2 * (c) 2017 Stefano Stabellini <stefano@aporeto.com>
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.
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.
15 #include <linux/module.h>
16 #include <linux/net.h>
17 #include <linux/socket.h>
21 #include <xen/events.h>
22 #include <xen/grant_table.h>
24 #include <xen/xenbus.h>
25 #include <xen/interface/io/pvcalls.h>
27 #include "pvcalls-front.h"
29 #define PVCALLS_INVALID_ID UINT_MAX
30 #define PVCALLS_RING_ORDER XENBUS_MAX_RING_GRANT_ORDER
31 #define PVCALLS_NR_RSP_PER_RING __CONST_RING_SIZE(xen_pvcalls, XEN_PAGE_SIZE)
32 #define PVCALLS_FRONT_MAX_SPIN 5000
34 static struct proto pvcalls_proto = {
37 .obj_size = sizeof(struct sock),
40 struct pvcalls_bedata {
41 struct xen_pvcalls_front_ring ring;
45 struct list_head socket_mappings;
46 spinlock_t socket_lock;
48 wait_queue_head_t inflight_req;
49 struct xen_pvcalls_response rsp[PVCALLS_NR_RSP_PER_RING];
51 /* Only one front/back connection supported. */
52 static struct xenbus_device *pvcalls_front_dev;
53 static atomic_t pvcalls_refcount;
55 /* first increment refcount, then proceed */
56 #define pvcalls_enter() { \
57 atomic_inc(&pvcalls_refcount); \
60 /* first complete other operations, then decrement refcount */
61 #define pvcalls_exit() { \
62 atomic_dec(&pvcalls_refcount); \
67 struct list_head list;
74 struct pvcalls_data_intf *ring;
75 struct pvcalls_data data;
76 struct mutex in_mutex;
77 struct mutex out_mutex;
79 wait_queue_head_t inflight_conn_req;
83 * Socket status, needs to be 64-bit aligned due to the
84 * test_and_* functions which have this requirement on arm64.
86 #define PVCALLS_STATUS_UNINITALIZED 0
87 #define PVCALLS_STATUS_BIND 1
88 #define PVCALLS_STATUS_LISTEN 2
89 uint8_t status __attribute__((aligned(8)));
91 * Internal state-machine flags.
92 * Only one accept operation can be inflight for a socket.
93 * Only one poll operation can be inflight for a given socket.
94 * flags needs to be 64-bit aligned due to the test_and_*
95 * functions which have this requirement on arm64.
97 #define PVCALLS_FLAG_ACCEPT_INFLIGHT 0
98 #define PVCALLS_FLAG_POLL_INFLIGHT 1
99 #define PVCALLS_FLAG_POLL_RET 2
100 uint8_t flags __attribute__((aligned(8)));
101 uint32_t inflight_req_id;
102 struct sock_mapping *accept_map;
103 wait_queue_head_t inflight_accept_req;
108 static inline struct sock_mapping *pvcalls_enter_sock(struct socket *sock)
110 struct sock_mapping *map;
112 if (!pvcalls_front_dev ||
113 dev_get_drvdata(&pvcalls_front_dev->dev) == NULL)
114 return ERR_PTR(-ENOTCONN);
116 map = (struct sock_mapping *)sock->sk->sk_send_head;
118 return ERR_PTR(-ENOTSOCK);
121 atomic_inc(&map->refcount);
125 static inline void pvcalls_exit_sock(struct socket *sock)
127 struct sock_mapping *map;
129 map = (struct sock_mapping *)sock->sk->sk_send_head;
130 atomic_dec(&map->refcount);
134 static inline int get_request(struct pvcalls_bedata *bedata, int *req_id)
136 *req_id = bedata->ring.req_prod_pvt & (RING_SIZE(&bedata->ring) - 1);
137 if (RING_FULL(&bedata->ring) ||
138 bedata->rsp[*req_id].req_id != PVCALLS_INVALID_ID)
143 static bool pvcalls_front_write_todo(struct sock_mapping *map)
145 struct pvcalls_data_intf *intf = map->active.ring;
146 RING_IDX cons, prod, size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
149 error = intf->out_error;
150 if (error == -ENOTCONN)
155 cons = intf->out_cons;
156 prod = intf->out_prod;
157 return !!(size - pvcalls_queued(prod, cons, size));
160 static bool pvcalls_front_read_todo(struct sock_mapping *map)
162 struct pvcalls_data_intf *intf = map->active.ring;
166 cons = intf->in_cons;
167 prod = intf->in_prod;
168 error = intf->in_error;
169 return (error != 0 ||
170 pvcalls_queued(prod, cons,
171 XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER)) != 0);
174 static irqreturn_t pvcalls_front_event_handler(int irq, void *dev_id)
176 struct xenbus_device *dev = dev_id;
177 struct pvcalls_bedata *bedata;
178 struct xen_pvcalls_response *rsp;
180 int req_id = 0, more = 0, done = 0;
186 bedata = dev_get_drvdata(&dev->dev);
187 if (bedata == NULL) {
193 while (RING_HAS_UNCONSUMED_RESPONSES(&bedata->ring)) {
194 rsp = RING_GET_RESPONSE(&bedata->ring, bedata->ring.rsp_cons);
196 req_id = rsp->req_id;
197 if (rsp->cmd == PVCALLS_POLL) {
198 struct sock_mapping *map = (struct sock_mapping *)(uintptr_t)
201 clear_bit(PVCALLS_FLAG_POLL_INFLIGHT,
202 (void *)&map->passive.flags);
204 * clear INFLIGHT, then set RET. It pairs with
205 * the checks at the beginning of
206 * pvcalls_front_poll_passive.
209 set_bit(PVCALLS_FLAG_POLL_RET,
210 (void *)&map->passive.flags);
212 dst = (uint8_t *)&bedata->rsp[req_id] +
214 src = (uint8_t *)rsp + sizeof(rsp->req_id);
215 memcpy(dst, src, sizeof(*rsp) - sizeof(rsp->req_id));
217 * First copy the rest of the data, then req_id. It is
218 * paired with the barrier when accessing bedata->rsp.
221 bedata->rsp[req_id].req_id = req_id;
225 bedata->ring.rsp_cons++;
228 RING_FINAL_CHECK_FOR_RESPONSES(&bedata->ring, more);
232 wake_up(&bedata->inflight_req);
237 static void pvcalls_front_free_map(struct pvcalls_bedata *bedata,
238 struct sock_mapping *map)
242 unbind_from_irqhandler(map->active.irq, map);
244 spin_lock(&bedata->socket_lock);
245 if (!list_empty(&map->list))
246 list_del_init(&map->list);
247 spin_unlock(&bedata->socket_lock);
249 for (i = 0; i < (1 << PVCALLS_RING_ORDER); i++)
250 gnttab_end_foreign_access(map->active.ring->ref[i], 0, 0);
251 gnttab_end_foreign_access(map->active.ref, 0, 0);
252 free_page((unsigned long)map->active.ring);
257 static irqreturn_t pvcalls_front_conn_handler(int irq, void *sock_map)
259 struct sock_mapping *map = sock_map;
264 wake_up_interruptible(&map->active.inflight_conn_req);
269 int pvcalls_front_socket(struct socket *sock)
271 struct pvcalls_bedata *bedata;
272 struct sock_mapping *map = NULL;
273 struct xen_pvcalls_request *req;
274 int notify, req_id, ret;
277 * PVCalls only supports domain AF_INET,
278 * type SOCK_STREAM and protocol 0 sockets for now.
280 * Check socket type here, AF_INET and protocol checks are done
283 if (sock->type != SOCK_STREAM)
287 if (!pvcalls_front_dev) {
291 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
293 map = kzalloc(sizeof(*map), GFP_KERNEL);
299 spin_lock(&bedata->socket_lock);
301 ret = get_request(bedata, &req_id);
304 spin_unlock(&bedata->socket_lock);
310 * sock->sk->sk_send_head is not used for ip sockets: reuse the
311 * field to store a pointer to the struct sock_mapping
312 * corresponding to the socket. This way, we can easily get the
313 * struct sock_mapping from the struct socket.
315 sock->sk->sk_send_head = (void *)map;
316 list_add_tail(&map->list, &bedata->socket_mappings);
318 req = RING_GET_REQUEST(&bedata->ring, req_id);
319 req->req_id = req_id;
320 req->cmd = PVCALLS_SOCKET;
321 req->u.socket.id = (uintptr_t) map;
322 req->u.socket.domain = AF_INET;
323 req->u.socket.type = SOCK_STREAM;
324 req->u.socket.protocol = IPPROTO_IP;
326 bedata->ring.req_prod_pvt++;
327 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
328 spin_unlock(&bedata->socket_lock);
330 notify_remote_via_irq(bedata->irq);
332 wait_event(bedata->inflight_req,
333 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
335 /* read req_id, then the content */
337 ret = bedata->rsp[req_id].ret;
338 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
344 static void free_active_ring(struct sock_mapping *map)
346 free_pages((unsigned long)map->active.data.in,
347 map->active.ring->ring_order);
348 free_page((unsigned long)map->active.ring);
351 static int alloc_active_ring(struct sock_mapping *map)
355 map->active.ring = (struct pvcalls_data_intf *)
356 get_zeroed_page(GFP_KERNEL);
357 if (!map->active.ring)
360 map->active.ring->ring_order = PVCALLS_RING_ORDER;
361 bytes = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
366 map->active.data.in = bytes;
367 map->active.data.out = bytes +
368 XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
373 free_active_ring(map);
377 static int create_active(struct sock_mapping *map, int *evtchn)
380 int ret = -ENOMEM, irq = -1, i;
383 init_waitqueue_head(&map->active.inflight_conn_req);
385 bytes = map->active.data.in;
386 for (i = 0; i < (1 << PVCALLS_RING_ORDER); i++)
387 map->active.ring->ref[i] = gnttab_grant_foreign_access(
388 pvcalls_front_dev->otherend_id,
389 pfn_to_gfn(virt_to_pfn(bytes) + i), 0);
391 map->active.ref = gnttab_grant_foreign_access(
392 pvcalls_front_dev->otherend_id,
393 pfn_to_gfn(virt_to_pfn((void *)map->active.ring)), 0);
395 ret = xenbus_alloc_evtchn(pvcalls_front_dev, evtchn);
398 irq = bind_evtchn_to_irqhandler(*evtchn, pvcalls_front_conn_handler,
399 0, "pvcalls-frontend", map);
405 map->active.irq = irq;
406 map->active_socket = true;
407 mutex_init(&map->active.in_mutex);
408 mutex_init(&map->active.out_mutex);
414 xenbus_free_evtchn(pvcalls_front_dev, *evtchn);
418 int pvcalls_front_connect(struct socket *sock, struct sockaddr *addr,
419 int addr_len, int flags)
421 struct pvcalls_bedata *bedata;
422 struct sock_mapping *map = NULL;
423 struct xen_pvcalls_request *req;
424 int notify, req_id, ret, evtchn;
426 if (addr->sa_family != AF_INET || sock->type != SOCK_STREAM)
429 map = pvcalls_enter_sock(sock);
433 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
434 ret = alloc_active_ring(map);
436 pvcalls_exit_sock(sock);
440 spin_lock(&bedata->socket_lock);
441 ret = get_request(bedata, &req_id);
443 spin_unlock(&bedata->socket_lock);
444 free_active_ring(map);
445 pvcalls_exit_sock(sock);
448 ret = create_active(map, &evtchn);
450 spin_unlock(&bedata->socket_lock);
451 free_active_ring(map);
452 pvcalls_exit_sock(sock);
456 req = RING_GET_REQUEST(&bedata->ring, req_id);
457 req->req_id = req_id;
458 req->cmd = PVCALLS_CONNECT;
459 req->u.connect.id = (uintptr_t)map;
460 req->u.connect.len = addr_len;
461 req->u.connect.flags = flags;
462 req->u.connect.ref = map->active.ref;
463 req->u.connect.evtchn = evtchn;
464 memcpy(req->u.connect.addr, addr, sizeof(*addr));
468 bedata->ring.req_prod_pvt++;
469 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
470 spin_unlock(&bedata->socket_lock);
473 notify_remote_via_irq(bedata->irq);
475 wait_event(bedata->inflight_req,
476 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
478 /* read req_id, then the content */
480 ret = bedata->rsp[req_id].ret;
481 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
482 pvcalls_exit_sock(sock);
486 static int __write_ring(struct pvcalls_data_intf *intf,
487 struct pvcalls_data *data,
488 struct iov_iter *msg_iter,
491 RING_IDX cons, prod, size, masked_prod, masked_cons;
492 RING_IDX array_size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
495 error = intf->out_error;
498 cons = intf->out_cons;
499 prod = intf->out_prod;
500 /* read indexes before continuing */
503 size = pvcalls_queued(prod, cons, array_size);
504 if (size > array_size)
506 if (size == array_size)
508 if (len > array_size - size)
509 len = array_size - size;
511 masked_prod = pvcalls_mask(prod, array_size);
512 masked_cons = pvcalls_mask(cons, array_size);
514 if (masked_prod < masked_cons) {
515 len = copy_from_iter(data->out + masked_prod, len, msg_iter);
517 if (len > array_size - masked_prod) {
518 int ret = copy_from_iter(data->out + masked_prod,
519 array_size - masked_prod, msg_iter);
520 if (ret != array_size - masked_prod) {
524 len = ret + copy_from_iter(data->out, len - ret, msg_iter);
526 len = copy_from_iter(data->out + masked_prod, len, msg_iter);
530 /* write to ring before updating pointer */
532 intf->out_prod += len;
537 int pvcalls_front_sendmsg(struct socket *sock, struct msghdr *msg,
540 struct pvcalls_bedata *bedata;
541 struct sock_mapping *map;
542 int sent, tot_sent = 0;
543 int count = 0, flags;
545 flags = msg->msg_flags;
546 if (flags & (MSG_CONFIRM|MSG_DONTROUTE|MSG_EOR|MSG_OOB))
549 map = pvcalls_enter_sock(sock);
552 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
554 mutex_lock(&map->active.out_mutex);
555 if ((flags & MSG_DONTWAIT) && !pvcalls_front_write_todo(map)) {
556 mutex_unlock(&map->active.out_mutex);
557 pvcalls_exit_sock(sock);
565 sent = __write_ring(map->active.ring,
566 &map->active.data, &msg->msg_iter,
571 notify_remote_via_irq(map->active.irq);
573 if (sent >= 0 && len > 0 && count < PVCALLS_FRONT_MAX_SPIN)
578 mutex_unlock(&map->active.out_mutex);
579 pvcalls_exit_sock(sock);
583 static int __read_ring(struct pvcalls_data_intf *intf,
584 struct pvcalls_data *data,
585 struct iov_iter *msg_iter,
586 size_t len, int flags)
588 RING_IDX cons, prod, size, masked_prod, masked_cons;
589 RING_IDX array_size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
592 cons = intf->in_cons;
593 prod = intf->in_prod;
594 error = intf->in_error;
595 /* get pointers before reading from the ring */
598 size = pvcalls_queued(prod, cons, array_size);
599 masked_prod = pvcalls_mask(prod, array_size);
600 masked_cons = pvcalls_mask(cons, array_size);
603 return error ?: size;
608 if (masked_prod > masked_cons) {
609 len = copy_to_iter(data->in + masked_cons, len, msg_iter);
611 if (len > (array_size - masked_cons)) {
612 int ret = copy_to_iter(data->in + masked_cons,
613 array_size - masked_cons, msg_iter);
614 if (ret != array_size - masked_cons) {
618 len = ret + copy_to_iter(data->in, len - ret, msg_iter);
620 len = copy_to_iter(data->in + masked_cons, len, msg_iter);
624 /* read data from the ring before increasing the index */
626 if (!(flags & MSG_PEEK))
627 intf->in_cons += len;
632 int pvcalls_front_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
635 struct pvcalls_bedata *bedata;
637 struct sock_mapping *map;
639 if (flags & (MSG_CMSG_CLOEXEC|MSG_ERRQUEUE|MSG_OOB|MSG_TRUNC))
642 map = pvcalls_enter_sock(sock);
645 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
647 mutex_lock(&map->active.in_mutex);
648 if (len > XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER))
649 len = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
651 while (!(flags & MSG_DONTWAIT) && !pvcalls_front_read_todo(map)) {
652 wait_event_interruptible(map->active.inflight_conn_req,
653 pvcalls_front_read_todo(map));
655 ret = __read_ring(map->active.ring, &map->active.data,
656 &msg->msg_iter, len, flags);
659 notify_remote_via_irq(map->active.irq);
661 ret = (flags & MSG_DONTWAIT) ? -EAGAIN : 0;
662 if (ret == -ENOTCONN)
665 mutex_unlock(&map->active.in_mutex);
666 pvcalls_exit_sock(sock);
670 int pvcalls_front_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
672 struct pvcalls_bedata *bedata;
673 struct sock_mapping *map = NULL;
674 struct xen_pvcalls_request *req;
675 int notify, req_id, ret;
677 if (addr->sa_family != AF_INET || sock->type != SOCK_STREAM)
680 map = pvcalls_enter_sock(sock);
683 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
685 spin_lock(&bedata->socket_lock);
686 ret = get_request(bedata, &req_id);
688 spin_unlock(&bedata->socket_lock);
689 pvcalls_exit_sock(sock);
692 req = RING_GET_REQUEST(&bedata->ring, req_id);
693 req->req_id = req_id;
695 req->cmd = PVCALLS_BIND;
696 req->u.bind.id = (uintptr_t)map;
697 memcpy(req->u.bind.addr, addr, sizeof(*addr));
698 req->u.bind.len = addr_len;
700 init_waitqueue_head(&map->passive.inflight_accept_req);
702 map->active_socket = false;
704 bedata->ring.req_prod_pvt++;
705 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
706 spin_unlock(&bedata->socket_lock);
708 notify_remote_via_irq(bedata->irq);
710 wait_event(bedata->inflight_req,
711 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
713 /* read req_id, then the content */
715 ret = bedata->rsp[req_id].ret;
716 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
718 map->passive.status = PVCALLS_STATUS_BIND;
719 pvcalls_exit_sock(sock);
723 int pvcalls_front_listen(struct socket *sock, int backlog)
725 struct pvcalls_bedata *bedata;
726 struct sock_mapping *map;
727 struct xen_pvcalls_request *req;
728 int notify, req_id, ret;
730 map = pvcalls_enter_sock(sock);
733 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
735 if (map->passive.status != PVCALLS_STATUS_BIND) {
736 pvcalls_exit_sock(sock);
740 spin_lock(&bedata->socket_lock);
741 ret = get_request(bedata, &req_id);
743 spin_unlock(&bedata->socket_lock);
744 pvcalls_exit_sock(sock);
747 req = RING_GET_REQUEST(&bedata->ring, req_id);
748 req->req_id = req_id;
749 req->cmd = PVCALLS_LISTEN;
750 req->u.listen.id = (uintptr_t) map;
751 req->u.listen.backlog = backlog;
753 bedata->ring.req_prod_pvt++;
754 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
755 spin_unlock(&bedata->socket_lock);
757 notify_remote_via_irq(bedata->irq);
759 wait_event(bedata->inflight_req,
760 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
762 /* read req_id, then the content */
764 ret = bedata->rsp[req_id].ret;
765 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
767 map->passive.status = PVCALLS_STATUS_LISTEN;
768 pvcalls_exit_sock(sock);
772 int pvcalls_front_accept(struct socket *sock, struct socket *newsock, int flags)
774 struct pvcalls_bedata *bedata;
775 struct sock_mapping *map;
776 struct sock_mapping *map2 = NULL;
777 struct xen_pvcalls_request *req;
778 int notify, req_id, ret, evtchn, nonblock;
780 map = pvcalls_enter_sock(sock);
783 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
785 if (map->passive.status != PVCALLS_STATUS_LISTEN) {
786 pvcalls_exit_sock(sock);
790 nonblock = flags & SOCK_NONBLOCK;
792 * Backend only supports 1 inflight accept request, will return
793 * errors for the others
795 if (test_and_set_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
796 (void *)&map->passive.flags)) {
797 req_id = READ_ONCE(map->passive.inflight_req_id);
798 if (req_id != PVCALLS_INVALID_ID &&
799 READ_ONCE(bedata->rsp[req_id].req_id) == req_id) {
800 map2 = map->passive.accept_map;
804 pvcalls_exit_sock(sock);
807 if (wait_event_interruptible(map->passive.inflight_accept_req,
808 !test_and_set_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
809 (void *)&map->passive.flags))) {
810 pvcalls_exit_sock(sock);
815 map2 = kzalloc(sizeof(*map2), GFP_KERNEL);
817 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
818 (void *)&map->passive.flags);
819 pvcalls_exit_sock(sock);
822 ret = alloc_active_ring(map2);
824 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
825 (void *)&map->passive.flags);
827 pvcalls_exit_sock(sock);
830 spin_lock(&bedata->socket_lock);
831 ret = get_request(bedata, &req_id);
833 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
834 (void *)&map->passive.flags);
835 spin_unlock(&bedata->socket_lock);
836 free_active_ring(map2);
838 pvcalls_exit_sock(sock);
842 ret = create_active(map2, &evtchn);
844 free_active_ring(map2);
846 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
847 (void *)&map->passive.flags);
848 spin_unlock(&bedata->socket_lock);
849 pvcalls_exit_sock(sock);
852 list_add_tail(&map2->list, &bedata->socket_mappings);
854 req = RING_GET_REQUEST(&bedata->ring, req_id);
855 req->req_id = req_id;
856 req->cmd = PVCALLS_ACCEPT;
857 req->u.accept.id = (uintptr_t) map;
858 req->u.accept.ref = map2->active.ref;
859 req->u.accept.id_new = (uintptr_t) map2;
860 req->u.accept.evtchn = evtchn;
861 map->passive.accept_map = map2;
863 bedata->ring.req_prod_pvt++;
864 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
865 spin_unlock(&bedata->socket_lock);
867 notify_remote_via_irq(bedata->irq);
868 /* We could check if we have received a response before returning. */
870 WRITE_ONCE(map->passive.inflight_req_id, req_id);
871 pvcalls_exit_sock(sock);
875 if (wait_event_interruptible(bedata->inflight_req,
876 READ_ONCE(bedata->rsp[req_id].req_id) == req_id)) {
877 pvcalls_exit_sock(sock);
880 /* read req_id, then the content */
884 map2->sock = newsock;
885 newsock->sk = sk_alloc(sock_net(sock->sk), PF_INET, GFP_KERNEL, &pvcalls_proto, false);
887 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
888 map->passive.inflight_req_id = PVCALLS_INVALID_ID;
889 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
890 (void *)&map->passive.flags);
891 pvcalls_front_free_map(bedata, map2);
892 pvcalls_exit_sock(sock);
895 newsock->sk->sk_send_head = (void *)map2;
897 ret = bedata->rsp[req_id].ret;
898 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
899 map->passive.inflight_req_id = PVCALLS_INVALID_ID;
901 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT, (void *)&map->passive.flags);
902 wake_up(&map->passive.inflight_accept_req);
904 pvcalls_exit_sock(sock);
908 static __poll_t pvcalls_front_poll_passive(struct file *file,
909 struct pvcalls_bedata *bedata,
910 struct sock_mapping *map,
913 int notify, req_id, ret;
914 struct xen_pvcalls_request *req;
916 if (test_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
917 (void *)&map->passive.flags)) {
918 uint32_t req_id = READ_ONCE(map->passive.inflight_req_id);
920 if (req_id != PVCALLS_INVALID_ID &&
921 READ_ONCE(bedata->rsp[req_id].req_id) == req_id)
922 return EPOLLIN | EPOLLRDNORM;
924 poll_wait(file, &map->passive.inflight_accept_req, wait);
928 if (test_and_clear_bit(PVCALLS_FLAG_POLL_RET,
929 (void *)&map->passive.flags))
930 return EPOLLIN | EPOLLRDNORM;
933 * First check RET, then INFLIGHT. No barriers necessary to
934 * ensure execution ordering because of the conditional
935 * instructions creating control dependencies.
938 if (test_and_set_bit(PVCALLS_FLAG_POLL_INFLIGHT,
939 (void *)&map->passive.flags)) {
940 poll_wait(file, &bedata->inflight_req, wait);
944 spin_lock(&bedata->socket_lock);
945 ret = get_request(bedata, &req_id);
947 spin_unlock(&bedata->socket_lock);
950 req = RING_GET_REQUEST(&bedata->ring, req_id);
951 req->req_id = req_id;
952 req->cmd = PVCALLS_POLL;
953 req->u.poll.id = (uintptr_t) map;
955 bedata->ring.req_prod_pvt++;
956 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
957 spin_unlock(&bedata->socket_lock);
959 notify_remote_via_irq(bedata->irq);
961 poll_wait(file, &bedata->inflight_req, wait);
965 static __poll_t pvcalls_front_poll_active(struct file *file,
966 struct pvcalls_bedata *bedata,
967 struct sock_mapping *map,
971 int32_t in_error, out_error;
972 struct pvcalls_data_intf *intf = map->active.ring;
974 out_error = intf->out_error;
975 in_error = intf->in_error;
977 poll_wait(file, &map->active.inflight_conn_req, wait);
978 if (pvcalls_front_write_todo(map))
979 mask |= EPOLLOUT | EPOLLWRNORM;
980 if (pvcalls_front_read_todo(map))
981 mask |= EPOLLIN | EPOLLRDNORM;
982 if (in_error != 0 || out_error != 0)
988 __poll_t pvcalls_front_poll(struct file *file, struct socket *sock,
991 struct pvcalls_bedata *bedata;
992 struct sock_mapping *map;
995 map = pvcalls_enter_sock(sock);
998 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
1000 if (map->active_socket)
1001 ret = pvcalls_front_poll_active(file, bedata, map, wait);
1003 ret = pvcalls_front_poll_passive(file, bedata, map, wait);
1004 pvcalls_exit_sock(sock);
1008 int pvcalls_front_release(struct socket *sock)
1010 struct pvcalls_bedata *bedata;
1011 struct sock_mapping *map;
1012 int req_id, notify, ret;
1013 struct xen_pvcalls_request *req;
1015 if (sock->sk == NULL)
1018 map = pvcalls_enter_sock(sock);
1020 if (PTR_ERR(map) == -ENOTCONN)
1025 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
1027 spin_lock(&bedata->socket_lock);
1028 ret = get_request(bedata, &req_id);
1030 spin_unlock(&bedata->socket_lock);
1031 pvcalls_exit_sock(sock);
1034 sock->sk->sk_send_head = NULL;
1036 req = RING_GET_REQUEST(&bedata->ring, req_id);
1037 req->req_id = req_id;
1038 req->cmd = PVCALLS_RELEASE;
1039 req->u.release.id = (uintptr_t)map;
1041 bedata->ring.req_prod_pvt++;
1042 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
1043 spin_unlock(&bedata->socket_lock);
1045 notify_remote_via_irq(bedata->irq);
1047 wait_event(bedata->inflight_req,
1048 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
1050 if (map->active_socket) {
1052 * Set in_error and wake up inflight_conn_req to force
1053 * recvmsg waiters to exit.
1055 map->active.ring->in_error = -EBADF;
1056 wake_up_interruptible(&map->active.inflight_conn_req);
1059 * We need to make sure that sendmsg/recvmsg on this socket have
1060 * not started before we've cleared sk_send_head here. The
1061 * easiest way to guarantee this is to see that no pvcalls
1062 * (other than us) is in progress on this socket.
1064 while (atomic_read(&map->refcount) > 1)
1067 pvcalls_front_free_map(bedata, map);
1069 wake_up(&bedata->inflight_req);
1070 wake_up(&map->passive.inflight_accept_req);
1072 while (atomic_read(&map->refcount) > 1)
1075 spin_lock(&bedata->socket_lock);
1076 list_del(&map->list);
1077 spin_unlock(&bedata->socket_lock);
1078 if (READ_ONCE(map->passive.inflight_req_id) != PVCALLS_INVALID_ID &&
1079 READ_ONCE(map->passive.inflight_req_id) != 0) {
1080 pvcalls_front_free_map(bedata,
1081 map->passive.accept_map);
1085 WRITE_ONCE(bedata->rsp[req_id].req_id, PVCALLS_INVALID_ID);
1091 static const struct xenbus_device_id pvcalls_front_ids[] = {
1096 static int pvcalls_front_remove(struct xenbus_device *dev)
1098 struct pvcalls_bedata *bedata;
1099 struct sock_mapping *map = NULL, *n;
1101 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
1102 dev_set_drvdata(&dev->dev, NULL);
1103 pvcalls_front_dev = NULL;
1104 if (bedata->irq >= 0)
1105 unbind_from_irqhandler(bedata->irq, dev);
1107 list_for_each_entry_safe(map, n, &bedata->socket_mappings, list) {
1108 map->sock->sk->sk_send_head = NULL;
1109 if (map->active_socket) {
1110 map->active.ring->in_error = -EBADF;
1111 wake_up_interruptible(&map->active.inflight_conn_req);
1116 while (atomic_read(&pvcalls_refcount) > 0)
1118 list_for_each_entry_safe(map, n, &bedata->socket_mappings, list) {
1119 if (map->active_socket) {
1120 /* No need to lock, refcount is 0 */
1121 pvcalls_front_free_map(bedata, map);
1123 list_del(&map->list);
1127 if (bedata->ref != -1)
1128 gnttab_end_foreign_access(bedata->ref, 0, 0);
1129 kfree(bedata->ring.sring);
1131 xenbus_switch_state(dev, XenbusStateClosed);
1135 static int pvcalls_front_probe(struct xenbus_device *dev,
1136 const struct xenbus_device_id *id)
1138 int ret = -ENOMEM, evtchn, i;
1139 unsigned int max_page_order, function_calls, len;
1141 grant_ref_t gref_head = 0;
1142 struct xenbus_transaction xbt;
1143 struct pvcalls_bedata *bedata = NULL;
1144 struct xen_pvcalls_sring *sring;
1146 if (pvcalls_front_dev != NULL) {
1147 dev_err(&dev->dev, "only one PV Calls connection supported\n");
1151 versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
1152 if (IS_ERR(versions))
1153 return PTR_ERR(versions);
1156 if (strcmp(versions, "1")) {
1161 max_page_order = xenbus_read_unsigned(dev->otherend,
1162 "max-page-order", 0);
1163 if (max_page_order < PVCALLS_RING_ORDER)
1165 function_calls = xenbus_read_unsigned(dev->otherend,
1166 "function-calls", 0);
1167 /* See XENBUS_FUNCTIONS_CALLS in pvcalls.h */
1168 if (function_calls != 1)
1170 pr_info("%s max-page-order is %u\n", __func__, max_page_order);
1172 bedata = kzalloc(sizeof(struct pvcalls_bedata), GFP_KERNEL);
1176 dev_set_drvdata(&dev->dev, bedata);
1177 pvcalls_front_dev = dev;
1178 init_waitqueue_head(&bedata->inflight_req);
1179 INIT_LIST_HEAD(&bedata->socket_mappings);
1180 spin_lock_init(&bedata->socket_lock);
1184 for (i = 0; i < PVCALLS_NR_RSP_PER_RING; i++)
1185 bedata->rsp[i].req_id = PVCALLS_INVALID_ID;
1187 sring = (struct xen_pvcalls_sring *) __get_free_page(GFP_KERNEL |
1191 SHARED_RING_INIT(sring);
1192 FRONT_RING_INIT(&bedata->ring, sring, XEN_PAGE_SIZE);
1194 ret = xenbus_alloc_evtchn(dev, &evtchn);
1198 bedata->irq = bind_evtchn_to_irqhandler(evtchn,
1199 pvcalls_front_event_handler,
1200 0, "pvcalls-frontend", dev);
1201 if (bedata->irq < 0) {
1206 ret = gnttab_alloc_grant_references(1, &gref_head);
1209 ret = gnttab_claim_grant_reference(&gref_head);
1213 gnttab_grant_foreign_access_ref(bedata->ref, dev->otherend_id,
1214 virt_to_gfn((void *)sring), 0);
1217 ret = xenbus_transaction_start(&xbt);
1219 xenbus_dev_fatal(dev, ret, "starting transaction");
1222 ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
1225 ret = xenbus_printf(xbt, dev->nodename, "ring-ref", "%d", bedata->ref);
1228 ret = xenbus_printf(xbt, dev->nodename, "port", "%u",
1232 ret = xenbus_transaction_end(xbt, 0);
1236 xenbus_dev_fatal(dev, ret, "completing transaction");
1239 xenbus_switch_state(dev, XenbusStateInitialised);
1244 xenbus_transaction_end(xbt, 1);
1245 xenbus_dev_fatal(dev, ret, "writing xenstore");
1247 pvcalls_front_remove(dev);
1251 static void pvcalls_front_changed(struct xenbus_device *dev,
1252 enum xenbus_state backend_state)
1254 switch (backend_state) {
1255 case XenbusStateReconfiguring:
1256 case XenbusStateReconfigured:
1257 case XenbusStateInitialising:
1258 case XenbusStateInitialised:
1259 case XenbusStateUnknown:
1262 case XenbusStateInitWait:
1265 case XenbusStateConnected:
1266 xenbus_switch_state(dev, XenbusStateConnected);
1269 case XenbusStateClosed:
1270 if (dev->state == XenbusStateClosed)
1272 /* Missed the backend's CLOSING state */
1274 case XenbusStateClosing:
1275 xenbus_frontend_closed(dev);
1280 static struct xenbus_driver pvcalls_front_driver = {
1281 .ids = pvcalls_front_ids,
1282 .probe = pvcalls_front_probe,
1283 .remove = pvcalls_front_remove,
1284 .otherend_changed = pvcalls_front_changed,
1287 static int __init pvcalls_frontend_init(void)
1292 pr_info("Initialising Xen pvcalls frontend driver\n");
1294 return xenbus_register_frontend(&pvcalls_front_driver);
1297 module_init(pvcalls_frontend_init);
1299 MODULE_DESCRIPTION("Xen PV Calls frontend driver");
1300 MODULE_AUTHOR("Stefano Stabellini <sstabellini@kernel.org>");
1301 MODULE_LICENSE("GPL");