1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * (c) 2017 Stefano Stabellini <stefano@aporeto.com>
6 #include <linux/inet.h>
7 #include <linux/kthread.h>
8 #include <linux/list.h>
9 #include <linux/radix-tree.h>
10 #include <linux/module.h>
11 #include <linux/semaphore.h>
12 #include <linux/wait.h>
14 #include <net/inet_common.h>
15 #include <net/inet_connection_sock.h>
16 #include <net/request_sock.h>
18 #include <xen/events.h>
19 #include <xen/grant_table.h>
21 #include <xen/xenbus.h>
22 #include <xen/interface/io/pvcalls.h>
24 #define PVCALLS_VERSIONS "1"
25 #define MAX_RING_ORDER XENBUS_MAX_RING_GRANT_ORDER
27 struct pvcalls_back_global {
28 struct list_head frontends;
29 struct semaphore frontends_lock;
30 } pvcalls_back_global;
33 * Per-frontend data structure. It contains pointers to the command
34 * ring, its event channel, a list of active sockets and a tree of
37 struct pvcalls_fedata {
38 struct list_head list;
39 struct xenbus_device *dev;
40 struct xen_pvcalls_sring *sring;
41 struct xen_pvcalls_back_ring ring;
43 struct list_head socket_mappings;
44 struct radix_tree_root socketpass_mappings;
45 struct semaphore socket_lock;
48 struct pvcalls_ioworker {
49 struct work_struct register_work;
50 struct workqueue_struct *wq;
54 struct list_head list;
55 struct pvcalls_fedata *fedata;
56 struct sockpass_mapping *sockpass;
60 struct pvcalls_data_intf *ring;
62 struct pvcalls_data data;
69 void (*saved_data_ready)(struct sock *sk);
70 struct pvcalls_ioworker ioworker;
73 struct sockpass_mapping {
74 struct list_head list;
75 struct pvcalls_fedata *fedata;
78 struct xen_pvcalls_request reqcopy;
80 struct workqueue_struct *wq;
81 struct work_struct register_work;
82 void (*saved_data_ready)(struct sock *sk);
85 static irqreturn_t pvcalls_back_conn_event(int irq, void *sock_map);
86 static int pvcalls_back_release_active(struct xenbus_device *dev,
87 struct pvcalls_fedata *fedata,
88 struct sock_mapping *map);
90 static void pvcalls_conn_back_read(void *opaque)
92 struct sock_mapping *map = (struct sock_mapping *)opaque;
95 RING_IDX cons, prod, size, wanted, array_size, masked_prod, masked_cons;
97 struct pvcalls_data_intf *intf = map->ring;
98 struct pvcalls_data *data = &map->data;
102 array_size = XEN_FLEX_RING_SIZE(map->ring_order);
103 cons = intf->in_cons;
104 prod = intf->in_prod;
105 error = intf->in_error;
106 /* read the indexes first, then deal with the data */
112 size = pvcalls_queued(prod, cons, array_size);
113 if (size >= array_size)
115 spin_lock_irqsave(&map->sock->sk->sk_receive_queue.lock, flags);
116 if (skb_queue_empty(&map->sock->sk->sk_receive_queue)) {
117 atomic_set(&map->read, 0);
118 spin_unlock_irqrestore(&map->sock->sk->sk_receive_queue.lock,
122 spin_unlock_irqrestore(&map->sock->sk->sk_receive_queue.lock, flags);
123 wanted = array_size - size;
124 masked_prod = pvcalls_mask(prod, array_size);
125 masked_cons = pvcalls_mask(cons, array_size);
127 memset(&msg, 0, sizeof(msg));
128 if (masked_prod < masked_cons) {
129 vec[0].iov_base = data->in + masked_prod;
130 vec[0].iov_len = wanted;
131 iov_iter_kvec(&msg.msg_iter, WRITE, vec, 1, wanted);
133 vec[0].iov_base = data->in + masked_prod;
134 vec[0].iov_len = array_size - masked_prod;
135 vec[1].iov_base = data->in;
136 vec[1].iov_len = wanted - vec[0].iov_len;
137 iov_iter_kvec(&msg.msg_iter, WRITE, vec, 2, wanted);
140 atomic_set(&map->read, 0);
141 ret = inet_recvmsg(map->sock, &msg, wanted, MSG_DONTWAIT);
142 WARN_ON(ret > wanted);
143 if (ret == -EAGAIN) /* shouldn't happen */
147 spin_lock_irqsave(&map->sock->sk->sk_receive_queue.lock, flags);
148 if (ret > 0 && !skb_queue_empty(&map->sock->sk->sk_receive_queue))
149 atomic_inc(&map->read);
150 spin_unlock_irqrestore(&map->sock->sk->sk_receive_queue.lock, flags);
152 /* write the data, then modify the indexes */
155 atomic_set(&map->read, 0);
156 intf->in_error = ret;
158 intf->in_prod = prod + ret;
159 /* update the indexes, then notify the other end */
161 notify_remote_via_irq(map->irq);
166 static void pvcalls_conn_back_write(struct sock_mapping *map)
168 struct pvcalls_data_intf *intf = map->ring;
169 struct pvcalls_data *data = &map->data;
172 RING_IDX cons, prod, size, array_size;
175 cons = intf->out_cons;
176 prod = intf->out_prod;
177 /* read the indexes before dealing with the data */
180 array_size = XEN_FLEX_RING_SIZE(map->ring_order);
181 size = pvcalls_queued(prod, cons, array_size);
185 memset(&msg, 0, sizeof(msg));
186 msg.msg_flags |= MSG_DONTWAIT;
187 if (pvcalls_mask(prod, array_size) > pvcalls_mask(cons, array_size)) {
188 vec[0].iov_base = data->out + pvcalls_mask(cons, array_size);
189 vec[0].iov_len = size;
190 iov_iter_kvec(&msg.msg_iter, READ, vec, 1, size);
192 vec[0].iov_base = data->out + pvcalls_mask(cons, array_size);
193 vec[0].iov_len = array_size - pvcalls_mask(cons, array_size);
194 vec[1].iov_base = data->out;
195 vec[1].iov_len = size - vec[0].iov_len;
196 iov_iter_kvec(&msg.msg_iter, READ, vec, 2, size);
199 atomic_set(&map->write, 0);
200 ret = inet_sendmsg(map->sock, &msg, size);
201 if (ret == -EAGAIN || (ret >= 0 && ret < size)) {
202 atomic_inc(&map->write);
203 atomic_inc(&map->io);
208 /* write the data, then update the indexes */
211 intf->out_error = ret;
214 intf->out_cons = cons + ret;
215 prod = intf->out_prod;
217 /* update the indexes, then notify the other end */
219 if (prod != cons + ret)
220 atomic_inc(&map->write);
221 notify_remote_via_irq(map->irq);
224 static void pvcalls_back_ioworker(struct work_struct *work)
226 struct pvcalls_ioworker *ioworker = container_of(work,
227 struct pvcalls_ioworker, register_work);
228 struct sock_mapping *map = container_of(ioworker, struct sock_mapping,
231 while (atomic_read(&map->io) > 0) {
232 if (atomic_read(&map->release) > 0) {
233 atomic_set(&map->release, 0);
237 if (atomic_read(&map->read) > 0)
238 pvcalls_conn_back_read(map);
239 if (atomic_read(&map->write) > 0)
240 pvcalls_conn_back_write(map);
242 atomic_dec(&map->io);
246 static int pvcalls_back_socket(struct xenbus_device *dev,
247 struct xen_pvcalls_request *req)
249 struct pvcalls_fedata *fedata;
251 struct xen_pvcalls_response *rsp;
253 fedata = dev_get_drvdata(&dev->dev);
255 if (req->u.socket.domain != AF_INET ||
256 req->u.socket.type != SOCK_STREAM ||
257 (req->u.socket.protocol != IPPROTO_IP &&
258 req->u.socket.protocol != AF_INET))
263 /* leave the actual socket allocation for later */
265 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
266 rsp->req_id = req->req_id;
268 rsp->u.socket.id = req->u.socket.id;
274 static void pvcalls_sk_state_change(struct sock *sock)
276 struct sock_mapping *map = sock->sk_user_data;
281 atomic_inc(&map->read);
282 notify_remote_via_irq(map->irq);
285 static void pvcalls_sk_data_ready(struct sock *sock)
287 struct sock_mapping *map = sock->sk_user_data;
288 struct pvcalls_ioworker *iow;
293 iow = &map->ioworker;
294 atomic_inc(&map->read);
295 atomic_inc(&map->io);
296 queue_work(iow->wq, &iow->register_work);
299 static struct sock_mapping *pvcalls_new_active_socket(
300 struct pvcalls_fedata *fedata,
307 struct sock_mapping *map;
310 map = kzalloc(sizeof(*map), GFP_KERNEL);
314 map->fedata = fedata;
319 ret = xenbus_map_ring_valloc(fedata->dev, &ref, 1, &page);
323 map->ring_order = map->ring->ring_order;
324 /* first read the order, then map the data ring */
326 if (map->ring_order > MAX_RING_ORDER) {
327 pr_warn("%s frontend requested ring_order %u, which is > MAX (%u)\n",
328 __func__, map->ring_order, MAX_RING_ORDER);
331 ret = xenbus_map_ring_valloc(fedata->dev, map->ring->ref,
332 (1 << map->ring_order), &page);
337 ret = bind_interdomain_evtchn_to_irqhandler(fedata->dev->otherend_id,
339 pvcalls_back_conn_event,
347 map->data.in = map->bytes;
348 map->data.out = map->bytes + XEN_FLEX_RING_SIZE(map->ring_order);
350 map->ioworker.wq = alloc_workqueue("pvcalls_io", WQ_UNBOUND, 1);
351 if (!map->ioworker.wq)
353 atomic_set(&map->io, 1);
354 INIT_WORK(&map->ioworker.register_work, pvcalls_back_ioworker);
356 down(&fedata->socket_lock);
357 list_add_tail(&map->list, &fedata->socket_mappings);
358 up(&fedata->socket_lock);
360 write_lock_bh(&map->sock->sk->sk_callback_lock);
361 map->saved_data_ready = map->sock->sk->sk_data_ready;
362 map->sock->sk->sk_user_data = map;
363 map->sock->sk->sk_data_ready = pvcalls_sk_data_ready;
364 map->sock->sk->sk_state_change = pvcalls_sk_state_change;
365 write_unlock_bh(&map->sock->sk->sk_callback_lock);
369 down(&fedata->socket_lock);
370 list_del(&map->list);
371 pvcalls_back_release_active(fedata->dev, fedata, map);
372 up(&fedata->socket_lock);
376 static int pvcalls_back_connect(struct xenbus_device *dev,
377 struct xen_pvcalls_request *req)
379 struct pvcalls_fedata *fedata;
382 struct sock_mapping *map;
383 struct xen_pvcalls_response *rsp;
384 struct sockaddr *sa = (struct sockaddr *)&req->u.connect.addr;
386 fedata = dev_get_drvdata(&dev->dev);
388 if (req->u.connect.len < sizeof(sa->sa_family) ||
389 req->u.connect.len > sizeof(req->u.connect.addr) ||
390 sa->sa_family != AF_INET)
393 ret = sock_create(AF_INET, SOCK_STREAM, 0, &sock);
396 ret = inet_stream_connect(sock, sa, req->u.connect.len, 0);
402 map = pvcalls_new_active_socket(fedata,
405 req->u.connect.evtchn,
413 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
414 rsp->req_id = req->req_id;
416 rsp->u.connect.id = req->u.connect.id;
422 static int pvcalls_back_release_active(struct xenbus_device *dev,
423 struct pvcalls_fedata *fedata,
424 struct sock_mapping *map)
426 disable_irq(map->irq);
427 if (map->sock->sk != NULL) {
428 write_lock_bh(&map->sock->sk->sk_callback_lock);
429 map->sock->sk->sk_user_data = NULL;
430 map->sock->sk->sk_data_ready = map->saved_data_ready;
431 write_unlock_bh(&map->sock->sk->sk_callback_lock);
434 atomic_set(&map->release, 1);
435 flush_work(&map->ioworker.register_work);
437 xenbus_unmap_ring_vfree(dev, map->bytes);
438 xenbus_unmap_ring_vfree(dev, (void *)map->ring);
439 unbind_from_irqhandler(map->irq, map);
441 sock_release(map->sock);
447 static int pvcalls_back_release_passive(struct xenbus_device *dev,
448 struct pvcalls_fedata *fedata,
449 struct sockpass_mapping *mappass)
451 if (mappass->sock->sk != NULL) {
452 write_lock_bh(&mappass->sock->sk->sk_callback_lock);
453 mappass->sock->sk->sk_user_data = NULL;
454 mappass->sock->sk->sk_data_ready = mappass->saved_data_ready;
455 write_unlock_bh(&mappass->sock->sk->sk_callback_lock);
457 sock_release(mappass->sock);
458 flush_workqueue(mappass->wq);
459 destroy_workqueue(mappass->wq);
465 static int pvcalls_back_release(struct xenbus_device *dev,
466 struct xen_pvcalls_request *req)
468 struct pvcalls_fedata *fedata;
469 struct sock_mapping *map, *n;
470 struct sockpass_mapping *mappass;
472 struct xen_pvcalls_response *rsp;
474 fedata = dev_get_drvdata(&dev->dev);
476 down(&fedata->socket_lock);
477 list_for_each_entry_safe(map, n, &fedata->socket_mappings, list) {
478 if (map->id == req->u.release.id) {
479 list_del(&map->list);
480 up(&fedata->socket_lock);
481 ret = pvcalls_back_release_active(dev, fedata, map);
485 mappass = radix_tree_lookup(&fedata->socketpass_mappings,
487 if (mappass != NULL) {
488 radix_tree_delete(&fedata->socketpass_mappings, mappass->id);
489 up(&fedata->socket_lock);
490 ret = pvcalls_back_release_passive(dev, fedata, mappass);
492 up(&fedata->socket_lock);
495 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
496 rsp->req_id = req->req_id;
497 rsp->u.release.id = req->u.release.id;
503 static void __pvcalls_back_accept(struct work_struct *work)
505 struct sockpass_mapping *mappass = container_of(
506 work, struct sockpass_mapping, register_work);
507 struct sock_mapping *map;
508 struct pvcalls_ioworker *iow;
509 struct pvcalls_fedata *fedata;
511 struct xen_pvcalls_response *rsp;
512 struct xen_pvcalls_request *req;
517 fedata = mappass->fedata;
519 * __pvcalls_back_accept can race against pvcalls_back_accept.
520 * We only need to check the value of "cmd" on read. It could be
521 * done atomically, but to simplify the code on the write side, we
524 spin_lock_irqsave(&mappass->copy_lock, flags);
525 req = &mappass->reqcopy;
526 if (req->cmd != PVCALLS_ACCEPT) {
527 spin_unlock_irqrestore(&mappass->copy_lock, flags);
530 spin_unlock_irqrestore(&mappass->copy_lock, flags);
535 sock->type = mappass->sock->type;
536 sock->ops = mappass->sock->ops;
538 ret = inet_accept(mappass->sock, sock, O_NONBLOCK, true);
539 if (ret == -EAGAIN) {
544 map = pvcalls_new_active_socket(fedata,
545 req->u.accept.id_new,
547 req->u.accept.evtchn,
555 map->sockpass = mappass;
556 iow = &map->ioworker;
557 atomic_inc(&map->read);
558 atomic_inc(&map->io);
559 queue_work(iow->wq, &iow->register_work);
562 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
563 rsp->req_id = req->req_id;
565 rsp->u.accept.id = req->u.accept.id;
567 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&fedata->ring, notify);
569 notify_remote_via_irq(fedata->irq);
571 mappass->reqcopy.cmd = 0;
574 static void pvcalls_pass_sk_data_ready(struct sock *sock)
576 struct sockpass_mapping *mappass = sock->sk_user_data;
577 struct pvcalls_fedata *fedata;
578 struct xen_pvcalls_response *rsp;
585 fedata = mappass->fedata;
586 spin_lock_irqsave(&mappass->copy_lock, flags);
587 if (mappass->reqcopy.cmd == PVCALLS_POLL) {
588 rsp = RING_GET_RESPONSE(&fedata->ring,
589 fedata->ring.rsp_prod_pvt++);
590 rsp->req_id = mappass->reqcopy.req_id;
591 rsp->u.poll.id = mappass->reqcopy.u.poll.id;
592 rsp->cmd = mappass->reqcopy.cmd;
595 mappass->reqcopy.cmd = 0;
596 spin_unlock_irqrestore(&mappass->copy_lock, flags);
598 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&fedata->ring, notify);
600 notify_remote_via_irq(mappass->fedata->irq);
602 spin_unlock_irqrestore(&mappass->copy_lock, flags);
603 queue_work(mappass->wq, &mappass->register_work);
607 static int pvcalls_back_bind(struct xenbus_device *dev,
608 struct xen_pvcalls_request *req)
610 struct pvcalls_fedata *fedata;
612 struct sockpass_mapping *map;
613 struct xen_pvcalls_response *rsp;
615 fedata = dev_get_drvdata(&dev->dev);
617 map = kzalloc(sizeof(*map), GFP_KERNEL);
623 INIT_WORK(&map->register_work, __pvcalls_back_accept);
624 spin_lock_init(&map->copy_lock);
625 map->wq = alloc_workqueue("pvcalls_wq", WQ_UNBOUND, 1);
631 ret = sock_create(AF_INET, SOCK_STREAM, 0, &map->sock);
635 ret = inet_bind(map->sock, (struct sockaddr *)&req->u.bind.addr,
640 map->fedata = fedata;
641 map->id = req->u.bind.id;
643 down(&fedata->socket_lock);
644 ret = radix_tree_insert(&fedata->socketpass_mappings, map->id,
646 up(&fedata->socket_lock);
650 write_lock_bh(&map->sock->sk->sk_callback_lock);
651 map->saved_data_ready = map->sock->sk->sk_data_ready;
652 map->sock->sk->sk_user_data = map;
653 map->sock->sk->sk_data_ready = pvcalls_pass_sk_data_ready;
654 write_unlock_bh(&map->sock->sk->sk_callback_lock);
658 if (map && map->sock)
659 sock_release(map->sock);
661 destroy_workqueue(map->wq);
664 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
665 rsp->req_id = req->req_id;
667 rsp->u.bind.id = req->u.bind.id;
672 static int pvcalls_back_listen(struct xenbus_device *dev,
673 struct xen_pvcalls_request *req)
675 struct pvcalls_fedata *fedata;
677 struct sockpass_mapping *map;
678 struct xen_pvcalls_response *rsp;
680 fedata = dev_get_drvdata(&dev->dev);
682 down(&fedata->socket_lock);
683 map = radix_tree_lookup(&fedata->socketpass_mappings, req->u.listen.id);
684 up(&fedata->socket_lock);
688 ret = inet_listen(map->sock, req->u.listen.backlog);
691 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
692 rsp->req_id = req->req_id;
694 rsp->u.listen.id = req->u.listen.id;
699 static int pvcalls_back_accept(struct xenbus_device *dev,
700 struct xen_pvcalls_request *req)
702 struct pvcalls_fedata *fedata;
703 struct sockpass_mapping *mappass;
705 struct xen_pvcalls_response *rsp;
708 fedata = dev_get_drvdata(&dev->dev);
710 down(&fedata->socket_lock);
711 mappass = radix_tree_lookup(&fedata->socketpass_mappings,
713 up(&fedata->socket_lock);
718 * Limitation of the current implementation: only support one
719 * concurrent accept or poll call on one socket.
721 spin_lock_irqsave(&mappass->copy_lock, flags);
722 if (mappass->reqcopy.cmd != 0) {
723 spin_unlock_irqrestore(&mappass->copy_lock, flags);
728 mappass->reqcopy = *req;
729 spin_unlock_irqrestore(&mappass->copy_lock, flags);
730 queue_work(mappass->wq, &mappass->register_work);
732 /* Tell the caller we don't need to send back a notification yet */
736 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
737 rsp->req_id = req->req_id;
739 rsp->u.accept.id = req->u.accept.id;
744 static int pvcalls_back_poll(struct xenbus_device *dev,
745 struct xen_pvcalls_request *req)
747 struct pvcalls_fedata *fedata;
748 struct sockpass_mapping *mappass;
749 struct xen_pvcalls_response *rsp;
750 struct inet_connection_sock *icsk;
751 struct request_sock_queue *queue;
756 fedata = dev_get_drvdata(&dev->dev);
758 down(&fedata->socket_lock);
759 mappass = radix_tree_lookup(&fedata->socketpass_mappings,
761 up(&fedata->socket_lock);
766 * Limitation of the current implementation: only support one
767 * concurrent accept or poll call on one socket.
769 spin_lock_irqsave(&mappass->copy_lock, flags);
770 if (mappass->reqcopy.cmd != 0) {
775 mappass->reqcopy = *req;
776 icsk = inet_csk(mappass->sock->sk);
777 queue = &icsk->icsk_accept_queue;
778 data = READ_ONCE(queue->rskq_accept_head) != NULL;
780 mappass->reqcopy.cmd = 0;
784 spin_unlock_irqrestore(&mappass->copy_lock, flags);
786 /* Tell the caller we don't need to send back a notification yet */
790 spin_unlock_irqrestore(&mappass->copy_lock, flags);
792 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
793 rsp->req_id = req->req_id;
795 rsp->u.poll.id = req->u.poll.id;
800 static int pvcalls_back_handle_cmd(struct xenbus_device *dev,
801 struct xen_pvcalls_request *req)
807 ret = pvcalls_back_socket(dev, req);
809 case PVCALLS_CONNECT:
810 ret = pvcalls_back_connect(dev, req);
812 case PVCALLS_RELEASE:
813 ret = pvcalls_back_release(dev, req);
816 ret = pvcalls_back_bind(dev, req);
819 ret = pvcalls_back_listen(dev, req);
822 ret = pvcalls_back_accept(dev, req);
825 ret = pvcalls_back_poll(dev, req);
829 struct pvcalls_fedata *fedata;
830 struct xen_pvcalls_response *rsp;
832 fedata = dev_get_drvdata(&dev->dev);
833 rsp = RING_GET_RESPONSE(
834 &fedata->ring, fedata->ring.rsp_prod_pvt++);
835 rsp->req_id = req->req_id;
837 rsp->ret = -ENOTSUPP;
844 static void pvcalls_back_work(struct pvcalls_fedata *fedata)
846 int notify, notify_all = 0, more = 1;
847 struct xen_pvcalls_request req;
848 struct xenbus_device *dev = fedata->dev;
851 while (RING_HAS_UNCONSUMED_REQUESTS(&fedata->ring)) {
852 RING_COPY_REQUEST(&fedata->ring,
853 fedata->ring.req_cons++,
856 if (!pvcalls_back_handle_cmd(dev, &req)) {
857 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(
858 &fedata->ring, notify);
859 notify_all += notify;
864 notify_remote_via_irq(fedata->irq);
868 RING_FINAL_CHECK_FOR_REQUESTS(&fedata->ring, more);
872 static irqreturn_t pvcalls_back_event(int irq, void *dev_id)
874 struct xenbus_device *dev = dev_id;
875 struct pvcalls_fedata *fedata = NULL;
880 fedata = dev_get_drvdata(&dev->dev);
884 pvcalls_back_work(fedata);
888 static irqreturn_t pvcalls_back_conn_event(int irq, void *sock_map)
890 struct sock_mapping *map = sock_map;
891 struct pvcalls_ioworker *iow;
893 if (map == NULL || map->sock == NULL || map->sock->sk == NULL ||
894 map->sock->sk->sk_user_data != map)
897 iow = &map->ioworker;
899 atomic_inc(&map->write);
900 atomic_inc(&map->io);
901 queue_work(iow->wq, &iow->register_work);
906 static int backend_connect(struct xenbus_device *dev)
909 grant_ref_t ring_ref;
910 struct pvcalls_fedata *fedata = NULL;
912 fedata = kzalloc(sizeof(struct pvcalls_fedata), GFP_KERNEL);
917 err = xenbus_scanf(XBT_NIL, dev->otherend, "port", "%u",
921 xenbus_dev_fatal(dev, err, "reading %s/event-channel",
926 err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-ref", "%u", &ring_ref);
929 xenbus_dev_fatal(dev, err, "reading %s/ring-ref",
934 err = bind_interdomain_evtchn_to_irq(dev->otherend_id, evtchn);
939 err = request_threaded_irq(fedata->irq, NULL, pvcalls_back_event,
940 IRQF_ONESHOT, "pvcalls-back", dev);
944 err = xenbus_map_ring_valloc(dev, &ring_ref, 1,
945 (void **)&fedata->sring);
949 BACK_RING_INIT(&fedata->ring, fedata->sring, XEN_PAGE_SIZE * 1);
952 INIT_LIST_HEAD(&fedata->socket_mappings);
953 INIT_RADIX_TREE(&fedata->socketpass_mappings, GFP_KERNEL);
954 sema_init(&fedata->socket_lock, 1);
955 dev_set_drvdata(&dev->dev, fedata);
957 down(&pvcalls_back_global.frontends_lock);
958 list_add_tail(&fedata->list, &pvcalls_back_global.frontends);
959 up(&pvcalls_back_global.frontends_lock);
964 if (fedata->irq >= 0)
965 unbind_from_irqhandler(fedata->irq, dev);
966 if (fedata->sring != NULL)
967 xenbus_unmap_ring_vfree(dev, fedata->sring);
972 static int backend_disconnect(struct xenbus_device *dev)
974 struct pvcalls_fedata *fedata;
975 struct sock_mapping *map, *n;
976 struct sockpass_mapping *mappass;
977 struct radix_tree_iter iter;
981 fedata = dev_get_drvdata(&dev->dev);
983 down(&fedata->socket_lock);
984 list_for_each_entry_safe(map, n, &fedata->socket_mappings, list) {
985 list_del(&map->list);
986 pvcalls_back_release_active(dev, fedata, map);
989 radix_tree_for_each_slot(slot, &fedata->socketpass_mappings, &iter, 0) {
990 mappass = radix_tree_deref_slot(slot);
993 if (radix_tree_exception(mappass)) {
994 if (radix_tree_deref_retry(mappass))
995 slot = radix_tree_iter_retry(&iter);
997 radix_tree_delete(&fedata->socketpass_mappings,
999 pvcalls_back_release_passive(dev, fedata, mappass);
1002 up(&fedata->socket_lock);
1004 unbind_from_irqhandler(fedata->irq, dev);
1005 xenbus_unmap_ring_vfree(dev, fedata->sring);
1007 list_del(&fedata->list);
1009 dev_set_drvdata(&dev->dev, NULL);
1014 static int pvcalls_back_probe(struct xenbus_device *dev,
1015 const struct xenbus_device_id *id)
1018 struct xenbus_transaction xbt;
1023 err = xenbus_transaction_start(&xbt);
1025 pr_warn("%s cannot create xenstore transaction\n", __func__);
1029 err = xenbus_printf(xbt, dev->nodename, "versions", "%s",
1032 pr_warn("%s write out 'versions' failed\n", __func__);
1036 err = xenbus_printf(xbt, dev->nodename, "max-page-order", "%u",
1039 pr_warn("%s write out 'max-page-order' failed\n", __func__);
1043 err = xenbus_printf(xbt, dev->nodename, "function-calls",
1044 XENBUS_FUNCTIONS_CALLS);
1046 pr_warn("%s write out 'function-calls' failed\n", __func__);
1052 err = xenbus_transaction_end(xbt, abort);
1054 if (err == -EAGAIN && !abort)
1056 pr_warn("%s cannot complete xenstore transaction\n", __func__);
1063 xenbus_switch_state(dev, XenbusStateInitWait);
1068 static void set_backend_state(struct xenbus_device *dev,
1069 enum xenbus_state state)
1071 while (dev->state != state) {
1072 switch (dev->state) {
1073 case XenbusStateClosed:
1075 case XenbusStateInitWait:
1076 case XenbusStateConnected:
1077 xenbus_switch_state(dev, XenbusStateInitWait);
1079 case XenbusStateClosing:
1080 xenbus_switch_state(dev, XenbusStateClosing);
1086 case XenbusStateInitWait:
1087 case XenbusStateInitialised:
1089 case XenbusStateConnected:
1090 backend_connect(dev);
1091 xenbus_switch_state(dev, XenbusStateConnected);
1093 case XenbusStateClosing:
1094 case XenbusStateClosed:
1095 xenbus_switch_state(dev, XenbusStateClosing);
1101 case XenbusStateConnected:
1103 case XenbusStateInitWait:
1104 case XenbusStateClosing:
1105 case XenbusStateClosed:
1106 down(&pvcalls_back_global.frontends_lock);
1107 backend_disconnect(dev);
1108 up(&pvcalls_back_global.frontends_lock);
1109 xenbus_switch_state(dev, XenbusStateClosing);
1115 case XenbusStateClosing:
1117 case XenbusStateInitWait:
1118 case XenbusStateConnected:
1119 case XenbusStateClosed:
1120 xenbus_switch_state(dev, XenbusStateClosed);
1132 static void pvcalls_back_changed(struct xenbus_device *dev,
1133 enum xenbus_state frontend_state)
1135 switch (frontend_state) {
1136 case XenbusStateInitialising:
1137 set_backend_state(dev, XenbusStateInitWait);
1140 case XenbusStateInitialised:
1141 case XenbusStateConnected:
1142 set_backend_state(dev, XenbusStateConnected);
1145 case XenbusStateClosing:
1146 set_backend_state(dev, XenbusStateClosing);
1149 case XenbusStateClosed:
1150 set_backend_state(dev, XenbusStateClosed);
1151 if (xenbus_dev_is_online(dev))
1153 device_unregister(&dev->dev);
1155 case XenbusStateUnknown:
1156 set_backend_state(dev, XenbusStateClosed);
1157 device_unregister(&dev->dev);
1161 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
1167 static int pvcalls_back_remove(struct xenbus_device *dev)
1172 static int pvcalls_back_uevent(struct xenbus_device *xdev,
1173 struct kobj_uevent_env *env)
1178 static const struct xenbus_device_id pvcalls_back_ids[] = {
1183 static struct xenbus_driver pvcalls_back_driver = {
1184 .ids = pvcalls_back_ids,
1185 .probe = pvcalls_back_probe,
1186 .remove = pvcalls_back_remove,
1187 .uevent = pvcalls_back_uevent,
1188 .otherend_changed = pvcalls_back_changed,
1191 static int __init pvcalls_back_init(void)
1198 ret = xenbus_register_backend(&pvcalls_back_driver);
1202 sema_init(&pvcalls_back_global.frontends_lock, 1);
1203 INIT_LIST_HEAD(&pvcalls_back_global.frontends);
1206 module_init(pvcalls_back_init);
1208 static void __exit pvcalls_back_fin(void)
1210 struct pvcalls_fedata *fedata, *nfedata;
1212 down(&pvcalls_back_global.frontends_lock);
1213 list_for_each_entry_safe(fedata, nfedata,
1214 &pvcalls_back_global.frontends, list) {
1215 backend_disconnect(fedata->dev);
1217 up(&pvcalls_back_global.frontends_lock);
1219 xenbus_unregister_driver(&pvcalls_back_driver);
1222 module_exit(pvcalls_back_fin);
1224 MODULE_DESCRIPTION("Xen PV Calls backend driver");
1225 MODULE_AUTHOR("Stefano Stabellini <sstabellini@kernel.org>");
1226 MODULE_LICENSE("GPL");