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 	atomic_t eoi;
79 	void (*saved_data_ready)(struct sock *sk);
80 	struct pvcalls_ioworker ioworker;
81 };
82 
83 struct sockpass_mapping {
84 	struct list_head list;
85 	struct pvcalls_fedata *fedata;
86 	struct socket *sock;
87 	uint64_t id;
88 	struct xen_pvcalls_request reqcopy;
89 	spinlock_t copy_lock;
90 	struct workqueue_struct *wq;
91 	struct work_struct register_work;
92 	void (*saved_data_ready)(struct sock *sk);
93 };
94 
95 static irqreturn_t pvcalls_back_conn_event(int irq, void *sock_map);
96 static int pvcalls_back_release_active(struct xenbus_device *dev,
97 				       struct pvcalls_fedata *fedata,
98 				       struct sock_mapping *map);
99 
pvcalls_conn_back_read(void * opaque)100 static bool pvcalls_conn_back_read(void *opaque)
101 {
102 	struct sock_mapping *map = (struct sock_mapping *)opaque;
103 	struct msghdr msg;
104 	struct kvec vec[2];
105 	RING_IDX cons, prod, size, wanted, array_size, masked_prod, masked_cons;
106 	int32_t error;
107 	struct pvcalls_data_intf *intf = map->ring;
108 	struct pvcalls_data *data = &map->data;
109 	unsigned long flags;
110 	int ret;
111 
112 	array_size = XEN_FLEX_RING_SIZE(map->ring_order);
113 	cons = intf->in_cons;
114 	prod = intf->in_prod;
115 	error = intf->in_error;
116 	/* read the indexes first, then deal with the data */
117 	virt_mb();
118 
119 	if (error)
120 		return false;
121 
122 	size = pvcalls_queued(prod, cons, array_size);
123 	if (size >= array_size)
124 		return false;
125 	spin_lock_irqsave(&map->sock->sk->sk_receive_queue.lock, flags);
126 	if (skb_queue_empty(&map->sock->sk->sk_receive_queue)) {
127 		atomic_set(&map->read, 0);
128 		spin_unlock_irqrestore(&map->sock->sk->sk_receive_queue.lock,
129 				flags);
130 		return true;
131 	}
132 	spin_unlock_irqrestore(&map->sock->sk->sk_receive_queue.lock, flags);
133 	wanted = array_size - size;
134 	masked_prod = pvcalls_mask(prod, array_size);
135 	masked_cons = pvcalls_mask(cons, array_size);
136 
137 	memset(&msg, 0, sizeof(msg));
138 	if (masked_prod < masked_cons) {
139 		vec[0].iov_base = data->in + masked_prod;
140 		vec[0].iov_len = wanted;
141 		iov_iter_kvec(&msg.msg_iter, ITER_KVEC|WRITE, vec, 1, wanted);
142 	} else {
143 		vec[0].iov_base = data->in + masked_prod;
144 		vec[0].iov_len = array_size - masked_prod;
145 		vec[1].iov_base = data->in;
146 		vec[1].iov_len = wanted - vec[0].iov_len;
147 		iov_iter_kvec(&msg.msg_iter, ITER_KVEC|WRITE, vec, 2, wanted);
148 	}
149 
150 	atomic_set(&map->read, 0);
151 	ret = inet_recvmsg(map->sock, &msg, wanted, MSG_DONTWAIT);
152 	WARN_ON(ret > wanted);
153 	if (ret == -EAGAIN) /* shouldn't happen */
154 		return true;
155 	if (!ret)
156 		ret = -ENOTCONN;
157 	spin_lock_irqsave(&map->sock->sk->sk_receive_queue.lock, flags);
158 	if (ret > 0 && !skb_queue_empty(&map->sock->sk->sk_receive_queue))
159 		atomic_inc(&map->read);
160 	spin_unlock_irqrestore(&map->sock->sk->sk_receive_queue.lock, flags);
161 
162 	/* write the data, then modify the indexes */
163 	virt_wmb();
164 	if (ret < 0) {
165 		atomic_set(&map->read, 0);
166 		intf->in_error = ret;
167 	} else
168 		intf->in_prod = prod + ret;
169 	/* update the indexes, then notify the other end */
170 	virt_wmb();
171 	notify_remote_via_irq(map->irq);
172 
173 	return true;
174 }
175 
pvcalls_conn_back_write(struct sock_mapping * map)176 static bool pvcalls_conn_back_write(struct sock_mapping *map)
177 {
178 	struct pvcalls_data_intf *intf = map->ring;
179 	struct pvcalls_data *data = &map->data;
180 	struct msghdr msg;
181 	struct kvec vec[2];
182 	RING_IDX cons, prod, size, array_size;
183 	int ret;
184 
185 	cons = intf->out_cons;
186 	prod = intf->out_prod;
187 	/* read the indexes before dealing with the data */
188 	virt_mb();
189 
190 	array_size = XEN_FLEX_RING_SIZE(map->ring_order);
191 	size = pvcalls_queued(prod, cons, array_size);
192 	if (size == 0)
193 		return false;
194 
195 	memset(&msg, 0, sizeof(msg));
196 	msg.msg_flags |= MSG_DONTWAIT;
197 	if (pvcalls_mask(prod, array_size) > pvcalls_mask(cons, array_size)) {
198 		vec[0].iov_base = data->out + pvcalls_mask(cons, array_size);
199 		vec[0].iov_len = size;
200 		iov_iter_kvec(&msg.msg_iter, ITER_KVEC|READ, vec, 1, size);
201 	} else {
202 		vec[0].iov_base = data->out + pvcalls_mask(cons, array_size);
203 		vec[0].iov_len = array_size - pvcalls_mask(cons, array_size);
204 		vec[1].iov_base = data->out;
205 		vec[1].iov_len = size - vec[0].iov_len;
206 		iov_iter_kvec(&msg.msg_iter, ITER_KVEC|READ, vec, 2, size);
207 	}
208 
209 	atomic_set(&map->write, 0);
210 	ret = inet_sendmsg(map->sock, &msg, size);
211 	if (ret == -EAGAIN) {
212 		atomic_inc(&map->write);
213 		atomic_inc(&map->io);
214 		return true;
215 	}
216 
217 	/* write the data, then update the indexes */
218 	virt_wmb();
219 	if (ret < 0) {
220 		intf->out_error = ret;
221 	} else {
222 		intf->out_error = 0;
223 		intf->out_cons = cons + ret;
224 		prod = intf->out_prod;
225 	}
226 	/* update the indexes, then notify the other end */
227 	virt_wmb();
228 	if (prod != cons + ret) {
229 		atomic_inc(&map->write);
230 		atomic_inc(&map->io);
231 	}
232 	notify_remote_via_irq(map->irq);
233 
234 	return true;
235 }
236 
pvcalls_back_ioworker(struct work_struct * work)237 static void pvcalls_back_ioworker(struct work_struct *work)
238 {
239 	struct pvcalls_ioworker *ioworker = container_of(work,
240 		struct pvcalls_ioworker, register_work);
241 	struct sock_mapping *map = container_of(ioworker, struct sock_mapping,
242 		ioworker);
243 	unsigned int eoi_flags = XEN_EOI_FLAG_SPURIOUS;
244 
245 	while (atomic_read(&map->io) > 0) {
246 		if (atomic_read(&map->release) > 0) {
247 			atomic_set(&map->release, 0);
248 			return;
249 		}
250 
251 		if (atomic_read(&map->read) > 0 &&
252 		    pvcalls_conn_back_read(map))
253 			eoi_flags = 0;
254 		if (atomic_read(&map->write) > 0 &&
255 		    pvcalls_conn_back_write(map))
256 			eoi_flags = 0;
257 
258 		if (atomic_read(&map->eoi) > 0 && !atomic_read(&map->write)) {
259 			atomic_set(&map->eoi, 0);
260 			xen_irq_lateeoi(map->irq, eoi_flags);
261 			eoi_flags = XEN_EOI_FLAG_SPURIOUS;
262 		}
263 
264 		atomic_dec(&map->io);
265 	}
266 }
267 
pvcalls_back_socket(struct xenbus_device * dev,struct xen_pvcalls_request * req)268 static int pvcalls_back_socket(struct xenbus_device *dev,
269 		struct xen_pvcalls_request *req)
270 {
271 	struct pvcalls_fedata *fedata;
272 	int ret;
273 	struct xen_pvcalls_response *rsp;
274 
275 	fedata = dev_get_drvdata(&dev->dev);
276 
277 	if (req->u.socket.domain != AF_INET ||
278 	    req->u.socket.type != SOCK_STREAM ||
279 	    (req->u.socket.protocol != IPPROTO_IP &&
280 	     req->u.socket.protocol != AF_INET))
281 		ret = -EAFNOSUPPORT;
282 	else
283 		ret = 0;
284 
285 	/* leave the actual socket allocation for later */
286 
287 	rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
288 	rsp->req_id = req->req_id;
289 	rsp->cmd = req->cmd;
290 	rsp->u.socket.id = req->u.socket.id;
291 	rsp->ret = ret;
292 
293 	return 0;
294 }
295 
pvcalls_sk_state_change(struct sock * sock)296 static void pvcalls_sk_state_change(struct sock *sock)
297 {
298 	struct sock_mapping *map = sock->sk_user_data;
299 
300 	if (map == NULL)
301 		return;
302 
303 	atomic_inc(&map->read);
304 	notify_remote_via_irq(map->irq);
305 }
306 
pvcalls_sk_data_ready(struct sock * sock)307 static void pvcalls_sk_data_ready(struct sock *sock)
308 {
309 	struct sock_mapping *map = sock->sk_user_data;
310 	struct pvcalls_ioworker *iow;
311 
312 	if (map == NULL)
313 		return;
314 
315 	iow = &map->ioworker;
316 	atomic_inc(&map->read);
317 	atomic_inc(&map->io);
318 	queue_work(iow->wq, &iow->register_work);
319 }
320 
pvcalls_new_active_socket(struct pvcalls_fedata * fedata,uint64_t id,grant_ref_t ref,uint32_t evtchn,struct socket * sock)321 static struct sock_mapping *pvcalls_new_active_socket(
322 		struct pvcalls_fedata *fedata,
323 		uint64_t id,
324 		grant_ref_t ref,
325 		uint32_t evtchn,
326 		struct socket *sock)
327 {
328 	int ret;
329 	struct sock_mapping *map;
330 	void *page;
331 
332 	map = kzalloc(sizeof(*map), GFP_KERNEL);
333 	if (map == NULL) {
334 		sock_release(sock);
335 		return NULL;
336 	}
337 
338 	map->fedata = fedata;
339 	map->sock = sock;
340 	map->id = id;
341 	map->ref = ref;
342 
343 	ret = xenbus_map_ring_valloc(fedata->dev, &ref, 1, &page);
344 	if (ret < 0)
345 		goto out;
346 	map->ring = page;
347 	map->ring_order = map->ring->ring_order;
348 	/* first read the order, then map the data ring */
349 	virt_rmb();
350 	if (map->ring_order > MAX_RING_ORDER) {
351 		pr_warn("%s frontend requested ring_order %u, which is > MAX (%u)\n",
352 				__func__, map->ring_order, MAX_RING_ORDER);
353 		goto out;
354 	}
355 	ret = xenbus_map_ring_valloc(fedata->dev, map->ring->ref,
356 				     (1 << map->ring_order), &page);
357 	if (ret < 0)
358 		goto out;
359 	map->bytes = page;
360 
361 	ret = bind_interdomain_evtchn_to_irqhandler_lateeoi(
362 			fedata->dev->otherend_id, evtchn,
363 			pvcalls_back_conn_event, 0, "pvcalls-backend", map);
364 	if (ret < 0)
365 		goto out;
366 	map->irq = ret;
367 
368 	map->data.in = map->bytes;
369 	map->data.out = map->bytes + XEN_FLEX_RING_SIZE(map->ring_order);
370 
371 	map->ioworker.wq = alloc_workqueue("pvcalls_io", WQ_UNBOUND, 1);
372 	if (!map->ioworker.wq)
373 		goto out;
374 	atomic_set(&map->io, 1);
375 	INIT_WORK(&map->ioworker.register_work,	pvcalls_back_ioworker);
376 
377 	down(&fedata->socket_lock);
378 	list_add_tail(&map->list, &fedata->socket_mappings);
379 	up(&fedata->socket_lock);
380 
381 	write_lock_bh(&map->sock->sk->sk_callback_lock);
382 	map->saved_data_ready = map->sock->sk->sk_data_ready;
383 	map->sock->sk->sk_user_data = map;
384 	map->sock->sk->sk_data_ready = pvcalls_sk_data_ready;
385 	map->sock->sk->sk_state_change = pvcalls_sk_state_change;
386 	write_unlock_bh(&map->sock->sk->sk_callback_lock);
387 
388 	return map;
389 out:
390 	down(&fedata->socket_lock);
391 	list_del(&map->list);
392 	pvcalls_back_release_active(fedata->dev, fedata, map);
393 	up(&fedata->socket_lock);
394 	return NULL;
395 }
396 
pvcalls_back_connect(struct xenbus_device * dev,struct xen_pvcalls_request * req)397 static int pvcalls_back_connect(struct xenbus_device *dev,
398 				struct xen_pvcalls_request *req)
399 {
400 	struct pvcalls_fedata *fedata;
401 	int ret = -EINVAL;
402 	struct socket *sock;
403 	struct sock_mapping *map;
404 	struct xen_pvcalls_response *rsp;
405 	struct sockaddr *sa = (struct sockaddr *)&req->u.connect.addr;
406 
407 	fedata = dev_get_drvdata(&dev->dev);
408 
409 	if (req->u.connect.len < sizeof(sa->sa_family) ||
410 	    req->u.connect.len > sizeof(req->u.connect.addr) ||
411 	    sa->sa_family != AF_INET)
412 		goto out;
413 
414 	ret = sock_create(AF_INET, SOCK_STREAM, 0, &sock);
415 	if (ret < 0)
416 		goto out;
417 	ret = inet_stream_connect(sock, sa, req->u.connect.len, 0);
418 	if (ret < 0) {
419 		sock_release(sock);
420 		goto out;
421 	}
422 
423 	map = pvcalls_new_active_socket(fedata,
424 					req->u.connect.id,
425 					req->u.connect.ref,
426 					req->u.connect.evtchn,
427 					sock);
428 	if (!map)
429 		ret = -EFAULT;
430 
431 out:
432 	rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
433 	rsp->req_id = req->req_id;
434 	rsp->cmd = req->cmd;
435 	rsp->u.connect.id = req->u.connect.id;
436 	rsp->ret = ret;
437 
438 	return 0;
439 }
440 
pvcalls_back_release_active(struct xenbus_device * dev,struct pvcalls_fedata * fedata,struct sock_mapping * map)441 static int pvcalls_back_release_active(struct xenbus_device *dev,
442 				       struct pvcalls_fedata *fedata,
443 				       struct sock_mapping *map)
444 {
445 	disable_irq(map->irq);
446 	if (map->sock->sk != NULL) {
447 		write_lock_bh(&map->sock->sk->sk_callback_lock);
448 		map->sock->sk->sk_user_data = NULL;
449 		map->sock->sk->sk_data_ready = map->saved_data_ready;
450 		write_unlock_bh(&map->sock->sk->sk_callback_lock);
451 	}
452 
453 	atomic_set(&map->release, 1);
454 	flush_work(&map->ioworker.register_work);
455 
456 	xenbus_unmap_ring_vfree(dev, map->bytes);
457 	xenbus_unmap_ring_vfree(dev, (void *)map->ring);
458 	unbind_from_irqhandler(map->irq, map);
459 
460 	sock_release(map->sock);
461 	kfree(map);
462 
463 	return 0;
464 }
465 
pvcalls_back_release_passive(struct xenbus_device * dev,struct pvcalls_fedata * fedata,struct sockpass_mapping * mappass)466 static int pvcalls_back_release_passive(struct xenbus_device *dev,
467 					struct pvcalls_fedata *fedata,
468 					struct sockpass_mapping *mappass)
469 {
470 	if (mappass->sock->sk != NULL) {
471 		write_lock_bh(&mappass->sock->sk->sk_callback_lock);
472 		mappass->sock->sk->sk_user_data = NULL;
473 		mappass->sock->sk->sk_data_ready = mappass->saved_data_ready;
474 		write_unlock_bh(&mappass->sock->sk->sk_callback_lock);
475 	}
476 	sock_release(mappass->sock);
477 	flush_workqueue(mappass->wq);
478 	destroy_workqueue(mappass->wq);
479 	kfree(mappass);
480 
481 	return 0;
482 }
483 
pvcalls_back_release(struct xenbus_device * dev,struct xen_pvcalls_request * req)484 static int pvcalls_back_release(struct xenbus_device *dev,
485 				struct xen_pvcalls_request *req)
486 {
487 	struct pvcalls_fedata *fedata;
488 	struct sock_mapping *map, *n;
489 	struct sockpass_mapping *mappass;
490 	int ret = 0;
491 	struct xen_pvcalls_response *rsp;
492 
493 	fedata = dev_get_drvdata(&dev->dev);
494 
495 	down(&fedata->socket_lock);
496 	list_for_each_entry_safe(map, n, &fedata->socket_mappings, list) {
497 		if (map->id == req->u.release.id) {
498 			list_del(&map->list);
499 			up(&fedata->socket_lock);
500 			ret = pvcalls_back_release_active(dev, fedata, map);
501 			goto out;
502 		}
503 	}
504 	mappass = radix_tree_lookup(&fedata->socketpass_mappings,
505 				    req->u.release.id);
506 	if (mappass != NULL) {
507 		radix_tree_delete(&fedata->socketpass_mappings, mappass->id);
508 		up(&fedata->socket_lock);
509 		ret = pvcalls_back_release_passive(dev, fedata, mappass);
510 	} else
511 		up(&fedata->socket_lock);
512 
513 out:
514 	rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
515 	rsp->req_id = req->req_id;
516 	rsp->u.release.id = req->u.release.id;
517 	rsp->cmd = req->cmd;
518 	rsp->ret = ret;
519 	return 0;
520 }
521 
__pvcalls_back_accept(struct work_struct * work)522 static void __pvcalls_back_accept(struct work_struct *work)
523 {
524 	struct sockpass_mapping *mappass = container_of(
525 		work, struct sockpass_mapping, register_work);
526 	struct sock_mapping *map;
527 	struct pvcalls_ioworker *iow;
528 	struct pvcalls_fedata *fedata;
529 	struct socket *sock;
530 	struct xen_pvcalls_response *rsp;
531 	struct xen_pvcalls_request *req;
532 	int notify;
533 	int ret = -EINVAL;
534 	unsigned long flags;
535 
536 	fedata = mappass->fedata;
537 	/*
538 	 * __pvcalls_back_accept can race against pvcalls_back_accept.
539 	 * We only need to check the value of "cmd" on read. It could be
540 	 * done atomically, but to simplify the code on the write side, we
541 	 * use a spinlock.
542 	 */
543 	spin_lock_irqsave(&mappass->copy_lock, flags);
544 	req = &mappass->reqcopy;
545 	if (req->cmd != PVCALLS_ACCEPT) {
546 		spin_unlock_irqrestore(&mappass->copy_lock, flags);
547 		return;
548 	}
549 	spin_unlock_irqrestore(&mappass->copy_lock, flags);
550 
551 	sock = sock_alloc();
552 	if (sock == NULL)
553 		goto out_error;
554 	sock->type = mappass->sock->type;
555 	sock->ops = mappass->sock->ops;
556 
557 	ret = inet_accept(mappass->sock, sock, O_NONBLOCK, true);
558 	if (ret == -EAGAIN) {
559 		sock_release(sock);
560 		return;
561 	}
562 
563 	map = pvcalls_new_active_socket(fedata,
564 					req->u.accept.id_new,
565 					req->u.accept.ref,
566 					req->u.accept.evtchn,
567 					sock);
568 	if (!map) {
569 		ret = -EFAULT;
570 		goto out_error;
571 	}
572 
573 	map->sockpass = mappass;
574 	iow = &map->ioworker;
575 	atomic_inc(&map->read);
576 	atomic_inc(&map->io);
577 	queue_work(iow->wq, &iow->register_work);
578 
579 out_error:
580 	rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
581 	rsp->req_id = req->req_id;
582 	rsp->cmd = req->cmd;
583 	rsp->u.accept.id = req->u.accept.id;
584 	rsp->ret = ret;
585 	RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&fedata->ring, notify);
586 	if (notify)
587 		notify_remote_via_irq(fedata->irq);
588 
589 	mappass->reqcopy.cmd = 0;
590 }
591 
pvcalls_pass_sk_data_ready(struct sock * sock)592 static void pvcalls_pass_sk_data_ready(struct sock *sock)
593 {
594 	struct sockpass_mapping *mappass = sock->sk_user_data;
595 	struct pvcalls_fedata *fedata;
596 	struct xen_pvcalls_response *rsp;
597 	unsigned long flags;
598 	int notify;
599 
600 	if (mappass == NULL)
601 		return;
602 
603 	fedata = mappass->fedata;
604 	spin_lock_irqsave(&mappass->copy_lock, flags);
605 	if (mappass->reqcopy.cmd == PVCALLS_POLL) {
606 		rsp = RING_GET_RESPONSE(&fedata->ring,
607 					fedata->ring.rsp_prod_pvt++);
608 		rsp->req_id = mappass->reqcopy.req_id;
609 		rsp->u.poll.id = mappass->reqcopy.u.poll.id;
610 		rsp->cmd = mappass->reqcopy.cmd;
611 		rsp->ret = 0;
612 
613 		mappass->reqcopy.cmd = 0;
614 		spin_unlock_irqrestore(&mappass->copy_lock, flags);
615 
616 		RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&fedata->ring, notify);
617 		if (notify)
618 			notify_remote_via_irq(mappass->fedata->irq);
619 	} else {
620 		spin_unlock_irqrestore(&mappass->copy_lock, flags);
621 		queue_work(mappass->wq, &mappass->register_work);
622 	}
623 }
624 
pvcalls_back_bind(struct xenbus_device * dev,struct xen_pvcalls_request * req)625 static int pvcalls_back_bind(struct xenbus_device *dev,
626 			     struct xen_pvcalls_request *req)
627 {
628 	struct pvcalls_fedata *fedata;
629 	int ret;
630 	struct sockpass_mapping *map;
631 	struct xen_pvcalls_response *rsp;
632 
633 	fedata = dev_get_drvdata(&dev->dev);
634 
635 	map = kzalloc(sizeof(*map), GFP_KERNEL);
636 	if (map == NULL) {
637 		ret = -ENOMEM;
638 		goto out;
639 	}
640 
641 	INIT_WORK(&map->register_work, __pvcalls_back_accept);
642 	spin_lock_init(&map->copy_lock);
643 	map->wq = alloc_workqueue("pvcalls_wq", WQ_UNBOUND, 1);
644 	if (!map->wq) {
645 		ret = -ENOMEM;
646 		goto out;
647 	}
648 
649 	ret = sock_create(AF_INET, SOCK_STREAM, 0, &map->sock);
650 	if (ret < 0)
651 		goto out;
652 
653 	ret = inet_bind(map->sock, (struct sockaddr *)&req->u.bind.addr,
654 			req->u.bind.len);
655 	if (ret < 0)
656 		goto out;
657 
658 	map->fedata = fedata;
659 	map->id = req->u.bind.id;
660 
661 	down(&fedata->socket_lock);
662 	ret = radix_tree_insert(&fedata->socketpass_mappings, map->id,
663 				map);
664 	up(&fedata->socket_lock);
665 	if (ret)
666 		goto out;
667 
668 	write_lock_bh(&map->sock->sk->sk_callback_lock);
669 	map->saved_data_ready = map->sock->sk->sk_data_ready;
670 	map->sock->sk->sk_user_data = map;
671 	map->sock->sk->sk_data_ready = pvcalls_pass_sk_data_ready;
672 	write_unlock_bh(&map->sock->sk->sk_callback_lock);
673 
674 out:
675 	if (ret) {
676 		if (map && map->sock)
677 			sock_release(map->sock);
678 		if (map && map->wq)
679 			destroy_workqueue(map->wq);
680 		kfree(map);
681 	}
682 	rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
683 	rsp->req_id = req->req_id;
684 	rsp->cmd = req->cmd;
685 	rsp->u.bind.id = req->u.bind.id;
686 	rsp->ret = ret;
687 	return 0;
688 }
689 
pvcalls_back_listen(struct xenbus_device * dev,struct xen_pvcalls_request * req)690 static int pvcalls_back_listen(struct xenbus_device *dev,
691 			       struct xen_pvcalls_request *req)
692 {
693 	struct pvcalls_fedata *fedata;
694 	int ret = -EINVAL;
695 	struct sockpass_mapping *map;
696 	struct xen_pvcalls_response *rsp;
697 
698 	fedata = dev_get_drvdata(&dev->dev);
699 
700 	down(&fedata->socket_lock);
701 	map = radix_tree_lookup(&fedata->socketpass_mappings, req->u.listen.id);
702 	up(&fedata->socket_lock);
703 	if (map == NULL)
704 		goto out;
705 
706 	ret = inet_listen(map->sock, req->u.listen.backlog);
707 
708 out:
709 	rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
710 	rsp->req_id = req->req_id;
711 	rsp->cmd = req->cmd;
712 	rsp->u.listen.id = req->u.listen.id;
713 	rsp->ret = ret;
714 	return 0;
715 }
716 
pvcalls_back_accept(struct xenbus_device * dev,struct xen_pvcalls_request * req)717 static int pvcalls_back_accept(struct xenbus_device *dev,
718 			       struct xen_pvcalls_request *req)
719 {
720 	struct pvcalls_fedata *fedata;
721 	struct sockpass_mapping *mappass;
722 	int ret = -EINVAL;
723 	struct xen_pvcalls_response *rsp;
724 	unsigned long flags;
725 
726 	fedata = dev_get_drvdata(&dev->dev);
727 
728 	down(&fedata->socket_lock);
729 	mappass = radix_tree_lookup(&fedata->socketpass_mappings,
730 		req->u.accept.id);
731 	up(&fedata->socket_lock);
732 	if (mappass == NULL)
733 		goto out_error;
734 
735 	/*
736 	 * Limitation of the current implementation: only support one
737 	 * concurrent accept or poll call on one socket.
738 	 */
739 	spin_lock_irqsave(&mappass->copy_lock, flags);
740 	if (mappass->reqcopy.cmd != 0) {
741 		spin_unlock_irqrestore(&mappass->copy_lock, flags);
742 		ret = -EINTR;
743 		goto out_error;
744 	}
745 
746 	mappass->reqcopy = *req;
747 	spin_unlock_irqrestore(&mappass->copy_lock, flags);
748 	queue_work(mappass->wq, &mappass->register_work);
749 
750 	/* Tell the caller we don't need to send back a notification yet */
751 	return -1;
752 
753 out_error:
754 	rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
755 	rsp->req_id = req->req_id;
756 	rsp->cmd = req->cmd;
757 	rsp->u.accept.id = req->u.accept.id;
758 	rsp->ret = ret;
759 	return 0;
760 }
761 
pvcalls_back_poll(struct xenbus_device * dev,struct xen_pvcalls_request * req)762 static int pvcalls_back_poll(struct xenbus_device *dev,
763 			     struct xen_pvcalls_request *req)
764 {
765 	struct pvcalls_fedata *fedata;
766 	struct sockpass_mapping *mappass;
767 	struct xen_pvcalls_response *rsp;
768 	struct inet_connection_sock *icsk;
769 	struct request_sock_queue *queue;
770 	unsigned long flags;
771 	int ret;
772 	bool data;
773 
774 	fedata = dev_get_drvdata(&dev->dev);
775 
776 	down(&fedata->socket_lock);
777 	mappass = radix_tree_lookup(&fedata->socketpass_mappings,
778 				    req->u.poll.id);
779 	up(&fedata->socket_lock);
780 	if (mappass == NULL)
781 		return -EINVAL;
782 
783 	/*
784 	 * Limitation of the current implementation: only support one
785 	 * concurrent accept or poll call on one socket.
786 	 */
787 	spin_lock_irqsave(&mappass->copy_lock, flags);
788 	if (mappass->reqcopy.cmd != 0) {
789 		ret = -EINTR;
790 		goto out;
791 	}
792 
793 	mappass->reqcopy = *req;
794 	icsk = inet_csk(mappass->sock->sk);
795 	queue = &icsk->icsk_accept_queue;
796 	data = READ_ONCE(queue->rskq_accept_head) != NULL;
797 	if (data) {
798 		mappass->reqcopy.cmd = 0;
799 		ret = 0;
800 		goto out;
801 	}
802 	spin_unlock_irqrestore(&mappass->copy_lock, flags);
803 
804 	/* Tell the caller we don't need to send back a notification yet */
805 	return -1;
806 
807 out:
808 	spin_unlock_irqrestore(&mappass->copy_lock, flags);
809 
810 	rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
811 	rsp->req_id = req->req_id;
812 	rsp->cmd = req->cmd;
813 	rsp->u.poll.id = req->u.poll.id;
814 	rsp->ret = ret;
815 	return 0;
816 }
817 
pvcalls_back_handle_cmd(struct xenbus_device * dev,struct xen_pvcalls_request * req)818 static int pvcalls_back_handle_cmd(struct xenbus_device *dev,
819 				   struct xen_pvcalls_request *req)
820 {
821 	int ret = 0;
822 
823 	switch (req->cmd) {
824 	case PVCALLS_SOCKET:
825 		ret = pvcalls_back_socket(dev, req);
826 		break;
827 	case PVCALLS_CONNECT:
828 		ret = pvcalls_back_connect(dev, req);
829 		break;
830 	case PVCALLS_RELEASE:
831 		ret = pvcalls_back_release(dev, req);
832 		break;
833 	case PVCALLS_BIND:
834 		ret = pvcalls_back_bind(dev, req);
835 		break;
836 	case PVCALLS_LISTEN:
837 		ret = pvcalls_back_listen(dev, req);
838 		break;
839 	case PVCALLS_ACCEPT:
840 		ret = pvcalls_back_accept(dev, req);
841 		break;
842 	case PVCALLS_POLL:
843 		ret = pvcalls_back_poll(dev, req);
844 		break;
845 	default:
846 	{
847 		struct pvcalls_fedata *fedata;
848 		struct xen_pvcalls_response *rsp;
849 
850 		fedata = dev_get_drvdata(&dev->dev);
851 		rsp = RING_GET_RESPONSE(
852 				&fedata->ring, fedata->ring.rsp_prod_pvt++);
853 		rsp->req_id = req->req_id;
854 		rsp->cmd = req->cmd;
855 		rsp->ret = -ENOTSUPP;
856 		break;
857 	}
858 	}
859 	return ret;
860 }
861 
pvcalls_back_work(struct pvcalls_fedata * fedata)862 static void pvcalls_back_work(struct pvcalls_fedata *fedata)
863 {
864 	int notify, notify_all = 0, more = 1;
865 	struct xen_pvcalls_request req;
866 	struct xenbus_device *dev = fedata->dev;
867 
868 	while (more) {
869 		while (RING_HAS_UNCONSUMED_REQUESTS(&fedata->ring)) {
870 			RING_COPY_REQUEST(&fedata->ring,
871 					  fedata->ring.req_cons++,
872 					  &req);
873 
874 			if (!pvcalls_back_handle_cmd(dev, &req)) {
875 				RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(
876 					&fedata->ring, notify);
877 				notify_all += notify;
878 			}
879 		}
880 
881 		if (notify_all) {
882 			notify_remote_via_irq(fedata->irq);
883 			notify_all = 0;
884 		}
885 
886 		RING_FINAL_CHECK_FOR_REQUESTS(&fedata->ring, more);
887 	}
888 }
889 
pvcalls_back_event(int irq,void * dev_id)890 static irqreturn_t pvcalls_back_event(int irq, void *dev_id)
891 {
892 	struct xenbus_device *dev = dev_id;
893 	struct pvcalls_fedata *fedata = NULL;
894 	unsigned int eoi_flags = XEN_EOI_FLAG_SPURIOUS;
895 
896 	if (dev) {
897 		fedata = dev_get_drvdata(&dev->dev);
898 		if (fedata) {
899 			pvcalls_back_work(fedata);
900 			eoi_flags = 0;
901 		}
902 	}
903 
904 	xen_irq_lateeoi(irq, eoi_flags);
905 
906 	return IRQ_HANDLED;
907 }
908 
pvcalls_back_conn_event(int irq,void * sock_map)909 static irqreturn_t pvcalls_back_conn_event(int irq, void *sock_map)
910 {
911 	struct sock_mapping *map = sock_map;
912 	struct pvcalls_ioworker *iow;
913 
914 	if (map == NULL || map->sock == NULL || map->sock->sk == NULL ||
915 		map->sock->sk->sk_user_data != map) {
916 		xen_irq_lateeoi(irq, 0);
917 		return IRQ_HANDLED;
918 	}
919 
920 	iow = &map->ioworker;
921 
922 	atomic_inc(&map->write);
923 	atomic_inc(&map->eoi);
924 	atomic_inc(&map->io);
925 	queue_work(iow->wq, &iow->register_work);
926 
927 	return IRQ_HANDLED;
928 }
929 
backend_connect(struct xenbus_device * dev)930 static int backend_connect(struct xenbus_device *dev)
931 {
932 	int err, evtchn;
933 	grant_ref_t ring_ref;
934 	struct pvcalls_fedata *fedata = NULL;
935 
936 	fedata = kzalloc(sizeof(struct pvcalls_fedata), GFP_KERNEL);
937 	if (!fedata)
938 		return -ENOMEM;
939 
940 	fedata->irq = -1;
941 	err = xenbus_scanf(XBT_NIL, dev->otherend, "port", "%u",
942 			   &evtchn);
943 	if (err != 1) {
944 		err = -EINVAL;
945 		xenbus_dev_fatal(dev, err, "reading %s/event-channel",
946 				 dev->otherend);
947 		goto error;
948 	}
949 
950 	err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-ref", "%u", &ring_ref);
951 	if (err != 1) {
952 		err = -EINVAL;
953 		xenbus_dev_fatal(dev, err, "reading %s/ring-ref",
954 				 dev->otherend);
955 		goto error;
956 	}
957 
958 	err = bind_interdomain_evtchn_to_irq_lateeoi(dev->otherend_id, evtchn);
959 	if (err < 0)
960 		goto error;
961 	fedata->irq = err;
962 
963 	err = request_threaded_irq(fedata->irq, NULL, pvcalls_back_event,
964 				   IRQF_ONESHOT, "pvcalls-back", dev);
965 	if (err < 0)
966 		goto error;
967 
968 	err = xenbus_map_ring_valloc(dev, &ring_ref, 1,
969 				     (void **)&fedata->sring);
970 	if (err < 0)
971 		goto error;
972 
973 	BACK_RING_INIT(&fedata->ring, fedata->sring, XEN_PAGE_SIZE * 1);
974 	fedata->dev = dev;
975 
976 	INIT_LIST_HEAD(&fedata->socket_mappings);
977 	INIT_RADIX_TREE(&fedata->socketpass_mappings, GFP_KERNEL);
978 	sema_init(&fedata->socket_lock, 1);
979 	dev_set_drvdata(&dev->dev, fedata);
980 
981 	down(&pvcalls_back_global.frontends_lock);
982 	list_add_tail(&fedata->list, &pvcalls_back_global.frontends);
983 	up(&pvcalls_back_global.frontends_lock);
984 
985 	return 0;
986 
987  error:
988 	if (fedata->irq >= 0)
989 		unbind_from_irqhandler(fedata->irq, dev);
990 	if (fedata->sring != NULL)
991 		xenbus_unmap_ring_vfree(dev, fedata->sring);
992 	kfree(fedata);
993 	return err;
994 }
995 
backend_disconnect(struct xenbus_device * dev)996 static int backend_disconnect(struct xenbus_device *dev)
997 {
998 	struct pvcalls_fedata *fedata;
999 	struct sock_mapping *map, *n;
1000 	struct sockpass_mapping *mappass;
1001 	struct radix_tree_iter iter;
1002 	void **slot;
1003 
1004 
1005 	fedata = dev_get_drvdata(&dev->dev);
1006 
1007 	down(&fedata->socket_lock);
1008 	list_for_each_entry_safe(map, n, &fedata->socket_mappings, list) {
1009 		list_del(&map->list);
1010 		pvcalls_back_release_active(dev, fedata, map);
1011 	}
1012 
1013 	radix_tree_for_each_slot(slot, &fedata->socketpass_mappings, &iter, 0) {
1014 		mappass = radix_tree_deref_slot(slot);
1015 		if (!mappass)
1016 			continue;
1017 		if (radix_tree_exception(mappass)) {
1018 			if (radix_tree_deref_retry(mappass))
1019 				slot = radix_tree_iter_retry(&iter);
1020 		} else {
1021 			radix_tree_delete(&fedata->socketpass_mappings,
1022 					  mappass->id);
1023 			pvcalls_back_release_passive(dev, fedata, mappass);
1024 		}
1025 	}
1026 	up(&fedata->socket_lock);
1027 
1028 	unbind_from_irqhandler(fedata->irq, dev);
1029 	xenbus_unmap_ring_vfree(dev, fedata->sring);
1030 
1031 	list_del(&fedata->list);
1032 	kfree(fedata);
1033 	dev_set_drvdata(&dev->dev, NULL);
1034 
1035 	return 0;
1036 }
1037 
pvcalls_back_probe(struct xenbus_device * dev,const struct xenbus_device_id * id)1038 static int pvcalls_back_probe(struct xenbus_device *dev,
1039 			      const struct xenbus_device_id *id)
1040 {
1041 	int err, abort;
1042 	struct xenbus_transaction xbt;
1043 
1044 again:
1045 	abort = 1;
1046 
1047 	err = xenbus_transaction_start(&xbt);
1048 	if (err) {
1049 		pr_warn("%s cannot create xenstore transaction\n", __func__);
1050 		return err;
1051 	}
1052 
1053 	err = xenbus_printf(xbt, dev->nodename, "versions", "%s",
1054 			    PVCALLS_VERSIONS);
1055 	if (err) {
1056 		pr_warn("%s write out 'versions' failed\n", __func__);
1057 		goto abort;
1058 	}
1059 
1060 	err = xenbus_printf(xbt, dev->nodename, "max-page-order", "%u",
1061 			    MAX_RING_ORDER);
1062 	if (err) {
1063 		pr_warn("%s write out 'max-page-order' failed\n", __func__);
1064 		goto abort;
1065 	}
1066 
1067 	err = xenbus_printf(xbt, dev->nodename, "function-calls",
1068 			    XENBUS_FUNCTIONS_CALLS);
1069 	if (err) {
1070 		pr_warn("%s write out 'function-calls' failed\n", __func__);
1071 		goto abort;
1072 	}
1073 
1074 	abort = 0;
1075 abort:
1076 	err = xenbus_transaction_end(xbt, abort);
1077 	if (err) {
1078 		if (err == -EAGAIN && !abort)
1079 			goto again;
1080 		pr_warn("%s cannot complete xenstore transaction\n", __func__);
1081 		return err;
1082 	}
1083 
1084 	if (abort)
1085 		return -EFAULT;
1086 
1087 	xenbus_switch_state(dev, XenbusStateInitWait);
1088 
1089 	return 0;
1090 }
1091 
set_backend_state(struct xenbus_device * dev,enum xenbus_state state)1092 static void set_backend_state(struct xenbus_device *dev,
1093 			      enum xenbus_state state)
1094 {
1095 	while (dev->state != state) {
1096 		switch (dev->state) {
1097 		case XenbusStateClosed:
1098 			switch (state) {
1099 			case XenbusStateInitWait:
1100 			case XenbusStateConnected:
1101 				xenbus_switch_state(dev, XenbusStateInitWait);
1102 				break;
1103 			case XenbusStateClosing:
1104 				xenbus_switch_state(dev, XenbusStateClosing);
1105 				break;
1106 			default:
1107 				WARN_ON(1);
1108 			}
1109 			break;
1110 		case XenbusStateInitWait:
1111 		case XenbusStateInitialised:
1112 			switch (state) {
1113 			case XenbusStateConnected:
1114 				if (backend_connect(dev))
1115 					return;
1116 				xenbus_switch_state(dev, XenbusStateConnected);
1117 				break;
1118 			case XenbusStateClosing:
1119 			case XenbusStateClosed:
1120 				xenbus_switch_state(dev, XenbusStateClosing);
1121 				break;
1122 			default:
1123 				WARN_ON(1);
1124 			}
1125 			break;
1126 		case XenbusStateConnected:
1127 			switch (state) {
1128 			case XenbusStateInitWait:
1129 			case XenbusStateClosing:
1130 			case XenbusStateClosed:
1131 				down(&pvcalls_back_global.frontends_lock);
1132 				backend_disconnect(dev);
1133 				up(&pvcalls_back_global.frontends_lock);
1134 				xenbus_switch_state(dev, XenbusStateClosing);
1135 				break;
1136 			default:
1137 				WARN_ON(1);
1138 			}
1139 			break;
1140 		case XenbusStateClosing:
1141 			switch (state) {
1142 			case XenbusStateInitWait:
1143 			case XenbusStateConnected:
1144 			case XenbusStateClosed:
1145 				xenbus_switch_state(dev, XenbusStateClosed);
1146 				break;
1147 			default:
1148 				WARN_ON(1);
1149 			}
1150 			break;
1151 		default:
1152 			WARN_ON(1);
1153 		}
1154 	}
1155 }
1156 
pvcalls_back_changed(struct xenbus_device * dev,enum xenbus_state frontend_state)1157 static void pvcalls_back_changed(struct xenbus_device *dev,
1158 				 enum xenbus_state frontend_state)
1159 {
1160 	switch (frontend_state) {
1161 	case XenbusStateInitialising:
1162 		set_backend_state(dev, XenbusStateInitWait);
1163 		break;
1164 
1165 	case XenbusStateInitialised:
1166 	case XenbusStateConnected:
1167 		set_backend_state(dev, XenbusStateConnected);
1168 		break;
1169 
1170 	case XenbusStateClosing:
1171 		set_backend_state(dev, XenbusStateClosing);
1172 		break;
1173 
1174 	case XenbusStateClosed:
1175 		set_backend_state(dev, XenbusStateClosed);
1176 		if (xenbus_dev_is_online(dev))
1177 			break;
1178 		device_unregister(&dev->dev);
1179 		break;
1180 	case XenbusStateUnknown:
1181 		set_backend_state(dev, XenbusStateClosed);
1182 		device_unregister(&dev->dev);
1183 		break;
1184 
1185 	default:
1186 		xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
1187 				 frontend_state);
1188 		break;
1189 	}
1190 }
1191 
pvcalls_back_remove(struct xenbus_device * dev)1192 static int pvcalls_back_remove(struct xenbus_device *dev)
1193 {
1194 	return 0;
1195 }
1196 
pvcalls_back_uevent(struct xenbus_device * xdev,struct kobj_uevent_env * env)1197 static int pvcalls_back_uevent(struct xenbus_device *xdev,
1198 			       struct kobj_uevent_env *env)
1199 {
1200 	return 0;
1201 }
1202 
1203 static const struct xenbus_device_id pvcalls_back_ids[] = {
1204 	{ "pvcalls" },
1205 	{ "" }
1206 };
1207 
1208 static struct xenbus_driver pvcalls_back_driver = {
1209 	.ids = pvcalls_back_ids,
1210 	.probe = pvcalls_back_probe,
1211 	.remove = pvcalls_back_remove,
1212 	.uevent = pvcalls_back_uevent,
1213 	.otherend_changed = pvcalls_back_changed,
1214 };
1215 
pvcalls_back_init(void)1216 static int __init pvcalls_back_init(void)
1217 {
1218 	int ret;
1219 
1220 	if (!xen_domain())
1221 		return -ENODEV;
1222 
1223 	ret = xenbus_register_backend(&pvcalls_back_driver);
1224 	if (ret < 0)
1225 		return ret;
1226 
1227 	sema_init(&pvcalls_back_global.frontends_lock, 1);
1228 	INIT_LIST_HEAD(&pvcalls_back_global.frontends);
1229 	return 0;
1230 }
1231 module_init(pvcalls_back_init);
1232 
pvcalls_back_fin(void)1233 static void __exit pvcalls_back_fin(void)
1234 {
1235 	struct pvcalls_fedata *fedata, *nfedata;
1236 
1237 	down(&pvcalls_back_global.frontends_lock);
1238 	list_for_each_entry_safe(fedata, nfedata,
1239 				 &pvcalls_back_global.frontends, list) {
1240 		backend_disconnect(fedata->dev);
1241 	}
1242 	up(&pvcalls_back_global.frontends_lock);
1243 
1244 	xenbus_unregister_driver(&pvcalls_back_driver);
1245 }
1246 
1247 module_exit(pvcalls_back_fin);
1248 
1249 MODULE_DESCRIPTION("Xen PV Calls backend driver");
1250 MODULE_AUTHOR("Stefano Stabellini <sstabellini@kernel.org>");
1251 MODULE_LICENSE("GPL");
1252