drivers/xen/pvcalls-front.c

   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>
  16 #include <linux/net.h>
  17 #include <linux/socket.h>
  18
  19 #include <net/sock.h>
  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
  27 #include "pvcalls-front.h"
  28
  29 #define PVCALLS_INVALID_ID UINT_MAX
  30 #define PVCALLS_RING_ORDER XENBUS_MAX_RING_GRANT_ORDER
  31 #define PVCALLS_NR_RSP_PER_RING __CONST_RING_SIZE(xen_pvcalls, XEN_PAGE_SIZE)
  32 #define PVCALLS_FRONT_MAX_SPIN 5000
  33
  34 struct pvcalls_bedata {
  35         struct xen_pvcalls_front_ring ring;
  36         grant_ref_t ref;
  37         int irq;
  38
  39         struct list_head socket_mappings;
  40         spinlock_t socket_lock;
  41
  42         wait_queue_head_t inflight_req;
  43         struct xen_pvcalls_response rsp[PVCALLS_NR_RSP_PER_RING];
  44 };
  45 /* Only one front/back connection supported. */
  46 static struct xenbus_device *pvcalls_front_dev;
  47 static atomic_t pvcalls_refcount;
  48
  49 /* first increment refcount, then proceed */
  50 #define pvcalls_enter() {               \
  51         atomic_inc(&pvcalls_refcount);      \
  52 }
  53
  54 /* first complete other operations, then decrement refcount */
  55 #define pvcalls_exit() {                \
  56         atomic_dec(&pvcalls_refcount);      \
  57 }
  58
  59 struct sock_mapping {
  60         bool active_socket;
  61         struct list_head list;
  62         struct socket *sock;
  63         union {
  64                 struct {
  65                         int irq;
  66                         grant_ref_t ref;
  67                         struct pvcalls_data_intf *ring;
  68                         struct pvcalls_data data;
  69                         struct mutex in_mutex;
  70                         struct mutex out_mutex;
  71
  72                         wait_queue_head_t inflight_conn_req;
  73                 } active;
  74                 struct {
  75                 /* Socket status */
  76 #define PVCALLS_STATUS_UNINITALIZED  0
  77 #define PVCALLS_STATUS_BIND          1
  78 #define PVCALLS_STATUS_LISTEN        2
  79                         uint8_t status;
  80                 /*
  81                  * Internal state-machine flags.
  82                  * Only one accept operation can be inflight for a socket.
  83                  * Only one poll operation can be inflight for a given socket.
  84                  */
  85 #define PVCALLS_FLAG_ACCEPT_INFLIGHT 0
  86 #define PVCALLS_FLAG_POLL_INFLIGHT   1
  87 #define PVCALLS_FLAG_POLL_RET        2
  88                         uint8_t flags;
  89                         uint32_t inflight_req_id;
  90                         struct sock_mapping *accept_map;
  91                         wait_queue_head_t inflight_accept_req;
  92                 } passive;
  93         };
  94 };
  95
  96 static inline int get_request(struct pvcalls_bedata *bedata, int *req_id)
  97 {
  98         *req_id = bedata->ring.req_prod_pvt & (RING_SIZE(&bedata->ring) - 1);
  99         if (RING_FULL(&bedata->ring) ||
 100             bedata->rsp[*req_id].req_id != PVCALLS_INVALID_ID)
 101                 return -EAGAIN;
 102         return 0;
 103 }
 104
 105 static bool pvcalls_front_write_todo(struct sock_mapping *map)
 106 {
 107         struct pvcalls_data_intf *intf = map->active.ring;
 108         RING_IDX cons, prod, size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
 109         int32_t error;
 110
 111         error = intf->out_error;
 112         if (error == -ENOTCONN)
 113                 return false;
 114         if (error != 0)
 115                 return true;
 116
 117         cons = intf->out_cons;
 118         prod = intf->out_prod;
 119         return !!(size - pvcalls_queued(prod, cons, size));
 120 }
 121
 122 static bool pvcalls_front_read_todo(struct sock_mapping *map)
 123 {
 124         struct pvcalls_data_intf *intf = map->active.ring;
 125         RING_IDX cons, prod;
 126         int32_t error;
 127
 128         cons = intf->in_cons;
 129         prod = intf->in_prod;
 130         error = intf->in_error;
 131         return (error != 0 ||
 132                 pvcalls_queued(prod, cons,
 133                                XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER)) != 0);
 134 }
 135
 136 static irqreturn_t pvcalls_front_event_handler(int irq, void *dev_id)
 137 {
 138         struct xenbus_device *dev = dev_id;
 139         struct pvcalls_bedata *bedata;
 140         struct xen_pvcalls_response *rsp;
 141         uint8_t *src, *dst;
 142         int req_id = 0, more = 0, done = 0;
 143
 144         if (dev == NULL)
 145                 return IRQ_HANDLED;
 146
 147         pvcalls_enter();
 148         bedata = dev_get_drvdata(&dev->dev);
 149         if (bedata == NULL) {
 150                 pvcalls_exit();
 151                 return IRQ_HANDLED;
 152         }
 153
 154 again:
 155         while (RING_HAS_UNCONSUMED_RESPONSES(&bedata->ring)) {
 156                 rsp = RING_GET_RESPONSE(&bedata->ring, bedata->ring.rsp_cons);
 157
 158                 req_id = rsp->req_id;
 159                 if (rsp->cmd == PVCALLS_POLL) {
 160                         struct sock_mapping *map = (struct sock_mapping *)(uintptr_t)
 161                                                    rsp->u.poll.id;
 162
 163                         clear_bit(PVCALLS_FLAG_POLL_INFLIGHT,
 164                                   (void *)&map->passive.flags);
 165                         /*
 166                          * clear INFLIGHT, then set RET. It pairs with
 167                          * the checks at the beginning of
 168                          * pvcalls_front_poll_passive.
 169                          */
 170                         smp_wmb();
 171                         set_bit(PVCALLS_FLAG_POLL_RET,
 172                                 (void *)&map->passive.flags);
 173                 } else {
 174                         dst = (uint8_t *)&bedata->rsp[req_id] +
 175                               sizeof(rsp->req_id);
 176                         src = (uint8_t *)rsp + sizeof(rsp->req_id);
 177                         memcpy(dst, src, sizeof(*rsp) - sizeof(rsp->req_id));
 178                         /*
 179                          * First copy the rest of the data, then req_id. It is
 180                          * paired with the barrier when accessing bedata->rsp.
 181                          */
 182                         smp_wmb();
 183                         bedata->rsp[req_id].req_id = req_id;
 184                 }
 185
 186                 done = 1;
 187                 bedata->ring.rsp_cons++;
 188         }
 189
 190         RING_FINAL_CHECK_FOR_RESPONSES(&bedata->ring, more);
 191         if (more)
 192                 goto again;
 193         if (done)
 194                 wake_up(&bedata->inflight_req);
 195         pvcalls_exit();
 196         return IRQ_HANDLED;
 197 }
 198
 199 static void pvcalls_front_free_map(struct pvcalls_bedata *bedata,
 200                                    struct sock_mapping *map)
 201 {
 202         int i;
 203
 204         unbind_from_irqhandler(map->active.irq, map);
 205
 206         spin_lock(&bedata->socket_lock);
 207         if (!list_empty(&map->list))
 208                 list_del_init(&map->list);
 209         spin_unlock(&bedata->socket_lock);
 210
 211         for (i = 0; i < (1 << PVCALLS_RING_ORDER); i++)
 212                 gnttab_end_foreign_access(map->active.ring->ref[i], 0, 0);
 213         gnttab_end_foreign_access(map->active.ref, 0, 0);
 214         free_page((unsigned long)map->active.ring);
 215
 216         kfree(map);
 217 }
 218
 219 static irqreturn_t pvcalls_front_conn_handler(int irq, void *sock_map)
 220 {
 221         struct sock_mapping *map = sock_map;
 222
 223         if (map == NULL)
 224                 return IRQ_HANDLED;
 225
 226         wake_up_interruptible(&map->active.inflight_conn_req);
 227
 228         return IRQ_HANDLED;
 229 }
 230
 231 int pvcalls_front_socket(struct socket *sock)
 232 {
 233         struct pvcalls_bedata *bedata;
 234         struct sock_mapping *map = NULL;
 235         struct xen_pvcalls_request *req;
 236         int notify, req_id, ret;
 237
 238         /*
 239          * PVCalls only supports domain AF_INET,
 240          * type SOCK_STREAM and protocol 0 sockets for now.
 241          *
 242          * Check socket type here, AF_INET and protocol checks are done
 243          * by the caller.
 244          */
 245         if (sock->type != SOCK_STREAM)
 246                 return -EOPNOTSUPP;
 247
 248         pvcalls_enter();
 249         if (!pvcalls_front_dev) {
 250                 pvcalls_exit();
 251                 return -EACCES;
 252         }
 253         bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
 254
 255         map = kzalloc(sizeof(*map), GFP_KERNEL);
 256         if (map == NULL) {
 257                 pvcalls_exit();
 258                 return -ENOMEM;
 259         }
 260
 261         spin_lock(&bedata->socket_lock);
 262
 263         ret = get_request(bedata, &req_id);
 264         if (ret < 0) {
 265                 kfree(map);
 266                 spin_unlock(&bedata->socket_lock);
 267                 pvcalls_exit();
 268                 return ret;
 269         }
 270
 271         /*
 272          * sock->sk->sk_send_head is not used for ip sockets: reuse the
 273          * field to store a pointer to the struct sock_mapping
 274          * corresponding to the socket. This way, we can easily get the
 275          * struct sock_mapping from the struct socket.
 276          */
 277         sock->sk->sk_send_head = (void *)map;
 278         list_add_tail(&map->list, &bedata->socket_mappings);
 279
 280         req = RING_GET_REQUEST(&bedata->ring, req_id);
 281         req->req_id = req_id;
 282         req->cmd = PVCALLS_SOCKET;
 283         req->u.socket.id = (uintptr_t) map;
 284         req->u.socket.domain = AF_INET;
 285         req->u.socket.type = SOCK_STREAM;
 286         req->u.socket.protocol = IPPROTO_IP;
 287
 288         bedata->ring.req_prod_pvt++;
 289         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
 290         spin_unlock(&bedata->socket_lock);
 291         if (notify)
 292                 notify_remote_via_irq(bedata->irq);
 293
 294         wait_event(bedata->inflight_req,
 295                    READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
 296
 297         /* read req_id, then the content */
 298         smp_rmb();
 299         ret = bedata->rsp[req_id].ret;
 300         bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
 301
 302         pvcalls_exit();
 303         return ret;
 304 }
 305
 306 static int create_active(struct sock_mapping *map, int *evtchn)
 307 {
 308         void *bytes;
 309         int ret = -ENOMEM, irq = -1, i;
 310
 311         *evtchn = -1;
 312         init_waitqueue_head(&map->active.inflight_conn_req);
 313
 314         map->active.ring = (struct pvcalls_data_intf *)
 315                 __get_free_page(GFP_KERNEL | __GFP_ZERO);
 316         if (map->active.ring == NULL)
 317                 goto out_error;
 318         map->active.ring->ring_order = PVCALLS_RING_ORDER;
 319         bytes = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
 320                                         PVCALLS_RING_ORDER);
 321         if (bytes == NULL)
 322                 goto out_error;
 323         for (i = 0; i < (1 << PVCALLS_RING_ORDER); i++)
 324                 map->active.ring->ref[i] = gnttab_grant_foreign_access(
 325                         pvcalls_front_dev->otherend_id,
 326                         pfn_to_gfn(virt_to_pfn(bytes) + i), 0);
 327
 328         map->active.ref = gnttab_grant_foreign_access(
 329                 pvcalls_front_dev->otherend_id,
 330                 pfn_to_gfn(virt_to_pfn((void *)map->active.ring)), 0);
 331
 332         map->active.data.in = bytes;
 333         map->active.data.out = bytes +
 334                 XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
 335
 336         ret = xenbus_alloc_evtchn(pvcalls_front_dev, evtchn);
 337         if (ret)
 338                 goto out_error;
 339         irq = bind_evtchn_to_irqhandler(*evtchn, pvcalls_front_conn_handler,
 340                                         0, "pvcalls-frontend", map);
 341         if (irq < 0) {
 342                 ret = irq;
 343                 goto out_error;
 344         }
 345
 346         map->active.irq = irq;
 347         map->active_socket = true;
 348         mutex_init(&map->active.in_mutex);
 349         mutex_init(&map->active.out_mutex);
 350
 351         return 0;
 352
 353 out_error:
 354         if (*evtchn >= 0)
 355                 xenbus_free_evtchn(pvcalls_front_dev, *evtchn);
 356         kfree(map->active.data.in);
 357         kfree(map->active.ring);
 358         return ret;
 359 }
 360
 361 int pvcalls_front_connect(struct socket *sock, struct sockaddr *addr,
 362                                 int addr_len, int flags)
 363 {
 364         struct pvcalls_bedata *bedata;
 365         struct sock_mapping *map = NULL;
 366         struct xen_pvcalls_request *req;
 367         int notify, req_id, ret, evtchn;
 368
 369         if (addr->sa_family != AF_INET || sock->type != SOCK_STREAM)
 370                 return -EOPNOTSUPP;
 371
 372         pvcalls_enter();
 373         if (!pvcalls_front_dev) {
 374                 pvcalls_exit();
 375                 return -ENOTCONN;
 376         }
 377
 378         bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
 379
 380         map = (struct sock_mapping *)sock->sk->sk_send_head;
 381         if (!map) {
 382                 pvcalls_exit();
 383                 return -ENOTSOCK;
 384         }
 385
 386         spin_lock(&bedata->socket_lock);
 387         ret = get_request(bedata, &req_id);
 388         if (ret < 0) {
 389                 spin_unlock(&bedata->socket_lock);
 390                 pvcalls_exit();
 391                 return ret;
 392         }
 393         ret = create_active(map, &evtchn);
 394         if (ret < 0) {
 395                 spin_unlock(&bedata->socket_lock);
 396                 pvcalls_exit();
 397                 return ret;
 398         }
 399
 400         req = RING_GET_REQUEST(&bedata->ring, req_id);
 401         req->req_id = req_id;
 402         req->cmd = PVCALLS_CONNECT;
 403         req->u.connect.id = (uintptr_t)map;
 404         req->u.connect.len = addr_len;
 405         req->u.connect.flags = flags;
 406         req->u.connect.ref = map->active.ref;
 407         req->u.connect.evtchn = evtchn;
 408         memcpy(req->u.connect.addr, addr, sizeof(*addr));
 409
 410         map->sock = sock;
 411
 412         bedata->ring.req_prod_pvt++;
 413         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
 414         spin_unlock(&bedata->socket_lock);
 415
 416         if (notify)
 417                 notify_remote_via_irq(bedata->irq);
 418
 419         wait_event(bedata->inflight_req,
 420                    READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
 421
 422         /* read req_id, then the content */
 423         smp_rmb();
 424         ret = bedata->rsp[req_id].ret;
 425         bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
 426         pvcalls_exit();
 427         return ret;
 428 }
 429
 430 static int __write_ring(struct pvcalls_data_intf *intf,
 431                         struct pvcalls_data *data,
 432                         struct iov_iter *msg_iter,
 433                         int len)
 434 {
 435         RING_IDX cons, prod, size, masked_prod, masked_cons;
 436         RING_IDX array_size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
 437         int32_t error;
 438
 439         error = intf->out_error;
 440         if (error < 0)
 441                 return error;
 442         cons = intf->out_cons;
 443         prod = intf->out_prod;
 444         /* read indexes before continuing */
 445         virt_mb();
 446
 447         size = pvcalls_queued(prod, cons, array_size);
 448         if (size >= array_size)
 449                 return -EINVAL;
 450         if (len > array_size - size)
 451                 len = array_size - size;
 452
 453         masked_prod = pvcalls_mask(prod, array_size);
 454         masked_cons = pvcalls_mask(cons, array_size);
 455
 456         if (masked_prod < masked_cons) {
 457                 len = copy_from_iter(data->out + masked_prod, len, msg_iter);
 458         } else {
 459                 if (len > array_size - masked_prod) {
 460                         int ret = copy_from_iter(data->out + masked_prod,
 461                                        array_size - masked_prod, msg_iter);
 462                         if (ret != array_size - masked_prod) {
 463                                 len = ret;
 464                                 goto out;
 465                         }
 466                         len = ret + copy_from_iter(data->out, len - ret, msg_iter);
 467                 } else {
 468                         len = copy_from_iter(data->out + masked_prod, len, msg_iter);
 469                 }
 470         }
 471 out:
 472         /* write to ring before updating pointer */
 473         virt_wmb();
 474         intf->out_prod += len;
 475
 476         return len;
 477 }
 478
 479 int pvcalls_front_sendmsg(struct socket *sock, struct msghdr *msg,
 480                           size_t len)
 481 {
 482         struct pvcalls_bedata *bedata;
 483         struct sock_mapping *map;
 484         int sent, tot_sent = 0;
 485         int count = 0, flags;
 486
 487         flags = msg->msg_flags;
 488         if (flags & (MSG_CONFIRM|MSG_DONTROUTE|MSG_EOR|MSG_OOB))
 489                 return -EOPNOTSUPP;
 490
 491         pvcalls_enter();
 492         if (!pvcalls_front_dev) {
 493                 pvcalls_exit();
 494                 return -ENOTCONN;
 495         }
 496         bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
 497
 498         map = (struct sock_mapping *) sock->sk->sk_send_head;
 499         if (!map) {
 500                 pvcalls_exit();
 501                 return -ENOTSOCK;
 502         }
 503
 504         mutex_lock(&map->active.out_mutex);
 505         if ((flags & MSG_DONTWAIT) && !pvcalls_front_write_todo(map)) {
 506                 mutex_unlock(&map->active.out_mutex);
 507                 pvcalls_exit();
 508                 return -EAGAIN;
 509         }
 510         if (len > INT_MAX)
 511                 len = INT_MAX;
 512
 513 again:
 514         count++;
 515         sent = __write_ring(map->active.ring,
 516                             &map->active.data, &msg->msg_iter,
 517                             len);
 518         if (sent > 0) {
 519                 len -= sent;
 520                 tot_sent += sent;
 521                 notify_remote_via_irq(map->active.irq);
 522         }
 523         if (sent >= 0 && len > 0 && count < PVCALLS_FRONT_MAX_SPIN)
 524                 goto again;
 525         if (sent < 0)
 526                 tot_sent = sent;
 527
 528         mutex_unlock(&map->active.out_mutex);
 529         pvcalls_exit();
 530         return tot_sent;
 531 }
 532
 533 static int __read_ring(struct pvcalls_data_intf *intf,
 534                        struct pvcalls_data *data,
 535                        struct iov_iter *msg_iter,
 536                        size_t len, int flags)
 537 {
 538         RING_IDX cons, prod, size, masked_prod, masked_cons;
 539         RING_IDX array_size = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
 540         int32_t error;
 541
 542         cons = intf->in_cons;
 543         prod = intf->in_prod;
 544         error = intf->in_error;
 545         /* get pointers before reading from the ring */
 546         virt_rmb();
 547         if (error < 0)
 548                 return error;
 549
 550         size = pvcalls_queued(prod, cons, array_size);
 551         masked_prod = pvcalls_mask(prod, array_size);
 552         masked_cons = pvcalls_mask(cons, array_size);
 553
 554         if (size == 0)
 555                 return 0;
 556
 557         if (len > size)
 558                 len = size;
 559
 560         if (masked_prod > masked_cons) {
 561                 len = copy_to_iter(data->in + masked_cons, len, msg_iter);
 562         } else {
 563                 if (len > (array_size - masked_cons)) {
 564                         int ret = copy_to_iter(data->in + masked_cons,
 565                                      array_size - masked_cons, msg_iter);
 566                         if (ret != array_size - masked_cons) {
 567                                 len = ret;
 568                                 goto out;
 569                         }
 570                         len = ret + copy_to_iter(data->in, len - ret, msg_iter);
 571                 } else {
 572                         len = copy_to_iter(data->in + masked_cons, len, msg_iter);
 573                 }
 574         }
 575 out:
 576         /* read data from the ring before increasing the index */
 577         virt_mb();
 578         if (!(flags & MSG_PEEK))
 579                 intf->in_cons += len;
 580
 581         return len;
 582 }
 583
 584 int pvcalls_front_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
 585                      int flags)
 586 {
 587         struct pvcalls_bedata *bedata;
 588         int ret;
 589         struct sock_mapping *map;
 590
 591         if (flags & (MSG_CMSG_CLOEXEC|MSG_ERRQUEUE|MSG_OOB|MSG_TRUNC))
 592                 return -EOPNOTSUPP;
 593
 594         pvcalls_enter();
 595         if (!pvcalls_front_dev) {
 596                 pvcalls_exit();
 597                 return -ENOTCONN;
 598         }
 599         bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
 600
 601         map = (struct sock_mapping *) sock->sk->sk_send_head;
 602         if (!map) {
 603                 pvcalls_exit();
 604                 return -ENOTSOCK;
 605         }
 606
 607         mutex_lock(&map->active.in_mutex);
 608         if (len > XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER))
 609                 len = XEN_FLEX_RING_SIZE(PVCALLS_RING_ORDER);
 610
 611         while (!(flags & MSG_DONTWAIT) && !pvcalls_front_read_todo(map)) {
 612                 wait_event_interruptible(map->active.inflight_conn_req,
 613                                          pvcalls_front_read_todo(map));
 614         }
 615         ret = __read_ring(map->active.ring, &map->active.data,
 616                           &msg->msg_iter, len, flags);
 617
 618         if (ret > 0)
 619                 notify_remote_via_irq(map->active.irq);
 620         if (ret == 0)
 621                 ret = (flags & MSG_DONTWAIT) ? -EAGAIN : 0;
 622         if (ret == -ENOTCONN)
 623                 ret = 0;
 624
 625         mutex_unlock(&map->active.in_mutex);
 626         pvcalls_exit();
 627         return ret;
 628 }
 629
 630 int pvcalls_front_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 631 {
 632         struct pvcalls_bedata *bedata;
 633         struct sock_mapping *map = NULL;
 634         struct xen_pvcalls_request *req;
 635         int notify, req_id, ret;
 636
 637         if (addr->sa_family != AF_INET || sock->type != SOCK_STREAM)
 638                 return -EOPNOTSUPP;
 639
 640         pvcalls_enter();
 641         if (!pvcalls_front_dev) {
 642                 pvcalls_exit();
 643                 return -ENOTCONN;
 644         }
 645         bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
 646
 647         map = (struct sock_mapping *) sock->sk->sk_send_head;
 648         if (map == NULL) {
 649                 pvcalls_exit();
 650                 return -ENOTSOCK;
 651         }
 652
 653         spin_lock(&bedata->socket_lock);
 654         ret = get_request(bedata, &req_id);
 655         if (ret < 0) {
 656                 spin_unlock(&bedata->socket_lock);
 657                 pvcalls_exit();
 658                 return ret;
 659         }
 660         req = RING_GET_REQUEST(&bedata->ring, req_id);
 661         req->req_id = req_id;
 662         map->sock = sock;
 663         req->cmd = PVCALLS_BIND;
 664         req->u.bind.id = (uintptr_t)map;
 665         memcpy(req->u.bind.addr, addr, sizeof(*addr));
 666         req->u.bind.len = addr_len;
 667
 668         init_waitqueue_head(&map->passive.inflight_accept_req);
 669
 670         map->active_socket = false;
 671
 672         bedata->ring.req_prod_pvt++;
 673         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
 674         spin_unlock(&bedata->socket_lock);
 675         if (notify)
 676                 notify_remote_via_irq(bedata->irq);
 677
 678         wait_event(bedata->inflight_req,
 679                    READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
 680
 681         /* read req_id, then the content */
 682         smp_rmb();
 683         ret = bedata->rsp[req_id].ret;
 684         bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
 685
 686         map->passive.status = PVCALLS_STATUS_BIND;
 687         pvcalls_exit();
 688         return 0;
 689 }
 690
 691 int pvcalls_front_listen(struct socket *sock, int backlog)
 692 {
 693         struct pvcalls_bedata *bedata;
 694         struct sock_mapping *map;
 695         struct xen_pvcalls_request *req;
 696         int notify, req_id, ret;
 697
 698         pvcalls_enter();
 699         if (!pvcalls_front_dev) {
 700                 pvcalls_exit();
 701                 return -ENOTCONN;
 702         }
 703         bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
 704
 705         map = (struct sock_mapping *) sock->sk->sk_send_head;
 706         if (!map) {
 707                 pvcalls_exit();
 708                 return -ENOTSOCK;
 709         }
 710
 711         if (map->passive.status != PVCALLS_STATUS_BIND) {
 712                 pvcalls_exit();
 713                 return -EOPNOTSUPP;
 714         }
 715
 716         spin_lock(&bedata->socket_lock);
 717         ret = get_request(bedata, &req_id);
 718         if (ret < 0) {
 719                 spin_unlock(&bedata->socket_lock);
 720                 pvcalls_exit();
 721                 return ret;
 722         }
 723         req = RING_GET_REQUEST(&bedata->ring, req_id);
 724         req->req_id = req_id;
 725         req->cmd = PVCALLS_LISTEN;
 726         req->u.listen.id = (uintptr_t) map;
 727         req->u.listen.backlog = backlog;
 728
 729         bedata->ring.req_prod_pvt++;
 730         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
 731         spin_unlock(&bedata->socket_lock);
 732         if (notify)
 733                 notify_remote_via_irq(bedata->irq);
 734
 735         wait_event(bedata->inflight_req,
 736                    READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
 737
 738         /* read req_id, then the content */
 739         smp_rmb();
 740         ret = bedata->rsp[req_id].ret;
 741         bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
 742
 743         map->passive.status = PVCALLS_STATUS_LISTEN;
 744         pvcalls_exit();
 745         return ret;
 746 }
 747
 748 int pvcalls_front_accept(struct socket *sock, struct socket *newsock, int flags)
 749 {
 750         struct pvcalls_bedata *bedata;
 751         struct sock_mapping *map;
 752         struct sock_mapping *map2 = NULL;
 753         struct xen_pvcalls_request *req;
 754         int notify, req_id, ret, evtchn, nonblock;
 755
 756         pvcalls_enter();
 757         if (!pvcalls_front_dev) {
 758                 pvcalls_exit();
 759                 return -ENOTCONN;
 760         }
 761         bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
 762
 763         map = (struct sock_mapping *) sock->sk->sk_send_head;
 764         if (!map) {
 765                 pvcalls_exit();
 766                 return -ENOTSOCK;
 767         }
 768
 769         if (map->passive.status != PVCALLS_STATUS_LISTEN) {
 770                 pvcalls_exit();
 771                 return -EINVAL;
 772         }
 773
 774         nonblock = flags & SOCK_NONBLOCK;
 775         /*
 776          * Backend only supports 1 inflight accept request, will return
 777          * errors for the others
 778          */
 779         if (test_and_set_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
 780                              (void *)&map->passive.flags)) {
 781                 req_id = READ_ONCE(map->passive.inflight_req_id);
 782                 if (req_id != PVCALLS_INVALID_ID &&
 783                     READ_ONCE(bedata->rsp[req_id].req_id) == req_id) {
 784                         map2 = map->passive.accept_map;
 785                         goto received;
 786                 }
 787                 if (nonblock) {
 788                         pvcalls_exit();
 789                         return -EAGAIN;
 790                 }
 791                 if (wait_event_interruptible(map->passive.inflight_accept_req,
 792                         !test_and_set_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
 793                                           (void *)&map->passive.flags))) {
 794                         pvcalls_exit();
 795                         return -EINTR;
 796                 }
 797         }
 798
 799         spin_lock(&bedata->socket_lock);
 800         ret = get_request(bedata, &req_id);
 801         if (ret < 0) {
 802                 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
 803                           (void *)&map->passive.flags);
 804                 spin_unlock(&bedata->socket_lock);
 805                 pvcalls_exit();
 806                 return ret;
 807         }
 808         map2 = kzalloc(sizeof(*map2), GFP_ATOMIC);
 809         if (map2 == NULL) {
 810                 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
 811                           (void *)&map->passive.flags);
 812                 spin_unlock(&bedata->socket_lock);
 813                 pvcalls_exit();
 814                 return -ENOMEM;
 815         }
 816         ret = create_active(map2, &evtchn);
 817         if (ret < 0) {
 818                 kfree(map2);
 819                 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
 820                           (void *)&map->passive.flags);
 821                 spin_unlock(&bedata->socket_lock);
 822                 pvcalls_exit();
 823                 return ret;
 824         }
 825         list_add_tail(&map2->list, &bedata->socket_mappings);
 826
 827         req = RING_GET_REQUEST(&bedata->ring, req_id);
 828         req->req_id = req_id;
 829         req->cmd = PVCALLS_ACCEPT;
 830         req->u.accept.id = (uintptr_t) map;
 831         req->u.accept.ref = map2->active.ref;
 832         req->u.accept.id_new = (uintptr_t) map2;
 833         req->u.accept.evtchn = evtchn;
 834         map->passive.accept_map = map2;
 835
 836         bedata->ring.req_prod_pvt++;
 837         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
 838         spin_unlock(&bedata->socket_lock);
 839         if (notify)
 840                 notify_remote_via_irq(bedata->irq);
 841         /* We could check if we have received a response before returning. */
 842         if (nonblock) {
 843                 WRITE_ONCE(map->passive.inflight_req_id, req_id);
 844                 pvcalls_exit();
 845                 return -EAGAIN;
 846         }
 847
 848         if (wait_event_interruptible(bedata->inflight_req,
 849                 READ_ONCE(bedata->rsp[req_id].req_id) == req_id)) {
 850                 pvcalls_exit();
 851                 return -EINTR;
 852         }
 853         /* read req_id, then the content */
 854         smp_rmb();
 855
 856 received:
 857         map2->sock = newsock;
 858         newsock->sk = kzalloc(sizeof(*newsock->sk), GFP_KERNEL);
 859         if (!newsock->sk) {
 860                 bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
 861                 map->passive.inflight_req_id = PVCALLS_INVALID_ID;
 862                 clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
 863                           (void *)&map->passive.flags);
 864                 pvcalls_front_free_map(bedata, map2);
 865                 pvcalls_exit();
 866                 return -ENOMEM;
 867         }
 868         newsock->sk->sk_send_head = (void *)map2;
 869
 870         ret = bedata->rsp[req_id].ret;
 871         bedata->rsp[req_id].req_id = PVCALLS_INVALID_ID;
 872         map->passive.inflight_req_id = PVCALLS_INVALID_ID;
 873
 874         clear_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT, (void *)&map->passive.flags);
 875         wake_up(&map->passive.inflight_accept_req);
 876
 877         pvcalls_exit();
 878         return ret;
 879 }
 880
 881 static unsigned int pvcalls_front_poll_passive(struct file *file,
 882                                                struct pvcalls_bedata *bedata,
 883                                                struct sock_mapping *map,
 884                                                poll_table *wait)
 885 {
 886         int notify, req_id, ret;
 887         struct xen_pvcalls_request *req;
 888
 889         if (test_bit(PVCALLS_FLAG_ACCEPT_INFLIGHT,
 890                      (void *)&map->passive.flags)) {
 891                 uint32_t req_id = READ_ONCE(map->passive.inflight_req_id);
 892
 893                 if (req_id != PVCALLS_INVALID_ID &&
 894                     READ_ONCE(bedata->rsp[req_id].req_id) == req_id)
 895                         return POLLIN | POLLRDNORM;
 896
 897                 poll_wait(file, &map->passive.inflight_accept_req, wait);
 898                 return 0;
 899         }
 900
 901         if (test_and_clear_bit(PVCALLS_FLAG_POLL_RET,
 902                                (void *)&map->passive.flags))
 903                 return POLLIN | POLLRDNORM;
 904
 905         /*
 906          * First check RET, then INFLIGHT. No barriers necessary to
 907          * ensure execution ordering because of the conditional
 908          * instructions creating control dependencies.
 909          */
 910
 911         if (test_and_set_bit(PVCALLS_FLAG_POLL_INFLIGHT,
 912                              (void *)&map->passive.flags)) {
 913                 poll_wait(file, &bedata->inflight_req, wait);
 914                 return 0;
 915         }
 916
 917         spin_lock(&bedata->socket_lock);
 918         ret = get_request(bedata, &req_id);
 919         if (ret < 0) {
 920                 spin_unlock(&bedata->socket_lock);
 921                 return ret;
 922         }
 923         req = RING_GET_REQUEST(&bedata->ring, req_id);
 924         req->req_id = req_id;
 925         req->cmd = PVCALLS_POLL;
 926         req->u.poll.id = (uintptr_t) map;
 927
 928         bedata->ring.req_prod_pvt++;
 929         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
 930         spin_unlock(&bedata->socket_lock);
 931         if (notify)
 932                 notify_remote_via_irq(bedata->irq);
 933
 934         poll_wait(file, &bedata->inflight_req, wait);
 935         return 0;
 936 }
 937
 938 static unsigned int pvcalls_front_poll_active(struct file *file,
 939                                               struct pvcalls_bedata *bedata,
 940                                               struct sock_mapping *map,
 941                                               poll_table *wait)
 942 {
 943         unsigned int mask = 0;
 944         int32_t in_error, out_error;
 945         struct pvcalls_data_intf *intf = map->active.ring;
 946
 947         out_error = intf->out_error;
 948         in_error = intf->in_error;
 949
 950         poll_wait(file, &map->active.inflight_conn_req, wait);
 951         if (pvcalls_front_write_todo(map))
 952                 mask |= POLLOUT | POLLWRNORM;
 953         if (pvcalls_front_read_todo(map))
 954                 mask |= POLLIN | POLLRDNORM;
 955         if (in_error != 0 || out_error != 0)
 956                 mask |= POLLERR;
 957
 958         return mask;
 959 }
 960
 961 unsigned int pvcalls_front_poll(struct file *file, struct socket *sock,
 962                                poll_table *wait)
 963 {
 964         struct pvcalls_bedata *bedata;
 965         struct sock_mapping *map;
 966         int ret;
 967
 968         pvcalls_enter();
 969         if (!pvcalls_front_dev) {
 970                 pvcalls_exit();
 971                 return POLLNVAL;
 972         }
 973         bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
 974
 975         map = (struct sock_mapping *) sock->sk->sk_send_head;
 976         if (!map) {
 977                 pvcalls_exit();
 978                 return POLLNVAL;
 979         }
 980         if (map->active_socket)
 981                 ret = pvcalls_front_poll_active(file, bedata, map, wait);
 982         else
 983                 ret = pvcalls_front_poll_passive(file, bedata, map, wait);
 984         pvcalls_exit();
 985         return ret;
 986 }
 987
 988 int pvcalls_front_release(struct socket *sock)
 989 {
 990         struct pvcalls_bedata *bedata;
 991         struct sock_mapping *map;
 992         int req_id, notify, ret;
 993         struct xen_pvcalls_request *req;
 994
 995         if (sock->sk == NULL)
 996                 return 0;
 997
 998         pvcalls_enter();
 999         if (!pvcalls_front_dev) {
1000                 pvcalls_exit();
1001                 return -EIO;
1002         }
1003
1004         bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
1005
1006         map = (struct sock_mapping *) sock->sk->sk_send_head;
1007         if (map == NULL) {
1008                 pvcalls_exit();
1009                 return 0;
1010         }
1011
1012         spin_lock(&bedata->socket_lock);
1013         ret = get_request(bedata, &req_id);
1014         if (ret < 0) {
1015                 spin_unlock(&bedata->socket_lock);
1016                 pvcalls_exit();
1017                 return ret;
1018         }
1019         sock->sk->sk_send_head = NULL;
1020
1021         req = RING_GET_REQUEST(&bedata->ring, req_id);
1022         req->req_id = req_id;
1023         req->cmd = PVCALLS_RELEASE;
1024         req->u.release.id = (uintptr_t)map;
1025
1026         bedata->ring.req_prod_pvt++;
1027         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&bedata->ring, notify);
1028         spin_unlock(&bedata->socket_lock);
1029         if (notify)
1030                 notify_remote_via_irq(bedata->irq);
1031
1032         wait_event(bedata->inflight_req,
1033                    READ_ONCE(bedata->rsp[req_id].req_id) == req_id);
1034
1035         if (map->active_socket) {
1036                 /*
1037                  * Set in_error and wake up inflight_conn_req to force
1038                  * recvmsg waiters to exit.
1039                  */
1040                 map->active.ring->in_error = -EBADF;
1041                 wake_up_interruptible(&map->active.inflight_conn_req);
1042
1043                 /*
1044                  * We need to make sure that sendmsg/recvmsg on this socket have
1045                  * not started before we've cleared sk_send_head here. The
1046                  * easiest (though not optimal) way to guarantee this is to see
1047                  * that no pvcall (other than us) is in progress.
1048                  */
1049                 while (atomic_read(&pvcalls_refcount) > 1)
1050                         cpu_relax();
1051
1052                 pvcalls_front_free_map(bedata, map);
1053         } else {
1054                 spin_lock(&bedata->socket_lock);
1055                 list_del(&map->list);
1056                 spin_unlock(&bedata->socket_lock);
1057                 if (READ_ONCE(map->passive.inflight_req_id) !=
1058                     PVCALLS_INVALID_ID) {
1059                         pvcalls_front_free_map(bedata,
1060                                                map->passive.accept_map);
1061                 }
1062                 kfree(map);
1063         }
1064         WRITE_ONCE(bedata->rsp[req_id].req_id, PVCALLS_INVALID_ID);
1065
1066         pvcalls_exit();
1067         return 0;
1068 }
1069
1070 static const struct xenbus_device_id pvcalls_front_ids[] = {
1071         { "pvcalls" },
1072         { "" }
1073 };
1074
1075 static int pvcalls_front_remove(struct xenbus_device *dev)
1076 {
1077         struct pvcalls_bedata *bedata;
1078         struct sock_mapping *map = NULL, *n;
1079
1080         bedata = dev_get_drvdata(&pvcalls_front_dev->dev);
1081         dev_set_drvdata(&dev->dev, NULL);
1082         pvcalls_front_dev = NULL;
1083         if (bedata->irq >= 0)
1084                 unbind_from_irqhandler(bedata->irq, dev);
1085
1086         list_for_each_entry_safe(map, n, &bedata->socket_mappings, list) {
1087                 map->sock->sk->sk_send_head = NULL;
1088                 if (map->active_socket) {
1089                         map->active.ring->in_error = -EBADF;
1090                         wake_up_interruptible(&map->active.inflight_conn_req);
1091                 }
1092         }
1093
1094         smp_mb();
1095         while (atomic_read(&pvcalls_refcount) > 0)
1096                 cpu_relax();
1097         list_for_each_entry_safe(map, n, &bedata->socket_mappings, list) {
1098                 if (map->active_socket) {
1099                         /* No need to lock, refcount is 0 */
1100                         pvcalls_front_free_map(bedata, map);
1101                 } else {
1102                         list_del(&map->list);
1103                         kfree(map);
1104                 }
1105         }
1106         if (bedata->ref != -1)
1107                 gnttab_end_foreign_access(bedata->ref, 0, 0);
1108         kfree(bedata->ring.sring);
1109         kfree(bedata);
1110         xenbus_switch_state(dev, XenbusStateClosed);
1111         return 0;
1112 }
1113
1114 static int pvcalls_front_probe(struct xenbus_device *dev,
1115                           const struct xenbus_device_id *id)
1116 {
1117         int ret = -ENOMEM, evtchn, i;
1118         unsigned int max_page_order, function_calls, len;
1119         char *versions;
1120         grant_ref_t gref_head = 0;
1121         struct xenbus_transaction xbt;
1122         struct pvcalls_bedata *bedata = NULL;
1123         struct xen_pvcalls_sring *sring;
1124
1125         if (pvcalls_front_dev != NULL) {
1126                 dev_err(&dev->dev, "only one PV Calls connection supported\n");
1127                 return -EINVAL;
1128         }
1129
1130         versions = xenbus_read(XBT_NIL, dev->otherend, "versions", &len);
1131         if (IS_ERR(versions))
1132                 return PTR_ERR(versions);
1133         if (!len)
1134                 return -EINVAL;
1135         if (strcmp(versions, "1")) {
1136                 kfree(versions);
1137                 return -EINVAL;
1138         }
1139         kfree(versions);
1140         max_page_order = xenbus_read_unsigned(dev->otherend,
1141                                               "max-page-order", 0);
1142         if (max_page_order < PVCALLS_RING_ORDER)
1143                 return -ENODEV;
1144         function_calls = xenbus_read_unsigned(dev->otherend,
1145                                               "function-calls", 0);
1146         /* See XENBUS_FUNCTIONS_CALLS in pvcalls.h */
1147         if (function_calls != 1)
1148                 return -ENODEV;
1149         pr_info("%s max-page-order is %u\n", __func__, max_page_order);
1150
1151         bedata = kzalloc(sizeof(struct pvcalls_bedata), GFP_KERNEL);
1152         if (!bedata)
1153                 return -ENOMEM;
1154
1155         dev_set_drvdata(&dev->dev, bedata);
1156         pvcalls_front_dev = dev;
1157         init_waitqueue_head(&bedata->inflight_req);
1158         INIT_LIST_HEAD(&bedata->socket_mappings);
1159         spin_lock_init(&bedata->socket_lock);
1160         bedata->irq = -1;
1161         bedata->ref = -1;
1162
1163         for (i = 0; i < PVCALLS_NR_RSP_PER_RING; i++)
1164                 bedata->rsp[i].req_id = PVCALLS_INVALID_ID;
1165
1166         sring = (struct xen_pvcalls_sring *) __get_free_page(GFP_KERNEL |
1167                                                              __GFP_ZERO);
1168         if (!sring)
1169                 goto error;
1170         SHARED_RING_INIT(sring);
1171         FRONT_RING_INIT(&bedata->ring, sring, XEN_PAGE_SIZE);
1172
1173         ret = xenbus_alloc_evtchn(dev, &evtchn);
1174         if (ret)
1175                 goto error;
1176
1177         bedata->irq = bind_evtchn_to_irqhandler(evtchn,
1178                                                 pvcalls_front_event_handler,
1179                                                 0, "pvcalls-frontend", dev);
1180         if (bedata->irq < 0) {
1181                 ret = bedata->irq;
1182                 goto error;
1183         }
1184
1185         ret = gnttab_alloc_grant_references(1, &gref_head);
1186         if (ret < 0)
1187                 goto error;
1188         ret = gnttab_claim_grant_reference(&gref_head);
1189         if (ret < 0)
1190                 goto error;
1191         bedata->ref = ret;
1192         gnttab_grant_foreign_access_ref(bedata->ref, dev->otherend_id,
1193                                         virt_to_gfn((void *)sring), 0);
1194
1195  again:
1196         ret = xenbus_transaction_start(&xbt);
1197         if (ret) {
1198                 xenbus_dev_fatal(dev, ret, "starting transaction");
1199                 goto error;
1200         }
1201         ret = xenbus_printf(xbt, dev->nodename, "version", "%u", 1);
1202         if (ret)
1203                 goto error_xenbus;
1204         ret = xenbus_printf(xbt, dev->nodename, "ring-ref", "%d", bedata->ref);
1205         if (ret)
1206                 goto error_xenbus;
1207         ret = xenbus_printf(xbt, dev->nodename, "port", "%u",
1208                             evtchn);
1209         if (ret)
1210                 goto error_xenbus;
1211         ret = xenbus_transaction_end(xbt, 0);
1212         if (ret) {
1213                 if (ret == -EAGAIN)
1214                         goto again;
1215                 xenbus_dev_fatal(dev, ret, "completing transaction");
1216                 goto error;
1217         }
1218         xenbus_switch_state(dev, XenbusStateInitialised);
1219
1220         return 0;
1221
1222  error_xenbus:
1223         xenbus_transaction_end(xbt, 1);
1224         xenbus_dev_fatal(dev, ret, "writing xenstore");
1225  error:
1226         pvcalls_front_remove(dev);
1227         return ret;
1228 }
1229
1230 static void pvcalls_front_changed(struct xenbus_device *dev,
1231                             enum xenbus_state backend_state)
1232 {
1233         switch (backend_state) {
1234         case XenbusStateReconfiguring:
1235         case XenbusStateReconfigured:
1236         case XenbusStateInitialising:
1237         case XenbusStateInitialised:
1238         case XenbusStateUnknown:
1239                 break;
1240
1241         case XenbusStateInitWait:
1242                 break;
1243
1244         case XenbusStateConnected:
1245                 xenbus_switch_state(dev, XenbusStateConnected);
1246                 break;
1247
1248         case XenbusStateClosed:
1249                 if (dev->state == XenbusStateClosed)
1250                         break;
1251                 /* Missed the backend's CLOSING state */
1252                 /* fall through */
1253         case XenbusStateClosing:
1254                 xenbus_frontend_closed(dev);
1255                 break;
1256         }
1257 }
1258
1259 static struct xenbus_driver pvcalls_front_driver = {
1260         .ids = pvcalls_front_ids,
1261         .probe = pvcalls_front_probe,
1262         .remove = pvcalls_front_remove,
1263         .otherend_changed = pvcalls_front_changed,
1264 };
1265
1266 static int __init pvcalls_frontend_init(void)
1267 {
1268         if (!xen_domain())
1269                 return -ENODEV;
1270
1271         pr_info("Initialising Xen pvcalls frontend driver\n");
1272
1273         return xenbus_register_frontend(&pvcalls_front_driver);
1274 }
1275
1276 module_init(pvcalls_frontend_init);
1277
1278 MODULE_DESCRIPTION("Xen PV Calls frontend driver");
1279 MODULE_AUTHOR("Stefano Stabellini <sstabellini@kernel.org>");
1280 MODULE_LICENSE("GPL");