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 int create_active(struct sock_mapping *map, int *evtchn)
347 int ret = -ENOMEM, irq = -1, i;
350 init_waitqueue_head(&map->active.inflight_conn_req);
352 map->active.ring = (struct pvcalls_data_intf *)
353 __get_free_page(GFP_KERNEL | __GFP_ZERO);
354 if (map->active.ring == NULL)
356 map->active.ring->ring_order = PVCALLS_RING_ORDER;
357 bytes = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
361 for (i = 0; i < (1 << PVCALLS_RING_ORDER); i++)
362 map->active.ring->ref[i] = gnttab_grant_foreign_access(
363 pvcalls_front_dev->otherend_id,
364 pfn_to_gfn(virt_to_pfn(bytes) + i), 0);
366 map->active.ref = gnttab_grant_foreign_access(
367 pvcalls_front_dev->otherend_id,
368 pfn_to_gfn(virt_to_pfn((void *)map->active.ring)), 0);
370 map->active.data.in = bytes;
371 map->active.data.out = bytes +
372 XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
374 ret = xenbus_alloc_evtchn(pvcalls_front_dev, evtchn);
377 irq = bind_evtchn_to_irqhandler(*evtchn, pvcalls_front_conn_handler,
378 0, "pvcalls-frontend", map);
384 map->active.irq = irq;
385 map->active_socket = true;
386 mutex_init(&map->active.in_mutex);
387 mutex_init(&map->active.out_mutex);
393 xenbus_free_evtchn(pvcalls_front_dev, *evtchn);
394 free_pages((unsigned long)map->active.data.in, PVCALLS_RING_ORDER);
395 free_page((unsigned long)map->active.ring);
399 int pvcalls_front_connect(struct socket *sock, struct sockaddr *addr,
400 int addr_len, int flags)
402 struct pvcalls_bedata *bedata;
403 struct sock_mapping *map = NULL;
404 struct xen_pvcalls_request *req;
405 int notify, req_id, ret, evtchn;
407 if (addr->sa_family != AF_INET || sock->type != SOCK_STREAM)
410 map = pvcalls_enter_sock(sock);
414 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
416 spin_lock(&bedata->socket_lock);
417 ret = get_request(bedata, &req_id);
419 spin_unlock(&bedata->socket_lock);
420 pvcalls_exit_sock(sock);
423 ret = create_active(map, &evtchn);
425 spin_unlock(&bedata->socket_lock);
426 pvcalls_exit_sock(sock);
430 req = RING_GET_REQUEST(&bedata->ring, req_id);
431 req->req_id = req_id;
432 req->cmd = PVCALLS_CONNECT;
433 req->u.connect.id = (uintptr_t)map;
434 req->u.connect.len = addr_len;
435 req->u.connect.flags = flags;
436 req->u.connect.ref = map->active.ref;
437 req->u.connect.evtchn = evtchn;
438 memcpy(req->u.connect.addr, addr, sizeof(*addr));
442 bedata->ring.req_prod_pvt++;
443 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
444 spin_unlock(&bedata->socket_lock);
447 notify_remote_via_irq(bedata->irq);
449 wait_event(bedata->inflight_req,
450 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
452 /* read req_id, then the content */
454 ret = bedata->rsp[req_id].ret;
455 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
456 pvcalls_exit_sock(sock);
460 static int __write_ring(struct pvcalls_data_intf *intf,
461 struct pvcalls_data *data,
462 struct iov_iter *msg_iter,
465 RING_IDX cons, prod, size, masked_prod, masked_cons;
466 RING_IDX array_size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
469 error = intf->out_error;
472 cons = intf->out_cons;
473 prod = intf->out_prod;
474 /* read indexes before continuing */
477 size = pvcalls_queued(prod, cons, array_size);
478 if (size >= array_size)
480 if (len > array_size - size)
481 len = array_size - size;
483 masked_prod = pvcalls_mask(prod, array_size);
484 masked_cons = pvcalls_mask(cons, array_size);
486 if (masked_prod < masked_cons) {
487 len = copy_from_iter(data->out + masked_prod, len, msg_iter);
489 if (len > array_size - masked_prod) {
490 int ret = copy_from_iter(data->out + masked_prod,
491 array_size - masked_prod, msg_iter);
492 if (ret != array_size - masked_prod) {
496 len = ret + copy_from_iter(data->out, len - ret, msg_iter);
498 len = copy_from_iter(data->out + masked_prod, len, msg_iter);
502 /* write to ring before updating pointer */
504 intf->out_prod += len;
509 int pvcalls_front_sendmsg(struct socket *sock, struct msghdr *msg,
512 struct pvcalls_bedata *bedata;
513 struct sock_mapping *map;
514 int sent, tot_sent = 0;
515 int count = 0, flags;
517 flags = msg->msg_flags;
518 if (flags & (MSG_CONFIRM|MSG_DONTROUTE|MSG_EOR|MSG_OOB))
521 map = pvcalls_enter_sock(sock);
524 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
526 mutex_lock(&map->active.out_mutex);
527 if ((flags & MSG_DONTWAIT) && !pvcalls_front_write_todo(map)) {
528 mutex_unlock(&map->active.out_mutex);
529 pvcalls_exit_sock(sock);
537 sent = __write_ring(map->active.ring,
538 &map->active.data, &msg->msg_iter,
543 notify_remote_via_irq(map->active.irq);
545 if (sent >= 0 && len > 0 && count < PVCALLS_FRONT_MAX_SPIN)
550 mutex_unlock(&map->active.out_mutex);
551 pvcalls_exit_sock(sock);
555 static int __read_ring(struct pvcalls_data_intf *intf,
556 struct pvcalls_data *data,
557 struct iov_iter *msg_iter,
558 size_t len, int flags)
560 RING_IDX cons, prod, size, masked_prod, masked_cons;
561 RING_IDX array_size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
564 cons = intf->in_cons;
565 prod = intf->in_prod;
566 error = intf->in_error;
567 /* get pointers before reading from the ring */
570 size = pvcalls_queued(prod, cons, array_size);
571 masked_prod = pvcalls_mask(prod, array_size);
572 masked_cons = pvcalls_mask(cons, array_size);
575 return error ?: size;
580 if (masked_prod > masked_cons) {
581 len = copy_to_iter(data->in + masked_cons, len, msg_iter);
583 if (len > (array_size - masked_cons)) {
584 int ret = copy_to_iter(data->in + masked_cons,
585 array_size - masked_cons, msg_iter);
586 if (ret != array_size - masked_cons) {
590 len = ret + copy_to_iter(data->in, len - ret, msg_iter);
592 len = copy_to_iter(data->in + masked_cons, len, msg_iter);
596 /* read data from the ring before increasing the index */
598 if (!(flags & MSG_PEEK))
599 intf->in_cons += len;
604 int pvcalls_front_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
607 struct pvcalls_bedata *bedata;
609 struct sock_mapping *map;
611 if (flags & (MSG_CMSG_CLOEXEC|MSG_ERRQUEUE|MSG_OOB|MSG_TRUNC))
614 map = pvcalls_enter_sock(sock);
617 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
619 mutex_lock(&map->active.in_mutex);
620 if (len > XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER))
621 len = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
623 while (!(flags & MSG_DONTWAIT) && !pvcalls_front_read_todo(map)) {
624 wait_event_interruptible(map->active.inflight_conn_req,
625 pvcalls_front_read_todo(map));
627 ret = __read_ring(map->active.ring, &map->active.data,
628 &msg->msg_iter, len, flags);
631 notify_remote_via_irq(map->active.irq);
633 ret = (flags & MSG_DONTWAIT) ? -EAGAIN : 0;
634 if (ret == -ENOTCONN)
637 mutex_unlock(&map->active.in_mutex);
638 pvcalls_exit_sock(sock);
642 int pvcalls_front_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
644 struct pvcalls_bedata *bedata;
645 struct sock_mapping *map = NULL;
646 struct xen_pvcalls_request *req;
647 int notify, req_id, ret;
649 if (addr->sa_family != AF_INET || sock->type != SOCK_STREAM)
652 map = pvcalls_enter_sock(sock);
655 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
657 spin_lock(&bedata->socket_lock);
658 ret = get_request(bedata, &req_id);
660 spin_unlock(&bedata->socket_lock);
661 pvcalls_exit_sock(sock);
664 req = RING_GET_REQUEST(&bedata->ring, req_id);
665 req->req_id = req_id;
667 req->cmd = PVCALLS_BIND;
668 req->u.bind.id = (uintptr_t)map;
669 memcpy(req->u.bind.addr, addr, sizeof(*addr));
670 req->u.bind.len = addr_len;
672 init_waitqueue_head(&map->passive.inflight_accept_req);
674 map->active_socket = false;
676 bedata->ring.req_prod_pvt++;
677 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
678 spin_unlock(&bedata->socket_lock);
680 notify_remote_via_irq(bedata->irq);
682 wait_event(bedata->inflight_req,
683 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
685 /* read req_id, then the content */
687 ret = bedata->rsp[req_id].ret;
688 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
690 map->passive.status = PVCALLS_STATUS_BIND;
691 pvcalls_exit_sock(sock);
695 int pvcalls_front_listen(struct socket *sock, int backlog)
697 struct pvcalls_bedata *bedata;
698 struct sock_mapping *map;
699 struct xen_pvcalls_request *req;
700 int notify, req_id, ret;
702 map = pvcalls_enter_sock(sock);
705 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
707 if (map->passive.status != PVCALLS_STATUS_BIND) {
708 pvcalls_exit_sock(sock);
712 spin_lock(&bedata->socket_lock);
713 ret = get_request(bedata, &req_id);
715 spin_unlock(&bedata->socket_lock);
716 pvcalls_exit_sock(sock);
719 req = RING_GET_REQUEST(&bedata->ring, req_id);
720 req->req_id = req_id;
721 req->cmd = PVCALLS_LISTEN;
722 req->u.listen.id = (uintptr_t) map;
723 req->u.listen.backlog = backlog;
725 bedata->ring.req_prod_pvt++;
726 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
727 spin_unlock(&bedata->socket_lock);
729 notify_remote_via_irq(bedata->irq);
731 wait_event(bedata->inflight_req,
732 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
734 /* read req_id, then the content */
736 ret = bedata->rsp[req_id].ret;
737 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
739 map->passive.status = PVCALLS_STATUS_LISTEN;
740 pvcalls_exit_sock(sock);
744 int pvcalls_front_accept(struct socket *sock, struct socket *newsock, int flags)
746 struct pvcalls_bedata *bedata;
747 struct sock_mapping *map;
748 struct sock_mapping *map2 = NULL;
749 struct xen_pvcalls_request *req;
750 int notify, req_id, ret, evtchn, nonblock;
752 map = pvcalls_enter_sock(sock);
755 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
757 if (map->passive.status != PVCALLS_STATUS_LISTEN) {
758 pvcalls_exit_sock(sock);
762 nonblock = flags & SOCK_NONBLOCK;
764 * Backend only supports 1 inflight accept request, will return
765 * errors for the others
767 if (test_and_set_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
768 (void *)&map->passive.flags)) {
769 req_id = READ_ONCE(map->passive.inflight_req_id);
770 if (req_id != PVCALLS_INVALID_ID &&
771 READ_ONCE(bedata->rsp[req_id].req_id) == req_id) {
772 map2 = map->passive.accept_map;
776 pvcalls_exit_sock(sock);
779 if (wait_event_interruptible(map->passive.inflight_accept_req,
780 !test_and_set_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
781 (void *)&map->passive.flags))) {
782 pvcalls_exit_sock(sock);
787 spin_lock(&bedata->socket_lock);
788 ret = get_request(bedata, &req_id);
790 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
791 (void *)&map->passive.flags);
792 spin_unlock(&bedata->socket_lock);
793 pvcalls_exit_sock(sock);
796 map2 = kzalloc(sizeof(*map2), GFP_ATOMIC);
798 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
799 (void *)&map->passive.flags);
800 spin_unlock(&bedata->socket_lock);
801 pvcalls_exit_sock(sock);
804 ret = create_active(map2, &evtchn);
807 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
808 (void *)&map->passive.flags);
809 spin_unlock(&bedata->socket_lock);
810 pvcalls_exit_sock(sock);
813 list_add_tail(&map2->list, &bedata->socket_mappings);
815 req = RING_GET_REQUEST(&bedata->ring, req_id);
816 req->req_id = req_id;
817 req->cmd = PVCALLS_ACCEPT;
818 req->u.accept.id = (uintptr_t) map;
819 req->u.accept.ref = map2->active.ref;
820 req->u.accept.id_new = (uintptr_t) map2;
821 req->u.accept.evtchn = evtchn;
822 map->passive.accept_map = map2;
824 bedata->ring.req_prod_pvt++;
825 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
826 spin_unlock(&bedata->socket_lock);
828 notify_remote_via_irq(bedata->irq);
829 /* We could check if we have received a response before returning. */
831 WRITE_ONCE(map->passive.inflight_req_id, req_id);
832 pvcalls_exit_sock(sock);
836 if (wait_event_interruptible(bedata->inflight_req,
837 READ_ONCE(bedata->rsp[req_id].req_id) == req_id)) {
838 pvcalls_exit_sock(sock);
841 /* read req_id, then the content */
845 map2->sock = newsock;
846 newsock->sk = sk_alloc(sock_net(sock->sk), PF_INET, GFP_KERNEL, &pvcalls_proto, false);
848 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
849 map->passive.inflight_req_id = PVCALLS_INVALID_ID;
850 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
851 (void *)&map->passive.flags);
852 pvcalls_front_free_map(bedata, map2);
853 pvcalls_exit_sock(sock);
856 newsock->sk->sk_send_head = (void *)map2;
858 ret = bedata->rsp[req_id].ret;
859 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
860 map->passive.inflight_req_id = PVCALLS_INVALID_ID;
862 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT, (void *)&map->passive.flags);
863 wake_up(&map->passive.inflight_accept_req);
865 pvcalls_exit_sock(sock);
869 static __poll_t pvcalls_front_poll_passive(struct file *file,
870 struct pvcalls_bedata *bedata,
871 struct sock_mapping *map,
874 int notify, req_id, ret;
875 struct xen_pvcalls_request *req;
877 if (test_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
878 (void *)&map->passive.flags)) {
879 uint32_t req_id = READ_ONCE(map->passive.inflight_req_id);
881 if (req_id != PVCALLS_INVALID_ID &&
882 READ_ONCE(bedata->rsp[req_id].req_id) == req_id)
883 return EPOLLIN | EPOLLRDNORM;
885 poll_wait(file, &map->passive.inflight_accept_req, wait);
889 if (test_and_clear_bit(PVCALLS_FLAG_POLL_RET,
890 (void *)&map->passive.flags))
891 return EPOLLIN | EPOLLRDNORM;
894 * First check RET, then INFLIGHT. No barriers necessary to
895 * ensure execution ordering because of the conditional
896 * instructions creating control dependencies.
899 if (test_and_set_bit(PVCALLS_FLAG_POLL_INFLIGHT,
900 (void *)&map->passive.flags)) {
901 poll_wait(file, &bedata->inflight_req, wait);
905 spin_lock(&bedata->socket_lock);
906 ret = get_request(bedata, &req_id);
908 spin_unlock(&bedata->socket_lock);
911 req = RING_GET_REQUEST(&bedata->ring, req_id);
912 req->req_id = req_id;
913 req->cmd = PVCALLS_POLL;
914 req->u.poll.id = (uintptr_t) map;
916 bedata->ring.req_prod_pvt++;
917 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
918 spin_unlock(&bedata->socket_lock);
920 notify_remote_via_irq(bedata->irq);
922 poll_wait(file, &bedata->inflight_req, wait);
926 static __poll_t pvcalls_front_poll_active(struct file *file,
927 struct pvcalls_bedata *bedata,
928 struct sock_mapping *map,
932 int32_t in_error, out_error;
933 struct pvcalls_data_intf *intf = map->active.ring;
935 out_error = intf->out_error;
936 in_error = intf->in_error;
938 poll_wait(file, &map->active.inflight_conn_req, wait);
939 if (pvcalls_front_write_todo(map))
940 mask |= EPOLLOUT | EPOLLWRNORM;
941 if (pvcalls_front_read_todo(map))
942 mask |= EPOLLIN | EPOLLRDNORM;
943 if (in_error != 0 || out_error != 0)
949 __poll_t pvcalls_front_poll(struct file *file, struct socket *sock,
952 struct pvcalls_bedata *bedata;
953 struct sock_mapping *map;
956 map = pvcalls_enter_sock(sock);
959 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
961 if (map->active_socket)
962 ret = pvcalls_front_poll_active(file, bedata, map, wait);
964 ret = pvcalls_front_poll_passive(file, bedata, map, wait);
965 pvcalls_exit_sock(sock);
969 int pvcalls_front_release(struct socket *sock)
971 struct pvcalls_bedata *bedata;
972 struct sock_mapping *map;
973 int req_id, notify, ret;
974 struct xen_pvcalls_request *req;
976 if (sock->sk == NULL)
979 map = pvcalls_enter_sock(sock);
981 if (PTR_ERR(map) == -ENOTCONN)
986 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
988 spin_lock(&bedata->socket_lock);
989 ret = get_request(bedata, &req_id);
991 spin_unlock(&bedata->socket_lock);
992 pvcalls_exit_sock(sock);
995 sock->sk->sk_send_head = NULL;
997 req = RING_GET_REQUEST(&bedata->ring, req_id);
998 req->req_id = req_id;
999 req->cmd = PVCALLS_RELEASE;
1000 req->u.release.id = (uintptr_t)map;
1002 bedata->ring.req_prod_pvt++;
1003 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
1004 spin_unlock(&bedata->socket_lock);
1006 notify_remote_via_irq(bedata->irq);
1008 wait_event(bedata->inflight_req,
1009 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
1011 if (map->active_socket) {
1013 * Set in_error and wake up inflight_conn_req to force
1014 * recvmsg waiters to exit.
1016 map->active.ring->in_error = -EBADF;
1017 wake_up_interruptible(&map->active.inflight_conn_req);
1020 * We need to make sure that sendmsg/recvmsg on this socket have
1021 * not started before we've cleared sk_send_head here. The
1022 * easiest way to guarantee this is to see that no pvcalls
1023 * (other than us) is in progress on this socket.
1025 while (atomic_read(&map->refcount) > 1)
1028 pvcalls_front_free_map(bedata, map);
1030 wake_up(&bedata->inflight_req);
1031 wake_up(&map->passive.inflight_accept_req);
1033 while (atomic_read(&map->refcount) > 1)
1036 spin_lock(&bedata->socket_lock);
1037 list_del(&map->list);
1038 spin_unlock(&bedata->socket_lock);
1039 if (READ_ONCE(map->passive.inflight_req_id) != PVCALLS_INVALID_ID &&
1040 READ_ONCE(map->passive.inflight_req_id) != 0) {
1041 pvcalls_front_free_map(bedata,
1042 map->passive.accept_map);
1046 WRITE_ONCE(bedata->rsp[req_id].req_id, PVCALLS_INVALID_ID);
1052 static const struct xenbus_device_id pvcalls_front_ids[] = {
1057 static int pvcalls_front_remove(struct xenbus_device *dev)
1059 struct pvcalls_bedata *bedata;
1060 struct sock_mapping *map = NULL, *n;
1062 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
1063 dev_set_drvdata(&dev->dev, NULL);
1064 pvcalls_front_dev = NULL;
1065 if (bedata->irq >= 0)
1066 unbind_from_irqhandler(bedata->irq, dev);
1068 list_for_each_entry_safe(map, n, &bedata->socket_mappings, list) {
1069 map->sock->sk->sk_send_head = NULL;
1070 if (map->active_socket) {
1071 map->active.ring->in_error = -EBADF;
1072 wake_up_interruptible(&map->active.inflight_conn_req);
1077 while (atomic_read(&pvcalls_refcount) > 0)
1079 list_for_each_entry_safe(map, n, &bedata->socket_mappings, list) {
1080 if (map->active_socket) {
1081 /* No need to lock, refcount is 0 */
1082 pvcalls_front_free_map(bedata, map);
1084 list_del(&map->list);
1088 if (bedata->ref != -1)
1089 gnttab_end_foreign_access(bedata->ref, 0, 0);
1090 kfree(bedata->ring.sring);
1092 xenbus_switch_state(dev, XenbusStateClosed);
1096 static int pvcalls_front_probe(struct xenbus_device *dev,
1097 const struct xenbus_device_id *id)
1099 int ret = -ENOMEM, evtchn, i;
1100 unsigned int max_page_order, function_calls, len;
1102 grant_ref_t gref_head = 0;
1103 struct xenbus_transaction xbt;
1104 struct pvcalls_bedata *bedata = NULL;
1105 struct xen_pvcalls_sring *sring;
1107 if (pvcalls_front_dev != NULL) {
1108 dev_err(&dev->dev, "only one PV Calls connection supported\n");
1112 versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
1113 if (IS_ERR(versions))
1114 return PTR_ERR(versions);
1117 if (strcmp(versions, "1")) {
1122 max_page_order = xenbus_read_unsigned(dev->otherend,
1123 "max-page-order", 0);
1124 if (max_page_order < PVCALLS_RING_ORDER)
1126 function_calls = xenbus_read_unsigned(dev->otherend,
1127 "function-calls", 0);
1128 /* See XENBUS_FUNCTIONS_CALLS in pvcalls.h */
1129 if (function_calls != 1)
1131 pr_info("%s max-page-order is %u\n", __func__, max_page_order);
1133 bedata = kzalloc(sizeof(struct pvcalls_bedata), GFP_KERNEL);
1137 dev_set_drvdata(&dev->dev, bedata);
1138 pvcalls_front_dev = dev;
1139 init_waitqueue_head(&bedata->inflight_req);
1140 INIT_LIST_HEAD(&bedata->socket_mappings);
1141 spin_lock_init(&bedata->socket_lock);
1145 for (i = 0; i < PVCALLS_NR_RSP_PER_RING; i++)
1146 bedata->rsp[i].req_id = PVCALLS_INVALID_ID;
1148 sring = (struct xen_pvcalls_sring *) __get_free_page(GFP_KERNEL |
1152 SHARED_RING_INIT(sring);
1153 FRONT_RING_INIT(&bedata->ring, sring, XEN_PAGE_SIZE);
1155 ret = xenbus_alloc_evtchn(dev, &evtchn);
1159 bedata->irq = bind_evtchn_to_irqhandler(evtchn,
1160 pvcalls_front_event_handler,
1161 0, "pvcalls-frontend", dev);
1162 if (bedata->irq < 0) {
1167 ret = gnttab_alloc_grant_references(1, &gref_head);
1170 ret = gnttab_claim_grant_reference(&gref_head);
1174 gnttab_grant_foreign_access_ref(bedata->ref, dev->otherend_id,
1175 virt_to_gfn((void *)sring), 0);
1178 ret = xenbus_transaction_start(&xbt);
1180 xenbus_dev_fatal(dev, ret, "starting transaction");
1183 ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
1186 ret = xenbus_printf(xbt, dev->nodename, "ring-ref", "%d", bedata->ref);
1189 ret = xenbus_printf(xbt, dev->nodename, "port", "%u",
1193 ret = xenbus_transaction_end(xbt, 0);
1197 xenbus_dev_fatal(dev, ret, "completing transaction");
1200 xenbus_switch_state(dev, XenbusStateInitialised);
1205 xenbus_transaction_end(xbt, 1);
1206 xenbus_dev_fatal(dev, ret, "writing xenstore");
1208 pvcalls_front_remove(dev);
1212 static void pvcalls_front_changed(struct xenbus_device *dev,
1213 enum xenbus_state backend_state)
1215 switch (backend_state) {
1216 case XenbusStateReconfiguring:
1217 case XenbusStateReconfigured:
1218 case XenbusStateInitialising:
1219 case XenbusStateInitialised:
1220 case XenbusStateUnknown:
1223 case XenbusStateInitWait:
1226 case XenbusStateConnected:
1227 xenbus_switch_state(dev, XenbusStateConnected);
1230 case XenbusStateClosed:
1231 if (dev->state == XenbusStateClosed)
1233 /* Missed the backend's CLOSING state */
1235 case XenbusStateClosing:
1236 xenbus_frontend_closed(dev);
1241 static struct xenbus_driver pvcalls_front_driver = {
1242 .ids = pvcalls_front_ids,
1243 .probe = pvcalls_front_probe,
1244 .remove = pvcalls_front_remove,
1245 .otherend_changed = pvcalls_front_changed,
1248 static int __init pvcalls_frontend_init(void)
1253 pr_info("Initialising Xen pvcalls frontend driver\n");
1255 return xenbus_register_frontend(&pvcalls_front_driver);
1258 module_init(pvcalls_frontend_init);
1260 MODULE_DESCRIPTION("Xen PV Calls frontend driver");
1261 MODULE_AUTHOR("Stefano Stabellini <sstabellini@kernel.org>");
1262 MODULE_LICENSE("GPL");