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 27static 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 */ 37struct 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 48struct pvcalls_ioworker { 49 struct work_struct register_work; 50 struct workqueue_struct *wq; 51}; 52 53struct 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 atomic_t eoi; 70 void (*saved_data_ready)(struct sock *sk); 71 struct pvcalls_ioworker ioworker; 72}; 73 74struct sockpass_mapping { 75 struct list_head list; 76 struct pvcalls_fedata *fedata; 77 struct socket *sock; 78 uint64_t id; 79 struct xen_pvcalls_request reqcopy; 80 spinlock_t copy_lock; 81 struct workqueue_struct *wq; 82 struct work_struct register_work; 83 void (*saved_data_ready)(struct sock *sk); 84}; 85 86static irqreturn_t pvcalls_back_conn_event(int irq, void *sock_map); 87static int pvcalls_back_release_active(struct xenbus_device *dev, 88 struct pvcalls_fedata *fedata, 89 struct sock_mapping *map); 90 91static bool pvcalls_conn_back_read(void *opaque) 92{ 93 struct sock_mapping *map = (struct sock_mapping *)opaque; 94 struct msghdr msg; 95 struct kvec vec[2]; 96 RING_IDX cons, prod, size, wanted, array_size, masked_prod, masked_cons; 97 int32_t error; 98 struct pvcalls_data_intf *intf = map->ring; 99 struct pvcalls_data *data = &map->data; 100 unsigned long flags; 101 int ret; 102 103 array_size = XEN_FLEX_RING_SIZE(map->ring_order); 104 cons = intf->in_cons; 105 prod = intf->in_prod; 106 error = intf->in_error; 107 /* read the indexes first, then deal with the data */ 108 virt_mb(); 109 110 if (error) 111 return false; 112 113 size = pvcalls_queued(prod, cons, array_size); 114 if (size >= array_size) 115 return false; 116 spin_lock_irqsave(&map->sock->sk->sk_receive_queue.lock, flags); 117 if (skb_queue_empty(&map->sock->sk->sk_receive_queue)) { 118 atomic_set(&map->read, 0); 119 spin_unlock_irqrestore(&map->sock->sk->sk_receive_queue.lock, 120 flags); 121 return true; 122 } 123 spin_unlock_irqrestore(&map->sock->sk->sk_receive_queue.lock, flags); 124 wanted = array_size - size; 125 masked_prod = pvcalls_mask(prod, array_size); 126 masked_cons = pvcalls_mask(cons, array_size); 127 128 memset(&msg, 0, sizeof(msg)); 129 if (masked_prod < masked_cons) { 130 vec[0].iov_base = data->in + masked_prod; 131 vec[0].iov_len = wanted; 132 iov_iter_kvec(&msg.msg_iter, READ, vec, 1, wanted); 133 } else { 134 vec[0].iov_base = data->in + masked_prod; 135 vec[0].iov_len = array_size - masked_prod; 136 vec[1].iov_base = data->in; 137 vec[1].iov_len = wanted - vec[0].iov_len; 138 iov_iter_kvec(&msg.msg_iter, READ, vec, 2, wanted); 139 } 140 141 atomic_set(&map->read, 0); 142 ret = inet_recvmsg(map->sock, &msg, wanted, MSG_DONTWAIT); 143 WARN_ON(ret > wanted); 144 if (ret == -EAGAIN) /* shouldn't happen */ 145 return true; 146 if (!ret) 147 ret = -ENOTCONN; 148 spin_lock_irqsave(&map->sock->sk->sk_receive_queue.lock, flags); 149 if (ret > 0 && !skb_queue_empty(&map->sock->sk->sk_receive_queue)) 150 atomic_inc(&map->read); 151 spin_unlock_irqrestore(&map->sock->sk->sk_receive_queue.lock, flags); 152 153 /* write the data, then modify the indexes */ 154 virt_wmb(); 155 if (ret < 0) { 156 atomic_set(&map->read, 0); 157 intf->in_error = ret; 158 } else 159 intf->in_prod = prod + ret; 160 /* update the indexes, then notify the other end */ 161 virt_wmb(); 162 notify_remote_via_irq(map->irq); 163 164 return true; 165} 166 167static bool pvcalls_conn_back_write(struct sock_mapping *map) 168{ 169 struct pvcalls_data_intf *intf = map->ring; 170 struct pvcalls_data *data = &map->data; 171 struct msghdr msg; 172 struct kvec vec[2]; 173 RING_IDX cons, prod, size, array_size; 174 int ret; 175 176 cons = intf->out_cons; 177 prod = intf->out_prod; 178 /* read the indexes before dealing with the data */ 179 virt_mb(); 180 181 array_size = XEN_FLEX_RING_SIZE(map->ring_order); 182 size = pvcalls_queued(prod, cons, array_size); 183 if (size == 0) 184 return false; 185 186 memset(&msg, 0, sizeof(msg)); 187 msg.msg_flags |= MSG_DONTWAIT; 188 if (pvcalls_mask(prod, array_size) > pvcalls_mask(cons, array_size)) { 189 vec[0].iov_base = data->out + pvcalls_mask(cons, array_size); 190 vec[0].iov_len = size; 191 iov_iter_kvec(&msg.msg_iter, WRITE, vec, 1, size); 192 } else { 193 vec[0].iov_base = data->out + pvcalls_mask(cons, array_size); 194 vec[0].iov_len = array_size - pvcalls_mask(cons, array_size); 195 vec[1].iov_base = data->out; 196 vec[1].iov_len = size - vec[0].iov_len; 197 iov_iter_kvec(&msg.msg_iter, WRITE, vec, 2, size); 198 } 199 200 atomic_set(&map->write, 0); 201 ret = inet_sendmsg(map->sock, &msg, size); 202 if (ret == -EAGAIN) { 203 atomic_inc(&map->write); 204 atomic_inc(&map->io); 205 return true; 206 } 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 atomic_inc(&map->io); 222 } 223 notify_remote_via_irq(map->irq); 224 225 return true; 226} 227 228static void pvcalls_back_ioworker(struct work_struct *work) 229{ 230 struct pvcalls_ioworker *ioworker = container_of(work, 231 struct pvcalls_ioworker, register_work); 232 struct sock_mapping *map = container_of(ioworker, struct sock_mapping, 233 ioworker); 234 unsigned int eoi_flags = XEN_EOI_FLAG_SPURIOUS; 235 236 while (atomic_read(&map->io) > 0) { 237 if (atomic_read(&map->release) > 0) { 238 atomic_set(&map->release, 0); 239 return; 240 } 241 242 if (atomic_read(&map->read) > 0 && 243 pvcalls_conn_back_read(map)) 244 eoi_flags = 0; 245 if (atomic_read(&map->write) > 0 && 246 pvcalls_conn_back_write(map)) 247 eoi_flags = 0; 248 249 if (atomic_read(&map->eoi) > 0 && !atomic_read(&map->write)) { 250 atomic_set(&map->eoi, 0); 251 xen_irq_lateeoi(map->irq, eoi_flags); 252 eoi_flags = XEN_EOI_FLAG_SPURIOUS; 253 } 254 255 atomic_dec(&map->io); 256 } 257} 258 259static int pvcalls_back_socket(struct xenbus_device *dev, 260 struct xen_pvcalls_request *req) 261{ 262 struct pvcalls_fedata *fedata; 263 int ret; 264 struct xen_pvcalls_response *rsp; 265 266 fedata = dev_get_drvdata(&dev->dev); 267 268 if (req->u.socket.domain != AF_INET || 269 req->u.socket.type != SOCK_STREAM || 270 (req->u.socket.protocol != IPPROTO_IP && 271 req->u.socket.protocol != AF_INET)) 272 ret = -EAFNOSUPPORT; 273 else 274 ret = 0; 275 276 /* leave the actual socket allocation for later */ 277 278 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++); 279 rsp->req_id = req->req_id; 280 rsp->cmd = req->cmd; 281 rsp->u.socket.id = req->u.socket.id; 282 rsp->ret = ret; 283 284 return 0; 285} 286 287static void pvcalls_sk_state_change(struct sock *sock) 288{ 289 struct sock_mapping *map = sock->sk_user_data; 290 291 if (map == NULL) 292 return; 293 294 atomic_inc(&map->read); 295 notify_remote_via_irq(map->irq); 296} 297 298static void pvcalls_sk_data_ready(struct sock *sock) 299{ 300 struct sock_mapping *map = sock->sk_user_data; 301 struct pvcalls_ioworker *iow; 302 303 if (map == NULL) 304 return; 305 306 iow = &map->ioworker; 307 atomic_inc(&map->read); 308 atomic_inc(&map->io); 309 queue_work(iow->wq, &iow->register_work); 310} 311 312static struct sock_mapping *pvcalls_new_active_socket( 313 struct pvcalls_fedata *fedata, 314 uint64_t id, 315 grant_ref_t ref, 316 evtchn_port_t evtchn, 317 struct socket *sock) 318{ 319 int ret; 320 struct sock_mapping *map; 321 void *page; 322 323 map = kzalloc(sizeof(*map), GFP_KERNEL); 324 if (map == NULL) { 325 sock_release(sock); 326 return NULL; 327 } 328 329 map->fedata = fedata; 330 map->sock = sock; 331 map->id = id; 332 map->ref = ref; 333 334 ret = xenbus_map_ring_valloc(fedata->dev, &ref, 1, &page); 335 if (ret < 0) 336 goto out; 337 map->ring = page; 338 map->ring_order = map->ring->ring_order; 339 /* first read the order, then map the data ring */ 340 virt_rmb(); 341 if (map->ring_order > MAX_RING_ORDER) { 342 pr_warn("%s frontend requested ring_order %u, which is > MAX (%u)\n", 343 __func__, map->ring_order, MAX_RING_ORDER); 344 goto out; 345 } 346 ret = xenbus_map_ring_valloc(fedata->dev, map->ring->ref, 347 (1 << map->ring_order), &page); 348 if (ret < 0) 349 goto out; 350 map->bytes = page; 351 352 ret = bind_interdomain_evtchn_to_irqhandler_lateeoi( 353 fedata->dev->otherend_id, evtchn, 354 pvcalls_back_conn_event, 0, "pvcalls-backend", map); 355 if (ret < 0) 356 goto out; 357 map->irq = ret; 358 359 map->data.in = map->bytes; 360 map->data.out = map->bytes + XEN_FLEX_RING_SIZE(map->ring_order); 361 362 map->ioworker.wq = alloc_workqueue("pvcalls_io", WQ_UNBOUND, 1); 363 if (!map->ioworker.wq) 364 goto out; 365 atomic_set(&map->io, 1); 366 INIT_WORK(&map->ioworker.register_work, pvcalls_back_ioworker); 367 368 down(&fedata->socket_lock); 369 list_add_tail(&map->list, &fedata->socket_mappings); 370 up(&fedata->socket_lock); 371 372 write_lock_bh(&map->sock->sk->sk_callback_lock); 373 map->saved_data_ready = map->sock->sk->sk_data_ready; 374 map->sock->sk->sk_user_data = map; 375 map->sock->sk->sk_data_ready = pvcalls_sk_data_ready; 376 map->sock->sk->sk_state_change = pvcalls_sk_state_change; 377 write_unlock_bh(&map->sock->sk->sk_callback_lock); 378 379 return map; 380out: 381 down(&fedata->socket_lock); 382 list_del(&map->list); 383 pvcalls_back_release_active(fedata->dev, fedata, map); 384 up(&fedata->socket_lock); 385 return NULL; 386} 387 388static int pvcalls_back_connect(struct xenbus_device *dev, 389 struct xen_pvcalls_request *req) 390{ 391 struct pvcalls_fedata *fedata; 392 int ret = -EINVAL; 393 struct socket *sock; 394 struct sock_mapping *map; 395 struct xen_pvcalls_response *rsp; 396 struct sockaddr *sa = (struct sockaddr *)&req->u.connect.addr; 397 398 fedata = dev_get_drvdata(&dev->dev); 399 400 if (req->u.connect.len < sizeof(sa->sa_family) || 401 req->u.connect.len > sizeof(req->u.connect.addr) || 402 sa->sa_family != AF_INET) 403 goto out; 404 405 ret = sock_create(AF_INET, SOCK_STREAM, 0, &sock); 406 if (ret < 0) 407 goto out; 408 ret = inet_stream_connect(sock, sa, req->u.connect.len, 0); 409 if (ret < 0) { 410 sock_release(sock); 411 goto out; 412 } 413 414 map = pvcalls_new_active_socket(fedata, 415 req->u.connect.id, 416 req->u.connect.ref, 417 req->u.connect.evtchn, 418 sock); 419 if (!map) 420 ret = -EFAULT; 421 422out: 423 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++); 424 rsp->req_id = req->req_id; 425 rsp->cmd = req->cmd; 426 rsp->u.connect.id = req->u.connect.id; 427 rsp->ret = ret; 428 429 return 0; 430} 431 432static int pvcalls_back_release_active(struct xenbus_device *dev, 433 struct pvcalls_fedata *fedata, 434 struct sock_mapping *map) 435{ 436 disable_irq(map->irq); 437 if (map->sock->sk != NULL) { 438 write_lock_bh(&map->sock->sk->sk_callback_lock); 439 map->sock->sk->sk_user_data = NULL; 440 map->sock->sk->sk_data_ready = map->saved_data_ready; 441 write_unlock_bh(&map->sock->sk->sk_callback_lock); 442 } 443 444 atomic_set(&map->release, 1); 445 flush_work(&map->ioworker.register_work); 446 447 xenbus_unmap_ring_vfree(dev, map->bytes); 448 xenbus_unmap_ring_vfree(dev, (void *)map->ring); 449 unbind_from_irqhandler(map->irq, map); 450 451 sock_release(map->sock); 452 kfree(map); 453 454 return 0; 455} 456 457static int pvcalls_back_release_passive(struct xenbus_device *dev, 458 struct pvcalls_fedata *fedata, 459 struct sockpass_mapping *mappass) 460{ 461 if (mappass->sock->sk != NULL) { 462 write_lock_bh(&mappass->sock->sk->sk_callback_lock); 463 mappass->sock->sk->sk_user_data = NULL; 464 mappass->sock->sk->sk_data_ready = mappass->saved_data_ready; 465 write_unlock_bh(&mappass->sock->sk->sk_callback_lock); 466 } 467 sock_release(mappass->sock); 468 flush_workqueue(mappass->wq); 469 destroy_workqueue(mappass->wq); 470 kfree(mappass); 471 472 return 0; 473} 474 475static int pvcalls_back_release(struct xenbus_device *dev, 476 struct xen_pvcalls_request *req) 477{ 478 struct pvcalls_fedata *fedata; 479 struct sock_mapping *map, *n; 480 struct sockpass_mapping *mappass; 481 int ret = 0; 482 struct xen_pvcalls_response *rsp; 483 484 fedata = dev_get_drvdata(&dev->dev); 485 486 down(&fedata->socket_lock); 487 list_for_each_entry_safe(map, n, &fedata->socket_mappings, list) { 488 if (map->id == req->u.release.id) { 489 list_del(&map->list); 490 up(&fedata->socket_lock); 491 ret = pvcalls_back_release_active(dev, fedata, map); 492 goto out; 493 } 494 } 495 mappass = radix_tree_lookup(&fedata->socketpass_mappings, 496 req->u.release.id); 497 if (mappass != NULL) { 498 radix_tree_delete(&fedata->socketpass_mappings, mappass->id); 499 up(&fedata->socket_lock); 500 ret = pvcalls_back_release_passive(dev, fedata, mappass); 501 } else 502 up(&fedata->socket_lock); 503 504out: 505 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++); 506 rsp->req_id = req->req_id; 507 rsp->u.release.id = req->u.release.id; 508 rsp->cmd = req->cmd; 509 rsp->ret = ret; 510 return 0; 511} 512 513static void __pvcalls_back_accept(struct work_struct *work) 514{ 515 struct sockpass_mapping *mappass = container_of( 516 work, struct sockpass_mapping, register_work); 517 struct sock_mapping *map; 518 struct pvcalls_ioworker *iow; 519 struct pvcalls_fedata *fedata; 520 struct socket *sock; 521 struct xen_pvcalls_response *rsp; 522 struct xen_pvcalls_request *req; 523 int notify; 524 int ret = -EINVAL; 525 unsigned long flags; 526 527 fedata = mappass->fedata; 528 /* 529 * __pvcalls_back_accept can race against pvcalls_back_accept. 530 * We only need to check the value of "cmd" on read. It could be 531 * done atomically, but to simplify the code on the write side, we 532 * use a spinlock. 533 */ 534 spin_lock_irqsave(&mappass->copy_lock, flags); 535 req = &mappass->reqcopy; 536 if (req->cmd != PVCALLS_ACCEPT) { 537 spin_unlock_irqrestore(&mappass->copy_lock, flags); 538 return; 539 } 540 spin_unlock_irqrestore(&mappass->copy_lock, flags); 541 542 sock = sock_alloc(); 543 if (sock == NULL) 544 goto out_error; 545 sock->type = mappass->sock->type; 546 sock->ops = mappass->sock->ops; 547 548 ret = inet_accept(mappass->sock, sock, O_NONBLOCK, true); 549 if (ret == -EAGAIN) { 550 sock_release(sock); 551 return; 552 } 553 554 map = pvcalls_new_active_socket(fedata, 555 req->u.accept.id_new, 556 req->u.accept.ref, 557 req->u.accept.evtchn, 558 sock); 559 if (!map) { 560 ret = -EFAULT; 561 goto out_error; 562 } 563 564 map->sockpass = mappass; 565 iow = &map->ioworker; 566 atomic_inc(&map->read); 567 atomic_inc(&map->io); 568 queue_work(iow->wq, &iow->register_work); 569 570out_error: 571 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++); 572 rsp->req_id = req->req_id; 573 rsp->cmd = req->cmd; 574 rsp->u.accept.id = req->u.accept.id; 575 rsp->ret = ret; 576 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&fedata->ring, notify); 577 if (notify) 578 notify_remote_via_irq(fedata->irq); 579 580 mappass->reqcopy.cmd = 0; 581} 582 583static void pvcalls_pass_sk_data_ready(struct sock *sock) 584{ 585 struct sockpass_mapping *mappass = sock->sk_user_data; 586 struct pvcalls_fedata *fedata; 587 struct xen_pvcalls_response *rsp; 588 unsigned long flags; 589 int notify; 590 591 if (mappass == NULL) 592 return; 593 594 fedata = mappass->fedata; 595 spin_lock_irqsave(&mappass->copy_lock, flags); 596 if (mappass->reqcopy.cmd == PVCALLS_POLL) { 597 rsp = RING_GET_RESPONSE(&fedata->ring, 598 fedata->ring.rsp_prod_pvt++); 599 rsp->req_id = mappass->reqcopy.req_id; 600 rsp->u.poll.id = mappass->reqcopy.u.poll.id; 601 rsp->cmd = mappass->reqcopy.cmd; 602 rsp->ret = 0; 603 604 mappass->reqcopy.cmd = 0; 605 spin_unlock_irqrestore(&mappass->copy_lock, flags); 606 607 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&fedata->ring, notify); 608 if (notify) 609 notify_remote_via_irq(mappass->fedata->irq); 610 } else { 611 spin_unlock_irqrestore(&mappass->copy_lock, flags); 612 queue_work(mappass->wq, &mappass->register_work); 613 } 614} 615 616static int pvcalls_back_bind(struct xenbus_device *dev, 617 struct xen_pvcalls_request *req) 618{ 619 struct pvcalls_fedata *fedata; 620 int ret; 621 struct sockpass_mapping *map; 622 struct xen_pvcalls_response *rsp; 623 624 fedata = dev_get_drvdata(&dev->dev); 625 626 map = kzalloc(sizeof(*map), GFP_KERNEL); 627 if (map == NULL) { 628 ret = -ENOMEM; 629 goto out; 630 } 631 632 INIT_WORK(&map->register_work, __pvcalls_back_accept); 633 spin_lock_init(&map->copy_lock); 634 map->wq = alloc_workqueue("pvcalls_wq", WQ_UNBOUND, 1); 635 if (!map->wq) { 636 ret = -ENOMEM; 637 goto out; 638 } 639 640 ret = sock_create(AF_INET, SOCK_STREAM, 0, &map->sock); 641 if (ret < 0) 642 goto out; 643 644 ret = inet_bind(map->sock, (struct sockaddr *)&req->u.bind.addr, 645 req->u.bind.len); 646 if (ret < 0) 647 goto out; 648 649 map->fedata = fedata; 650 map->id = req->u.bind.id; 651 652 down(&fedata->socket_lock); 653 ret = radix_tree_insert(&fedata->socketpass_mappings, map->id, 654 map); 655 up(&fedata->socket_lock); 656 if (ret) 657 goto out; 658 659 write_lock_bh(&map->sock->sk->sk_callback_lock); 660 map->saved_data_ready = map->sock->sk->sk_data_ready; 661 map->sock->sk->sk_user_data = map; 662 map->sock->sk->sk_data_ready = pvcalls_pass_sk_data_ready; 663 write_unlock_bh(&map->sock->sk->sk_callback_lock); 664 665out: 666 if (ret) { 667 if (map && map->sock) 668 sock_release(map->sock); 669 if (map && map->wq) 670 destroy_workqueue(map->wq); 671 kfree(map); 672 } 673 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++); 674 rsp->req_id = req->req_id; 675 rsp->cmd = req->cmd; 676 rsp->u.bind.id = req->u.bind.id; 677 rsp->ret = ret; 678 return 0; 679} 680 681static int pvcalls_back_listen(struct xenbus_device *dev, 682 struct xen_pvcalls_request *req) 683{ 684 struct pvcalls_fedata *fedata; 685 int ret = -EINVAL; 686 struct sockpass_mapping *map; 687 struct xen_pvcalls_response *rsp; 688 689 fedata = dev_get_drvdata(&dev->dev); 690 691 down(&fedata->socket_lock); 692 map = radix_tree_lookup(&fedata->socketpass_mappings, req->u.listen.id); 693 up(&fedata->socket_lock); 694 if (map == NULL) 695 goto out; 696 697 ret = inet_listen(map->sock, req->u.listen.backlog); 698 699out: 700 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++); 701 rsp->req_id = req->req_id; 702 rsp->cmd = req->cmd; 703 rsp->u.listen.id = req->u.listen.id; 704 rsp->ret = ret; 705 return 0; 706} 707 708static int pvcalls_back_accept(struct xenbus_device *dev, 709 struct xen_pvcalls_request *req) 710{ 711 struct pvcalls_fedata *fedata; 712 struct sockpass_mapping *mappass; 713 int ret = -EINVAL; 714 struct xen_pvcalls_response *rsp; 715 unsigned long flags; 716 717 fedata = dev_get_drvdata(&dev->dev); 718 719 down(&fedata->socket_lock); 720 mappass = radix_tree_lookup(&fedata->socketpass_mappings, 721 req->u.accept.id); 722 up(&fedata->socket_lock); 723 if (mappass == NULL) 724 goto out_error; 725 726 /* 727 * Limitation of the current implementation: only support one 728 * concurrent accept or poll call on one socket. 729 */ 730 spin_lock_irqsave(&mappass->copy_lock, flags); 731 if (mappass->reqcopy.cmd != 0) { 732 spin_unlock_irqrestore(&mappass->copy_lock, flags); 733 ret = -EINTR; 734 goto out_error; 735 } 736 737 mappass->reqcopy = *req; 738 spin_unlock_irqrestore(&mappass->copy_lock, flags); 739 queue_work(mappass->wq, &mappass->register_work); 740 741 /* Tell the caller we don't need to send back a notification yet */ 742 return -1; 743 744out_error: 745 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++); 746 rsp->req_id = req->req_id; 747 rsp->cmd = req->cmd; 748 rsp->u.accept.id = req->u.accept.id; 749 rsp->ret = ret; 750 return 0; 751} 752 753static int pvcalls_back_poll(struct xenbus_device *dev, 754 struct xen_pvcalls_request *req) 755{ 756 struct pvcalls_fedata *fedata; 757 struct sockpass_mapping *mappass; 758 struct xen_pvcalls_response *rsp; 759 struct inet_connection_sock *icsk; 760 struct request_sock_queue *queue; 761 unsigned long flags; 762 int ret; 763 bool data; 764 765 fedata = dev_get_drvdata(&dev->dev); 766 767 down(&fedata->socket_lock); 768 mappass = radix_tree_lookup(&fedata->socketpass_mappings, 769 req->u.poll.id); 770 up(&fedata->socket_lock); 771 if (mappass == NULL) 772 return -EINVAL; 773 774 /* 775 * Limitation of the current implementation: only support one 776 * concurrent accept or poll call on one socket. 777 */ 778 spin_lock_irqsave(&mappass->copy_lock, flags); 779 if (mappass->reqcopy.cmd != 0) { 780 ret = -EINTR; 781 goto out; 782 } 783 784 mappass->reqcopy = *req; 785 icsk = inet_csk(mappass->sock->sk); 786 queue = &icsk->icsk_accept_queue; 787 data = READ_ONCE(queue->rskq_accept_head) != NULL; 788 if (data) { 789 mappass->reqcopy.cmd = 0; 790 ret = 0; 791 goto out; 792 } 793 spin_unlock_irqrestore(&mappass->copy_lock, flags); 794 795 /* Tell the caller we don't need to send back a notification yet */ 796 return -1; 797 798out: 799 spin_unlock_irqrestore(&mappass->copy_lock, flags); 800 801 rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++); 802 rsp->req_id = req->req_id; 803 rsp->cmd = req->cmd; 804 rsp->u.poll.id = req->u.poll.id; 805 rsp->ret = ret; 806 return 0; 807} 808 809static int pvcalls_back_handle_cmd(struct xenbus_device *dev, 810 struct xen_pvcalls_request *req) 811{ 812 int ret = 0; 813 814 switch (req->cmd) { 815 case PVCALLS_SOCKET: 816 ret = pvcalls_back_socket(dev, req); 817 break; 818 case PVCALLS_CONNECT: 819 ret = pvcalls_back_connect(dev, req); 820 break; 821 case PVCALLS_RELEASE: 822 ret = pvcalls_back_release(dev, req); 823 break; 824 case PVCALLS_BIND: 825 ret = pvcalls_back_bind(dev, req); 826 break; 827 case PVCALLS_LISTEN: 828 ret = pvcalls_back_listen(dev, req); 829 break; 830 case PVCALLS_ACCEPT: 831 ret = pvcalls_back_accept(dev, req); 832 break; 833 case PVCALLS_POLL: 834 ret = pvcalls_back_poll(dev, req); 835 break; 836 default: 837 { 838 struct pvcalls_fedata *fedata; 839 struct xen_pvcalls_response *rsp; 840 841 fedata = dev_get_drvdata(&dev->dev); 842 rsp = RING_GET_RESPONSE( 843 &fedata->ring, fedata->ring.rsp_prod_pvt++); 844 rsp->req_id = req->req_id; 845 rsp->cmd = req->cmd; 846 rsp->ret = -ENOTSUPP; 847 break; 848 } 849 } 850 return ret; 851} 852 853static void pvcalls_back_work(struct pvcalls_fedata *fedata) 854{ 855 int notify, notify_all = 0, more = 1; 856 struct xen_pvcalls_request req; 857 struct xenbus_device *dev = fedata->dev; 858 859 while (more) { 860 while (RING_HAS_UNCONSUMED_REQUESTS(&fedata->ring)) { 861 RING_COPY_REQUEST(&fedata->ring, 862 fedata->ring.req_cons++, 863 &req); 864 865 if (!pvcalls_back_handle_cmd(dev, &req)) { 866 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY( 867 &fedata->ring, notify); 868 notify_all += notify; 869 } 870 } 871 872 if (notify_all) { 873 notify_remote_via_irq(fedata->irq); 874 notify_all = 0; 875 } 876 877 RING_FINAL_CHECK_FOR_REQUESTS(&fedata->ring, more); 878 } 879} 880 881static irqreturn_t pvcalls_back_event(int irq, void *dev_id) 882{ 883 struct xenbus_device *dev = dev_id; 884 struct pvcalls_fedata *fedata = NULL; 885 unsigned int eoi_flags = XEN_EOI_FLAG_SPURIOUS; 886 887 if (dev) { 888 fedata = dev_get_drvdata(&dev->dev); 889 if (fedata) { 890 pvcalls_back_work(fedata); 891 eoi_flags = 0; 892 } 893 } 894 895 xen_irq_lateeoi(irq, eoi_flags); 896 897 return IRQ_HANDLED; 898} 899 900static irqreturn_t pvcalls_back_conn_event(int irq, void *sock_map) 901{ 902 struct sock_mapping *map = sock_map; 903 struct pvcalls_ioworker *iow; 904 905 if (map == NULL || map->sock == NULL || map->sock->sk == NULL || 906 map->sock->sk->sk_user_data != map) { 907 xen_irq_lateeoi(irq, 0); 908 return IRQ_HANDLED; 909 } 910 911 iow = &map->ioworker; 912 913 atomic_inc(&map->write); 914 atomic_inc(&map->eoi); 915 atomic_inc(&map->io); 916 queue_work(iow->wq, &iow->register_work); 917 918 return IRQ_HANDLED; 919} 920 921static int backend_connect(struct xenbus_device *dev) 922{ 923 int err; 924 evtchn_port_t evtchn; 925 grant_ref_t ring_ref; 926 struct pvcalls_fedata *fedata = NULL; 927 928 fedata = kzalloc(sizeof(struct pvcalls_fedata), GFP_KERNEL); 929 if (!fedata) 930 return -ENOMEM; 931 932 fedata->irq = -1; 933 err = xenbus_scanf(XBT_NIL, dev->otherend, "port", "%u", 934 &evtchn); 935 if (err != 1) { 936 err = -EINVAL; 937 xenbus_dev_fatal(dev, err, "reading %s/event-channel", 938 dev->otherend); 939 goto error; 940 } 941 942 err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-ref", "%u", &ring_ref); 943 if (err != 1) { 944 err = -EINVAL; 945 xenbus_dev_fatal(dev, err, "reading %s/ring-ref", 946 dev->otherend); 947 goto error; 948 } 949 950 err = bind_interdomain_evtchn_to_irq_lateeoi(dev->otherend_id, evtchn); 951 if (err < 0) 952 goto error; 953 fedata->irq = err; 954 955 err = request_threaded_irq(fedata->irq, NULL, pvcalls_back_event, 956 IRQF_ONESHOT, "pvcalls-back", dev); 957 if (err < 0) 958 goto error; 959 960 err = xenbus_map_ring_valloc(dev, &ring_ref, 1, 961 (void **)&fedata->sring); 962 if (err < 0) 963 goto error; 964 965 BACK_RING_INIT(&fedata->ring, fedata->sring, XEN_PAGE_SIZE * 1); 966 fedata->dev = dev; 967 968 INIT_LIST_HEAD(&fedata->socket_mappings); 969 INIT_RADIX_TREE(&fedata->socketpass_mappings, GFP_KERNEL); 970 sema_init(&fedata->socket_lock, 1); 971 dev_set_drvdata(&dev->dev, fedata); 972 973 down(&pvcalls_back_global.frontends_lock); 974 list_add_tail(&fedata->list, &pvcalls_back_global.frontends); 975 up(&pvcalls_back_global.frontends_lock); 976 977 return 0; 978 979 error: 980 if (fedata->irq >= 0) 981 unbind_from_irqhandler(fedata->irq, dev); 982 if (fedata->sring != NULL) 983 xenbus_unmap_ring_vfree(dev, fedata->sring); 984 kfree(fedata); 985 return err; 986} 987 988static int backend_disconnect(struct xenbus_device *dev) 989{ 990 struct pvcalls_fedata *fedata; 991 struct sock_mapping *map, *n; 992 struct sockpass_mapping *mappass; 993 struct radix_tree_iter iter; 994 void **slot; 995 996 997 fedata = dev_get_drvdata(&dev->dev); 998 999 down(&fedata->socket_lock); 1000 list_for_each_entry_safe(map, n, &fedata->socket_mappings, list) { 1001 list_del(&map->list); 1002 pvcalls_back_release_active(dev, fedata, map); 1003 } 1004 1005 radix_tree_for_each_slot(slot, &fedata->socketpass_mappings, &iter, 0) { 1006 mappass = radix_tree_deref_slot(slot); 1007 if (!mappass) 1008 continue; 1009 if (radix_tree_exception(mappass)) { 1010 if (radix_tree_deref_retry(mappass)) 1011 slot = radix_tree_iter_retry(&iter); 1012 } else { 1013 radix_tree_delete(&fedata->socketpass_mappings, 1014 mappass->id); 1015 pvcalls_back_release_passive(dev, fedata, mappass); 1016 } 1017 } 1018 up(&fedata->socket_lock); 1019 1020 unbind_from_irqhandler(fedata->irq, dev); 1021 xenbus_unmap_ring_vfree(dev, fedata->sring); 1022 1023 list_del(&fedata->list); 1024 kfree(fedata); 1025 dev_set_drvdata(&dev->dev, NULL); 1026 1027 return 0; 1028} 1029 1030static int pvcalls_back_probe(struct xenbus_device *dev, 1031 const struct xenbus_device_id *id) 1032{ 1033 int err, abort; 1034 struct xenbus_transaction xbt; 1035 1036again: 1037 abort = 1; 1038 1039 err = xenbus_transaction_start(&xbt); 1040 if (err) { 1041 pr_warn("%s cannot create xenstore transaction\n", __func__); 1042 return err; 1043 } 1044 1045 err = xenbus_printf(xbt, dev->nodename, "versions", "%s", 1046 PVCALLS_VERSIONS); 1047 if (err) { 1048 pr_warn("%s write out 'versions' failed\n", __func__); 1049 goto abort; 1050 } 1051 1052 err = xenbus_printf(xbt, dev->nodename, "max-page-order", "%u", 1053 MAX_RING_ORDER); 1054 if (err) { 1055 pr_warn("%s write out 'max-page-order' failed\n", __func__); 1056 goto abort; 1057 } 1058 1059 err = xenbus_printf(xbt, dev->nodename, "function-calls", 1060 XENBUS_FUNCTIONS_CALLS); 1061 if (err) { 1062 pr_warn("%s write out 'function-calls' failed\n", __func__); 1063 goto abort; 1064 } 1065 1066 abort = 0; 1067abort: 1068 err = xenbus_transaction_end(xbt, abort); 1069 if (err) { 1070 if (err == -EAGAIN && !abort) 1071 goto again; 1072 pr_warn("%s cannot complete xenstore transaction\n", __func__); 1073 return err; 1074 } 1075 1076 if (abort) 1077 return -EFAULT; 1078 1079 xenbus_switch_state(dev, XenbusStateInitWait); 1080 1081 return 0; 1082} 1083 1084static void set_backend_state(struct xenbus_device *dev, 1085 enum xenbus_state state) 1086{ 1087 while (dev->state != state) { 1088 switch (dev->state) { 1089 case XenbusStateClosed: 1090 switch (state) { 1091 case XenbusStateInitWait: 1092 case XenbusStateConnected: 1093 xenbus_switch_state(dev, XenbusStateInitWait); 1094 break; 1095 case XenbusStateClosing: 1096 xenbus_switch_state(dev, XenbusStateClosing); 1097 break; 1098 default: 1099 WARN_ON(1); 1100 } 1101 break; 1102 case XenbusStateInitWait: 1103 case XenbusStateInitialised: 1104 switch (state) { 1105 case XenbusStateConnected: 1106 if (backend_connect(dev)) 1107 return; 1108 xenbus_switch_state(dev, XenbusStateConnected); 1109 break; 1110 case XenbusStateClosing: 1111 case XenbusStateClosed: 1112 xenbus_switch_state(dev, XenbusStateClosing); 1113 break; 1114 default: 1115 WARN_ON(1); 1116 } 1117 break; 1118 case XenbusStateConnected: 1119 switch (state) { 1120 case XenbusStateInitWait: 1121 case XenbusStateClosing: 1122 case XenbusStateClosed: 1123 down(&pvcalls_back_global.frontends_lock); 1124 backend_disconnect(dev); 1125 up(&pvcalls_back_global.frontends_lock); 1126 xenbus_switch_state(dev, XenbusStateClosing); 1127 break; 1128 default: 1129 WARN_ON(1); 1130 } 1131 break; 1132 case XenbusStateClosing: 1133 switch (state) { 1134 case XenbusStateInitWait: 1135 case XenbusStateConnected: 1136 case XenbusStateClosed: 1137 xenbus_switch_state(dev, XenbusStateClosed); 1138 break; 1139 default: 1140 WARN_ON(1); 1141 } 1142 break; 1143 default: 1144 WARN_ON(1); 1145 } 1146 } 1147} 1148 1149static void pvcalls_back_changed(struct xenbus_device *dev, 1150 enum xenbus_state frontend_state) 1151{ 1152 switch (frontend_state) { 1153 case XenbusStateInitialising: 1154 set_backend_state(dev, XenbusStateInitWait); 1155 break; 1156 1157 case XenbusStateInitialised: 1158 case XenbusStateConnected: 1159 set_backend_state(dev, XenbusStateConnected); 1160 break; 1161 1162 case XenbusStateClosing: 1163 set_backend_state(dev, XenbusStateClosing); 1164 break; 1165 1166 case XenbusStateClosed: 1167 set_backend_state(dev, XenbusStateClosed); 1168 if (xenbus_dev_is_online(dev)) 1169 break; 1170 device_unregister(&dev->dev); 1171 break; 1172 case XenbusStateUnknown: 1173 set_backend_state(dev, XenbusStateClosed); 1174 device_unregister(&dev->dev); 1175 break; 1176 1177 default: 1178 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend", 1179 frontend_state); 1180 break; 1181 } 1182} 1183 1184static int pvcalls_back_remove(struct xenbus_device *dev) 1185{ 1186 return 0; 1187} 1188 1189static int pvcalls_back_uevent(struct xenbus_device *xdev, 1190 struct kobj_uevent_env *env) 1191{ 1192 return 0; 1193} 1194 1195static const struct xenbus_device_id pvcalls_back_ids[] = { 1196 { "pvcalls" }, 1197 { "" } 1198}; 1199 1200static struct xenbus_driver pvcalls_back_driver = { 1201 .ids = pvcalls_back_ids, 1202 .probe = pvcalls_back_probe, 1203 .remove = pvcalls_back_remove, 1204 .uevent = pvcalls_back_uevent, 1205 .otherend_changed = pvcalls_back_changed, 1206}; 1207 1208static int __init pvcalls_back_init(void) 1209{ 1210 int ret; 1211 1212 if (!xen_domain()) 1213 return -ENODEV; 1214 1215 ret = xenbus_register_backend(&pvcalls_back_driver); 1216 if (ret < 0) 1217 return ret; 1218 1219 sema_init(&pvcalls_back_global.frontends_lock, 1); 1220 INIT_LIST_HEAD(&pvcalls_back_global.frontends); 1221 return 0; 1222} 1223module_init(pvcalls_back_init); 1224 1225static void __exit pvcalls_back_fin(void) 1226{ 1227 struct pvcalls_fedata *fedata, *nfedata; 1228 1229 down(&pvcalls_back_global.frontends_lock); 1230 list_for_each_entry_safe(fedata, nfedata, 1231 &pvcalls_back_global.frontends, list) { 1232 backend_disconnect(fedata->dev); 1233 } 1234 up(&pvcalls_back_global.frontends_lock); 1235 1236 xenbus_unregister_driver(&pvcalls_back_driver); 1237} 1238 1239module_exit(pvcalls_back_fin); 1240 1241MODULE_DESCRIPTION("Xen PV Calls backend driver"); 1242MODULE_AUTHOR("Stefano Stabellini <sstabellini@kernel.org>"); 1243MODULE_LICENSE("GPL"); 1244