xref: /linux/drivers/xen/pvcalls-back.c (revision fd639726bf15fca8ee1a00dce8e0096d0ad9bd18)
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/inet.h>
16 #include <linux/kthread.h>
17 #include <linux/list.h>
18 #include <linux/radix-tree.h>
19 #include <linux/module.h>
20 #include <linux/semaphore.h>
21 #include <linux/wait.h>
22 #include <net/sock.h>
23 #include <net/inet_common.h>
24 #include <net/inet_connection_sock.h>
25 #include <net/request_sock.h>
26 
27 #include <xen/events.h>
28 #include <xen/grant_table.h>
29 #include <xen/xen.h>
30 #include <xen/xenbus.h>
31 #include <xen/interface/io/pvcalls.h>
32 
33 #define PVCALLS_VERSIONS "1"
34 #define MAX_RING_ORDER XENBUS_MAX_RING_GRANT_ORDER
35 
36 struct pvcalls_back_global {
37 	struct list_head frontends;
38 	struct semaphore frontends_lock;
39 } pvcalls_back_global;
40 
41 /*
42  * Per-frontend data structure. It contains pointers to the command
43  * ring, its event channel, a list of active sockets and a tree of
44  * passive sockets.
45  */
46 struct pvcalls_fedata {
47 	struct list_head list;
48 	struct xenbus_device *dev;
49 	struct xen_pvcalls_sring *sring;
50 	struct xen_pvcalls_back_ring ring;
51 	int irq;
52 	struct list_head socket_mappings;
53 	struct radix_tree_root socketpass_mappings;
54 	struct semaphore socket_lock;
55 };
56 
57 struct pvcalls_ioworker {
58 	struct work_struct register_work;
59 	struct workqueue_struct *wq;
60 };
61 
62 struct sock_mapping {
63 	struct list_head list;
64 	struct pvcalls_fedata *fedata;
65 	struct sockpass_mapping *sockpass;
66 	struct socket *sock;
67 	uint64_t id;
68 	grant_ref_t ref;
69 	struct pvcalls_data_intf *ring;
70 	void *bytes;
71 	struct pvcalls_data data;
72 	uint32_t ring_order;
73 	int irq;
74 	atomic_t read;
75 	atomic_t write;
76 	atomic_t io;
77 	atomic_t release;
78 	void (*saved_data_ready)(struct sock *sk);
79 	struct pvcalls_ioworker ioworker;
80 };
81 
82 struct sockpass_mapping {
83 	struct list_head list;
84 	struct pvcalls_fedata *fedata;
85 	struct socket *sock;
86 	uint64_t id;
87 	struct xen_pvcalls_request reqcopy;
88 	spinlock_t copy_lock;
89 	struct workqueue_struct *wq;
90 	struct work_struct register_work;
91 	void (*saved_data_ready)(struct sock *sk);
92 };
93 
94 static irqreturn_t pvcalls_back_conn_event(int irq, void *sock_map);
95 static int pvcalls_back_release_active(struct xenbus_device *dev,
96 				       struct pvcalls_fedata *fedata,
97 				       struct sock_mapping *map);
98 
99 static void pvcalls_conn_back_read(void *opaque)
100 {
101 	struct sock_mapping *map = (struct sock_mapping *)opaque;
102 	struct msghdr msg;
103 	struct kvec vec[2];
104 	RING_IDX cons, prod, size, wanted, array_size, masked_prod, masked_cons;
105 	int32_t error;
106 	struct pvcalls_data_intf *intf = map->ring;
107 	struct pvcalls_data *data = &map->data;
108 	unsigned long flags;
109 	int ret;
110 
111 	array_size = XEN_FLEX_RING_SIZE(map->ring_order);
112 	cons = intf->in_cons;
113 	prod = intf->in_prod;
114 	error = intf->in_error;
115 	/* read the indexes first, then deal with the data */
116 	virt_mb();
117 
118 	if (error)
119 		return;
120 
121 	size = pvcalls_queued(prod, cons, array_size);
122 	if (size >= array_size)
123 		return;
124 	spin_lock_irqsave(&map->sock->sk->sk_receive_queue.lock, flags);
125 	if (skb_queue_empty(&map->sock->sk->sk_receive_queue)) {
126 		atomic_set(&map->read, 0);
127 		spin_unlock_irqrestore(&map->sock->sk->sk_receive_queue.lock,
128 				flags);
129 		return;
130 	}
131 	spin_unlock_irqrestore(&map->sock->sk->sk_receive_queue.lock, flags);
132 	wanted = array_size - size;
133 	masked_prod = pvcalls_mask(prod, array_size);
134 	masked_cons = pvcalls_mask(cons, array_size);
135 
136 	memset(&msg, 0, sizeof(msg));
137 	if (masked_prod < masked_cons) {
138 		vec[0].iov_base = data->in + masked_prod;
139 		vec[0].iov_len = wanted;
140 		iov_iter_kvec(&msg.msg_iter, ITER_KVEC|WRITE, vec, 1, wanted);
141 	} else {
142 		vec[0].iov_base = data->in + masked_prod;
143 		vec[0].iov_len = array_size - masked_prod;
144 		vec[1].iov_base = data->in;
145 		vec[1].iov_len = wanted - vec[0].iov_len;
146 		iov_iter_kvec(&msg.msg_iter, ITER_KVEC|WRITE, vec, 2, wanted);
147 	}
148 
149 	atomic_set(&map->read, 0);
150 	ret = inet_recvmsg(map->sock, &msg, wanted, MSG_DONTWAIT);
151 	WARN_ON(ret > wanted);
152 	if (ret == -EAGAIN) /* shouldn't happen */
153 		return;
154 	if (!ret)
155 		ret = -ENOTCONN;
156 	spin_lock_irqsave(&map->sock->sk->sk_receive_queue.lock, flags);
157 	if (ret > 0 && !skb_queue_empty(&map->sock->sk->sk_receive_queue))
158 		atomic_inc(&map->read);
159 	spin_unlock_irqrestore(&map->sock->sk->sk_receive_queue.lock, flags);
160 
161 	/* write the data, then modify the indexes */
162 	virt_wmb();
163 	if (ret < 0)
164 		intf->in_error = ret;
165 	else
166 		intf->in_prod = prod + ret;
167 	/* update the indexes, then notify the other end */
168 	virt_wmb();
169 	notify_remote_via_irq(map->irq);
170 
171 	return;
172 }
173 
174 static void pvcalls_conn_back_write(struct sock_mapping *map)
175 {
176 	struct pvcalls_data_intf *intf = map->ring;
177 	struct pvcalls_data *data = &map->data;
178 	struct msghdr msg;
179 	struct kvec vec[2];
180 	RING_IDX cons, prod, size, array_size;
181 	int ret;
182 
183 	cons = intf->out_cons;
184 	prod = intf->out_prod;
185 	/* read the indexes before dealing with the data */
186 	virt_mb();
187 
188 	array_size = XEN_FLEX_RING_SIZE(map->ring_order);
189 	size = pvcalls_queued(prod, cons, array_size);
190 	if (size == 0)
191 		return;
192 
193 	memset(&msg, 0, sizeof(msg));
194 	msg.msg_flags |= MSG_DONTWAIT;
195 	if (pvcalls_mask(prod, array_size) > pvcalls_mask(cons, array_size)) {
196 		vec[0].iov_base = data->out + pvcalls_mask(cons, array_size);
197 		vec[0].iov_len = size;
198 		iov_iter_kvec(&msg.msg_iter, ITER_KVEC|READ, vec, 1, size);
199 	} else {
200 		vec[0].iov_base = data->out + pvcalls_mask(cons, array_size);
201 		vec[0].iov_len = array_size - pvcalls_mask(cons, array_size);
202 		vec[1].iov_base = data->out;
203 		vec[1].iov_len = size - vec[0].iov_len;
204 		iov_iter_kvec(&msg.msg_iter, ITER_KVEC|READ, vec, 2, size);
205 	}
206 
207 	atomic_set(&map->write, 0);
208 	ret = inet_sendmsg(map->sock, &msg, size);
209 	if (ret == -EAGAIN || (ret >= 0 && ret < size)) {
210 		atomic_inc(&map->write);
211 		atomic_inc(&map->io);
212 	}
213 	if (ret == -EAGAIN)
214 		return;
215 
216 	/* write the data, then update the indexes */
217 	virt_wmb();
218 	if (ret < 0) {
219 		intf->out_error = ret;
220 	} else {
221 		intf->out_error = 0;
222 		intf->out_cons = cons + ret;
223 		prod = intf->out_prod;
224 	}
225 	/* update the indexes, then notify the other end */
226 	virt_wmb();
227 	if (prod != cons + ret)
228 		atomic_inc(&map->write);
229 	notify_remote_via_irq(map->irq);
230 }
231 
232 static void pvcalls_back_ioworker(struct work_struct *work)
233 {
234 	struct pvcalls_ioworker *ioworker = container_of(work,
235 		struct pvcalls_ioworker, register_work);
236 	struct sock_mapping *map = container_of(ioworker, struct sock_mapping,
237 		ioworker);
238 
239 	while (atomic_read(&map->io) > 0) {
240 		if (atomic_read(&map->release) > 0) {
241 			atomic_set(&map->release, 0);
242 			return;
243 		}
244 
245 		if (atomic_read(&map->read) > 0)
246 			pvcalls_conn_back_read(map);
247 		if (atomic_read(&map->write) > 0)
248 			pvcalls_conn_back_write(map);
249 
250 		atomic_dec(&map->io);
251 	}
252 }
253 
254 static int pvcalls_back_socket(struct xenbus_device *dev,
255 		struct xen_pvcalls_request *req)
256 {
257 	struct pvcalls_fedata *fedata;
258 	int ret;
259 	struct xen_pvcalls_response *rsp;
260 
261 	fedata = dev_get_drvdata(&dev->dev);
262 
263 	if (req->u.socket.domain != AF_INET ||
264 	    req->u.socket.type != SOCK_STREAM ||
265 	    (req->u.socket.protocol != IPPROTO_IP &&
266 	     req->u.socket.protocol != AF_INET))
267 		ret = -EAFNOSUPPORT;
268 	else
269 		ret = 0;
270 
271 	/* leave the actual socket allocation for later */
272 
273 	rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
274 	rsp->req_id = req->req_id;
275 	rsp->cmd = req->cmd;
276 	rsp->u.socket.id = req->u.socket.id;
277 	rsp->ret = ret;
278 
279 	return 0;
280 }
281 
282 static void pvcalls_sk_state_change(struct sock *sock)
283 {
284 	struct sock_mapping *map = sock->sk_user_data;
285 	struct pvcalls_data_intf *intf;
286 
287 	if (map == NULL)
288 		return;
289 
290 	intf = map->ring;
291 	intf->in_error = -ENOTCONN;
292 	notify_remote_via_irq(map->irq);
293 }
294 
295 static void pvcalls_sk_data_ready(struct sock *sock)
296 {
297 	struct sock_mapping *map = sock->sk_user_data;
298 	struct pvcalls_ioworker *iow;
299 
300 	if (map == NULL)
301 		return;
302 
303 	iow = &map->ioworker;
304 	atomic_inc(&map->read);
305 	atomic_inc(&map->io);
306 	queue_work(iow->wq, &iow->register_work);
307 }
308 
309 static struct sock_mapping *pvcalls_new_active_socket(
310 		struct pvcalls_fedata *fedata,
311 		uint64_t id,
312 		grant_ref_t ref,
313 		uint32_t evtchn,
314 		struct socket *sock)
315 {
316 	int ret;
317 	struct sock_mapping *map;
318 	void *page;
319 
320 	map = kzalloc(sizeof(*map), GFP_KERNEL);
321 	if (map == NULL)
322 		return NULL;
323 
324 	map->fedata = fedata;
325 	map->sock = sock;
326 	map->id = id;
327 	map->ref = ref;
328 
329 	ret = xenbus_map_ring_valloc(fedata->dev, &ref, 1, &page);
330 	if (ret < 0)
331 		goto out;
332 	map->ring = page;
333 	map->ring_order = map->ring->ring_order;
334 	/* first read the order, then map the data ring */
335 	virt_rmb();
336 	if (map->ring_order > MAX_RING_ORDER) {
337 		pr_warn("%s frontend requested ring_order %u, which is > MAX (%u)\n",
338 				__func__, map->ring_order, MAX_RING_ORDER);
339 		goto out;
340 	}
341 	ret = xenbus_map_ring_valloc(fedata->dev, map->ring->ref,
342 				     (1 << map->ring_order), &page);
343 	if (ret < 0)
344 		goto out;
345 	map->bytes = page;
346 
347 	ret = bind_interdomain_evtchn_to_irqhandler(fedata->dev->otherend_id,
348 						    evtchn,
349 						    pvcalls_back_conn_event,
350 						    0,
351 						    "pvcalls-backend",
352 						    map);
353 	if (ret < 0)
354 		goto out;
355 	map->irq = ret;
356 
357 	map->data.in = map->bytes;
358 	map->data.out = map->bytes + XEN_FLEX_RING_SIZE(map->ring_order);
359 
360 	map->ioworker.wq = alloc_workqueue("pvcalls_io", WQ_UNBOUND, 1);
361 	if (!map->ioworker.wq)
362 		goto out;
363 	atomic_set(&map->io, 1);
364 	INIT_WORK(&map->ioworker.register_work,	pvcalls_back_ioworker);
365 
366 	down(&fedata->socket_lock);
367 	list_add_tail(&map->list, &fedata->socket_mappings);
368 	up(&fedata->socket_lock);
369 
370 	write_lock_bh(&map->sock->sk->sk_callback_lock);
371 	map->saved_data_ready = map->sock->sk->sk_data_ready;
372 	map->sock->sk->sk_user_data = map;
373 	map->sock->sk->sk_data_ready = pvcalls_sk_data_ready;
374 	map->sock->sk->sk_state_change = pvcalls_sk_state_change;
375 	write_unlock_bh(&map->sock->sk->sk_callback_lock);
376 
377 	return map;
378 out:
379 	down(&fedata->socket_lock);
380 	list_del(&map->list);
381 	pvcalls_back_release_active(fedata->dev, fedata, map);
382 	up(&fedata->socket_lock);
383 	return NULL;
384 }
385 
386 static int pvcalls_back_connect(struct xenbus_device *dev,
387 				struct xen_pvcalls_request *req)
388 {
389 	struct pvcalls_fedata *fedata;
390 	int ret = -EINVAL;
391 	struct socket *sock;
392 	struct sock_mapping *map;
393 	struct xen_pvcalls_response *rsp;
394 	struct sockaddr *sa = (struct sockaddr *)&req->u.connect.addr;
395 
396 	fedata = dev_get_drvdata(&dev->dev);
397 
398 	if (req->u.connect.len < sizeof(sa->sa_family) ||
399 	    req->u.connect.len > sizeof(req->u.connect.addr) ||
400 	    sa->sa_family != AF_INET)
401 		goto out;
402 
403 	ret = sock_create(AF_INET, SOCK_STREAM, 0, &sock);
404 	if (ret < 0)
405 		goto out;
406 	ret = inet_stream_connect(sock, sa, req->u.connect.len, 0);
407 	if (ret < 0) {
408 		sock_release(sock);
409 		goto out;
410 	}
411 
412 	map = pvcalls_new_active_socket(fedata,
413 					req->u.connect.id,
414 					req->u.connect.ref,
415 					req->u.connect.evtchn,
416 					sock);
417 	if (!map) {
418 		ret = -EFAULT;
419 		sock_release(map->sock);
420 	}
421 
422 out:
423 	rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
424 	rsp->req_id = req->req_id;
425 	rsp->cmd = req->cmd;
426 	rsp->u.connect.id = req->u.connect.id;
427 	rsp->ret = ret;
428 
429 	return 0;
430 }
431 
432 static int pvcalls_back_release_active(struct xenbus_device *dev,
433 				       struct pvcalls_fedata *fedata,
434 				       struct sock_mapping *map)
435 {
436 	disable_irq(map->irq);
437 	if (map->sock->sk != NULL) {
438 		write_lock_bh(&map->sock->sk->sk_callback_lock);
439 		map->sock->sk->sk_user_data = NULL;
440 		map->sock->sk->sk_data_ready = map->saved_data_ready;
441 		write_unlock_bh(&map->sock->sk->sk_callback_lock);
442 	}
443 
444 	atomic_set(&map->release, 1);
445 	flush_work(&map->ioworker.register_work);
446 
447 	xenbus_unmap_ring_vfree(dev, map->bytes);
448 	xenbus_unmap_ring_vfree(dev, (void *)map->ring);
449 	unbind_from_irqhandler(map->irq, map);
450 
451 	sock_release(map->sock);
452 	kfree(map);
453 
454 	return 0;
455 }
456 
457 static int pvcalls_back_release_passive(struct xenbus_device *dev,
458 					struct pvcalls_fedata *fedata,
459 					struct sockpass_mapping *mappass)
460 {
461 	if (mappass->sock->sk != NULL) {
462 		write_lock_bh(&mappass->sock->sk->sk_callback_lock);
463 		mappass->sock->sk->sk_user_data = NULL;
464 		mappass->sock->sk->sk_data_ready = mappass->saved_data_ready;
465 		write_unlock_bh(&mappass->sock->sk->sk_callback_lock);
466 	}
467 	sock_release(mappass->sock);
468 	flush_workqueue(mappass->wq);
469 	destroy_workqueue(mappass->wq);
470 	kfree(mappass);
471 
472 	return 0;
473 }
474 
475 static int pvcalls_back_release(struct xenbus_device *dev,
476 				struct xen_pvcalls_request *req)
477 {
478 	struct pvcalls_fedata *fedata;
479 	struct sock_mapping *map, *n;
480 	struct sockpass_mapping *mappass;
481 	int ret = 0;
482 	struct xen_pvcalls_response *rsp;
483 
484 	fedata = dev_get_drvdata(&dev->dev);
485 
486 	down(&fedata->socket_lock);
487 	list_for_each_entry_safe(map, n, &fedata->socket_mappings, list) {
488 		if (map->id == req->u.release.id) {
489 			list_del(&map->list);
490 			up(&fedata->socket_lock);
491 			ret = pvcalls_back_release_active(dev, fedata, map);
492 			goto out;
493 		}
494 	}
495 	mappass = radix_tree_lookup(&fedata->socketpass_mappings,
496 				    req->u.release.id);
497 	if (mappass != NULL) {
498 		radix_tree_delete(&fedata->socketpass_mappings, mappass->id);
499 		up(&fedata->socket_lock);
500 		ret = pvcalls_back_release_passive(dev, fedata, mappass);
501 	} else
502 		up(&fedata->socket_lock);
503 
504 out:
505 	rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
506 	rsp->req_id = req->req_id;
507 	rsp->u.release.id = req->u.release.id;
508 	rsp->cmd = req->cmd;
509 	rsp->ret = ret;
510 	return 0;
511 }
512 
513 static void __pvcalls_back_accept(struct work_struct *work)
514 {
515 	struct sockpass_mapping *mappass = container_of(
516 		work, struct sockpass_mapping, register_work);
517 	struct sock_mapping *map;
518 	struct pvcalls_ioworker *iow;
519 	struct pvcalls_fedata *fedata;
520 	struct socket *sock;
521 	struct xen_pvcalls_response *rsp;
522 	struct xen_pvcalls_request *req;
523 	int notify;
524 	int ret = -EINVAL;
525 	unsigned long flags;
526 
527 	fedata = mappass->fedata;
528 	/*
529 	 * __pvcalls_back_accept can race against pvcalls_back_accept.
530 	 * We only need to check the value of "cmd" on read. It could be
531 	 * done atomically, but to simplify the code on the write side, we
532 	 * use a spinlock.
533 	 */
534 	spin_lock_irqsave(&mappass->copy_lock, flags);
535 	req = &mappass->reqcopy;
536 	if (req->cmd != PVCALLS_ACCEPT) {
537 		spin_unlock_irqrestore(&mappass->copy_lock, flags);
538 		return;
539 	}
540 	spin_unlock_irqrestore(&mappass->copy_lock, flags);
541 
542 	sock = sock_alloc();
543 	if (sock == NULL)
544 		goto out_error;
545 	sock->type = mappass->sock->type;
546 	sock->ops = mappass->sock->ops;
547 
548 	ret = inet_accept(mappass->sock, sock, O_NONBLOCK, true);
549 	if (ret == -EAGAIN) {
550 		sock_release(sock);
551 		goto out_error;
552 	}
553 
554 	map = pvcalls_new_active_socket(fedata,
555 					req->u.accept.id_new,
556 					req->u.accept.ref,
557 					req->u.accept.evtchn,
558 					sock);
559 	if (!map) {
560 		ret = -EFAULT;
561 		sock_release(sock);
562 		goto out_error;
563 	}
564 
565 	map->sockpass = mappass;
566 	iow = &map->ioworker;
567 	atomic_inc(&map->read);
568 	atomic_inc(&map->io);
569 	queue_work(iow->wq, &iow->register_work);
570 
571 out_error:
572 	rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
573 	rsp->req_id = req->req_id;
574 	rsp->cmd = req->cmd;
575 	rsp->u.accept.id = req->u.accept.id;
576 	rsp->ret = ret;
577 	RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&fedata->ring, notify);
578 	if (notify)
579 		notify_remote_via_irq(fedata->irq);
580 
581 	mappass->reqcopy.cmd = 0;
582 }
583 
584 static void pvcalls_pass_sk_data_ready(struct sock *sock)
585 {
586 	struct sockpass_mapping *mappass = sock->sk_user_data;
587 	struct pvcalls_fedata *fedata;
588 	struct xen_pvcalls_response *rsp;
589 	unsigned long flags;
590 	int notify;
591 
592 	if (mappass == NULL)
593 		return;
594 
595 	fedata = mappass->fedata;
596 	spin_lock_irqsave(&mappass->copy_lock, flags);
597 	if (mappass->reqcopy.cmd == PVCALLS_POLL) {
598 		rsp = RING_GET_RESPONSE(&fedata->ring,
599 					fedata->ring.rsp_prod_pvt++);
600 		rsp->req_id = mappass->reqcopy.req_id;
601 		rsp->u.poll.id = mappass->reqcopy.u.poll.id;
602 		rsp->cmd = mappass->reqcopy.cmd;
603 		rsp->ret = 0;
604 
605 		mappass->reqcopy.cmd = 0;
606 		spin_unlock_irqrestore(&mappass->copy_lock, flags);
607 
608 		RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&fedata->ring, notify);
609 		if (notify)
610 			notify_remote_via_irq(mappass->fedata->irq);
611 	} else {
612 		spin_unlock_irqrestore(&mappass->copy_lock, flags);
613 		queue_work(mappass->wq, &mappass->register_work);
614 	}
615 }
616 
617 static int pvcalls_back_bind(struct xenbus_device *dev,
618 			     struct xen_pvcalls_request *req)
619 {
620 	struct pvcalls_fedata *fedata;
621 	int ret;
622 	struct sockpass_mapping *map;
623 	struct xen_pvcalls_response *rsp;
624 
625 	fedata = dev_get_drvdata(&dev->dev);
626 
627 	map = kzalloc(sizeof(*map), GFP_KERNEL);
628 	if (map == NULL) {
629 		ret = -ENOMEM;
630 		goto out;
631 	}
632 
633 	INIT_WORK(&map->register_work, __pvcalls_back_accept);
634 	spin_lock_init(&map->copy_lock);
635 	map->wq = alloc_workqueue("pvcalls_wq", WQ_UNBOUND, 1);
636 	if (!map->wq) {
637 		ret = -ENOMEM;
638 		goto out;
639 	}
640 
641 	ret = sock_create(AF_INET, SOCK_STREAM, 0, &map->sock);
642 	if (ret < 0)
643 		goto out;
644 
645 	ret = inet_bind(map->sock, (struct sockaddr *)&req->u.bind.addr,
646 			req->u.bind.len);
647 	if (ret < 0)
648 		goto out;
649 
650 	map->fedata = fedata;
651 	map->id = req->u.bind.id;
652 
653 	down(&fedata->socket_lock);
654 	ret = radix_tree_insert(&fedata->socketpass_mappings, map->id,
655 				map);
656 	up(&fedata->socket_lock);
657 	if (ret)
658 		goto out;
659 
660 	write_lock_bh(&map->sock->sk->sk_callback_lock);
661 	map->saved_data_ready = map->sock->sk->sk_data_ready;
662 	map->sock->sk->sk_user_data = map;
663 	map->sock->sk->sk_data_ready = pvcalls_pass_sk_data_ready;
664 	write_unlock_bh(&map->sock->sk->sk_callback_lock);
665 
666 out:
667 	if (ret) {
668 		if (map && map->sock)
669 			sock_release(map->sock);
670 		if (map && map->wq)
671 			destroy_workqueue(map->wq);
672 		kfree(map);
673 	}
674 	rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
675 	rsp->req_id = req->req_id;
676 	rsp->cmd = req->cmd;
677 	rsp->u.bind.id = req->u.bind.id;
678 	rsp->ret = ret;
679 	return 0;
680 }
681 
682 static int pvcalls_back_listen(struct xenbus_device *dev,
683 			       struct xen_pvcalls_request *req)
684 {
685 	struct pvcalls_fedata *fedata;
686 	int ret = -EINVAL;
687 	struct sockpass_mapping *map;
688 	struct xen_pvcalls_response *rsp;
689 
690 	fedata = dev_get_drvdata(&dev->dev);
691 
692 	down(&fedata->socket_lock);
693 	map = radix_tree_lookup(&fedata->socketpass_mappings, req->u.listen.id);
694 	up(&fedata->socket_lock);
695 	if (map == NULL)
696 		goto out;
697 
698 	ret = inet_listen(map->sock, req->u.listen.backlog);
699 
700 out:
701 	rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
702 	rsp->req_id = req->req_id;
703 	rsp->cmd = req->cmd;
704 	rsp->u.listen.id = req->u.listen.id;
705 	rsp->ret = ret;
706 	return 0;
707 }
708 
709 static int pvcalls_back_accept(struct xenbus_device *dev,
710 			       struct xen_pvcalls_request *req)
711 {
712 	struct pvcalls_fedata *fedata;
713 	struct sockpass_mapping *mappass;
714 	int ret = -EINVAL;
715 	struct xen_pvcalls_response *rsp;
716 	unsigned long flags;
717 
718 	fedata = dev_get_drvdata(&dev->dev);
719 
720 	down(&fedata->socket_lock);
721 	mappass = radix_tree_lookup(&fedata->socketpass_mappings,
722 		req->u.accept.id);
723 	up(&fedata->socket_lock);
724 	if (mappass == NULL)
725 		goto out_error;
726 
727 	/*
728 	 * Limitation of the current implementation: only support one
729 	 * concurrent accept or poll call on one socket.
730 	 */
731 	spin_lock_irqsave(&mappass->copy_lock, flags);
732 	if (mappass->reqcopy.cmd != 0) {
733 		spin_unlock_irqrestore(&mappass->copy_lock, flags);
734 		ret = -EINTR;
735 		goto out_error;
736 	}
737 
738 	mappass->reqcopy = *req;
739 	spin_unlock_irqrestore(&mappass->copy_lock, flags);
740 	queue_work(mappass->wq, &mappass->register_work);
741 
742 	/* Tell the caller we don't need to send back a notification yet */
743 	return -1;
744 
745 out_error:
746 	rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
747 	rsp->req_id = req->req_id;
748 	rsp->cmd = req->cmd;
749 	rsp->u.accept.id = req->u.accept.id;
750 	rsp->ret = ret;
751 	return 0;
752 }
753 
754 static int pvcalls_back_poll(struct xenbus_device *dev,
755 			     struct xen_pvcalls_request *req)
756 {
757 	struct pvcalls_fedata *fedata;
758 	struct sockpass_mapping *mappass;
759 	struct xen_pvcalls_response *rsp;
760 	struct inet_connection_sock *icsk;
761 	struct request_sock_queue *queue;
762 	unsigned long flags;
763 	int ret;
764 	bool data;
765 
766 	fedata = dev_get_drvdata(&dev->dev);
767 
768 	down(&fedata->socket_lock);
769 	mappass = radix_tree_lookup(&fedata->socketpass_mappings,
770 				    req->u.poll.id);
771 	up(&fedata->socket_lock);
772 	if (mappass == NULL)
773 		return -EINVAL;
774 
775 	/*
776 	 * Limitation of the current implementation: only support one
777 	 * concurrent accept or poll call on one socket.
778 	 */
779 	spin_lock_irqsave(&mappass->copy_lock, flags);
780 	if (mappass->reqcopy.cmd != 0) {
781 		ret = -EINTR;
782 		goto out;
783 	}
784 
785 	mappass->reqcopy = *req;
786 	icsk = inet_csk(mappass->sock->sk);
787 	queue = &icsk->icsk_accept_queue;
788 	data = queue->rskq_accept_head != NULL;
789 	if (data) {
790 		mappass->reqcopy.cmd = 0;
791 		ret = 0;
792 		goto out;
793 	}
794 	spin_unlock_irqrestore(&mappass->copy_lock, flags);
795 
796 	/* Tell the caller we don't need to send back a notification yet */
797 	return -1;
798 
799 out:
800 	spin_unlock_irqrestore(&mappass->copy_lock, flags);
801 
802 	rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
803 	rsp->req_id = req->req_id;
804 	rsp->cmd = req->cmd;
805 	rsp->u.poll.id = req->u.poll.id;
806 	rsp->ret = ret;
807 	return 0;
808 }
809 
810 static int pvcalls_back_handle_cmd(struct xenbus_device *dev,
811 				   struct xen_pvcalls_request *req)
812 {
813 	int ret = 0;
814 
815 	switch (req->cmd) {
816 	case PVCALLS_SOCKET:
817 		ret = pvcalls_back_socket(dev, req);
818 		break;
819 	case PVCALLS_CONNECT:
820 		ret = pvcalls_back_connect(dev, req);
821 		break;
822 	case PVCALLS_RELEASE:
823 		ret = pvcalls_back_release(dev, req);
824 		break;
825 	case PVCALLS_BIND:
826 		ret = pvcalls_back_bind(dev, req);
827 		break;
828 	case PVCALLS_LISTEN:
829 		ret = pvcalls_back_listen(dev, req);
830 		break;
831 	case PVCALLS_ACCEPT:
832 		ret = pvcalls_back_accept(dev, req);
833 		break;
834 	case PVCALLS_POLL:
835 		ret = pvcalls_back_poll(dev, req);
836 		break;
837 	default:
838 	{
839 		struct pvcalls_fedata *fedata;
840 		struct xen_pvcalls_response *rsp;
841 
842 		fedata = dev_get_drvdata(&dev->dev);
843 		rsp = RING_GET_RESPONSE(
844 				&fedata->ring, fedata->ring.rsp_prod_pvt++);
845 		rsp->req_id = req->req_id;
846 		rsp->cmd = req->cmd;
847 		rsp->ret = -ENOTSUPP;
848 		break;
849 	}
850 	}
851 	return ret;
852 }
853 
854 static void pvcalls_back_work(struct pvcalls_fedata *fedata)
855 {
856 	int notify, notify_all = 0, more = 1;
857 	struct xen_pvcalls_request req;
858 	struct xenbus_device *dev = fedata->dev;
859 
860 	while (more) {
861 		while (RING_HAS_UNCONSUMED_REQUESTS(&fedata->ring)) {
862 			RING_COPY_REQUEST(&fedata->ring,
863 					  fedata->ring.req_cons++,
864 					  &req);
865 
866 			if (!pvcalls_back_handle_cmd(dev, &req)) {
867 				RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(
868 					&fedata->ring, notify);
869 				notify_all += notify;
870 			}
871 		}
872 
873 		if (notify_all) {
874 			notify_remote_via_irq(fedata->irq);
875 			notify_all = 0;
876 		}
877 
878 		RING_FINAL_CHECK_FOR_REQUESTS(&fedata->ring, more);
879 	}
880 }
881 
882 static irqreturn_t pvcalls_back_event(int irq, void *dev_id)
883 {
884 	struct xenbus_device *dev = dev_id;
885 	struct pvcalls_fedata *fedata = NULL;
886 
887 	if (dev == NULL)
888 		return IRQ_HANDLED;
889 
890 	fedata = dev_get_drvdata(&dev->dev);
891 	if (fedata == NULL)
892 		return IRQ_HANDLED;
893 
894 	pvcalls_back_work(fedata);
895 	return IRQ_HANDLED;
896 }
897 
898 static irqreturn_t pvcalls_back_conn_event(int irq, void *sock_map)
899 {
900 	struct sock_mapping *map = sock_map;
901 	struct pvcalls_ioworker *iow;
902 
903 	if (map == NULL || map->sock == NULL || map->sock->sk == NULL ||
904 		map->sock->sk->sk_user_data != map)
905 		return IRQ_HANDLED;
906 
907 	iow = &map->ioworker;
908 
909 	atomic_inc(&map->write);
910 	atomic_inc(&map->io);
911 	queue_work(iow->wq, &iow->register_work);
912 
913 	return IRQ_HANDLED;
914 }
915 
916 static int backend_connect(struct xenbus_device *dev)
917 {
918 	int err, evtchn;
919 	grant_ref_t ring_ref;
920 	struct pvcalls_fedata *fedata = NULL;
921 
922 	fedata = kzalloc(sizeof(struct pvcalls_fedata), GFP_KERNEL);
923 	if (!fedata)
924 		return -ENOMEM;
925 
926 	fedata->irq = -1;
927 	err = xenbus_scanf(XBT_NIL, dev->otherend, "port", "%u",
928 			   &evtchn);
929 	if (err != 1) {
930 		err = -EINVAL;
931 		xenbus_dev_fatal(dev, err, "reading %s/event-channel",
932 				 dev->otherend);
933 		goto error;
934 	}
935 
936 	err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-ref", "%u", &ring_ref);
937 	if (err != 1) {
938 		err = -EINVAL;
939 		xenbus_dev_fatal(dev, err, "reading %s/ring-ref",
940 				 dev->otherend);
941 		goto error;
942 	}
943 
944 	err = bind_interdomain_evtchn_to_irq(dev->otherend_id, evtchn);
945 	if (err < 0)
946 		goto error;
947 	fedata->irq = err;
948 
949 	err = request_threaded_irq(fedata->irq, NULL, pvcalls_back_event,
950 				   IRQF_ONESHOT, "pvcalls-back", dev);
951 	if (err < 0)
952 		goto error;
953 
954 	err = xenbus_map_ring_valloc(dev, &ring_ref, 1,
955 				     (void **)&fedata->sring);
956 	if (err < 0)
957 		goto error;
958 
959 	BACK_RING_INIT(&fedata->ring, fedata->sring, XEN_PAGE_SIZE * 1);
960 	fedata->dev = dev;
961 
962 	INIT_LIST_HEAD(&fedata->socket_mappings);
963 	INIT_RADIX_TREE(&fedata->socketpass_mappings, GFP_KERNEL);
964 	sema_init(&fedata->socket_lock, 1);
965 	dev_set_drvdata(&dev->dev, fedata);
966 
967 	down(&pvcalls_back_global.frontends_lock);
968 	list_add_tail(&fedata->list, &pvcalls_back_global.frontends);
969 	up(&pvcalls_back_global.frontends_lock);
970 
971 	return 0;
972 
973  error:
974 	if (fedata->irq >= 0)
975 		unbind_from_irqhandler(fedata->irq, dev);
976 	if (fedata->sring != NULL)
977 		xenbus_unmap_ring_vfree(dev, fedata->sring);
978 	kfree(fedata);
979 	return err;
980 }
981 
982 static int backend_disconnect(struct xenbus_device *dev)
983 {
984 	struct pvcalls_fedata *fedata;
985 	struct sock_mapping *map, *n;
986 	struct sockpass_mapping *mappass;
987 	struct radix_tree_iter iter;
988 	void **slot;
989 
990 
991 	fedata = dev_get_drvdata(&dev->dev);
992 
993 	down(&fedata->socket_lock);
994 	list_for_each_entry_safe(map, n, &fedata->socket_mappings, list) {
995 		list_del(&map->list);
996 		pvcalls_back_release_active(dev, fedata, map);
997 	}
998 
999 	radix_tree_for_each_slot(slot, &fedata->socketpass_mappings, &iter, 0) {
1000 		mappass = radix_tree_deref_slot(slot);
1001 		if (!mappass)
1002 			continue;
1003 		if (radix_tree_exception(mappass)) {
1004 			if (radix_tree_deref_retry(mappass))
1005 				slot = radix_tree_iter_retry(&iter);
1006 		} else {
1007 			radix_tree_delete(&fedata->socketpass_mappings,
1008 					  mappass->id);
1009 			pvcalls_back_release_passive(dev, fedata, mappass);
1010 		}
1011 	}
1012 	up(&fedata->socket_lock);
1013 
1014 	unbind_from_irqhandler(fedata->irq, dev);
1015 	xenbus_unmap_ring_vfree(dev, fedata->sring);
1016 
1017 	list_del(&fedata->list);
1018 	kfree(fedata);
1019 	dev_set_drvdata(&dev->dev, NULL);
1020 
1021 	return 0;
1022 }
1023 
1024 static int pvcalls_back_probe(struct xenbus_device *dev,
1025 			      const struct xenbus_device_id *id)
1026 {
1027 	int err, abort;
1028 	struct xenbus_transaction xbt;
1029 
1030 again:
1031 	abort = 1;
1032 
1033 	err = xenbus_transaction_start(&xbt);
1034 	if (err) {
1035 		pr_warn("%s cannot create xenstore transaction\n", __func__);
1036 		return err;
1037 	}
1038 
1039 	err = xenbus_printf(xbt, dev->nodename, "versions", "%s",
1040 			    PVCALLS_VERSIONS);
1041 	if (err) {
1042 		pr_warn("%s write out 'versions' failed\n", __func__);
1043 		goto abort;
1044 	}
1045 
1046 	err = xenbus_printf(xbt, dev->nodename, "max-page-order", "%u",
1047 			    MAX_RING_ORDER);
1048 	if (err) {
1049 		pr_warn("%s write out 'max-page-order' failed\n", __func__);
1050 		goto abort;
1051 	}
1052 
1053 	err = xenbus_printf(xbt, dev->nodename, "function-calls",
1054 			    XENBUS_FUNCTIONS_CALLS);
1055 	if (err) {
1056 		pr_warn("%s write out 'function-calls' failed\n", __func__);
1057 		goto abort;
1058 	}
1059 
1060 	abort = 0;
1061 abort:
1062 	err = xenbus_transaction_end(xbt, abort);
1063 	if (err) {
1064 		if (err == -EAGAIN && !abort)
1065 			goto again;
1066 		pr_warn("%s cannot complete xenstore transaction\n", __func__);
1067 		return err;
1068 	}
1069 
1070 	if (abort)
1071 		return -EFAULT;
1072 
1073 	xenbus_switch_state(dev, XenbusStateInitWait);
1074 
1075 	return 0;
1076 }
1077 
1078 static void set_backend_state(struct xenbus_device *dev,
1079 			      enum xenbus_state state)
1080 {
1081 	while (dev->state != state) {
1082 		switch (dev->state) {
1083 		case XenbusStateClosed:
1084 			switch (state) {
1085 			case XenbusStateInitWait:
1086 			case XenbusStateConnected:
1087 				xenbus_switch_state(dev, XenbusStateInitWait);
1088 				break;
1089 			case XenbusStateClosing:
1090 				xenbus_switch_state(dev, XenbusStateClosing);
1091 				break;
1092 			default:
1093 				WARN_ON(1);
1094 			}
1095 			break;
1096 		case XenbusStateInitWait:
1097 		case XenbusStateInitialised:
1098 			switch (state) {
1099 			case XenbusStateConnected:
1100 				backend_connect(dev);
1101 				xenbus_switch_state(dev, XenbusStateConnected);
1102 				break;
1103 			case XenbusStateClosing:
1104 			case XenbusStateClosed:
1105 				xenbus_switch_state(dev, XenbusStateClosing);
1106 				break;
1107 			default:
1108 				WARN_ON(1);
1109 			}
1110 			break;
1111 		case XenbusStateConnected:
1112 			switch (state) {
1113 			case XenbusStateInitWait:
1114 			case XenbusStateClosing:
1115 			case XenbusStateClosed:
1116 				down(&pvcalls_back_global.frontends_lock);
1117 				backend_disconnect(dev);
1118 				up(&pvcalls_back_global.frontends_lock);
1119 				xenbus_switch_state(dev, XenbusStateClosing);
1120 				break;
1121 			default:
1122 				WARN_ON(1);
1123 			}
1124 			break;
1125 		case XenbusStateClosing:
1126 			switch (state) {
1127 			case XenbusStateInitWait:
1128 			case XenbusStateConnected:
1129 			case XenbusStateClosed:
1130 				xenbus_switch_state(dev, XenbusStateClosed);
1131 				break;
1132 			default:
1133 				WARN_ON(1);
1134 			}
1135 			break;
1136 		default:
1137 			WARN_ON(1);
1138 		}
1139 	}
1140 }
1141 
1142 static void pvcalls_back_changed(struct xenbus_device *dev,
1143 				 enum xenbus_state frontend_state)
1144 {
1145 	switch (frontend_state) {
1146 	case XenbusStateInitialising:
1147 		set_backend_state(dev, XenbusStateInitWait);
1148 		break;
1149 
1150 	case XenbusStateInitialised:
1151 	case XenbusStateConnected:
1152 		set_backend_state(dev, XenbusStateConnected);
1153 		break;
1154 
1155 	case XenbusStateClosing:
1156 		set_backend_state(dev, XenbusStateClosing);
1157 		break;
1158 
1159 	case XenbusStateClosed:
1160 		set_backend_state(dev, XenbusStateClosed);
1161 		if (xenbus_dev_is_online(dev))
1162 			break;
1163 		device_unregister(&dev->dev);
1164 		break;
1165 	case XenbusStateUnknown:
1166 		set_backend_state(dev, XenbusStateClosed);
1167 		device_unregister(&dev->dev);
1168 		break;
1169 
1170 	default:
1171 		xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
1172 				 frontend_state);
1173 		break;
1174 	}
1175 }
1176 
1177 static int pvcalls_back_remove(struct xenbus_device *dev)
1178 {
1179 	return 0;
1180 }
1181 
1182 static int pvcalls_back_uevent(struct xenbus_device *xdev,
1183 			       struct kobj_uevent_env *env)
1184 {
1185 	return 0;
1186 }
1187 
1188 static const struct xenbus_device_id pvcalls_back_ids[] = {
1189 	{ "pvcalls" },
1190 	{ "" }
1191 };
1192 
1193 static struct xenbus_driver pvcalls_back_driver = {
1194 	.ids = pvcalls_back_ids,
1195 	.probe = pvcalls_back_probe,
1196 	.remove = pvcalls_back_remove,
1197 	.uevent = pvcalls_back_uevent,
1198 	.otherend_changed = pvcalls_back_changed,
1199 };
1200 
1201 static int __init pvcalls_back_init(void)
1202 {
1203 	int ret;
1204 
1205 	if (!xen_domain())
1206 		return -ENODEV;
1207 
1208 	ret = xenbus_register_backend(&pvcalls_back_driver);
1209 	if (ret < 0)
1210 		return ret;
1211 
1212 	sema_init(&pvcalls_back_global.frontends_lock, 1);
1213 	INIT_LIST_HEAD(&pvcalls_back_global.frontends);
1214 	return 0;
1215 }
1216 module_init(pvcalls_back_init);
1217 
1218 static void __exit pvcalls_back_fin(void)
1219 {
1220 	struct pvcalls_fedata *fedata, *nfedata;
1221 
1222 	down(&pvcalls_back_global.frontends_lock);
1223 	list_for_each_entry_safe(fedata, nfedata,
1224 				 &pvcalls_back_global.frontends, list) {
1225 		backend_disconnect(fedata->dev);
1226 	}
1227 	up(&pvcalls_back_global.frontends_lock);
1228 
1229 	xenbus_unregister_driver(&pvcalls_back_driver);
1230 }
1231 
1232 module_exit(pvcalls_back_fin);
1233 
1234 MODULE_DESCRIPTION("Xen PV Calls backend driver");
1235 MODULE_AUTHOR("Stefano Stabellini <sstabellini@kernel.org>");
1236 MODULE_LICENSE("GPL");
1237