drivers/xen/pvcalls-back.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * (c) 2017 Stefano Stabellini <stefano@aporeto.com>
   4  */
   5
   6 #include <linux/inet.h>
   7 #include <linux/kthread.h>
   8 #include <linux/list.h>
   9 #include <linux/radix-tree.h>
  10 #include <linux/module.h>
  11 #include <linux/semaphore.h>
  12 #include <linux/wait.h>
  13 #include <net/sock.h>
  14 #include <net/inet_common.h>
  15 #include <net/inet_connection_sock.h>
  16 #include <net/request_sock.h>
  17
  18 #include <xen/events.h>
  19 #include <xen/grant_table.h>
  20 #include <xen/xen.h>
  21 #include <xen/xenbus.h>
  22 #include <xen/interface/io/pvcalls.h>
  23
  24 #define PVCALLS_VERSIONS "1"
  25 #define MAX_RING_ORDER XENBUS_MAX_RING_GRANT_ORDER
  26
  27 struct pvcalls_back_global {
  28         struct list_head frontends;
  29         struct semaphore frontends_lock;
  30 } pvcalls_back_global;
  31
  32 /*
  33  * Per-frontend data structure. It contains pointers to the command
  34  * ring, its event channel, a list of active sockets and a tree of
  35  * passive sockets.
  36  */
  37 struct pvcalls_fedata {
  38         struct list_head list;
  39         struct xenbus_device *dev;
  40         struct xen_pvcalls_sring *sring;
  41         struct xen_pvcalls_back_ring ring;
  42         int irq;
  43         struct list_head socket_mappings;
  44         struct radix_tree_root socketpass_mappings;
  45         struct semaphore socket_lock;
  46 };
  47
  48 struct pvcalls_ioworker {
  49         struct work_struct register_work;
  50         struct workqueue_struct *wq;
  51 };
  52
  53 struct sock_mapping {
  54         struct list_head list;
  55         struct pvcalls_fedata *fedata;
  56         struct sockpass_mapping *sockpass;
  57         struct socket *sock;
  58         uint64_t id;
  59         grant_ref_t ref;
  60         struct pvcalls_data_intf *ring;
  61         void *bytes;
  62         struct pvcalls_data data;
  63         uint32_t ring_order;
  64         int irq;
  65         atomic_t read;
  66         atomic_t write;
  67         atomic_t io;
  68         atomic_t release;
  69         void (*saved_data_ready)(struct sock *sk);
  70         struct pvcalls_ioworker ioworker;
  71 };
  72
  73 struct sockpass_mapping {
  74         struct list_head list;
  75         struct pvcalls_fedata *fedata;
  76         struct socket *sock;
  77         uint64_t id;
  78         struct xen_pvcalls_request reqcopy;
  79         spinlock_t copy_lock;
  80         struct workqueue_struct *wq;
  81         struct work_struct register_work;
  82         void (*saved_data_ready)(struct sock *sk);
  83 };
  84
  85 static irqreturn_t pvcalls_back_conn_event(int irq, void *sock_map);
  86 static int pvcalls_back_release_active(struct xenbus_device *dev,
  87                                        struct pvcalls_fedata *fedata,
  88                                        struct sock_mapping *map);
  89
  90 static void pvcalls_conn_back_read(void *opaque)
  91 {
  92         struct sock_mapping *map = (struct sock_mapping *)opaque;
  93         struct msghdr msg;
  94         struct kvec vec[2];
  95         RING_IDX cons, prod, size, wanted, array_size, masked_prod, masked_cons;
  96         int32_t error;
  97         struct pvcalls_data_intf *intf = map->ring;
  98         struct pvcalls_data *data = &map->data;
  99         unsigned long flags;
 100         int ret;
 101
 102         array_size = XEN_FLEX_RING_SIZE(map->ring_order);
 103         cons = intf->in_cons;
 104         prod = intf->in_prod;
 105         error = intf->in_error;
 106         /* read the indexes first, then deal with the data */
 107         virt_mb();
 108
 109         if (error)
 110                 return;
 111
 112         size = pvcalls_queued(prod, cons, array_size);
 113         if (size >= array_size)
 114                 return;
 115         spin_lock_irqsave(&map->sock->sk->sk_receive_queue.lock, flags);
 116         if (skb_queue_empty(&map->sock->sk->sk_receive_queue)) {
 117                 atomic_set(&map->read, 0);
 118                 spin_unlock_irqrestore(&map->sock->sk->sk_receive_queue.lock,
 119                                 flags);
 120                 return;
 121         }
 122         spin_unlock_irqrestore(&map->sock->sk->sk_receive_queue.lock, flags);
 123         wanted = array_size - size;
 124         masked_prod = pvcalls_mask(prod, array_size);
 125         masked_cons = pvcalls_mask(cons, array_size);
 126
 127         memset(&msg, 0, sizeof(msg));
 128         if (masked_prod < masked_cons) {
 129                 vec[0].iov_base = data->in + masked_prod;
 130                 vec[0].iov_len = wanted;
 131                 iov_iter_kvec(&msg.msg_iter, WRITE, vec, 1, wanted);
 132         } else {
 133                 vec[0].iov_base = data->in + masked_prod;
 134                 vec[0].iov_len = array_size - masked_prod;
 135                 vec[1].iov_base = data->in;
 136                 vec[1].iov_len = wanted - vec[0].iov_len;
 137                 iov_iter_kvec(&msg.msg_iter, WRITE, vec, 2, wanted);
 138         }
 139
 140         atomic_set(&map->read, 0);
 141         ret = inet_recvmsg(map->sock, &msg, wanted, MSG_DONTWAIT);
 142         WARN_ON(ret > wanted);
 143         if (ret == -EAGAIN) /* shouldn't happen */
 144                 return;
 145         if (!ret)
 146                 ret = -ENOTCONN;
 147         spin_lock_irqsave(&map->sock->sk->sk_receive_queue.lock, flags);
 148         if (ret > 0 && !skb_queue_empty(&map->sock->sk->sk_receive_queue))
 149                 atomic_inc(&map->read);
 150         spin_unlock_irqrestore(&map->sock->sk->sk_receive_queue.lock, flags);
 151
 152         /* write the data, then modify the indexes */
 153         virt_wmb();
 154         if (ret < 0) {
 155                 atomic_set(&map->read, 0);
 156                 intf->in_error = ret;
 157         } else
 158                 intf->in_prod = prod + ret;
 159         /* update the indexes, then notify the other end */
 160         virt_wmb();
 161         notify_remote_via_irq(map->irq);
 162
 163         return;
 164 }
 165
 166 static void pvcalls_conn_back_write(struct sock_mapping *map)
 167 {
 168         struct pvcalls_data_intf *intf = map->ring;
 169         struct pvcalls_data *data = &map->data;
 170         struct msghdr msg;
 171         struct kvec vec[2];
 172         RING_IDX cons, prod, size, array_size;
 173         int ret;
 174
 175         cons = intf->out_cons;
 176         prod = intf->out_prod;
 177         /* read the indexes before dealing with the data */
 178         virt_mb();
 179
 180         array_size = XEN_FLEX_RING_SIZE(map->ring_order);
 181         size = pvcalls_queued(prod, cons, array_size);
 182         if (size == 0)
 183                 return;
 184
 185         memset(&msg, 0, sizeof(msg));
 186         msg.msg_flags |= MSG_DONTWAIT;
 187         if (pvcalls_mask(prod, array_size) > pvcalls_mask(cons, array_size)) {
 188                 vec[0].iov_base = data->out + pvcalls_mask(cons, array_size);
 189                 vec[0].iov_len = size;
 190                 iov_iter_kvec(&msg.msg_iter, READ, vec, 1, size);
 191         } else {
 192                 vec[0].iov_base = data->out + pvcalls_mask(cons, array_size);
 193                 vec[0].iov_len = array_size - pvcalls_mask(cons, array_size);
 194                 vec[1].iov_base = data->out;
 195                 vec[1].iov_len = size - vec[0].iov_len;
 196                 iov_iter_kvec(&msg.msg_iter, READ, vec, 2, size);
 197         }
 198
 199         atomic_set(&map->write, 0);
 200         ret = inet_sendmsg(map->sock, &msg, size);
 201         if (ret == -EAGAIN || (ret >= 0 && ret < size)) {
 202                 atomic_inc(&map->write);
 203                 atomic_inc(&map->io);
 204         }
 205         if (ret == -EAGAIN)
 206                 return;
 207
 208         /* write the data, then update the indexes */
 209         virt_wmb();
 210         if (ret < 0) {
 211                 intf->out_error = ret;
 212         } else {
 213                 intf->out_error = 0;
 214                 intf->out_cons = cons + ret;
 215                 prod = intf->out_prod;
 216         }
 217         /* update the indexes, then notify the other end */
 218         virt_wmb();
 219         if (prod != cons + ret)
 220                 atomic_inc(&map->write);
 221         notify_remote_via_irq(map->irq);
 222 }
 223
 224 static void pvcalls_back_ioworker(struct work_struct *work)
 225 {
 226         struct pvcalls_ioworker *ioworker = container_of(work,
 227                 struct pvcalls_ioworker, register_work);
 228         struct sock_mapping *map = container_of(ioworker, struct sock_mapping,
 229                 ioworker);
 230
 231         while (atomic_read(&map->io) > 0) {
 232                 if (atomic_read(&map->release) > 0) {
 233                         atomic_set(&map->release, 0);
 234                         return;
 235                 }
 236
 237                 if (atomic_read(&map->read) > 0)
 238                         pvcalls_conn_back_read(map);
 239                 if (atomic_read(&map->write) > 0)
 240                         pvcalls_conn_back_write(map);
 241
 242                 atomic_dec(&map->io);
 243         }
 244 }
 245
 246 static int pvcalls_back_socket(struct xenbus_device *dev,
 247                 struct xen_pvcalls_request *req)
 248 {
 249         struct pvcalls_fedata *fedata;
 250         int ret;
 251         struct xen_pvcalls_response *rsp;
 252
 253         fedata = dev_get_drvdata(&dev->dev);
 254
 255         if (req->u.socket.domain != AF_INET ||
 256             req->u.socket.type != SOCK_STREAM ||
 257             (req->u.socket.protocol != IPPROTO_IP &&
 258              req->u.socket.protocol != AF_INET))
 259                 ret = -EAFNOSUPPORT;
 260         else
 261                 ret = 0;
 262
 263         /* leave the actual socket allocation for later */
 264
 265         rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
 266         rsp->req_id = req->req_id;
 267         rsp->cmd = req->cmd;
 268         rsp->u.socket.id = req->u.socket.id;
 269         rsp->ret = ret;
 270
 271         return 0;
 272 }
 273
 274 static void pvcalls_sk_state_change(struct sock *sock)
 275 {
 276         struct sock_mapping *map = sock->sk_user_data;
 277
 278         if (map == NULL)
 279                 return;
 280
 281         atomic_inc(&map->read);
 282         notify_remote_via_irq(map->irq);
 283 }
 284
 285 static void pvcalls_sk_data_ready(struct sock *sock)
 286 {
 287         struct sock_mapping *map = sock->sk_user_data;
 288         struct pvcalls_ioworker *iow;
 289
 290         if (map == NULL)
 291                 return;
 292
 293         iow = &map->ioworker;
 294         atomic_inc(&map->read);
 295         atomic_inc(&map->io);
 296         queue_work(iow->wq, &iow->register_work);
 297 }
 298
 299 static struct sock_mapping *pvcalls_new_active_socket(
 300                 struct pvcalls_fedata *fedata,
 301                 uint64_t id,
 302                 grant_ref_t ref,
 303                 evtchn_port_t evtchn,
 304                 struct socket *sock)
 305 {
 306         int ret;
 307         struct sock_mapping *map;
 308         void *page;
 309
 310         map = kzalloc(sizeof(*map), GFP_KERNEL);
 311         if (map == NULL)
 312                 return NULL;
 313
 314         map->fedata = fedata;
 315         map->sock = sock;
 316         map->id = id;
 317         map->ref = ref;
 318
 319         ret = xenbus_map_ring_valloc(fedata->dev, &ref, 1, &page);
 320         if (ret < 0)
 321                 goto out;
 322         map->ring = page;
 323         map->ring_order = map->ring->ring_order;
 324         /* first read the order, then map the data ring */
 325         virt_rmb();
 326         if (map->ring_order > MAX_RING_ORDER) {
 327                 pr_warn("%s frontend requested ring_order %u, which is > MAX (%u)\n",
 328                                 __func__, map->ring_order, MAX_RING_ORDER);
 329                 goto out;
 330         }
 331         ret = xenbus_map_ring_valloc(fedata->dev, map->ring->ref,
 332                                      (1 << map->ring_order), &page);
 333         if (ret < 0)
 334                 goto out;
 335         map->bytes = page;
 336
 337         ret = bind_interdomain_evtchn_to_irqhandler(fedata->dev->otherend_id,
 338                                                     evtchn,
 339                                                     pvcalls_back_conn_event,
 340                                                     0,
 341                                                     "pvcalls-backend",
 342                                                     map);
 343         if (ret < 0)
 344                 goto out;
 345         map->irq = ret;
 346
 347         map->data.in = map->bytes;
 348         map->data.out = map->bytes + XEN_FLEX_RING_SIZE(map->ring_order);
 349
 350         map->ioworker.wq = alloc_workqueue("pvcalls_io", WQ_UNBOUND, 1);
 351         if (!map->ioworker.wq)
 352                 goto out;
 353         atomic_set(&map->io, 1);
 354         INIT_WORK(&map->ioworker.register_work, pvcalls_back_ioworker);
 355
 356         down(&fedata->socket_lock);
 357         list_add_tail(&map->list, &fedata->socket_mappings);
 358         up(&fedata->socket_lock);
 359
 360         write_lock_bh(&map->sock->sk->sk_callback_lock);
 361         map->saved_data_ready = map->sock->sk->sk_data_ready;
 362         map->sock->sk->sk_user_data = map;
 363         map->sock->sk->sk_data_ready = pvcalls_sk_data_ready;
 364         map->sock->sk->sk_state_change = pvcalls_sk_state_change;
 365         write_unlock_bh(&map->sock->sk->sk_callback_lock);
 366
 367         return map;
 368 out:
 369         down(&fedata->socket_lock);
 370         list_del(&map->list);
 371         pvcalls_back_release_active(fedata->dev, fedata, map);
 372         up(&fedata->socket_lock);
 373         return NULL;
 374 }
 375
 376 static int pvcalls_back_connect(struct xenbus_device *dev,
 377                                 struct xen_pvcalls_request *req)
 378 {
 379         struct pvcalls_fedata *fedata;
 380         int ret = -EINVAL;
 381         struct socket *sock;
 382         struct sock_mapping *map;
 383         struct xen_pvcalls_response *rsp;
 384         struct sockaddr *sa = (struct sockaddr *)&req->u.connect.addr;
 385
 386         fedata = dev_get_drvdata(&dev->dev);
 387
 388         if (req->u.connect.len < sizeof(sa->sa_family) ||
 389             req->u.connect.len > sizeof(req->u.connect.addr) ||
 390             sa->sa_family != AF_INET)
 391                 goto out;
 392
 393         ret = sock_create(AF_INET, SOCK_STREAM, 0, &sock);
 394         if (ret < 0)
 395                 goto out;
 396         ret = inet_stream_connect(sock, sa, req->u.connect.len, 0);
 397         if (ret < 0) {
 398                 sock_release(sock);
 399                 goto out;
 400         }
 401
 402         map = pvcalls_new_active_socket(fedata,
 403                                         req->u.connect.id,
 404                                         req->u.connect.ref,
 405                                         req->u.connect.evtchn,
 406                                         sock);
 407         if (!map) {
 408                 ret = -EFAULT;
 409                 sock_release(sock);
 410         }
 411
 412 out:
 413         rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
 414         rsp->req_id = req->req_id;
 415         rsp->cmd = req->cmd;
 416         rsp->u.connect.id = req->u.connect.id;
 417         rsp->ret = ret;
 418
 419         return 0;
 420 }
 421
 422 static int pvcalls_back_release_active(struct xenbus_device *dev,
 423                                        struct pvcalls_fedata *fedata,
 424                                        struct sock_mapping *map)
 425 {
 426         disable_irq(map->irq);
 427         if (map->sock->sk != NULL) {
 428                 write_lock_bh(&map->sock->sk->sk_callback_lock);
 429                 map->sock->sk->sk_user_data = NULL;
 430                 map->sock->sk->sk_data_ready = map->saved_data_ready;
 431                 write_unlock_bh(&map->sock->sk->sk_callback_lock);
 432         }
 433
 434         atomic_set(&map->release, 1);
 435         flush_work(&map->ioworker.register_work);
 436
 437         xenbus_unmap_ring_vfree(dev, map->bytes);
 438         xenbus_unmap_ring_vfree(dev, (void *)map->ring);
 439         unbind_from_irqhandler(map->irq, map);
 440
 441         sock_release(map->sock);
 442         kfree(map);
 443
 444         return 0;
 445 }
 446
 447 static int pvcalls_back_release_passive(struct xenbus_device *dev,
 448                                         struct pvcalls_fedata *fedata,
 449                                         struct sockpass_mapping *mappass)
 450 {
 451         if (mappass->sock->sk != NULL) {
 452                 write_lock_bh(&mappass->sock->sk->sk_callback_lock);
 453                 mappass->sock->sk->sk_user_data = NULL;
 454                 mappass->sock->sk->sk_data_ready = mappass->saved_data_ready;
 455                 write_unlock_bh(&mappass->sock->sk->sk_callback_lock);
 456         }
 457         sock_release(mappass->sock);
 458         flush_workqueue(mappass->wq);
 459         destroy_workqueue(mappass->wq);
 460         kfree(mappass);
 461
 462         return 0;
 463 }
 464
 465 static int pvcalls_back_release(struct xenbus_device *dev,
 466                                 struct xen_pvcalls_request *req)
 467 {
 468         struct pvcalls_fedata *fedata;
 469         struct sock_mapping *map, *n;
 470         struct sockpass_mapping *mappass;
 471         int ret = 0;
 472         struct xen_pvcalls_response *rsp;
 473
 474         fedata = dev_get_drvdata(&dev->dev);
 475
 476         down(&fedata->socket_lock);
 477         list_for_each_entry_safe(map, n, &fedata->socket_mappings, list) {
 478                 if (map->id == req->u.release.id) {
 479                         list_del(&map->list);
 480                         up(&fedata->socket_lock);
 481                         ret = pvcalls_back_release_active(dev, fedata, map);
 482                         goto out;
 483                 }
 484         }
 485         mappass = radix_tree_lookup(&fedata->socketpass_mappings,
 486                                     req->u.release.id);
 487         if (mappass != NULL) {
 488                 radix_tree_delete(&fedata->socketpass_mappings, mappass->id);
 489                 up(&fedata->socket_lock);
 490                 ret = pvcalls_back_release_passive(dev, fedata, mappass);
 491         } else
 492                 up(&fedata->socket_lock);
 493
 494 out:
 495         rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
 496         rsp->req_id = req->req_id;
 497         rsp->u.release.id = req->u.release.id;
 498         rsp->cmd = req->cmd;
 499         rsp->ret = ret;
 500         return 0;
 501 }
 502
 503 static void __pvcalls_back_accept(struct work_struct *work)
 504 {
 505         struct sockpass_mapping *mappass = container_of(
 506                 work, struct sockpass_mapping, register_work);
 507         struct sock_mapping *map;
 508         struct pvcalls_ioworker *iow;
 509         struct pvcalls_fedata *fedata;
 510         struct socket *sock;
 511         struct xen_pvcalls_response *rsp;
 512         struct xen_pvcalls_request *req;
 513         int notify;
 514         int ret = -EINVAL;
 515         unsigned long flags;
 516
 517         fedata = mappass->fedata;
 518         /*
 519          * __pvcalls_back_accept can race against pvcalls_back_accept.
 520          * We only need to check the value of "cmd" on read. It could be
 521          * done atomically, but to simplify the code on the write side, we
 522          * use a spinlock.
 523          */
 524         spin_lock_irqsave(&mappass->copy_lock, flags);
 525         req = &mappass->reqcopy;
 526         if (req->cmd != PVCALLS_ACCEPT) {
 527                 spin_unlock_irqrestore(&mappass->copy_lock, flags);
 528                 return;
 529         }
 530         spin_unlock_irqrestore(&mappass->copy_lock, flags);
 531
 532         sock = sock_alloc();
 533         if (sock == NULL)
 534                 goto out_error;
 535         sock->type = mappass->sock->type;
 536         sock->ops = mappass->sock->ops;
 537
 538         ret = inet_accept(mappass->sock, sock, O_NONBLOCK, true);
 539         if (ret == -EAGAIN) {
 540                 sock_release(sock);
 541                 return;
 542         }
 543
 544         map = pvcalls_new_active_socket(fedata,
 545                                         req->u.accept.id_new,
 546                                         req->u.accept.ref,
 547                                         req->u.accept.evtchn,
 548                                         sock);
 549         if (!map) {
 550                 ret = -EFAULT;
 551                 sock_release(sock);
 552                 goto out_error;
 553         }
 554
 555         map->sockpass = mappass;
 556         iow = &map->ioworker;
 557         atomic_inc(&map->read);
 558         atomic_inc(&map->io);
 559         queue_work(iow->wq, &iow->register_work);
 560
 561 out_error:
 562         rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
 563         rsp->req_id = req->req_id;
 564         rsp->cmd = req->cmd;
 565         rsp->u.accept.id = req->u.accept.id;
 566         rsp->ret = ret;
 567         RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&fedata->ring, notify);
 568         if (notify)
 569                 notify_remote_via_irq(fedata->irq);
 570
 571         mappass->reqcopy.cmd = 0;
 572 }
 573
 574 static void pvcalls_pass_sk_data_ready(struct sock *sock)
 575 {
 576         struct sockpass_mapping *mappass = sock->sk_user_data;
 577         struct pvcalls_fedata *fedata;
 578         struct xen_pvcalls_response *rsp;
 579         unsigned long flags;
 580         int notify;
 581
 582         if (mappass == NULL)
 583                 return;
 584
 585         fedata = mappass->fedata;
 586         spin_lock_irqsave(&mappass->copy_lock, flags);
 587         if (mappass->reqcopy.cmd == PVCALLS_POLL) {
 588                 rsp = RING_GET_RESPONSE(&fedata->ring,
 589                                         fedata->ring.rsp_prod_pvt++);
 590                 rsp->req_id = mappass->reqcopy.req_id;
 591                 rsp->u.poll.id = mappass->reqcopy.u.poll.id;
 592                 rsp->cmd = mappass->reqcopy.cmd;
 593                 rsp->ret = 0;
 594
 595                 mappass->reqcopy.cmd = 0;
 596                 spin_unlock_irqrestore(&mappass->copy_lock, flags);
 597
 598                 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&fedata->ring, notify);
 599                 if (notify)
 600                         notify_remote_via_irq(mappass->fedata->irq);
 601         } else {
 602                 spin_unlock_irqrestore(&mappass->copy_lock, flags);
 603                 queue_work(mappass->wq, &mappass->register_work);
 604         }
 605 }
 606
 607 static int pvcalls_back_bind(struct xenbus_device *dev,
 608                              struct xen_pvcalls_request *req)
 609 {
 610         struct pvcalls_fedata *fedata;
 611         int ret;
 612         struct sockpass_mapping *map;
 613         struct xen_pvcalls_response *rsp;
 614
 615         fedata = dev_get_drvdata(&dev->dev);
 616
 617         map = kzalloc(sizeof(*map), GFP_KERNEL);
 618         if (map == NULL) {
 619                 ret = -ENOMEM;
 620                 goto out;
 621         }
 622
 623         INIT_WORK(&map->register_work, __pvcalls_back_accept);
 624         spin_lock_init(&map->copy_lock);
 625         map->wq = alloc_workqueue("pvcalls_wq", WQ_UNBOUND, 1);
 626         if (!map->wq) {
 627                 ret = -ENOMEM;
 628                 goto out;
 629         }
 630
 631         ret = sock_create(AF_INET, SOCK_STREAM, 0, &map->sock);
 632         if (ret < 0)
 633                 goto out;
 634
 635         ret = inet_bind(map->sock, (struct sockaddr *)&req->u.bind.addr,
 636                         req->u.bind.len);
 637         if (ret < 0)
 638                 goto out;
 639
 640         map->fedata = fedata;
 641         map->id = req->u.bind.id;
 642
 643         down(&fedata->socket_lock);
 644         ret = radix_tree_insert(&fedata->socketpass_mappings, map->id,
 645                                 map);
 646         up(&fedata->socket_lock);
 647         if (ret)
 648                 goto out;
 649
 650         write_lock_bh(&map->sock->sk->sk_callback_lock);
 651         map->saved_data_ready = map->sock->sk->sk_data_ready;
 652         map->sock->sk->sk_user_data = map;
 653         map->sock->sk->sk_data_ready = pvcalls_pass_sk_data_ready;
 654         write_unlock_bh(&map->sock->sk->sk_callback_lock);
 655
 656 out:
 657         if (ret) {
 658                 if (map && map->sock)
 659                         sock_release(map->sock);
 660                 if (map && map->wq)
 661                         destroy_workqueue(map->wq);
 662                 kfree(map);
 663         }
 664         rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
 665         rsp->req_id = req->req_id;
 666         rsp->cmd = req->cmd;
 667         rsp->u.bind.id = req->u.bind.id;
 668         rsp->ret = ret;
 669         return 0;
 670 }
 671
 672 static int pvcalls_back_listen(struct xenbus_device *dev,
 673                                struct xen_pvcalls_request *req)
 674 {
 675         struct pvcalls_fedata *fedata;
 676         int ret = -EINVAL;
 677         struct sockpass_mapping *map;
 678         struct xen_pvcalls_response *rsp;
 679
 680         fedata = dev_get_drvdata(&dev->dev);
 681
 682         down(&fedata->socket_lock);
 683         map = radix_tree_lookup(&fedata->socketpass_mappings, req->u.listen.id);
 684         up(&fedata->socket_lock);
 685         if (map == NULL)
 686                 goto out;
 687
 688         ret = inet_listen(map->sock, req->u.listen.backlog);
 689
 690 out:
 691         rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
 692         rsp->req_id = req->req_id;
 693         rsp->cmd = req->cmd;
 694         rsp->u.listen.id = req->u.listen.id;
 695         rsp->ret = ret;
 696         return 0;
 697 }
 698
 699 static int pvcalls_back_accept(struct xenbus_device *dev,
 700                                struct xen_pvcalls_request *req)
 701 {
 702         struct pvcalls_fedata *fedata;
 703         struct sockpass_mapping *mappass;
 704         int ret = -EINVAL;
 705         struct xen_pvcalls_response *rsp;
 706         unsigned long flags;
 707
 708         fedata = dev_get_drvdata(&dev->dev);
 709
 710         down(&fedata->socket_lock);
 711         mappass = radix_tree_lookup(&fedata->socketpass_mappings,
 712                 req->u.accept.id);
 713         up(&fedata->socket_lock);
 714         if (mappass == NULL)
 715                 goto out_error;
 716
 717         /*
 718          * Limitation of the current implementation: only support one
 719          * concurrent accept or poll call on one socket.
 720          */
 721         spin_lock_irqsave(&mappass->copy_lock, flags);
 722         if (mappass->reqcopy.cmd != 0) {
 723                 spin_unlock_irqrestore(&mappass->copy_lock, flags);
 724                 ret = -EINTR;
 725                 goto out_error;
 726         }
 727
 728         mappass->reqcopy = *req;
 729         spin_unlock_irqrestore(&mappass->copy_lock, flags);
 730         queue_work(mappass->wq, &mappass->register_work);
 731
 732         /* Tell the caller we don't need to send back a notification yet */
 733         return -1;
 734
 735 out_error:
 736         rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
 737         rsp->req_id = req->req_id;
 738         rsp->cmd = req->cmd;
 739         rsp->u.accept.id = req->u.accept.id;
 740         rsp->ret = ret;
 741         return 0;
 742 }
 743
 744 static int pvcalls_back_poll(struct xenbus_device *dev,
 745                              struct xen_pvcalls_request *req)
 746 {
 747         struct pvcalls_fedata *fedata;
 748         struct sockpass_mapping *mappass;
 749         struct xen_pvcalls_response *rsp;
 750         struct inet_connection_sock *icsk;
 751         struct request_sock_queue *queue;
 752         unsigned long flags;
 753         int ret;
 754         bool data;
 755
 756         fedata = dev_get_drvdata(&dev->dev);
 757
 758         down(&fedata->socket_lock);
 759         mappass = radix_tree_lookup(&fedata->socketpass_mappings,
 760                                     req->u.poll.id);
 761         up(&fedata->socket_lock);
 762         if (mappass == NULL)
 763                 return -EINVAL;
 764
 765         /*
 766          * Limitation of the current implementation: only support one
 767          * concurrent accept or poll call on one socket.
 768          */
 769         spin_lock_irqsave(&mappass->copy_lock, flags);
 770         if (mappass->reqcopy.cmd != 0) {
 771                 ret = -EINTR;
 772                 goto out;
 773         }
 774
 775         mappass->reqcopy = *req;
 776         icsk = inet_csk(mappass->sock->sk);
 777         queue = &icsk->icsk_accept_queue;
 778         data = READ_ONCE(queue->rskq_accept_head) != NULL;
 779         if (data) {
 780                 mappass->reqcopy.cmd = 0;
 781                 ret = 0;
 782                 goto out;
 783         }
 784         spin_unlock_irqrestore(&mappass->copy_lock, flags);
 785
 786         /* Tell the caller we don't need to send back a notification yet */
 787         return -1;
 788
 789 out:
 790         spin_unlock_irqrestore(&mappass->copy_lock, flags);
 791
 792         rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
 793         rsp->req_id = req->req_id;
 794         rsp->cmd = req->cmd;
 795         rsp->u.poll.id = req->u.poll.id;
 796         rsp->ret = ret;
 797         return 0;
 798 }
 799
 800 static int pvcalls_back_handle_cmd(struct xenbus_device *dev,
 801                                    struct xen_pvcalls_request *req)
 802 {
 803         int ret = 0;
 804
 805         switch (req->cmd) {
 806         case PVCALLS_SOCKET:
 807                 ret = pvcalls_back_socket(dev, req);
 808                 break;
 809         case PVCALLS_CONNECT:
 810                 ret = pvcalls_back_connect(dev, req);
 811                 break;
 812         case PVCALLS_RELEASE:
 813                 ret = pvcalls_back_release(dev, req);
 814                 break;
 815         case PVCALLS_BIND:
 816                 ret = pvcalls_back_bind(dev, req);
 817                 break;
 818         case PVCALLS_LISTEN:
 819                 ret = pvcalls_back_listen(dev, req);
 820                 break;
 821         case PVCALLS_ACCEPT:
 822                 ret = pvcalls_back_accept(dev, req);
 823                 break;
 824         case PVCALLS_POLL:
 825                 ret = pvcalls_back_poll(dev, req);
 826                 break;
 827         default:
 828         {
 829                 struct pvcalls_fedata *fedata;
 830                 struct xen_pvcalls_response *rsp;
 831
 832                 fedata = dev_get_drvdata(&dev->dev);
 833                 rsp = RING_GET_RESPONSE(
 834                                 &fedata->ring, fedata->ring.rsp_prod_pvt++);
 835                 rsp->req_id = req->req_id;
 836                 rsp->cmd = req->cmd;
 837                 rsp->ret = -ENOTSUPP;
 838                 break;
 839         }
 840         }
 841         return ret;
 842 }
 843
 844 static void pvcalls_back_work(struct pvcalls_fedata *fedata)
 845 {
 846         int notify, notify_all = 0, more = 1;
 847         struct xen_pvcalls_request req;
 848         struct xenbus_device *dev = fedata->dev;
 849
 850         while (more) {
 851                 while (RING_HAS_UNCONSUMED_REQUESTS(&fedata->ring)) {
 852                         RING_COPY_REQUEST(&fedata->ring,
 853                                           fedata->ring.req_cons++,
 854                                           &req);
 855
 856                         if (!pvcalls_back_handle_cmd(dev, &req)) {
 857                                 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(
 858                                         &fedata->ring, notify);
 859                                 notify_all += notify;
 860                         }
 861                 }
 862
 863                 if (notify_all) {
 864                         notify_remote_via_irq(fedata->irq);
 865                         notify_all = 0;
 866                 }
 867
 868                 RING_FINAL_CHECK_FOR_REQUESTS(&fedata->ring, more);
 869         }
 870 }
 871
 872 static irqreturn_t pvcalls_back_event(int irq, void *dev_id)
 873 {
 874         struct xenbus_device *dev = dev_id;
 875         struct pvcalls_fedata *fedata = NULL;
 876
 877         if (dev == NULL)
 878                 return IRQ_HANDLED;
 879
 880         fedata = dev_get_drvdata(&dev->dev);
 881         if (fedata == NULL)
 882                 return IRQ_HANDLED;
 883
 884         pvcalls_back_work(fedata);
 885         return IRQ_HANDLED;
 886 }
 887
 888 static irqreturn_t pvcalls_back_conn_event(int irq, void *sock_map)
 889 {
 890         struct sock_mapping *map = sock_map;
 891         struct pvcalls_ioworker *iow;
 892
 893         if (map == NULL || map->sock == NULL || map->sock->sk == NULL ||
 894                 map->sock->sk->sk_user_data != map)
 895                 return IRQ_HANDLED;
 896
 897         iow = &map->ioworker;
 898
 899         atomic_inc(&map->write);
 900         atomic_inc(&map->io);
 901         queue_work(iow->wq, &iow->register_work);
 902
 903         return IRQ_HANDLED;
 904 }
 905
 906 static int backend_connect(struct xenbus_device *dev)
 907 {
 908         int err;
 909         evtchn_port_t evtchn;
 910         grant_ref_t ring_ref;
 911         struct pvcalls_fedata *fedata = NULL;
 912
 913         fedata = kzalloc(sizeof(struct pvcalls_fedata), GFP_KERNEL);
 914         if (!fedata)
 915                 return -ENOMEM;
 916
 917         fedata->irq = -1;
 918         err = xenbus_scanf(XBT_NIL, dev->otherend, "port", "%u",
 919                            &evtchn);
 920         if (err != 1) {
 921                 err = -EINVAL;
 922                 xenbus_dev_fatal(dev, err, "reading %s/event-channel",
 923                                  dev->otherend);
 924                 goto error;
 925         }
 926
 927         err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-ref", "%u", &ring_ref);
 928         if (err != 1) {
 929                 err = -EINVAL;
 930                 xenbus_dev_fatal(dev, err, "reading %s/ring-ref",
 931                                  dev->otherend);
 932                 goto error;
 933         }
 934
 935         err = bind_interdomain_evtchn_to_irq(dev->otherend_id, evtchn);
 936         if (err < 0)
 937                 goto error;
 938         fedata->irq = err;
 939
 940         err = request_threaded_irq(fedata->irq, NULL, pvcalls_back_event,
 941                                    IRQF_ONESHOT, "pvcalls-back", dev);
 942         if (err < 0)
 943                 goto error;
 944
 945         err = xenbus_map_ring_valloc(dev, &ring_ref, 1,
 946                                      (void **)&fedata->sring);
 947         if (err < 0)
 948                 goto error;
 949
 950         BACK_RING_INIT(&fedata->ring, fedata->sring, XEN_PAGE_SIZE * 1);
 951         fedata->dev = dev;
 952
 953         INIT_LIST_HEAD(&fedata->socket_mappings);
 954         INIT_RADIX_TREE(&fedata->socketpass_mappings, GFP_KERNEL);
 955         sema_init(&fedata->socket_lock, 1);
 956         dev_set_drvdata(&dev->dev, fedata);
 957
 958         down(&pvcalls_back_global.frontends_lock);
 959         list_add_tail(&fedata->list, &pvcalls_back_global.frontends);
 960         up(&pvcalls_back_global.frontends_lock);
 961
 962         return 0;
 963
 964  error:
 965         if (fedata->irq >= 0)
 966                 unbind_from_irqhandler(fedata->irq, dev);
 967         if (fedata->sring != NULL)
 968                 xenbus_unmap_ring_vfree(dev, fedata->sring);
 969         kfree(fedata);
 970         return err;
 971 }
 972
 973 static int backend_disconnect(struct xenbus_device *dev)
 974 {
 975         struct pvcalls_fedata *fedata;
 976         struct sock_mapping *map, *n;
 977         struct sockpass_mapping *mappass;
 978         struct radix_tree_iter iter;
 979         void **slot;
 980
 981
 982         fedata = dev_get_drvdata(&dev->dev);
 983
 984         down(&fedata->socket_lock);
 985         list_for_each_entry_safe(map, n, &fedata->socket_mappings, list) {
 986                 list_del(&map->list);
 987                 pvcalls_back_release_active(dev, fedata, map);
 988         }
 989
 990         radix_tree_for_each_slot(slot, &fedata->socketpass_mappings, &iter, 0) {
 991                 mappass = radix_tree_deref_slot(slot);
 992                 if (!mappass)
 993                         continue;
 994                 if (radix_tree_exception(mappass)) {
 995                         if (radix_tree_deref_retry(mappass))
 996                                 slot = radix_tree_iter_retry(&iter);
 997                 } else {
 998                         radix_tree_delete(&fedata->socketpass_mappings,
 999                                           mappass->id);
1000                         pvcalls_back_release_passive(dev, fedata, mappass);
1001                 }
1002         }
1003         up(&fedata->socket_lock);
1004
1005         unbind_from_irqhandler(fedata->irq, dev);
1006         xenbus_unmap_ring_vfree(dev, fedata->sring);
1007
1008         list_del(&fedata->list);
1009         kfree(fedata);
1010         dev_set_drvdata(&dev->dev, NULL);
1011
1012         return 0;
1013 }
1014
1015 static int pvcalls_back_probe(struct xenbus_device *dev,
1016                               const struct xenbus_device_id *id)
1017 {
1018         int err, abort;
1019         struct xenbus_transaction xbt;
1020
1021 again:
1022         abort = 1;
1023
1024         err = xenbus_transaction_start(&xbt);
1025         if (err) {
1026                 pr_warn("%s cannot create xenstore transaction\n", __func__);
1027                 return err;
1028         }
1029
1030         err = xenbus_printf(xbt, dev->nodename, "versions", "%s",
1031                             PVCALLS_VERSIONS);
1032         if (err) {
1033                 pr_warn("%s write out 'versions' failed\n", __func__);
1034                 goto abort;
1035         }
1036
1037         err = xenbus_printf(xbt, dev->nodename, "max-page-order", "%u",
1038                             MAX_RING_ORDER);
1039         if (err) {
1040                 pr_warn("%s write out 'max-page-order' failed\n", __func__);
1041                 goto abort;
1042         }
1043
1044         err = xenbus_printf(xbt, dev->nodename, "function-calls",
1045                             XENBUS_FUNCTIONS_CALLS);
1046         if (err) {
1047                 pr_warn("%s write out 'function-calls' failed\n", __func__);
1048                 goto abort;
1049         }
1050
1051         abort = 0;
1052 abort:
1053         err = xenbus_transaction_end(xbt, abort);
1054         if (err) {
1055                 if (err == -EAGAIN && !abort)
1056                         goto again;
1057                 pr_warn("%s cannot complete xenstore transaction\n", __func__);
1058                 return err;
1059         }
1060
1061         if (abort)
1062                 return -EFAULT;
1063
1064         xenbus_switch_state(dev, XenbusStateInitWait);
1065
1066         return 0;
1067 }
1068
1069 static void set_backend_state(struct xenbus_device *dev,
1070                               enum xenbus_state state)
1071 {
1072         while (dev->state != state) {
1073                 switch (dev->state) {
1074                 case XenbusStateClosed:
1075                         switch (state) {
1076                         case XenbusStateInitWait:
1077                         case XenbusStateConnected:
1078                                 xenbus_switch_state(dev, XenbusStateInitWait);
1079                                 break;
1080                         case XenbusStateClosing:
1081                                 xenbus_switch_state(dev, XenbusStateClosing);
1082                                 break;
1083                         default:
1084                                 WARN_ON(1);
1085                         }
1086                         break;
1087                 case XenbusStateInitWait:
1088                 case XenbusStateInitialised:
1089                         switch (state) {
1090                         case XenbusStateConnected:
1091                                 if (backend_connect(dev))
1092                                         return;
1093                                 xenbus_switch_state(dev, XenbusStateConnected);
1094                                 break;
1095                         case XenbusStateClosing:
1096                         case XenbusStateClosed:
1097                                 xenbus_switch_state(dev, XenbusStateClosing);
1098                                 break;
1099                         default:
1100                                 WARN_ON(1);
1101                         }
1102                         break;
1103                 case XenbusStateConnected:
1104                         switch (state) {
1105                         case XenbusStateInitWait:
1106                         case XenbusStateClosing:
1107                         case XenbusStateClosed:
1108                                 down(&pvcalls_back_global.frontends_lock);
1109                                 backend_disconnect(dev);
1110                                 up(&pvcalls_back_global.frontends_lock);
1111                                 xenbus_switch_state(dev, XenbusStateClosing);
1112                                 break;
1113                         default:
1114                                 WARN_ON(1);
1115                         }
1116                         break;
1117                 case XenbusStateClosing:
1118                         switch (state) {
1119                         case XenbusStateInitWait:
1120                         case XenbusStateConnected:
1121                         case XenbusStateClosed:
1122                                 xenbus_switch_state(dev, XenbusStateClosed);
1123                                 break;
1124                         default:
1125                                 WARN_ON(1);
1126                         }
1127                         break;
1128                 default:
1129                         WARN_ON(1);
1130                 }
1131         }
1132 }
1133
1134 static void pvcalls_back_changed(struct xenbus_device *dev,
1135                                  enum xenbus_state frontend_state)
1136 {
1137         switch (frontend_state) {
1138         case XenbusStateInitialising:
1139                 set_backend_state(dev, XenbusStateInitWait);
1140                 break;
1141
1142         case XenbusStateInitialised:
1143         case XenbusStateConnected:
1144                 set_backend_state(dev, XenbusStateConnected);
1145                 break;
1146
1147         case XenbusStateClosing:
1148                 set_backend_state(dev, XenbusStateClosing);
1149                 break;
1150
1151         case XenbusStateClosed:
1152                 set_backend_state(dev, XenbusStateClosed);
1153                 if (xenbus_dev_is_online(dev))
1154                         break;
1155                 device_unregister(&dev->dev);
1156                 break;
1157         case XenbusStateUnknown:
1158                 set_backend_state(dev, XenbusStateClosed);
1159                 device_unregister(&dev->dev);
1160                 break;
1161
1162         default:
1163                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
1164                                  frontend_state);
1165                 break;
1166         }
1167 }
1168
1169 static int pvcalls_back_remove(struct xenbus_device *dev)
1170 {
1171         return 0;
1172 }
1173
1174 static int pvcalls_back_uevent(struct xenbus_device *xdev,
1175                                struct kobj_uevent_env *env)
1176 {
1177         return 0;
1178 }
1179
1180 static const struct xenbus_device_id pvcalls_back_ids[] = {
1181         { "pvcalls" },
1182         { "" }
1183 };
1184
1185 static struct xenbus_driver pvcalls_back_driver = {
1186         .ids = pvcalls_back_ids,
1187         .probe = pvcalls_back_probe,
1188         .remove = pvcalls_back_remove,
1189         .uevent = pvcalls_back_uevent,
1190         .otherend_changed = pvcalls_back_changed,
1191 };
1192
1193 static int __init pvcalls_back_init(void)
1194 {
1195         int ret;
1196
1197         if (!xen_domain())
1198                 return -ENODEV;
1199
1200         ret = xenbus_register_backend(&pvcalls_back_driver);
1201         if (ret < 0)
1202                 return ret;
1203
1204         sema_init(&pvcalls_back_global.frontends_lock, 1);
1205         INIT_LIST_HEAD(&pvcalls_back_global.frontends);
1206         return 0;
1207 }
1208 module_init(pvcalls_back_init);
1209
1210 static void __exit pvcalls_back_fin(void)
1211 {
1212         struct pvcalls_fedata *fedata, *nfedata;
1213
1214         down(&pvcalls_back_global.frontends_lock);
1215         list_for_each_entry_safe(fedata, nfedata,
1216                                  &pvcalls_back_global.frontends, list) {
1217                 backend_disconnect(fedata->dev);
1218         }
1219         up(&pvcalls_back_global.frontends_lock);
1220
1221         xenbus_unregister_driver(&pvcalls_back_driver);
1222 }
1223
1224 module_exit(pvcalls_back_fin);
1225
1226 MODULE_DESCRIPTION("Xen PV Calls backend driver");
1227 MODULE_AUTHOR("Stefano Stabellini <sstabellini@kernel.org>");
1228 MODULE_LICENSE("GPL");