1 // SPDX-License-Identifier: GPL-2.0-or-later
6 #include <linux/module.h>
8 #include <linux/socket.h>
12 #include <xen/events.h>
13 #include <xen/grant_table.h>
15 #include <xen/xenbus.h>
16 #include <xen/interface/io/pvcalls.h>
18 #include "pvcalls-front.h"
20 #define PVCALLS_INVALID_ID UINT_MAX
21 #define PVCALLS_RING_ORDER XENBUS_MAX_RING_GRANT_ORDER
22 #define PVCALLS_NR_RSP_PER_RING __CONST_RING_SIZE(xen_pvcalls, XEN_PAGE_SIZE)
23 #define PVCALLS_FRONT_MAX_SPIN 5000
25 static struct proto pvcalls_proto = {
28 .obj_size = sizeof(struct sock),
31 struct pvcalls_bedata {
32 struct xen_pvcalls_front_ring ring;
36 struct list_head socket_mappings;
37 spinlock_t socket_lock;
39 wait_queue_head_t inflight_req;
40 struct xen_pvcalls_response rsp[PVCALLS_NR_RSP_PER_RING];
42 /* Only one front/back connection supported. */
43 static struct xenbus_device *pvcalls_front_dev;
44 static atomic_t pvcalls_refcount;
46 /* first increment refcount, then proceed */
47 #define pvcalls_enter() { \
48 atomic_inc(&pvcalls_refcount); \
51 /* first complete other operations, then decrement refcount */
52 #define pvcalls_exit() { \
53 atomic_dec(&pvcalls_refcount); \
58 struct list_head list;
65 struct pvcalls_data_intf *ring;
66 struct pvcalls_data data;
67 struct mutex in_mutex;
68 struct mutex out_mutex;
70 wait_queue_head_t inflight_conn_req;
74 * Socket status, needs to be 64-bit aligned due to the
75 * test_and_* functions which have this requirement on arm64.
77 #define PVCALLS_STATUS_UNINITALIZED 0
78 #define PVCALLS_STATUS_BIND 1
79 #define PVCALLS_STATUS_LISTEN 2
80 uint8_t status __attribute__((aligned(8)));
82 * Internal state-machine flags.
83 * Only one accept operation can be inflight for a socket.
84 * Only one poll operation can be inflight for a given socket.
85 * flags needs to be 64-bit aligned due to the test_and_*
86 * functions which have this requirement on arm64.
88 #define PVCALLS_FLAG_ACCEPT_INFLIGHT 0
89 #define PVCALLS_FLAG_POLL_INFLIGHT 1
90 #define PVCALLS_FLAG_POLL_RET 2
91 uint8_t flags __attribute__((aligned(8)));
92 uint32_t inflight_req_id;
93 struct sock_mapping *accept_map;
94 wait_queue_head_t inflight_accept_req;
99 static inline struct sock_mapping *pvcalls_enter_sock(struct socket *sock)
101 struct sock_mapping *map;
103 if (!pvcalls_front_dev ||
104 dev_get_drvdata(&pvcalls_front_dev->dev) == NULL)
105 return ERR_PTR(-ENOTCONN);
107 map = (struct sock_mapping *)sock->sk->sk_send_head;
109 return ERR_PTR(-ENOTSOCK);
112 atomic_inc(&map->refcount);
116 static inline void pvcalls_exit_sock(struct socket *sock)
118 struct sock_mapping *map;
120 map = (struct sock_mapping *)sock->sk->sk_send_head;
121 atomic_dec(&map->refcount);
125 static inline int get_request(struct pvcalls_bedata *bedata, int *req_id)
127 *req_id = bedata->ring.req_prod_pvt & (RING_SIZE(&bedata->ring) - 1);
128 if (RING_FULL(&bedata->ring) ||
129 bedata->rsp[*req_id].req_id != PVCALLS_INVALID_ID)
134 static bool pvcalls_front_write_todo(struct sock_mapping *map)
136 struct pvcalls_data_intf *intf = map->active.ring;
137 RING_IDX cons, prod, size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
140 error = intf->out_error;
141 if (error == -ENOTCONN)
146 cons = intf->out_cons;
147 prod = intf->out_prod;
148 return !!(size - pvcalls_queued(prod, cons, size));
151 static bool pvcalls_front_read_todo(struct sock_mapping *map)
153 struct pvcalls_data_intf *intf = map->active.ring;
157 cons = intf->in_cons;
158 prod = intf->in_prod;
159 error = intf->in_error;
160 return (error != 0 ||
161 pvcalls_queued(prod, cons,
162 XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER)) != 0);
165 static irqreturn_t pvcalls_front_event_handler(int irq, void *dev_id)
167 struct xenbus_device *dev = dev_id;
168 struct pvcalls_bedata *bedata;
169 struct xen_pvcalls_response *rsp;
171 int req_id = 0, more = 0, done = 0;
177 bedata = dev_get_drvdata(&dev->dev);
178 if (bedata == NULL) {
184 while (RING_HAS_UNCONSUMED_RESPONSES(&bedata->ring)) {
185 rsp = RING_GET_RESPONSE(&bedata->ring, bedata->ring.rsp_cons);
187 req_id = rsp->req_id;
188 if (rsp->cmd == PVCALLS_POLL) {
189 struct sock_mapping *map = (struct sock_mapping *)(uintptr_t)
192 clear_bit(PVCALLS_FLAG_POLL_INFLIGHT,
193 (void *)&map->passive.flags);
195 * clear INFLIGHT, then set RET. It pairs with
196 * the checks at the beginning of
197 * pvcalls_front_poll_passive.
200 set_bit(PVCALLS_FLAG_POLL_RET,
201 (void *)&map->passive.flags);
203 dst = (uint8_t *)&bedata->rsp[req_id] +
205 src = (uint8_t *)rsp + sizeof(rsp->req_id);
206 memcpy(dst, src, sizeof(*rsp) - sizeof(rsp->req_id));
208 * First copy the rest of the data, then req_id. It is
209 * paired with the barrier when accessing bedata->rsp.
212 bedata->rsp[req_id].req_id = req_id;
216 bedata->ring.rsp_cons++;
219 RING_FINAL_CHECK_FOR_RESPONSES(&bedata->ring, more);
223 wake_up(&bedata->inflight_req);
228 static void pvcalls_front_free_map(struct pvcalls_bedata *bedata,
229 struct sock_mapping *map)
233 unbind_from_irqhandler(map->active.irq, map);
235 spin_lock(&bedata->socket_lock);
236 if (!list_empty(&map->list))
237 list_del_init(&map->list);
238 spin_unlock(&bedata->socket_lock);
240 for (i = 0; i < (1 << PVCALLS_RING_ORDER); i++)
241 gnttab_end_foreign_access(map->active.ring->ref[i], 0, 0);
242 gnttab_end_foreign_access(map->active.ref, 0, 0);
243 free_page((unsigned long)map->active.ring);
248 static irqreturn_t pvcalls_front_conn_handler(int irq, void *sock_map)
250 struct sock_mapping *map = sock_map;
255 wake_up_interruptible(&map->active.inflight_conn_req);
260 int pvcalls_front_socket(struct socket *sock)
262 struct pvcalls_bedata *bedata;
263 struct sock_mapping *map = NULL;
264 struct xen_pvcalls_request *req;
265 int notify, req_id, ret;
268 * PVCalls only supports domain AF_INET,
269 * type SOCK_STREAM and protocol 0 sockets for now.
271 * Check socket type here, AF_INET and protocol checks are done
274 if (sock->type != SOCK_STREAM)
278 if (!pvcalls_front_dev) {
282 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
284 map = kzalloc(sizeof(*map), GFP_KERNEL);
290 spin_lock(&bedata->socket_lock);
292 ret = get_request(bedata, &req_id);
295 spin_unlock(&bedata->socket_lock);
301 * sock->sk->sk_send_head is not used for ip sockets: reuse the
302 * field to store a pointer to the struct sock_mapping
303 * corresponding to the socket. This way, we can easily get the
304 * struct sock_mapping from the struct socket.
306 sock->sk->sk_send_head = (void *)map;
307 list_add_tail(&map->list, &bedata->socket_mappings);
309 req = RING_GET_REQUEST(&bedata->ring, req_id);
310 req->req_id = req_id;
311 req->cmd = PVCALLS_SOCKET;
312 req->u.socket.id = (uintptr_t) map;
313 req->u.socket.domain = AF_INET;
314 req->u.socket.type = SOCK_STREAM;
315 req->u.socket.protocol = IPPROTO_IP;
317 bedata->ring.req_prod_pvt++;
318 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
319 spin_unlock(&bedata->socket_lock);
321 notify_remote_via_irq(bedata->irq);
323 wait_event(bedata->inflight_req,
324 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
326 /* read req_id, then the content */
328 ret = bedata->rsp[req_id].ret;
329 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
335 static void free_active_ring(struct sock_mapping *map)
337 if (!map->active.ring)
340 free_pages((unsigned long)map->active.data.in,
341 map->active.ring->ring_order);
342 free_page((unsigned long)map->active.ring);
345 static int alloc_active_ring(struct sock_mapping *map)
349 map->active.ring = (struct pvcalls_data_intf *)
350 get_zeroed_page(GFP_KERNEL);
351 if (!map->active.ring)
354 map->active.ring->ring_order = PVCALLS_RING_ORDER;
355 bytes = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
360 map->active.data.in = bytes;
361 map->active.data.out = bytes +
362 XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
367 free_active_ring(map);
371 static int create_active(struct sock_mapping *map, int *evtchn)
374 int ret = -ENOMEM, irq = -1, i;
377 init_waitqueue_head(&map->active.inflight_conn_req);
379 bytes = map->active.data.in;
380 for (i = 0; i < (1 << PVCALLS_RING_ORDER); i++)
381 map->active.ring->ref[i] = gnttab_grant_foreign_access(
382 pvcalls_front_dev->otherend_id,
383 pfn_to_gfn(virt_to_pfn(bytes) + i), 0);
385 map->active.ref = gnttab_grant_foreign_access(
386 pvcalls_front_dev->otherend_id,
387 pfn_to_gfn(virt_to_pfn((void *)map->active.ring)), 0);
389 ret = xenbus_alloc_evtchn(pvcalls_front_dev, evtchn);
392 irq = bind_evtchn_to_irqhandler(*evtchn, pvcalls_front_conn_handler,
393 0, "pvcalls-frontend", map);
399 map->active.irq = irq;
400 map->active_socket = true;
401 mutex_init(&map->active.in_mutex);
402 mutex_init(&map->active.out_mutex);
408 xenbus_free_evtchn(pvcalls_front_dev, *evtchn);
412 int pvcalls_front_connect(struct socket *sock, struct sockaddr *addr,
413 int addr_len, int flags)
415 struct pvcalls_bedata *bedata;
416 struct sock_mapping *map = NULL;
417 struct xen_pvcalls_request *req;
418 int notify, req_id, ret, evtchn;
420 if (addr->sa_family != AF_INET || sock->type != SOCK_STREAM)
423 map = pvcalls_enter_sock(sock);
427 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
428 ret = alloc_active_ring(map);
430 pvcalls_exit_sock(sock);
434 spin_lock(&bedata->socket_lock);
435 ret = get_request(bedata, &req_id);
437 spin_unlock(&bedata->socket_lock);
438 free_active_ring(map);
439 pvcalls_exit_sock(sock);
442 ret = create_active(map, &evtchn);
444 spin_unlock(&bedata->socket_lock);
445 free_active_ring(map);
446 pvcalls_exit_sock(sock);
450 req = RING_GET_REQUEST(&bedata->ring, req_id);
451 req->req_id = req_id;
452 req->cmd = PVCALLS_CONNECT;
453 req->u.connect.id = (uintptr_t)map;
454 req->u.connect.len = addr_len;
455 req->u.connect.flags = flags;
456 req->u.connect.ref = map->active.ref;
457 req->u.connect.evtchn = evtchn;
458 memcpy(req->u.connect.addr, addr, sizeof(*addr));
462 bedata->ring.req_prod_pvt++;
463 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
464 spin_unlock(&bedata->socket_lock);
467 notify_remote_via_irq(bedata->irq);
469 wait_event(bedata->inflight_req,
470 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
472 /* read req_id, then the content */
474 ret = bedata->rsp[req_id].ret;
475 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
476 pvcalls_exit_sock(sock);
480 static int __write_ring(struct pvcalls_data_intf *intf,
481 struct pvcalls_data *data,
482 struct iov_iter *msg_iter,
485 RING_IDX cons, prod, size, masked_prod, masked_cons;
486 RING_IDX array_size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
489 error = intf->out_error;
492 cons = intf->out_cons;
493 prod = intf->out_prod;
494 /* read indexes before continuing */
497 size = pvcalls_queued(prod, cons, array_size);
498 if (size > array_size)
500 if (size == array_size)
502 if (len > array_size - size)
503 len = array_size - size;
505 masked_prod = pvcalls_mask(prod, array_size);
506 masked_cons = pvcalls_mask(cons, array_size);
508 if (masked_prod < masked_cons) {
509 len = copy_from_iter(data->out + masked_prod, len, msg_iter);
511 if (len > array_size - masked_prod) {
512 int ret = copy_from_iter(data->out + masked_prod,
513 array_size - masked_prod, msg_iter);
514 if (ret != array_size - masked_prod) {
518 len = ret + copy_from_iter(data->out, len - ret, msg_iter);
520 len = copy_from_iter(data->out + masked_prod, len, msg_iter);
524 /* write to ring before updating pointer */
526 intf->out_prod += len;
531 int pvcalls_front_sendmsg(struct socket *sock, struct msghdr *msg,
534 struct sock_mapping *map;
535 int sent, tot_sent = 0;
536 int count = 0, flags;
538 flags = msg->msg_flags;
539 if (flags & (MSG_CONFIRM|MSG_DONTROUTE|MSG_EOR|MSG_OOB))
542 map = pvcalls_enter_sock(sock);
546 mutex_lock(&map->active.out_mutex);
547 if ((flags & MSG_DONTWAIT) && !pvcalls_front_write_todo(map)) {
548 mutex_unlock(&map->active.out_mutex);
549 pvcalls_exit_sock(sock);
557 sent = __write_ring(map->active.ring,
558 &map->active.data, &msg->msg_iter,
563 notify_remote_via_irq(map->active.irq);
565 if (sent >= 0 && len > 0 && count < PVCALLS_FRONT_MAX_SPIN)
570 mutex_unlock(&map->active.out_mutex);
571 pvcalls_exit_sock(sock);
575 static int __read_ring(struct pvcalls_data_intf *intf,
576 struct pvcalls_data *data,
577 struct iov_iter *msg_iter,
578 size_t len, int flags)
580 RING_IDX cons, prod, size, masked_prod, masked_cons;
581 RING_IDX array_size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
584 cons = intf->in_cons;
585 prod = intf->in_prod;
586 error = intf->in_error;
587 /* get pointers before reading from the ring */
590 size = pvcalls_queued(prod, cons, array_size);
591 masked_prod = pvcalls_mask(prod, array_size);
592 masked_cons = pvcalls_mask(cons, array_size);
595 return error ?: size;
600 if (masked_prod > masked_cons) {
601 len = copy_to_iter(data->in + masked_cons, len, msg_iter);
603 if (len > (array_size - masked_cons)) {
604 int ret = copy_to_iter(data->in + masked_cons,
605 array_size - masked_cons, msg_iter);
606 if (ret != array_size - masked_cons) {
610 len = ret + copy_to_iter(data->in, len - ret, msg_iter);
612 len = copy_to_iter(data->in + masked_cons, len, msg_iter);
616 /* read data from the ring before increasing the index */
618 if (!(flags & MSG_PEEK))
619 intf->in_cons += len;
624 int pvcalls_front_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
628 struct sock_mapping *map;
630 if (flags & (MSG_CMSG_CLOEXEC|MSG_ERRQUEUE|MSG_OOB|MSG_TRUNC))
633 map = pvcalls_enter_sock(sock);
637 mutex_lock(&map->active.in_mutex);
638 if (len > XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER))
639 len = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
641 while (!(flags & MSG_DONTWAIT) && !pvcalls_front_read_todo(map)) {
642 wait_event_interruptible(map->active.inflight_conn_req,
643 pvcalls_front_read_todo(map));
645 ret = __read_ring(map->active.ring, &map->active.data,
646 &msg->msg_iter, len, flags);
649 notify_remote_via_irq(map->active.irq);
651 ret = (flags & MSG_DONTWAIT) ? -EAGAIN : 0;
652 if (ret == -ENOTCONN)
655 mutex_unlock(&map->active.in_mutex);
656 pvcalls_exit_sock(sock);
660 int pvcalls_front_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
662 struct pvcalls_bedata *bedata;
663 struct sock_mapping *map = NULL;
664 struct xen_pvcalls_request *req;
665 int notify, req_id, ret;
667 if (addr->sa_family != AF_INET || sock->type != SOCK_STREAM)
670 map = pvcalls_enter_sock(sock);
673 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
675 spin_lock(&bedata->socket_lock);
676 ret = get_request(bedata, &req_id);
678 spin_unlock(&bedata->socket_lock);
679 pvcalls_exit_sock(sock);
682 req = RING_GET_REQUEST(&bedata->ring, req_id);
683 req->req_id = req_id;
685 req->cmd = PVCALLS_BIND;
686 req->u.bind.id = (uintptr_t)map;
687 memcpy(req->u.bind.addr, addr, sizeof(*addr));
688 req->u.bind.len = addr_len;
690 init_waitqueue_head(&map->passive.inflight_accept_req);
692 map->active_socket = false;
694 bedata->ring.req_prod_pvt++;
695 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
696 spin_unlock(&bedata->socket_lock);
698 notify_remote_via_irq(bedata->irq);
700 wait_event(bedata->inflight_req,
701 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
703 /* read req_id, then the content */
705 ret = bedata->rsp[req_id].ret;
706 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
708 map->passive.status = PVCALLS_STATUS_BIND;
709 pvcalls_exit_sock(sock);
713 int pvcalls_front_listen(struct socket *sock, int backlog)
715 struct pvcalls_bedata *bedata;
716 struct sock_mapping *map;
717 struct xen_pvcalls_request *req;
718 int notify, req_id, ret;
720 map = pvcalls_enter_sock(sock);
723 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
725 if (map->passive.status != PVCALLS_STATUS_BIND) {
726 pvcalls_exit_sock(sock);
730 spin_lock(&bedata->socket_lock);
731 ret = get_request(bedata, &req_id);
733 spin_unlock(&bedata->socket_lock);
734 pvcalls_exit_sock(sock);
737 req = RING_GET_REQUEST(&bedata->ring, req_id);
738 req->req_id = req_id;
739 req->cmd = PVCALLS_LISTEN;
740 req->u.listen.id = (uintptr_t) map;
741 req->u.listen.backlog = backlog;
743 bedata->ring.req_prod_pvt++;
744 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
745 spin_unlock(&bedata->socket_lock);
747 notify_remote_via_irq(bedata->irq);
749 wait_event(bedata->inflight_req,
750 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
752 /* read req_id, then the content */
754 ret = bedata->rsp[req_id].ret;
755 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
757 map->passive.status = PVCALLS_STATUS_LISTEN;
758 pvcalls_exit_sock(sock);
762 int pvcalls_front_accept(struct socket *sock, struct socket *newsock, int flags)
764 struct pvcalls_bedata *bedata;
765 struct sock_mapping *map;
766 struct sock_mapping *map2 = NULL;
767 struct xen_pvcalls_request *req;
768 int notify, req_id, ret, evtchn, nonblock;
770 map = pvcalls_enter_sock(sock);
773 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
775 if (map->passive.status != PVCALLS_STATUS_LISTEN) {
776 pvcalls_exit_sock(sock);
780 nonblock = flags & SOCK_NONBLOCK;
782 * Backend only supports 1 inflight accept request, will return
783 * errors for the others
785 if (test_and_set_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
786 (void *)&map->passive.flags)) {
787 req_id = READ_ONCE(map->passive.inflight_req_id);
788 if (req_id != PVCALLS_INVALID_ID &&
789 READ_ONCE(bedata->rsp[req_id].req_id) == req_id) {
790 map2 = map->passive.accept_map;
794 pvcalls_exit_sock(sock);
797 if (wait_event_interruptible(map->passive.inflight_accept_req,
798 !test_and_set_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
799 (void *)&map->passive.flags))) {
800 pvcalls_exit_sock(sock);
805 map2 = kzalloc(sizeof(*map2), GFP_KERNEL);
807 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
808 (void *)&map->passive.flags);
809 pvcalls_exit_sock(sock);
812 ret = alloc_active_ring(map2);
814 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
815 (void *)&map->passive.flags);
817 pvcalls_exit_sock(sock);
820 spin_lock(&bedata->socket_lock);
821 ret = get_request(bedata, &req_id);
823 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
824 (void *)&map->passive.flags);
825 spin_unlock(&bedata->socket_lock);
826 free_active_ring(map2);
828 pvcalls_exit_sock(sock);
832 ret = create_active(map2, &evtchn);
834 free_active_ring(map2);
836 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
837 (void *)&map->passive.flags);
838 spin_unlock(&bedata->socket_lock);
839 pvcalls_exit_sock(sock);
842 list_add_tail(&map2->list, &bedata->socket_mappings);
844 req = RING_GET_REQUEST(&bedata->ring, req_id);
845 req->req_id = req_id;
846 req->cmd = PVCALLS_ACCEPT;
847 req->u.accept.id = (uintptr_t) map;
848 req->u.accept.ref = map2->active.ref;
849 req->u.accept.id_new = (uintptr_t) map2;
850 req->u.accept.evtchn = evtchn;
851 map->passive.accept_map = map2;
853 bedata->ring.req_prod_pvt++;
854 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
855 spin_unlock(&bedata->socket_lock);
857 notify_remote_via_irq(bedata->irq);
858 /* We could check if we have received a response before returning. */
860 WRITE_ONCE(map->passive.inflight_req_id, req_id);
861 pvcalls_exit_sock(sock);
865 if (wait_event_interruptible(bedata->inflight_req,
866 READ_ONCE(bedata->rsp[req_id].req_id) == req_id)) {
867 pvcalls_exit_sock(sock);
870 /* read req_id, then the content */
874 map2->sock = newsock;
875 newsock->sk = sk_alloc(sock_net(sock->sk), PF_INET, GFP_KERNEL, &pvcalls_proto, false);
877 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
878 map->passive.inflight_req_id = PVCALLS_INVALID_ID;
879 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
880 (void *)&map->passive.flags);
881 pvcalls_front_free_map(bedata, map2);
882 pvcalls_exit_sock(sock);
885 newsock->sk->sk_send_head = (void *)map2;
887 ret = bedata->rsp[req_id].ret;
888 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
889 map->passive.inflight_req_id = PVCALLS_INVALID_ID;
891 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT, (void *)&map->passive.flags);
892 wake_up(&map->passive.inflight_accept_req);
894 pvcalls_exit_sock(sock);
898 static __poll_t pvcalls_front_poll_passive(struct file *file,
899 struct pvcalls_bedata *bedata,
900 struct sock_mapping *map,
903 int notify, req_id, ret;
904 struct xen_pvcalls_request *req;
906 if (test_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
907 (void *)&map->passive.flags)) {
908 uint32_t req_id = READ_ONCE(map->passive.inflight_req_id);
910 if (req_id != PVCALLS_INVALID_ID &&
911 READ_ONCE(bedata->rsp[req_id].req_id) == req_id)
912 return EPOLLIN | EPOLLRDNORM;
914 poll_wait(file, &map->passive.inflight_accept_req, wait);
918 if (test_and_clear_bit(PVCALLS_FLAG_POLL_RET,
919 (void *)&map->passive.flags))
920 return EPOLLIN | EPOLLRDNORM;
923 * First check RET, then INFLIGHT. No barriers necessary to
924 * ensure execution ordering because of the conditional
925 * instructions creating control dependencies.
928 if (test_and_set_bit(PVCALLS_FLAG_POLL_INFLIGHT,
929 (void *)&map->passive.flags)) {
930 poll_wait(file, &bedata->inflight_req, wait);
934 spin_lock(&bedata->socket_lock);
935 ret = get_request(bedata, &req_id);
937 spin_unlock(&bedata->socket_lock);
940 req = RING_GET_REQUEST(&bedata->ring, req_id);
941 req->req_id = req_id;
942 req->cmd = PVCALLS_POLL;
943 req->u.poll.id = (uintptr_t) map;
945 bedata->ring.req_prod_pvt++;
946 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
947 spin_unlock(&bedata->socket_lock);
949 notify_remote_via_irq(bedata->irq);
951 poll_wait(file, &bedata->inflight_req, wait);
955 static __poll_t pvcalls_front_poll_active(struct file *file,
956 struct pvcalls_bedata *bedata,
957 struct sock_mapping *map,
961 int32_t in_error, out_error;
962 struct pvcalls_data_intf *intf = map->active.ring;
964 out_error = intf->out_error;
965 in_error = intf->in_error;
967 poll_wait(file, &map->active.inflight_conn_req, wait);
968 if (pvcalls_front_write_todo(map))
969 mask |= EPOLLOUT | EPOLLWRNORM;
970 if (pvcalls_front_read_todo(map))
971 mask |= EPOLLIN | EPOLLRDNORM;
972 if (in_error != 0 || out_error != 0)
978 __poll_t pvcalls_front_poll(struct file *file, struct socket *sock,
981 struct pvcalls_bedata *bedata;
982 struct sock_mapping *map;
985 map = pvcalls_enter_sock(sock);
988 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
990 if (map->active_socket)
991 ret = pvcalls_front_poll_active(file, bedata, map, wait);
993 ret = pvcalls_front_poll_passive(file, bedata, map, wait);
994 pvcalls_exit_sock(sock);
998 int pvcalls_front_release(struct socket *sock)
1000 struct pvcalls_bedata *bedata;
1001 struct sock_mapping *map;
1002 int req_id, notify, ret;
1003 struct xen_pvcalls_request *req;
1005 if (sock->sk == NULL)
1008 map = pvcalls_enter_sock(sock);
1010 if (PTR_ERR(map) == -ENOTCONN)
1015 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
1017 spin_lock(&bedata->socket_lock);
1018 ret = get_request(bedata, &req_id);
1020 spin_unlock(&bedata->socket_lock);
1021 pvcalls_exit_sock(sock);
1024 sock->sk->sk_send_head = NULL;
1026 req = RING_GET_REQUEST(&bedata->ring, req_id);
1027 req->req_id = req_id;
1028 req->cmd = PVCALLS_RELEASE;
1029 req->u.release.id = (uintptr_t)map;
1031 bedata->ring.req_prod_pvt++;
1032 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
1033 spin_unlock(&bedata->socket_lock);
1035 notify_remote_via_irq(bedata->irq);
1037 wait_event(bedata->inflight_req,
1038 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
1040 if (map->active_socket) {
1042 * Set in_error and wake up inflight_conn_req to force
1043 * recvmsg waiters to exit.
1045 map->active.ring->in_error = -EBADF;
1046 wake_up_interruptible(&map->active.inflight_conn_req);
1049 * We need to make sure that sendmsg/recvmsg on this socket have
1050 * not started before we've cleared sk_send_head here. The
1051 * easiest way to guarantee this is to see that no pvcalls
1052 * (other than us) is in progress on this socket.
1054 while (atomic_read(&map->refcount) > 1)
1057 pvcalls_front_free_map(bedata, map);
1059 wake_up(&bedata->inflight_req);
1060 wake_up(&map->passive.inflight_accept_req);
1062 while (atomic_read(&map->refcount) > 1)
1065 spin_lock(&bedata->socket_lock);
1066 list_del(&map->list);
1067 spin_unlock(&bedata->socket_lock);
1068 if (READ_ONCE(map->passive.inflight_req_id) != PVCALLS_INVALID_ID &&
1069 READ_ONCE(map->passive.inflight_req_id) != 0) {
1070 pvcalls_front_free_map(bedata,
1071 map->passive.accept_map);
1075 WRITE_ONCE(bedata->rsp[req_id].req_id, PVCALLS_INVALID_ID);
1081 static const struct xenbus_device_id pvcalls_front_ids[] = {
1086 static int pvcalls_front_remove(struct xenbus_device *dev)
1088 struct pvcalls_bedata *bedata;
1089 struct sock_mapping *map = NULL, *n;
1091 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
1092 dev_set_drvdata(&dev->dev, NULL);
1093 pvcalls_front_dev = NULL;
1094 if (bedata->irq >= 0)
1095 unbind_from_irqhandler(bedata->irq, dev);
1097 list_for_each_entry_safe(map, n, &bedata->socket_mappings, list) {
1098 map->sock->sk->sk_send_head = NULL;
1099 if (map->active_socket) {
1100 map->active.ring->in_error = -EBADF;
1101 wake_up_interruptible(&map->active.inflight_conn_req);
1106 while (atomic_read(&pvcalls_refcount) > 0)
1108 list_for_each_entry_safe(map, n, &bedata->socket_mappings, list) {
1109 if (map->active_socket) {
1110 /* No need to lock, refcount is 0 */
1111 pvcalls_front_free_map(bedata, map);
1113 list_del(&map->list);
1117 if (bedata->ref != -1)
1118 gnttab_end_foreign_access(bedata->ref, 0, 0);
1119 kfree(bedata->ring.sring);
1121 xenbus_switch_state(dev, XenbusStateClosed);
1125 static int pvcalls_front_probe(struct xenbus_device *dev,
1126 const struct xenbus_device_id *id)
1128 int ret = -ENOMEM, evtchn, i;
1129 unsigned int max_page_order, function_calls, len;
1131 grant_ref_t gref_head = 0;
1132 struct xenbus_transaction xbt;
1133 struct pvcalls_bedata *bedata = NULL;
1134 struct xen_pvcalls_sring *sring;
1136 if (pvcalls_front_dev != NULL) {
1137 dev_err(&dev->dev, "only one PV Calls connection supported\n");
1141 versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
1142 if (IS_ERR(versions))
1143 return PTR_ERR(versions);
1146 if (strcmp(versions, "1")) {
1151 max_page_order = xenbus_read_unsigned(dev->otherend,
1152 "max-page-order", 0);
1153 if (max_page_order < PVCALLS_RING_ORDER)
1155 function_calls = xenbus_read_unsigned(dev->otherend,
1156 "function-calls", 0);
1157 /* See XENBUS_FUNCTIONS_CALLS in pvcalls.h */
1158 if (function_calls != 1)
1160 pr_info("%s max-page-order is %u\n", __func__, max_page_order);
1162 bedata = kzalloc(sizeof(struct pvcalls_bedata), GFP_KERNEL);
1166 dev_set_drvdata(&dev->dev, bedata);
1167 pvcalls_front_dev = dev;
1168 init_waitqueue_head(&bedata->inflight_req);
1169 INIT_LIST_HEAD(&bedata->socket_mappings);
1170 spin_lock_init(&bedata->socket_lock);
1174 for (i = 0; i < PVCALLS_NR_RSP_PER_RING; i++)
1175 bedata->rsp[i].req_id = PVCALLS_INVALID_ID;
1177 sring = (struct xen_pvcalls_sring *) __get_free_page(GFP_KERNEL |
1181 SHARED_RING_INIT(sring);
1182 FRONT_RING_INIT(&bedata->ring, sring, XEN_PAGE_SIZE);
1184 ret = xenbus_alloc_evtchn(dev, &evtchn);
1188 bedata->irq = bind_evtchn_to_irqhandler(evtchn,
1189 pvcalls_front_event_handler,
1190 0, "pvcalls-frontend", dev);
1191 if (bedata->irq < 0) {
1196 ret = gnttab_alloc_grant_references(1, &gref_head);
1199 ret = gnttab_claim_grant_reference(&gref_head);
1203 gnttab_grant_foreign_access_ref(bedata->ref, dev->otherend_id,
1204 virt_to_gfn((void *)sring), 0);
1207 ret = xenbus_transaction_start(&xbt);
1209 xenbus_dev_fatal(dev, ret, "starting transaction");
1212 ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
1215 ret = xenbus_printf(xbt, dev->nodename, "ring-ref", "%d", bedata->ref);
1218 ret = xenbus_printf(xbt, dev->nodename, "port", "%u",
1222 ret = xenbus_transaction_end(xbt, 0);
1226 xenbus_dev_fatal(dev, ret, "completing transaction");
1229 xenbus_switch_state(dev, XenbusStateInitialised);
1234 xenbus_transaction_end(xbt, 1);
1235 xenbus_dev_fatal(dev, ret, "writing xenstore");
1237 pvcalls_front_remove(dev);
1241 static void pvcalls_front_changed(struct xenbus_device *dev,
1242 enum xenbus_state backend_state)
1244 switch (backend_state) {
1245 case XenbusStateReconfiguring:
1246 case XenbusStateReconfigured:
1247 case XenbusStateInitialising:
1248 case XenbusStateInitialised:
1249 case XenbusStateUnknown:
1252 case XenbusStateInitWait:
1255 case XenbusStateConnected:
1256 xenbus_switch_state(dev, XenbusStateConnected);
1259 case XenbusStateClosed:
1260 if (dev->state == XenbusStateClosed)
1262 /* Missed the backend's CLOSING state */
1264 case XenbusStateClosing:
1265 xenbus_frontend_closed(dev);
1270 static struct xenbus_driver pvcalls_front_driver = {
1271 .ids = pvcalls_front_ids,
1272 .probe = pvcalls_front_probe,
1273 .remove = pvcalls_front_remove,
1274 .otherend_changed = pvcalls_front_changed,
1277 static int __init pvcalls_frontend_init(void)
1282 pr_info("Initialising Xen pvcalls frontend driver\n");
1284 return xenbus_register_frontend(&pvcalls_front_driver);
1287 module_init(pvcalls_frontend_init);
1289 MODULE_DESCRIPTION("Xen PV Calls frontend driver");
1291 MODULE_LICENSE("GPL");