1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * (c) 2017 Stefano Stabellini <stefano@aporeto.com>
4 */
5
6 #include <linux/module.h>
7 #include <linux/net.h>
8 #include <linux/socket.h>
9
10 #include <net/sock.h>
11
12 #include <xen/events.h>
13 #include <xen/grant_table.h>
14 #include <xen/xen.h>
15 #include <xen/xenbus.h>
16 #include <xen/interface/io/pvcalls.h>
17
18 #include "pvcalls-front.h"
19
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
24
25 static struct proto pvcalls_proto = {
26 .name = "PVCalls",
27 .owner = THIS_MODULE,
28 .obj_size = sizeof(struct sock),
29 };
30
31 struct pvcalls_bedata {
32 struct xen_pvcalls_front_ring ring;
33 grant_ref_t ref;
34 int irq;
35
36 struct list_head socket_mappings;
37 spinlock_t socket_lock;
38
39 wait_queue_head_t inflight_req;
40 struct xen_pvcalls_response rsp[PVCALLS_NR_RSP_PER_RING];
41 };
42 /* Only one front/back connection supported. */
43 static struct xenbus_device *pvcalls_front_dev;
44 static atomic_t pvcalls_refcount;
45
46 /* first increment refcount, then proceed */
47 #define pvcalls_enter() { \
48 atomic_inc(&pvcalls_refcount); \
49 }
50
51 /* first complete other operations, then decrement refcount */
52 #define pvcalls_exit() { \
53 atomic_dec(&pvcalls_refcount); \
54 }
55
56 struct sock_mapping {
57 bool active_socket;
58 struct list_head list;
59 struct socket *sock;
60 atomic_t refcount;
61 union {
62 struct {
63 int irq;
64 grant_ref_t ref;
65 struct pvcalls_data_intf *ring;
66 struct pvcalls_data data;
67 struct mutex in_mutex;
68 struct mutex out_mutex;
69
70 wait_queue_head_t inflight_conn_req;
71 } active;
72 struct {
73 /*
74 * Socket status, needs to be 64-bit aligned due to the
75 * test_and_* functions which have this requirement on arm64.
76 */
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)));
81 /*
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.
87 */
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;
95 } passive;
96 };
97 };
98
pvcalls_enter_sock(struct socket * sock)99 static inline struct sock_mapping *pvcalls_enter_sock(struct socket *sock)
100 {
101 struct sock_mapping *map;
102
103 if (!pvcalls_front_dev ||
104 dev_get_drvdata(&pvcalls_front_dev->dev) == NULL)
105 return ERR_PTR(-ENOTCONN);
106
107 map = (struct sock_mapping *)sock->sk->sk_send_head;
108 if (map == NULL)
109 return ERR_PTR(-ENOTSOCK);
110
111 pvcalls_enter();
112 atomic_inc(&map->refcount);
113 return map;
114 }
115
pvcalls_exit_sock(struct socket * sock)116 static inline void pvcalls_exit_sock(struct socket *sock)
117 {
118 struct sock_mapping *map;
119
120 map = (struct sock_mapping *)sock->sk->sk_send_head;
121 atomic_dec(&map->refcount);
122 pvcalls_exit();
123 }
124
get_request(struct pvcalls_bedata * bedata,int * req_id)125 static inline int get_request(struct pvcalls_bedata *bedata, int *req_id)
126 {
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)
130 return -EAGAIN;
131 return 0;
132 }
133
pvcalls_front_write_todo(struct sock_mapping * map)134 static bool pvcalls_front_write_todo(struct sock_mapping *map)
135 {
136 struct pvcalls_data_intf *intf = map->active.ring;
137 RING_IDX cons, prod, size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
138 int32_t error;
139
140 error = intf->out_error;
141 if (error == -ENOTCONN)
142 return false;
143 if (error != 0)
144 return true;
145
146 cons = intf->out_cons;
147 prod = intf->out_prod;
148 return !!(size - pvcalls_queued(prod, cons, size));
149 }
150
pvcalls_front_read_todo(struct sock_mapping * map)151 static bool pvcalls_front_read_todo(struct sock_mapping *map)
152 {
153 struct pvcalls_data_intf *intf = map->active.ring;
154 RING_IDX cons, prod;
155 int32_t error;
156
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);
163 }
164
pvcalls_front_event_handler(int irq,void * dev_id)165 static irqreturn_t pvcalls_front_event_handler(int irq, void *dev_id)
166 {
167 struct xenbus_device *dev = dev_id;
168 struct pvcalls_bedata *bedata;
169 struct xen_pvcalls_response *rsp;
170 uint8_t *src, *dst;
171 int req_id = 0, more = 0, done = 0;
172
173 if (dev == NULL)
174 return IRQ_HANDLED;
175
176 pvcalls_enter();
177 bedata = dev_get_drvdata(&dev->dev);
178 if (bedata == NULL) {
179 pvcalls_exit();
180 return IRQ_HANDLED;
181 }
182
183 again:
184 while (RING_HAS_UNCONSUMED_RESPONSES(&bedata->ring)) {
185 rsp = RING_GET_RESPONSE(&bedata->ring, bedata->ring.rsp_cons);
186
187 req_id = rsp->req_id;
188 if (rsp->cmd == PVCALLS_POLL) {
189 struct sock_mapping *map = (struct sock_mapping *)(uintptr_t)
190 rsp->u.poll.id;
191
192 clear_bit(PVCALLS_FLAG_POLL_INFLIGHT,
193 (void *)&map->passive.flags);
194 /*
195 * clear INFLIGHT, then set RET. It pairs with
196 * the checks at the beginning of
197 * pvcalls_front_poll_passive.
198 */
199 smp_wmb();
200 set_bit(PVCALLS_FLAG_POLL_RET,
201 (void *)&map->passive.flags);
202 } else {
203 dst = (uint8_t *)&bedata->rsp[req_id] +
204 sizeof(rsp->req_id);
205 src = (uint8_t *)rsp + sizeof(rsp->req_id);
206 memcpy(dst, src, sizeof(*rsp) - sizeof(rsp->req_id));
207 /*
208 * First copy the rest of the data, then req_id. It is
209 * paired with the barrier when accessing bedata->rsp.
210 */
211 smp_wmb();
212 bedata->rsp[req_id].req_id = req_id;
213 }
214
215 done = 1;
216 bedata->ring.rsp_cons++;
217 }
218
219 RING_FINAL_CHECK_FOR_RESPONSES(&bedata->ring, more);
220 if (more)
221 goto again;
222 if (done)
223 wake_up(&bedata->inflight_req);
224 pvcalls_exit();
225 return IRQ_HANDLED;
226 }
227
pvcalls_front_free_map(struct pvcalls_bedata * bedata,struct sock_mapping * map)228 static void pvcalls_front_free_map(struct pvcalls_bedata *bedata,
229 struct sock_mapping *map)
230 {
231 int i;
232
233 unbind_from_irqhandler(map->active.irq, map);
234
235 spin_lock(&bedata->socket_lock);
236 if (!list_empty(&map->list))
237 list_del_init(&map->list);
238 spin_unlock(&bedata->socket_lock);
239
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);
244
245 kfree(map);
246 }
247
pvcalls_front_conn_handler(int irq,void * sock_map)248 static irqreturn_t pvcalls_front_conn_handler(int irq, void *sock_map)
249 {
250 struct sock_mapping *map = sock_map;
251
252 if (map == NULL)
253 return IRQ_HANDLED;
254
255 wake_up_interruptible(&map->active.inflight_conn_req);
256
257 return IRQ_HANDLED;
258 }
259
pvcalls_front_socket(struct socket * sock)260 int pvcalls_front_socket(struct socket *sock)
261 {
262 struct pvcalls_bedata *bedata;
263 struct sock_mapping *map = NULL;
264 struct xen_pvcalls_request *req;
265 int notify, req_id, ret;
266
267 /*
268 * PVCalls only supports domain AF_INET,
269 * type SOCK_STREAM and protocol 0 sockets for now.
270 *
271 * Check socket type here, AF_INET and protocol checks are done
272 * by the caller.
273 */
274 if (sock->type != SOCK_STREAM)
275 return -EOPNOTSUPP;
276
277 pvcalls_enter();
278 if (!pvcalls_front_dev) {
279 pvcalls_exit();
280 return -EACCES;
281 }
282 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
283
284 map = kzalloc(sizeof(*map), GFP_KERNEL);
285 if (map == NULL) {
286 pvcalls_exit();
287 return -ENOMEM;
288 }
289
290 spin_lock(&bedata->socket_lock);
291
292 ret = get_request(bedata, &req_id);
293 if (ret < 0) {
294 kfree(map);
295 spin_unlock(&bedata->socket_lock);
296 pvcalls_exit();
297 return ret;
298 }
299
300 /*
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.
305 */
306 sock->sk->sk_send_head = (void *)map;
307 list_add_tail(&map->list, &bedata->socket_mappings);
308
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;
316
317 bedata->ring.req_prod_pvt++;
318 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
319 spin_unlock(&bedata->socket_lock);
320 if (notify)
321 notify_remote_via_irq(bedata->irq);
322
323 wait_event(bedata->inflight_req,
324 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
325
326 /* read req_id, then the content */
327 smp_rmb();
328 ret = bedata->rsp[req_id].ret;
329 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
330
331 pvcalls_exit();
332 return ret;
333 }
334
free_active_ring(struct sock_mapping * map)335 static void free_active_ring(struct sock_mapping *map)
336 {
337 if (!map->active.ring)
338 return;
339
340 free_pages_exact(map->active.data.in,
341 PAGE_SIZE << map->active.ring->ring_order);
342 free_page((unsigned long)map->active.ring);
343 }
344
alloc_active_ring(struct sock_mapping * map)345 static int alloc_active_ring(struct sock_mapping *map)
346 {
347 void *bytes;
348
349 map->active.ring = (struct pvcalls_data_intf *)
350 get_zeroed_page(GFP_KERNEL);
351 if (!map->active.ring)
352 goto out;
353
354 map->active.ring->ring_order = PVCALLS_RING_ORDER;
355 bytes = alloc_pages_exact(PAGE_SIZE << PVCALLS_RING_ORDER,
356 GFP_KERNEL | __GFP_ZERO);
357 if (!bytes)
358 goto out;
359
360 map->active.data.in = bytes;
361 map->active.data.out = bytes +
362 XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
363
364 return 0;
365
366 out:
367 free_active_ring(map);
368 return -ENOMEM;
369 }
370
create_active(struct sock_mapping * map,evtchn_port_t * evtchn)371 static int create_active(struct sock_mapping *map, evtchn_port_t *evtchn)
372 {
373 void *bytes;
374 int ret, irq = -1, i;
375
376 *evtchn = 0;
377 init_waitqueue_head(&map->active.inflight_conn_req);
378
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);
384
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);
388
389 ret = xenbus_alloc_evtchn(pvcalls_front_dev, evtchn);
390 if (ret)
391 goto out_error;
392 irq = bind_evtchn_to_irqhandler(*evtchn, pvcalls_front_conn_handler,
393 0, "pvcalls-frontend", map);
394 if (irq < 0) {
395 ret = irq;
396 goto out_error;
397 }
398
399 map->active.irq = irq;
400 map->active_socket = true;
401 mutex_init(&map->active.in_mutex);
402 mutex_init(&map->active.out_mutex);
403
404 return 0;
405
406 out_error:
407 if (*evtchn > 0)
408 xenbus_free_evtchn(pvcalls_front_dev, *evtchn);
409 return ret;
410 }
411
pvcalls_front_connect(struct socket * sock,struct sockaddr * addr,int addr_len,int flags)412 int pvcalls_front_connect(struct socket *sock, struct sockaddr *addr,
413 int addr_len, int flags)
414 {
415 struct pvcalls_bedata *bedata;
416 struct sock_mapping *map = NULL;
417 struct xen_pvcalls_request *req;
418 int notify, req_id, ret;
419 evtchn_port_t evtchn;
420
421 if (addr->sa_family != AF_INET || sock->type != SOCK_STREAM)
422 return -EOPNOTSUPP;
423
424 map = pvcalls_enter_sock(sock);
425 if (IS_ERR(map))
426 return PTR_ERR(map);
427
428 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
429 ret = alloc_active_ring(map);
430 if (ret < 0) {
431 pvcalls_exit_sock(sock);
432 return ret;
433 }
434
435 spin_lock(&bedata->socket_lock);
436 ret = get_request(bedata, &req_id);
437 if (ret < 0) {
438 spin_unlock(&bedata->socket_lock);
439 free_active_ring(map);
440 pvcalls_exit_sock(sock);
441 return ret;
442 }
443 ret = create_active(map, &evtchn);
444 if (ret < 0) {
445 spin_unlock(&bedata->socket_lock);
446 free_active_ring(map);
447 pvcalls_exit_sock(sock);
448 return ret;
449 }
450
451 req = RING_GET_REQUEST(&bedata->ring, req_id);
452 req->req_id = req_id;
453 req->cmd = PVCALLS_CONNECT;
454 req->u.connect.id = (uintptr_t)map;
455 req->u.connect.len = addr_len;
456 req->u.connect.flags = flags;
457 req->u.connect.ref = map->active.ref;
458 req->u.connect.evtchn = evtchn;
459 memcpy(req->u.connect.addr, addr, sizeof(*addr));
460
461 map->sock = sock;
462
463 bedata->ring.req_prod_pvt++;
464 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
465 spin_unlock(&bedata->socket_lock);
466
467 if (notify)
468 notify_remote_via_irq(bedata->irq);
469
470 wait_event(bedata->inflight_req,
471 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
472
473 /* read req_id, then the content */
474 smp_rmb();
475 ret = bedata->rsp[req_id].ret;
476 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
477 pvcalls_exit_sock(sock);
478 return ret;
479 }
480
__write_ring(struct pvcalls_data_intf * intf,struct pvcalls_data * data,struct iov_iter * msg_iter,int len)481 static int __write_ring(struct pvcalls_data_intf *intf,
482 struct pvcalls_data *data,
483 struct iov_iter *msg_iter,
484 int len)
485 {
486 RING_IDX cons, prod, size, masked_prod, masked_cons;
487 RING_IDX array_size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
488 int32_t error;
489
490 error = intf->out_error;
491 if (error < 0)
492 return error;
493 cons = intf->out_cons;
494 prod = intf->out_prod;
495 /* read indexes before continuing */
496 virt_mb();
497
498 size = pvcalls_queued(prod, cons, array_size);
499 if (size > array_size)
500 return -EINVAL;
501 if (size == array_size)
502 return 0;
503 if (len > array_size - size)
504 len = array_size - size;
505
506 masked_prod = pvcalls_mask(prod, array_size);
507 masked_cons = pvcalls_mask(cons, array_size);
508
509 if (masked_prod < masked_cons) {
510 len = copy_from_iter(data->out + masked_prod, len, msg_iter);
511 } else {
512 if (len > array_size - masked_prod) {
513 int ret = copy_from_iter(data->out + masked_prod,
514 array_size - masked_prod, msg_iter);
515 if (ret != array_size - masked_prod) {
516 len = ret;
517 goto out;
518 }
519 len = ret + copy_from_iter(data->out, len - ret, msg_iter);
520 } else {
521 len = copy_from_iter(data->out + masked_prod, len, msg_iter);
522 }
523 }
524 out:
525 /* write to ring before updating pointer */
526 virt_wmb();
527 intf->out_prod += len;
528
529 return len;
530 }
531
pvcalls_front_sendmsg(struct socket * sock,struct msghdr * msg,size_t len)532 int pvcalls_front_sendmsg(struct socket *sock, struct msghdr *msg,
533 size_t len)
534 {
535 struct sock_mapping *map;
536 int sent, tot_sent = 0;
537 int count = 0, flags;
538
539 flags = msg->msg_flags;
540 if (flags & (MSG_CONFIRM|MSG_DONTROUTE|MSG_EOR|MSG_OOB))
541 return -EOPNOTSUPP;
542
543 map = pvcalls_enter_sock(sock);
544 if (IS_ERR(map))
545 return PTR_ERR(map);
546
547 mutex_lock(&map->active.out_mutex);
548 if ((flags & MSG_DONTWAIT) && !pvcalls_front_write_todo(map)) {
549 mutex_unlock(&map->active.out_mutex);
550 pvcalls_exit_sock(sock);
551 return -EAGAIN;
552 }
553 if (len > INT_MAX)
554 len = INT_MAX;
555
556 again:
557 count++;
558 sent = __write_ring(map->active.ring,
559 &map->active.data, &msg->msg_iter,
560 len);
561 if (sent > 0) {
562 len -= sent;
563 tot_sent += sent;
564 notify_remote_via_irq(map->active.irq);
565 }
566 if (sent >= 0 && len > 0 && count < PVCALLS_FRONT_MAX_SPIN)
567 goto again;
568 if (sent < 0)
569 tot_sent = sent;
570
571 mutex_unlock(&map->active.out_mutex);
572 pvcalls_exit_sock(sock);
573 return tot_sent;
574 }
575
__read_ring(struct pvcalls_data_intf * intf,struct pvcalls_data * data,struct iov_iter * msg_iter,size_t len,int flags)576 static int __read_ring(struct pvcalls_data_intf *intf,
577 struct pvcalls_data *data,
578 struct iov_iter *msg_iter,
579 size_t len, int flags)
580 {
581 RING_IDX cons, prod, size, masked_prod, masked_cons;
582 RING_IDX array_size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
583 int32_t error;
584
585 cons = intf->in_cons;
586 prod = intf->in_prod;
587 error = intf->in_error;
588 /* get pointers before reading from the ring */
589 virt_rmb();
590
591 size = pvcalls_queued(prod, cons, array_size);
592 masked_prod = pvcalls_mask(prod, array_size);
593 masked_cons = pvcalls_mask(cons, array_size);
594
595 if (size == 0)
596 return error ?: size;
597
598 if (len > size)
599 len = size;
600
601 if (masked_prod > masked_cons) {
602 len = copy_to_iter(data->in + masked_cons, len, msg_iter);
603 } else {
604 if (len > (array_size - masked_cons)) {
605 int ret = copy_to_iter(data->in + masked_cons,
606 array_size - masked_cons, msg_iter);
607 if (ret != array_size - masked_cons) {
608 len = ret;
609 goto out;
610 }
611 len = ret + copy_to_iter(data->in, len - ret, msg_iter);
612 } else {
613 len = copy_to_iter(data->in + masked_cons, len, msg_iter);
614 }
615 }
616 out:
617 /* read data from the ring before increasing the index */
618 virt_mb();
619 if (!(flags & MSG_PEEK))
620 intf->in_cons += len;
621
622 return len;
623 }
624
pvcalls_front_recvmsg(struct socket * sock,struct msghdr * msg,size_t len,int flags)625 int pvcalls_front_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
626 int flags)
627 {
628 int ret;
629 struct sock_mapping *map;
630
631 if (flags & (MSG_CMSG_CLOEXEC|MSG_ERRQUEUE|MSG_OOB|MSG_TRUNC))
632 return -EOPNOTSUPP;
633
634 map = pvcalls_enter_sock(sock);
635 if (IS_ERR(map))
636 return PTR_ERR(map);
637
638 mutex_lock(&map->active.in_mutex);
639 if (len > XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER))
640 len = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
641
642 while (!(flags & MSG_DONTWAIT) && !pvcalls_front_read_todo(map)) {
643 wait_event_interruptible(map->active.inflight_conn_req,
644 pvcalls_front_read_todo(map));
645 }
646 ret = __read_ring(map->active.ring, &map->active.data,
647 &msg->msg_iter, len, flags);
648
649 if (ret > 0)
650 notify_remote_via_irq(map->active.irq);
651 if (ret == 0)
652 ret = (flags & MSG_DONTWAIT) ? -EAGAIN : 0;
653 if (ret == -ENOTCONN)
654 ret = 0;
655
656 mutex_unlock(&map->active.in_mutex);
657 pvcalls_exit_sock(sock);
658 return ret;
659 }
660
pvcalls_front_bind(struct socket * sock,struct sockaddr * addr,int addr_len)661 int pvcalls_front_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
662 {
663 struct pvcalls_bedata *bedata;
664 struct sock_mapping *map = NULL;
665 struct xen_pvcalls_request *req;
666 int notify, req_id, ret;
667
668 if (addr->sa_family != AF_INET || sock->type != SOCK_STREAM)
669 return -EOPNOTSUPP;
670
671 map = pvcalls_enter_sock(sock);
672 if (IS_ERR(map))
673 return PTR_ERR(map);
674 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
675
676 spin_lock(&bedata->socket_lock);
677 ret = get_request(bedata, &req_id);
678 if (ret < 0) {
679 spin_unlock(&bedata->socket_lock);
680 pvcalls_exit_sock(sock);
681 return ret;
682 }
683 req = RING_GET_REQUEST(&bedata->ring, req_id);
684 req->req_id = req_id;
685 map->sock = sock;
686 req->cmd = PVCALLS_BIND;
687 req->u.bind.id = (uintptr_t)map;
688 memcpy(req->u.bind.addr, addr, sizeof(*addr));
689 req->u.bind.len = addr_len;
690
691 init_waitqueue_head(&map->passive.inflight_accept_req);
692
693 map->active_socket = false;
694
695 bedata->ring.req_prod_pvt++;
696 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
697 spin_unlock(&bedata->socket_lock);
698 if (notify)
699 notify_remote_via_irq(bedata->irq);
700
701 wait_event(bedata->inflight_req,
702 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
703
704 /* read req_id, then the content */
705 smp_rmb();
706 ret = bedata->rsp[req_id].ret;
707 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
708
709 map->passive.status = PVCALLS_STATUS_BIND;
710 pvcalls_exit_sock(sock);
711 return 0;
712 }
713
pvcalls_front_listen(struct socket * sock,int backlog)714 int pvcalls_front_listen(struct socket *sock, int backlog)
715 {
716 struct pvcalls_bedata *bedata;
717 struct sock_mapping *map;
718 struct xen_pvcalls_request *req;
719 int notify, req_id, ret;
720
721 map = pvcalls_enter_sock(sock);
722 if (IS_ERR(map))
723 return PTR_ERR(map);
724 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
725
726 if (map->passive.status != PVCALLS_STATUS_BIND) {
727 pvcalls_exit_sock(sock);
728 return -EOPNOTSUPP;
729 }
730
731 spin_lock(&bedata->socket_lock);
732 ret = get_request(bedata, &req_id);
733 if (ret < 0) {
734 spin_unlock(&bedata->socket_lock);
735 pvcalls_exit_sock(sock);
736 return ret;
737 }
738 req = RING_GET_REQUEST(&bedata->ring, req_id);
739 req->req_id = req_id;
740 req->cmd = PVCALLS_LISTEN;
741 req->u.listen.id = (uintptr_t) map;
742 req->u.listen.backlog = backlog;
743
744 bedata->ring.req_prod_pvt++;
745 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
746 spin_unlock(&bedata->socket_lock);
747 if (notify)
748 notify_remote_via_irq(bedata->irq);
749
750 wait_event(bedata->inflight_req,
751 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
752
753 /* read req_id, then the content */
754 smp_rmb();
755 ret = bedata->rsp[req_id].ret;
756 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
757
758 map->passive.status = PVCALLS_STATUS_LISTEN;
759 pvcalls_exit_sock(sock);
760 return ret;
761 }
762
pvcalls_front_accept(struct socket * sock,struct socket * newsock,int flags)763 int pvcalls_front_accept(struct socket *sock, struct socket *newsock, int flags)
764 {
765 struct pvcalls_bedata *bedata;
766 struct sock_mapping *map;
767 struct sock_mapping *map2 = NULL;
768 struct xen_pvcalls_request *req;
769 int notify, req_id, ret, nonblock;
770 evtchn_port_t evtchn;
771
772 map = pvcalls_enter_sock(sock);
773 if (IS_ERR(map))
774 return PTR_ERR(map);
775 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
776
777 if (map->passive.status != PVCALLS_STATUS_LISTEN) {
778 pvcalls_exit_sock(sock);
779 return -EINVAL;
780 }
781
782 nonblock = flags & SOCK_NONBLOCK;
783 /*
784 * Backend only supports 1 inflight accept request, will return
785 * errors for the others
786 */
787 if (test_and_set_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
788 (void *)&map->passive.flags)) {
789 req_id = READ_ONCE(map->passive.inflight_req_id);
790 if (req_id != PVCALLS_INVALID_ID &&
791 READ_ONCE(bedata->rsp[req_id].req_id) == req_id) {
792 map2 = map->passive.accept_map;
793 goto received;
794 }
795 if (nonblock) {
796 pvcalls_exit_sock(sock);
797 return -EAGAIN;
798 }
799 if (wait_event_interruptible(map->passive.inflight_accept_req,
800 !test_and_set_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
801 (void *)&map->passive.flags))) {
802 pvcalls_exit_sock(sock);
803 return -EINTR;
804 }
805 }
806
807 map2 = kzalloc(sizeof(*map2), GFP_KERNEL);
808 if (map2 == NULL) {
809 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
810 (void *)&map->passive.flags);
811 pvcalls_exit_sock(sock);
812 return -ENOMEM;
813 }
814 ret = alloc_active_ring(map2);
815 if (ret < 0) {
816 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
817 (void *)&map->passive.flags);
818 kfree(map2);
819 pvcalls_exit_sock(sock);
820 return ret;
821 }
822 spin_lock(&bedata->socket_lock);
823 ret = get_request(bedata, &req_id);
824 if (ret < 0) {
825 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
826 (void *)&map->passive.flags);
827 spin_unlock(&bedata->socket_lock);
828 free_active_ring(map2);
829 kfree(map2);
830 pvcalls_exit_sock(sock);
831 return ret;
832 }
833
834 ret = create_active(map2, &evtchn);
835 if (ret < 0) {
836 free_active_ring(map2);
837 kfree(map2);
838 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
839 (void *)&map->passive.flags);
840 spin_unlock(&bedata->socket_lock);
841 pvcalls_exit_sock(sock);
842 return ret;
843 }
844 list_add_tail(&map2->list, &bedata->socket_mappings);
845
846 req = RING_GET_REQUEST(&bedata->ring, req_id);
847 req->req_id = req_id;
848 req->cmd = PVCALLS_ACCEPT;
849 req->u.accept.id = (uintptr_t) map;
850 req->u.accept.ref = map2->active.ref;
851 req->u.accept.id_new = (uintptr_t) map2;
852 req->u.accept.evtchn = evtchn;
853 map->passive.accept_map = map2;
854
855 bedata->ring.req_prod_pvt++;
856 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
857 spin_unlock(&bedata->socket_lock);
858 if (notify)
859 notify_remote_via_irq(bedata->irq);
860 /* We could check if we have received a response before returning. */
861 if (nonblock) {
862 WRITE_ONCE(map->passive.inflight_req_id, req_id);
863 pvcalls_exit_sock(sock);
864 return -EAGAIN;
865 }
866
867 if (wait_event_interruptible(bedata->inflight_req,
868 READ_ONCE(bedata->rsp[req_id].req_id) == req_id)) {
869 pvcalls_exit_sock(sock);
870 return -EINTR;
871 }
872 /* read req_id, then the content */
873 smp_rmb();
874
875 received:
876 map2->sock = newsock;
877 newsock->sk = sk_alloc(sock_net(sock->sk), PF_INET, GFP_KERNEL, &pvcalls_proto, false);
878 if (!newsock->sk) {
879 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
880 map->passive.inflight_req_id = PVCALLS_INVALID_ID;
881 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
882 (void *)&map->passive.flags);
883 pvcalls_front_free_map(bedata, map2);
884 pvcalls_exit_sock(sock);
885 return -ENOMEM;
886 }
887 newsock->sk->sk_send_head = (void *)map2;
888
889 ret = bedata->rsp[req_id].ret;
890 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
891 map->passive.inflight_req_id = PVCALLS_INVALID_ID;
892
893 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT, (void *)&map->passive.flags);
894 wake_up(&map->passive.inflight_accept_req);
895
896 pvcalls_exit_sock(sock);
897 return ret;
898 }
899
pvcalls_front_poll_passive(struct file * file,struct pvcalls_bedata * bedata,struct sock_mapping * map,poll_table * wait)900 static __poll_t pvcalls_front_poll_passive(struct file *file,
901 struct pvcalls_bedata *bedata,
902 struct sock_mapping *map,
903 poll_table *wait)
904 {
905 int notify, req_id, ret;
906 struct xen_pvcalls_request *req;
907
908 if (test_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
909 (void *)&map->passive.flags)) {
910 uint32_t req_id = READ_ONCE(map->passive.inflight_req_id);
911
912 if (req_id != PVCALLS_INVALID_ID &&
913 READ_ONCE(bedata->rsp[req_id].req_id) == req_id)
914 return EPOLLIN | EPOLLRDNORM;
915
916 poll_wait(file, &map->passive.inflight_accept_req, wait);
917 return 0;
918 }
919
920 if (test_and_clear_bit(PVCALLS_FLAG_POLL_RET,
921 (void *)&map->passive.flags))
922 return EPOLLIN | EPOLLRDNORM;
923
924 /*
925 * First check RET, then INFLIGHT. No barriers necessary to
926 * ensure execution ordering because of the conditional
927 * instructions creating control dependencies.
928 */
929
930 if (test_and_set_bit(PVCALLS_FLAG_POLL_INFLIGHT,
931 (void *)&map->passive.flags)) {
932 poll_wait(file, &bedata->inflight_req, wait);
933 return 0;
934 }
935
936 spin_lock(&bedata->socket_lock);
937 ret = get_request(bedata, &req_id);
938 if (ret < 0) {
939 spin_unlock(&bedata->socket_lock);
940 return ret;
941 }
942 req = RING_GET_REQUEST(&bedata->ring, req_id);
943 req->req_id = req_id;
944 req->cmd = PVCALLS_POLL;
945 req->u.poll.id = (uintptr_t) map;
946
947 bedata->ring.req_prod_pvt++;
948 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
949 spin_unlock(&bedata->socket_lock);
950 if (notify)
951 notify_remote_via_irq(bedata->irq);
952
953 poll_wait(file, &bedata->inflight_req, wait);
954 return 0;
955 }
956
pvcalls_front_poll_active(struct file * file,struct pvcalls_bedata * bedata,struct sock_mapping * map,poll_table * wait)957 static __poll_t pvcalls_front_poll_active(struct file *file,
958 struct pvcalls_bedata *bedata,
959 struct sock_mapping *map,
960 poll_table *wait)
961 {
962 __poll_t mask = 0;
963 int32_t in_error, out_error;
964 struct pvcalls_data_intf *intf = map->active.ring;
965
966 out_error = intf->out_error;
967 in_error = intf->in_error;
968
969 poll_wait(file, &map->active.inflight_conn_req, wait);
970 if (pvcalls_front_write_todo(map))
971 mask |= EPOLLOUT | EPOLLWRNORM;
972 if (pvcalls_front_read_todo(map))
973 mask |= EPOLLIN | EPOLLRDNORM;
974 if (in_error != 0 || out_error != 0)
975 mask |= EPOLLERR;
976
977 return mask;
978 }
979
pvcalls_front_poll(struct file * file,struct socket * sock,poll_table * wait)980 __poll_t pvcalls_front_poll(struct file *file, struct socket *sock,
981 poll_table *wait)
982 {
983 struct pvcalls_bedata *bedata;
984 struct sock_mapping *map;
985 __poll_t ret;
986
987 map = pvcalls_enter_sock(sock);
988 if (IS_ERR(map))
989 return EPOLLNVAL;
990 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
991
992 if (map->active_socket)
993 ret = pvcalls_front_poll_active(file, bedata, map, wait);
994 else
995 ret = pvcalls_front_poll_passive(file, bedata, map, wait);
996 pvcalls_exit_sock(sock);
997 return ret;
998 }
999
pvcalls_front_release(struct socket * sock)1000 int pvcalls_front_release(struct socket *sock)
1001 {
1002 struct pvcalls_bedata *bedata;
1003 struct sock_mapping *map;
1004 int req_id, notify, ret;
1005 struct xen_pvcalls_request *req;
1006
1007 if (sock->sk == NULL)
1008 return 0;
1009
1010 map = pvcalls_enter_sock(sock);
1011 if (IS_ERR(map)) {
1012 if (PTR_ERR(map) == -ENOTCONN)
1013 return -EIO;
1014 else
1015 return 0;
1016 }
1017 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
1018
1019 spin_lock(&bedata->socket_lock);
1020 ret = get_request(bedata, &req_id);
1021 if (ret < 0) {
1022 spin_unlock(&bedata->socket_lock);
1023 pvcalls_exit_sock(sock);
1024 return ret;
1025 }
1026 sock->sk->sk_send_head = NULL;
1027
1028 req = RING_GET_REQUEST(&bedata->ring, req_id);
1029 req->req_id = req_id;
1030 req->cmd = PVCALLS_RELEASE;
1031 req->u.release.id = (uintptr_t)map;
1032
1033 bedata->ring.req_prod_pvt++;
1034 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
1035 spin_unlock(&bedata->socket_lock);
1036 if (notify)
1037 notify_remote_via_irq(bedata->irq);
1038
1039 wait_event(bedata->inflight_req,
1040 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
1041
1042 if (map->active_socket) {
1043 /*
1044 * Set in_error and wake up inflight_conn_req to force
1045 * recvmsg waiters to exit.
1046 */
1047 map->active.ring->in_error = -EBADF;
1048 wake_up_interruptible(&map->active.inflight_conn_req);
1049
1050 /*
1051 * We need to make sure that sendmsg/recvmsg on this socket have
1052 * not started before we've cleared sk_send_head here. The
1053 * easiest way to guarantee this is to see that no pvcalls
1054 * (other than us) is in progress on this socket.
1055 */
1056 while (atomic_read(&map->refcount) > 1)
1057 cpu_relax();
1058
1059 pvcalls_front_free_map(bedata, map);
1060 } else {
1061 wake_up(&bedata->inflight_req);
1062 wake_up(&map->passive.inflight_accept_req);
1063
1064 while (atomic_read(&map->refcount) > 1)
1065 cpu_relax();
1066
1067 spin_lock(&bedata->socket_lock);
1068 list_del(&map->list);
1069 spin_unlock(&bedata->socket_lock);
1070 if (READ_ONCE(map->passive.inflight_req_id) != PVCALLS_INVALID_ID &&
1071 READ_ONCE(map->passive.inflight_req_id) != 0) {
1072 pvcalls_front_free_map(bedata,
1073 map->passive.accept_map);
1074 }
1075 kfree(map);
1076 }
1077 WRITE_ONCE(bedata->rsp[req_id].req_id, PVCALLS_INVALID_ID);
1078
1079 pvcalls_exit();
1080 return 0;
1081 }
1082
1083 static const struct xenbus_device_id pvcalls_front_ids[] = {
1084 { "pvcalls" },
1085 { "" }
1086 };
1087
pvcalls_front_remove(struct xenbus_device * dev)1088 static int pvcalls_front_remove(struct xenbus_device *dev)
1089 {
1090 struct pvcalls_bedata *bedata;
1091 struct sock_mapping *map = NULL, *n;
1092
1093 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
1094 dev_set_drvdata(&dev->dev, NULL);
1095 pvcalls_front_dev = NULL;
1096 if (bedata->irq >= 0)
1097 unbind_from_irqhandler(bedata->irq, dev);
1098
1099 list_for_each_entry_safe(map, n, &bedata->socket_mappings, list) {
1100 map->sock->sk->sk_send_head = NULL;
1101 if (map->active_socket) {
1102 map->active.ring->in_error = -EBADF;
1103 wake_up_interruptible(&map->active.inflight_conn_req);
1104 }
1105 }
1106
1107 smp_mb();
1108 while (atomic_read(&pvcalls_refcount) > 0)
1109 cpu_relax();
1110 list_for_each_entry_safe(map, n, &bedata->socket_mappings, list) {
1111 if (map->active_socket) {
1112 /* No need to lock, refcount is 0 */
1113 pvcalls_front_free_map(bedata, map);
1114 } else {
1115 list_del(&map->list);
1116 kfree(map);
1117 }
1118 }
1119 if (bedata->ref != -1)
1120 gnttab_end_foreign_access(bedata->ref, 0, 0);
1121 kfree(bedata->ring.sring);
1122 kfree(bedata);
1123 xenbus_switch_state(dev, XenbusStateClosed);
1124 return 0;
1125 }
1126
pvcalls_front_probe(struct xenbus_device * dev,const struct xenbus_device_id * id)1127 static int pvcalls_front_probe(struct xenbus_device *dev,
1128 const struct xenbus_device_id *id)
1129 {
1130 int ret = -ENOMEM, i;
1131 evtchn_port_t evtchn;
1132 unsigned int max_page_order, function_calls, len;
1133 char *versions;
1134 grant_ref_t gref_head = 0;
1135 struct xenbus_transaction xbt;
1136 struct pvcalls_bedata *bedata = NULL;
1137 struct xen_pvcalls_sring *sring;
1138
1139 if (pvcalls_front_dev != NULL) {
1140 dev_err(&dev->dev, "only one PV Calls connection supported\n");
1141 return -EINVAL;
1142 }
1143
1144 versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
1145 if (IS_ERR(versions))
1146 return PTR_ERR(versions);
1147 if (!len)
1148 return -EINVAL;
1149 if (strcmp(versions, "1")) {
1150 kfree(versions);
1151 return -EINVAL;
1152 }
1153 kfree(versions);
1154 max_page_order = xenbus_read_unsigned(dev->otherend,
1155 "max-page-order", 0);
1156 if (max_page_order < PVCALLS_RING_ORDER)
1157 return -ENODEV;
1158 function_calls = xenbus_read_unsigned(dev->otherend,
1159 "function-calls", 0);
1160 /* See XENBUS_FUNCTIONS_CALLS in pvcalls.h */
1161 if (function_calls != 1)
1162 return -ENODEV;
1163 pr_info("%s max-page-order is %u\n", __func__, max_page_order);
1164
1165 bedata = kzalloc(sizeof(struct pvcalls_bedata), GFP_KERNEL);
1166 if (!bedata)
1167 return -ENOMEM;
1168
1169 dev_set_drvdata(&dev->dev, bedata);
1170 pvcalls_front_dev = dev;
1171 init_waitqueue_head(&bedata->inflight_req);
1172 INIT_LIST_HEAD(&bedata->socket_mappings);
1173 spin_lock_init(&bedata->socket_lock);
1174 bedata->irq = -1;
1175 bedata->ref = -1;
1176
1177 for (i = 0; i < PVCALLS_NR_RSP_PER_RING; i++)
1178 bedata->rsp[i].req_id = PVCALLS_INVALID_ID;
1179
1180 sring = (struct xen_pvcalls_sring *) __get_free_page(GFP_KERNEL |
1181 __GFP_ZERO);
1182 if (!sring)
1183 goto error;
1184 SHARED_RING_INIT(sring);
1185 FRONT_RING_INIT(&bedata->ring, sring, XEN_PAGE_SIZE);
1186
1187 ret = xenbus_alloc_evtchn(dev, &evtchn);
1188 if (ret)
1189 goto error;
1190
1191 bedata->irq = bind_evtchn_to_irqhandler(evtchn,
1192 pvcalls_front_event_handler,
1193 0, "pvcalls-frontend", dev);
1194 if (bedata->irq < 0) {
1195 ret = bedata->irq;
1196 goto error;
1197 }
1198
1199 ret = gnttab_alloc_grant_references(1, &gref_head);
1200 if (ret < 0)
1201 goto error;
1202 ret = gnttab_claim_grant_reference(&gref_head);
1203 if (ret < 0)
1204 goto error;
1205 bedata->ref = ret;
1206 gnttab_grant_foreign_access_ref(bedata->ref, dev->otherend_id,
1207 virt_to_gfn((void *)sring), 0);
1208
1209 again:
1210 ret = xenbus_transaction_start(&xbt);
1211 if (ret) {
1212 xenbus_dev_fatal(dev, ret, "starting transaction");
1213 goto error;
1214 }
1215 ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
1216 if (ret)
1217 goto error_xenbus;
1218 ret = xenbus_printf(xbt, dev->nodename, "ring-ref", "%d", bedata->ref);
1219 if (ret)
1220 goto error_xenbus;
1221 ret = xenbus_printf(xbt, dev->nodename, "port", "%u",
1222 evtchn);
1223 if (ret)
1224 goto error_xenbus;
1225 ret = xenbus_transaction_end(xbt, 0);
1226 if (ret) {
1227 if (ret == -EAGAIN)
1228 goto again;
1229 xenbus_dev_fatal(dev, ret, "completing transaction");
1230 goto error;
1231 }
1232 xenbus_switch_state(dev, XenbusStateInitialised);
1233
1234 return 0;
1235
1236 error_xenbus:
1237 xenbus_transaction_end(xbt, 1);
1238 xenbus_dev_fatal(dev, ret, "writing xenstore");
1239 error:
1240 pvcalls_front_remove(dev);
1241 return ret;
1242 }
1243
pvcalls_front_changed(struct xenbus_device * dev,enum xenbus_state backend_state)1244 static void pvcalls_front_changed(struct xenbus_device *dev,
1245 enum xenbus_state backend_state)
1246 {
1247 switch (backend_state) {
1248 case XenbusStateReconfiguring:
1249 case XenbusStateReconfigured:
1250 case XenbusStateInitialising:
1251 case XenbusStateInitialised:
1252 case XenbusStateUnknown:
1253 break;
1254
1255 case XenbusStateInitWait:
1256 break;
1257
1258 case XenbusStateConnected:
1259 xenbus_switch_state(dev, XenbusStateConnected);
1260 break;
1261
1262 case XenbusStateClosed:
1263 if (dev->state == XenbusStateClosed)
1264 break;
1265 /* Missed the backend's CLOSING state */
1266 fallthrough;
1267 case XenbusStateClosing:
1268 xenbus_frontend_closed(dev);
1269 break;
1270 }
1271 }
1272
1273 static struct xenbus_driver pvcalls_front_driver = {
1274 .ids = pvcalls_front_ids,
1275 .probe = pvcalls_front_probe,
1276 .remove = pvcalls_front_remove,
1277 .otherend_changed = pvcalls_front_changed,
1278 };
1279
pvcalls_frontend_init(void)1280 static int __init pvcalls_frontend_init(void)
1281 {
1282 if (!xen_domain())
1283 return -ENODEV;
1284
1285 pr_info("Initialising Xen pvcalls frontend driver\n");
1286
1287 return xenbus_register_frontend(&pvcalls_front_driver);
1288 }
1289
1290 module_init(pvcalls_frontend_init);
1291
1292 MODULE_DESCRIPTION("Xen PV Calls frontend driver");
1293 MODULE_AUTHOR("Stefano Stabellini <sstabellini@kernel.org>");
1294 MODULE_LICENSE("GPL");
1295