xref: /openbmc/linux/drivers/xen/pvcalls-front.c (revision 9977a8c3497a8f7f7f951994f298a8e4d961234f)
1 /*
2  * (c) 2017 Stefano Stabellini <stefano@aporeto.com>
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License as published by
6  * the Free Software Foundation; either version 2 of the License, or
7  * (at your option) any later version.
8  *
9  * This program is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  */
14 
15 #include <linux/module.h>
16 #include <linux/net.h>
17 #include <linux/socket.h>
18 
19 #include <net/sock.h>
20 
21 #include <xen/events.h>
22 #include <xen/grant_table.h>
23 #include <xen/xen.h>
24 #include <xen/xenbus.h>
25 #include <xen/interface/io/pvcalls.h>
26 
27 #include "pvcalls-front.h"
28 
29 #define PVCALLS_INVALID_ID UINT_MAX
30 #define PVCALLS_RING_ORDER XENBUS_MAX_RING_GRANT_ORDER
31 #define PVCALLS_NR_RSP_PER_RING __CONST_RING_SIZE(xen_pvcalls, XEN_PAGE_SIZE)
32 #define PVCALLS_FRONT_MAX_SPIN 5000
33 
34 struct pvcalls_bedata {
35 	struct xen_pvcalls_front_ring ring;
36 	grant_ref_t ref;
37 	int irq;
38 
39 	struct list_head socket_mappings;
40 	spinlock_t socket_lock;
41 
42 	wait_queue_head_t inflight_req;
43 	struct xen_pvcalls_response rsp[PVCALLS_NR_RSP_PER_RING];
44 };
45 /* Only one front/back connection supported. */
46 static struct xenbus_device *pvcalls_front_dev;
47 static atomic_t pvcalls_refcount;
48 
49 /* first increment refcount, then proceed */
50 #define pvcalls_enter() {               \
51 	atomic_inc(&pvcalls_refcount);      \
52 }
53 
54 /* first complete other operations, then decrement refcount */
55 #define pvcalls_exit() {                \
56 	atomic_dec(&pvcalls_refcount);      \
57 }
58 
59 struct sock_mapping {
60 	bool active_socket;
61 	struct list_head list;
62 	struct socket *sock;
63 	atomic_t refcount;
64 	union {
65 		struct {
66 			int irq;
67 			grant_ref_t ref;
68 			struct pvcalls_data_intf *ring;
69 			struct pvcalls_data data;
70 			struct mutex in_mutex;
71 			struct mutex out_mutex;
72 
73 			wait_queue_head_t inflight_conn_req;
74 		} active;
75 		struct {
76 		/* Socket status */
77 #define PVCALLS_STATUS_UNINITALIZED  0
78 #define PVCALLS_STATUS_BIND          1
79 #define PVCALLS_STATUS_LISTEN        2
80 			uint8_t status;
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 		 */
86 #define PVCALLS_FLAG_ACCEPT_INFLIGHT 0
87 #define PVCALLS_FLAG_POLL_INFLIGHT   1
88 #define PVCALLS_FLAG_POLL_RET        2
89 			uint8_t flags;
90 			uint32_t inflight_req_id;
91 			struct sock_mapping *accept_map;
92 			wait_queue_head_t inflight_accept_req;
93 		} passive;
94 	};
95 };
96 
97 static inline struct sock_mapping *pvcalls_enter_sock(struct socket *sock)
98 {
99 	struct sock_mapping *map;
100 
101 	if (!pvcalls_front_dev ||
102 		dev_get_drvdata(&pvcalls_front_dev->dev) == NULL)
103 		return ERR_PTR(-ENOTCONN);
104 
105 	map = (struct sock_mapping *)sock->sk->sk_send_head;
106 	if (map == NULL)
107 		return ERR_PTR(-ENOTSOCK);
108 
109 	pvcalls_enter();
110 	atomic_inc(&map->refcount);
111 	return map;
112 }
113 
114 static inline void pvcalls_exit_sock(struct socket *sock)
115 {
116 	struct sock_mapping *map;
117 
118 	map = (struct sock_mapping *)sock->sk->sk_send_head;
119 	atomic_dec(&map->refcount);
120 	pvcalls_exit();
121 }
122 
123 static inline int get_request(struct pvcalls_bedata *bedata, int *req_id)
124 {
125 	*req_id = bedata->ring.req_prod_pvt & (RING_SIZE(&bedata->ring) - 1);
126 	if (RING_FULL(&bedata->ring) ||
127 	    bedata->rsp[*req_id].req_id != PVCALLS_INVALID_ID)
128 		return -EAGAIN;
129 	return 0;
130 }
131 
132 static bool pvcalls_front_write_todo(struct sock_mapping *map)
133 {
134 	struct pvcalls_data_intf *intf = map->active.ring;
135 	RING_IDX cons, prod, size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
136 	int32_t error;
137 
138 	error = intf->out_error;
139 	if (error == -ENOTCONN)
140 		return false;
141 	if (error != 0)
142 		return true;
143 
144 	cons = intf->out_cons;
145 	prod = intf->out_prod;
146 	return !!(size - pvcalls_queued(prod, cons, size));
147 }
148 
149 static bool pvcalls_front_read_todo(struct sock_mapping *map)
150 {
151 	struct pvcalls_data_intf *intf = map->active.ring;
152 	RING_IDX cons, prod;
153 	int32_t error;
154 
155 	cons = intf->in_cons;
156 	prod = intf->in_prod;
157 	error = intf->in_error;
158 	return (error != 0 ||
159 		pvcalls_queued(prod, cons,
160 			       XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER)) != 0);
161 }
162 
163 static irqreturn_t pvcalls_front_event_handler(int irq, void *dev_id)
164 {
165 	struct xenbus_device *dev = dev_id;
166 	struct pvcalls_bedata *bedata;
167 	struct xen_pvcalls_response *rsp;
168 	uint8_t *src, *dst;
169 	int req_id = 0, more = 0, done = 0;
170 
171 	if (dev == NULL)
172 		return IRQ_HANDLED;
173 
174 	pvcalls_enter();
175 	bedata = dev_get_drvdata(&dev->dev);
176 	if (bedata == NULL) {
177 		pvcalls_exit();
178 		return IRQ_HANDLED;
179 	}
180 
181 again:
182 	while (RING_HAS_UNCONSUMED_RESPONSES(&bedata->ring)) {
183 		rsp = RING_GET_RESPONSE(&bedata->ring, bedata->ring.rsp_cons);
184 
185 		req_id = rsp->req_id;
186 		if (rsp->cmd == PVCALLS_POLL) {
187 			struct sock_mapping *map = (struct sock_mapping *)(uintptr_t)
188 						   rsp->u.poll.id;
189 
190 			clear_bit(PVCALLS_FLAG_POLL_INFLIGHT,
191 				  (void *)&map->passive.flags);
192 			/*
193 			 * clear INFLIGHT, then set RET. It pairs with
194 			 * the checks at the beginning of
195 			 * pvcalls_front_poll_passive.
196 			 */
197 			smp_wmb();
198 			set_bit(PVCALLS_FLAG_POLL_RET,
199 				(void *)&map->passive.flags);
200 		} else {
201 			dst = (uint8_t *)&bedata->rsp[req_id] +
202 			      sizeof(rsp->req_id);
203 			src = (uint8_t *)rsp + sizeof(rsp->req_id);
204 			memcpy(dst, src, sizeof(*rsp) - sizeof(rsp->req_id));
205 			/*
206 			 * First copy the rest of the data, then req_id. It is
207 			 * paired with the barrier when accessing bedata->rsp.
208 			 */
209 			smp_wmb();
210 			bedata->rsp[req_id].req_id = req_id;
211 		}
212 
213 		done = 1;
214 		bedata->ring.rsp_cons++;
215 	}
216 
217 	RING_FINAL_CHECK_FOR_RESPONSES(&bedata->ring, more);
218 	if (more)
219 		goto again;
220 	if (done)
221 		wake_up(&bedata->inflight_req);
222 	pvcalls_exit();
223 	return IRQ_HANDLED;
224 }
225 
226 static void pvcalls_front_free_map(struct pvcalls_bedata *bedata,
227 				   struct sock_mapping *map)
228 {
229 	int i;
230 
231 	unbind_from_irqhandler(map->active.irq, map);
232 
233 	spin_lock(&bedata->socket_lock);
234 	if (!list_empty(&map->list))
235 		list_del_init(&map->list);
236 	spin_unlock(&bedata->socket_lock);
237 
238 	for (i = 0; i < (1 << PVCALLS_RING_ORDER); i++)
239 		gnttab_end_foreign_access(map->active.ring->ref[i], 0, 0);
240 	gnttab_end_foreign_access(map->active.ref, 0, 0);
241 	free_page((unsigned long)map->active.ring);
242 
243 	kfree(map);
244 }
245 
246 static irqreturn_t pvcalls_front_conn_handler(int irq, void *sock_map)
247 {
248 	struct sock_mapping *map = sock_map;
249 
250 	if (map == NULL)
251 		return IRQ_HANDLED;
252 
253 	wake_up_interruptible(&map->active.inflight_conn_req);
254 
255 	return IRQ_HANDLED;
256 }
257 
258 int pvcalls_front_socket(struct socket *sock)
259 {
260 	struct pvcalls_bedata *bedata;
261 	struct sock_mapping *map = NULL;
262 	struct xen_pvcalls_request *req;
263 	int notify, req_id, ret;
264 
265 	/*
266 	 * PVCalls only supports domain AF_INET,
267 	 * type SOCK_STREAM and protocol 0 sockets for now.
268 	 *
269 	 * Check socket type here, AF_INET and protocol checks are done
270 	 * by the caller.
271 	 */
272 	if (sock->type != SOCK_STREAM)
273 		return -EOPNOTSUPP;
274 
275 	pvcalls_enter();
276 	if (!pvcalls_front_dev) {
277 		pvcalls_exit();
278 		return -EACCES;
279 	}
280 	bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
281 
282 	map = kzalloc(sizeof(*map), GFP_KERNEL);
283 	if (map == NULL) {
284 		pvcalls_exit();
285 		return -ENOMEM;
286 	}
287 
288 	spin_lock(&bedata->socket_lock);
289 
290 	ret = get_request(bedata, &req_id);
291 	if (ret < 0) {
292 		kfree(map);
293 		spin_unlock(&bedata->socket_lock);
294 		pvcalls_exit();
295 		return ret;
296 	}
297 
298 	/*
299 	 * sock->sk->sk_send_head is not used for ip sockets: reuse the
300 	 * field to store a pointer to the struct sock_mapping
301 	 * corresponding to the socket. This way, we can easily get the
302 	 * struct sock_mapping from the struct socket.
303 	 */
304 	sock->sk->sk_send_head = (void *)map;
305 	list_add_tail(&map->list, &bedata->socket_mappings);
306 
307 	req = RING_GET_REQUEST(&bedata->ring, req_id);
308 	req->req_id = req_id;
309 	req->cmd = PVCALLS_SOCKET;
310 	req->u.socket.id = (uintptr_t) map;
311 	req->u.socket.domain = AF_INET;
312 	req->u.socket.type = SOCK_STREAM;
313 	req->u.socket.protocol = IPPROTO_IP;
314 
315 	bedata->ring.req_prod_pvt++;
316 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
317 	spin_unlock(&bedata->socket_lock);
318 	if (notify)
319 		notify_remote_via_irq(bedata->irq);
320 
321 	wait_event(bedata->inflight_req,
322 		   READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
323 
324 	/* read req_id, then the content */
325 	smp_rmb();
326 	ret = bedata->rsp[req_id].ret;
327 	bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
328 
329 	pvcalls_exit();
330 	return ret;
331 }
332 
333 static int create_active(struct sock_mapping *map, int *evtchn)
334 {
335 	void *bytes;
336 	int ret = -ENOMEM, irq = -1, i;
337 
338 	*evtchn = -1;
339 	init_waitqueue_head(&map->active.inflight_conn_req);
340 
341 	map->active.ring = (struct pvcalls_data_intf *)
342 		__get_free_page(GFP_KERNEL | __GFP_ZERO);
343 	if (map->active.ring == NULL)
344 		goto out_error;
345 	map->active.ring->ring_order = PVCALLS_RING_ORDER;
346 	bytes = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
347 					PVCALLS_RING_ORDER);
348 	if (bytes == NULL)
349 		goto out_error;
350 	for (i = 0; i < (1 << PVCALLS_RING_ORDER); i++)
351 		map->active.ring->ref[i] = gnttab_grant_foreign_access(
352 			pvcalls_front_dev->otherend_id,
353 			pfn_to_gfn(virt_to_pfn(bytes) + i), 0);
354 
355 	map->active.ref = gnttab_grant_foreign_access(
356 		pvcalls_front_dev->otherend_id,
357 		pfn_to_gfn(virt_to_pfn((void *)map->active.ring)), 0);
358 
359 	map->active.data.in = bytes;
360 	map->active.data.out = bytes +
361 		XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
362 
363 	ret = xenbus_alloc_evtchn(pvcalls_front_dev, evtchn);
364 	if (ret)
365 		goto out_error;
366 	irq = bind_evtchn_to_irqhandler(*evtchn, pvcalls_front_conn_handler,
367 					0, "pvcalls-frontend", map);
368 	if (irq < 0) {
369 		ret = irq;
370 		goto out_error;
371 	}
372 
373 	map->active.irq = irq;
374 	map->active_socket = true;
375 	mutex_init(&map->active.in_mutex);
376 	mutex_init(&map->active.out_mutex);
377 
378 	return 0;
379 
380 out_error:
381 	if (*evtchn >= 0)
382 		xenbus_free_evtchn(pvcalls_front_dev, *evtchn);
383 	kfree(map->active.data.in);
384 	kfree(map->active.ring);
385 	return ret;
386 }
387 
388 int pvcalls_front_connect(struct socket *sock, struct sockaddr *addr,
389 				int addr_len, int flags)
390 {
391 	struct pvcalls_bedata *bedata;
392 	struct sock_mapping *map = NULL;
393 	struct xen_pvcalls_request *req;
394 	int notify, req_id, ret, evtchn;
395 
396 	if (addr->sa_family != AF_INET || sock->type != SOCK_STREAM)
397 		return -EOPNOTSUPP;
398 
399 	map = pvcalls_enter_sock(sock);
400 	if (IS_ERR(map))
401 		return PTR_ERR(map);
402 
403 	bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
404 
405 	spin_lock(&bedata->socket_lock);
406 	ret = get_request(bedata, &req_id);
407 	if (ret < 0) {
408 		spin_unlock(&bedata->socket_lock);
409 		pvcalls_exit_sock(sock);
410 		return ret;
411 	}
412 	ret = create_active(map, &evtchn);
413 	if (ret < 0) {
414 		spin_unlock(&bedata->socket_lock);
415 		pvcalls_exit_sock(sock);
416 		return ret;
417 	}
418 
419 	req = RING_GET_REQUEST(&bedata->ring, req_id);
420 	req->req_id = req_id;
421 	req->cmd = PVCALLS_CONNECT;
422 	req->u.connect.id = (uintptr_t)map;
423 	req->u.connect.len = addr_len;
424 	req->u.connect.flags = flags;
425 	req->u.connect.ref = map->active.ref;
426 	req->u.connect.evtchn = evtchn;
427 	memcpy(req->u.connect.addr, addr, sizeof(*addr));
428 
429 	map->sock = sock;
430 
431 	bedata->ring.req_prod_pvt++;
432 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
433 	spin_unlock(&bedata->socket_lock);
434 
435 	if (notify)
436 		notify_remote_via_irq(bedata->irq);
437 
438 	wait_event(bedata->inflight_req,
439 		   READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
440 
441 	/* read req_id, then the content */
442 	smp_rmb();
443 	ret = bedata->rsp[req_id].ret;
444 	bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
445 	pvcalls_exit_sock(sock);
446 	return ret;
447 }
448 
449 static int __write_ring(struct pvcalls_data_intf *intf,
450 			struct pvcalls_data *data,
451 			struct iov_iter *msg_iter,
452 			int len)
453 {
454 	RING_IDX cons, prod, size, masked_prod, masked_cons;
455 	RING_IDX array_size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
456 	int32_t error;
457 
458 	error = intf->out_error;
459 	if (error < 0)
460 		return error;
461 	cons = intf->out_cons;
462 	prod = intf->out_prod;
463 	/* read indexes before continuing */
464 	virt_mb();
465 
466 	size = pvcalls_queued(prod, cons, array_size);
467 	if (size >= array_size)
468 		return -EINVAL;
469 	if (len > array_size - size)
470 		len = array_size - size;
471 
472 	masked_prod = pvcalls_mask(prod, array_size);
473 	masked_cons = pvcalls_mask(cons, array_size);
474 
475 	if (masked_prod < masked_cons) {
476 		len = copy_from_iter(data->out + masked_prod, len, msg_iter);
477 	} else {
478 		if (len > array_size - masked_prod) {
479 			int ret = copy_from_iter(data->out + masked_prod,
480 				       array_size - masked_prod, msg_iter);
481 			if (ret != array_size - masked_prod) {
482 				len = ret;
483 				goto out;
484 			}
485 			len = ret + copy_from_iter(data->out, len - ret, msg_iter);
486 		} else {
487 			len = copy_from_iter(data->out + masked_prod, len, msg_iter);
488 		}
489 	}
490 out:
491 	/* write to ring before updating pointer */
492 	virt_wmb();
493 	intf->out_prod += len;
494 
495 	return len;
496 }
497 
498 int pvcalls_front_sendmsg(struct socket *sock, struct msghdr *msg,
499 			  size_t len)
500 {
501 	struct pvcalls_bedata *bedata;
502 	struct sock_mapping *map;
503 	int sent, tot_sent = 0;
504 	int count = 0, flags;
505 
506 	flags = msg->msg_flags;
507 	if (flags & (MSG_CONFIRM|MSG_DONTROUTE|MSG_EOR|MSG_OOB))
508 		return -EOPNOTSUPP;
509 
510 	map = pvcalls_enter_sock(sock);
511 	if (IS_ERR(map))
512 		return PTR_ERR(map);
513 	bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
514 
515 	mutex_lock(&map->active.out_mutex);
516 	if ((flags & MSG_DONTWAIT) && !pvcalls_front_write_todo(map)) {
517 		mutex_unlock(&map->active.out_mutex);
518 		pvcalls_exit_sock(sock);
519 		return -EAGAIN;
520 	}
521 	if (len > INT_MAX)
522 		len = INT_MAX;
523 
524 again:
525 	count++;
526 	sent = __write_ring(map->active.ring,
527 			    &map->active.data, &msg->msg_iter,
528 			    len);
529 	if (sent > 0) {
530 		len -= sent;
531 		tot_sent += sent;
532 		notify_remote_via_irq(map->active.irq);
533 	}
534 	if (sent >= 0 && len > 0 && count < PVCALLS_FRONT_MAX_SPIN)
535 		goto again;
536 	if (sent < 0)
537 		tot_sent = sent;
538 
539 	mutex_unlock(&map->active.out_mutex);
540 	pvcalls_exit_sock(sock);
541 	return tot_sent;
542 }
543 
544 static int __read_ring(struct pvcalls_data_intf *intf,
545 		       struct pvcalls_data *data,
546 		       struct iov_iter *msg_iter,
547 		       size_t len, int flags)
548 {
549 	RING_IDX cons, prod, size, masked_prod, masked_cons;
550 	RING_IDX array_size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
551 	int32_t error;
552 
553 	cons = intf->in_cons;
554 	prod = intf->in_prod;
555 	error = intf->in_error;
556 	/* get pointers before reading from the ring */
557 	virt_rmb();
558 	if (error < 0)
559 		return error;
560 
561 	size = pvcalls_queued(prod, cons, array_size);
562 	masked_prod = pvcalls_mask(prod, array_size);
563 	masked_cons = pvcalls_mask(cons, array_size);
564 
565 	if (size == 0)
566 		return 0;
567 
568 	if (len > size)
569 		len = size;
570 
571 	if (masked_prod > masked_cons) {
572 		len = copy_to_iter(data->in + masked_cons, len, msg_iter);
573 	} else {
574 		if (len > (array_size - masked_cons)) {
575 			int ret = copy_to_iter(data->in + masked_cons,
576 				     array_size - masked_cons, msg_iter);
577 			if (ret != array_size - masked_cons) {
578 				len = ret;
579 				goto out;
580 			}
581 			len = ret + copy_to_iter(data->in, len - ret, msg_iter);
582 		} else {
583 			len = copy_to_iter(data->in + masked_cons, len, msg_iter);
584 		}
585 	}
586 out:
587 	/* read data from the ring before increasing the index */
588 	virt_mb();
589 	if (!(flags & MSG_PEEK))
590 		intf->in_cons += len;
591 
592 	return len;
593 }
594 
595 int pvcalls_front_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
596 		     int flags)
597 {
598 	struct pvcalls_bedata *bedata;
599 	int ret;
600 	struct sock_mapping *map;
601 
602 	if (flags & (MSG_CMSG_CLOEXEC|MSG_ERRQUEUE|MSG_OOB|MSG_TRUNC))
603 		return -EOPNOTSUPP;
604 
605 	map = pvcalls_enter_sock(sock);
606 	if (IS_ERR(map))
607 		return PTR_ERR(map);
608 	bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
609 
610 	mutex_lock(&map->active.in_mutex);
611 	if (len > XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER))
612 		len = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
613 
614 	while (!(flags & MSG_DONTWAIT) && !pvcalls_front_read_todo(map)) {
615 		wait_event_interruptible(map->active.inflight_conn_req,
616 					 pvcalls_front_read_todo(map));
617 	}
618 	ret = __read_ring(map->active.ring, &map->active.data,
619 			  &msg->msg_iter, len, flags);
620 
621 	if (ret > 0)
622 		notify_remote_via_irq(map->active.irq);
623 	if (ret == 0)
624 		ret = (flags & MSG_DONTWAIT) ? -EAGAIN : 0;
625 	if (ret == -ENOTCONN)
626 		ret = 0;
627 
628 	mutex_unlock(&map->active.in_mutex);
629 	pvcalls_exit_sock(sock);
630 	return ret;
631 }
632 
633 int pvcalls_front_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
634 {
635 	struct pvcalls_bedata *bedata;
636 	struct sock_mapping *map = NULL;
637 	struct xen_pvcalls_request *req;
638 	int notify, req_id, ret;
639 
640 	if (addr->sa_family != AF_INET || sock->type != SOCK_STREAM)
641 		return -EOPNOTSUPP;
642 
643 	map = pvcalls_enter_sock(sock);
644 	if (IS_ERR(map))
645 		return PTR_ERR(map);
646 	bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
647 
648 	spin_lock(&bedata->socket_lock);
649 	ret = get_request(bedata, &req_id);
650 	if (ret < 0) {
651 		spin_unlock(&bedata->socket_lock);
652 		pvcalls_exit_sock(sock);
653 		return ret;
654 	}
655 	req = RING_GET_REQUEST(&bedata->ring, req_id);
656 	req->req_id = req_id;
657 	map->sock = sock;
658 	req->cmd = PVCALLS_BIND;
659 	req->u.bind.id = (uintptr_t)map;
660 	memcpy(req->u.bind.addr, addr, sizeof(*addr));
661 	req->u.bind.len = addr_len;
662 
663 	init_waitqueue_head(&map->passive.inflight_accept_req);
664 
665 	map->active_socket = false;
666 
667 	bedata->ring.req_prod_pvt++;
668 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
669 	spin_unlock(&bedata->socket_lock);
670 	if (notify)
671 		notify_remote_via_irq(bedata->irq);
672 
673 	wait_event(bedata->inflight_req,
674 		   READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
675 
676 	/* read req_id, then the content */
677 	smp_rmb();
678 	ret = bedata->rsp[req_id].ret;
679 	bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
680 
681 	map->passive.status = PVCALLS_STATUS_BIND;
682 	pvcalls_exit_sock(sock);
683 	return 0;
684 }
685 
686 int pvcalls_front_listen(struct socket *sock, int backlog)
687 {
688 	struct pvcalls_bedata *bedata;
689 	struct sock_mapping *map;
690 	struct xen_pvcalls_request *req;
691 	int notify, req_id, ret;
692 
693 	map = pvcalls_enter_sock(sock);
694 	if (IS_ERR(map))
695 		return PTR_ERR(map);
696 	bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
697 
698 	if (map->passive.status != PVCALLS_STATUS_BIND) {
699 		pvcalls_exit_sock(sock);
700 		return -EOPNOTSUPP;
701 	}
702 
703 	spin_lock(&bedata->socket_lock);
704 	ret = get_request(bedata, &req_id);
705 	if (ret < 0) {
706 		spin_unlock(&bedata->socket_lock);
707 		pvcalls_exit_sock(sock);
708 		return ret;
709 	}
710 	req = RING_GET_REQUEST(&bedata->ring, req_id);
711 	req->req_id = req_id;
712 	req->cmd = PVCALLS_LISTEN;
713 	req->u.listen.id = (uintptr_t) map;
714 	req->u.listen.backlog = backlog;
715 
716 	bedata->ring.req_prod_pvt++;
717 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
718 	spin_unlock(&bedata->socket_lock);
719 	if (notify)
720 		notify_remote_via_irq(bedata->irq);
721 
722 	wait_event(bedata->inflight_req,
723 		   READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
724 
725 	/* read req_id, then the content */
726 	smp_rmb();
727 	ret = bedata->rsp[req_id].ret;
728 	bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
729 
730 	map->passive.status = PVCALLS_STATUS_LISTEN;
731 	pvcalls_exit_sock(sock);
732 	return ret;
733 }
734 
735 int pvcalls_front_accept(struct socket *sock, struct socket *newsock, int flags)
736 {
737 	struct pvcalls_bedata *bedata;
738 	struct sock_mapping *map;
739 	struct sock_mapping *map2 = NULL;
740 	struct xen_pvcalls_request *req;
741 	int notify, req_id, ret, evtchn, nonblock;
742 
743 	map = pvcalls_enter_sock(sock);
744 	if (IS_ERR(map))
745 		return PTR_ERR(map);
746 	bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
747 
748 	if (map->passive.status != PVCALLS_STATUS_LISTEN) {
749 		pvcalls_exit_sock(sock);
750 		return -EINVAL;
751 	}
752 
753 	nonblock = flags & SOCK_NONBLOCK;
754 	/*
755 	 * Backend only supports 1 inflight accept request, will return
756 	 * errors for the others
757 	 */
758 	if (test_and_set_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
759 			     (void *)&map->passive.flags)) {
760 		req_id = READ_ONCE(map->passive.inflight_req_id);
761 		if (req_id != PVCALLS_INVALID_ID &&
762 		    READ_ONCE(bedata->rsp[req_id].req_id) == req_id) {
763 			map2 = map->passive.accept_map;
764 			goto received;
765 		}
766 		if (nonblock) {
767 			pvcalls_exit_sock(sock);
768 			return -EAGAIN;
769 		}
770 		if (wait_event_interruptible(map->passive.inflight_accept_req,
771 			!test_and_set_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
772 					  (void *)&map->passive.flags))) {
773 			pvcalls_exit_sock(sock);
774 			return -EINTR;
775 		}
776 	}
777 
778 	spin_lock(&bedata->socket_lock);
779 	ret = get_request(bedata, &req_id);
780 	if (ret < 0) {
781 		clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
782 			  (void *)&map->passive.flags);
783 		spin_unlock(&bedata->socket_lock);
784 		pvcalls_exit_sock(sock);
785 		return ret;
786 	}
787 	map2 = kzalloc(sizeof(*map2), GFP_ATOMIC);
788 	if (map2 == NULL) {
789 		clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
790 			  (void *)&map->passive.flags);
791 		spin_unlock(&bedata->socket_lock);
792 		pvcalls_exit_sock(sock);
793 		return -ENOMEM;
794 	}
795 	ret = create_active(map2, &evtchn);
796 	if (ret < 0) {
797 		kfree(map2);
798 		clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
799 			  (void *)&map->passive.flags);
800 		spin_unlock(&bedata->socket_lock);
801 		pvcalls_exit_sock(sock);
802 		return ret;
803 	}
804 	list_add_tail(&map2->list, &bedata->socket_mappings);
805 
806 	req = RING_GET_REQUEST(&bedata->ring, req_id);
807 	req->req_id = req_id;
808 	req->cmd = PVCALLS_ACCEPT;
809 	req->u.accept.id = (uintptr_t) map;
810 	req->u.accept.ref = map2->active.ref;
811 	req->u.accept.id_new = (uintptr_t) map2;
812 	req->u.accept.evtchn = evtchn;
813 	map->passive.accept_map = map2;
814 
815 	bedata->ring.req_prod_pvt++;
816 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
817 	spin_unlock(&bedata->socket_lock);
818 	if (notify)
819 		notify_remote_via_irq(bedata->irq);
820 	/* We could check if we have received a response before returning. */
821 	if (nonblock) {
822 		WRITE_ONCE(map->passive.inflight_req_id, req_id);
823 		pvcalls_exit_sock(sock);
824 		return -EAGAIN;
825 	}
826 
827 	if (wait_event_interruptible(bedata->inflight_req,
828 		READ_ONCE(bedata->rsp[req_id].req_id) == req_id)) {
829 		pvcalls_exit_sock(sock);
830 		return -EINTR;
831 	}
832 	/* read req_id, then the content */
833 	smp_rmb();
834 
835 received:
836 	map2->sock = newsock;
837 	newsock->sk = kzalloc(sizeof(*newsock->sk), GFP_KERNEL);
838 	if (!newsock->sk) {
839 		bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
840 		map->passive.inflight_req_id = PVCALLS_INVALID_ID;
841 		clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
842 			  (void *)&map->passive.flags);
843 		pvcalls_front_free_map(bedata, map2);
844 		pvcalls_exit_sock(sock);
845 		return -ENOMEM;
846 	}
847 	newsock->sk->sk_send_head = (void *)map2;
848 
849 	ret = bedata->rsp[req_id].ret;
850 	bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
851 	map->passive.inflight_req_id = PVCALLS_INVALID_ID;
852 
853 	clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT, (void *)&map->passive.flags);
854 	wake_up(&map->passive.inflight_accept_req);
855 
856 	pvcalls_exit_sock(sock);
857 	return ret;
858 }
859 
860 static __poll_t pvcalls_front_poll_passive(struct file *file,
861 					       struct pvcalls_bedata *bedata,
862 					       struct sock_mapping *map,
863 					       poll_table *wait)
864 {
865 	int notify, req_id, ret;
866 	struct xen_pvcalls_request *req;
867 
868 	if (test_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
869 		     (void *)&map->passive.flags)) {
870 		uint32_t req_id = READ_ONCE(map->passive.inflight_req_id);
871 
872 		if (req_id != PVCALLS_INVALID_ID &&
873 		    READ_ONCE(bedata->rsp[req_id].req_id) == req_id)
874 			return EPOLLIN | EPOLLRDNORM;
875 
876 		poll_wait(file, &map->passive.inflight_accept_req, wait);
877 		return 0;
878 	}
879 
880 	if (test_and_clear_bit(PVCALLS_FLAG_POLL_RET,
881 			       (void *)&map->passive.flags))
882 		return EPOLLIN | EPOLLRDNORM;
883 
884 	/*
885 	 * First check RET, then INFLIGHT. No barriers necessary to
886 	 * ensure execution ordering because of the conditional
887 	 * instructions creating control dependencies.
888 	 */
889 
890 	if (test_and_set_bit(PVCALLS_FLAG_POLL_INFLIGHT,
891 			     (void *)&map->passive.flags)) {
892 		poll_wait(file, &bedata->inflight_req, wait);
893 		return 0;
894 	}
895 
896 	spin_lock(&bedata->socket_lock);
897 	ret = get_request(bedata, &req_id);
898 	if (ret < 0) {
899 		spin_unlock(&bedata->socket_lock);
900 		return ret;
901 	}
902 	req = RING_GET_REQUEST(&bedata->ring, req_id);
903 	req->req_id = req_id;
904 	req->cmd = PVCALLS_POLL;
905 	req->u.poll.id = (uintptr_t) map;
906 
907 	bedata->ring.req_prod_pvt++;
908 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
909 	spin_unlock(&bedata->socket_lock);
910 	if (notify)
911 		notify_remote_via_irq(bedata->irq);
912 
913 	poll_wait(file, &bedata->inflight_req, wait);
914 	return 0;
915 }
916 
917 static __poll_t pvcalls_front_poll_active(struct file *file,
918 					      struct pvcalls_bedata *bedata,
919 					      struct sock_mapping *map,
920 					      poll_table *wait)
921 {
922 	__poll_t mask = 0;
923 	int32_t in_error, out_error;
924 	struct pvcalls_data_intf *intf = map->active.ring;
925 
926 	out_error = intf->out_error;
927 	in_error = intf->in_error;
928 
929 	poll_wait(file, &map->active.inflight_conn_req, wait);
930 	if (pvcalls_front_write_todo(map))
931 		mask |= EPOLLOUT | EPOLLWRNORM;
932 	if (pvcalls_front_read_todo(map))
933 		mask |= EPOLLIN | EPOLLRDNORM;
934 	if (in_error != 0 || out_error != 0)
935 		mask |= EPOLLERR;
936 
937 	return mask;
938 }
939 
940 __poll_t pvcalls_front_poll(struct file *file, struct socket *sock,
941 			       poll_table *wait)
942 {
943 	struct pvcalls_bedata *bedata;
944 	struct sock_mapping *map;
945 	__poll_t ret;
946 
947 	map = pvcalls_enter_sock(sock);
948 	if (IS_ERR(map))
949 		return EPOLLNVAL;
950 	bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
951 
952 	if (map->active_socket)
953 		ret = pvcalls_front_poll_active(file, bedata, map, wait);
954 	else
955 		ret = pvcalls_front_poll_passive(file, bedata, map, wait);
956 	pvcalls_exit_sock(sock);
957 	return ret;
958 }
959 
960 int pvcalls_front_release(struct socket *sock)
961 {
962 	struct pvcalls_bedata *bedata;
963 	struct sock_mapping *map;
964 	int req_id, notify, ret;
965 	struct xen_pvcalls_request *req;
966 
967 	if (sock->sk == NULL)
968 		return 0;
969 
970 	map = pvcalls_enter_sock(sock);
971 	if (IS_ERR(map)) {
972 		if (PTR_ERR(map) == -ENOTCONN)
973 			return -EIO;
974 		else
975 			return 0;
976 	}
977 	bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
978 
979 	spin_lock(&bedata->socket_lock);
980 	ret = get_request(bedata, &req_id);
981 	if (ret < 0) {
982 		spin_unlock(&bedata->socket_lock);
983 		pvcalls_exit_sock(sock);
984 		return ret;
985 	}
986 	sock->sk->sk_send_head = NULL;
987 
988 	req = RING_GET_REQUEST(&bedata->ring, req_id);
989 	req->req_id = req_id;
990 	req->cmd = PVCALLS_RELEASE;
991 	req->u.release.id = (uintptr_t)map;
992 
993 	bedata->ring.req_prod_pvt++;
994 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
995 	spin_unlock(&bedata->socket_lock);
996 	if (notify)
997 		notify_remote_via_irq(bedata->irq);
998 
999 	wait_event(bedata->inflight_req,
1000 		   READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
1001 
1002 	if (map->active_socket) {
1003 		/*
1004 		 * Set in_error and wake up inflight_conn_req to force
1005 		 * recvmsg waiters to exit.
1006 		 */
1007 		map->active.ring->in_error = -EBADF;
1008 		wake_up_interruptible(&map->active.inflight_conn_req);
1009 
1010 		/*
1011 		 * We need to make sure that sendmsg/recvmsg on this socket have
1012 		 * not started before we've cleared sk_send_head here. The
1013 		 * easiest way to guarantee this is to see that no pvcalls
1014 		 * (other than us) is in progress on this socket.
1015 		 */
1016 		while (atomic_read(&map->refcount) > 1)
1017 			cpu_relax();
1018 
1019 		pvcalls_front_free_map(bedata, map);
1020 	} else {
1021 		wake_up(&bedata->inflight_req);
1022 		wake_up(&map->passive.inflight_accept_req);
1023 
1024 		while (atomic_read(&map->refcount) > 1)
1025 			cpu_relax();
1026 
1027 		spin_lock(&bedata->socket_lock);
1028 		list_del(&map->list);
1029 		spin_unlock(&bedata->socket_lock);
1030 		if (READ_ONCE(map->passive.inflight_req_id) !=
1031 		    PVCALLS_INVALID_ID) {
1032 			pvcalls_front_free_map(bedata,
1033 					       map->passive.accept_map);
1034 		}
1035 		kfree(map);
1036 	}
1037 	WRITE_ONCE(bedata->rsp[req_id].req_id, PVCALLS_INVALID_ID);
1038 
1039 	pvcalls_exit();
1040 	return 0;
1041 }
1042 
1043 static const struct xenbus_device_id pvcalls_front_ids[] = {
1044 	{ "pvcalls" },
1045 	{ "" }
1046 };
1047 
1048 static int pvcalls_front_remove(struct xenbus_device *dev)
1049 {
1050 	struct pvcalls_bedata *bedata;
1051 	struct sock_mapping *map = NULL, *n;
1052 
1053 	bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
1054 	dev_set_drvdata(&dev->dev, NULL);
1055 	pvcalls_front_dev = NULL;
1056 	if (bedata->irq >= 0)
1057 		unbind_from_irqhandler(bedata->irq, dev);
1058 
1059 	list_for_each_entry_safe(map, n, &bedata->socket_mappings, list) {
1060 		map->sock->sk->sk_send_head = NULL;
1061 		if (map->active_socket) {
1062 			map->active.ring->in_error = -EBADF;
1063 			wake_up_interruptible(&map->active.inflight_conn_req);
1064 		}
1065 	}
1066 
1067 	smp_mb();
1068 	while (atomic_read(&pvcalls_refcount) > 0)
1069 		cpu_relax();
1070 	list_for_each_entry_safe(map, n, &bedata->socket_mappings, list) {
1071 		if (map->active_socket) {
1072 			/* No need to lock, refcount is 0 */
1073 			pvcalls_front_free_map(bedata, map);
1074 		} else {
1075 			list_del(&map->list);
1076 			kfree(map);
1077 		}
1078 	}
1079 	if (bedata->ref != -1)
1080 		gnttab_end_foreign_access(bedata->ref, 0, 0);
1081 	kfree(bedata->ring.sring);
1082 	kfree(bedata);
1083 	xenbus_switch_state(dev, XenbusStateClosed);
1084 	return 0;
1085 }
1086 
1087 static int pvcalls_front_probe(struct xenbus_device *dev,
1088 			  const struct xenbus_device_id *id)
1089 {
1090 	int ret = -ENOMEM, evtchn, i;
1091 	unsigned int max_page_order, function_calls, len;
1092 	char *versions;
1093 	grant_ref_t gref_head = 0;
1094 	struct xenbus_transaction xbt;
1095 	struct pvcalls_bedata *bedata = NULL;
1096 	struct xen_pvcalls_sring *sring;
1097 
1098 	if (pvcalls_front_dev != NULL) {
1099 		dev_err(&dev->dev, "only one PV Calls connection supported\n");
1100 		return -EINVAL;
1101 	}
1102 
1103 	versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
1104 	if (IS_ERR(versions))
1105 		return PTR_ERR(versions);
1106 	if (!len)
1107 		return -EINVAL;
1108 	if (strcmp(versions, "1")) {
1109 		kfree(versions);
1110 		return -EINVAL;
1111 	}
1112 	kfree(versions);
1113 	max_page_order = xenbus_read_unsigned(dev->otherend,
1114 					      "max-page-order", 0);
1115 	if (max_page_order < PVCALLS_RING_ORDER)
1116 		return -ENODEV;
1117 	function_calls = xenbus_read_unsigned(dev->otherend,
1118 					      "function-calls", 0);
1119 	/* See XENBUS_FUNCTIONS_CALLS in pvcalls.h */
1120 	if (function_calls != 1)
1121 		return -ENODEV;
1122 	pr_info("%s max-page-order is %u\n", __func__, max_page_order);
1123 
1124 	bedata = kzalloc(sizeof(struct pvcalls_bedata), GFP_KERNEL);
1125 	if (!bedata)
1126 		return -ENOMEM;
1127 
1128 	dev_set_drvdata(&dev->dev, bedata);
1129 	pvcalls_front_dev = dev;
1130 	init_waitqueue_head(&bedata->inflight_req);
1131 	INIT_LIST_HEAD(&bedata->socket_mappings);
1132 	spin_lock_init(&bedata->socket_lock);
1133 	bedata->irq = -1;
1134 	bedata->ref = -1;
1135 
1136 	for (i = 0; i < PVCALLS_NR_RSP_PER_RING; i++)
1137 		bedata->rsp[i].req_id = PVCALLS_INVALID_ID;
1138 
1139 	sring = (struct xen_pvcalls_sring *) __get_free_page(GFP_KERNEL |
1140 							     __GFP_ZERO);
1141 	if (!sring)
1142 		goto error;
1143 	SHARED_RING_INIT(sring);
1144 	FRONT_RING_INIT(&bedata->ring, sring, XEN_PAGE_SIZE);
1145 
1146 	ret = xenbus_alloc_evtchn(dev, &evtchn);
1147 	if (ret)
1148 		goto error;
1149 
1150 	bedata->irq = bind_evtchn_to_irqhandler(evtchn,
1151 						pvcalls_front_event_handler,
1152 						0, "pvcalls-frontend", dev);
1153 	if (bedata->irq < 0) {
1154 		ret = bedata->irq;
1155 		goto error;
1156 	}
1157 
1158 	ret = gnttab_alloc_grant_references(1, &gref_head);
1159 	if (ret < 0)
1160 		goto error;
1161 	ret = gnttab_claim_grant_reference(&gref_head);
1162 	if (ret < 0)
1163 		goto error;
1164 	bedata->ref = ret;
1165 	gnttab_grant_foreign_access_ref(bedata->ref, dev->otherend_id,
1166 					virt_to_gfn((void *)sring), 0);
1167 
1168  again:
1169 	ret = xenbus_transaction_start(&xbt);
1170 	if (ret) {
1171 		xenbus_dev_fatal(dev, ret, "starting transaction");
1172 		goto error;
1173 	}
1174 	ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
1175 	if (ret)
1176 		goto error_xenbus;
1177 	ret = xenbus_printf(xbt, dev->nodename, "ring-ref", "%d", bedata->ref);
1178 	if (ret)
1179 		goto error_xenbus;
1180 	ret = xenbus_printf(xbt, dev->nodename, "port", "%u",
1181 			    evtchn);
1182 	if (ret)
1183 		goto error_xenbus;
1184 	ret = xenbus_transaction_end(xbt, 0);
1185 	if (ret) {
1186 		if (ret == -EAGAIN)
1187 			goto again;
1188 		xenbus_dev_fatal(dev, ret, "completing transaction");
1189 		goto error;
1190 	}
1191 	xenbus_switch_state(dev, XenbusStateInitialised);
1192 
1193 	return 0;
1194 
1195  error_xenbus:
1196 	xenbus_transaction_end(xbt, 1);
1197 	xenbus_dev_fatal(dev, ret, "writing xenstore");
1198  error:
1199 	pvcalls_front_remove(dev);
1200 	return ret;
1201 }
1202 
1203 static void pvcalls_front_changed(struct xenbus_device *dev,
1204 			    enum xenbus_state backend_state)
1205 {
1206 	switch (backend_state) {
1207 	case XenbusStateReconfiguring:
1208 	case XenbusStateReconfigured:
1209 	case XenbusStateInitialising:
1210 	case XenbusStateInitialised:
1211 	case XenbusStateUnknown:
1212 		break;
1213 
1214 	case XenbusStateInitWait:
1215 		break;
1216 
1217 	case XenbusStateConnected:
1218 		xenbus_switch_state(dev, XenbusStateConnected);
1219 		break;
1220 
1221 	case XenbusStateClosed:
1222 		if (dev->state == XenbusStateClosed)
1223 			break;
1224 		/* Missed the backend's CLOSING state */
1225 		/* fall through */
1226 	case XenbusStateClosing:
1227 		xenbus_frontend_closed(dev);
1228 		break;
1229 	}
1230 }
1231 
1232 static struct xenbus_driver pvcalls_front_driver = {
1233 	.ids = pvcalls_front_ids,
1234 	.probe = pvcalls_front_probe,
1235 	.remove = pvcalls_front_remove,
1236 	.otherend_changed = pvcalls_front_changed,
1237 };
1238 
1239 static int __init pvcalls_frontend_init(void)
1240 {
1241 	if (!xen_domain())
1242 		return -ENODEV;
1243 
1244 	pr_info("Initialising Xen pvcalls frontend driver\n");
1245 
1246 	return xenbus_register_frontend(&pvcalls_front_driver);
1247 }
1248 
1249 module_init(pvcalls_frontend_init);
1250 
1251 MODULE_DESCRIPTION("Xen PV Calls frontend driver");
1252 MODULE_AUTHOR("Stefano Stabellini <sstabellini@kernel.org>");
1253 MODULE_LICENSE("GPL");
1254