xref: /linux/drivers/hv/hv_snapshot.c (revision 4f2c0a4acffbec01079c28f839422e64ddeff004)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * An implementation of host initiated guest snapshot.
4  *
5  * Copyright (C) 2013, Microsoft, Inc.
6  * Author : K. Y. Srinivasan <kys@microsoft.com>
7  */
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 
10 #include <linux/net.h>
11 #include <linux/nls.h>
12 #include <linux/connector.h>
13 #include <linux/workqueue.h>
14 #include <linux/hyperv.h>
15 #include <asm/hyperv-tlfs.h>
16 
17 #include "hyperv_vmbus.h"
18 #include "hv_utils_transport.h"
19 
20 #define VSS_MAJOR  5
21 #define VSS_MINOR  0
22 #define VSS_VERSION    (VSS_MAJOR << 16 | VSS_MINOR)
23 
24 #define VSS_VER_COUNT 1
25 static const int vss_versions[] = {
26 	VSS_VERSION
27 };
28 
29 #define FW_VER_COUNT 1
30 static const int fw_versions[] = {
31 	UTIL_FW_VERSION
32 };
33 
34 /* See comment with struct hv_vss_msg regarding the max VMbus packet size */
35 #define VSS_MAX_PKT_SIZE (HV_HYP_PAGE_SIZE * 2)
36 
37 /*
38  * Timeout values are based on expecations from host
39  */
40 #define VSS_FREEZE_TIMEOUT (15 * 60)
41 
42 /*
43  * Global state maintained for transaction that is being processed. For a class
44  * of integration services, including the "VSS service", the specified protocol
45  * is a "request/response" protocol which means that there can only be single
46  * outstanding transaction from the host at any given point in time. We use
47  * this to simplify memory management in this driver - we cache and process
48  * only one message at a time.
49  *
50  * While the request/response protocol is guaranteed by the host, we further
51  * ensure this by serializing packet processing in this driver - we do not
52  * read additional packets from the VMBUs until the current packet is fully
53  * handled.
54  */
55 
56 static struct {
57 	int state;   /* hvutil_device_state */
58 	int recv_len; /* number of bytes received. */
59 	struct vmbus_channel *recv_channel; /* chn we got the request */
60 	u64 recv_req_id; /* request ID. */
61 	struct hv_vss_msg  *msg; /* current message */
62 } vss_transaction;
63 
64 
65 static void vss_respond_to_host(int error);
66 
67 /*
68  * This state maintains the version number registered by the daemon.
69  */
70 static int dm_reg_value;
71 
72 static const char vss_devname[] = "vmbus/hv_vss";
73 static __u8 *recv_buffer;
74 static struct hvutil_transport *hvt;
75 
76 static void vss_timeout_func(struct work_struct *dummy);
77 static void vss_handle_request(struct work_struct *dummy);
78 
79 static DECLARE_DELAYED_WORK(vss_timeout_work, vss_timeout_func);
80 static DECLARE_WORK(vss_handle_request_work, vss_handle_request);
81 
vss_poll_wrapper(void * channel)82 static void vss_poll_wrapper(void *channel)
83 {
84 	/* Transaction is finished, reset the state here to avoid races. */
85 	vss_transaction.state = HVUTIL_READY;
86 	tasklet_schedule(&((struct vmbus_channel *)channel)->callback_event);
87 }
88 
89 /*
90  * Callback when data is received from user mode.
91  */
92 
vss_timeout_func(struct work_struct * dummy)93 static void vss_timeout_func(struct work_struct *dummy)
94 {
95 	/*
96 	 * Timeout waiting for userspace component to reply happened.
97 	 */
98 	pr_warn("VSS: timeout waiting for daemon to reply\n");
99 	vss_respond_to_host(HV_E_FAIL);
100 
101 	hv_poll_channel(vss_transaction.recv_channel, vss_poll_wrapper);
102 }
103 
vss_register_done(void)104 static void vss_register_done(void)
105 {
106 	hv_poll_channel(vss_transaction.recv_channel, vss_poll_wrapper);
107 	pr_debug("VSS: userspace daemon registered\n");
108 }
109 
vss_handle_handshake(struct hv_vss_msg * vss_msg)110 static int vss_handle_handshake(struct hv_vss_msg *vss_msg)
111 {
112 	u32 our_ver = VSS_OP_REGISTER1;
113 
114 	switch (vss_msg->vss_hdr.operation) {
115 	case VSS_OP_REGISTER:
116 		/* Daemon doesn't expect us to reply */
117 		dm_reg_value = VSS_OP_REGISTER;
118 		break;
119 	case VSS_OP_REGISTER1:
120 		/* Daemon expects us to reply with our own version */
121 		if (hvutil_transport_send(hvt, &our_ver, sizeof(our_ver),
122 					  vss_register_done))
123 			return -EFAULT;
124 		dm_reg_value = VSS_OP_REGISTER1;
125 		break;
126 	default:
127 		return -EINVAL;
128 	}
129 	pr_info("VSS: userspace daemon ver. %d connected\n", dm_reg_value);
130 	return 0;
131 }
132 
vss_on_msg(void * msg,int len)133 static int vss_on_msg(void *msg, int len)
134 {
135 	struct hv_vss_msg *vss_msg = (struct hv_vss_msg *)msg;
136 
137 	if (len != sizeof(*vss_msg)) {
138 		pr_debug("VSS: Message size does not match length\n");
139 		return -EINVAL;
140 	}
141 
142 	if (vss_msg->vss_hdr.operation == VSS_OP_REGISTER ||
143 	    vss_msg->vss_hdr.operation == VSS_OP_REGISTER1) {
144 		/*
145 		 * Don't process registration messages if we're in the middle
146 		 * of a transaction processing.
147 		 */
148 		if (vss_transaction.state > HVUTIL_READY) {
149 			pr_debug("VSS: Got unexpected registration request\n");
150 			return -EINVAL;
151 		}
152 
153 		return vss_handle_handshake(vss_msg);
154 	} else if (vss_transaction.state == HVUTIL_USERSPACE_REQ) {
155 		vss_transaction.state = HVUTIL_USERSPACE_RECV;
156 
157 		if (vss_msg->vss_hdr.operation == VSS_OP_HOT_BACKUP)
158 			vss_transaction.msg->vss_cf.flags =
159 				VSS_HBU_NO_AUTO_RECOVERY;
160 
161 		if (cancel_delayed_work_sync(&vss_timeout_work)) {
162 			vss_respond_to_host(vss_msg->error);
163 			/* Transaction is finished, reset the state. */
164 			hv_poll_channel(vss_transaction.recv_channel,
165 					vss_poll_wrapper);
166 		}
167 	} else {
168 		/* This is a spurious call! */
169 		pr_debug("VSS: Transaction not active\n");
170 		return -EINVAL;
171 	}
172 	return 0;
173 }
174 
vss_send_op(void)175 static void vss_send_op(void)
176 {
177 	int op = vss_transaction.msg->vss_hdr.operation;
178 	int rc;
179 	struct hv_vss_msg *vss_msg;
180 
181 	/* The transaction state is wrong. */
182 	if (vss_transaction.state != HVUTIL_HOSTMSG_RECEIVED) {
183 		pr_debug("VSS: Unexpected attempt to send to daemon\n");
184 		return;
185 	}
186 
187 	vss_msg = kzalloc(sizeof(*vss_msg), GFP_KERNEL);
188 	if (!vss_msg)
189 		return;
190 
191 	vss_msg->vss_hdr.operation = op;
192 
193 	vss_transaction.state = HVUTIL_USERSPACE_REQ;
194 
195 	schedule_delayed_work(&vss_timeout_work, op == VSS_OP_FREEZE ?
196 			VSS_FREEZE_TIMEOUT * HZ : HV_UTIL_TIMEOUT * HZ);
197 
198 	rc = hvutil_transport_send(hvt, vss_msg, sizeof(*vss_msg), NULL);
199 	if (rc) {
200 		pr_warn("VSS: failed to communicate to the daemon: %d\n", rc);
201 		if (cancel_delayed_work_sync(&vss_timeout_work)) {
202 			vss_respond_to_host(HV_E_FAIL);
203 			vss_transaction.state = HVUTIL_READY;
204 		}
205 	}
206 
207 	kfree(vss_msg);
208 }
209 
vss_handle_request(struct work_struct * dummy)210 static void vss_handle_request(struct work_struct *dummy)
211 {
212 	switch (vss_transaction.msg->vss_hdr.operation) {
213 	/*
214 	 * Initiate a "freeze/thaw" operation in the guest.
215 	 * We respond to the host once the operation is complete.
216 	 *
217 	 * We send the message to the user space daemon and the operation is
218 	 * performed in the daemon.
219 	 */
220 	case VSS_OP_THAW:
221 	case VSS_OP_FREEZE:
222 	case VSS_OP_HOT_BACKUP:
223 		if (vss_transaction.state < HVUTIL_READY) {
224 			/* Userspace is not registered yet */
225 			pr_debug("VSS: Not ready for request.\n");
226 			vss_respond_to_host(HV_E_FAIL);
227 			return;
228 		}
229 
230 		pr_debug("VSS: Received request for op code: %d\n",
231 			vss_transaction.msg->vss_hdr.operation);
232 		vss_transaction.state = HVUTIL_HOSTMSG_RECEIVED;
233 		vss_send_op();
234 		return;
235 	case VSS_OP_GET_DM_INFO:
236 		vss_transaction.msg->dm_info.flags = 0;
237 		break;
238 	default:
239 		break;
240 	}
241 
242 	vss_respond_to_host(0);
243 	hv_poll_channel(vss_transaction.recv_channel, vss_poll_wrapper);
244 }
245 
246 /*
247  * Send a response back to the host.
248  */
249 
250 static void
vss_respond_to_host(int error)251 vss_respond_to_host(int error)
252 {
253 	struct icmsg_hdr *icmsghdrp;
254 	u32	buf_len;
255 	struct vmbus_channel *channel;
256 	u64	req_id;
257 
258 	/*
259 	 * Copy the global state for completing the transaction. Note that
260 	 * only one transaction can be active at a time.
261 	 */
262 
263 	buf_len = vss_transaction.recv_len;
264 	channel = vss_transaction.recv_channel;
265 	req_id = vss_transaction.recv_req_id;
266 
267 	icmsghdrp = (struct icmsg_hdr *)
268 			&recv_buffer[sizeof(struct vmbuspipe_hdr)];
269 
270 	if (channel->onchannel_callback == NULL)
271 		/*
272 		 * We have raced with util driver being unloaded;
273 		 * silently return.
274 		 */
275 		return;
276 
277 	icmsghdrp->status = error;
278 
279 	icmsghdrp->icflags = ICMSGHDRFLAG_TRANSACTION | ICMSGHDRFLAG_RESPONSE;
280 
281 	vmbus_sendpacket(channel, recv_buffer, buf_len, req_id,
282 				VM_PKT_DATA_INBAND, 0);
283 
284 }
285 
286 /*
287  * This callback is invoked when we get a VSS message from the host.
288  * The host ensures that only one VSS transaction can be active at a time.
289  */
290 
hv_vss_onchannelcallback(void * context)291 void hv_vss_onchannelcallback(void *context)
292 {
293 	struct vmbus_channel *channel = context;
294 	u32 recvlen;
295 	u64 requestid;
296 	struct hv_vss_msg *vss_msg;
297 	int vss_srv_version;
298 
299 	struct icmsg_hdr *icmsghdrp;
300 
301 	if (vss_transaction.state > HVUTIL_READY)
302 		return;
303 
304 	if (vmbus_recvpacket(channel, recv_buffer, VSS_MAX_PKT_SIZE, &recvlen, &requestid)) {
305 		pr_err_ratelimited("VSS request received. Could not read into recv buf\n");
306 		return;
307 	}
308 
309 	if (!recvlen)
310 		return;
311 
312 	/* Ensure recvlen is big enough to read header data */
313 	if (recvlen < ICMSG_HDR) {
314 		pr_err_ratelimited("VSS request received. Packet length too small: %d\n",
315 				   recvlen);
316 		return;
317 	}
318 
319 	icmsghdrp = (struct icmsg_hdr *)&recv_buffer[sizeof(struct vmbuspipe_hdr)];
320 
321 	if (icmsghdrp->icmsgtype == ICMSGTYPE_NEGOTIATE) {
322 		if (vmbus_prep_negotiate_resp(icmsghdrp,
323 				recv_buffer, recvlen,
324 				fw_versions, FW_VER_COUNT,
325 				vss_versions, VSS_VER_COUNT,
326 				NULL, &vss_srv_version)) {
327 
328 			pr_info("VSS IC version %d.%d\n",
329 				vss_srv_version >> 16,
330 				vss_srv_version & 0xFFFF);
331 		}
332 	} else if (icmsghdrp->icmsgtype == ICMSGTYPE_VSS) {
333 		/* Ensure recvlen is big enough to contain hv_vss_msg */
334 		if (recvlen < ICMSG_HDR + sizeof(struct hv_vss_msg)) {
335 			pr_err_ratelimited("Invalid VSS msg. Packet length too small: %u\n",
336 					   recvlen);
337 			return;
338 		}
339 		vss_msg = (struct hv_vss_msg *)&recv_buffer[ICMSG_HDR];
340 
341 		/*
342 		 * Stash away this global state for completing the
343 		 * transaction; note transactions are serialized.
344 		 */
345 
346 		vss_transaction.recv_len = recvlen;
347 		vss_transaction.recv_req_id = requestid;
348 		vss_transaction.msg = (struct hv_vss_msg *)vss_msg;
349 
350 		schedule_work(&vss_handle_request_work);
351 		return;
352 	} else {
353 		pr_err_ratelimited("VSS request received. Invalid msg type: %d\n",
354 				   icmsghdrp->icmsgtype);
355 		return;
356 	}
357 
358 	icmsghdrp->icflags = ICMSGHDRFLAG_TRANSACTION |
359 		ICMSGHDRFLAG_RESPONSE;
360 	vmbus_sendpacket(channel, recv_buffer, recvlen, requestid,
361 			 VM_PKT_DATA_INBAND, 0);
362 }
363 
vss_on_reset(void)364 static void vss_on_reset(void)
365 {
366 	if (cancel_delayed_work_sync(&vss_timeout_work))
367 		vss_respond_to_host(HV_E_FAIL);
368 	vss_transaction.state = HVUTIL_DEVICE_INIT;
369 }
370 
371 int
hv_vss_init(struct hv_util_service * srv)372 hv_vss_init(struct hv_util_service *srv)
373 {
374 	if (vmbus_proto_version < VERSION_WIN8_1) {
375 		pr_warn("Integration service 'Backup (volume snapshot)'"
376 			" not supported on this host version.\n");
377 		return -ENOTSUPP;
378 	}
379 	recv_buffer = srv->recv_buffer;
380 	vss_transaction.recv_channel = srv->channel;
381 	vss_transaction.recv_channel->max_pkt_size = VSS_MAX_PKT_SIZE;
382 
383 	/*
384 	 * When this driver loads, the user level daemon that
385 	 * processes the host requests may not yet be running.
386 	 * Defer processing channel callbacks until the daemon
387 	 * has registered.
388 	 */
389 	vss_transaction.state = HVUTIL_DEVICE_INIT;
390 
391 	hvt = hvutil_transport_init(vss_devname, CN_VSS_IDX, CN_VSS_VAL,
392 				    vss_on_msg, vss_on_reset);
393 	if (!hvt) {
394 		pr_warn("VSS: Failed to initialize transport\n");
395 		return -EFAULT;
396 	}
397 
398 	return 0;
399 }
400 
hv_vss_cancel_work(void)401 static void hv_vss_cancel_work(void)
402 {
403 	cancel_delayed_work_sync(&vss_timeout_work);
404 	cancel_work_sync(&vss_handle_request_work);
405 }
406 
hv_vss_pre_suspend(void)407 int hv_vss_pre_suspend(void)
408 {
409 	struct vmbus_channel *channel = vss_transaction.recv_channel;
410 	struct hv_vss_msg *vss_msg;
411 
412 	/*
413 	 * Fake a THAW message for the user space daemon in case the daemon
414 	 * has frozen the file systems. It doesn't matter if there is already
415 	 * a message pending to be delivered to the user space since we force
416 	 * vss_transaction.state to be HVUTIL_READY, so the user space daemon's
417 	 * write() will fail with EINVAL (see vss_on_msg()), and the daemon
418 	 * will reset the device by closing and re-opening it.
419 	 */
420 	vss_msg = kzalloc(sizeof(*vss_msg), GFP_KERNEL);
421 	if (!vss_msg)
422 		return -ENOMEM;
423 
424 	tasklet_disable(&channel->callback_event);
425 
426 	vss_msg->vss_hdr.operation = VSS_OP_THAW;
427 
428 	/* Cancel any possible pending work. */
429 	hv_vss_cancel_work();
430 
431 	/* We don't care about the return value. */
432 	hvutil_transport_send(hvt, vss_msg, sizeof(*vss_msg), NULL);
433 
434 	kfree(vss_msg);
435 
436 	vss_transaction.state = HVUTIL_READY;
437 
438 	/* tasklet_enable() will be called in hv_vss_pre_resume(). */
439 	return 0;
440 }
441 
hv_vss_pre_resume(void)442 int hv_vss_pre_resume(void)
443 {
444 	struct vmbus_channel *channel = vss_transaction.recv_channel;
445 
446 	tasklet_enable(&channel->callback_event);
447 
448 	return 0;
449 }
450 
hv_vss_deinit(void)451 void hv_vss_deinit(void)
452 {
453 	vss_transaction.state = HVUTIL_DEVICE_DYING;
454 
455 	hv_vss_cancel_work();
456 
457 	hvutil_transport_destroy(hvt);
458 }
459