]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - drivers/xen/pvcalls-back.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
xen/pvcalls: implement read
[mirror_ubuntu-bionic-kernel.git] / drivers / xen / pvcalls-back.c
CommitLineData
72e59c30
SS		 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
fb029875 15#include <linux/inet.h>
72e59c30
SS		 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>
fb029875
SS		 22#include <net/sock.h>
23#include <net/inet_common.h>
24#include <net/inet_connection_sock.h>
25#include <net/request_sock.h>
72e59c30
SS		 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
0a9c75c2
SS		 33#define PVCALLS_VERSIONS "1"
34#define MAX_RING_ORDER XENBUS_MAX_RING_GRANT_ORDER
35
9be07334
SS		 36struct pvcalls_back_global {
37 struct list_head frontends;
38 struct semaphore frontends_lock;
39} pvcalls_back_global;
40
d0e4d560
SS		 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 */
46struct 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
5db4d286
SS		 57struct pvcalls_ioworker {
58 struct work_struct register_work;
59 struct workqueue_struct *wq;
60};
61
62struct sock_mapping {
63 struct list_head list;
64 struct pvcalls_fedata *fedata;
6f474e71 65 struct sockpass_mapping *sockpass;
5db4d286
SS		 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
331a63e6
SS		 82struct 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
5db4d286
SS		 94static irqreturn_t pvcalls_back_conn_event(int irq, void *sock_map);
95static int pvcalls_back_release_active(struct xenbus_device *dev,
96 struct pvcalls_fedata *fedata,
97 struct sock_mapping *map);
98
5d520d85
SS		 99static void pvcalls_conn_back_read(void *opaque)
100{
b3f9f773
SS		 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 msg.msg_iter.type = ITER_KVEC|WRITE;
138 msg.msg_iter.count = wanted;
139 if (masked_prod < masked_cons) {
140 vec[0].iov_base = data->in + masked_prod;
141 vec[0].iov_len = wanted;
142 msg.msg_iter.kvec = vec;
143 msg.msg_iter.nr_segs = 1;
144 } else {
145 vec[0].iov_base = data->in + masked_prod;
146 vec[0].iov_len = array_size - masked_prod;
147 vec[1].iov_base = data->in;
148 vec[1].iov_len = wanted - vec[0].iov_len;
149 msg.msg_iter.kvec = vec;
150 msg.msg_iter.nr_segs = 2;
151 }
152
153 atomic_set(&map->read, 0);
154 ret = inet_recvmsg(map->sock, &msg, wanted, MSG_DONTWAIT);
155 WARN_ON(ret > wanted);
156 if (ret == -EAGAIN) /* shouldn't happen */
157 return;
158 if (!ret)
159 ret = -ENOTCONN;
160 spin_lock_irqsave(&map->sock->sk->sk_receive_queue.lock, flags);
161 if (ret > 0 && !skb_queue_empty(&map->sock->sk->sk_receive_queue))
162 atomic_inc(&map->read);
163 spin_unlock_irqrestore(&map->sock->sk->sk_receive_queue.lock, flags);
164
165 /* write the data, then modify the indexes */
166 virt_wmb();
167 if (ret < 0)
168 intf->in_error = ret;
169 else
170 intf->in_prod = prod + ret;
171 /* update the indexes, then notify the other end */
172 virt_wmb();
173 notify_remote_via_irq(map->irq);
174
175 return;
5d520d85
SS		 176}
177
178static void pvcalls_conn_back_write(struct sock_mapping *map)
179{
180}
181
5db4d286
SS		 182static void pvcalls_back_ioworker(struct work_struct *work)
183{
5d520d85
SS		 184 struct pvcalls_ioworker *ioworker = container_of(work,
185 struct pvcalls_ioworker, register_work);
186 struct sock_mapping *map = container_of(ioworker, struct sock_mapping,
187 ioworker);
188
189 while (atomic_read(&map->io) > 0) {
190 if (atomic_read(&map->release) > 0) {
191 atomic_set(&map->release, 0);
192 return;
193 }
194
195 if (atomic_read(&map->read) > 0)
196 pvcalls_conn_back_read(map);
197 if (atomic_read(&map->write) > 0)
198 pvcalls_conn_back_write(map);
199
200 atomic_dec(&map->io);
201 }
5db4d286
SS		 202}
203
b1efa693
SS		 204static int pvcalls_back_socket(struct xenbus_device *dev,
205 struct xen_pvcalls_request *req)
206{
fb029875
SS		 207 struct pvcalls_fedata *fedata;
208 int ret;
209 struct xen_pvcalls_response *rsp;
210
211 fedata = dev_get_drvdata(&dev->dev);
212
213 if (req->u.socket.domain != AF_INET ||
214 req->u.socket.type != SOCK_STREAM ||
215 (req->u.socket.protocol != IPPROTO_IP &&
216 req->u.socket.protocol != AF_INET))
217 ret = -EAFNOSUPPORT;
218 else
219 ret = 0;
220
221 /* leave the actual socket allocation for later */
222
223 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
224 rsp->req_id = req->req_id;
225 rsp->cmd = req->cmd;
226 rsp->u.socket.id = req->u.socket.id;
227 rsp->ret = ret;
228
b1efa693
SS		 229 return 0;
230}
231
5db4d286
SS		 232static void pvcalls_sk_state_change(struct sock *sock)
233{
234 struct sock_mapping *map = sock->sk_user_data;
235 struct pvcalls_data_intf *intf;
236
237 if (map == NULL)
238 return;
239
240 intf = map->ring;
241 intf->in_error = -ENOTCONN;
242 notify_remote_via_irq(map->irq);
243}
244
245static void pvcalls_sk_data_ready(struct sock *sock)
246{
b3f9f773
SS		 247 struct sock_mapping *map = sock->sk_user_data;
248 struct pvcalls_ioworker *iow;
249
250 if (map == NULL)
251 return;
252
253 iow = &map->ioworker;
254 atomic_inc(&map->read);
255 atomic_inc(&map->io);
256 queue_work(iow->wq, &iow->register_work);
5db4d286
SS		 257}
258
259static struct sock_mapping *pvcalls_new_active_socket(
260 struct pvcalls_fedata *fedata,
261 uint64_t id,
262 grant_ref_t ref,
263 uint32_t evtchn,
264 struct socket *sock)
265{
266 int ret;
267 struct sock_mapping *map;
268 void *page;
269
270 map = kzalloc(sizeof(*map), GFP_KERNEL);
271 if (map == NULL)
272 return NULL;
273
274 map->fedata = fedata;
275 map->sock = sock;
276 map->id = id;
277 map->ref = ref;
278
279 ret = xenbus_map_ring_valloc(fedata->dev, &ref, 1, &page);
280 if (ret < 0)
281 goto out;
282 map->ring = page;
283 map->ring_order = map->ring->ring_order;
284 /* first read the order, then map the data ring */
285 virt_rmb();
286 if (map->ring_order > MAX_RING_ORDER) {
287 pr_warn("%s frontend requested ring_order %u, which is > MAX (%u)\n",
288 __func__, map->ring_order, MAX_RING_ORDER);
289 goto out;
290 }
291 ret = xenbus_map_ring_valloc(fedata->dev, map->ring->ref,
292 (1 << map->ring_order), &page);
293 if (ret < 0)
294 goto out;
295 map->bytes = page;
296
297 ret = bind_interdomain_evtchn_to_irqhandler(fedata->dev->otherend_id,
298 evtchn,
299 pvcalls_back_conn_event,
300 0,
301 "pvcalls-backend",
302 map);
303 if (ret < 0)
304 goto out;
305 map->irq = ret;
306
307 map->data.in = map->bytes;
308 map->data.out = map->bytes + XEN_FLEX_RING_SIZE(map->ring_order);
309
310 map->ioworker.wq = alloc_workqueue("pvcalls_io", WQ_UNBOUND, 1);
311 if (!map->ioworker.wq)
312 goto out;
313 atomic_set(&map->io, 1);
314 INIT_WORK(&map->ioworker.register_work, pvcalls_back_ioworker);
315
316 down(&fedata->socket_lock);
317 list_add_tail(&map->list, &fedata->socket_mappings);
318 up(&fedata->socket_lock);
319
320 write_lock_bh(&map->sock->sk->sk_callback_lock);
321 map->saved_data_ready = map->sock->sk->sk_data_ready;
322 map->sock->sk->sk_user_data = map;
323 map->sock->sk->sk_data_ready = pvcalls_sk_data_ready;
324 map->sock->sk->sk_state_change = pvcalls_sk_state_change;
325 write_unlock_bh(&map->sock->sk->sk_callback_lock);
326
327 return map;
328out:
329 down(&fedata->socket_lock);
330 list_del(&map->list);
331 pvcalls_back_release_active(fedata->dev, fedata, map);
332 up(&fedata->socket_lock);
333 return NULL;
334}
335
b1efa693
SS		 336static int pvcalls_back_connect(struct xenbus_device *dev,
337 struct xen_pvcalls_request *req)
5db4d286
SS		 338{
339 struct pvcalls_fedata *fedata;
340 int ret = -EINVAL;
341 struct socket *sock;
342 struct sock_mapping *map;
343 struct xen_pvcalls_response *rsp;
344 struct sockaddr *sa = (struct sockaddr *)&req->u.connect.addr;
345
346 fedata = dev_get_drvdata(&dev->dev);
347
348 if (req->u.connect.len < sizeof(sa->sa_family) ||
349 req->u.connect.len > sizeof(req->u.connect.addr) ||
350 sa->sa_family != AF_INET)
351 goto out;
352
353 ret = sock_create(AF_INET, SOCK_STREAM, 0, &sock);
354 if (ret < 0)
355 goto out;
356 ret = inet_stream_connect(sock, sa, req->u.connect.len, 0);
357 if (ret < 0) {
358 sock_release(sock);
359 goto out;
360 }
361
362 map = pvcalls_new_active_socket(fedata,
363 req->u.connect.id,
364 req->u.connect.ref,
365 req->u.connect.evtchn,
366 sock);
367 if (!map) {
368 ret = -EFAULT;
369 sock_release(map->sock);
370 }
371
372out:
373 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
374 rsp->req_id = req->req_id;
375 rsp->cmd = req->cmd;
376 rsp->u.connect.id = req->u.connect.id;
377 rsp->ret = ret;
378
379 return 0;
380}
381
382static int pvcalls_back_release_active(struct xenbus_device *dev,
383 struct pvcalls_fedata *fedata,
384 struct sock_mapping *map)
b1efa693 385{
a51729cb
SS		 386 disable_irq(map->irq);
387 if (map->sock->sk != NULL) {
388 write_lock_bh(&map->sock->sk->sk_callback_lock);
389 map->sock->sk->sk_user_data = NULL;
390 map->sock->sk->sk_data_ready = map->saved_data_ready;
391 write_unlock_bh(&map->sock->sk->sk_callback_lock);
392 }
393
394 atomic_set(&map->release, 1);
395 flush_work(&map->ioworker.register_work);
396
397 xenbus_unmap_ring_vfree(dev, map->bytes);
398 xenbus_unmap_ring_vfree(dev, (void *)map->ring);
399 unbind_from_irqhandler(map->irq, map);
400
401 sock_release(map->sock);
402 kfree(map);
403
404 return 0;
405}
406
407static int pvcalls_back_release_passive(struct xenbus_device *dev,
408 struct pvcalls_fedata *fedata,
409 struct sockpass_mapping *mappass)
410{
411 if (mappass->sock->sk != NULL) {
412 write_lock_bh(&mappass->sock->sk->sk_callback_lock);
413 mappass->sock->sk->sk_user_data = NULL;
414 mappass->sock->sk->sk_data_ready = mappass->saved_data_ready;
415 write_unlock_bh(&mappass->sock->sk->sk_callback_lock);
416 }
417 sock_release(mappass->sock);
418 flush_workqueue(mappass->wq);
419 destroy_workqueue(mappass->wq);
420 kfree(mappass);
421
b1efa693
SS		 422 return 0;
423}
424
425static int pvcalls_back_release(struct xenbus_device *dev,
426 struct xen_pvcalls_request *req)
427{
a51729cb
SS		 428 struct pvcalls_fedata *fedata;
429 struct sock_mapping *map, *n;
430 struct sockpass_mapping *mappass;
431 int ret = 0;
432 struct xen_pvcalls_response *rsp;
433
434 fedata = dev_get_drvdata(&dev->dev);
435
436 down(&fedata->socket_lock);
437 list_for_each_entry_safe(map, n, &fedata->socket_mappings, list) {
438 if (map->id == req->u.release.id) {
439 list_del(&map->list);
440 up(&fedata->socket_lock);
441 ret = pvcalls_back_release_active(dev, fedata, map);
442 goto out;
443 }
444 }
445 mappass = radix_tree_lookup(&fedata->socketpass_mappings,
446 req->u.release.id);
447 if (mappass != NULL) {
448 radix_tree_delete(&fedata->socketpass_mappings, mappass->id);
449 up(&fedata->socket_lock);
450 ret = pvcalls_back_release_passive(dev, fedata, mappass);
451 } else
452 up(&fedata->socket_lock);
453
454out:
455 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
456 rsp->req_id = req->req_id;
457 rsp->u.release.id = req->u.release.id;
458 rsp->cmd = req->cmd;
459 rsp->ret = ret;
b1efa693
SS		 460 return 0;
461}
462
331a63e6
SS		 463static void __pvcalls_back_accept(struct work_struct *work)
464{
6f474e71
SS		 465 struct sockpass_mapping *mappass = container_of(
466 work, struct sockpass_mapping, register_work);
467 struct sock_mapping *map;
468 struct pvcalls_ioworker *iow;
469 struct pvcalls_fedata *fedata;
470 struct socket *sock;
471 struct xen_pvcalls_response *rsp;
472 struct xen_pvcalls_request *req;
473 int notify;
474 int ret = -EINVAL;
475 unsigned long flags;
476
477 fedata = mappass->fedata;
478 /*
479 * __pvcalls_back_accept can race against pvcalls_back_accept.
480 * We only need to check the value of "cmd" on read. It could be
481 * done atomically, but to simplify the code on the write side, we
482 * use a spinlock.
483 */
484 spin_lock_irqsave(&mappass->copy_lock, flags);
485 req = &mappass->reqcopy;
486 if (req->cmd != PVCALLS_ACCEPT) {
487 spin_unlock_irqrestore(&mappass->copy_lock, flags);
488 return;
489 }
490 spin_unlock_irqrestore(&mappass->copy_lock, flags);
491
492 sock = sock_alloc();
493 if (sock == NULL)
494 goto out_error;
495 sock->type = mappass->sock->type;
496 sock->ops = mappass->sock->ops;
497
498 ret = inet_accept(mappass->sock, sock, O_NONBLOCK, true);
499 if (ret == -EAGAIN) {
500 sock_release(sock);
501 goto out_error;
502 }
503
504 map = pvcalls_new_active_socket(fedata,
505 req->u.accept.id_new,
506 req->u.accept.ref,
507 req->u.accept.evtchn,
508 sock);
509 if (!map) {
510 ret = -EFAULT;
511 sock_release(sock);
512 goto out_error;
513 }
514
515 map->sockpass = mappass;
516 iow = &map->ioworker;
517 atomic_inc(&map->read);
518 atomic_inc(&map->io);
519 queue_work(iow->wq, &iow->register_work);
520
521out_error:
522 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
523 rsp->req_id = req->req_id;
524 rsp->cmd = req->cmd;
525 rsp->u.accept.id = req->u.accept.id;
526 rsp->ret = ret;
527 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&fedata->ring, notify);
528 if (notify)
529 notify_remote_via_irq(fedata->irq);
530
531 mappass->reqcopy.cmd = 0;
331a63e6
SS		 532}
533
534static void pvcalls_pass_sk_data_ready(struct sock *sock)
535{
6f474e71 536 struct sockpass_mapping *mappass = sock->sk_user_data;
3cf33a58
SS		 537 struct pvcalls_fedata *fedata;
538 struct xen_pvcalls_response *rsp;
539 unsigned long flags;
540 int notify;
6f474e71
SS		 541
542 if (mappass == NULL)
543 return;
544
3cf33a58
SS		 545 fedata = mappass->fedata;
546 spin_lock_irqsave(&mappass->copy_lock, flags);
547 if (mappass->reqcopy.cmd == PVCALLS_POLL) {
548 rsp = RING_GET_RESPONSE(&fedata->ring,
549 fedata->ring.rsp_prod_pvt++);
550 rsp->req_id = mappass->reqcopy.req_id;
551 rsp->u.poll.id = mappass->reqcopy.u.poll.id;
552 rsp->cmd = mappass->reqcopy.cmd;
553 rsp->ret = 0;
554
555 mappass->reqcopy.cmd = 0;
556 spin_unlock_irqrestore(&mappass->copy_lock, flags);
557
558 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&fedata->ring, notify);
559 if (notify)
560 notify_remote_via_irq(mappass->fedata->irq);
561 } else {
562 spin_unlock_irqrestore(&mappass->copy_lock, flags);
563 queue_work(mappass->wq, &mappass->register_work);
564 }
331a63e6
SS		 565}
566
b1efa693
SS		 567static int pvcalls_back_bind(struct xenbus_device *dev,
568 struct xen_pvcalls_request *req)
569{
331a63e6
SS		 570 struct pvcalls_fedata *fedata;
571 int ret;
572 struct sockpass_mapping *map;
573 struct xen_pvcalls_response *rsp;
574
575 fedata = dev_get_drvdata(&dev->dev);
576
577 map = kzalloc(sizeof(*map), GFP_KERNEL);
578 if (map == NULL) {
579 ret = -ENOMEM;
580 goto out;
581 }
582
583 INIT_WORK(&map->register_work, __pvcalls_back_accept);
584 spin_lock_init(&map->copy_lock);
585 map->wq = alloc_workqueue("pvcalls_wq", WQ_UNBOUND, 1);
586 if (!map->wq) {
587 ret = -ENOMEM;
588 goto out;
589 }
590
591 ret = sock_create(AF_INET, SOCK_STREAM, 0, &map->sock);
592 if (ret < 0)
593 goto out;
594
595 ret = inet_bind(map->sock, (struct sockaddr *)&req->u.bind.addr,
596 req->u.bind.len);
597 if (ret < 0)
598 goto out;
599
600 map->fedata = fedata;
601 map->id = req->u.bind.id;
602
603 down(&fedata->socket_lock);
604 ret = radix_tree_insert(&fedata->socketpass_mappings, map->id,
605 map);
606 up(&fedata->socket_lock);
607 if (ret)
608 goto out;
609
610 write_lock_bh(&map->sock->sk->sk_callback_lock);
611 map->saved_data_ready = map->sock->sk->sk_data_ready;
612 map->sock->sk->sk_user_data = map;
613 map->sock->sk->sk_data_ready = pvcalls_pass_sk_data_ready;
614 write_unlock_bh(&map->sock->sk->sk_callback_lock);
615
616out:
617 if (ret) {
618 if (map && map->sock)
619 sock_release(map->sock);
620 if (map && map->wq)
621 destroy_workqueue(map->wq);
622 kfree(map);
623 }
624 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
625 rsp->req_id = req->req_id;
626 rsp->cmd = req->cmd;
627 rsp->u.bind.id = req->u.bind.id;
628 rsp->ret = ret;
b1efa693
SS		 629 return 0;
630}
631
632static int pvcalls_back_listen(struct xenbus_device *dev,
633 struct xen_pvcalls_request *req)
634{
8ce3f762
SS		 635 struct pvcalls_fedata *fedata;
636 int ret = -EINVAL;
637 struct sockpass_mapping *map;
638 struct xen_pvcalls_response *rsp;
639
640 fedata = dev_get_drvdata(&dev->dev);
641
642 down(&fedata->socket_lock);
643 map = radix_tree_lookup(&fedata->socketpass_mappings, req->u.listen.id);
644 up(&fedata->socket_lock);
645 if (map == NULL)
646 goto out;
647
648 ret = inet_listen(map->sock, req->u.listen.backlog);
649
650out:
651 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
652 rsp->req_id = req->req_id;
653 rsp->cmd = req->cmd;
654 rsp->u.listen.id = req->u.listen.id;
655 rsp->ret = ret;
b1efa693
SS		 656 return 0;
657}
658
659static int pvcalls_back_accept(struct xenbus_device *dev,
660 struct xen_pvcalls_request *req)
661{
6f474e71
SS		 662 struct pvcalls_fedata *fedata;
663 struct sockpass_mapping *mappass;
664 int ret = -EINVAL;
665 struct xen_pvcalls_response *rsp;
666 unsigned long flags;
667
668 fedata = dev_get_drvdata(&dev->dev);
669
670 down(&fedata->socket_lock);
671 mappass = radix_tree_lookup(&fedata->socketpass_mappings,
672 req->u.accept.id);
673 up(&fedata->socket_lock);
674 if (mappass == NULL)
675 goto out_error;
676
677 /*
678 * Limitation of the current implementation: only support one
679 * concurrent accept or poll call on one socket.
680 */
681 spin_lock_irqsave(&mappass->copy_lock, flags);
682 if (mappass->reqcopy.cmd != 0) {
683 spin_unlock_irqrestore(&mappass->copy_lock, flags);
684 ret = -EINTR;
685 goto out_error;
686 }
687
688 mappass->reqcopy = *req;
689 spin_unlock_irqrestore(&mappass->copy_lock, flags);
690 queue_work(mappass->wq, &mappass->register_work);
691
692 /* Tell the caller we don't need to send back a notification yet */
693 return -1;
694
695out_error:
696 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
697 rsp->req_id = req->req_id;
698 rsp->cmd = req->cmd;
699 rsp->u.accept.id = req->u.accept.id;
700 rsp->ret = ret;
b1efa693
SS		 701 return 0;
702}
703
704static int pvcalls_back_poll(struct xenbus_device *dev,
705 struct xen_pvcalls_request *req)
706{
3cf33a58
SS		 707 struct pvcalls_fedata *fedata;
708 struct sockpass_mapping *mappass;
709 struct xen_pvcalls_response *rsp;
710 struct inet_connection_sock *icsk;
711 struct request_sock_queue *queue;
712 unsigned long flags;
713 int ret;
714 bool data;
715
716 fedata = dev_get_drvdata(&dev->dev);
717
718 down(&fedata->socket_lock);
719 mappass = radix_tree_lookup(&fedata->socketpass_mappings,
720 req->u.poll.id);
721 up(&fedata->socket_lock);
722 if (mappass == NULL)
723 return -EINVAL;
724
725 /*
726 * Limitation of the current implementation: only support one
727 * concurrent accept or poll call on one socket.
728 */
729 spin_lock_irqsave(&mappass->copy_lock, flags);
730 if (mappass->reqcopy.cmd != 0) {
731 ret = -EINTR;
732 goto out;
733 }
734
735 mappass->reqcopy = *req;
736 icsk = inet_csk(mappass->sock->sk);
737 queue = &icsk->icsk_accept_queue;
738 data = queue->rskq_accept_head != NULL;
739 if (data) {
740 mappass->reqcopy.cmd = 0;
741 ret = 0;
742 goto out;
743 }
744 spin_unlock_irqrestore(&mappass->copy_lock, flags);
745
746 /* Tell the caller we don't need to send back a notification yet */
747 return -1;
748
749out:
750 spin_unlock_irqrestore(&mappass->copy_lock, flags);
751
752 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
753 rsp->req_id = req->req_id;
754 rsp->cmd = req->cmd;
755 rsp->u.poll.id = req->u.poll.id;
756 rsp->ret = ret;
b1efa693
SS		 757 return 0;
758}
759
760static int pvcalls_back_handle_cmd(struct xenbus_device *dev,
761 struct xen_pvcalls_request *req)
762{
763 int ret = 0;
764
765 switch (req->cmd) {
766 case PVCALLS_SOCKET:
767 ret = pvcalls_back_socket(dev, req);
768 break;
769 case PVCALLS_CONNECT:
770 ret = pvcalls_back_connect(dev, req);
771 break;
772 case PVCALLS_RELEASE:
773 ret = pvcalls_back_release(dev, req);
774 break;
775 case PVCALLS_BIND:
776 ret = pvcalls_back_bind(dev, req);
777 break;
778 case PVCALLS_LISTEN:
779 ret = pvcalls_back_listen(dev, req);
780 break;
781 case PVCALLS_ACCEPT:
782 ret = pvcalls_back_accept(dev, req);
783 break;
784 case PVCALLS_POLL:
785 ret = pvcalls_back_poll(dev, req);
786 break;
787 default:
788 {
789 struct pvcalls_fedata *fedata;
790 struct xen_pvcalls_response *rsp;
791
792 fedata = dev_get_drvdata(&dev->dev);
793 rsp = RING_GET_RESPONSE(
794 &fedata->ring, fedata->ring.rsp_prod_pvt++);
795 rsp->req_id = req->req_id;
796 rsp->cmd = req->cmd;
797 rsp->ret = -ENOTSUPP;
798 break;
799 }
800 }
801 return ret;
802}
803
804static void pvcalls_back_work(struct pvcalls_fedata *fedata)
805{
806 int notify, notify_all = 0, more = 1;
807 struct xen_pvcalls_request req;
808 struct xenbus_device *dev = fedata->dev;
809
810 while (more) {
811 while (RING_HAS_UNCONSUMED_REQUESTS(&fedata->ring)) {
812 RING_COPY_REQUEST(&fedata->ring,
813 fedata->ring.req_cons++,
814 &req);
815
816 if (!pvcalls_back_handle_cmd(dev, &req)) {
817 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(
818 &fedata->ring, notify);
819 notify_all += notify;
820 }
821 }
822
823 if (notify_all) {
824 notify_remote_via_irq(fedata->irq);
825 notify_all = 0;
826 }
827
828 RING_FINAL_CHECK_FOR_REQUESTS(&fedata->ring, more);
829 }
830}
831
d0e4d560
SS		 832static irqreturn_t pvcalls_back_event(int irq, void *dev_id)
833{
b1efa693
SS		 834 struct xenbus_device *dev = dev_id;
835 struct pvcalls_fedata *fedata = NULL;
836
837 if (dev == NULL)
838 return IRQ_HANDLED;
839
840 fedata = dev_get_drvdata(&dev->dev);
841 if (fedata == NULL)
842 return IRQ_HANDLED;
843
844 pvcalls_back_work(fedata);
d0e4d560
SS		 845 return IRQ_HANDLED;
846}
847
5db4d286
SS		 848static irqreturn_t pvcalls_back_conn_event(int irq, void *sock_map)
849{
850 return IRQ_HANDLED;
851}
852
0a9c75c2
SS		 853static int backend_connect(struct xenbus_device *dev)
854{
d0e4d560
SS		 855 int err, evtchn;
856 grant_ref_t ring_ref;
857 struct pvcalls_fedata *fedata = NULL;
858
859 fedata = kzalloc(sizeof(struct pvcalls_fedata), GFP_KERNEL);
860 if (!fedata)
861 return -ENOMEM;
862
863 fedata->irq = -1;
864 err = xenbus_scanf(XBT_NIL, dev->otherend, "port", "%u",
865 &evtchn);
866 if (err != 1) {
867 err = -EINVAL;
868 xenbus_dev_fatal(dev, err, "reading %s/event-channel",
869 dev->otherend);
870 goto error;
871 }
872
873 err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-ref", "%u", &ring_ref);
874 if (err != 1) {
875 err = -EINVAL;
876 xenbus_dev_fatal(dev, err, "reading %s/ring-ref",
877 dev->otherend);
878 goto error;
879 }
880
881 err = bind_interdomain_evtchn_to_irq(dev->otherend_id, evtchn);
882 if (err < 0)
883 goto error;
884 fedata->irq = err;
885
886 err = request_threaded_irq(fedata->irq, NULL, pvcalls_back_event,
887 IRQF_ONESHOT, "pvcalls-back", dev);
888 if (err < 0)
889 goto error;
890
891 err = xenbus_map_ring_valloc(dev, &ring_ref, 1,
892 (void **)&fedata->sring);
893 if (err < 0)
894 goto error;
895
896 BACK_RING_INIT(&fedata->ring, fedata->sring, XEN_PAGE_SIZE * 1);
897 fedata->dev = dev;
898
899 INIT_LIST_HEAD(&fedata->socket_mappings);
900 INIT_RADIX_TREE(&fedata->socketpass_mappings, GFP_KERNEL);
901 sema_init(&fedata->socket_lock, 1);
902 dev_set_drvdata(&dev->dev, fedata);
903
904 down(&pvcalls_back_global.frontends_lock);
905 list_add_tail(&fedata->list, &pvcalls_back_global.frontends);
906 up(&pvcalls_back_global.frontends_lock);
907
0a9c75c2 908 return 0;
d0e4d560
SS		 909
910 error:
911 if (fedata->irq >= 0)
912 unbind_from_irqhandler(fedata->irq, dev);
913 if (fedata->sring != NULL)
914 xenbus_unmap_ring_vfree(dev, fedata->sring);
915 kfree(fedata);
916 return err;
0a9c75c2
SS		 917}
918
919static int backend_disconnect(struct xenbus_device *dev)
920{
0a85d23b
SS		 921 struct pvcalls_fedata *fedata;
922 struct sock_mapping *map, *n;
923 struct sockpass_mapping *mappass;
924 struct radix_tree_iter iter;
925 void **slot;
926
927
928 fedata = dev_get_drvdata(&dev->dev);
929
930 down(&fedata->socket_lock);
931 list_for_each_entry_safe(map, n, &fedata->socket_mappings, list) {
932 list_del(&map->list);
933 pvcalls_back_release_active(dev, fedata, map);
934 }
935
936 radix_tree_for_each_slot(slot, &fedata->socketpass_mappings, &iter, 0) {
937 mappass = radix_tree_deref_slot(slot);
938 if (!mappass)
939 continue;
940 if (radix_tree_exception(mappass)) {
941 if (radix_tree_deref_retry(mappass))
942 slot = radix_tree_iter_retry(&iter);
943 } else {
944 radix_tree_delete(&fedata->socketpass_mappings,
945 mappass->id);
946 pvcalls_back_release_passive(dev, fedata, mappass);
947 }
948 }
949 up(&fedata->socket_lock);
950
951 unbind_from_irqhandler(fedata->irq, dev);
952 xenbus_unmap_ring_vfree(dev, fedata->sring);
953
954 list_del(&fedata->list);
955 kfree(fedata);
956 dev_set_drvdata(&dev->dev, NULL);
957
0a9c75c2
SS		 958 return 0;
959}
960
72e59c30
SS		 961static int pvcalls_back_probe(struct xenbus_device *dev,
962 const struct xenbus_device_id *id)
963{
0a9c75c2
SS		 964 int err, abort;
965 struct xenbus_transaction xbt;
966
967again:
968 abort = 1;
969
970 err = xenbus_transaction_start(&xbt);
971 if (err) {
972 pr_warn("%s cannot create xenstore transaction\n", __func__);
973 return err;
974 }
975
976 err = xenbus_printf(xbt, dev->nodename, "versions", "%s",
977 PVCALLS_VERSIONS);
978 if (err) {
979 pr_warn("%s write out 'versions' failed\n", __func__);
980 goto abort;
981 }
982
983 err = xenbus_printf(xbt, dev->nodename, "max-page-order", "%u",
984 MAX_RING_ORDER);
985 if (err) {
986 pr_warn("%s write out 'max-page-order' failed\n", __func__);
987 goto abort;
988 }
989
990 err = xenbus_printf(xbt, dev->nodename, "function-calls",
991 XENBUS_FUNCTIONS_CALLS);
992 if (err) {
993 pr_warn("%s write out 'function-calls' failed\n", __func__);
994 goto abort;
995 }
996
997 abort = 0;
998abort:
999 err = xenbus_transaction_end(xbt, abort);
1000 if (err) {
1001 if (err == -EAGAIN && !abort)
1002 goto again;
1003 pr_warn("%s cannot complete xenstore transaction\n", __func__);
1004 return err;
1005 }
1006
1007 if (abort)
1008 return -EFAULT;
1009
1010 xenbus_switch_state(dev, XenbusStateInitWait);
1011
72e59c30
SS		 1012 return 0;
1013}
1014
0a9c75c2
SS		 1015static void set_backend_state(struct xenbus_device *dev,
1016 enum xenbus_state state)
1017{
1018 while (dev->state != state) {
1019 switch (dev->state) {
1020 case XenbusStateClosed:
1021 switch (state) {
1022 case XenbusStateInitWait:
1023 case XenbusStateConnected:
1024 xenbus_switch_state(dev, XenbusStateInitWait);
1025 break;
1026 case XenbusStateClosing:
1027 xenbus_switch_state(dev, XenbusStateClosing);
1028 break;
1029 default:
1030 __WARN();
1031 }
1032 break;
1033 case XenbusStateInitWait:
1034 case XenbusStateInitialised:
1035 switch (state) {
1036 case XenbusStateConnected:
1037 backend_connect(dev);
1038 xenbus_switch_state(dev, XenbusStateConnected);
1039 break;
1040 case XenbusStateClosing:
1041 case XenbusStateClosed:
1042 xenbus_switch_state(dev, XenbusStateClosing);
1043 break;
1044 default:
1045 __WARN();
1046 }
1047 break;
1048 case XenbusStateConnected:
1049 switch (state) {
1050 case XenbusStateInitWait:
1051 case XenbusStateClosing:
1052 case XenbusStateClosed:
1053 down(&pvcalls_back_global.frontends_lock);
1054 backend_disconnect(dev);
1055 up(&pvcalls_back_global.frontends_lock);
1056 xenbus_switch_state(dev, XenbusStateClosing);
1057 break;
1058 default:
1059 __WARN();
1060 }
1061 break;
1062 case XenbusStateClosing:
1063 switch (state) {
1064 case XenbusStateInitWait:
1065 case XenbusStateConnected:
1066 case XenbusStateClosed:
1067 xenbus_switch_state(dev, XenbusStateClosed);
1068 break;
1069 default:
1070 __WARN();
1071 }
1072 break;
1073 default:
1074 __WARN();
1075 }
1076 }
1077}
1078
72e59c30
SS		 1079static void pvcalls_back_changed(struct xenbus_device *dev,
1080 enum xenbus_state frontend_state)
1081{
0a9c75c2
SS		 1082 switch (frontend_state) {
1083 case XenbusStateInitialising:
1084 set_backend_state(dev, XenbusStateInitWait);
1085 break;
1086
1087 case XenbusStateInitialised:
1088 case XenbusStateConnected:
1089 set_backend_state(dev, XenbusStateConnected);
1090 break;
1091
1092 case XenbusStateClosing:
1093 set_backend_state(dev, XenbusStateClosing);
1094 break;
1095
1096 case XenbusStateClosed:
1097 set_backend_state(dev, XenbusStateClosed);
1098 if (xenbus_dev_is_online(dev))
1099 break;
1100 device_unregister(&dev->dev);
1101 break;
1102 case XenbusStateUnknown:
1103 set_backend_state(dev, XenbusStateClosed);
1104 device_unregister(&dev->dev);
1105 break;
1106
1107 default:
1108 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
1109 frontend_state);
1110 break;
1111 }
72e59c30
SS		 1112}
1113
1114static int pvcalls_back_remove(struct xenbus_device *dev)
1115{
1116 return 0;
1117}
1118
1119static int pvcalls_back_uevent(struct xenbus_device *xdev,
1120 struct kobj_uevent_env *env)
1121{
1122 return 0;
1123}
1124
1125static const struct xenbus_device_id pvcalls_back_ids[] = {
1126 { "pvcalls" },
1127 { "" }
1128};
1129
1130static struct xenbus_driver pvcalls_back_driver = {
1131 .ids = pvcalls_back_ids,
1132 .probe = pvcalls_back_probe,
1133 .remove = pvcalls_back_remove,
1134 .uevent = pvcalls_back_uevent,
1135 .otherend_changed = pvcalls_back_changed,
1136};
9be07334
SS		 1137
1138static int __init pvcalls_back_init(void)
1139{
1140 int ret;
1141
1142 if (!xen_domain())
1143 return -ENODEV;
1144
1145 ret = xenbus_register_backend(&pvcalls_back_driver);
1146 if (ret < 0)
1147 return ret;
1148
1149 sema_init(&pvcalls_back_global.frontends_lock, 1);
1150 INIT_LIST_HEAD(&pvcalls_back_global.frontends);
1151 return 0;
1152}
1153module_init(pvcalls_back_init);
0a85d23b
SS		 1154
1155static void __exit pvcalls_back_fin(void)
1156{
1157 struct pvcalls_fedata *fedata, *nfedata;
1158
1159 down(&pvcalls_back_global.frontends_lock);
1160 list_for_each_entry_safe(fedata, nfedata,
1161 &pvcalls_back_global.frontends, list) {
1162 backend_disconnect(fedata->dev);
1163 }
1164 up(&pvcalls_back_global.frontends_lock);
1165
1166 xenbus_unregister_driver(&pvcalls_back_driver);
1167}
1168
1169module_exit(pvcalls_back_fin);
ubuntu-bionic-kernel mirror