]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - drivers/xen/pvcalls-front.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
pvcalls-front: properly allocate sk
[mirror_ubuntu-bionic-kernel.git] / drivers / xen / pvcalls-front.c
CommitLineData
416efba0
SS		 1/*
2 * (c) 2017 Stefano Stabellini <stefano@aporeto.com>
3 *
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.
8 *
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.
13 */
14
15#include <linux/module.h>
2195046b
SS		 16#include <linux/net.h>
17#include <linux/socket.h>
18
19#include <net/sock.h>
416efba0
SS		 20
21#include <xen/events.h>
22#include <xen/grant_table.h>
23#include <xen/xen.h>
24#include <xen/xenbus.h>
25#include <xen/interface/io/pvcalls.h>
26
2195046b
SS		 27#include "pvcalls-front.h"
28
aa7ba376
SS		 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)
45ddce21 32#define PVCALLS_FRONT_MAX_SPIN 5000
aa7ba376 33
63be5d01
SS		 34static struct proto pvcalls_proto = {
35 .name = "PVCalls",
36 .owner = THIS_MODULE,
37 .obj_size = sizeof(struct sock),
38};
39
aa7ba376
SS		 40struct pvcalls_bedata {
41 struct xen_pvcalls_front_ring ring;
42 grant_ref_t ref;
43 int irq;
44
45 struct list_head socket_mappings;
46 spinlock_t socket_lock;
47
48 wait_queue_head_t inflight_req;
49 struct xen_pvcalls_response rsp[PVCALLS_NR_RSP_PER_RING];
50};
51/* Only one front/back connection supported. */
52static struct xenbus_device *pvcalls_front_dev;
53static atomic_t pvcalls_refcount;
54
55/* first increment refcount, then proceed */
56#define pvcalls_enter() { \
57 atomic_inc(&pvcalls_refcount); \
58}
59
60/* first complete other operations, then decrement refcount */
61#define pvcalls_exit() { \
62 atomic_dec(&pvcalls_refcount); \
63}
64
65struct sock_mapping {
66 bool active_socket;
67 struct list_head list;
68 struct socket *sock;
cb1c7d9b
SS		 69 union {
70 struct {
71 int irq;
72 grant_ref_t ref;
73 struct pvcalls_data_intf *ring;
74 struct pvcalls_data data;
75 struct mutex in_mutex;
76 struct mutex out_mutex;
77
78 wait_queue_head_t inflight_conn_req;
79 } active;
67ea9893
SS		 80 struct {
81 /* Socket status */
82#define PVCALLS_STATUS_UNINITALIZED 0
83#define PVCALLS_STATUS_BIND 1
84#define PVCALLS_STATUS_LISTEN 2
85 uint8_t status;
9774c6cc
SS		 86 /*
87 * Internal state-machine flags.
88 * Only one accept operation can be inflight for a socket.
89 * Only one poll operation can be inflight for a given socket.
90 */
91#define PVCALLS_FLAG_ACCEPT_INFLIGHT 0
5842c835
SS		 92#define PVCALLS_FLAG_POLL_INFLIGHT 1
93#define PVCALLS_FLAG_POLL_RET 2
9774c6cc
SS		 94 uint8_t flags;
95 uint32_t inflight_req_id;
96 struct sock_mapping *accept_map;
97 wait_queue_head_t inflight_accept_req;
67ea9893 98 } passive;
cb1c7d9b 99 };
aa7ba376
SS		 100};
101
2195046b
SS		 102static inline int get_request(struct pvcalls_bedata *bedata, int *req_id)
103{
104 *req_id = bedata->ring.req_prod_pvt & (RING_SIZE(&bedata->ring) - 1);
105 if (RING_FULL(&bedata->ring) ||
106 bedata->rsp[*req_id].req_id != PVCALLS_INVALID_ID)
107 return -EAGAIN;
108 return 0;
109}
110
45ddce21
SS		 111static bool pvcalls_front_write_todo(struct sock_mapping *map)
112{
113 struct pvcalls_data_intf *intf = map->active.ring;
114 RING_IDX cons, prod, size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
115 int32_t error;
116
117 error = intf->out_error;
118 if (error == -ENOTCONN)
119 return false;
120 if (error != 0)
121 return true;
122
123 cons = intf->out_cons;
124 prod = intf->out_prod;
125 return !!(size - pvcalls_queued(prod, cons, size));
126}
127
ae0d0405
SS		 128static bool pvcalls_front_read_todo(struct sock_mapping *map)
129{
130 struct pvcalls_data_intf *intf = map->active.ring;
131 RING_IDX cons, prod;
132 int32_t error;
133
134 cons = intf->in_cons;
135 prod = intf->in_prod;
136 error = intf->in_error;
137 return (error != 0 ||
138 pvcalls_queued(prod, cons,
139 XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER)) != 0);
140}
141
aa7ba376
SS		 142static irqreturn_t pvcalls_front_event_handler(int irq, void *dev_id)
143{
2195046b
SS		 144 struct xenbus_device *dev = dev_id;
145 struct pvcalls_bedata *bedata;
146 struct xen_pvcalls_response *rsp;
147 uint8_t *src, *dst;
148 int req_id = 0, more = 0, done = 0;
149
150 if (dev == NULL)
151 return IRQ_HANDLED;
152
153 pvcalls_enter();
154 bedata = dev_get_drvdata(&dev->dev);
155 if (bedata == NULL) {
156 pvcalls_exit();
157 return IRQ_HANDLED;
158 }
159
160again:
161 while (RING_HAS_UNCONSUMED_RESPONSES(&bedata->ring)) {
162 rsp = RING_GET_RESPONSE(&bedata->ring, bedata->ring.rsp_cons);
163
164 req_id = rsp->req_id;
5842c835
SS		 165 if (rsp->cmd == PVCALLS_POLL) {
166 struct sock_mapping *map = (struct sock_mapping *)(uintptr_t)
167 rsp->u.poll.id;
168
169 clear_bit(PVCALLS_FLAG_POLL_INFLIGHT,
170 (void *)&map->passive.flags);
171 /*
172 * clear INFLIGHT, then set RET. It pairs with
173 * the checks at the beginning of
174 * pvcalls_front_poll_passive.
175 */
176 smp_wmb();
177 set_bit(PVCALLS_FLAG_POLL_RET,
178 (void *)&map->passive.flags);
179 } else {
180 dst = (uint8_t *)&bedata->rsp[req_id] +
181 sizeof(rsp->req_id);
182 src = (uint8_t *)rsp + sizeof(rsp->req_id);
183 memcpy(dst, src, sizeof(*rsp) - sizeof(rsp->req_id));
184 /*
185 * First copy the rest of the data, then req_id. It is
186 * paired with the barrier when accessing bedata->rsp.
187 */
188 smp_wmb();
189 bedata->rsp[req_id].req_id = req_id;
190 }
2195046b
SS		 191
192 done = 1;
193 bedata->ring.rsp_cons++;
194 }
195
196 RING_FINAL_CHECK_FOR_RESPONSES(&bedata->ring, more);
197 if (more)
198 goto again;
199 if (done)
200 wake_up(&bedata->inflight_req);
201 pvcalls_exit();
aa7ba376
SS		 202 return IRQ_HANDLED;
203}
204
205static void pvcalls_front_free_map(struct pvcalls_bedata *bedata,
206 struct sock_mapping *map)
207{
235a71c5
SS		 208 int i;
209
210 unbind_from_irqhandler(map->active.irq, map);
211
212 spin_lock(&bedata->socket_lock);
213 if (!list_empty(&map->list))
214 list_del_init(&map->list);
215 spin_unlock(&bedata->socket_lock);
216
217 for (i = 0; i < (1 << PVCALLS_RING_ORDER); i++)
218 gnttab_end_foreign_access(map->active.ring->ref[i], 0, 0);
219 gnttab_end_foreign_access(map->active.ref, 0, 0);
220 free_page((unsigned long)map->active.ring);
221
222 kfree(map);
aa7ba376
SS		 223}
224
cb1c7d9b
SS		 225static irqreturn_t pvcalls_front_conn_handler(int irq, void *sock_map)
226{
227 struct sock_mapping *map = sock_map;
228
229 if (map == NULL)
230 return IRQ_HANDLED;
231
232 wake_up_interruptible(&map->active.inflight_conn_req);
233
234 return IRQ_HANDLED;
235}
236
2195046b
SS		 237int pvcalls_front_socket(struct socket *sock)
238{
239 struct pvcalls_bedata *bedata;
240 struct sock_mapping *map = NULL;
241 struct xen_pvcalls_request *req;
242 int notify, req_id, ret;
243
244 /*
245 * PVCalls only supports domain AF_INET,
246 * type SOCK_STREAM and protocol 0 sockets for now.
247 *
248 * Check socket type here, AF_INET and protocol checks are done
249 * by the caller.
250 */
251 if (sock->type != SOCK_STREAM)
252 return -EOPNOTSUPP;
253
254 pvcalls_enter();
255 if (!pvcalls_front_dev) {
256 pvcalls_exit();
257 return -EACCES;
258 }
259 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
260
261 map = kzalloc(sizeof(*map), GFP_KERNEL);
262 if (map == NULL) {
263 pvcalls_exit();
264 return -ENOMEM;
265 }
266
267 spin_lock(&bedata->socket_lock);
268
269 ret = get_request(bedata, &req_id);
270 if (ret < 0) {
271 kfree(map);
272 spin_unlock(&bedata->socket_lock);
273 pvcalls_exit();
274 return ret;
275 }
276
277 /*
278 * sock->sk->sk_send_head is not used for ip sockets: reuse the
279 * field to store a pointer to the struct sock_mapping
280 * corresponding to the socket. This way, we can easily get the
281 * struct sock_mapping from the struct socket.
282 */
283 sock->sk->sk_send_head = (void *)map;
284 list_add_tail(&map->list, &bedata->socket_mappings);
285
286 req = RING_GET_REQUEST(&bedata->ring, req_id);
287 req->req_id = req_id;
288 req->cmd = PVCALLS_SOCKET;
289 req->u.socket.id = (uintptr_t) map;
290 req->u.socket.domain = AF_INET;
291 req->u.socket.type = SOCK_STREAM;
292 req->u.socket.protocol = IPPROTO_IP;
293
294 bedata->ring.req_prod_pvt++;
295 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
296 spin_unlock(&bedata->socket_lock);
297 if (notify)
298 notify_remote_via_irq(bedata->irq);
299
300 wait_event(bedata->inflight_req,
301 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
302
303 /* read req_id, then the content */
304 smp_rmb();
305 ret = bedata->rsp[req_id].ret;
306 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
307
308 pvcalls_exit();
309 return ret;
310}
311
cb1c7d9b
SS		 312static int create_active(struct sock_mapping *map, int *evtchn)
313{
314 void *bytes;
315 int ret = -ENOMEM, irq = -1, i;
316
317 *evtchn = -1;
318 init_waitqueue_head(&map->active.inflight_conn_req);
319
320 map->active.ring = (struct pvcalls_data_intf *)
321 __get_free_page(GFP_KERNEL | __GFP_ZERO);
322 if (map->active.ring == NULL)
323 goto out_error;
324 map->active.ring->ring_order = PVCALLS_RING_ORDER;
325 bytes = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
326 PVCALLS_RING_ORDER);
327 if (bytes == NULL)
328 goto out_error;
329 for (i = 0; i < (1 << PVCALLS_RING_ORDER); i++)
330 map->active.ring->ref[i] = gnttab_grant_foreign_access(
331 pvcalls_front_dev->otherend_id,
332 pfn_to_gfn(virt_to_pfn(bytes) + i), 0);
333
334 map->active.ref = gnttab_grant_foreign_access(
335 pvcalls_front_dev->otherend_id,
336 pfn_to_gfn(virt_to_pfn((void *)map->active.ring)), 0);
337
338 map->active.data.in = bytes;
339 map->active.data.out = bytes +
340 XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
341
342 ret = xenbus_alloc_evtchn(pvcalls_front_dev, evtchn);
343 if (ret)
344 goto out_error;
345 irq = bind_evtchn_to_irqhandler(*evtchn, pvcalls_front_conn_handler,
346 0, "pvcalls-frontend", map);
347 if (irq < 0) {
348 ret = irq;
349 goto out_error;
350 }
351
352 map->active.irq = irq;
353 map->active_socket = true;
354 mutex_init(&map->active.in_mutex);
355 mutex_init(&map->active.out_mutex);
356
357 return 0;
358
359out_error:
773aaadc 360 if (*evtchn >= 0)
cb1c7d9b 361 xenbus_free_evtchn(pvcalls_front_dev, *evtchn);
e7881bba
PB		 362 free_pages((unsigned long)map->active.data.in, PVCALLS_RING_ORDER);
363 free_page((unsigned long)map->active.ring);
cb1c7d9b
SS		 364 return ret;
365}
366
367int pvcalls_front_connect(struct socket *sock, struct sockaddr *addr,
368 int addr_len, int flags)
369{
370 struct pvcalls_bedata *bedata;
371 struct sock_mapping *map = NULL;
372 struct xen_pvcalls_request *req;
373 int notify, req_id, ret, evtchn;
374
375 if (addr->sa_family != AF_INET || sock->type != SOCK_STREAM)
376 return -EOPNOTSUPP;
377
378 pvcalls_enter();
379 if (!pvcalls_front_dev) {
380 pvcalls_exit();
381 return -ENOTCONN;
382 }
383
384 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
385
386 map = (struct sock_mapping *)sock->sk->sk_send_head;
387 if (!map) {
388 pvcalls_exit();
389 return -ENOTSOCK;
390 }
391
392 spin_lock(&bedata->socket_lock);
393 ret = get_request(bedata, &req_id);
394 if (ret < 0) {
395 spin_unlock(&bedata->socket_lock);
396 pvcalls_exit();
397 return ret;
398 }
399 ret = create_active(map, &evtchn);
400 if (ret < 0) {
401 spin_unlock(&bedata->socket_lock);
402 pvcalls_exit();
403 return ret;
404 }
405
406 req = RING_GET_REQUEST(&bedata->ring, req_id);
407 req->req_id = req_id;
408 req->cmd = PVCALLS_CONNECT;
409 req->u.connect.id = (uintptr_t)map;
410 req->u.connect.len = addr_len;
411 req->u.connect.flags = flags;
412 req->u.connect.ref = map->active.ref;
413 req->u.connect.evtchn = evtchn;
414 memcpy(req->u.connect.addr, addr, sizeof(*addr));
415
416 map->sock = sock;
417
418 bedata->ring.req_prod_pvt++;
419 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
420 spin_unlock(&bedata->socket_lock);
421
422 if (notify)
423 notify_remote_via_irq(bedata->irq);
424
425 wait_event(bedata->inflight_req,
426 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
427
428 /* read req_id, then the content */
429 smp_rmb();
430 ret = bedata->rsp[req_id].ret;
431 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
432 pvcalls_exit();
433 return ret;
434}
435
45ddce21
SS		 436static int __write_ring(struct pvcalls_data_intf *intf,
437 struct pvcalls_data *data,
438 struct iov_iter *msg_iter,
439 int len)
440{
441 RING_IDX cons, prod, size, masked_prod, masked_cons;
442 RING_IDX array_size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
443 int32_t error;
444
445 error = intf->out_error;
446 if (error < 0)
447 return error;
448 cons = intf->out_cons;
449 prod = intf->out_prod;
450 /* read indexes before continuing */
451 virt_mb();
452
453 size = pvcalls_queued(prod, cons, array_size);
454 if (size >= array_size)
455 return -EINVAL;
456 if (len > array_size - size)
457 len = array_size - size;
458
459 masked_prod = pvcalls_mask(prod, array_size);
460 masked_cons = pvcalls_mask(cons, array_size);
461
462 if (masked_prod < masked_cons) {
463 len = copy_from_iter(data->out + masked_prod, len, msg_iter);
464 } else {
465 if (len > array_size - masked_prod) {
466 int ret = copy_from_iter(data->out + masked_prod,
467 array_size - masked_prod, msg_iter);
468 if (ret != array_size - masked_prod) {
469 len = ret;
470 goto out;
471 }
472 len = ret + copy_from_iter(data->out, len - ret, msg_iter);
473 } else {
474 len = copy_from_iter(data->out + masked_prod, len, msg_iter);
475 }
476 }
477out:
478 /* write to ring before updating pointer */
479 virt_wmb();
480 intf->out_prod += len;
481
482 return len;
483}
484
485int pvcalls_front_sendmsg(struct socket *sock, struct msghdr *msg,
486 size_t len)
487{
488 struct pvcalls_bedata *bedata;
489 struct sock_mapping *map;
490 int sent, tot_sent = 0;
491 int count = 0, flags;
492
493 flags = msg->msg_flags;
494 if (flags & (MSG_CONFIRM|MSG_DONTROUTE|MSG_EOR|MSG_OOB))
495 return -EOPNOTSUPP;
496
497 pvcalls_enter();
498 if (!pvcalls_front_dev) {
499 pvcalls_exit();
500 return -ENOTCONN;
501 }
502 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
503
504 map = (struct sock_mapping *) sock->sk->sk_send_head;
505 if (!map) {
506 pvcalls_exit();
507 return -ENOTSOCK;
508 }
509
510 mutex_lock(&map->active.out_mutex);
511 if ((flags & MSG_DONTWAIT) && !pvcalls_front_write_todo(map)) {
512 mutex_unlock(&map->active.out_mutex);
513 pvcalls_exit();
514 return -EAGAIN;
515 }
516 if (len > INT_MAX)
517 len = INT_MAX;
518
519again:
520 count++;
521 sent = __write_ring(map->active.ring,
522 &map->active.data, &msg->msg_iter,
523 len);
524 if (sent > 0) {
525 len -= sent;
526 tot_sent += sent;
527 notify_remote_via_irq(map->active.irq);
528 }
529 if (sent >= 0 && len > 0 && count < PVCALLS_FRONT_MAX_SPIN)
530 goto again;
531 if (sent < 0)
532 tot_sent = sent;
533
534 mutex_unlock(&map->active.out_mutex);
535 pvcalls_exit();
536 return tot_sent;
537}
538
ae0d0405
SS		 539static int __read_ring(struct pvcalls_data_intf *intf,
540 struct pvcalls_data *data,
541 struct iov_iter *msg_iter,
542 size_t len, int flags)
543{
544 RING_IDX cons, prod, size, masked_prod, masked_cons;
545 RING_IDX array_size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
546 int32_t error;
547
548 cons = intf->in_cons;
549 prod = intf->in_prod;
550 error = intf->in_error;
551 /* get pointers before reading from the ring */
552 virt_rmb();
ae0d0405
SS		 553
554 size = pvcalls_queued(prod, cons, array_size);
555 masked_prod = pvcalls_mask(prod, array_size);
556 masked_cons = pvcalls_mask(cons, array_size);
557
558 if (size == 0)
ee88f289 559 return error ?: size;
ae0d0405
SS		 560
561 if (len > size)
562 len = size;
563
564 if (masked_prod > masked_cons) {
565 len = copy_to_iter(data->in + masked_cons, len, msg_iter);
566 } else {
567 if (len > (array_size - masked_cons)) {
568 int ret = copy_to_iter(data->in + masked_cons,
569 array_size - masked_cons, msg_iter);
570 if (ret != array_size - masked_cons) {
571 len = ret;
572 goto out;
573 }
574 len = ret + copy_to_iter(data->in, len - ret, msg_iter);
575 } else {
576 len = copy_to_iter(data->in + masked_cons, len, msg_iter);
577 }
578 }
579out:
580 /* read data from the ring before increasing the index */
581 virt_mb();
582 if (!(flags & MSG_PEEK))
583 intf->in_cons += len;
584
585 return len;
586}
587
588int pvcalls_front_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
589 int flags)
590{
591 struct pvcalls_bedata *bedata;
592 int ret;
593 struct sock_mapping *map;
594
595 if (flags & (MSG_CMSG_CLOEXEC|MSG_ERRQUEUE|MSG_OOB|MSG_TRUNC))
596 return -EOPNOTSUPP;
597
598 pvcalls_enter();
599 if (!pvcalls_front_dev) {
600 pvcalls_exit();
601 return -ENOTCONN;
602 }
603 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
604
605 map = (struct sock_mapping *) sock->sk->sk_send_head;
606 if (!map) {
607 pvcalls_exit();
608 return -ENOTSOCK;
609 }
610
611 mutex_lock(&map->active.in_mutex);
612 if (len > XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER))
613 len = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
614
615 while (!(flags & MSG_DONTWAIT) && !pvcalls_front_read_todo(map)) {
616 wait_event_interruptible(map->active.inflight_conn_req,
617 pvcalls_front_read_todo(map));
618 }
619 ret = __read_ring(map->active.ring, &map->active.data,
620 &msg->msg_iter, len, flags);
621
622 if (ret > 0)
623 notify_remote_via_irq(map->active.irq);
624 if (ret == 0)
625 ret = (flags & MSG_DONTWAIT) ? -EAGAIN : 0;
626 if (ret == -ENOTCONN)
627 ret = 0;
628
629 mutex_unlock(&map->active.in_mutex);
630 pvcalls_exit();
631 return ret;
632}
633
67ea9893
SS		 634int pvcalls_front_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
635{
636 struct pvcalls_bedata *bedata;
637 struct sock_mapping *map = NULL;
638 struct xen_pvcalls_request *req;
639 int notify, req_id, ret;
640
641 if (addr->sa_family != AF_INET || sock->type != SOCK_STREAM)
642 return -EOPNOTSUPP;
643
644 pvcalls_enter();
645 if (!pvcalls_front_dev) {
646 pvcalls_exit();
647 return -ENOTCONN;
648 }
649 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
650
651 map = (struct sock_mapping *) sock->sk->sk_send_head;
652 if (map == NULL) {
653 pvcalls_exit();
654 return -ENOTSOCK;
655 }
656
657 spin_lock(&bedata->socket_lock);
658 ret = get_request(bedata, &req_id);
659 if (ret < 0) {
660 spin_unlock(&bedata->socket_lock);
661 pvcalls_exit();
662 return ret;
663 }
664 req = RING_GET_REQUEST(&bedata->ring, req_id);
665 req->req_id = req_id;
666 map->sock = sock;
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;
671
9774c6cc
SS		 672 init_waitqueue_head(&map->passive.inflight_accept_req);
673
67ea9893
SS		 674 map->active_socket = false;
675
676 bedata->ring.req_prod_pvt++;
677 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
678 spin_unlock(&bedata->socket_lock);
679 if (notify)
680 notify_remote_via_irq(bedata->irq);
681
682 wait_event(bedata->inflight_req,
683 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
684
685 /* read req_id, then the content */
686 smp_rmb();
687 ret = bedata->rsp[req_id].ret;
688 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
689
690 map->passive.status = PVCALLS_STATUS_BIND;
691 pvcalls_exit();
692 return 0;
693}
694
1853f11d
SS		 695int pvcalls_front_listen(struct socket *sock, int backlog)
696{
697 struct pvcalls_bedata *bedata;
698 struct sock_mapping *map;
699 struct xen_pvcalls_request *req;
700 int notify, req_id, ret;
701
702 pvcalls_enter();
703 if (!pvcalls_front_dev) {
704 pvcalls_exit();
705 return -ENOTCONN;
706 }
707 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
708
709 map = (struct sock_mapping *) sock->sk->sk_send_head;
710 if (!map) {
711 pvcalls_exit();
712 return -ENOTSOCK;
713 }
714
715 if (map->passive.status != PVCALLS_STATUS_BIND) {
716 pvcalls_exit();
717 return -EOPNOTSUPP;
718 }
719
720 spin_lock(&bedata->socket_lock);
721 ret = get_request(bedata, &req_id);
722 if (ret < 0) {
723 spin_unlock(&bedata->socket_lock);
724 pvcalls_exit();
725 return ret;
726 }
727 req = RING_GET_REQUEST(&bedata->ring, req_id);
728 req->req_id = req_id;
729 req->cmd = PVCALLS_LISTEN;
730 req->u.listen.id = (uintptr_t) map;
731 req->u.listen.backlog = backlog;
732
733 bedata->ring.req_prod_pvt++;
734 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
735 spin_unlock(&bedata->socket_lock);
736 if (notify)
737 notify_remote_via_irq(bedata->irq);
738
739 wait_event(bedata->inflight_req,
740 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
741
742 /* read req_id, then the content */
743 smp_rmb();
744 ret = bedata->rsp[req_id].ret;
745 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
746
747 map->passive.status = PVCALLS_STATUS_LISTEN;
748 pvcalls_exit();
749 return ret;
750}
751
9774c6cc
SS		 752int pvcalls_front_accept(struct socket *sock, struct socket *newsock, int flags)
753{
754 struct pvcalls_bedata *bedata;
755 struct sock_mapping *map;
756 struct sock_mapping *map2 = NULL;
757 struct xen_pvcalls_request *req;
758 int notify, req_id, ret, evtchn, nonblock;
759
760 pvcalls_enter();
761 if (!pvcalls_front_dev) {
762 pvcalls_exit();
763 return -ENOTCONN;
764 }
765 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
766
767 map = (struct sock_mapping *) sock->sk->sk_send_head;
768 if (!map) {
769 pvcalls_exit();
770 return -ENOTSOCK;
771 }
772
773 if (map->passive.status != PVCALLS_STATUS_LISTEN) {
774 pvcalls_exit();
775 return -EINVAL;
776 }
777
778 nonblock = flags & SOCK_NONBLOCK;
779 /*
780 * Backend only supports 1 inflight accept request, will return
781 * errors for the others
782 */
783 if (test_and_set_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
784 (void *)&map->passive.flags)) {
785 req_id = READ_ONCE(map->passive.inflight_req_id);
786 if (req_id != PVCALLS_INVALID_ID &&
787 READ_ONCE(bedata->rsp[req_id].req_id) == req_id) {
788 map2 = map->passive.accept_map;
789 goto received;
790 }
791 if (nonblock) {
792 pvcalls_exit();
793 return -EAGAIN;
794 }
795 if (wait_event_interruptible(map->passive.inflight_accept_req,
796 !test_and_set_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
797 (void *)&map->passive.flags))) {
798 pvcalls_exit();
799 return -EINTR;
800 }
801 }
802
803 spin_lock(&bedata->socket_lock);
804 ret = get_request(bedata, &req_id);
805 if (ret < 0) {
806 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
807 (void *)&map->passive.flags);
808 spin_unlock(&bedata->socket_lock);
809 pvcalls_exit();
810 return ret;
811 }
4aac2caf 812 map2 = kzalloc(sizeof(*map2), GFP_ATOMIC);
9774c6cc
SS		 813 if (map2 == NULL) {
814 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
815 (void *)&map->passive.flags);
816 spin_unlock(&bedata->socket_lock);
817 pvcalls_exit();
818 return -ENOMEM;
819 }
820 ret = create_active(map2, &evtchn);
821 if (ret < 0) {
822 kfree(map2);
823 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
824 (void *)&map->passive.flags);
825 spin_unlock(&bedata->socket_lock);
826 pvcalls_exit();
827 return ret;
828 }
829 list_add_tail(&map2->list, &bedata->socket_mappings);
830
831 req = RING_GET_REQUEST(&bedata->ring, req_id);
832 req->req_id = req_id;
833 req->cmd = PVCALLS_ACCEPT;
834 req->u.accept.id = (uintptr_t) map;
835 req->u.accept.ref = map2->active.ref;
836 req->u.accept.id_new = (uintptr_t) map2;
837 req->u.accept.evtchn = evtchn;
838 map->passive.accept_map = map2;
839
840 bedata->ring.req_prod_pvt++;
841 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
842 spin_unlock(&bedata->socket_lock);
843 if (notify)
844 notify_remote_via_irq(bedata->irq);
845 /* We could check if we have received a response before returning. */
846 if (nonblock) {
847 WRITE_ONCE(map->passive.inflight_req_id, req_id);
848 pvcalls_exit();
849 return -EAGAIN;
850 }
851
852 if (wait_event_interruptible(bedata->inflight_req,
853 READ_ONCE(bedata->rsp[req_id].req_id) == req_id)) {
854 pvcalls_exit();
855 return -EINTR;
856 }
857 /* read req_id, then the content */
858 smp_rmb();
859
860received:
861 map2->sock = newsock;
63be5d01 862 newsock->sk = sk_alloc(sock_net(sock->sk), PF_INET, GFP_KERNEL, &pvcalls_proto, false);
9774c6cc
SS		 863 if (!newsock->sk) {
864 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
865 map->passive.inflight_req_id = PVCALLS_INVALID_ID;
866 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
867 (void *)&map->passive.flags);
868 pvcalls_front_free_map(bedata, map2);
869 pvcalls_exit();
870 return -ENOMEM;
871 }
872 newsock->sk->sk_send_head = (void *)map2;
873
874 ret = bedata->rsp[req_id].ret;
875 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
876 map->passive.inflight_req_id = PVCALLS_INVALID_ID;
877
878 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT, (void *)&map->passive.flags);
879 wake_up(&map->passive.inflight_accept_req);
880
881 pvcalls_exit();
882 return ret;
883}
884
5842c835
SS		 885static unsigned int pvcalls_front_poll_passive(struct file *file,
886 struct pvcalls_bedata *bedata,
887 struct sock_mapping *map,
888 poll_table *wait)
889{
890 int notify, req_id, ret;
891 struct xen_pvcalls_request *req;
892
893 if (test_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
894 (void *)&map->passive.flags)) {
895 uint32_t req_id = READ_ONCE(map->passive.inflight_req_id);
896
897 if (req_id != PVCALLS_INVALID_ID &&
898 READ_ONCE(bedata->rsp[req_id].req_id) == req_id)
899 return POLLIN | POLLRDNORM;
900
901 poll_wait(file, &map->passive.inflight_accept_req, wait);
902 return 0;
903 }
904
905 if (test_and_clear_bit(PVCALLS_FLAG_POLL_RET,
906 (void *)&map->passive.flags))
907 return POLLIN | POLLRDNORM;
908
909 /*
910 * First check RET, then INFLIGHT. No barriers necessary to
911 * ensure execution ordering because of the conditional
912 * instructions creating control dependencies.
913 */
914
915 if (test_and_set_bit(PVCALLS_FLAG_POLL_INFLIGHT,
916 (void *)&map->passive.flags)) {
917 poll_wait(file, &bedata->inflight_req, wait);
918 return 0;
919 }
920
921 spin_lock(&bedata->socket_lock);
922 ret = get_request(bedata, &req_id);
923 if (ret < 0) {
924 spin_unlock(&bedata->socket_lock);
925 return ret;
926 }
927 req = RING_GET_REQUEST(&bedata->ring, req_id);
928 req->req_id = req_id;
929 req->cmd = PVCALLS_POLL;
930 req->u.poll.id = (uintptr_t) map;
931
932 bedata->ring.req_prod_pvt++;
933 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
934 spin_unlock(&bedata->socket_lock);
935 if (notify)
936 notify_remote_via_irq(bedata->irq);
937
938 poll_wait(file, &bedata->inflight_req, wait);
939 return 0;
940}
941
942static unsigned int pvcalls_front_poll_active(struct file *file,
943 struct pvcalls_bedata *bedata,
944 struct sock_mapping *map,
945 poll_table *wait)
946{
947 unsigned int mask = 0;
948 int32_t in_error, out_error;
949 struct pvcalls_data_intf *intf = map->active.ring;
950
951 out_error = intf->out_error;
952 in_error = intf->in_error;
953
954 poll_wait(file, &map->active.inflight_conn_req, wait);
955 if (pvcalls_front_write_todo(map))
956 mask |= POLLOUT | POLLWRNORM;
957 if (pvcalls_front_read_todo(map))
958 mask |= POLLIN | POLLRDNORM;
959 if (in_error != 0 || out_error != 0)
960 mask |= POLLERR;
961
962 return mask;
963}
964
965unsigned int pvcalls_front_poll(struct file *file, struct socket *sock,
966 poll_table *wait)
967{
968 struct pvcalls_bedata *bedata;
969 struct sock_mapping *map;
970 int ret;
971
972 pvcalls_enter();
973 if (!pvcalls_front_dev) {
974 pvcalls_exit();
975 return POLLNVAL;
976 }
977 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
978
979 map = (struct sock_mapping *) sock->sk->sk_send_head;
980 if (!map) {
981 pvcalls_exit();
982 return POLLNVAL;
983 }
984 if (map->active_socket)
985 ret = pvcalls_front_poll_active(file, bedata, map, wait);
986 else
987 ret = pvcalls_front_poll_passive(file, bedata, map, wait);
988 pvcalls_exit();
989 return ret;
990}
991
235a71c5
SS		 992int pvcalls_front_release(struct socket *sock)
993{
994 struct pvcalls_bedata *bedata;
995 struct sock_mapping *map;
996 int req_id, notify, ret;
997 struct xen_pvcalls_request *req;
998
999 if (sock->sk == NULL)
1000 return 0;
1001
1002 pvcalls_enter();
1003 if (!pvcalls_front_dev) {
1004 pvcalls_exit();
1005 return -EIO;
1006 }
1007
1008 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
1009
1010 map = (struct sock_mapping *) sock->sk->sk_send_head;
1011 if (map == NULL) {
1012 pvcalls_exit();
1013 return 0;
1014 }
1015
1016 spin_lock(&bedata->socket_lock);
1017 ret = get_request(bedata, &req_id);
1018 if (ret < 0) {
1019 spin_unlock(&bedata->socket_lock);
1020 pvcalls_exit();
1021 return ret;
1022 }
1023 sock->sk->sk_send_head = NULL;
1024
1025 req = RING_GET_REQUEST(&bedata->ring, req_id);
1026 req->req_id = req_id;
1027 req->cmd = PVCALLS_RELEASE;
1028 req->u.release.id = (uintptr_t)map;
1029
1030 bedata->ring.req_prod_pvt++;
1031 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
1032 spin_unlock(&bedata->socket_lock);
1033 if (notify)
1034 notify_remote_via_irq(bedata->irq);
1035
1036 wait_event(bedata->inflight_req,
1037 READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
1038
1039 if (map->active_socket) {
1040 /*
1041 * Set in_error and wake up inflight_conn_req to force
1042 * recvmsg waiters to exit.
1043 */
1044 map->active.ring->in_error = -EBADF;
1045 wake_up_interruptible(&map->active.inflight_conn_req);
1046
1047 /*
646d944c
SS		 1048 * We need to make sure that sendmsg/recvmsg on this socket have
1049 * not started before we've cleared sk_send_head here. The
1050 * easiest (though not optimal) way to guarantee this is to see
1051 * that no pvcall (other than us) is in progress.
235a71c5 1052 */
646d944c 1053 while (atomic_read(&pvcalls_refcount) > 1)
235a71c5
SS		 1054 cpu_relax();
1055
1056 pvcalls_front_free_map(bedata, map);
1057 } else {
1058 spin_lock(&bedata->socket_lock);
1059 list_del(&map->list);
1060 spin_unlock(&bedata->socket_lock);
c2c273f9
SS		 1061 if (READ_ONCE(map->passive.inflight_req_id) != PVCALLS_INVALID_ID &&
1062 READ_ONCE(map->passive.inflight_req_id) != 0) {
235a71c5
SS		 1063 pvcalls_front_free_map(bedata,
1064 map->passive.accept_map);
1065 }
1066 kfree(map);
1067 }
1068 WRITE_ONCE(bedata->rsp[req_id].req_id, PVCALLS_INVALID_ID);
1069
1070 pvcalls_exit();
1071 return 0;
1072}
1073
416efba0
SS		 1074static const struct xenbus_device_id pvcalls_front_ids[] = {
1075 { "pvcalls" },
1076 { "" }
1077};
1078
1079static int pvcalls_front_remove(struct xenbus_device *dev)
1080{
aa7ba376
SS		 1081 struct pvcalls_bedata *bedata;
1082 struct sock_mapping *map = NULL, *n;
1083
1084 bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
1085 dev_set_drvdata(&dev->dev, NULL);
1086 pvcalls_front_dev = NULL;
1087 if (bedata->irq >= 0)
1088 unbind_from_irqhandler(bedata->irq, dev);
1089
cb1c7d9b
SS		 1090 list_for_each_entry_safe(map, n, &bedata->socket_mappings, list) {
1091 map->sock->sk->sk_send_head = NULL;
1092 if (map->active_socket) {
1093 map->active.ring->in_error = -EBADF;
1094 wake_up_interruptible(&map->active.inflight_conn_req);
1095 }
1096 }
1097
aa7ba376
SS		 1098 smp_mb();
1099 while (atomic_read(&pvcalls_refcount) > 0)
1100 cpu_relax();
1101 list_for_each_entry_safe(map, n, &bedata->socket_mappings, list) {
1102 if (map->active_socket) {
1103 /* No need to lock, refcount is 0 */
1104 pvcalls_front_free_map(bedata, map);
1105 } else {
1106 list_del(&map->list);
1107 kfree(map);
1108 }
1109 }
1ab134ca 1110 if (bedata->ref != -1)
aa7ba376
SS		 1111 gnttab_end_foreign_access(bedata->ref, 0, 0);
1112 kfree(bedata->ring.sring);
1113 kfree(bedata);
1114 xenbus_switch_state(dev, XenbusStateClosed);
416efba0
SS		 1115 return 0;
1116}
1117
1118static int pvcalls_front_probe(struct xenbus_device *dev,
1119 const struct xenbus_device_id *id)
1120{
21968190
SS		 1121 int ret = -ENOMEM, evtchn, i;
1122 unsigned int max_page_order, function_calls, len;
1123 char *versions;
1124 grant_ref_t gref_head = 0;
1125 struct xenbus_transaction xbt;
1126 struct pvcalls_bedata *bedata = NULL;
1127 struct xen_pvcalls_sring *sring;
1128
1129 if (pvcalls_front_dev != NULL) {
1130 dev_err(&dev->dev, "only one PV Calls connection supported\n");
1131 return -EINVAL;
1132 }
1133
1134 versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
8c71fa88
DC		 1135 if (IS_ERR(versions))
1136 return PTR_ERR(versions);
21968190
SS		 1137 if (!len)
1138 return -EINVAL;
1139 if (strcmp(versions, "1")) {
1140 kfree(versions);
1141 return -EINVAL;
1142 }
1143 kfree(versions);
1144 max_page_order = xenbus_read_unsigned(dev->otherend,
1145 "max-page-order", 0);
1146 if (max_page_order < PVCALLS_RING_ORDER)
1147 return -ENODEV;
1148 function_calls = xenbus_read_unsigned(dev->otherend,
1149 "function-calls", 0);
1150 /* See XENBUS_FUNCTIONS_CALLS in pvcalls.h */
1151 if (function_calls != 1)
1152 return -ENODEV;
1153 pr_info("%s max-page-order is %u\n", __func__, max_page_order);
1154
1155 bedata = kzalloc(sizeof(struct pvcalls_bedata), GFP_KERNEL);
1156 if (!bedata)
1157 return -ENOMEM;
1158
1159 dev_set_drvdata(&dev->dev, bedata);
1160 pvcalls_front_dev = dev;
1161 init_waitqueue_head(&bedata->inflight_req);
1162 INIT_LIST_HEAD(&bedata->socket_mappings);
1163 spin_lock_init(&bedata->socket_lock);
1164 bedata->irq = -1;
1165 bedata->ref = -1;
1166
1167 for (i = 0; i < PVCALLS_NR_RSP_PER_RING; i++)
1168 bedata->rsp[i].req_id = PVCALLS_INVALID_ID;
1169
1170 sring = (struct xen_pvcalls_sring *) __get_free_page(GFP_KERNEL |
1171 __GFP_ZERO);
1172 if (!sring)
1173 goto error;
1174 SHARED_RING_INIT(sring);
1175 FRONT_RING_INIT(&bedata->ring, sring, XEN_PAGE_SIZE);
1176
1177 ret = xenbus_alloc_evtchn(dev, &evtchn);
1178 if (ret)
1179 goto error;
1180
1181 bedata->irq = bind_evtchn_to_irqhandler(evtchn,
1182 pvcalls_front_event_handler,
1183 0, "pvcalls-frontend", dev);
1184 if (bedata->irq < 0) {
1185 ret = bedata->irq;
1186 goto error;
1187 }
1188
1189 ret = gnttab_alloc_grant_references(1, &gref_head);
1190 if (ret < 0)
1191 goto error;
95110ac8
CIK		 1192 ret = gnttab_claim_grant_reference(&gref_head);
1193 if (ret < 0)
21968190 1194 goto error;
95110ac8 1195 bedata->ref = ret;
21968190
SS		 1196 gnttab_grant_foreign_access_ref(bedata->ref, dev->otherend_id,
1197 virt_to_gfn((void *)sring), 0);
1198
1199 again:
1200 ret = xenbus_transaction_start(&xbt);
1201 if (ret) {
1202 xenbus_dev_fatal(dev, ret, "starting transaction");
1203 goto error;
1204 }
1205 ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
1206 if (ret)
1207 goto error_xenbus;
1208 ret = xenbus_printf(xbt, dev->nodename, "ring-ref", "%d", bedata->ref);
1209 if (ret)
1210 goto error_xenbus;
1211 ret = xenbus_printf(xbt, dev->nodename, "port", "%u",
1212 evtchn);
1213 if (ret)
1214 goto error_xenbus;
1215 ret = xenbus_transaction_end(xbt, 0);
1216 if (ret) {
1217 if (ret == -EAGAIN)
1218 goto again;
1219 xenbus_dev_fatal(dev, ret, "completing transaction");
1220 goto error;
1221 }
1222 xenbus_switch_state(dev, XenbusStateInitialised);
1223
416efba0 1224 return 0;
21968190
SS		 1225
1226 error_xenbus:
1227 xenbus_transaction_end(xbt, 1);
1228 xenbus_dev_fatal(dev, ret, "writing xenstore");
1229 error:
1230 pvcalls_front_remove(dev);
1231 return ret;
416efba0
SS		 1232}
1233
1234static void pvcalls_front_changed(struct xenbus_device *dev,
1235 enum xenbus_state backend_state)
1236{
21968190
SS		 1237 switch (backend_state) {
1238 case XenbusStateReconfiguring:
1239 case XenbusStateReconfigured:
1240 case XenbusStateInitialising:
1241 case XenbusStateInitialised:
1242 case XenbusStateUnknown:
1243 break;
1244
1245 case XenbusStateInitWait:
1246 break;
1247
1248 case XenbusStateConnected:
1249 xenbus_switch_state(dev, XenbusStateConnected);
1250 break;
1251
1252 case XenbusStateClosed:
1253 if (dev->state == XenbusStateClosed)
1254 break;
3d8765d4
GS		 1255 /* Missed the backend's CLOSING state */
1256 /* fall through */
21968190
SS		 1257 case XenbusStateClosing:
1258 xenbus_frontend_closed(dev);
1259 break;
1260 }
416efba0
SS		 1261}
1262
1263static struct xenbus_driver pvcalls_front_driver = {
1264 .ids = pvcalls_front_ids,
1265 .probe = pvcalls_front_probe,
1266 .remove = pvcalls_front_remove,
1267 .otherend_changed = pvcalls_front_changed,
1268};
1269
1270static int __init pvcalls_frontend_init(void)
1271{
1272 if (!xen_domain())
1273 return -ENODEV;
1274
1275 pr_info("Initialising Xen pvcalls frontend driver\n");
1276
1277 return xenbus_register_frontend(&pvcalls_front_driver);
1278}
1279
1280module_init(pvcalls_frontend_init);
24e7f84d
BO		 1281
1282MODULE_DESCRIPTION("Xen PV Calls frontend driver");
1283MODULE_AUTHOR("Stefano Stabellini <sstabellini@kernel.org>");
1284MODULE_LICENSE("GPL");
ubuntu-bionic-kernel mirror