hv_vmbus_icmsg_hdr)];
292
293		switch (shutdown_msg->flags) {
294		    case 0:
295		    case 1:
296			icmsghdrp->status = HV_S_OK;
297			execute_shutdown = 1;
298			if(bootverbose)
299			    printf("Shutdown request received -"
300				    " graceful shutdown initiated\n");
301			break;
302		    default:
303			icmsghdrp->status = HV_E_FAIL;
304			execute_shutdown = 0;
305			printf("Shutdown request received -"
306			    " Invalid request\n");
307			break;
308		    }
309	    }
310
311	    icmsghdrp->icflags = HV_ICMSGHDRFLAG_TRANSACTION |
312				 HV_ICMSGHDRFLAG_RESPONSE;
313
314	    hv_vmbus_channel_send_packet(channel, buf,
315					recv_len, request_id,
316					HV_VMBUS_PACKET_TYPE_DATA_IN_BAND, 0);
317	}
318
319	if (execute_shutdown)
320	    shutdown_nice(RB_POWEROFF);
321}
322
323/**
324 * Process heartbeat message
325 */
326static void
327hv_heartbeat_cb(void *context)
328{
329	uint8_t*		buf;
330	hv_vmbus_channel*	channel = context;
331	uint32_t		recvlen;
332	uint64_t		requestid;
333	int			ret;
334
335	struct hv_vmbus_heartbeat_msg_data*	heartbeat_msg;
336	struct hv_vmbus_icmsg_hdr*		icmsghdrp;
337
338	buf = receive_buffer[HV_HEART_BEAT];
339
340	ret = hv_vmbus_channel_recv_packet(channel, buf, PAGE_SIZE, &recvlen,
341					    &requestid);
342
343	if ((ret == 0) && recvlen > 0) {
344
345	    icmsghdrp = (struct hv_vmbus_icmsg_hdr *)
346		&buf[sizeof(struct hv_vmbus_pipe_hdr)];
347
348	    if (icmsghdrp->icmsgtype == HV_ICMSGTYPE_NEGOTIATE) {
349		hv_negotiate_version(icmsghdrp, NULL, buf);
350
351	    } else {
352		heartbeat_msg =
353		    (struct hv_vmbus_heartbeat_msg_data *)
354			&buf[sizeof(struct hv_vmbus_pipe_hdr) +
355			     sizeof(struct hv_vmbus_icmsg_hdr)];
356
357		heartbeat_msg->seq_num += 1;
358	    }
359
360	    icmsghdrp->icflags = HV_ICMSGHDRFLAG_TRANSACTION |
361				 HV_ICMSGHDRFLAG_RESPONSE;
362
363	    hv_vmbus_channel_send_packet(channel, buf, recvlen, requestid,
364		HV_VMBUS_PACKET_TYPE_DATA_IN_BAND, 0);
365	}
366}
367
368
369static int
370hv_util_probe(device_t dev)
371{
372	int i;
373	int rtn_value = ENXIO;
374
375	for (i = 0; i < HV_MAX_UTIL_SERVICES; i++) {
376	    const char *p = vmbus_get_type(dev);
377	    if (service_table[i].enabled && !memcmp(p, &service_table[i].guid, sizeof(hv_guid))) {
378		device_set_softc(dev, (void *) (&service_table[i]));
379		rtn_value = 0;
380	    }
381	}
382
383	return rtn_value;
384}
385
386static int
387hv_util_attach(device_t dev)
388{
389	struct hv_device*		hv_dev;
390	struct hv_vmbus_service*	service;
391	int				ret;
392	size_t				receive_buffer_offset;
393
394	hv_dev = vmbus_get_devctx(dev);
395	service = device_get_softc(dev);
396	receive_buffer_offset = service - &service_table[0];
397	device_printf(dev, "Hyper-V Service attaching: %s\n", service->name);
398	receive_buffer[receive_buffer_offset] =
399		malloc(4 * PAGE_SIZE, M_DEVBUF, M_WAITOK | M_ZERO);
400
401	if (service->init != NULL) {
402	    ret = service->init(service);
403	    if (ret) {
404		ret = ENODEV;
405		goto error0;
406	    }
407	}
408
409	ret = hv_vmbus_channel_open(hv_dev->channel, 4 * PAGE_SIZE,
410		    4 * PAGE_SIZE, NULL, 0,
411		    service->callback, hv_dev->channel);
412
413	if (ret)
414	    goto error0;
415
416	return (0);
417
418	error0:
419
420	    free(receive_buffer[receive_buffer_offset], M_DEVBUF);
421	    receive_buffer[receive_buffer_offset] = NULL;
422
423	return (ret);
424}
425
426static int
427hv_util_detach(device_t dev)
428{
429	struct hv_device*		hv_dev;
430	struct hv_vmbus_service*	service;
431	size_t				receive_buffer_offset;
432
433	hv_dev = vmbus_get_devctx(dev);
434
435	hv_vmbus_channel_close(hv_dev->channel);
436	service = device_get_softc(dev);
437	receive_buffer_offset = service - &service_table[0];
438
439	if (service->work_queue != NULL)
440	    hv_work_queue_close(service->work_queue);
441
442	free(receive_buffer[receive_buffer_offset], M_DEVBUF);
443	receive_buffer[receive_buffer_offset] = NULL;
444
445	return (0);
446}
447
448static void hv_util_init(void)
449{
450}
451
452static int hv_util_modevent(module_t mod, int event, void *arg)
453{
454	switch (event) {
455        case MOD_LOAD:
456                break;
457        case MOD_UNLOAD:
458                break;
459	default:
460		break;
461        }
462        return (0);
463}
464
465static device_method_t util_methods[] = {
466	/* Device interface */
467	DEVMETHOD(device_probe, hv_util_probe),
468	DEVMETHOD(device_attach, hv_util_attach),
469	DEVMETHOD(device_detach, hv_util_detach),
470	DEVMETHOD(device_shutdown, bus_generic_shutdown),
471	{ 0, 0 } }
472;
473
474static driver_t util_driver = { "hyperv-utils", util_methods, 0 };
475
476static devclass_t util_devclass;
477
478DRIVER_MODULE(hv_utils, vmbus, util_driver, util_devclass, hv_util_modevent, 0);
479MODULE_VERSION(hv_utils, 1);
480MODULE_DEPEND(hv_utils, vmbus, 1, 1, 1);
481
482SYSINIT(hv_util_initx, SI_SUB_KTHREAD_IDLE, SI_ORDER_MIDDLE + 1,
483	hv_util_init, NULL);
484