Merge branch 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
[linux-2.6-microblaze.git] / drivers / xen / pvcalls-front.c
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
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
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
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
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
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
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
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
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
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
335 static void free_active_ring(struct sock_mapping *map)
336 {
337         if (!map->active.ring)
338                 return;
339
340         free_pages((unsigned long)map->active.data.in,
341                         map->active.ring->ring_order);
342         free_page((unsigned long)map->active.ring);
343 }
344
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 = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
356                                         PVCALLS_RING_ORDER);
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
371 static int create_active(struct sock_mapping *map, int *evtchn)
372 {
373         void *bytes;
374         int ret = -ENOMEM, irq = -1, i;
375
376         *evtchn = -1;
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
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, evtchn;
419
420         if (addr->sa_family != AF_INET || sock->type != SOCK_STREAM)
421                 return -EOPNOTSUPP;
422
423         map = pvcalls_enter_sock(sock);
424         if (IS_ERR(map))
425                 return PTR_ERR(map);
426
427         bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
428         ret = alloc_active_ring(map);
429         if (ret < 0) {
430                 pvcalls_exit_sock(sock);
431                 return ret;
432         }
433
434         spin_lock(&bedata->socket_lock);
435         ret = get_request(bedata, &req_id);
436         if (ret < 0) {
437                 spin_unlock(&bedata->socket_lock);
438                 free_active_ring(map);
439                 pvcalls_exit_sock(sock);
440                 return ret;
441         }
442         ret = create_active(map, &evtchn);
443         if (ret < 0) {
444                 spin_unlock(&bedata->socket_lock);
445                 free_active_ring(map);
446                 pvcalls_exit_sock(sock);
447                 return ret;
448         }
449
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));
459
460         map->sock = sock;
461
462         bedata->ring.req_prod_pvt++;
463         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
464         spin_unlock(&bedata->socket_lock);
465
466         if (notify)
467                 notify_remote_via_irq(bedata->irq);
468
469         wait_event(bedata->inflight_req,
470                    READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
471
472         /* read req_id, then the content */
473         smp_rmb();
474         ret = bedata->rsp[req_id].ret;
475         bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
476         pvcalls_exit_sock(sock);
477         return ret;
478 }
479
480 static int __write_ring(struct pvcalls_data_intf *intf,
481                         struct pvcalls_data *data,
482                         struct iov_iter *msg_iter,
483                         int len)
484 {
485         RING_IDX cons, prod, size, masked_prod, masked_cons;
486         RING_IDX array_size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
487         int32_t error;
488
489         error = intf->out_error;
490         if (error < 0)
491                 return error;
492         cons = intf->out_cons;
493         prod = intf->out_prod;
494         /* read indexes before continuing */
495         virt_mb();
496
497         size = pvcalls_queued(prod, cons, array_size);
498         if (size > array_size)
499                 return -EINVAL;
500         if (size == array_size)
501                 return 0;
502         if (len > array_size - size)
503                 len = array_size - size;
504
505         masked_prod = pvcalls_mask(prod, array_size);
506         masked_cons = pvcalls_mask(cons, array_size);
507
508         if (masked_prod < masked_cons) {
509                 len = copy_from_iter(data->out + masked_prod, len, msg_iter);
510         } else {
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) {
515                                 len = ret;
516                                 goto out;
517                         }
518                         len = ret + copy_from_iter(data->out, len - ret, msg_iter);
519                 } else {
520                         len = copy_from_iter(data->out + masked_prod, len, msg_iter);
521                 }
522         }
523 out:
524         /* write to ring before updating pointer */
525         virt_wmb();
526         intf->out_prod += len;
527
528         return len;
529 }
530
531 int pvcalls_front_sendmsg(struct socket *sock, struct msghdr *msg,
532                           size_t len)
533 {
534         struct sock_mapping *map;
535         int sent, tot_sent = 0;
536         int count = 0, flags;
537
538         flags = msg->msg_flags;
539         if (flags & (MSG_CONFIRM|MSG_DONTROUTE|MSG_EOR|MSG_OOB))
540                 return -EOPNOTSUPP;
541
542         map = pvcalls_enter_sock(sock);
543         if (IS_ERR(map))
544                 return PTR_ERR(map);
545
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);
550                 return -EAGAIN;
551         }
552         if (len > INT_MAX)
553                 len = INT_MAX;
554
555 again:
556         count++;
557         sent = __write_ring(map->active.ring,
558                             &map->active.data, &msg->msg_iter,
559                             len);
560         if (sent > 0) {
561                 len -= sent;
562                 tot_sent += sent;
563                 notify_remote_via_irq(map->active.irq);
564         }
565         if (sent >= 0 && len > 0 && count < PVCALLS_FRONT_MAX_SPIN)
566                 goto again;
567         if (sent < 0)
568                 tot_sent = sent;
569
570         mutex_unlock(&map->active.out_mutex);
571         pvcalls_exit_sock(sock);
572         return tot_sent;
573 }
574
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)
579 {
580         RING_IDX cons, prod, size, masked_prod, masked_cons;
581         RING_IDX array_size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
582         int32_t error;
583
584         cons = intf->in_cons;
585         prod = intf->in_prod;
586         error = intf->in_error;
587         /* get pointers before reading from the ring */
588         virt_rmb();
589
590         size = pvcalls_queued(prod, cons, array_size);
591         masked_prod = pvcalls_mask(prod, array_size);
592         masked_cons = pvcalls_mask(cons, array_size);
593
594         if (size == 0)
595                 return error ?: size;
596
597         if (len > size)
598                 len = size;
599
600         if (masked_prod > masked_cons) {
601                 len = copy_to_iter(data->in + masked_cons, len, msg_iter);
602         } else {
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) {
607                                 len = ret;
608                                 goto out;
609                         }
610                         len = ret + copy_to_iter(data->in, len - ret, msg_iter);
611                 } else {
612                         len = copy_to_iter(data->in + masked_cons, len, msg_iter);
613                 }
614         }
615 out:
616         /* read data from the ring before increasing the index */
617         virt_mb();
618         if (!(flags & MSG_PEEK))
619                 intf->in_cons += len;
620
621         return len;
622 }
623
624 int pvcalls_front_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
625                      int flags)
626 {
627         int ret;
628         struct sock_mapping *map;
629
630         if (flags & (MSG_CMSG_CLOEXEC|MSG_ERRQUEUE|MSG_OOB|MSG_TRUNC))
631                 return -EOPNOTSUPP;
632
633         map = pvcalls_enter_sock(sock);
634         if (IS_ERR(map))
635                 return PTR_ERR(map);
636
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);
640
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));
644         }
645         ret = __read_ring(map->active.ring, &map->active.data,
646                           &msg->msg_iter, len, flags);
647
648         if (ret > 0)
649                 notify_remote_via_irq(map->active.irq);
650         if (ret == 0)
651                 ret = (flags & MSG_DONTWAIT) ? -EAGAIN : 0;
652         if (ret == -ENOTCONN)
653                 ret = 0;
654
655         mutex_unlock(&map->active.in_mutex);
656         pvcalls_exit_sock(sock);
657         return ret;
658 }
659
660 int pvcalls_front_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
661 {
662         struct pvcalls_bedata *bedata;
663         struct sock_mapping *map = NULL;
664         struct xen_pvcalls_request *req;
665         int notify, req_id, ret;
666
667         if (addr->sa_family != AF_INET || sock->type != SOCK_STREAM)
668                 return -EOPNOTSUPP;
669
670         map = pvcalls_enter_sock(sock);
671         if (IS_ERR(map))
672                 return PTR_ERR(map);
673         bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
674
675         spin_lock(&bedata->socket_lock);
676         ret = get_request(bedata, &req_id);
677         if (ret < 0) {
678                 spin_unlock(&bedata->socket_lock);
679                 pvcalls_exit_sock(sock);
680                 return ret;
681         }
682         req = RING_GET_REQUEST(&bedata->ring, req_id);
683         req->req_id = req_id;
684         map->sock = sock;
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;
689
690         init_waitqueue_head(&map->passive.inflight_accept_req);
691
692         map->active_socket = false;
693
694         bedata->ring.req_prod_pvt++;
695         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
696         spin_unlock(&bedata->socket_lock);
697         if (notify)
698                 notify_remote_via_irq(bedata->irq);
699
700         wait_event(bedata->inflight_req,
701                    READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
702
703         /* read req_id, then the content */
704         smp_rmb();
705         ret = bedata->rsp[req_id].ret;
706         bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
707
708         map->passive.status = PVCALLS_STATUS_BIND;
709         pvcalls_exit_sock(sock);
710         return 0;
711 }
712
713 int pvcalls_front_listen(struct socket *sock, int backlog)
714 {
715         struct pvcalls_bedata *bedata;
716         struct sock_mapping *map;
717         struct xen_pvcalls_request *req;
718         int notify, req_id, ret;
719
720         map = pvcalls_enter_sock(sock);
721         if (IS_ERR(map))
722                 return PTR_ERR(map);
723         bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
724
725         if (map->passive.status != PVCALLS_STATUS_BIND) {
726                 pvcalls_exit_sock(sock);
727                 return -EOPNOTSUPP;
728         }
729
730         spin_lock(&bedata->socket_lock);
731         ret = get_request(bedata, &req_id);
732         if (ret < 0) {
733                 spin_unlock(&bedata->socket_lock);
734                 pvcalls_exit_sock(sock);
735                 return ret;
736         }
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;
742
743         bedata->ring.req_prod_pvt++;
744         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
745         spin_unlock(&bedata->socket_lock);
746         if (notify)
747                 notify_remote_via_irq(bedata->irq);
748
749         wait_event(bedata->inflight_req,
750                    READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
751
752         /* read req_id, then the content */
753         smp_rmb();
754         ret = bedata->rsp[req_id].ret;
755         bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
756
757         map->passive.status = PVCALLS_STATUS_LISTEN;
758         pvcalls_exit_sock(sock);
759         return ret;
760 }
761
762 int pvcalls_front_accept(struct socket *sock, struct socket *newsock, int flags)
763 {
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;
769
770         map = pvcalls_enter_sock(sock);
771         if (IS_ERR(map))
772                 return PTR_ERR(map);
773         bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
774
775         if (map->passive.status != PVCALLS_STATUS_LISTEN) {
776                 pvcalls_exit_sock(sock);
777                 return -EINVAL;
778         }
779
780         nonblock = flags & SOCK_NONBLOCK;
781         /*
782          * Backend only supports 1 inflight accept request, will return
783          * errors for the others
784          */
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;
791                         goto received;
792                 }
793                 if (nonblock) {
794                         pvcalls_exit_sock(sock);
795                         return -EAGAIN;
796                 }
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);
801                         return -EINTR;
802                 }
803         }
804
805         map2 = kzalloc(sizeof(*map2), GFP_KERNEL);
806         if (map2 == NULL) {
807                 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
808                           (void *)&map->passive.flags);
809                 pvcalls_exit_sock(sock);
810                 return -ENOMEM;
811         }
812         ret = alloc_active_ring(map2);
813         if (ret < 0) {
814                 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
815                                 (void *)&map->passive.flags);
816                 kfree(map2);
817                 pvcalls_exit_sock(sock);
818                 return ret;
819         }
820         spin_lock(&bedata->socket_lock);
821         ret = get_request(bedata, &req_id);
822         if (ret < 0) {
823                 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
824                           (void *)&map->passive.flags);
825                 spin_unlock(&bedata->socket_lock);
826                 free_active_ring(map2);
827                 kfree(map2);
828                 pvcalls_exit_sock(sock);
829                 return ret;
830         }
831
832         ret = create_active(map2, &evtchn);
833         if (ret < 0) {
834                 free_active_ring(map2);
835                 kfree(map2);
836                 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
837                           (void *)&map->passive.flags);
838                 spin_unlock(&bedata->socket_lock);
839                 pvcalls_exit_sock(sock);
840                 return ret;
841         }
842         list_add_tail(&map2->list, &bedata->socket_mappings);
843
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;
852
853         bedata->ring.req_prod_pvt++;
854         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
855         spin_unlock(&bedata->socket_lock);
856         if (notify)
857                 notify_remote_via_irq(bedata->irq);
858         /* We could check if we have received a response before returning. */
859         if (nonblock) {
860                 WRITE_ONCE(map->passive.inflight_req_id, req_id);
861                 pvcalls_exit_sock(sock);
862                 return -EAGAIN;
863         }
864
865         if (wait_event_interruptible(bedata->inflight_req,
866                 READ_ONCE(bedata->rsp[req_id].req_id) == req_id)) {
867                 pvcalls_exit_sock(sock);
868                 return -EINTR;
869         }
870         /* read req_id, then the content */
871         smp_rmb();
872
873 received:
874         map2->sock = newsock;
875         newsock->sk = sk_alloc(sock_net(sock->sk), PF_INET, GFP_KERNEL, &pvcalls_proto, false);
876         if (!newsock->sk) {
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);
883                 return -ENOMEM;
884         }
885         newsock->sk->sk_send_head = (void *)map2;
886
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;
890
891         clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT, (void *)&map->passive.flags);
892         wake_up(&map->passive.inflight_accept_req);
893
894         pvcalls_exit_sock(sock);
895         return ret;
896 }
897
898 static __poll_t pvcalls_front_poll_passive(struct file *file,
899                                                struct pvcalls_bedata *bedata,
900                                                struct sock_mapping *map,
901                                                poll_table *wait)
902 {
903         int notify, req_id, ret;
904         struct xen_pvcalls_request *req;
905
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);
909
910                 if (req_id != PVCALLS_INVALID_ID &&
911                     READ_ONCE(bedata->rsp[req_id].req_id) == req_id)
912                         return EPOLLIN | EPOLLRDNORM;
913
914                 poll_wait(file, &map->passive.inflight_accept_req, wait);
915                 return 0;
916         }
917
918         if (test_and_clear_bit(PVCALLS_FLAG_POLL_RET,
919                                (void *)&map->passive.flags))
920                 return EPOLLIN | EPOLLRDNORM;
921
922         /*
923          * First check RET, then INFLIGHT. No barriers necessary to
924          * ensure execution ordering because of the conditional
925          * instructions creating control dependencies.
926          */
927
928         if (test_and_set_bit(PVCALLS_FLAG_POLL_INFLIGHT,
929                              (void *)&map->passive.flags)) {
930                 poll_wait(file, &bedata->inflight_req, wait);
931                 return 0;
932         }
933
934         spin_lock(&bedata->socket_lock);
935         ret = get_request(bedata, &req_id);
936         if (ret < 0) {
937                 spin_unlock(&bedata->socket_lock);
938                 return ret;
939         }
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;
944
945         bedata->ring.req_prod_pvt++;
946         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
947         spin_unlock(&bedata->socket_lock);
948         if (notify)
949                 notify_remote_via_irq(bedata->irq);
950
951         poll_wait(file, &bedata->inflight_req, wait);
952         return 0;
953 }
954
955 static __poll_t pvcalls_front_poll_active(struct file *file,
956                                               struct pvcalls_bedata *bedata,
957                                               struct sock_mapping *map,
958                                               poll_table *wait)
959 {
960         __poll_t mask = 0;
961         int32_t in_error, out_error;
962         struct pvcalls_data_intf *intf = map->active.ring;
963
964         out_error = intf->out_error;
965         in_error = intf->in_error;
966
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)
973                 mask |= EPOLLERR;
974
975         return mask;
976 }
977
978 __poll_t pvcalls_front_poll(struct file *file, struct socket *sock,
979                                poll_table *wait)
980 {
981         struct pvcalls_bedata *bedata;
982         struct sock_mapping *map;
983         __poll_t ret;
984
985         map = pvcalls_enter_sock(sock);
986         if (IS_ERR(map))
987                 return EPOLLNVAL;
988         bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
989
990         if (map->active_socket)
991                 ret = pvcalls_front_poll_active(file, bedata, map, wait);
992         else
993                 ret = pvcalls_front_poll_passive(file, bedata, map, wait);
994         pvcalls_exit_sock(sock);
995         return ret;
996 }
997
998 int pvcalls_front_release(struct socket *sock)
999 {
1000         struct pvcalls_bedata *bedata;
1001         struct sock_mapping *map;
1002         int req_id, notify, ret;
1003         struct xen_pvcalls_request *req;
1004
1005         if (sock->sk == NULL)
1006                 return 0;
1007
1008         map = pvcalls_enter_sock(sock);
1009         if (IS_ERR(map)) {
1010                 if (PTR_ERR(map) == -ENOTCONN)
1011                         return -EIO;
1012                 else
1013                         return 0;
1014         }
1015         bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
1016
1017         spin_lock(&bedata->socket_lock);
1018         ret = get_request(bedata, &req_id);
1019         if (ret < 0) {
1020                 spin_unlock(&bedata->socket_lock);
1021                 pvcalls_exit_sock(sock);
1022                 return ret;
1023         }
1024         sock->sk->sk_send_head = NULL;
1025
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;
1030
1031         bedata->ring.req_prod_pvt++;
1032         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
1033         spin_unlock(&bedata->socket_lock);
1034         if (notify)
1035                 notify_remote_via_irq(bedata->irq);
1036
1037         wait_event(bedata->inflight_req,
1038                    READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
1039
1040         if (map->active_socket) {
1041                 /*
1042                  * Set in_error and wake up inflight_conn_req to force
1043                  * recvmsg waiters to exit.
1044                  */
1045                 map->active.ring->in_error = -EBADF;
1046                 wake_up_interruptible(&map->active.inflight_conn_req);
1047
1048                 /*
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.
1053                  */
1054                 while (atomic_read(&map->refcount) > 1)
1055                         cpu_relax();
1056
1057                 pvcalls_front_free_map(bedata, map);
1058         } else {
1059                 wake_up(&bedata->inflight_req);
1060                 wake_up(&map->passive.inflight_accept_req);
1061
1062                 while (atomic_read(&map->refcount) > 1)
1063                         cpu_relax();
1064
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);
1072                 }
1073                 kfree(map);
1074         }
1075         WRITE_ONCE(bedata->rsp[req_id].req_id, PVCALLS_INVALID_ID);
1076
1077         pvcalls_exit();
1078         return 0;
1079 }
1080
1081 static const struct xenbus_device_id pvcalls_front_ids[] = {
1082         { "pvcalls" },
1083         { "" }
1084 };
1085
1086 static int pvcalls_front_remove(struct xenbus_device *dev)
1087 {
1088         struct pvcalls_bedata *bedata;
1089         struct sock_mapping *map = NULL, *n;
1090
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);
1096
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);
1102                 }
1103         }
1104
1105         smp_mb();
1106         while (atomic_read(&pvcalls_refcount) > 0)
1107                 cpu_relax();
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);
1112                 } else {
1113                         list_del(&map->list);
1114                         kfree(map);
1115                 }
1116         }
1117         if (bedata->ref != -1)
1118                 gnttab_end_foreign_access(bedata->ref, 0, 0);
1119         kfree(bedata->ring.sring);
1120         kfree(bedata);
1121         xenbus_switch_state(dev, XenbusStateClosed);
1122         return 0;
1123 }
1124
1125 static int pvcalls_front_probe(struct xenbus_device *dev,
1126                           const struct xenbus_device_id *id)
1127 {
1128         int ret = -ENOMEM, evtchn, i;
1129         unsigned int max_page_order, function_calls, len;
1130         char *versions;
1131         grant_ref_t gref_head = 0;
1132         struct xenbus_transaction xbt;
1133         struct pvcalls_bedata *bedata = NULL;
1134         struct xen_pvcalls_sring *sring;
1135
1136         if (pvcalls_front_dev != NULL) {
1137                 dev_err(&dev->dev, "only one PV Calls connection supported\n");
1138                 return -EINVAL;
1139         }
1140
1141         versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
1142         if (IS_ERR(versions))
1143                 return PTR_ERR(versions);
1144         if (!len)
1145                 return -EINVAL;
1146         if (strcmp(versions, "1")) {
1147                 kfree(versions);
1148                 return -EINVAL;
1149         }
1150         kfree(versions);
1151         max_page_order = xenbus_read_unsigned(dev->otherend,
1152                                               "max-page-order", 0);
1153         if (max_page_order < PVCALLS_RING_ORDER)
1154                 return -ENODEV;
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)
1159                 return -ENODEV;
1160         pr_info("%s max-page-order is %u\n", __func__, max_page_order);
1161
1162         bedata = kzalloc(sizeof(struct pvcalls_bedata), GFP_KERNEL);
1163         if (!bedata)
1164                 return -ENOMEM;
1165
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);
1171         bedata->irq = -1;
1172         bedata->ref = -1;
1173
1174         for (i = 0; i < PVCALLS_NR_RSP_PER_RING; i++)
1175                 bedata->rsp[i].req_id = PVCALLS_INVALID_ID;
1176
1177         sring = (struct xen_pvcalls_sring *) __get_free_page(GFP_KERNEL |
1178                                                              __GFP_ZERO);
1179         if (!sring)
1180                 goto error;
1181         SHARED_RING_INIT(sring);
1182         FRONT_RING_INIT(&bedata->ring, sring, XEN_PAGE_SIZE);
1183
1184         ret = xenbus_alloc_evtchn(dev, &evtchn);
1185         if (ret)
1186                 goto error;
1187
1188         bedata->irq = bind_evtchn_to_irqhandler(evtchn,
1189                                                 pvcalls_front_event_handler,
1190                                                 0, "pvcalls-frontend", dev);
1191         if (bedata->irq < 0) {
1192                 ret = bedata->irq;
1193                 goto error;
1194         }
1195
1196         ret = gnttab_alloc_grant_references(1, &gref_head);
1197         if (ret < 0)
1198                 goto error;
1199         ret = gnttab_claim_grant_reference(&gref_head);
1200         if (ret < 0)
1201                 goto error;
1202         bedata->ref = ret;
1203         gnttab_grant_foreign_access_ref(bedata->ref, dev->otherend_id,
1204                                         virt_to_gfn((void *)sring), 0);
1205
1206  again:
1207         ret = xenbus_transaction_start(&xbt);
1208         if (ret) {
1209                 xenbus_dev_fatal(dev, ret, "starting transaction");
1210                 goto error;
1211         }
1212         ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
1213         if (ret)
1214                 goto error_xenbus;
1215         ret = xenbus_printf(xbt, dev->nodename, "ring-ref", "%d", bedata->ref);
1216         if (ret)
1217                 goto error_xenbus;
1218         ret = xenbus_printf(xbt, dev->nodename, "port", "%u",
1219                             evtchn);
1220         if (ret)
1221                 goto error_xenbus;
1222         ret = xenbus_transaction_end(xbt, 0);
1223         if (ret) {
1224                 if (ret == -EAGAIN)
1225                         goto again;
1226                 xenbus_dev_fatal(dev, ret, "completing transaction");
1227                 goto error;
1228         }
1229         xenbus_switch_state(dev, XenbusStateInitialised);
1230
1231         return 0;
1232
1233  error_xenbus:
1234         xenbus_transaction_end(xbt, 1);
1235         xenbus_dev_fatal(dev, ret, "writing xenstore");
1236  error:
1237         pvcalls_front_remove(dev);
1238         return ret;
1239 }
1240
1241 static void pvcalls_front_changed(struct xenbus_device *dev,
1242                             enum xenbus_state backend_state)
1243 {
1244         switch (backend_state) {
1245         case XenbusStateReconfiguring:
1246         case XenbusStateReconfigured:
1247         case XenbusStateInitialising:
1248         case XenbusStateInitialised:
1249         case XenbusStateUnknown:
1250                 break;
1251
1252         case XenbusStateInitWait:
1253                 break;
1254
1255         case XenbusStateConnected:
1256                 xenbus_switch_state(dev, XenbusStateConnected);
1257                 break;
1258
1259         case XenbusStateClosed:
1260                 if (dev->state == XenbusStateClosed)
1261                         break;
1262                 /* Missed the backend's CLOSING state */
1263                 /* fall through */
1264         case XenbusStateClosing:
1265                 xenbus_frontend_closed(dev);
1266                 break;
1267         }
1268 }
1269
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,
1275 };
1276
1277 static int __init pvcalls_frontend_init(void)
1278 {
1279         if (!xen_domain())
1280                 return -ENODEV;
1281
1282         pr_info("Initialising Xen pvcalls frontend driver\n");
1283
1284         return xenbus_register_frontend(&pvcalls_front_driver);
1285 }
1286
1287 module_init(pvcalls_frontend_init);
1288
1289 MODULE_DESCRIPTION("Xen PV Calls frontend driver");
1290 MODULE_AUTHOR("Stefano Stabellini <sstabellini@kernel.org>");
1291 MODULE_LICENSE("GPL");