1 /* 2 * Vhost User library 3 * 4 * Copyright IBM, Corp. 2007 5 * Copyright (c) 2016 Red Hat, Inc. 6 * 7 * Authors: 8 * Anthony Liguori <aliguori@us.ibm.com> 9 * Marc-André Lureau <mlureau@redhat.com> 10 * Victor Kaplansky <victork@redhat.com> 11 * 12 * This work is licensed under the terms of the GNU GPL, version 2 or 13 * later. See the COPYING file in the top-level directory. 14 */ 15 16 #ifndef _GNU_SOURCE 17 #define _GNU_SOURCE 18 #endif 19 20 /* this code avoids GLib dependency */ 21 #include <stdlib.h> 22 #include <stdio.h> 23 #include <unistd.h> 24 #include <stdarg.h> 25 #include <errno.h> 26 #include <string.h> 27 #include <assert.h> 28 #include <inttypes.h> 29 #include <sys/types.h> 30 #include <sys/socket.h> 31 #include <sys/eventfd.h> 32 #include <sys/mman.h> 33 #include <endian.h> 34 35 #if defined(__linux__) 36 #include <sys/syscall.h> 37 #include <fcntl.h> 38 #include <sys/ioctl.h> 39 #include <linux/vhost.h> 40 41 #ifdef __NR_userfaultfd 42 #include <linux/userfaultfd.h> 43 #endif 44 45 #endif 46 47 #include "include/atomic.h" 48 49 #include "libvhost-user.h" 50 51 /* usually provided by GLib */ 52 #if __GNUC__ > 2 || (__GNUC__ == 2 && __GNUC_MINOR__ > 4) 53 #if !defined(__clang__) && (__GNUC__ == 4 && __GNUC_MINOR__ == 4) 54 #define G_GNUC_PRINTF(format_idx, arg_idx) \ 55 __attribute__((__format__(gnu_printf, format_idx, arg_idx))) 56 #else 57 #define G_GNUC_PRINTF(format_idx, arg_idx) \ 58 __attribute__((__format__(__printf__, format_idx, arg_idx))) 59 #endif 60 #else /* !__GNUC__ */ 61 #define G_GNUC_PRINTF(format_idx, arg_idx) 62 #endif /* !__GNUC__ */ 63 #ifndef MIN 64 #define MIN(x, y) ({ \ 65 __typeof__(x) _min1 = (x); \ 66 __typeof__(y) _min2 = (y); \ 67 (void) (&_min1 == &_min2); \ 68 _min1 < _min2 ? _min1 : _min2; }) 69 #endif 70 71 /* Round number down to multiple */ 72 #define ALIGN_DOWN(n, m) ((n) / (m) * (m)) 73 74 /* Round number up to multiple */ 75 #define ALIGN_UP(n, m) ALIGN_DOWN((n) + (m) - 1, (m)) 76 77 #ifndef unlikely 78 #define unlikely(x) __builtin_expect(!!(x), 0) 79 #endif 80 81 /* Align each region to cache line size in inflight buffer */ 82 #define INFLIGHT_ALIGNMENT 64 83 84 /* The version of inflight buffer */ 85 #define INFLIGHT_VERSION 1 86 87 /* The version of the protocol we support */ 88 #define VHOST_USER_VERSION 1 89 #define LIBVHOST_USER_DEBUG 0 90 91 #define DPRINT(...) \ 92 do { \ 93 if (LIBVHOST_USER_DEBUG) { \ 94 fprintf(stderr, __VA_ARGS__); \ 95 } \ 96 } while (0) 97 98 static inline 99 bool has_feature(uint64_t features, unsigned int fbit) 100 { 101 assert(fbit < 64); 102 return !!(features & (1ULL << fbit)); 103 } 104 105 static inline 106 bool vu_has_feature(VuDev *dev, 107 unsigned int fbit) 108 { 109 return has_feature(dev->features, fbit); 110 } 111 112 static inline bool vu_has_protocol_feature(VuDev *dev, unsigned int fbit) 113 { 114 return has_feature(dev->protocol_features, fbit); 115 } 116 117 const char * 118 vu_request_to_string(unsigned int req) 119 { 120 #define REQ(req) [req] = #req 121 static const char *vu_request_str[] = { 122 REQ(VHOST_USER_NONE), 123 REQ(VHOST_USER_GET_FEATURES), 124 REQ(VHOST_USER_SET_FEATURES), 125 REQ(VHOST_USER_SET_OWNER), 126 REQ(VHOST_USER_RESET_OWNER), 127 REQ(VHOST_USER_SET_MEM_TABLE), 128 REQ(VHOST_USER_SET_LOG_BASE), 129 REQ(VHOST_USER_SET_LOG_FD), 130 REQ(VHOST_USER_SET_VRING_NUM), 131 REQ(VHOST_USER_SET_VRING_ADDR), 132 REQ(VHOST_USER_SET_VRING_BASE), 133 REQ(VHOST_USER_GET_VRING_BASE), 134 REQ(VHOST_USER_SET_VRING_KICK), 135 REQ(VHOST_USER_SET_VRING_CALL), 136 REQ(VHOST_USER_SET_VRING_ERR), 137 REQ(VHOST_USER_GET_PROTOCOL_FEATURES), 138 REQ(VHOST_USER_SET_PROTOCOL_FEATURES), 139 REQ(VHOST_USER_GET_QUEUE_NUM), 140 REQ(VHOST_USER_SET_VRING_ENABLE), 141 REQ(VHOST_USER_SEND_RARP), 142 REQ(VHOST_USER_NET_SET_MTU), 143 REQ(VHOST_USER_SET_SLAVE_REQ_FD), 144 REQ(VHOST_USER_IOTLB_MSG), 145 REQ(VHOST_USER_SET_VRING_ENDIAN), 146 REQ(VHOST_USER_GET_CONFIG), 147 REQ(VHOST_USER_SET_CONFIG), 148 REQ(VHOST_USER_POSTCOPY_ADVISE), 149 REQ(VHOST_USER_POSTCOPY_LISTEN), 150 REQ(VHOST_USER_POSTCOPY_END), 151 REQ(VHOST_USER_GET_INFLIGHT_FD), 152 REQ(VHOST_USER_SET_INFLIGHT_FD), 153 REQ(VHOST_USER_GPU_SET_SOCKET), 154 REQ(VHOST_USER_VRING_KICK), 155 REQ(VHOST_USER_GET_MAX_MEM_SLOTS), 156 REQ(VHOST_USER_ADD_MEM_REG), 157 REQ(VHOST_USER_REM_MEM_REG), 158 REQ(VHOST_USER_MAX), 159 }; 160 #undef REQ 161 162 if (req < VHOST_USER_MAX) { 163 return vu_request_str[req]; 164 } else { 165 return "unknown"; 166 } 167 } 168 169 static void G_GNUC_PRINTF(2, 3) 170 vu_panic(VuDev *dev, const char *msg, ...) 171 { 172 char *buf = NULL; 173 va_list ap; 174 175 va_start(ap, msg); 176 if (vasprintf(&buf, msg, ap) < 0) { 177 buf = NULL; 178 } 179 va_end(ap); 180 181 dev->broken = true; 182 dev->panic(dev, buf); 183 free(buf); 184 185 /* 186 * FIXME: 187 * find a way to call virtio_error, or perhaps close the connection? 188 */ 189 } 190 191 /* Translate guest physical address to our virtual address. */ 192 void * 193 vu_gpa_to_va(VuDev *dev, uint64_t *plen, uint64_t guest_addr) 194 { 195 unsigned int i; 196 197 if (*plen == 0) { 198 return NULL; 199 } 200 201 /* Find matching memory region. */ 202 for (i = 0; i < dev->nregions; i++) { 203 VuDevRegion *r = &dev->regions[i]; 204 205 if ((guest_addr >= r->gpa) && (guest_addr < (r->gpa + r->size))) { 206 if ((guest_addr + *plen) > (r->gpa + r->size)) { 207 *plen = r->gpa + r->size - guest_addr; 208 } 209 return (void *)(uintptr_t) 210 guest_addr - r->gpa + r->mmap_addr + r->mmap_offset; 211 } 212 } 213 214 return NULL; 215 } 216 217 /* Translate qemu virtual address to our virtual address. */ 218 static void * 219 qva_to_va(VuDev *dev, uint64_t qemu_addr) 220 { 221 unsigned int i; 222 223 /* Find matching memory region. */ 224 for (i = 0; i < dev->nregions; i++) { 225 VuDevRegion *r = &dev->regions[i]; 226 227 if ((qemu_addr >= r->qva) && (qemu_addr < (r->qva + r->size))) { 228 return (void *)(uintptr_t) 229 qemu_addr - r->qva + r->mmap_addr + r->mmap_offset; 230 } 231 } 232 233 return NULL; 234 } 235 236 static void 237 vmsg_close_fds(VhostUserMsg *vmsg) 238 { 239 int i; 240 241 for (i = 0; i < vmsg->fd_num; i++) { 242 close(vmsg->fds[i]); 243 } 244 } 245 246 /* Set reply payload.u64 and clear request flags and fd_num */ 247 static void vmsg_set_reply_u64(VhostUserMsg *vmsg, uint64_t val) 248 { 249 vmsg->flags = 0; /* defaults will be set by vu_send_reply() */ 250 vmsg->size = sizeof(vmsg->payload.u64); 251 vmsg->payload.u64 = val; 252 vmsg->fd_num = 0; 253 } 254 255 /* A test to see if we have userfault available */ 256 static bool 257 have_userfault(void) 258 { 259 #if defined(__linux__) && defined(__NR_userfaultfd) &&\ 260 defined(UFFD_FEATURE_MISSING_SHMEM) &&\ 261 defined(UFFD_FEATURE_MISSING_HUGETLBFS) 262 /* Now test the kernel we're running on really has the features */ 263 int ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK); 264 struct uffdio_api api_struct; 265 if (ufd < 0) { 266 return false; 267 } 268 269 api_struct.api = UFFD_API; 270 api_struct.features = UFFD_FEATURE_MISSING_SHMEM | 271 UFFD_FEATURE_MISSING_HUGETLBFS; 272 if (ioctl(ufd, UFFDIO_API, &api_struct)) { 273 close(ufd); 274 return false; 275 } 276 close(ufd); 277 return true; 278 279 #else 280 return false; 281 #endif 282 } 283 284 static bool 285 vu_message_read_default(VuDev *dev, int conn_fd, VhostUserMsg *vmsg) 286 { 287 char control[CMSG_SPACE(VHOST_MEMORY_BASELINE_NREGIONS * sizeof(int))] = {}; 288 struct iovec iov = { 289 .iov_base = (char *)vmsg, 290 .iov_len = VHOST_USER_HDR_SIZE, 291 }; 292 struct msghdr msg = { 293 .msg_iov = &iov, 294 .msg_iovlen = 1, 295 .msg_control = control, 296 .msg_controllen = sizeof(control), 297 }; 298 size_t fd_size; 299 struct cmsghdr *cmsg; 300 int rc; 301 302 do { 303 rc = recvmsg(conn_fd, &msg, 0); 304 } while (rc < 0 && (errno == EINTR || errno == EAGAIN)); 305 306 if (rc < 0) { 307 vu_panic(dev, "Error while recvmsg: %s", strerror(errno)); 308 return false; 309 } 310 311 vmsg->fd_num = 0; 312 for (cmsg = CMSG_FIRSTHDR(&msg); 313 cmsg != NULL; 314 cmsg = CMSG_NXTHDR(&msg, cmsg)) 315 { 316 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) { 317 fd_size = cmsg->cmsg_len - CMSG_LEN(0); 318 vmsg->fd_num = fd_size / sizeof(int); 319 memcpy(vmsg->fds, CMSG_DATA(cmsg), fd_size); 320 break; 321 } 322 } 323 324 if (vmsg->size > sizeof(vmsg->payload)) { 325 vu_panic(dev, 326 "Error: too big message request: %d, size: vmsg->size: %u, " 327 "while sizeof(vmsg->payload) = %zu\n", 328 vmsg->request, vmsg->size, sizeof(vmsg->payload)); 329 goto fail; 330 } 331 332 if (vmsg->size) { 333 do { 334 rc = read(conn_fd, &vmsg->payload, vmsg->size); 335 } while (rc < 0 && (errno == EINTR || errno == EAGAIN)); 336 337 if (rc <= 0) { 338 vu_panic(dev, "Error while reading: %s", strerror(errno)); 339 goto fail; 340 } 341 342 assert((uint32_t)rc == vmsg->size); 343 } 344 345 return true; 346 347 fail: 348 vmsg_close_fds(vmsg); 349 350 return false; 351 } 352 353 static bool 354 vu_message_write(VuDev *dev, int conn_fd, VhostUserMsg *vmsg) 355 { 356 int rc; 357 uint8_t *p = (uint8_t *)vmsg; 358 char control[CMSG_SPACE(VHOST_MEMORY_BASELINE_NREGIONS * sizeof(int))] = {}; 359 struct iovec iov = { 360 .iov_base = (char *)vmsg, 361 .iov_len = VHOST_USER_HDR_SIZE, 362 }; 363 struct msghdr msg = { 364 .msg_iov = &iov, 365 .msg_iovlen = 1, 366 .msg_control = control, 367 }; 368 struct cmsghdr *cmsg; 369 370 memset(control, 0, sizeof(control)); 371 assert(vmsg->fd_num <= VHOST_MEMORY_BASELINE_NREGIONS); 372 if (vmsg->fd_num > 0) { 373 size_t fdsize = vmsg->fd_num * sizeof(int); 374 msg.msg_controllen = CMSG_SPACE(fdsize); 375 cmsg = CMSG_FIRSTHDR(&msg); 376 cmsg->cmsg_len = CMSG_LEN(fdsize); 377 cmsg->cmsg_level = SOL_SOCKET; 378 cmsg->cmsg_type = SCM_RIGHTS; 379 memcpy(CMSG_DATA(cmsg), vmsg->fds, fdsize); 380 } else { 381 msg.msg_controllen = 0; 382 } 383 384 do { 385 rc = sendmsg(conn_fd, &msg, 0); 386 } while (rc < 0 && (errno == EINTR || errno == EAGAIN)); 387 388 if (vmsg->size) { 389 do { 390 if (vmsg->data) { 391 rc = write(conn_fd, vmsg->data, vmsg->size); 392 } else { 393 rc = write(conn_fd, p + VHOST_USER_HDR_SIZE, vmsg->size); 394 } 395 } while (rc < 0 && (errno == EINTR || errno == EAGAIN)); 396 } 397 398 if (rc <= 0) { 399 vu_panic(dev, "Error while writing: %s", strerror(errno)); 400 return false; 401 } 402 403 return true; 404 } 405 406 static bool 407 vu_send_reply(VuDev *dev, int conn_fd, VhostUserMsg *vmsg) 408 { 409 /* Set the version in the flags when sending the reply */ 410 vmsg->flags &= ~VHOST_USER_VERSION_MASK; 411 vmsg->flags |= VHOST_USER_VERSION; 412 vmsg->flags |= VHOST_USER_REPLY_MASK; 413 414 return vu_message_write(dev, conn_fd, vmsg); 415 } 416 417 /* 418 * Processes a reply on the slave channel. 419 * Entered with slave_mutex held and releases it before exit. 420 * Returns true on success. 421 */ 422 static bool 423 vu_process_message_reply(VuDev *dev, const VhostUserMsg *vmsg) 424 { 425 VhostUserMsg msg_reply; 426 bool result = false; 427 428 if ((vmsg->flags & VHOST_USER_NEED_REPLY_MASK) == 0) { 429 result = true; 430 goto out; 431 } 432 433 if (!vu_message_read_default(dev, dev->slave_fd, &msg_reply)) { 434 goto out; 435 } 436 437 if (msg_reply.request != vmsg->request) { 438 DPRINT("Received unexpected msg type. Expected %d received %d", 439 vmsg->request, msg_reply.request); 440 goto out; 441 } 442 443 result = msg_reply.payload.u64 == 0; 444 445 out: 446 pthread_mutex_unlock(&dev->slave_mutex); 447 return result; 448 } 449 450 /* Kick the log_call_fd if required. */ 451 static void 452 vu_log_kick(VuDev *dev) 453 { 454 if (dev->log_call_fd != -1) { 455 DPRINT("Kicking the QEMU's log...\n"); 456 if (eventfd_write(dev->log_call_fd, 1) < 0) { 457 vu_panic(dev, "Error writing eventfd: %s", strerror(errno)); 458 } 459 } 460 } 461 462 static void 463 vu_log_page(uint8_t *log_table, uint64_t page) 464 { 465 DPRINT("Logged dirty guest page: %"PRId64"\n", page); 466 qatomic_or(&log_table[page / 8], 1 << (page % 8)); 467 } 468 469 static void 470 vu_log_write(VuDev *dev, uint64_t address, uint64_t length) 471 { 472 uint64_t page; 473 474 if (!(dev->features & (1ULL << VHOST_F_LOG_ALL)) || 475 !dev->log_table || !length) { 476 return; 477 } 478 479 assert(dev->log_size > ((address + length - 1) / VHOST_LOG_PAGE / 8)); 480 481 page = address / VHOST_LOG_PAGE; 482 while (page * VHOST_LOG_PAGE < address + length) { 483 vu_log_page(dev->log_table, page); 484 page += 1; 485 } 486 487 vu_log_kick(dev); 488 } 489 490 static void 491 vu_kick_cb(VuDev *dev, int condition, void *data) 492 { 493 int index = (intptr_t)data; 494 VuVirtq *vq = &dev->vq[index]; 495 int sock = vq->kick_fd; 496 eventfd_t kick_data; 497 ssize_t rc; 498 499 rc = eventfd_read(sock, &kick_data); 500 if (rc == -1) { 501 vu_panic(dev, "kick eventfd_read(): %s", strerror(errno)); 502 dev->remove_watch(dev, dev->vq[index].kick_fd); 503 } else { 504 DPRINT("Got kick_data: %016"PRIx64" handler:%p idx:%d\n", 505 kick_data, vq->handler, index); 506 if (vq->handler) { 507 vq->handler(dev, index); 508 } 509 } 510 } 511 512 static bool 513 vu_get_features_exec(VuDev *dev, VhostUserMsg *vmsg) 514 { 515 vmsg->payload.u64 = 516 /* 517 * The following VIRTIO feature bits are supported by our virtqueue 518 * implementation: 519 */ 520 1ULL << VIRTIO_F_NOTIFY_ON_EMPTY | 521 1ULL << VIRTIO_RING_F_INDIRECT_DESC | 522 1ULL << VIRTIO_RING_F_EVENT_IDX | 523 1ULL << VIRTIO_F_VERSION_1 | 524 525 /* vhost-user feature bits */ 526 1ULL << VHOST_F_LOG_ALL | 527 1ULL << VHOST_USER_F_PROTOCOL_FEATURES; 528 529 if (dev->iface->get_features) { 530 vmsg->payload.u64 |= dev->iface->get_features(dev); 531 } 532 533 vmsg->size = sizeof(vmsg->payload.u64); 534 vmsg->fd_num = 0; 535 536 DPRINT("Sending back to guest u64: 0x%016"PRIx64"\n", vmsg->payload.u64); 537 538 return true; 539 } 540 541 static void 542 vu_set_enable_all_rings(VuDev *dev, bool enabled) 543 { 544 uint16_t i; 545 546 for (i = 0; i < dev->max_queues; i++) { 547 dev->vq[i].enable = enabled; 548 } 549 } 550 551 static bool 552 vu_set_features_exec(VuDev *dev, VhostUserMsg *vmsg) 553 { 554 DPRINT("u64: 0x%016"PRIx64"\n", vmsg->payload.u64); 555 556 dev->features = vmsg->payload.u64; 557 if (!vu_has_feature(dev, VIRTIO_F_VERSION_1)) { 558 /* 559 * We only support devices conforming to VIRTIO 1.0 or 560 * later 561 */ 562 vu_panic(dev, "virtio legacy devices aren't supported by libvhost-user"); 563 return false; 564 } 565 566 if (!(dev->features & VHOST_USER_F_PROTOCOL_FEATURES)) { 567 vu_set_enable_all_rings(dev, true); 568 } 569 570 if (dev->iface->set_features) { 571 dev->iface->set_features(dev, dev->features); 572 } 573 574 return false; 575 } 576 577 static bool 578 vu_set_owner_exec(VuDev *dev, VhostUserMsg *vmsg) 579 { 580 return false; 581 } 582 583 static void 584 vu_close_log(VuDev *dev) 585 { 586 if (dev->log_table) { 587 if (munmap(dev->log_table, dev->log_size) != 0) { 588 perror("close log munmap() error"); 589 } 590 591 dev->log_table = NULL; 592 } 593 if (dev->log_call_fd != -1) { 594 close(dev->log_call_fd); 595 dev->log_call_fd = -1; 596 } 597 } 598 599 static bool 600 vu_reset_device_exec(VuDev *dev, VhostUserMsg *vmsg) 601 { 602 vu_set_enable_all_rings(dev, false); 603 604 return false; 605 } 606 607 static bool 608 map_ring(VuDev *dev, VuVirtq *vq) 609 { 610 vq->vring.desc = qva_to_va(dev, vq->vra.desc_user_addr); 611 vq->vring.used = qva_to_va(dev, vq->vra.used_user_addr); 612 vq->vring.avail = qva_to_va(dev, vq->vra.avail_user_addr); 613 614 DPRINT("Setting virtq addresses:\n"); 615 DPRINT(" vring_desc at %p\n", vq->vring.desc); 616 DPRINT(" vring_used at %p\n", vq->vring.used); 617 DPRINT(" vring_avail at %p\n", vq->vring.avail); 618 619 return !(vq->vring.desc && vq->vring.used && vq->vring.avail); 620 } 621 622 static bool 623 generate_faults(VuDev *dev) { 624 unsigned int i; 625 for (i = 0; i < dev->nregions; i++) { 626 VuDevRegion *dev_region = &dev->regions[i]; 627 int ret; 628 #ifdef UFFDIO_REGISTER 629 struct uffdio_register reg_struct; 630 631 /* 632 * We should already have an open ufd. Mark each memory 633 * range as ufd. 634 * Discard any mapping we have here; note I can't use MADV_REMOVE 635 * or fallocate to make the hole since I don't want to lose 636 * data that's already arrived in the shared process. 637 * TODO: How to do hugepage 638 */ 639 ret = madvise((void *)(uintptr_t)dev_region->mmap_addr, 640 dev_region->size + dev_region->mmap_offset, 641 MADV_DONTNEED); 642 if (ret) { 643 fprintf(stderr, 644 "%s: Failed to madvise(DONTNEED) region %d: %s\n", 645 __func__, i, strerror(errno)); 646 } 647 /* 648 * Turn off transparent hugepages so we dont get lose wakeups 649 * in neighbouring pages. 650 * TODO: Turn this backon later. 651 */ 652 ret = madvise((void *)(uintptr_t)dev_region->mmap_addr, 653 dev_region->size + dev_region->mmap_offset, 654 MADV_NOHUGEPAGE); 655 if (ret) { 656 /* 657 * Note: This can happen legally on kernels that are configured 658 * without madvise'able hugepages 659 */ 660 fprintf(stderr, 661 "%s: Failed to madvise(NOHUGEPAGE) region %d: %s\n", 662 __func__, i, strerror(errno)); 663 } 664 665 reg_struct.range.start = (uintptr_t)dev_region->mmap_addr; 666 reg_struct.range.len = dev_region->size + dev_region->mmap_offset; 667 reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING; 668 669 if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER, ®_struct)) { 670 vu_panic(dev, "%s: Failed to userfault region %d " 671 "@%" PRIx64 " + size:%" PRIx64 " offset: %" PRIx64 672 ": (ufd=%d)%s\n", 673 __func__, i, 674 dev_region->mmap_addr, 675 dev_region->size, dev_region->mmap_offset, 676 dev->postcopy_ufd, strerror(errno)); 677 return false; 678 } 679 if (!(reg_struct.ioctls & ((__u64)1 << _UFFDIO_COPY))) { 680 vu_panic(dev, "%s Region (%d) doesn't support COPY", 681 __func__, i); 682 return false; 683 } 684 DPRINT("%s: region %d: Registered userfault for %" 685 PRIx64 " + %" PRIx64 "\n", __func__, i, 686 (uint64_t)reg_struct.range.start, 687 (uint64_t)reg_struct.range.len); 688 /* Now it's registered we can let the client at it */ 689 if (mprotect((void *)(uintptr_t)dev_region->mmap_addr, 690 dev_region->size + dev_region->mmap_offset, 691 PROT_READ | PROT_WRITE)) { 692 vu_panic(dev, "failed to mprotect region %d for postcopy (%s)", 693 i, strerror(errno)); 694 return false; 695 } 696 /* TODO: Stash 'zero' support flags somewhere */ 697 #endif 698 } 699 700 return true; 701 } 702 703 static bool 704 vu_add_mem_reg(VuDev *dev, VhostUserMsg *vmsg) { 705 int i; 706 bool track_ramblocks = dev->postcopy_listening; 707 VhostUserMemoryRegion m = vmsg->payload.memreg.region, *msg_region = &m; 708 VuDevRegion *dev_region = &dev->regions[dev->nregions]; 709 void *mmap_addr; 710 711 if (vmsg->fd_num != 1) { 712 vmsg_close_fds(vmsg); 713 vu_panic(dev, "VHOST_USER_ADD_MEM_REG received %d fds - only 1 fd " 714 "should be sent for this message type", vmsg->fd_num); 715 return false; 716 } 717 718 if (vmsg->size < VHOST_USER_MEM_REG_SIZE) { 719 close(vmsg->fds[0]); 720 vu_panic(dev, "VHOST_USER_ADD_MEM_REG requires a message size of at " 721 "least %zu bytes and only %d bytes were received", 722 VHOST_USER_MEM_REG_SIZE, vmsg->size); 723 return false; 724 } 725 726 if (dev->nregions == VHOST_USER_MAX_RAM_SLOTS) { 727 close(vmsg->fds[0]); 728 vu_panic(dev, "failing attempt to hot add memory via " 729 "VHOST_USER_ADD_MEM_REG message because the backend has " 730 "no free ram slots available"); 731 return false; 732 } 733 734 /* 735 * If we are in postcopy mode and we receive a u64 payload with a 0 value 736 * we know all the postcopy client bases have been received, and we 737 * should start generating faults. 738 */ 739 if (track_ramblocks && 740 vmsg->size == sizeof(vmsg->payload.u64) && 741 vmsg->payload.u64 == 0) { 742 (void)generate_faults(dev); 743 return false; 744 } 745 746 DPRINT("Adding region: %u\n", dev->nregions); 747 DPRINT(" guest_phys_addr: 0x%016"PRIx64"\n", 748 msg_region->guest_phys_addr); 749 DPRINT(" memory_size: 0x%016"PRIx64"\n", 750 msg_region->memory_size); 751 DPRINT(" userspace_addr 0x%016"PRIx64"\n", 752 msg_region->userspace_addr); 753 DPRINT(" mmap_offset 0x%016"PRIx64"\n", 754 msg_region->mmap_offset); 755 756 dev_region->gpa = msg_region->guest_phys_addr; 757 dev_region->size = msg_region->memory_size; 758 dev_region->qva = msg_region->userspace_addr; 759 dev_region->mmap_offset = msg_region->mmap_offset; 760 761 /* 762 * We don't use offset argument of mmap() since the 763 * mapped address has to be page aligned, and we use huge 764 * pages. 765 */ 766 if (track_ramblocks) { 767 /* 768 * In postcopy we're using PROT_NONE here to catch anyone 769 * accessing it before we userfault. 770 */ 771 mmap_addr = mmap(0, dev_region->size + dev_region->mmap_offset, 772 PROT_NONE, MAP_SHARED | MAP_NORESERVE, 773 vmsg->fds[0], 0); 774 } else { 775 mmap_addr = mmap(0, dev_region->size + dev_region->mmap_offset, 776 PROT_READ | PROT_WRITE, MAP_SHARED | MAP_NORESERVE, 777 vmsg->fds[0], 0); 778 } 779 780 if (mmap_addr == MAP_FAILED) { 781 vu_panic(dev, "region mmap error: %s", strerror(errno)); 782 } else { 783 dev_region->mmap_addr = (uint64_t)(uintptr_t)mmap_addr; 784 DPRINT(" mmap_addr: 0x%016"PRIx64"\n", 785 dev_region->mmap_addr); 786 } 787 788 close(vmsg->fds[0]); 789 790 if (track_ramblocks) { 791 /* 792 * Return the address to QEMU so that it can translate the ufd 793 * fault addresses back. 794 */ 795 msg_region->userspace_addr = (uintptr_t)(mmap_addr + 796 dev_region->mmap_offset); 797 798 /* Send the message back to qemu with the addresses filled in. */ 799 vmsg->fd_num = 0; 800 DPRINT("Successfully added new region in postcopy\n"); 801 dev->nregions++; 802 return true; 803 } else { 804 for (i = 0; i < dev->max_queues; i++) { 805 if (dev->vq[i].vring.desc) { 806 if (map_ring(dev, &dev->vq[i])) { 807 vu_panic(dev, "remapping queue %d for new memory region", 808 i); 809 } 810 } 811 } 812 813 DPRINT("Successfully added new region\n"); 814 dev->nregions++; 815 return false; 816 } 817 } 818 819 static inline bool reg_equal(VuDevRegion *vudev_reg, 820 VhostUserMemoryRegion *msg_reg) 821 { 822 if (vudev_reg->gpa == msg_reg->guest_phys_addr && 823 vudev_reg->qva == msg_reg->userspace_addr && 824 vudev_reg->size == msg_reg->memory_size) { 825 return true; 826 } 827 828 return false; 829 } 830 831 static bool 832 vu_rem_mem_reg(VuDev *dev, VhostUserMsg *vmsg) { 833 VhostUserMemoryRegion m = vmsg->payload.memreg.region, *msg_region = &m; 834 unsigned int i; 835 bool found = false; 836 837 if (vmsg->fd_num > 1) { 838 vmsg_close_fds(vmsg); 839 vu_panic(dev, "VHOST_USER_REM_MEM_REG received %d fds - at most 1 fd " 840 "should be sent for this message type", vmsg->fd_num); 841 return false; 842 } 843 844 if (vmsg->size < VHOST_USER_MEM_REG_SIZE) { 845 vmsg_close_fds(vmsg); 846 vu_panic(dev, "VHOST_USER_REM_MEM_REG requires a message size of at " 847 "least %zu bytes and only %d bytes were received", 848 VHOST_USER_MEM_REG_SIZE, vmsg->size); 849 return false; 850 } 851 852 DPRINT("Removing region:\n"); 853 DPRINT(" guest_phys_addr: 0x%016"PRIx64"\n", 854 msg_region->guest_phys_addr); 855 DPRINT(" memory_size: 0x%016"PRIx64"\n", 856 msg_region->memory_size); 857 DPRINT(" userspace_addr 0x%016"PRIx64"\n", 858 msg_region->userspace_addr); 859 DPRINT(" mmap_offset 0x%016"PRIx64"\n", 860 msg_region->mmap_offset); 861 862 for (i = 0; i < dev->nregions; i++) { 863 if (reg_equal(&dev->regions[i], msg_region)) { 864 VuDevRegion *r = &dev->regions[i]; 865 void *m = (void *) (uintptr_t) r->mmap_addr; 866 867 if (m) { 868 munmap(m, r->size + r->mmap_offset); 869 } 870 871 /* 872 * Shift all affected entries by 1 to close the hole at index i and 873 * zero out the last entry. 874 */ 875 memmove(dev->regions + i, dev->regions + i + 1, 876 sizeof(VuDevRegion) * (dev->nregions - i - 1)); 877 memset(dev->regions + dev->nregions - 1, 0, sizeof(VuDevRegion)); 878 DPRINT("Successfully removed a region\n"); 879 dev->nregions--; 880 i--; 881 882 found = true; 883 884 /* Continue the search for eventual duplicates. */ 885 } 886 } 887 888 if (!found) { 889 vu_panic(dev, "Specified region not found\n"); 890 } 891 892 vmsg_close_fds(vmsg); 893 894 return false; 895 } 896 897 static bool 898 vu_set_mem_table_exec_postcopy(VuDev *dev, VhostUserMsg *vmsg) 899 { 900 unsigned int i; 901 VhostUserMemory m = vmsg->payload.memory, *memory = &m; 902 dev->nregions = memory->nregions; 903 904 DPRINT("Nregions: %u\n", memory->nregions); 905 for (i = 0; i < dev->nregions; i++) { 906 void *mmap_addr; 907 VhostUserMemoryRegion *msg_region = &memory->regions[i]; 908 VuDevRegion *dev_region = &dev->regions[i]; 909 910 DPRINT("Region %d\n", i); 911 DPRINT(" guest_phys_addr: 0x%016"PRIx64"\n", 912 msg_region->guest_phys_addr); 913 DPRINT(" memory_size: 0x%016"PRIx64"\n", 914 msg_region->memory_size); 915 DPRINT(" userspace_addr 0x%016"PRIx64"\n", 916 msg_region->userspace_addr); 917 DPRINT(" mmap_offset 0x%016"PRIx64"\n", 918 msg_region->mmap_offset); 919 920 dev_region->gpa = msg_region->guest_phys_addr; 921 dev_region->size = msg_region->memory_size; 922 dev_region->qva = msg_region->userspace_addr; 923 dev_region->mmap_offset = msg_region->mmap_offset; 924 925 /* We don't use offset argument of mmap() since the 926 * mapped address has to be page aligned, and we use huge 927 * pages. 928 * In postcopy we're using PROT_NONE here to catch anyone 929 * accessing it before we userfault 930 */ 931 mmap_addr = mmap(0, dev_region->size + dev_region->mmap_offset, 932 PROT_NONE, MAP_SHARED | MAP_NORESERVE, 933 vmsg->fds[i], 0); 934 935 if (mmap_addr == MAP_FAILED) { 936 vu_panic(dev, "region mmap error: %s", strerror(errno)); 937 } else { 938 dev_region->mmap_addr = (uint64_t)(uintptr_t)mmap_addr; 939 DPRINT(" mmap_addr: 0x%016"PRIx64"\n", 940 dev_region->mmap_addr); 941 } 942 943 /* Return the address to QEMU so that it can translate the ufd 944 * fault addresses back. 945 */ 946 msg_region->userspace_addr = (uintptr_t)(mmap_addr + 947 dev_region->mmap_offset); 948 close(vmsg->fds[i]); 949 } 950 951 /* Send the message back to qemu with the addresses filled in */ 952 vmsg->fd_num = 0; 953 if (!vu_send_reply(dev, dev->sock, vmsg)) { 954 vu_panic(dev, "failed to respond to set-mem-table for postcopy"); 955 return false; 956 } 957 958 /* Wait for QEMU to confirm that it's registered the handler for the 959 * faults. 960 */ 961 if (!dev->read_msg(dev, dev->sock, vmsg) || 962 vmsg->size != sizeof(vmsg->payload.u64) || 963 vmsg->payload.u64 != 0) { 964 vu_panic(dev, "failed to receive valid ack for postcopy set-mem-table"); 965 return false; 966 } 967 968 /* OK, now we can go and register the memory and generate faults */ 969 (void)generate_faults(dev); 970 971 return false; 972 } 973 974 static bool 975 vu_set_mem_table_exec(VuDev *dev, VhostUserMsg *vmsg) 976 { 977 unsigned int i; 978 VhostUserMemory m = vmsg->payload.memory, *memory = &m; 979 980 for (i = 0; i < dev->nregions; i++) { 981 VuDevRegion *r = &dev->regions[i]; 982 void *m = (void *) (uintptr_t) r->mmap_addr; 983 984 if (m) { 985 munmap(m, r->size + r->mmap_offset); 986 } 987 } 988 dev->nregions = memory->nregions; 989 990 if (dev->postcopy_listening) { 991 return vu_set_mem_table_exec_postcopy(dev, vmsg); 992 } 993 994 DPRINT("Nregions: %u\n", memory->nregions); 995 for (i = 0; i < dev->nregions; i++) { 996 void *mmap_addr; 997 VhostUserMemoryRegion *msg_region = &memory->regions[i]; 998 VuDevRegion *dev_region = &dev->regions[i]; 999 1000 DPRINT("Region %d\n", i); 1001 DPRINT(" guest_phys_addr: 0x%016"PRIx64"\n", 1002 msg_region->guest_phys_addr); 1003 DPRINT(" memory_size: 0x%016"PRIx64"\n", 1004 msg_region->memory_size); 1005 DPRINT(" userspace_addr 0x%016"PRIx64"\n", 1006 msg_region->userspace_addr); 1007 DPRINT(" mmap_offset 0x%016"PRIx64"\n", 1008 msg_region->mmap_offset); 1009 1010 dev_region->gpa = msg_region->guest_phys_addr; 1011 dev_region->size = msg_region->memory_size; 1012 dev_region->qva = msg_region->userspace_addr; 1013 dev_region->mmap_offset = msg_region->mmap_offset; 1014 1015 /* We don't use offset argument of mmap() since the 1016 * mapped address has to be page aligned, and we use huge 1017 * pages. */ 1018 mmap_addr = mmap(0, dev_region->size + dev_region->mmap_offset, 1019 PROT_READ | PROT_WRITE, MAP_SHARED | MAP_NORESERVE, 1020 vmsg->fds[i], 0); 1021 1022 if (mmap_addr == MAP_FAILED) { 1023 vu_panic(dev, "region mmap error: %s", strerror(errno)); 1024 } else { 1025 dev_region->mmap_addr = (uint64_t)(uintptr_t)mmap_addr; 1026 DPRINT(" mmap_addr: 0x%016"PRIx64"\n", 1027 dev_region->mmap_addr); 1028 } 1029 1030 close(vmsg->fds[i]); 1031 } 1032 1033 for (i = 0; i < dev->max_queues; i++) { 1034 if (dev->vq[i].vring.desc) { 1035 if (map_ring(dev, &dev->vq[i])) { 1036 vu_panic(dev, "remapping queue %d during setmemtable", i); 1037 } 1038 } 1039 } 1040 1041 return false; 1042 } 1043 1044 static bool 1045 vu_set_log_base_exec(VuDev *dev, VhostUserMsg *vmsg) 1046 { 1047 int fd; 1048 uint64_t log_mmap_size, log_mmap_offset; 1049 void *rc; 1050 1051 if (vmsg->fd_num != 1 || 1052 vmsg->size != sizeof(vmsg->payload.log)) { 1053 vu_panic(dev, "Invalid log_base message"); 1054 return true; 1055 } 1056 1057 fd = vmsg->fds[0]; 1058 log_mmap_offset = vmsg->payload.log.mmap_offset; 1059 log_mmap_size = vmsg->payload.log.mmap_size; 1060 DPRINT("Log mmap_offset: %"PRId64"\n", log_mmap_offset); 1061 DPRINT("Log mmap_size: %"PRId64"\n", log_mmap_size); 1062 1063 rc = mmap(0, log_mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 1064 log_mmap_offset); 1065 close(fd); 1066 if (rc == MAP_FAILED) { 1067 perror("log mmap error"); 1068 } 1069 1070 if (dev->log_table) { 1071 munmap(dev->log_table, dev->log_size); 1072 } 1073 dev->log_table = rc; 1074 dev->log_size = log_mmap_size; 1075 1076 vmsg->size = sizeof(vmsg->payload.u64); 1077 vmsg->fd_num = 0; 1078 1079 return true; 1080 } 1081 1082 static bool 1083 vu_set_log_fd_exec(VuDev *dev, VhostUserMsg *vmsg) 1084 { 1085 if (vmsg->fd_num != 1) { 1086 vu_panic(dev, "Invalid log_fd message"); 1087 return false; 1088 } 1089 1090 if (dev->log_call_fd != -1) { 1091 close(dev->log_call_fd); 1092 } 1093 dev->log_call_fd = vmsg->fds[0]; 1094 DPRINT("Got log_call_fd: %d\n", vmsg->fds[0]); 1095 1096 return false; 1097 } 1098 1099 static bool 1100 vu_set_vring_num_exec(VuDev *dev, VhostUserMsg *vmsg) 1101 { 1102 unsigned int index = vmsg->payload.state.index; 1103 unsigned int num = vmsg->payload.state.num; 1104 1105 DPRINT("State.index: %u\n", index); 1106 DPRINT("State.num: %u\n", num); 1107 dev->vq[index].vring.num = num; 1108 1109 return false; 1110 } 1111 1112 static bool 1113 vu_set_vring_addr_exec(VuDev *dev, VhostUserMsg *vmsg) 1114 { 1115 struct vhost_vring_addr addr = vmsg->payload.addr, *vra = &addr; 1116 unsigned int index = vra->index; 1117 VuVirtq *vq = &dev->vq[index]; 1118 1119 DPRINT("vhost_vring_addr:\n"); 1120 DPRINT(" index: %d\n", vra->index); 1121 DPRINT(" flags: %d\n", vra->flags); 1122 DPRINT(" desc_user_addr: 0x%016" PRIx64 "\n", (uint64_t)vra->desc_user_addr); 1123 DPRINT(" used_user_addr: 0x%016" PRIx64 "\n", (uint64_t)vra->used_user_addr); 1124 DPRINT(" avail_user_addr: 0x%016" PRIx64 "\n", (uint64_t)vra->avail_user_addr); 1125 DPRINT(" log_guest_addr: 0x%016" PRIx64 "\n", (uint64_t)vra->log_guest_addr); 1126 1127 vq->vra = *vra; 1128 vq->vring.flags = vra->flags; 1129 vq->vring.log_guest_addr = vra->log_guest_addr; 1130 1131 1132 if (map_ring(dev, vq)) { 1133 vu_panic(dev, "Invalid vring_addr message"); 1134 return false; 1135 } 1136 1137 vq->used_idx = le16toh(vq->vring.used->idx); 1138 1139 if (vq->last_avail_idx != vq->used_idx) { 1140 bool resume = dev->iface->queue_is_processed_in_order && 1141 dev->iface->queue_is_processed_in_order(dev, index); 1142 1143 DPRINT("Last avail index != used index: %u != %u%s\n", 1144 vq->last_avail_idx, vq->used_idx, 1145 resume ? ", resuming" : ""); 1146 1147 if (resume) { 1148 vq->shadow_avail_idx = vq->last_avail_idx = vq->used_idx; 1149 } 1150 } 1151 1152 return false; 1153 } 1154 1155 static bool 1156 vu_set_vring_base_exec(VuDev *dev, VhostUserMsg *vmsg) 1157 { 1158 unsigned int index = vmsg->payload.state.index; 1159 unsigned int num = vmsg->payload.state.num; 1160 1161 DPRINT("State.index: %u\n", index); 1162 DPRINT("State.num: %u\n", num); 1163 dev->vq[index].shadow_avail_idx = dev->vq[index].last_avail_idx = num; 1164 1165 return false; 1166 } 1167 1168 static bool 1169 vu_get_vring_base_exec(VuDev *dev, VhostUserMsg *vmsg) 1170 { 1171 unsigned int index = vmsg->payload.state.index; 1172 1173 DPRINT("State.index: %u\n", index); 1174 vmsg->payload.state.num = dev->vq[index].last_avail_idx; 1175 vmsg->size = sizeof(vmsg->payload.state); 1176 1177 dev->vq[index].started = false; 1178 if (dev->iface->queue_set_started) { 1179 dev->iface->queue_set_started(dev, index, false); 1180 } 1181 1182 if (dev->vq[index].call_fd != -1) { 1183 close(dev->vq[index].call_fd); 1184 dev->vq[index].call_fd = -1; 1185 } 1186 if (dev->vq[index].kick_fd != -1) { 1187 dev->remove_watch(dev, dev->vq[index].kick_fd); 1188 close(dev->vq[index].kick_fd); 1189 dev->vq[index].kick_fd = -1; 1190 } 1191 1192 return true; 1193 } 1194 1195 static bool 1196 vu_check_queue_msg_file(VuDev *dev, VhostUserMsg *vmsg) 1197 { 1198 int index = vmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK; 1199 bool nofd = vmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK; 1200 1201 if (index >= dev->max_queues) { 1202 vmsg_close_fds(vmsg); 1203 vu_panic(dev, "Invalid queue index: %u", index); 1204 return false; 1205 } 1206 1207 if (nofd) { 1208 vmsg_close_fds(vmsg); 1209 return true; 1210 } 1211 1212 if (vmsg->fd_num != 1) { 1213 vmsg_close_fds(vmsg); 1214 vu_panic(dev, "Invalid fds in request: %d", vmsg->request); 1215 return false; 1216 } 1217 1218 return true; 1219 } 1220 1221 static int 1222 inflight_desc_compare(const void *a, const void *b) 1223 { 1224 VuVirtqInflightDesc *desc0 = (VuVirtqInflightDesc *)a, 1225 *desc1 = (VuVirtqInflightDesc *)b; 1226 1227 if (desc1->counter > desc0->counter && 1228 (desc1->counter - desc0->counter) < VIRTQUEUE_MAX_SIZE * 2) { 1229 return 1; 1230 } 1231 1232 return -1; 1233 } 1234 1235 static int 1236 vu_check_queue_inflights(VuDev *dev, VuVirtq *vq) 1237 { 1238 int i = 0; 1239 1240 if (!vu_has_protocol_feature(dev, VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)) { 1241 return 0; 1242 } 1243 1244 if (unlikely(!vq->inflight)) { 1245 return -1; 1246 } 1247 1248 if (unlikely(!vq->inflight->version)) { 1249 /* initialize the buffer */ 1250 vq->inflight->version = INFLIGHT_VERSION; 1251 return 0; 1252 } 1253 1254 vq->used_idx = le16toh(vq->vring.used->idx); 1255 vq->resubmit_num = 0; 1256 vq->resubmit_list = NULL; 1257 vq->counter = 0; 1258 1259 if (unlikely(vq->inflight->used_idx != vq->used_idx)) { 1260 vq->inflight->desc[vq->inflight->last_batch_head].inflight = 0; 1261 1262 barrier(); 1263 1264 vq->inflight->used_idx = vq->used_idx; 1265 } 1266 1267 for (i = 0; i < vq->inflight->desc_num; i++) { 1268 if (vq->inflight->desc[i].inflight == 1) { 1269 vq->inuse++; 1270 } 1271 } 1272 1273 vq->shadow_avail_idx = vq->last_avail_idx = vq->inuse + vq->used_idx; 1274 1275 if (vq->inuse) { 1276 vq->resubmit_list = calloc(vq->inuse, sizeof(VuVirtqInflightDesc)); 1277 if (!vq->resubmit_list) { 1278 return -1; 1279 } 1280 1281 for (i = 0; i < vq->inflight->desc_num; i++) { 1282 if (vq->inflight->desc[i].inflight) { 1283 vq->resubmit_list[vq->resubmit_num].index = i; 1284 vq->resubmit_list[vq->resubmit_num].counter = 1285 vq->inflight->desc[i].counter; 1286 vq->resubmit_num++; 1287 } 1288 } 1289 1290 if (vq->resubmit_num > 1) { 1291 qsort(vq->resubmit_list, vq->resubmit_num, 1292 sizeof(VuVirtqInflightDesc), inflight_desc_compare); 1293 } 1294 vq->counter = vq->resubmit_list[0].counter + 1; 1295 } 1296 1297 /* in case of I/O hang after reconnecting */ 1298 if (eventfd_write(vq->kick_fd, 1)) { 1299 return -1; 1300 } 1301 1302 return 0; 1303 } 1304 1305 static bool 1306 vu_set_vring_kick_exec(VuDev *dev, VhostUserMsg *vmsg) 1307 { 1308 int index = vmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK; 1309 bool nofd = vmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK; 1310 1311 DPRINT("u64: 0x%016"PRIx64"\n", vmsg->payload.u64); 1312 1313 if (!vu_check_queue_msg_file(dev, vmsg)) { 1314 return false; 1315 } 1316 1317 if (dev->vq[index].kick_fd != -1) { 1318 dev->remove_watch(dev, dev->vq[index].kick_fd); 1319 close(dev->vq[index].kick_fd); 1320 dev->vq[index].kick_fd = -1; 1321 } 1322 1323 dev->vq[index].kick_fd = nofd ? -1 : vmsg->fds[0]; 1324 DPRINT("Got kick_fd: %d for vq: %d\n", dev->vq[index].kick_fd, index); 1325 1326 dev->vq[index].started = true; 1327 if (dev->iface->queue_set_started) { 1328 dev->iface->queue_set_started(dev, index, true); 1329 } 1330 1331 if (dev->vq[index].kick_fd != -1 && dev->vq[index].handler) { 1332 dev->set_watch(dev, dev->vq[index].kick_fd, VU_WATCH_IN, 1333 vu_kick_cb, (void *)(long)index); 1334 1335 DPRINT("Waiting for kicks on fd: %d for vq: %d\n", 1336 dev->vq[index].kick_fd, index); 1337 } 1338 1339 if (vu_check_queue_inflights(dev, &dev->vq[index])) { 1340 vu_panic(dev, "Failed to check inflights for vq: %d\n", index); 1341 } 1342 1343 return false; 1344 } 1345 1346 void vu_set_queue_handler(VuDev *dev, VuVirtq *vq, 1347 vu_queue_handler_cb handler) 1348 { 1349 int qidx = vq - dev->vq; 1350 1351 vq->handler = handler; 1352 if (vq->kick_fd >= 0) { 1353 if (handler) { 1354 dev->set_watch(dev, vq->kick_fd, VU_WATCH_IN, 1355 vu_kick_cb, (void *)(long)qidx); 1356 } else { 1357 dev->remove_watch(dev, vq->kick_fd); 1358 } 1359 } 1360 } 1361 1362 bool vu_set_queue_host_notifier(VuDev *dev, VuVirtq *vq, int fd, 1363 int size, int offset) 1364 { 1365 int qidx = vq - dev->vq; 1366 int fd_num = 0; 1367 VhostUserMsg vmsg = { 1368 .request = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG, 1369 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK, 1370 .size = sizeof(vmsg.payload.area), 1371 .payload.area = { 1372 .u64 = qidx & VHOST_USER_VRING_IDX_MASK, 1373 .size = size, 1374 .offset = offset, 1375 }, 1376 }; 1377 1378 if (fd == -1) { 1379 vmsg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK; 1380 } else { 1381 vmsg.fds[fd_num++] = fd; 1382 } 1383 1384 vmsg.fd_num = fd_num; 1385 1386 if (!vu_has_protocol_feature(dev, VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) { 1387 return false; 1388 } 1389 1390 pthread_mutex_lock(&dev->slave_mutex); 1391 if (!vu_message_write(dev, dev->slave_fd, &vmsg)) { 1392 pthread_mutex_unlock(&dev->slave_mutex); 1393 return false; 1394 } 1395 1396 /* Also unlocks the slave_mutex */ 1397 return vu_process_message_reply(dev, &vmsg); 1398 } 1399 1400 static bool 1401 vu_set_vring_call_exec(VuDev *dev, VhostUserMsg *vmsg) 1402 { 1403 int index = vmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK; 1404 bool nofd = vmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK; 1405 1406 DPRINT("u64: 0x%016"PRIx64"\n", vmsg->payload.u64); 1407 1408 if (!vu_check_queue_msg_file(dev, vmsg)) { 1409 return false; 1410 } 1411 1412 if (dev->vq[index].call_fd != -1) { 1413 close(dev->vq[index].call_fd); 1414 dev->vq[index].call_fd = -1; 1415 } 1416 1417 dev->vq[index].call_fd = nofd ? -1 : vmsg->fds[0]; 1418 1419 /* in case of I/O hang after reconnecting */ 1420 if (dev->vq[index].call_fd != -1 && eventfd_write(vmsg->fds[0], 1)) { 1421 return -1; 1422 } 1423 1424 DPRINT("Got call_fd: %d for vq: %d\n", dev->vq[index].call_fd, index); 1425 1426 return false; 1427 } 1428 1429 static bool 1430 vu_set_vring_err_exec(VuDev *dev, VhostUserMsg *vmsg) 1431 { 1432 int index = vmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK; 1433 bool nofd = vmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK; 1434 1435 DPRINT("u64: 0x%016"PRIx64"\n", vmsg->payload.u64); 1436 1437 if (!vu_check_queue_msg_file(dev, vmsg)) { 1438 return false; 1439 } 1440 1441 if (dev->vq[index].err_fd != -1) { 1442 close(dev->vq[index].err_fd); 1443 dev->vq[index].err_fd = -1; 1444 } 1445 1446 dev->vq[index].err_fd = nofd ? -1 : vmsg->fds[0]; 1447 1448 return false; 1449 } 1450 1451 static bool 1452 vu_get_protocol_features_exec(VuDev *dev, VhostUserMsg *vmsg) 1453 { 1454 /* 1455 * Note that we support, but intentionally do not set, 1456 * VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS. This means that 1457 * a device implementation can return it in its callback 1458 * (get_protocol_features) if it wants to use this for 1459 * simulation, but it is otherwise not desirable (if even 1460 * implemented by the master.) 1461 */ 1462 uint64_t features = 1ULL << VHOST_USER_PROTOCOL_F_MQ | 1463 1ULL << VHOST_USER_PROTOCOL_F_LOG_SHMFD | 1464 1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ | 1465 1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER | 1466 1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD | 1467 1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK | 1468 1ULL << VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS; 1469 1470 if (have_userfault()) { 1471 features |= 1ULL << VHOST_USER_PROTOCOL_F_PAGEFAULT; 1472 } 1473 1474 if (dev->iface->get_config && dev->iface->set_config) { 1475 features |= 1ULL << VHOST_USER_PROTOCOL_F_CONFIG; 1476 } 1477 1478 if (dev->iface->get_protocol_features) { 1479 features |= dev->iface->get_protocol_features(dev); 1480 } 1481 1482 vmsg_set_reply_u64(vmsg, features); 1483 return true; 1484 } 1485 1486 static bool 1487 vu_set_protocol_features_exec(VuDev *dev, VhostUserMsg *vmsg) 1488 { 1489 uint64_t features = vmsg->payload.u64; 1490 1491 DPRINT("u64: 0x%016"PRIx64"\n", features); 1492 1493 dev->protocol_features = vmsg->payload.u64; 1494 1495 if (vu_has_protocol_feature(dev, 1496 VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS) && 1497 (!vu_has_protocol_feature(dev, VHOST_USER_PROTOCOL_F_SLAVE_REQ) || 1498 !vu_has_protocol_feature(dev, VHOST_USER_PROTOCOL_F_REPLY_ACK))) { 1499 /* 1500 * The use case for using messages for kick/call is simulation, to make 1501 * the kick and call synchronous. To actually get that behaviour, both 1502 * of the other features are required. 1503 * Theoretically, one could use only kick messages, or do them without 1504 * having F_REPLY_ACK, but too many (possibly pending) messages on the 1505 * socket will eventually cause the master to hang, to avoid this in 1506 * scenarios where not desired enforce that the settings are in a way 1507 * that actually enables the simulation case. 1508 */ 1509 vu_panic(dev, 1510 "F_IN_BAND_NOTIFICATIONS requires F_SLAVE_REQ && F_REPLY_ACK"); 1511 return false; 1512 } 1513 1514 if (dev->iface->set_protocol_features) { 1515 dev->iface->set_protocol_features(dev, features); 1516 } 1517 1518 return false; 1519 } 1520 1521 static bool 1522 vu_get_queue_num_exec(VuDev *dev, VhostUserMsg *vmsg) 1523 { 1524 vmsg_set_reply_u64(vmsg, dev->max_queues); 1525 return true; 1526 } 1527 1528 static bool 1529 vu_set_vring_enable_exec(VuDev *dev, VhostUserMsg *vmsg) 1530 { 1531 unsigned int index = vmsg->payload.state.index; 1532 unsigned int enable = vmsg->payload.state.num; 1533 1534 DPRINT("State.index: %u\n", index); 1535 DPRINT("State.enable: %u\n", enable); 1536 1537 if (index >= dev->max_queues) { 1538 vu_panic(dev, "Invalid vring_enable index: %u", index); 1539 return false; 1540 } 1541 1542 dev->vq[index].enable = enable; 1543 return false; 1544 } 1545 1546 static bool 1547 vu_set_slave_req_fd(VuDev *dev, VhostUserMsg *vmsg) 1548 { 1549 if (vmsg->fd_num != 1) { 1550 vu_panic(dev, "Invalid slave_req_fd message (%d fd's)", vmsg->fd_num); 1551 return false; 1552 } 1553 1554 if (dev->slave_fd != -1) { 1555 close(dev->slave_fd); 1556 } 1557 dev->slave_fd = vmsg->fds[0]; 1558 DPRINT("Got slave_fd: %d\n", vmsg->fds[0]); 1559 1560 return false; 1561 } 1562 1563 static bool 1564 vu_get_config(VuDev *dev, VhostUserMsg *vmsg) 1565 { 1566 int ret = -1; 1567 1568 if (dev->iface->get_config) { 1569 ret = dev->iface->get_config(dev, vmsg->payload.config.region, 1570 vmsg->payload.config.size); 1571 } 1572 1573 if (ret) { 1574 /* resize to zero to indicate an error to master */ 1575 vmsg->size = 0; 1576 } 1577 1578 return true; 1579 } 1580 1581 static bool 1582 vu_set_config(VuDev *dev, VhostUserMsg *vmsg) 1583 { 1584 int ret = -1; 1585 1586 if (dev->iface->set_config) { 1587 ret = dev->iface->set_config(dev, vmsg->payload.config.region, 1588 vmsg->payload.config.offset, 1589 vmsg->payload.config.size, 1590 vmsg->payload.config.flags); 1591 if (ret) { 1592 vu_panic(dev, "Set virtio configuration space failed"); 1593 } 1594 } 1595 1596 return false; 1597 } 1598 1599 static bool 1600 vu_set_postcopy_advise(VuDev *dev, VhostUserMsg *vmsg) 1601 { 1602 dev->postcopy_ufd = -1; 1603 #ifdef UFFDIO_API 1604 struct uffdio_api api_struct; 1605 1606 dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK); 1607 vmsg->size = 0; 1608 #endif 1609 1610 if (dev->postcopy_ufd == -1) { 1611 vu_panic(dev, "Userfaultfd not available: %s", strerror(errno)); 1612 goto out; 1613 } 1614 1615 #ifdef UFFDIO_API 1616 api_struct.api = UFFD_API; 1617 api_struct.features = 0; 1618 if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) { 1619 vu_panic(dev, "Failed UFFDIO_API: %s", strerror(errno)); 1620 close(dev->postcopy_ufd); 1621 dev->postcopy_ufd = -1; 1622 goto out; 1623 } 1624 /* TODO: Stash feature flags somewhere */ 1625 #endif 1626 1627 out: 1628 /* Return a ufd to the QEMU */ 1629 vmsg->fd_num = 1; 1630 vmsg->fds[0] = dev->postcopy_ufd; 1631 return true; /* = send a reply */ 1632 } 1633 1634 static bool 1635 vu_set_postcopy_listen(VuDev *dev, VhostUserMsg *vmsg) 1636 { 1637 if (dev->nregions) { 1638 vu_panic(dev, "Regions already registered at postcopy-listen"); 1639 vmsg_set_reply_u64(vmsg, -1); 1640 return true; 1641 } 1642 dev->postcopy_listening = true; 1643 1644 vmsg_set_reply_u64(vmsg, 0); 1645 return true; 1646 } 1647 1648 static bool 1649 vu_set_postcopy_end(VuDev *dev, VhostUserMsg *vmsg) 1650 { 1651 DPRINT("%s: Entry\n", __func__); 1652 dev->postcopy_listening = false; 1653 if (dev->postcopy_ufd > 0) { 1654 close(dev->postcopy_ufd); 1655 dev->postcopy_ufd = -1; 1656 DPRINT("%s: Done close\n", __func__); 1657 } 1658 1659 vmsg_set_reply_u64(vmsg, 0); 1660 DPRINT("%s: exit\n", __func__); 1661 return true; 1662 } 1663 1664 static inline uint64_t 1665 vu_inflight_queue_size(uint16_t queue_size) 1666 { 1667 return ALIGN_UP(sizeof(VuDescStateSplit) * queue_size + 1668 sizeof(uint16_t), INFLIGHT_ALIGNMENT); 1669 } 1670 1671 #ifdef MFD_ALLOW_SEALING 1672 static void * 1673 memfd_alloc(const char *name, size_t size, unsigned int flags, int *fd) 1674 { 1675 void *ptr; 1676 int ret; 1677 1678 *fd = memfd_create(name, MFD_ALLOW_SEALING); 1679 if (*fd < 0) { 1680 return NULL; 1681 } 1682 1683 ret = ftruncate(*fd, size); 1684 if (ret < 0) { 1685 close(*fd); 1686 return NULL; 1687 } 1688 1689 ret = fcntl(*fd, F_ADD_SEALS, flags); 1690 if (ret < 0) { 1691 close(*fd); 1692 return NULL; 1693 } 1694 1695 ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, *fd, 0); 1696 if (ptr == MAP_FAILED) { 1697 close(*fd); 1698 return NULL; 1699 } 1700 1701 return ptr; 1702 } 1703 #endif 1704 1705 static bool 1706 vu_get_inflight_fd(VuDev *dev, VhostUserMsg *vmsg) 1707 { 1708 int fd = -1; 1709 void *addr = NULL; 1710 uint64_t mmap_size; 1711 uint16_t num_queues, queue_size; 1712 1713 if (vmsg->size != sizeof(vmsg->payload.inflight)) { 1714 vu_panic(dev, "Invalid get_inflight_fd message:%d", vmsg->size); 1715 vmsg->payload.inflight.mmap_size = 0; 1716 return true; 1717 } 1718 1719 num_queues = vmsg->payload.inflight.num_queues; 1720 queue_size = vmsg->payload.inflight.queue_size; 1721 1722 DPRINT("set_inflight_fd num_queues: %"PRId16"\n", num_queues); 1723 DPRINT("set_inflight_fd queue_size: %"PRId16"\n", queue_size); 1724 1725 mmap_size = vu_inflight_queue_size(queue_size) * num_queues; 1726 1727 #ifdef MFD_ALLOW_SEALING 1728 addr = memfd_alloc("vhost-inflight", mmap_size, 1729 F_SEAL_GROW | F_SEAL_SHRINK | F_SEAL_SEAL, 1730 &fd); 1731 #else 1732 vu_panic(dev, "Not implemented: memfd support is missing"); 1733 #endif 1734 1735 if (!addr) { 1736 vu_panic(dev, "Failed to alloc vhost inflight area"); 1737 vmsg->payload.inflight.mmap_size = 0; 1738 return true; 1739 } 1740 1741 memset(addr, 0, mmap_size); 1742 1743 dev->inflight_info.addr = addr; 1744 dev->inflight_info.size = vmsg->payload.inflight.mmap_size = mmap_size; 1745 dev->inflight_info.fd = vmsg->fds[0] = fd; 1746 vmsg->fd_num = 1; 1747 vmsg->payload.inflight.mmap_offset = 0; 1748 1749 DPRINT("send inflight mmap_size: %"PRId64"\n", 1750 vmsg->payload.inflight.mmap_size); 1751 DPRINT("send inflight mmap offset: %"PRId64"\n", 1752 vmsg->payload.inflight.mmap_offset); 1753 1754 return true; 1755 } 1756 1757 static bool 1758 vu_set_inflight_fd(VuDev *dev, VhostUserMsg *vmsg) 1759 { 1760 int fd, i; 1761 uint64_t mmap_size, mmap_offset; 1762 uint16_t num_queues, queue_size; 1763 void *rc; 1764 1765 if (vmsg->fd_num != 1 || 1766 vmsg->size != sizeof(vmsg->payload.inflight)) { 1767 vu_panic(dev, "Invalid set_inflight_fd message size:%d fds:%d", 1768 vmsg->size, vmsg->fd_num); 1769 return false; 1770 } 1771 1772 fd = vmsg->fds[0]; 1773 mmap_size = vmsg->payload.inflight.mmap_size; 1774 mmap_offset = vmsg->payload.inflight.mmap_offset; 1775 num_queues = vmsg->payload.inflight.num_queues; 1776 queue_size = vmsg->payload.inflight.queue_size; 1777 1778 DPRINT("set_inflight_fd mmap_size: %"PRId64"\n", mmap_size); 1779 DPRINT("set_inflight_fd mmap_offset: %"PRId64"\n", mmap_offset); 1780 DPRINT("set_inflight_fd num_queues: %"PRId16"\n", num_queues); 1781 DPRINT("set_inflight_fd queue_size: %"PRId16"\n", queue_size); 1782 1783 rc = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, 1784 fd, mmap_offset); 1785 1786 if (rc == MAP_FAILED) { 1787 vu_panic(dev, "set_inflight_fd mmap error: %s", strerror(errno)); 1788 return false; 1789 } 1790 1791 if (dev->inflight_info.fd) { 1792 close(dev->inflight_info.fd); 1793 } 1794 1795 if (dev->inflight_info.addr) { 1796 munmap(dev->inflight_info.addr, dev->inflight_info.size); 1797 } 1798 1799 dev->inflight_info.fd = fd; 1800 dev->inflight_info.addr = rc; 1801 dev->inflight_info.size = mmap_size; 1802 1803 for (i = 0; i < num_queues; i++) { 1804 dev->vq[i].inflight = (VuVirtqInflight *)rc; 1805 dev->vq[i].inflight->desc_num = queue_size; 1806 rc = (void *)((char *)rc + vu_inflight_queue_size(queue_size)); 1807 } 1808 1809 return false; 1810 } 1811 1812 static bool 1813 vu_handle_vring_kick(VuDev *dev, VhostUserMsg *vmsg) 1814 { 1815 unsigned int index = vmsg->payload.state.index; 1816 1817 if (index >= dev->max_queues) { 1818 vu_panic(dev, "Invalid queue index: %u", index); 1819 return false; 1820 } 1821 1822 DPRINT("Got kick message: handler:%p idx:%u\n", 1823 dev->vq[index].handler, index); 1824 1825 if (!dev->vq[index].started) { 1826 dev->vq[index].started = true; 1827 1828 if (dev->iface->queue_set_started) { 1829 dev->iface->queue_set_started(dev, index, true); 1830 } 1831 } 1832 1833 if (dev->vq[index].handler) { 1834 dev->vq[index].handler(dev, index); 1835 } 1836 1837 return false; 1838 } 1839 1840 static bool vu_handle_get_max_memslots(VuDev *dev, VhostUserMsg *vmsg) 1841 { 1842 vmsg_set_reply_u64(vmsg, VHOST_USER_MAX_RAM_SLOTS); 1843 1844 DPRINT("u64: 0x%016"PRIx64"\n", (uint64_t) VHOST_USER_MAX_RAM_SLOTS); 1845 1846 return true; 1847 } 1848 1849 static bool 1850 vu_process_message(VuDev *dev, VhostUserMsg *vmsg) 1851 { 1852 int do_reply = 0; 1853 1854 /* Print out generic part of the request. */ 1855 DPRINT("================ Vhost user message ================\n"); 1856 DPRINT("Request: %s (%d)\n", vu_request_to_string(vmsg->request), 1857 vmsg->request); 1858 DPRINT("Flags: 0x%x\n", vmsg->flags); 1859 DPRINT("Size: %u\n", vmsg->size); 1860 1861 if (vmsg->fd_num) { 1862 int i; 1863 DPRINT("Fds:"); 1864 for (i = 0; i < vmsg->fd_num; i++) { 1865 DPRINT(" %d", vmsg->fds[i]); 1866 } 1867 DPRINT("\n"); 1868 } 1869 1870 if (dev->iface->process_msg && 1871 dev->iface->process_msg(dev, vmsg, &do_reply)) { 1872 return do_reply; 1873 } 1874 1875 switch (vmsg->request) { 1876 case VHOST_USER_GET_FEATURES: 1877 return vu_get_features_exec(dev, vmsg); 1878 case VHOST_USER_SET_FEATURES: 1879 return vu_set_features_exec(dev, vmsg); 1880 case VHOST_USER_GET_PROTOCOL_FEATURES: 1881 return vu_get_protocol_features_exec(dev, vmsg); 1882 case VHOST_USER_SET_PROTOCOL_FEATURES: 1883 return vu_set_protocol_features_exec(dev, vmsg); 1884 case VHOST_USER_SET_OWNER: 1885 return vu_set_owner_exec(dev, vmsg); 1886 case VHOST_USER_RESET_OWNER: 1887 return vu_reset_device_exec(dev, vmsg); 1888 case VHOST_USER_SET_MEM_TABLE: 1889 return vu_set_mem_table_exec(dev, vmsg); 1890 case VHOST_USER_SET_LOG_BASE: 1891 return vu_set_log_base_exec(dev, vmsg); 1892 case VHOST_USER_SET_LOG_FD: 1893 return vu_set_log_fd_exec(dev, vmsg); 1894 case VHOST_USER_SET_VRING_NUM: 1895 return vu_set_vring_num_exec(dev, vmsg); 1896 case VHOST_USER_SET_VRING_ADDR: 1897 return vu_set_vring_addr_exec(dev, vmsg); 1898 case VHOST_USER_SET_VRING_BASE: 1899 return vu_set_vring_base_exec(dev, vmsg); 1900 case VHOST_USER_GET_VRING_BASE: 1901 return vu_get_vring_base_exec(dev, vmsg); 1902 case VHOST_USER_SET_VRING_KICK: 1903 return vu_set_vring_kick_exec(dev, vmsg); 1904 case VHOST_USER_SET_VRING_CALL: 1905 return vu_set_vring_call_exec(dev, vmsg); 1906 case VHOST_USER_SET_VRING_ERR: 1907 return vu_set_vring_err_exec(dev, vmsg); 1908 case VHOST_USER_GET_QUEUE_NUM: 1909 return vu_get_queue_num_exec(dev, vmsg); 1910 case VHOST_USER_SET_VRING_ENABLE: 1911 return vu_set_vring_enable_exec(dev, vmsg); 1912 case VHOST_USER_SET_SLAVE_REQ_FD: 1913 return vu_set_slave_req_fd(dev, vmsg); 1914 case VHOST_USER_GET_CONFIG: 1915 return vu_get_config(dev, vmsg); 1916 case VHOST_USER_SET_CONFIG: 1917 return vu_set_config(dev, vmsg); 1918 case VHOST_USER_NONE: 1919 /* if you need processing before exit, override iface->process_msg */ 1920 exit(0); 1921 case VHOST_USER_POSTCOPY_ADVISE: 1922 return vu_set_postcopy_advise(dev, vmsg); 1923 case VHOST_USER_POSTCOPY_LISTEN: 1924 return vu_set_postcopy_listen(dev, vmsg); 1925 case VHOST_USER_POSTCOPY_END: 1926 return vu_set_postcopy_end(dev, vmsg); 1927 case VHOST_USER_GET_INFLIGHT_FD: 1928 return vu_get_inflight_fd(dev, vmsg); 1929 case VHOST_USER_SET_INFLIGHT_FD: 1930 return vu_set_inflight_fd(dev, vmsg); 1931 case VHOST_USER_VRING_KICK: 1932 return vu_handle_vring_kick(dev, vmsg); 1933 case VHOST_USER_GET_MAX_MEM_SLOTS: 1934 return vu_handle_get_max_memslots(dev, vmsg); 1935 case VHOST_USER_ADD_MEM_REG: 1936 return vu_add_mem_reg(dev, vmsg); 1937 case VHOST_USER_REM_MEM_REG: 1938 return vu_rem_mem_reg(dev, vmsg); 1939 default: 1940 vmsg_close_fds(vmsg); 1941 vu_panic(dev, "Unhandled request: %d", vmsg->request); 1942 } 1943 1944 return false; 1945 } 1946 1947 bool 1948 vu_dispatch(VuDev *dev) 1949 { 1950 VhostUserMsg vmsg = { 0, }; 1951 int reply_requested; 1952 bool need_reply, success = false; 1953 1954 if (!dev->read_msg(dev, dev->sock, &vmsg)) { 1955 goto end; 1956 } 1957 1958 need_reply = vmsg.flags & VHOST_USER_NEED_REPLY_MASK; 1959 1960 reply_requested = vu_process_message(dev, &vmsg); 1961 if (!reply_requested && need_reply) { 1962 vmsg_set_reply_u64(&vmsg, 0); 1963 reply_requested = 1; 1964 } 1965 1966 if (!reply_requested) { 1967 success = true; 1968 goto end; 1969 } 1970 1971 if (!vu_send_reply(dev, dev->sock, &vmsg)) { 1972 goto end; 1973 } 1974 1975 success = true; 1976 1977 end: 1978 free(vmsg.data); 1979 return success; 1980 } 1981 1982 void 1983 vu_deinit(VuDev *dev) 1984 { 1985 unsigned int i; 1986 1987 for (i = 0; i < dev->nregions; i++) { 1988 VuDevRegion *r = &dev->regions[i]; 1989 void *m = (void *) (uintptr_t) r->mmap_addr; 1990 if (m != MAP_FAILED) { 1991 munmap(m, r->size + r->mmap_offset); 1992 } 1993 } 1994 dev->nregions = 0; 1995 1996 for (i = 0; i < dev->max_queues; i++) { 1997 VuVirtq *vq = &dev->vq[i]; 1998 1999 if (vq->call_fd != -1) { 2000 close(vq->call_fd); 2001 vq->call_fd = -1; 2002 } 2003 2004 if (vq->kick_fd != -1) { 2005 dev->remove_watch(dev, vq->kick_fd); 2006 close(vq->kick_fd); 2007 vq->kick_fd = -1; 2008 } 2009 2010 if (vq->err_fd != -1) { 2011 close(vq->err_fd); 2012 vq->err_fd = -1; 2013 } 2014 2015 if (vq->resubmit_list) { 2016 free(vq->resubmit_list); 2017 vq->resubmit_list = NULL; 2018 } 2019 2020 vq->inflight = NULL; 2021 } 2022 2023 if (dev->inflight_info.addr) { 2024 munmap(dev->inflight_info.addr, dev->inflight_info.size); 2025 dev->inflight_info.addr = NULL; 2026 } 2027 2028 if (dev->inflight_info.fd > 0) { 2029 close(dev->inflight_info.fd); 2030 dev->inflight_info.fd = -1; 2031 } 2032 2033 vu_close_log(dev); 2034 if (dev->slave_fd != -1) { 2035 close(dev->slave_fd); 2036 dev->slave_fd = -1; 2037 } 2038 pthread_mutex_destroy(&dev->slave_mutex); 2039 2040 if (dev->sock != -1) { 2041 close(dev->sock); 2042 } 2043 2044 free(dev->vq); 2045 dev->vq = NULL; 2046 } 2047 2048 bool 2049 vu_init(VuDev *dev, 2050 uint16_t max_queues, 2051 int socket, 2052 vu_panic_cb panic, 2053 vu_read_msg_cb read_msg, 2054 vu_set_watch_cb set_watch, 2055 vu_remove_watch_cb remove_watch, 2056 const VuDevIface *iface) 2057 { 2058 uint16_t i; 2059 2060 assert(max_queues > 0); 2061 assert(socket >= 0); 2062 assert(set_watch); 2063 assert(remove_watch); 2064 assert(iface); 2065 assert(panic); 2066 2067 memset(dev, 0, sizeof(*dev)); 2068 2069 dev->sock = socket; 2070 dev->panic = panic; 2071 dev->read_msg = read_msg ? read_msg : vu_message_read_default; 2072 dev->set_watch = set_watch; 2073 dev->remove_watch = remove_watch; 2074 dev->iface = iface; 2075 dev->log_call_fd = -1; 2076 pthread_mutex_init(&dev->slave_mutex, NULL); 2077 dev->slave_fd = -1; 2078 dev->max_queues = max_queues; 2079 2080 dev->vq = malloc(max_queues * sizeof(dev->vq[0])); 2081 if (!dev->vq) { 2082 DPRINT("%s: failed to malloc virtqueues\n", __func__); 2083 return false; 2084 } 2085 2086 for (i = 0; i < max_queues; i++) { 2087 dev->vq[i] = (VuVirtq) { 2088 .call_fd = -1, .kick_fd = -1, .err_fd = -1, 2089 .notification = true, 2090 }; 2091 } 2092 2093 return true; 2094 } 2095 2096 VuVirtq * 2097 vu_get_queue(VuDev *dev, int qidx) 2098 { 2099 assert(qidx < dev->max_queues); 2100 return &dev->vq[qidx]; 2101 } 2102 2103 bool 2104 vu_queue_enabled(VuDev *dev, VuVirtq *vq) 2105 { 2106 return vq->enable; 2107 } 2108 2109 bool 2110 vu_queue_started(const VuDev *dev, const VuVirtq *vq) 2111 { 2112 return vq->started; 2113 } 2114 2115 static inline uint16_t 2116 vring_avail_flags(VuVirtq *vq) 2117 { 2118 return le16toh(vq->vring.avail->flags); 2119 } 2120 2121 static inline uint16_t 2122 vring_avail_idx(VuVirtq *vq) 2123 { 2124 vq->shadow_avail_idx = le16toh(vq->vring.avail->idx); 2125 2126 return vq->shadow_avail_idx; 2127 } 2128 2129 static inline uint16_t 2130 vring_avail_ring(VuVirtq *vq, int i) 2131 { 2132 return le16toh(vq->vring.avail->ring[i]); 2133 } 2134 2135 static inline uint16_t 2136 vring_get_used_event(VuVirtq *vq) 2137 { 2138 return vring_avail_ring(vq, vq->vring.num); 2139 } 2140 2141 static int 2142 virtqueue_num_heads(VuDev *dev, VuVirtq *vq, unsigned int idx) 2143 { 2144 uint16_t num_heads = vring_avail_idx(vq) - idx; 2145 2146 /* Check it isn't doing very strange things with descriptor numbers. */ 2147 if (num_heads > vq->vring.num) { 2148 vu_panic(dev, "Guest moved used index from %u to %u", 2149 idx, vq->shadow_avail_idx); 2150 return -1; 2151 } 2152 if (num_heads) { 2153 /* On success, callers read a descriptor at vq->last_avail_idx. 2154 * Make sure descriptor read does not bypass avail index read. */ 2155 smp_rmb(); 2156 } 2157 2158 return num_heads; 2159 } 2160 2161 static bool 2162 virtqueue_get_head(VuDev *dev, VuVirtq *vq, 2163 unsigned int idx, unsigned int *head) 2164 { 2165 /* Grab the next descriptor number they're advertising, and increment 2166 * the index we've seen. */ 2167 *head = vring_avail_ring(vq, idx % vq->vring.num); 2168 2169 /* If their number is silly, that's a fatal mistake. */ 2170 if (*head >= vq->vring.num) { 2171 vu_panic(dev, "Guest says index %u is available", *head); 2172 return false; 2173 } 2174 2175 return true; 2176 } 2177 2178 static int 2179 virtqueue_read_indirect_desc(VuDev *dev, struct vring_desc *desc, 2180 uint64_t addr, size_t len) 2181 { 2182 struct vring_desc *ori_desc; 2183 uint64_t read_len; 2184 2185 if (len > (VIRTQUEUE_MAX_SIZE * sizeof(struct vring_desc))) { 2186 return -1; 2187 } 2188 2189 if (len == 0) { 2190 return -1; 2191 } 2192 2193 while (len) { 2194 read_len = len; 2195 ori_desc = vu_gpa_to_va(dev, &read_len, addr); 2196 if (!ori_desc) { 2197 return -1; 2198 } 2199 2200 memcpy(desc, ori_desc, read_len); 2201 len -= read_len; 2202 addr += read_len; 2203 desc += read_len; 2204 } 2205 2206 return 0; 2207 } 2208 2209 enum { 2210 VIRTQUEUE_READ_DESC_ERROR = -1, 2211 VIRTQUEUE_READ_DESC_DONE = 0, /* end of chain */ 2212 VIRTQUEUE_READ_DESC_MORE = 1, /* more buffers in chain */ 2213 }; 2214 2215 static int 2216 virtqueue_read_next_desc(VuDev *dev, struct vring_desc *desc, 2217 int i, unsigned int max, unsigned int *next) 2218 { 2219 /* If this descriptor says it doesn't chain, we're done. */ 2220 if (!(le16toh(desc[i].flags) & VRING_DESC_F_NEXT)) { 2221 return VIRTQUEUE_READ_DESC_DONE; 2222 } 2223 2224 /* Check they're not leading us off end of descriptors. */ 2225 *next = le16toh(desc[i].next); 2226 /* Make sure compiler knows to grab that: we don't want it changing! */ 2227 smp_wmb(); 2228 2229 if (*next >= max) { 2230 vu_panic(dev, "Desc next is %u", *next); 2231 return VIRTQUEUE_READ_DESC_ERROR; 2232 } 2233 2234 return VIRTQUEUE_READ_DESC_MORE; 2235 } 2236 2237 void 2238 vu_queue_get_avail_bytes(VuDev *dev, VuVirtq *vq, unsigned int *in_bytes, 2239 unsigned int *out_bytes, 2240 unsigned max_in_bytes, unsigned max_out_bytes) 2241 { 2242 unsigned int idx; 2243 unsigned int total_bufs, in_total, out_total; 2244 int rc; 2245 2246 idx = vq->last_avail_idx; 2247 2248 total_bufs = in_total = out_total = 0; 2249 if (unlikely(dev->broken) || 2250 unlikely(!vq->vring.avail)) { 2251 goto done; 2252 } 2253 2254 while ((rc = virtqueue_num_heads(dev, vq, idx)) > 0) { 2255 unsigned int max, desc_len, num_bufs, indirect = 0; 2256 uint64_t desc_addr, read_len; 2257 struct vring_desc *desc; 2258 struct vring_desc desc_buf[VIRTQUEUE_MAX_SIZE]; 2259 unsigned int i; 2260 2261 max = vq->vring.num; 2262 num_bufs = total_bufs; 2263 if (!virtqueue_get_head(dev, vq, idx++, &i)) { 2264 goto err; 2265 } 2266 desc = vq->vring.desc; 2267 2268 if (le16toh(desc[i].flags) & VRING_DESC_F_INDIRECT) { 2269 if (le32toh(desc[i].len) % sizeof(struct vring_desc)) { 2270 vu_panic(dev, "Invalid size for indirect buffer table"); 2271 goto err; 2272 } 2273 2274 /* If we've got too many, that implies a descriptor loop. */ 2275 if (num_bufs >= max) { 2276 vu_panic(dev, "Looped descriptor"); 2277 goto err; 2278 } 2279 2280 /* loop over the indirect descriptor table */ 2281 indirect = 1; 2282 desc_addr = le64toh(desc[i].addr); 2283 desc_len = le32toh(desc[i].len); 2284 max = desc_len / sizeof(struct vring_desc); 2285 read_len = desc_len; 2286 desc = vu_gpa_to_va(dev, &read_len, desc_addr); 2287 if (unlikely(desc && read_len != desc_len)) { 2288 /* Failed to use zero copy */ 2289 desc = NULL; 2290 if (!virtqueue_read_indirect_desc(dev, desc_buf, 2291 desc_addr, 2292 desc_len)) { 2293 desc = desc_buf; 2294 } 2295 } 2296 if (!desc) { 2297 vu_panic(dev, "Invalid indirect buffer table"); 2298 goto err; 2299 } 2300 num_bufs = i = 0; 2301 } 2302 2303 do { 2304 /* If we've got too many, that implies a descriptor loop. */ 2305 if (++num_bufs > max) { 2306 vu_panic(dev, "Looped descriptor"); 2307 goto err; 2308 } 2309 2310 if (le16toh(desc[i].flags) & VRING_DESC_F_WRITE) { 2311 in_total += le32toh(desc[i].len); 2312 } else { 2313 out_total += le32toh(desc[i].len); 2314 } 2315 if (in_total >= max_in_bytes && out_total >= max_out_bytes) { 2316 goto done; 2317 } 2318 rc = virtqueue_read_next_desc(dev, desc, i, max, &i); 2319 } while (rc == VIRTQUEUE_READ_DESC_MORE); 2320 2321 if (rc == VIRTQUEUE_READ_DESC_ERROR) { 2322 goto err; 2323 } 2324 2325 if (!indirect) { 2326 total_bufs = num_bufs; 2327 } else { 2328 total_bufs++; 2329 } 2330 } 2331 if (rc < 0) { 2332 goto err; 2333 } 2334 done: 2335 if (in_bytes) { 2336 *in_bytes = in_total; 2337 } 2338 if (out_bytes) { 2339 *out_bytes = out_total; 2340 } 2341 return; 2342 2343 err: 2344 in_total = out_total = 0; 2345 goto done; 2346 } 2347 2348 bool 2349 vu_queue_avail_bytes(VuDev *dev, VuVirtq *vq, unsigned int in_bytes, 2350 unsigned int out_bytes) 2351 { 2352 unsigned int in_total, out_total; 2353 2354 vu_queue_get_avail_bytes(dev, vq, &in_total, &out_total, 2355 in_bytes, out_bytes); 2356 2357 return in_bytes <= in_total && out_bytes <= out_total; 2358 } 2359 2360 /* Fetch avail_idx from VQ memory only when we really need to know if 2361 * guest has added some buffers. */ 2362 bool 2363 vu_queue_empty(VuDev *dev, VuVirtq *vq) 2364 { 2365 if (unlikely(dev->broken) || 2366 unlikely(!vq->vring.avail)) { 2367 return true; 2368 } 2369 2370 if (vq->shadow_avail_idx != vq->last_avail_idx) { 2371 return false; 2372 } 2373 2374 return vring_avail_idx(vq) == vq->last_avail_idx; 2375 } 2376 2377 static bool 2378 vring_notify(VuDev *dev, VuVirtq *vq) 2379 { 2380 uint16_t old, new; 2381 bool v; 2382 2383 /* We need to expose used array entries before checking used event. */ 2384 smp_mb(); 2385 2386 /* Always notify when queue is empty (when feature acknowledge) */ 2387 if (vu_has_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY) && 2388 !vq->inuse && vu_queue_empty(dev, vq)) { 2389 return true; 2390 } 2391 2392 if (!vu_has_feature(dev, VIRTIO_RING_F_EVENT_IDX)) { 2393 return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT); 2394 } 2395 2396 v = vq->signalled_used_valid; 2397 vq->signalled_used_valid = true; 2398 old = vq->signalled_used; 2399 new = vq->signalled_used = vq->used_idx; 2400 return !v || vring_need_event(vring_get_used_event(vq), new, old); 2401 } 2402 2403 static void _vu_queue_notify(VuDev *dev, VuVirtq *vq, bool sync) 2404 { 2405 if (unlikely(dev->broken) || 2406 unlikely(!vq->vring.avail)) { 2407 return; 2408 } 2409 2410 if (!vring_notify(dev, vq)) { 2411 DPRINT("skipped notify...\n"); 2412 return; 2413 } 2414 2415 if (vq->call_fd < 0 && 2416 vu_has_protocol_feature(dev, 2417 VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS) && 2418 vu_has_protocol_feature(dev, VHOST_USER_PROTOCOL_F_SLAVE_REQ)) { 2419 VhostUserMsg vmsg = { 2420 .request = VHOST_USER_SLAVE_VRING_CALL, 2421 .flags = VHOST_USER_VERSION, 2422 .size = sizeof(vmsg.payload.state), 2423 .payload.state = { 2424 .index = vq - dev->vq, 2425 }, 2426 }; 2427 bool ack = sync && 2428 vu_has_protocol_feature(dev, 2429 VHOST_USER_PROTOCOL_F_REPLY_ACK); 2430 2431 if (ack) { 2432 vmsg.flags |= VHOST_USER_NEED_REPLY_MASK; 2433 } 2434 2435 vu_message_write(dev, dev->slave_fd, &vmsg); 2436 if (ack) { 2437 vu_message_read_default(dev, dev->slave_fd, &vmsg); 2438 } 2439 return; 2440 } 2441 2442 if (eventfd_write(vq->call_fd, 1) < 0) { 2443 vu_panic(dev, "Error writing eventfd: %s", strerror(errno)); 2444 } 2445 } 2446 2447 void vu_queue_notify(VuDev *dev, VuVirtq *vq) 2448 { 2449 _vu_queue_notify(dev, vq, false); 2450 } 2451 2452 void vu_queue_notify_sync(VuDev *dev, VuVirtq *vq) 2453 { 2454 _vu_queue_notify(dev, vq, true); 2455 } 2456 2457 static inline void 2458 vring_used_flags_set_bit(VuVirtq *vq, int mask) 2459 { 2460 uint16_t *flags; 2461 2462 flags = (uint16_t *)((char*)vq->vring.used + 2463 offsetof(struct vring_used, flags)); 2464 *flags = htole16(le16toh(*flags) | mask); 2465 } 2466 2467 static inline void 2468 vring_used_flags_unset_bit(VuVirtq *vq, int mask) 2469 { 2470 uint16_t *flags; 2471 2472 flags = (uint16_t *)((char*)vq->vring.used + 2473 offsetof(struct vring_used, flags)); 2474 *flags = htole16(le16toh(*flags) & ~mask); 2475 } 2476 2477 static inline void 2478 vring_set_avail_event(VuVirtq *vq, uint16_t val) 2479 { 2480 uint16_t *avail; 2481 2482 if (!vq->notification) { 2483 return; 2484 } 2485 2486 avail = (uint16_t *)&vq->vring.used->ring[vq->vring.num]; 2487 *avail = htole16(val); 2488 } 2489 2490 void 2491 vu_queue_set_notification(VuDev *dev, VuVirtq *vq, int enable) 2492 { 2493 vq->notification = enable; 2494 if (vu_has_feature(dev, VIRTIO_RING_F_EVENT_IDX)) { 2495 vring_set_avail_event(vq, vring_avail_idx(vq)); 2496 } else if (enable) { 2497 vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY); 2498 } else { 2499 vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY); 2500 } 2501 if (enable) { 2502 /* Expose avail event/used flags before caller checks the avail idx. */ 2503 smp_mb(); 2504 } 2505 } 2506 2507 static bool 2508 virtqueue_map_desc(VuDev *dev, 2509 unsigned int *p_num_sg, struct iovec *iov, 2510 unsigned int max_num_sg, bool is_write, 2511 uint64_t pa, size_t sz) 2512 { 2513 unsigned num_sg = *p_num_sg; 2514 2515 assert(num_sg <= max_num_sg); 2516 2517 if (!sz) { 2518 vu_panic(dev, "virtio: zero sized buffers are not allowed"); 2519 return false; 2520 } 2521 2522 while (sz) { 2523 uint64_t len = sz; 2524 2525 if (num_sg == max_num_sg) { 2526 vu_panic(dev, "virtio: too many descriptors in indirect table"); 2527 return false; 2528 } 2529 2530 iov[num_sg].iov_base = vu_gpa_to_va(dev, &len, pa); 2531 if (iov[num_sg].iov_base == NULL) { 2532 vu_panic(dev, "virtio: invalid address for buffers"); 2533 return false; 2534 } 2535 iov[num_sg].iov_len = len; 2536 num_sg++; 2537 sz -= len; 2538 pa += len; 2539 } 2540 2541 *p_num_sg = num_sg; 2542 return true; 2543 } 2544 2545 static void * 2546 virtqueue_alloc_element(size_t sz, 2547 unsigned out_num, unsigned in_num) 2548 { 2549 VuVirtqElement *elem; 2550 size_t in_sg_ofs = ALIGN_UP(sz, __alignof__(elem->in_sg[0])); 2551 size_t out_sg_ofs = in_sg_ofs + in_num * sizeof(elem->in_sg[0]); 2552 size_t out_sg_end = out_sg_ofs + out_num * sizeof(elem->out_sg[0]); 2553 2554 assert(sz >= sizeof(VuVirtqElement)); 2555 elem = malloc(out_sg_end); 2556 elem->out_num = out_num; 2557 elem->in_num = in_num; 2558 elem->in_sg = (void *)elem + in_sg_ofs; 2559 elem->out_sg = (void *)elem + out_sg_ofs; 2560 return elem; 2561 } 2562 2563 static void * 2564 vu_queue_map_desc(VuDev *dev, VuVirtq *vq, unsigned int idx, size_t sz) 2565 { 2566 struct vring_desc *desc = vq->vring.desc; 2567 uint64_t desc_addr, read_len; 2568 unsigned int desc_len; 2569 unsigned int max = vq->vring.num; 2570 unsigned int i = idx; 2571 VuVirtqElement *elem; 2572 unsigned int out_num = 0, in_num = 0; 2573 struct iovec iov[VIRTQUEUE_MAX_SIZE]; 2574 struct vring_desc desc_buf[VIRTQUEUE_MAX_SIZE]; 2575 int rc; 2576 2577 if (le16toh(desc[i].flags) & VRING_DESC_F_INDIRECT) { 2578 if (le32toh(desc[i].len) % sizeof(struct vring_desc)) { 2579 vu_panic(dev, "Invalid size for indirect buffer table"); 2580 return NULL; 2581 } 2582 2583 /* loop over the indirect descriptor table */ 2584 desc_addr = le64toh(desc[i].addr); 2585 desc_len = le32toh(desc[i].len); 2586 max = desc_len / sizeof(struct vring_desc); 2587 read_len = desc_len; 2588 desc = vu_gpa_to_va(dev, &read_len, desc_addr); 2589 if (unlikely(desc && read_len != desc_len)) { 2590 /* Failed to use zero copy */ 2591 desc = NULL; 2592 if (!virtqueue_read_indirect_desc(dev, desc_buf, 2593 desc_addr, 2594 desc_len)) { 2595 desc = desc_buf; 2596 } 2597 } 2598 if (!desc) { 2599 vu_panic(dev, "Invalid indirect buffer table"); 2600 return NULL; 2601 } 2602 i = 0; 2603 } 2604 2605 /* Collect all the descriptors */ 2606 do { 2607 if (le16toh(desc[i].flags) & VRING_DESC_F_WRITE) { 2608 if (!virtqueue_map_desc(dev, &in_num, iov + out_num, 2609 VIRTQUEUE_MAX_SIZE - out_num, true, 2610 le64toh(desc[i].addr), 2611 le32toh(desc[i].len))) { 2612 return NULL; 2613 } 2614 } else { 2615 if (in_num) { 2616 vu_panic(dev, "Incorrect order for descriptors"); 2617 return NULL; 2618 } 2619 if (!virtqueue_map_desc(dev, &out_num, iov, 2620 VIRTQUEUE_MAX_SIZE, false, 2621 le64toh(desc[i].addr), 2622 le32toh(desc[i].len))) { 2623 return NULL; 2624 } 2625 } 2626 2627 /* If we've got too many, that implies a descriptor loop. */ 2628 if ((in_num + out_num) > max) { 2629 vu_panic(dev, "Looped descriptor"); 2630 return NULL; 2631 } 2632 rc = virtqueue_read_next_desc(dev, desc, i, max, &i); 2633 } while (rc == VIRTQUEUE_READ_DESC_MORE); 2634 2635 if (rc == VIRTQUEUE_READ_DESC_ERROR) { 2636 vu_panic(dev, "read descriptor error"); 2637 return NULL; 2638 } 2639 2640 /* Now copy what we have collected and mapped */ 2641 elem = virtqueue_alloc_element(sz, out_num, in_num); 2642 elem->index = idx; 2643 for (i = 0; i < out_num; i++) { 2644 elem->out_sg[i] = iov[i]; 2645 } 2646 for (i = 0; i < in_num; i++) { 2647 elem->in_sg[i] = iov[out_num + i]; 2648 } 2649 2650 return elem; 2651 } 2652 2653 static int 2654 vu_queue_inflight_get(VuDev *dev, VuVirtq *vq, int desc_idx) 2655 { 2656 if (!vu_has_protocol_feature(dev, VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)) { 2657 return 0; 2658 } 2659 2660 if (unlikely(!vq->inflight)) { 2661 return -1; 2662 } 2663 2664 vq->inflight->desc[desc_idx].counter = vq->counter++; 2665 vq->inflight->desc[desc_idx].inflight = 1; 2666 2667 return 0; 2668 } 2669 2670 static int 2671 vu_queue_inflight_pre_put(VuDev *dev, VuVirtq *vq, int desc_idx) 2672 { 2673 if (!vu_has_protocol_feature(dev, VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)) { 2674 return 0; 2675 } 2676 2677 if (unlikely(!vq->inflight)) { 2678 return -1; 2679 } 2680 2681 vq->inflight->last_batch_head = desc_idx; 2682 2683 return 0; 2684 } 2685 2686 static int 2687 vu_queue_inflight_post_put(VuDev *dev, VuVirtq *vq, int desc_idx) 2688 { 2689 if (!vu_has_protocol_feature(dev, VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)) { 2690 return 0; 2691 } 2692 2693 if (unlikely(!vq->inflight)) { 2694 return -1; 2695 } 2696 2697 barrier(); 2698 2699 vq->inflight->desc[desc_idx].inflight = 0; 2700 2701 barrier(); 2702 2703 vq->inflight->used_idx = vq->used_idx; 2704 2705 return 0; 2706 } 2707 2708 void * 2709 vu_queue_pop(VuDev *dev, VuVirtq *vq, size_t sz) 2710 { 2711 int i; 2712 unsigned int head; 2713 VuVirtqElement *elem; 2714 2715 if (unlikely(dev->broken) || 2716 unlikely(!vq->vring.avail)) { 2717 return NULL; 2718 } 2719 2720 if (unlikely(vq->resubmit_list && vq->resubmit_num > 0)) { 2721 i = (--vq->resubmit_num); 2722 elem = vu_queue_map_desc(dev, vq, vq->resubmit_list[i].index, sz); 2723 2724 if (!vq->resubmit_num) { 2725 free(vq->resubmit_list); 2726 vq->resubmit_list = NULL; 2727 } 2728 2729 return elem; 2730 } 2731 2732 if (vu_queue_empty(dev, vq)) { 2733 return NULL; 2734 } 2735 /* 2736 * Needed after virtio_queue_empty(), see comment in 2737 * virtqueue_num_heads(). 2738 */ 2739 smp_rmb(); 2740 2741 if (vq->inuse >= vq->vring.num) { 2742 vu_panic(dev, "Virtqueue size exceeded"); 2743 return NULL; 2744 } 2745 2746 if (!virtqueue_get_head(dev, vq, vq->last_avail_idx++, &head)) { 2747 return NULL; 2748 } 2749 2750 if (vu_has_feature(dev, VIRTIO_RING_F_EVENT_IDX)) { 2751 vring_set_avail_event(vq, vq->last_avail_idx); 2752 } 2753 2754 elem = vu_queue_map_desc(dev, vq, head, sz); 2755 2756 if (!elem) { 2757 return NULL; 2758 } 2759 2760 vq->inuse++; 2761 2762 vu_queue_inflight_get(dev, vq, head); 2763 2764 return elem; 2765 } 2766 2767 static void 2768 vu_queue_detach_element(VuDev *dev, VuVirtq *vq, VuVirtqElement *elem, 2769 size_t len) 2770 { 2771 vq->inuse--; 2772 /* unmap, when DMA support is added */ 2773 } 2774 2775 void 2776 vu_queue_unpop(VuDev *dev, VuVirtq *vq, VuVirtqElement *elem, 2777 size_t len) 2778 { 2779 vq->last_avail_idx--; 2780 vu_queue_detach_element(dev, vq, elem, len); 2781 } 2782 2783 bool 2784 vu_queue_rewind(VuDev *dev, VuVirtq *vq, unsigned int num) 2785 { 2786 if (num > vq->inuse) { 2787 return false; 2788 } 2789 vq->last_avail_idx -= num; 2790 vq->inuse -= num; 2791 return true; 2792 } 2793 2794 static inline 2795 void vring_used_write(VuDev *dev, VuVirtq *vq, 2796 struct vring_used_elem *uelem, int i) 2797 { 2798 struct vring_used *used = vq->vring.used; 2799 2800 used->ring[i] = *uelem; 2801 vu_log_write(dev, vq->vring.log_guest_addr + 2802 offsetof(struct vring_used, ring[i]), 2803 sizeof(used->ring[i])); 2804 } 2805 2806 2807 static void 2808 vu_log_queue_fill(VuDev *dev, VuVirtq *vq, 2809 const VuVirtqElement *elem, 2810 unsigned int len) 2811 { 2812 struct vring_desc *desc = vq->vring.desc; 2813 unsigned int i, max, min, desc_len; 2814 uint64_t desc_addr, read_len; 2815 struct vring_desc desc_buf[VIRTQUEUE_MAX_SIZE]; 2816 unsigned num_bufs = 0; 2817 2818 max = vq->vring.num; 2819 i = elem->index; 2820 2821 if (le16toh(desc[i].flags) & VRING_DESC_F_INDIRECT) { 2822 if (le32toh(desc[i].len) % sizeof(struct vring_desc)) { 2823 vu_panic(dev, "Invalid size for indirect buffer table"); 2824 return; 2825 } 2826 2827 /* loop over the indirect descriptor table */ 2828 desc_addr = le64toh(desc[i].addr); 2829 desc_len = le32toh(desc[i].len); 2830 max = desc_len / sizeof(struct vring_desc); 2831 read_len = desc_len; 2832 desc = vu_gpa_to_va(dev, &read_len, desc_addr); 2833 if (unlikely(desc && read_len != desc_len)) { 2834 /* Failed to use zero copy */ 2835 desc = NULL; 2836 if (!virtqueue_read_indirect_desc(dev, desc_buf, 2837 desc_addr, 2838 desc_len)) { 2839 desc = desc_buf; 2840 } 2841 } 2842 if (!desc) { 2843 vu_panic(dev, "Invalid indirect buffer table"); 2844 return; 2845 } 2846 i = 0; 2847 } 2848 2849 do { 2850 if (++num_bufs > max) { 2851 vu_panic(dev, "Looped descriptor"); 2852 return; 2853 } 2854 2855 if (le16toh(desc[i].flags) & VRING_DESC_F_WRITE) { 2856 min = MIN(le32toh(desc[i].len), len); 2857 vu_log_write(dev, le64toh(desc[i].addr), min); 2858 len -= min; 2859 } 2860 2861 } while (len > 0 && 2862 (virtqueue_read_next_desc(dev, desc, i, max, &i) 2863 == VIRTQUEUE_READ_DESC_MORE)); 2864 } 2865 2866 void 2867 vu_queue_fill(VuDev *dev, VuVirtq *vq, 2868 const VuVirtqElement *elem, 2869 unsigned int len, unsigned int idx) 2870 { 2871 struct vring_used_elem uelem; 2872 2873 if (unlikely(dev->broken) || 2874 unlikely(!vq->vring.avail)) { 2875 return; 2876 } 2877 2878 vu_log_queue_fill(dev, vq, elem, len); 2879 2880 idx = (idx + vq->used_idx) % vq->vring.num; 2881 2882 uelem.id = htole32(elem->index); 2883 uelem.len = htole32(len); 2884 vring_used_write(dev, vq, &uelem, idx); 2885 } 2886 2887 static inline 2888 void vring_used_idx_set(VuDev *dev, VuVirtq *vq, uint16_t val) 2889 { 2890 vq->vring.used->idx = htole16(val); 2891 vu_log_write(dev, 2892 vq->vring.log_guest_addr + offsetof(struct vring_used, idx), 2893 sizeof(vq->vring.used->idx)); 2894 2895 vq->used_idx = val; 2896 } 2897 2898 void 2899 vu_queue_flush(VuDev *dev, VuVirtq *vq, unsigned int count) 2900 { 2901 uint16_t old, new; 2902 2903 if (unlikely(dev->broken) || 2904 unlikely(!vq->vring.avail)) { 2905 return; 2906 } 2907 2908 /* Make sure buffer is written before we update index. */ 2909 smp_wmb(); 2910 2911 old = vq->used_idx; 2912 new = old + count; 2913 vring_used_idx_set(dev, vq, new); 2914 vq->inuse -= count; 2915 if (unlikely((int16_t)(new - vq->signalled_used) < (uint16_t)(new - old))) { 2916 vq->signalled_used_valid = false; 2917 } 2918 } 2919 2920 void 2921 vu_queue_push(VuDev *dev, VuVirtq *vq, 2922 const VuVirtqElement *elem, unsigned int len) 2923 { 2924 vu_queue_fill(dev, vq, elem, len, 0); 2925 vu_queue_inflight_pre_put(dev, vq, elem->index); 2926 vu_queue_flush(dev, vq, 1); 2927 vu_queue_inflight_post_put(dev, vq, elem->index); 2928 } 2929