1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * (c) 2017 Stefano Stabellini <stefano@aporeto.com>
4  */
5 
6 #include <linux/module.h>
7 #include <linux/net.h>
8 #include <linux/socket.h>
9 
10 #include <net/sock.h>
11 
12 #include <xen/events.h>
13 #include <xen/grant_table.h>
14 #include <xen/xen.h>
15 #include <xen/xenbus.h>
16 #include <xen/interface/io/pvcalls.h>
17 
18 #include "pvcalls-front.h"
19 
20 #define PVCALLS_INVALID_ID UINT_MAX
21 #define PVCALLS_RING_ORDER XENBUS_MAX_RING_GRANT_ORDER
22 #define PVCALLS_NR_RSP_PER_RING __CONST_RING_SIZE(xen_pvcalls, XEN_PAGE_SIZE)
23 #define PVCALLS_FRONT_MAX_SPIN 5000
24 
25 static struct proto pvcalls_proto = {
26 	.name	= "PVCalls",
27 	.owner	= THIS_MODULE,
28 	.obj_size = sizeof(struct sock),
29 };
30 
31 struct pvcalls_bedata {
32 	struct xen_pvcalls_front_ring ring;
33 	grant_ref_t ref;
34 	int irq;
35 
36 	struct list_head socket_mappings;
37 	spinlock_t socket_lock;
38 
39 	wait_queue_head_t inflight_req;
40 	struct xen_pvcalls_response rsp[PVCALLS_NR_RSP_PER_RING];
41 };
42 /* Only one front/back connection supported. */
43 static struct xenbus_device *pvcalls_front_dev;
44 static atomic_t pvcalls_refcount;
45 
46 /* first increment refcount, then proceed */
47 #define pvcalls_enter() {               \
48 	atomic_inc(&pvcalls_refcount);      \
49 }
50 
51 /* first complete other operations, then decrement refcount */
52 #define pvcalls_exit() {                \
53 	atomic_dec(&pvcalls_refcount);      \
54 }
55 
56 struct sock_mapping {
57 	bool active_socket;
58 	struct list_head list;
59 	struct socket *sock;
60 	atomic_t refcount;
61 	union {
62 		struct {
63 			int irq;
64 			grant_ref_t ref;
65 			struct pvcalls_data_intf *ring;
66 			struct pvcalls_data data;
67 			struct mutex in_mutex;
68 			struct mutex out_mutex;
69 
70 			wait_queue_head_t inflight_conn_req;
71 		} active;
72 		struct {
73 		/*
74 		 * Socket status, needs to be 64-bit aligned due to the
75 		 * test_and_* functions which have this requirement on arm64.
76 		 */
77 #define PVCALLS_STATUS_UNINITALIZED  0
78 #define PVCALLS_STATUS_BIND          1
79 #define PVCALLS_STATUS_LISTEN        2
80 			uint8_t status __attribute__((aligned(8)));
81 		/*
82 		 * Internal state-machine flags.
83 		 * Only one accept operation can be inflight for a socket.
84 		 * Only one poll operation can be inflight for a given socket.
85 		 * flags needs to be 64-bit aligned due to the test_and_*
86 		 * functions which have this requirement on arm64.
87 		 */
88 #define PVCALLS_FLAG_ACCEPT_INFLIGHT 0
89 #define PVCALLS_FLAG_POLL_INFLIGHT   1
90 #define PVCALLS_FLAG_POLL_RET        2
91 			uint8_t flags __attribute__((aligned(8)));
92 			uint32_t inflight_req_id;
93 			struct sock_mapping *accept_map;
94 			wait_queue_head_t inflight_accept_req;
95 		} passive;
96 	};
97 };
98 
pvcalls_enter_sock(struct socket * sock)99 static inline struct sock_mapping *pvcalls_enter_sock(struct socket *sock)
100 {
101 	struct sock_mapping *map;
102 
103 	if (!pvcalls_front_dev ||
104 		dev_get_drvdata(&pvcalls_front_dev->dev) == NULL)
105 		return ERR_PTR(-ENOTCONN);
106 
107 	map = (struct sock_mapping *)sock->sk->sk_send_head;
108 	if (map == NULL)
109 		return ERR_PTR(-ENOTSOCK);
110 
111 	pvcalls_enter();
112 	atomic_inc(&map->refcount);
113 	return map;
114 }
115 
pvcalls_exit_sock(struct socket * sock)116 static inline void pvcalls_exit_sock(struct socket *sock)
117 {
118 	struct sock_mapping *map;
119 
120 	map = (struct sock_mapping *)sock->sk->sk_send_head;
121 	atomic_dec(&map->refcount);
122 	pvcalls_exit();
123 }
124 
get_request(struct pvcalls_bedata * bedata,int * req_id)125 static inline int get_request(struct pvcalls_bedata *bedata, int *req_id)
126 {
127 	*req_id = bedata->ring.req_prod_pvt & (RING_SIZE(&bedata->ring) - 1);
128 	if (RING_FULL(&bedata->ring) ||
129 	    bedata->rsp[*req_id].req_id != PVCALLS_INVALID_ID)
130 		return -EAGAIN;
131 	return 0;
132 }
133 
pvcalls_front_write_todo(struct sock_mapping * map)134 static bool pvcalls_front_write_todo(struct sock_mapping *map)
135 {
136 	struct pvcalls_data_intf *intf = map->active.ring;
137 	RING_IDX cons, prod, size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
138 	int32_t error;
139 
140 	error = intf->out_error;
141 	if (error == -ENOTCONN)
142 		return false;
143 	if (error != 0)
144 		return true;
145 
146 	cons = intf->out_cons;
147 	prod = intf->out_prod;
148 	return !!(size - pvcalls_queued(prod, cons, size));
149 }
150 
pvcalls_front_read_todo(struct sock_mapping * map)151 static bool pvcalls_front_read_todo(struct sock_mapping *map)
152 {
153 	struct pvcalls_data_intf *intf = map->active.ring;
154 	RING_IDX cons, prod;
155 	int32_t error;
156 
157 	cons = intf->in_cons;
158 	prod = intf->in_prod;
159 	error = intf->in_error;
160 	return (error != 0 ||
161 		pvcalls_queued(prod, cons,
162 			       XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER)) != 0);
163 }
164 
pvcalls_front_event_handler(int irq,void * dev_id)165 static irqreturn_t pvcalls_front_event_handler(int irq, void *dev_id)
166 {
167 	struct xenbus_device *dev = dev_id;
168 	struct pvcalls_bedata *bedata;
169 	struct xen_pvcalls_response *rsp;
170 	uint8_t *src, *dst;
171 	int req_id = 0, more = 0, done = 0;
172 
173 	if (dev == NULL)
174 		return IRQ_HANDLED;
175 
176 	pvcalls_enter();
177 	bedata = dev_get_drvdata(&dev->dev);
178 	if (bedata == NULL) {
179 		pvcalls_exit();
180 		return IRQ_HANDLED;
181 	}
182 
183 again:
184 	while (RING_HAS_UNCONSUMED_RESPONSES(&bedata->ring)) {
185 		rsp = RING_GET_RESPONSE(&bedata->ring, bedata->ring.rsp_cons);
186 
187 		req_id = rsp->req_id;
188 		if (rsp->cmd == PVCALLS_POLL) {
189 			struct sock_mapping *map = (struct sock_mapping *)(uintptr_t)
190 						   rsp->u.poll.id;
191 
192 			clear_bit(PVCALLS_FLAG_POLL_INFLIGHT,
193 				  (void *)&map->passive.flags);
194 			/*
195 			 * clear INFLIGHT, then set RET. It pairs with
196 			 * the checks at the beginning of
197 			 * pvcalls_front_poll_passive.
198 			 */
199 			smp_wmb();
200 			set_bit(PVCALLS_FLAG_POLL_RET,
201 				(void *)&map->passive.flags);
202 		} else {
203 			dst = (uint8_t *)&bedata->rsp[req_id] +
204 			      sizeof(rsp->req_id);
205 			src = (uint8_t *)rsp + sizeof(rsp->req_id);
206 			memcpy(dst, src, sizeof(*rsp) - sizeof(rsp->req_id));
207 			/*
208 			 * First copy the rest of the data, then req_id. It is
209 			 * paired with the barrier when accessing bedata->rsp.
210 			 */
211 			smp_wmb();
212 			bedata->rsp[req_id].req_id = req_id;
213 		}
214 
215 		done = 1;
216 		bedata->ring.rsp_cons++;
217 	}
218 
219 	RING_FINAL_CHECK_FOR_RESPONSES(&bedata->ring, more);
220 	if (more)
221 		goto again;
222 	if (done)
223 		wake_up(&bedata->inflight_req);
224 	pvcalls_exit();
225 	return IRQ_HANDLED;
226 }
227 
228 static void free_active_ring(struct sock_mapping *map);
229 
pvcalls_front_free_map(struct pvcalls_bedata * bedata,struct sock_mapping * map)230 static void pvcalls_front_free_map(struct pvcalls_bedata *bedata,
231 				   struct sock_mapping *map)
232 {
233 	int i;
234 
235 	unbind_from_irqhandler(map->active.irq, map);
236 
237 	spin_lock(&bedata->socket_lock);
238 	if (!list_empty(&map->list))
239 		list_del_init(&map->list);
240 	spin_unlock(&bedata->socket_lock);
241 
242 	for (i = 0; i < (1 << PVCALLS_RING_ORDER); i++)
243 		gnttab_end_foreign_access(map->active.ring->ref[i], NULL);
244 	gnttab_end_foreign_access(map->active.ref, NULL);
245 	free_active_ring(map);
246 
247 	kfree(map);
248 }
249 
pvcalls_front_conn_handler(int irq,void * sock_map)250 static irqreturn_t pvcalls_front_conn_handler(int irq, void *sock_map)
251 {
252 	struct sock_mapping *map = sock_map;
253 
254 	if (map == NULL)
255 		return IRQ_HANDLED;
256 
257 	wake_up_interruptible(&map->active.inflight_conn_req);
258 
259 	return IRQ_HANDLED;
260 }
261 
pvcalls_front_socket(struct socket * sock)262 int pvcalls_front_socket(struct socket *sock)
263 {
264 	struct pvcalls_bedata *bedata;
265 	struct sock_mapping *map = NULL;
266 	struct xen_pvcalls_request *req;
267 	int notify, req_id, ret;
268 
269 	/*
270 	 * PVCalls only supports domain AF_INET,
271 	 * type SOCK_STREAM and protocol 0 sockets for now.
272 	 *
273 	 * Check socket type here, AF_INET and protocol checks are done
274 	 * by the caller.
275 	 */
276 	if (sock->type != SOCK_STREAM)
277 		return -EOPNOTSUPP;
278 
279 	pvcalls_enter();
280 	if (!pvcalls_front_dev) {
281 		pvcalls_exit();
282 		return -EACCES;
283 	}
284 	bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
285 
286 	map = kzalloc(sizeof(*map), GFP_KERNEL);
287 	if (map == NULL) {
288 		pvcalls_exit();
289 		return -ENOMEM;
290 	}
291 
292 	spin_lock(&bedata->socket_lock);
293 
294 	ret = get_request(bedata, &req_id);
295 	if (ret < 0) {
296 		kfree(map);
297 		spin_unlock(&bedata->socket_lock);
298 		pvcalls_exit();
299 		return ret;
300 	}
301 
302 	/*
303 	 * sock->sk->sk_send_head is not used for ip sockets: reuse the
304 	 * field to store a pointer to the struct sock_mapping
305 	 * corresponding to the socket. This way, we can easily get the
306 	 * struct sock_mapping from the struct socket.
307 	 */
308 	sock->sk->sk_send_head = (void *)map;
309 	list_add_tail(&map->list, &bedata->socket_mappings);
310 
311 	req = RING_GET_REQUEST(&bedata->ring, req_id);
312 	req->req_id = req_id;
313 	req->cmd = PVCALLS_SOCKET;
314 	req->u.socket.id = (uintptr_t) map;
315 	req->u.socket.domain = AF_INET;
316 	req->u.socket.type = SOCK_STREAM;
317 	req->u.socket.protocol = IPPROTO_IP;
318 
319 	bedata->ring.req_prod_pvt++;
320 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
321 	spin_unlock(&bedata->socket_lock);
322 	if (notify)
323 		notify_remote_via_irq(bedata->irq);
324 
325 	wait_event(bedata->inflight_req,
326 		   READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
327 
328 	/* read req_id, then the content */
329 	smp_rmb();
330 	ret = bedata->rsp[req_id].ret;
331 	bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
332 
333 	pvcalls_exit();
334 	return ret;
335 }
336 
free_active_ring(struct sock_mapping * map)337 static void free_active_ring(struct sock_mapping *map)
338 {
339 	if (!map->active.ring)
340 		return;
341 
342 	free_pages_exact(map->active.data.in,
343 			 PAGE_SIZE << map->active.ring->ring_order);
344 	free_page((unsigned long)map->active.ring);
345 }
346 
alloc_active_ring(struct sock_mapping * map)347 static int alloc_active_ring(struct sock_mapping *map)
348 {
349 	void *bytes;
350 
351 	map->active.ring = (struct pvcalls_data_intf *)
352 		get_zeroed_page(GFP_KERNEL);
353 	if (!map->active.ring)
354 		goto out;
355 
356 	map->active.ring->ring_order = PVCALLS_RING_ORDER;
357 	bytes = alloc_pages_exact(PAGE_SIZE << PVCALLS_RING_ORDER,
358 				  GFP_KERNEL | __GFP_ZERO);
359 	if (!bytes)
360 		goto out;
361 
362 	map->active.data.in = bytes;
363 	map->active.data.out = bytes +
364 		XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
365 
366 	return 0;
367 
368 out:
369 	free_active_ring(map);
370 	return -ENOMEM;
371 }
372 
create_active(struct sock_mapping * map,evtchn_port_t * evtchn)373 static int create_active(struct sock_mapping *map, evtchn_port_t *evtchn)
374 {
375 	void *bytes;
376 	int ret, irq = -1, i;
377 
378 	*evtchn = 0;
379 	init_waitqueue_head(&map->active.inflight_conn_req);
380 
381 	bytes = map->active.data.in;
382 	for (i = 0; i < (1 << PVCALLS_RING_ORDER); i++)
383 		map->active.ring->ref[i] = gnttab_grant_foreign_access(
384 			pvcalls_front_dev->otherend_id,
385 			pfn_to_gfn(virt_to_pfn(bytes) + i), 0);
386 
387 	map->active.ref = gnttab_grant_foreign_access(
388 		pvcalls_front_dev->otherend_id,
389 		pfn_to_gfn(virt_to_pfn((void *)map->active.ring)), 0);
390 
391 	ret = xenbus_alloc_evtchn(pvcalls_front_dev, evtchn);
392 	if (ret)
393 		goto out_error;
394 	irq = bind_evtchn_to_irqhandler(*evtchn, pvcalls_front_conn_handler,
395 					0, "pvcalls-frontend", map);
396 	if (irq < 0) {
397 		ret = irq;
398 		goto out_error;
399 	}
400 
401 	map->active.irq = irq;
402 	map->active_socket = true;
403 	mutex_init(&map->active.in_mutex);
404 	mutex_init(&map->active.out_mutex);
405 
406 	return 0;
407 
408 out_error:
409 	if (*evtchn > 0)
410 		xenbus_free_evtchn(pvcalls_front_dev, *evtchn);
411 	return ret;
412 }
413 
pvcalls_front_connect(struct socket * sock,struct sockaddr * addr,int addr_len,int flags)414 int pvcalls_front_connect(struct socket *sock, struct sockaddr *addr,
415 				int addr_len, int flags)
416 {
417 	struct pvcalls_bedata *bedata;
418 	struct sock_mapping *map = NULL;
419 	struct xen_pvcalls_request *req;
420 	int notify, req_id, ret;
421 	evtchn_port_t evtchn;
422 
423 	if (addr->sa_family != AF_INET || sock->type != SOCK_STREAM)
424 		return -EOPNOTSUPP;
425 
426 	map = pvcalls_enter_sock(sock);
427 	if (IS_ERR(map))
428 		return PTR_ERR(map);
429 
430 	bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
431 	ret = alloc_active_ring(map);
432 	if (ret < 0) {
433 		pvcalls_exit_sock(sock);
434 		return ret;
435 	}
436 
437 	spin_lock(&bedata->socket_lock);
438 	ret = get_request(bedata, &req_id);
439 	if (ret < 0) {
440 		spin_unlock(&bedata->socket_lock);
441 		free_active_ring(map);
442 		pvcalls_exit_sock(sock);
443 		return ret;
444 	}
445 	ret = create_active(map, &evtchn);
446 	if (ret < 0) {
447 		spin_unlock(&bedata->socket_lock);
448 		free_active_ring(map);
449 		pvcalls_exit_sock(sock);
450 		return ret;
451 	}
452 
453 	req = RING_GET_REQUEST(&bedata->ring, req_id);
454 	req->req_id = req_id;
455 	req->cmd = PVCALLS_CONNECT;
456 	req->u.connect.id = (uintptr_t)map;
457 	req->u.connect.len = addr_len;
458 	req->u.connect.flags = flags;
459 	req->u.connect.ref = map->active.ref;
460 	req->u.connect.evtchn = evtchn;
461 	memcpy(req->u.connect.addr, addr, sizeof(*addr));
462 
463 	map->sock = sock;
464 
465 	bedata->ring.req_prod_pvt++;
466 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
467 	spin_unlock(&bedata->socket_lock);
468 
469 	if (notify)
470 		notify_remote_via_irq(bedata->irq);
471 
472 	wait_event(bedata->inflight_req,
473 		   READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
474 
475 	/* read req_id, then the content */
476 	smp_rmb();
477 	ret = bedata->rsp[req_id].ret;
478 	bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
479 	pvcalls_exit_sock(sock);
480 	return ret;
481 }
482 
__write_ring(struct pvcalls_data_intf * intf,struct pvcalls_data * data,struct iov_iter * msg_iter,int len)483 static int __write_ring(struct pvcalls_data_intf *intf,
484 			struct pvcalls_data *data,
485 			struct iov_iter *msg_iter,
486 			int len)
487 {
488 	RING_IDX cons, prod, size, masked_prod, masked_cons;
489 	RING_IDX array_size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
490 	int32_t error;
491 
492 	error = intf->out_error;
493 	if (error < 0)
494 		return error;
495 	cons = intf->out_cons;
496 	prod = intf->out_prod;
497 	/* read indexes before continuing */
498 	virt_mb();
499 
500 	size = pvcalls_queued(prod, cons, array_size);
501 	if (size > array_size)
502 		return -EINVAL;
503 	if (size == array_size)
504 		return 0;
505 	if (len > array_size - size)
506 		len = array_size - size;
507 
508 	masked_prod = pvcalls_mask(prod, array_size);
509 	masked_cons = pvcalls_mask(cons, array_size);
510 
511 	if (masked_prod < masked_cons) {
512 		len = copy_from_iter(data->out + masked_prod, len, msg_iter);
513 	} else {
514 		if (len > array_size - masked_prod) {
515 			int ret = copy_from_iter(data->out + masked_prod,
516 				       array_size - masked_prod, msg_iter);
517 			if (ret != array_size - masked_prod) {
518 				len = ret;
519 				goto out;
520 			}
521 			len = ret + copy_from_iter(data->out, len - ret, msg_iter);
522 		} else {
523 			len = copy_from_iter(data->out + masked_prod, len, msg_iter);
524 		}
525 	}
526 out:
527 	/* write to ring before updating pointer */
528 	virt_wmb();
529 	intf->out_prod += len;
530 
531 	return len;
532 }
533 
pvcalls_front_sendmsg(struct socket * sock,struct msghdr * msg,size_t len)534 int pvcalls_front_sendmsg(struct socket *sock, struct msghdr *msg,
535 			  size_t len)
536 {
537 	struct sock_mapping *map;
538 	int sent, tot_sent = 0;
539 	int count = 0, flags;
540 
541 	flags = msg->msg_flags;
542 	if (flags & (MSG_CONFIRM|MSG_DONTROUTE|MSG_EOR|MSG_OOB))
543 		return -EOPNOTSUPP;
544 
545 	map = pvcalls_enter_sock(sock);
546 	if (IS_ERR(map))
547 		return PTR_ERR(map);
548 
549 	mutex_lock(&map->active.out_mutex);
550 	if ((flags & MSG_DONTWAIT) && !pvcalls_front_write_todo(map)) {
551 		mutex_unlock(&map->active.out_mutex);
552 		pvcalls_exit_sock(sock);
553 		return -EAGAIN;
554 	}
555 	if (len > INT_MAX)
556 		len = INT_MAX;
557 
558 again:
559 	count++;
560 	sent = __write_ring(map->active.ring,
561 			    &map->active.data, &msg->msg_iter,
562 			    len);
563 	if (sent > 0) {
564 		len -= sent;
565 		tot_sent += sent;
566 		notify_remote_via_irq(map->active.irq);
567 	}
568 	if (sent >= 0 && len > 0 && count < PVCALLS_FRONT_MAX_SPIN)
569 		goto again;
570 	if (sent < 0)
571 		tot_sent = sent;
572 
573 	mutex_unlock(&map->active.out_mutex);
574 	pvcalls_exit_sock(sock);
575 	return tot_sent;
576 }
577 
__read_ring(struct pvcalls_data_intf * intf,struct pvcalls_data * data,struct iov_iter * msg_iter,size_t len,int flags)578 static int __read_ring(struct pvcalls_data_intf *intf,
579 		       struct pvcalls_data *data,
580 		       struct iov_iter *msg_iter,
581 		       size_t len, int flags)
582 {
583 	RING_IDX cons, prod, size, masked_prod, masked_cons;
584 	RING_IDX array_size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
585 	int32_t error;
586 
587 	cons = intf->in_cons;
588 	prod = intf->in_prod;
589 	error = intf->in_error;
590 	/* get pointers before reading from the ring */
591 	virt_rmb();
592 
593 	size = pvcalls_queued(prod, cons, array_size);
594 	masked_prod = pvcalls_mask(prod, array_size);
595 	masked_cons = pvcalls_mask(cons, array_size);
596 
597 	if (size == 0)
598 		return error ?: size;
599 
600 	if (len > size)
601 		len = size;
602 
603 	if (masked_prod > masked_cons) {
604 		len = copy_to_iter(data->in + masked_cons, len, msg_iter);
605 	} else {
606 		if (len > (array_size - masked_cons)) {
607 			int ret = copy_to_iter(data->in + masked_cons,
608 				     array_size - masked_cons, msg_iter);
609 			if (ret != array_size - masked_cons) {
610 				len = ret;
611 				goto out;
612 			}
613 			len = ret + copy_to_iter(data->in, len - ret, msg_iter);
614 		} else {
615 			len = copy_to_iter(data->in + masked_cons, len, msg_iter);
616 		}
617 	}
618 out:
619 	/* read data from the ring before increasing the index */
620 	virt_mb();
621 	if (!(flags & MSG_PEEK))
622 		intf->in_cons += len;
623 
624 	return len;
625 }
626 
pvcalls_front_recvmsg(struct socket * sock,struct msghdr * msg,size_t len,int flags)627 int pvcalls_front_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
628 		     int flags)
629 {
630 	int ret;
631 	struct sock_mapping *map;
632 
633 	if (flags & (MSG_CMSG_CLOEXEC|MSG_ERRQUEUE|MSG_OOB|MSG_TRUNC))
634 		return -EOPNOTSUPP;
635 
636 	map = pvcalls_enter_sock(sock);
637 	if (IS_ERR(map))
638 		return PTR_ERR(map);
639 
640 	mutex_lock(&map->active.in_mutex);
641 	if (len > XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER))
642 		len = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
643 
644 	while (!(flags & MSG_DONTWAIT) && !pvcalls_front_read_todo(map)) {
645 		wait_event_interruptible(map->active.inflight_conn_req,
646 					 pvcalls_front_read_todo(map));
647 	}
648 	ret = __read_ring(map->active.ring, &map->active.data,
649 			  &msg->msg_iter, len, flags);
650 
651 	if (ret > 0)
652 		notify_remote_via_irq(map->active.irq);
653 	if (ret == 0)
654 		ret = (flags & MSG_DONTWAIT) ? -EAGAIN : 0;
655 	if (ret == -ENOTCONN)
656 		ret = 0;
657 
658 	mutex_unlock(&map->active.in_mutex);
659 	pvcalls_exit_sock(sock);
660 	return ret;
661 }
662 
pvcalls_front_bind(struct socket * sock,struct sockaddr * addr,int addr_len)663 int pvcalls_front_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
664 {
665 	struct pvcalls_bedata *bedata;
666 	struct sock_mapping *map = NULL;
667 	struct xen_pvcalls_request *req;
668 	int notify, req_id, ret;
669 
670 	if (addr->sa_family != AF_INET || sock->type != SOCK_STREAM)
671 		return -EOPNOTSUPP;
672 
673 	map = pvcalls_enter_sock(sock);
674 	if (IS_ERR(map))
675 		return PTR_ERR(map);
676 	bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
677 
678 	spin_lock(&bedata->socket_lock);
679 	ret = get_request(bedata, &req_id);
680 	if (ret < 0) {
681 		spin_unlock(&bedata->socket_lock);
682 		pvcalls_exit_sock(sock);
683 		return ret;
684 	}
685 	req = RING_GET_REQUEST(&bedata->ring, req_id);
686 	req->req_id = req_id;
687 	map->sock = sock;
688 	req->cmd = PVCALLS_BIND;
689 	req->u.bind.id = (uintptr_t)map;
690 	memcpy(req->u.bind.addr, addr, sizeof(*addr));
691 	req->u.bind.len = addr_len;
692 
693 	init_waitqueue_head(&map->passive.inflight_accept_req);
694 
695 	map->active_socket = false;
696 
697 	bedata->ring.req_prod_pvt++;
698 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
699 	spin_unlock(&bedata->socket_lock);
700 	if (notify)
701 		notify_remote_via_irq(bedata->irq);
702 
703 	wait_event(bedata->inflight_req,
704 		   READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
705 
706 	/* read req_id, then the content */
707 	smp_rmb();
708 	ret = bedata->rsp[req_id].ret;
709 	bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
710 
711 	map->passive.status = PVCALLS_STATUS_BIND;
712 	pvcalls_exit_sock(sock);
713 	return 0;
714 }
715 
pvcalls_front_listen(struct socket * sock,int backlog)716 int pvcalls_front_listen(struct socket *sock, int backlog)
717 {
718 	struct pvcalls_bedata *bedata;
719 	struct sock_mapping *map;
720 	struct xen_pvcalls_request *req;
721 	int notify, req_id, ret;
722 
723 	map = pvcalls_enter_sock(sock);
724 	if (IS_ERR(map))
725 		return PTR_ERR(map);
726 	bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
727 
728 	if (map->passive.status != PVCALLS_STATUS_BIND) {
729 		pvcalls_exit_sock(sock);
730 		return -EOPNOTSUPP;
731 	}
732 
733 	spin_lock(&bedata->socket_lock);
734 	ret = get_request(bedata, &req_id);
735 	if (ret < 0) {
736 		spin_unlock(&bedata->socket_lock);
737 		pvcalls_exit_sock(sock);
738 		return ret;
739 	}
740 	req = RING_GET_REQUEST(&bedata->ring, req_id);
741 	req->req_id = req_id;
742 	req->cmd = PVCALLS_LISTEN;
743 	req->u.listen.id = (uintptr_t) map;
744 	req->u.listen.backlog = backlog;
745 
746 	bedata->ring.req_prod_pvt++;
747 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
748 	spin_unlock(&bedata->socket_lock);
749 	if (notify)
750 		notify_remote_via_irq(bedata->irq);
751 
752 	wait_event(bedata->inflight_req,
753 		   READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
754 
755 	/* read req_id, then the content */
756 	smp_rmb();
757 	ret = bedata->rsp[req_id].ret;
758 	bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
759 
760 	map->passive.status = PVCALLS_STATUS_LISTEN;
761 	pvcalls_exit_sock(sock);
762 	return ret;
763 }
764 
pvcalls_front_accept(struct socket * sock,struct socket * newsock,int flags)765 int pvcalls_front_accept(struct socket *sock, struct socket *newsock, int flags)
766 {
767 	struct pvcalls_bedata *bedata;
768 	struct sock_mapping *map;
769 	struct sock_mapping *map2 = NULL;
770 	struct xen_pvcalls_request *req;
771 	int notify, req_id, ret, nonblock;
772 	evtchn_port_t evtchn;
773 
774 	map = pvcalls_enter_sock(sock);
775 	if (IS_ERR(map))
776 		return PTR_ERR(map);
777 	bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
778 
779 	if (map->passive.status != PVCALLS_STATUS_LISTEN) {
780 		pvcalls_exit_sock(sock);
781 		return -EINVAL;
782 	}
783 
784 	nonblock = flags & SOCK_NONBLOCK;
785 	/*
786 	 * Backend only supports 1 inflight accept request, will return
787 	 * errors for the others
788 	 */
789 	if (test_and_set_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
790 			     (void *)&map->passive.flags)) {
791 		req_id = READ_ONCE(map->passive.inflight_req_id);
792 		if (req_id != PVCALLS_INVALID_ID &&
793 		    READ_ONCE(bedata->rsp[req_id].req_id) == req_id) {
794 			map2 = map->passive.accept_map;
795 			goto received;
796 		}
797 		if (nonblock) {
798 			pvcalls_exit_sock(sock);
799 			return -EAGAIN;
800 		}
801 		if (wait_event_interruptible(map->passive.inflight_accept_req,
802 			!test_and_set_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
803 					  (void *)&map->passive.flags))) {
804 			pvcalls_exit_sock(sock);
805 			return -EINTR;
806 		}
807 	}
808 
809 	map2 = kzalloc(sizeof(*map2), GFP_KERNEL);
810 	if (map2 == NULL) {
811 		clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
812 			  (void *)&map->passive.flags);
813 		pvcalls_exit_sock(sock);
814 		return -ENOMEM;
815 	}
816 	ret = alloc_active_ring(map2);
817 	if (ret < 0) {
818 		clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
819 				(void *)&map->passive.flags);
820 		kfree(map2);
821 		pvcalls_exit_sock(sock);
822 		return ret;
823 	}
824 	spin_lock(&bedata->socket_lock);
825 	ret = get_request(bedata, &req_id);
826 	if (ret < 0) {
827 		clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
828 			  (void *)&map->passive.flags);
829 		spin_unlock(&bedata->socket_lock);
830 		free_active_ring(map2);
831 		kfree(map2);
832 		pvcalls_exit_sock(sock);
833 		return ret;
834 	}
835 
836 	ret = create_active(map2, &evtchn);
837 	if (ret < 0) {
838 		free_active_ring(map2);
839 		kfree(map2);
840 		clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
841 			  (void *)&map->passive.flags);
842 		spin_unlock(&bedata->socket_lock);
843 		pvcalls_exit_sock(sock);
844 		return ret;
845 	}
846 	list_add_tail(&map2->list, &bedata->socket_mappings);
847 
848 	req = RING_GET_REQUEST(&bedata->ring, req_id);
849 	req->req_id = req_id;
850 	req->cmd = PVCALLS_ACCEPT;
851 	req->u.accept.id = (uintptr_t) map;
852 	req->u.accept.ref = map2->active.ref;
853 	req->u.accept.id_new = (uintptr_t) map2;
854 	req->u.accept.evtchn = evtchn;
855 	map->passive.accept_map = map2;
856 
857 	bedata->ring.req_prod_pvt++;
858 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
859 	spin_unlock(&bedata->socket_lock);
860 	if (notify)
861 		notify_remote_via_irq(bedata->irq);
862 	/* We could check if we have received a response before returning. */
863 	if (nonblock) {
864 		WRITE_ONCE(map->passive.inflight_req_id, req_id);
865 		pvcalls_exit_sock(sock);
866 		return -EAGAIN;
867 	}
868 
869 	if (wait_event_interruptible(bedata->inflight_req,
870 		READ_ONCE(bedata->rsp[req_id].req_id) == req_id)) {
871 		pvcalls_exit_sock(sock);
872 		return -EINTR;
873 	}
874 	/* read req_id, then the content */
875 	smp_rmb();
876 
877 received:
878 	map2->sock = newsock;
879 	newsock->sk = sk_alloc(sock_net(sock->sk), PF_INET, GFP_KERNEL, &pvcalls_proto, false);
880 	if (!newsock->sk) {
881 		bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
882 		map->passive.inflight_req_id = PVCALLS_INVALID_ID;
883 		clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
884 			  (void *)&map->passive.flags);
885 		pvcalls_front_free_map(bedata, map2);
886 		pvcalls_exit_sock(sock);
887 		return -ENOMEM;
888 	}
889 	newsock->sk->sk_send_head = (void *)map2;
890 
891 	ret = bedata->rsp[req_id].ret;
892 	bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
893 	map->passive.inflight_req_id = PVCALLS_INVALID_ID;
894 
895 	clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT, (void *)&map->passive.flags);
896 	wake_up(&map->passive.inflight_accept_req);
897 
898 	pvcalls_exit_sock(sock);
899 	return ret;
900 }
901 
pvcalls_front_poll_passive(struct file * file,struct pvcalls_bedata * bedata,struct sock_mapping * map,poll_table * wait)902 static __poll_t pvcalls_front_poll_passive(struct file *file,
903 					       struct pvcalls_bedata *bedata,
904 					       struct sock_mapping *map,
905 					       poll_table *wait)
906 {
907 	int notify, req_id, ret;
908 	struct xen_pvcalls_request *req;
909 
910 	if (test_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
911 		     (void *)&map->passive.flags)) {
912 		uint32_t req_id = READ_ONCE(map->passive.inflight_req_id);
913 
914 		if (req_id != PVCALLS_INVALID_ID &&
915 		    READ_ONCE(bedata->rsp[req_id].req_id) == req_id)
916 			return EPOLLIN | EPOLLRDNORM;
917 
918 		poll_wait(file, &map->passive.inflight_accept_req, wait);
919 		return 0;
920 	}
921 
922 	if (test_and_clear_bit(PVCALLS_FLAG_POLL_RET,
923 			       (void *)&map->passive.flags))
924 		return EPOLLIN | EPOLLRDNORM;
925 
926 	/*
927 	 * First check RET, then INFLIGHT. No barriers necessary to
928 	 * ensure execution ordering because of the conditional
929 	 * instructions creating control dependencies.
930 	 */
931 
932 	if (test_and_set_bit(PVCALLS_FLAG_POLL_INFLIGHT,
933 			     (void *)&map->passive.flags)) {
934 		poll_wait(file, &bedata->inflight_req, wait);
935 		return 0;
936 	}
937 
938 	spin_lock(&bedata->socket_lock);
939 	ret = get_request(bedata, &req_id);
940 	if (ret < 0) {
941 		spin_unlock(&bedata->socket_lock);
942 		return ret;
943 	}
944 	req = RING_GET_REQUEST(&bedata->ring, req_id);
945 	req->req_id = req_id;
946 	req->cmd = PVCALLS_POLL;
947 	req->u.poll.id = (uintptr_t) map;
948 
949 	bedata->ring.req_prod_pvt++;
950 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
951 	spin_unlock(&bedata->socket_lock);
952 	if (notify)
953 		notify_remote_via_irq(bedata->irq);
954 
955 	poll_wait(file, &bedata->inflight_req, wait);
956 	return 0;
957 }
958 
pvcalls_front_poll_active(struct file * file,struct pvcalls_bedata * bedata,struct sock_mapping * map,poll_table * wait)959 static __poll_t pvcalls_front_poll_active(struct file *file,
960 					      struct pvcalls_bedata *bedata,
961 					      struct sock_mapping *map,
962 					      poll_table *wait)
963 {
964 	__poll_t mask = 0;
965 	int32_t in_error, out_error;
966 	struct pvcalls_data_intf *intf = map->active.ring;
967 
968 	out_error = intf->out_error;
969 	in_error = intf->in_error;
970 
971 	poll_wait(file, &map->active.inflight_conn_req, wait);
972 	if (pvcalls_front_write_todo(map))
973 		mask |= EPOLLOUT | EPOLLWRNORM;
974 	if (pvcalls_front_read_todo(map))
975 		mask |= EPOLLIN | EPOLLRDNORM;
976 	if (in_error != 0 || out_error != 0)
977 		mask |= EPOLLERR;
978 
979 	return mask;
980 }
981 
pvcalls_front_poll(struct file * file,struct socket * sock,poll_table * wait)982 __poll_t pvcalls_front_poll(struct file *file, struct socket *sock,
983 			       poll_table *wait)
984 {
985 	struct pvcalls_bedata *bedata;
986 	struct sock_mapping *map;
987 	__poll_t ret;
988 
989 	map = pvcalls_enter_sock(sock);
990 	if (IS_ERR(map))
991 		return EPOLLNVAL;
992 	bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
993 
994 	if (map->active_socket)
995 		ret = pvcalls_front_poll_active(file, bedata, map, wait);
996 	else
997 		ret = pvcalls_front_poll_passive(file, bedata, map, wait);
998 	pvcalls_exit_sock(sock);
999 	return ret;
1000 }
1001 
pvcalls_front_release(struct socket * sock)1002 int pvcalls_front_release(struct socket *sock)
1003 {
1004 	struct pvcalls_bedata *bedata;
1005 	struct sock_mapping *map;
1006 	int req_id, notify, ret;
1007 	struct xen_pvcalls_request *req;
1008 
1009 	if (sock->sk == NULL)
1010 		return 0;
1011 
1012 	map = pvcalls_enter_sock(sock);
1013 	if (IS_ERR(map)) {
1014 		if (PTR_ERR(map) == -ENOTCONN)
1015 			return -EIO;
1016 		else
1017 			return 0;
1018 	}
1019 	bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
1020 
1021 	spin_lock(&bedata->socket_lock);
1022 	ret = get_request(bedata, &req_id);
1023 	if (ret < 0) {
1024 		spin_unlock(&bedata->socket_lock);
1025 		pvcalls_exit_sock(sock);
1026 		return ret;
1027 	}
1028 	sock->sk->sk_send_head = NULL;
1029 
1030 	req = RING_GET_REQUEST(&bedata->ring, req_id);
1031 	req->req_id = req_id;
1032 	req->cmd = PVCALLS_RELEASE;
1033 	req->u.release.id = (uintptr_t)map;
1034 
1035 	bedata->ring.req_prod_pvt++;
1036 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
1037 	spin_unlock(&bedata->socket_lock);
1038 	if (notify)
1039 		notify_remote_via_irq(bedata->irq);
1040 
1041 	wait_event(bedata->inflight_req,
1042 		   READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
1043 
1044 	if (map->active_socket) {
1045 		/*
1046 		 * Set in_error and wake up inflight_conn_req to force
1047 		 * recvmsg waiters to exit.
1048 		 */
1049 		map->active.ring->in_error = -EBADF;
1050 		wake_up_interruptible(&map->active.inflight_conn_req);
1051 
1052 		/*
1053 		 * We need to make sure that sendmsg/recvmsg on this socket have
1054 		 * not started before we've cleared sk_send_head here. The
1055 		 * easiest way to guarantee this is to see that no pvcalls
1056 		 * (other than us) is in progress on this socket.
1057 		 */
1058 		while (atomic_read(&map->refcount) > 1)
1059 			cpu_relax();
1060 
1061 		pvcalls_front_free_map(bedata, map);
1062 	} else {
1063 		wake_up(&bedata->inflight_req);
1064 		wake_up(&map->passive.inflight_accept_req);
1065 
1066 		while (atomic_read(&map->refcount) > 1)
1067 			cpu_relax();
1068 
1069 		spin_lock(&bedata->socket_lock);
1070 		list_del(&map->list);
1071 		spin_unlock(&bedata->socket_lock);
1072 		if (READ_ONCE(map->passive.inflight_req_id) != PVCALLS_INVALID_ID &&
1073 			READ_ONCE(map->passive.inflight_req_id) != 0) {
1074 			pvcalls_front_free_map(bedata,
1075 					       map->passive.accept_map);
1076 		}
1077 		kfree(map);
1078 	}
1079 	WRITE_ONCE(bedata->rsp[req_id].req_id, PVCALLS_INVALID_ID);
1080 
1081 	pvcalls_exit();
1082 	return 0;
1083 }
1084 
1085 static const struct xenbus_device_id pvcalls_front_ids[] = {
1086 	{ "pvcalls" },
1087 	{ "" }
1088 };
1089 
pvcalls_front_remove(struct xenbus_device * dev)1090 static int pvcalls_front_remove(struct xenbus_device *dev)
1091 {
1092 	struct pvcalls_bedata *bedata;
1093 	struct sock_mapping *map = NULL, *n;
1094 
1095 	bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
1096 	dev_set_drvdata(&dev->dev, NULL);
1097 	pvcalls_front_dev = NULL;
1098 	if (bedata->irq >= 0)
1099 		unbind_from_irqhandler(bedata->irq, dev);
1100 
1101 	list_for_each_entry_safe(map, n, &bedata->socket_mappings, list) {
1102 		map->sock->sk->sk_send_head = NULL;
1103 		if (map->active_socket) {
1104 			map->active.ring->in_error = -EBADF;
1105 			wake_up_interruptible(&map->active.inflight_conn_req);
1106 		}
1107 	}
1108 
1109 	smp_mb();
1110 	while (atomic_read(&pvcalls_refcount) > 0)
1111 		cpu_relax();
1112 	list_for_each_entry_safe(map, n, &bedata->socket_mappings, list) {
1113 		if (map->active_socket) {
1114 			/* No need to lock, refcount is 0 */
1115 			pvcalls_front_free_map(bedata, map);
1116 		} else {
1117 			list_del(&map->list);
1118 			kfree(map);
1119 		}
1120 	}
1121 	if (bedata->ref != -1)
1122 		gnttab_end_foreign_access(bedata->ref, NULL);
1123 	kfree(bedata->ring.sring);
1124 	kfree(bedata);
1125 	xenbus_switch_state(dev, XenbusStateClosed);
1126 	return 0;
1127 }
1128 
pvcalls_front_probe(struct xenbus_device * dev,const struct xenbus_device_id * id)1129 static int pvcalls_front_probe(struct xenbus_device *dev,
1130 			  const struct xenbus_device_id *id)
1131 {
1132 	int ret = -ENOMEM, i;
1133 	evtchn_port_t evtchn;
1134 	unsigned int max_page_order, function_calls, len;
1135 	char *versions;
1136 	grant_ref_t gref_head = 0;
1137 	struct xenbus_transaction xbt;
1138 	struct pvcalls_bedata *bedata = NULL;
1139 	struct xen_pvcalls_sring *sring;
1140 
1141 	if (pvcalls_front_dev != NULL) {
1142 		dev_err(&dev->dev, "only one PV Calls connection supported\n");
1143 		return -EINVAL;
1144 	}
1145 
1146 	versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
1147 	if (IS_ERR(versions))
1148 		return PTR_ERR(versions);
1149 	if (!len)
1150 		return -EINVAL;
1151 	if (strcmp(versions, "1")) {
1152 		kfree(versions);
1153 		return -EINVAL;
1154 	}
1155 	kfree(versions);
1156 	max_page_order = xenbus_read_unsigned(dev->otherend,
1157 					      "max-page-order", 0);
1158 	if (max_page_order < PVCALLS_RING_ORDER)
1159 		return -ENODEV;
1160 	function_calls = xenbus_read_unsigned(dev->otherend,
1161 					      "function-calls", 0);
1162 	/* See XENBUS_FUNCTIONS_CALLS in pvcalls.h */
1163 	if (function_calls != 1)
1164 		return -ENODEV;
1165 	pr_info("%s max-page-order is %u\n", __func__, max_page_order);
1166 
1167 	bedata = kzalloc(sizeof(struct pvcalls_bedata), GFP_KERNEL);
1168 	if (!bedata)
1169 		return -ENOMEM;
1170 
1171 	dev_set_drvdata(&dev->dev, bedata);
1172 	pvcalls_front_dev = dev;
1173 	init_waitqueue_head(&bedata->inflight_req);
1174 	INIT_LIST_HEAD(&bedata->socket_mappings);
1175 	spin_lock_init(&bedata->socket_lock);
1176 	bedata->irq = -1;
1177 	bedata->ref = -1;
1178 
1179 	for (i = 0; i < PVCALLS_NR_RSP_PER_RING; i++)
1180 		bedata->rsp[i].req_id = PVCALLS_INVALID_ID;
1181 
1182 	sring = (struct xen_pvcalls_sring *) __get_free_page(GFP_KERNEL |
1183 							     __GFP_ZERO);
1184 	if (!sring)
1185 		goto error;
1186 	SHARED_RING_INIT(sring);
1187 	FRONT_RING_INIT(&bedata->ring, sring, XEN_PAGE_SIZE);
1188 
1189 	ret = xenbus_alloc_evtchn(dev, &evtchn);
1190 	if (ret)
1191 		goto error;
1192 
1193 	bedata->irq = bind_evtchn_to_irqhandler(evtchn,
1194 						pvcalls_front_event_handler,
1195 						0, "pvcalls-frontend", dev);
1196 	if (bedata->irq < 0) {
1197 		ret = bedata->irq;
1198 		goto error;
1199 	}
1200 
1201 	ret = gnttab_alloc_grant_references(1, &gref_head);
1202 	if (ret < 0)
1203 		goto error;
1204 	ret = gnttab_claim_grant_reference(&gref_head);
1205 	if (ret < 0)
1206 		goto error;
1207 	bedata->ref = ret;
1208 	gnttab_grant_foreign_access_ref(bedata->ref, dev->otherend_id,
1209 					virt_to_gfn((void *)sring), 0);
1210 
1211  again:
1212 	ret = xenbus_transaction_start(&xbt);
1213 	if (ret) {
1214 		xenbus_dev_fatal(dev, ret, "starting transaction");
1215 		goto error;
1216 	}
1217 	ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
1218 	if (ret)
1219 		goto error_xenbus;
1220 	ret = xenbus_printf(xbt, dev->nodename, "ring-ref", "%d", bedata->ref);
1221 	if (ret)
1222 		goto error_xenbus;
1223 	ret = xenbus_printf(xbt, dev->nodename, "port", "%u",
1224 			    evtchn);
1225 	if (ret)
1226 		goto error_xenbus;
1227 	ret = xenbus_transaction_end(xbt, 0);
1228 	if (ret) {
1229 		if (ret == -EAGAIN)
1230 			goto again;
1231 		xenbus_dev_fatal(dev, ret, "completing transaction");
1232 		goto error;
1233 	}
1234 	xenbus_switch_state(dev, XenbusStateInitialised);
1235 
1236 	return 0;
1237 
1238  error_xenbus:
1239 	xenbus_transaction_end(xbt, 1);
1240 	xenbus_dev_fatal(dev, ret, "writing xenstore");
1241  error:
1242 	pvcalls_front_remove(dev);
1243 	return ret;
1244 }
1245 
pvcalls_front_changed(struct xenbus_device * dev,enum xenbus_state backend_state)1246 static void pvcalls_front_changed(struct xenbus_device *dev,
1247 			    enum xenbus_state backend_state)
1248 {
1249 	switch (backend_state) {
1250 	case XenbusStateReconfiguring:
1251 	case XenbusStateReconfigured:
1252 	case XenbusStateInitialising:
1253 	case XenbusStateInitialised:
1254 	case XenbusStateUnknown:
1255 		break;
1256 
1257 	case XenbusStateInitWait:
1258 		break;
1259 
1260 	case XenbusStateConnected:
1261 		xenbus_switch_state(dev, XenbusStateConnected);
1262 		break;
1263 
1264 	case XenbusStateClosed:
1265 		if (dev->state == XenbusStateClosed)
1266 			break;
1267 		/* Missed the backend's CLOSING state */
1268 		fallthrough;
1269 	case XenbusStateClosing:
1270 		xenbus_frontend_closed(dev);
1271 		break;
1272 	}
1273 }
1274 
1275 static struct xenbus_driver pvcalls_front_driver = {
1276 	.ids = pvcalls_front_ids,
1277 	.probe = pvcalls_front_probe,
1278 	.remove = pvcalls_front_remove,
1279 	.otherend_changed = pvcalls_front_changed,
1280 	.not_essential = true,
1281 };
1282 
pvcalls_frontend_init(void)1283 static int __init pvcalls_frontend_init(void)
1284 {
1285 	if (!xen_domain())
1286 		return -ENODEV;
1287 
1288 	pr_info("Initialising Xen pvcalls frontend driver\n");
1289 
1290 	return xenbus_register_frontend(&pvcalls_front_driver);
1291 }
1292 
1293 module_init(pvcalls_frontend_init);
1294 
1295 MODULE_DESCRIPTION("Xen PV Calls frontend driver");
1296 MODULE_AUTHOR("Stefano Stabellini <sstabellini@kernel.org>");
1297 MODULE_LICENSE("GPL");
1298