drivers/vhost/vhost.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Copyright (C) 2009 Red Hat, Inc.
   3  * Copyright (C) 2006 Rusty Russell IBM Corporation
   4  *
   5  * Author: Michael S. Tsirkin <mst@redhat.com>
   6  *
   7  * Inspiration, some code, and most witty comments come from
   8  * Documentation/virtual/lguest/lguest.c, by Rusty Russell
   9  *
  10  * Generic code for virtio server in host kernel.
  11  */
  12
  13 #include <linux/eventfd.h>
  14 #include <linux/vhost.h>
  15 #include <linux/uio.h>
  16 #include <linux/mm.h>
  17 #include <linux/mmu_context.h>
  18 #include <linux/miscdevice.h>
  19 #include <linux/mutex.h>
  20 #include <linux/poll.h>
  21 #include <linux/file.h>
  22 #include <linux/highmem.h>
  23 #include <linux/slab.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/kthread.h>
  26 #include <linux/cgroup.h>
  27 #include <linux/module.h>
  28 #include <linux/sort.h>
  29 #include <linux/sched/mm.h>
  30 #include <linux/sched/signal.h>
  31 #include <linux/interval_tree_generic.h>
  32 #include <linux/nospec.h>
  33
  34 #include "vhost.h"
  35
  36 static ushort max_mem_regions = 64;
  37 module_param(max_mem_regions, ushort, 0444);
  38 MODULE_PARM_DESC(max_mem_regions,
  39         "Maximum number of memory regions in memory map. (default: 64)");
  40 static int max_iotlb_entries = 2048;
  41 module_param(max_iotlb_entries, int, 0444);
  42 MODULE_PARM_DESC(max_iotlb_entries,
  43         "Maximum number of iotlb entries. (default: 2048)");
  44
  45 enum {
  46         VHOST_MEMORY_F_LOG = 0x1,
  47 };
  48
  49 #define vhost_used_event(vq) ((__virtio16 __user *)&vq->avail->ring[vq->num])
  50 #define vhost_avail_event(vq) ((__virtio16 __user *)&vq->used->ring[vq->num])
  51
  52 INTERVAL_TREE_DEFINE(struct vhost_umem_node,
  53                      rb, __u64, __subtree_last,
  54                      START, LAST, static inline, vhost_umem_interval_tree);
  55
  56 #ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
  57 static void vhost_disable_cross_endian(struct vhost_virtqueue *vq)
  58 {
  59         vq->user_be = !virtio_legacy_is_little_endian();
  60 }
  61
  62 static void vhost_enable_cross_endian_big(struct vhost_virtqueue *vq)
  63 {
  64         vq->user_be = true;
  65 }
  66
  67 static void vhost_enable_cross_endian_little(struct vhost_virtqueue *vq)
  68 {
  69         vq->user_be = false;
  70 }
  71
  72 static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp)
  73 {
  74         struct vhost_vring_state s;
  75
  76         if (vq->private_data)
  77                 return -EBUSY;
  78
  79         if (copy_from_user(&s, argp, sizeof(s)))
  80                 return -EFAULT;
  81
  82         if (s.num != VHOST_VRING_LITTLE_ENDIAN &&
  83             s.num != VHOST_VRING_BIG_ENDIAN)
  84                 return -EINVAL;
  85
  86         if (s.num == VHOST_VRING_BIG_ENDIAN)
  87                 vhost_enable_cross_endian_big(vq);
  88         else
  89                 vhost_enable_cross_endian_little(vq);
  90
  91         return 0;
  92 }
  93
  94 static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx,
  95                                    int __user *argp)
  96 {
  97         struct vhost_vring_state s = {
  98                 .index = idx,
  99                 .num = vq->user_be
 100         };
 101
 102         if (copy_to_user(argp, &s, sizeof(s)))
 103                 return -EFAULT;
 104
 105         return 0;
 106 }
 107
 108 static void vhost_init_is_le(struct vhost_virtqueue *vq)
 109 {
 110         /* Note for legacy virtio: user_be is initialized at reset time
 111          * according to the host endianness. If userspace does not set an
 112          * explicit endianness, the default behavior is native endian, as
 113          * expected by legacy virtio.
 114          */
 115         vq->is_le = vhost_has_feature(vq, VIRTIO_F_VERSION_1) || !vq->user_be;
 116 }
 117 #else
 118 static void vhost_disable_cross_endian(struct vhost_virtqueue *vq)
 119 {
 120 }
 121
 122 static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp)
 123 {
 124         return -ENOIOCTLCMD;
 125 }
 126
 127 static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx,
 128                                    int __user *argp)
 129 {
 130         return -ENOIOCTLCMD;
 131 }
 132
 133 static void vhost_init_is_le(struct vhost_virtqueue *vq)
 134 {
 135         vq->is_le = vhost_has_feature(vq, VIRTIO_F_VERSION_1)
 136                 || virtio_legacy_is_little_endian();
 137 }
 138 #endif /* CONFIG_VHOST_CROSS_ENDIAN_LEGACY */
 139
 140 static void vhost_reset_is_le(struct vhost_virtqueue *vq)
 141 {
 142         vhost_init_is_le(vq);
 143 }
 144
 145 struct vhost_flush_struct {
 146         struct vhost_work work;
 147         struct completion wait_event;
 148 };
 149
 150 static void vhost_flush_work(struct vhost_work *work)
 151 {
 152         struct vhost_flush_struct *s;
 153
 154         s = container_of(work, struct vhost_flush_struct, work);
 155         complete(&s->wait_event);
 156 }
 157
 158 static void vhost_poll_func(struct file *file, wait_queue_head_t *wqh,
 159                             poll_table *pt)
 160 {
 161         struct vhost_poll *poll;
 162
 163         poll = container_of(pt, struct vhost_poll, table);
 164         poll->wqh = wqh;
 165         add_wait_queue(wqh, &poll->wait);
 166 }
 167
 168 static int vhost_poll_wakeup(wait_queue_entry_t *wait, unsigned mode, int sync,
 169                              void *key)
 170 {
 171         struct vhost_poll *poll = container_of(wait, struct vhost_poll, wait);
 172
 173         if (!(key_to_poll(key) & poll->mask))
 174                 return 0;
 175
 176         vhost_poll_queue(poll);
 177         return 0;
 178 }
 179
 180 void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn)
 181 {
 182         clear_bit(VHOST_WORK_QUEUED, &work->flags);
 183         work->fn = fn;
 184 }
 185 EXPORT_SYMBOL_GPL(vhost_work_init);
 186
 187 /* Init poll structure */
 188 void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn,
 189                      __poll_t mask, struct vhost_dev *dev)
 190 {
 191         init_waitqueue_func_entry(&poll->wait, vhost_poll_wakeup);
 192         init_poll_funcptr(&poll->table, vhost_poll_func);
 193         poll->mask = mask;
 194         poll->dev = dev;
 195         poll->wqh = NULL;
 196
 197         vhost_work_init(&poll->work, fn);
 198 }
 199 EXPORT_SYMBOL_GPL(vhost_poll_init);
 200
 201 /* Start polling a file. We add ourselves to file's wait queue. The caller must
 202  * keep a reference to a file until after vhost_poll_stop is called. */
 203 int vhost_poll_start(struct vhost_poll *poll, struct file *file)
 204 {
 205         __poll_t mask;
 206         int ret = 0;
 207
 208         if (poll->wqh)
 209                 return 0;
 210
 211         mask = vfs_poll(file, &poll->table);
 212         if (mask)
 213                 vhost_poll_wakeup(&poll->wait, 0, 0, poll_to_key(mask));
 214         if (mask & EPOLLERR) {
 215                 vhost_poll_stop(poll);
 216                 ret = -EINVAL;
 217         }
 218
 219         return ret;
 220 }
 221 EXPORT_SYMBOL_GPL(vhost_poll_start);
 222
 223 /* Stop polling a file. After this function returns, it becomes safe to drop the
 224  * file reference. You must also flush afterwards. */
 225 void vhost_poll_stop(struct vhost_poll *poll)
 226 {
 227         if (poll->wqh) {
 228                 remove_wait_queue(poll->wqh, &poll->wait);
 229                 poll->wqh = NULL;
 230         }
 231 }
 232 EXPORT_SYMBOL_GPL(vhost_poll_stop);
 233
 234 void vhost_work_flush(struct vhost_dev *dev, struct vhost_work *work)
 235 {
 236         struct vhost_flush_struct flush;
 237
 238         if (dev->worker) {
 239                 init_completion(&flush.wait_event);
 240                 vhost_work_init(&flush.work, vhost_flush_work);
 241
 242                 vhost_work_queue(dev, &flush.work);
 243                 wait_for_completion(&flush.wait_event);
 244         }
 245 }
 246 EXPORT_SYMBOL_GPL(vhost_work_flush);
 247
 248 /* Flush any work that has been scheduled. When calling this, don't hold any
 249  * locks that are also used by the callback. */
 250 void vhost_poll_flush(struct vhost_poll *poll)
 251 {
 252         vhost_work_flush(poll->dev, &poll->work);
 253 }
 254 EXPORT_SYMBOL_GPL(vhost_poll_flush);
 255
 256 void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work)
 257 {
 258         if (!dev->worker)
 259                 return;
 260
 261         if (!test_and_set_bit(VHOST_WORK_QUEUED, &work->flags)) {
 262                 /* We can only add the work to the list after we're
 263                  * sure it was not in the list.
 264                  * test_and_set_bit() implies a memory barrier.
 265                  */
 266                 llist_add(&work->node, &dev->work_list);
 267                 wake_up_process(dev->worker);
 268         }
 269 }
 270 EXPORT_SYMBOL_GPL(vhost_work_queue);
 271
 272 /* A lockless hint for busy polling code to exit the loop */
 273 bool vhost_has_work(struct vhost_dev *dev)
 274 {
 275         return !llist_empty(&dev->work_list);
 276 }
 277 EXPORT_SYMBOL_GPL(vhost_has_work);
 278
 279 void vhost_poll_queue(struct vhost_poll *poll)
 280 {
 281         vhost_work_queue(poll->dev, &poll->work);
 282 }
 283 EXPORT_SYMBOL_GPL(vhost_poll_queue);
 284
 285 static void __vhost_vq_meta_reset(struct vhost_virtqueue *vq)
 286 {
 287         int j;
 288
 289         for (j = 0; j < VHOST_NUM_ADDRS; j++)
 290                 vq->meta_iotlb[j] = NULL;
 291 }
 292
 293 static void vhost_vq_meta_reset(struct vhost_dev *d)
 294 {
 295         int i;
 296
 297         for (i = 0; i < d->nvqs; ++i)
 298                 __vhost_vq_meta_reset(d->vqs[i]);
 299 }
 300
 301 #if VHOST_ARCH_CAN_ACCEL_UACCESS
 302 static void vhost_map_unprefetch(struct vhost_map *map)
 303 {
 304         kfree(map->pages);
 305         map->pages = NULL;
 306         map->npages = 0;
 307         map->addr = NULL;
 308 }
 309
 310 static void vhost_uninit_vq_maps(struct vhost_virtqueue *vq)
 311 {
 312         struct vhost_map *map[VHOST_NUM_ADDRS];
 313         int i;
 314
 315         spin_lock(&vq->mmu_lock);
 316         for (i = 0; i < VHOST_NUM_ADDRS; i++) {
 317                 map[i] = rcu_dereference_protected(vq->maps[i],
 318                                   lockdep_is_held(&vq->mmu_lock));
 319                 if (map[i])
 320                         rcu_assign_pointer(vq->maps[i], NULL);
 321         }
 322         spin_unlock(&vq->mmu_lock);
 323
 324         synchronize_rcu();
 325
 326         for (i = 0; i < VHOST_NUM_ADDRS; i++)
 327                 if (map[i])
 328                         vhost_map_unprefetch(map[i]);
 329
 330 }
 331
 332 static void vhost_reset_vq_maps(struct vhost_virtqueue *vq)
 333 {
 334         int i;
 335
 336         vhost_uninit_vq_maps(vq);
 337         for (i = 0; i < VHOST_NUM_ADDRS; i++)
 338                 vq->uaddrs[i].size = 0;
 339 }
 340
 341 static bool vhost_map_range_overlap(struct vhost_uaddr *uaddr,
 342                                      unsigned long start,
 343                                      unsigned long end)
 344 {
 345         if (unlikely(!uaddr->size))
 346                 return false;
 347
 348         return !(end < uaddr->uaddr || start > uaddr->uaddr - 1 + uaddr->size);
 349 }
 350
 351 static void vhost_invalidate_vq_start(struct vhost_virtqueue *vq,
 352                                       int index,
 353                                       unsigned long start,
 354                                       unsigned long end)
 355 {
 356         struct vhost_uaddr *uaddr = &vq->uaddrs[index];
 357         struct vhost_map *map;
 358         int i;
 359
 360         if (!vhost_map_range_overlap(uaddr, start, end))
 361                 return;
 362
 363         spin_lock(&vq->mmu_lock);
 364         ++vq->invalidate_count;
 365
 366         map = rcu_dereference_protected(vq->maps[index],
 367                                         lockdep_is_held(&vq->mmu_lock));
 368         if (map) {
 369                 if (uaddr->write) {
 370                         for (i = 0; i < map->npages; i++)
 371                                 set_page_dirty(map->pages[i]);
 372                 }
 373                 rcu_assign_pointer(vq->maps[index], NULL);
 374         }
 375         spin_unlock(&vq->mmu_lock);
 376
 377         if (map) {
 378                 synchronize_rcu();
 379                 vhost_map_unprefetch(map);
 380         }
 381 }
 382
 383 static void vhost_invalidate_vq_end(struct vhost_virtqueue *vq,
 384                                     int index,
 385                                     unsigned long start,
 386                                     unsigned long end)
 387 {
 388         if (!vhost_map_range_overlap(&vq->uaddrs[index], start, end))
 389                 return;
 390
 391         spin_lock(&vq->mmu_lock);
 392         --vq->invalidate_count;
 393         spin_unlock(&vq->mmu_lock);
 394 }
 395
 396 static int vhost_invalidate_range_start(struct mmu_notifier *mn,
 397                                         const struct mmu_notifier_range *range)
 398 {
 399         struct vhost_dev *dev = container_of(mn, struct vhost_dev,
 400                                              mmu_notifier);
 401         int i, j;
 402
 403         if (!mmu_notifier_range_blockable(range))
 404                 return -EAGAIN;
 405
 406         for (i = 0; i < dev->nvqs; i++) {
 407                 struct vhost_virtqueue *vq = dev->vqs[i];
 408
 409                 for (j = 0; j < VHOST_NUM_ADDRS; j++)
 410                         vhost_invalidate_vq_start(vq, j,
 411                                                   range->start,
 412                                                   range->end);
 413         }
 414
 415         return 0;
 416 }
 417
 418 static void vhost_invalidate_range_end(struct mmu_notifier *mn,
 419                                        const struct mmu_notifier_range *range)
 420 {
 421         struct vhost_dev *dev = container_of(mn, struct vhost_dev,
 422                                              mmu_notifier);
 423         int i, j;
 424
 425         for (i = 0; i < dev->nvqs; i++) {
 426                 struct vhost_virtqueue *vq = dev->vqs[i];
 427
 428                 for (j = 0; j < VHOST_NUM_ADDRS; j++)
 429                         vhost_invalidate_vq_end(vq, j,
 430                                                 range->start,
 431                                                 range->end);
 432         }
 433 }
 434
 435 static const struct mmu_notifier_ops vhost_mmu_notifier_ops = {
 436         .invalidate_range_start = vhost_invalidate_range_start,
 437         .invalidate_range_end = vhost_invalidate_range_end,
 438 };
 439
 440 static void vhost_init_maps(struct vhost_dev *dev)
 441 {
 442         struct vhost_virtqueue *vq;
 443         int i, j;
 444
 445         dev->mmu_notifier.ops = &vhost_mmu_notifier_ops;
 446
 447         for (i = 0; i < dev->nvqs; ++i) {
 448                 vq = dev->vqs[i];
 449                 for (j = 0; j < VHOST_NUM_ADDRS; j++)
 450                         RCU_INIT_POINTER(vq->maps[j], NULL);
 451         }
 452 }
 453 #endif
 454
 455 static void vhost_vq_reset(struct vhost_dev *dev,
 456                            struct vhost_virtqueue *vq)
 457 {
 458         vq->num = 1;
 459         vq->desc = NULL;
 460         vq->avail = NULL;
 461         vq->used = NULL;
 462         vq->last_avail_idx = 0;
 463         vq->avail_idx = 0;
 464         vq->last_used_idx = 0;
 465         vq->signalled_used = 0;
 466         vq->signalled_used_valid = false;
 467         vq->used_flags = 0;
 468         vq->log_used = false;
 469         vq->log_addr = -1ull;
 470         vq->private_data = NULL;
 471         vq->acked_features = 0;
 472         vq->acked_backend_features = 0;
 473         vq->log_base = NULL;
 474         vq->error_ctx = NULL;
 475         vq->kick = NULL;
 476         vq->call_ctx = NULL;
 477         vq->log_ctx = NULL;
 478         vhost_reset_is_le(vq);
 479         vhost_disable_cross_endian(vq);
 480         vq->busyloop_timeout = 0;
 481         vq->umem = NULL;
 482         vq->iotlb = NULL;
 483         vq->invalidate_count = 0;
 484         __vhost_vq_meta_reset(vq);
 485 #if VHOST_ARCH_CAN_ACCEL_UACCESS
 486         vhost_reset_vq_maps(vq);
 487 #endif
 488 }
 489
 490 static int vhost_worker(void *data)
 491 {
 492         struct vhost_dev *dev = data;
 493         struct vhost_work *work, *work_next;
 494         struct llist_node *node;
 495         mm_segment_t oldfs = get_fs();
 496
 497         set_fs(USER_DS);
 498         use_mm(dev->mm);
 499
 500         for (;;) {
 501                 /* mb paired w/ kthread_stop */
 502                 set_current_state(TASK_INTERRUPTIBLE);
 503
 504                 if (kthread_should_stop()) {
 505                         __set_current_state(TASK_RUNNING);
 506                         break;
 507                 }
 508
 509                 node = llist_del_all(&dev->work_list);
 510                 if (!node)
 511                         schedule();
 512
 513                 node = llist_reverse_order(node);
 514                 /* make sure flag is seen after deletion */
 515                 smp_wmb();
 516                 llist_for_each_entry_safe(work, work_next, node, node) {
 517                         clear_bit(VHOST_WORK_QUEUED, &work->flags);
 518                         __set_current_state(TASK_RUNNING);
 519                         work->fn(work);
 520                         if (need_resched())
 521                                 schedule();
 522                 }
 523         }
 524         unuse_mm(dev->mm);
 525         set_fs(oldfs);
 526         return 0;
 527 }
 528
 529 static void vhost_vq_free_iovecs(struct vhost_virtqueue *vq)
 530 {
 531         kfree(vq->indirect);
 532         vq->indirect = NULL;
 533         kfree(vq->log);
 534         vq->log = NULL;
 535         kfree(vq->heads);
 536         vq->heads = NULL;
 537 }
 538
 539 /* Helper to allocate iovec buffers for all vqs. */
 540 static long vhost_dev_alloc_iovecs(struct vhost_dev *dev)
 541 {
 542         struct vhost_virtqueue *vq;
 543         int i;
 544
 545         for (i = 0; i < dev->nvqs; ++i) {
 546                 vq = dev->vqs[i];
 547                 vq->indirect = kmalloc_array(UIO_MAXIOV,
 548                                              sizeof(*vq->indirect),
 549                                              GFP_KERNEL);
 550                 vq->log = kmalloc_array(dev->iov_limit, sizeof(*vq->log),
 551                                         GFP_KERNEL);
 552                 vq->heads = kmalloc_array(dev->iov_limit, sizeof(*vq->heads),
 553                                           GFP_KERNEL);
 554                 if (!vq->indirect || !vq->log || !vq->heads)
 555                         goto err_nomem;
 556         }
 557         return 0;
 558
 559 err_nomem:
 560         for (; i >= 0; --i)
 561                 vhost_vq_free_iovecs(dev->vqs[i]);
 562         return -ENOMEM;
 563 }
 564
 565 static void vhost_dev_free_iovecs(struct vhost_dev *dev)
 566 {
 567         int i;
 568
 569         for (i = 0; i < dev->nvqs; ++i)
 570                 vhost_vq_free_iovecs(dev->vqs[i]);
 571 }
 572
 573 bool vhost_exceeds_weight(struct vhost_virtqueue *vq,
 574                           int pkts, int total_len)
 575 {
 576         struct vhost_dev *dev = vq->dev;
 577
 578         if ((dev->byte_weight && total_len >= dev->byte_weight) ||
 579             pkts >= dev->weight) {
 580                 vhost_poll_queue(&vq->poll);
 581                 return true;
 582         }
 583
 584         return false;
 585 }
 586 EXPORT_SYMBOL_GPL(vhost_exceeds_weight);
 587
 588 static size_t vhost_get_avail_size(struct vhost_virtqueue *vq,
 589                                    unsigned int num)
 590 {
 591         size_t event __maybe_unused =
 592                vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
 593
 594         return sizeof(*vq->avail) +
 595                sizeof(*vq->avail->ring) * num + event;
 596 }
 597
 598 static size_t vhost_get_used_size(struct vhost_virtqueue *vq,
 599                                   unsigned int num)
 600 {
 601         size_t event __maybe_unused =
 602                vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
 603
 604         return sizeof(*vq->used) +
 605                sizeof(*vq->used->ring) * num + event;
 606 }
 607
 608 static size_t vhost_get_desc_size(struct vhost_virtqueue *vq,
 609                                   unsigned int num)
 610 {
 611         return sizeof(*vq->desc) * num;
 612 }
 613
 614 void vhost_dev_init(struct vhost_dev *dev,
 615                     struct vhost_virtqueue **vqs, int nvqs,
 616                     int iov_limit, int weight, int byte_weight)
 617 {
 618         struct vhost_virtqueue *vq;
 619         int i;
 620
 621         dev->vqs = vqs;
 622         dev->nvqs = nvqs;
 623         mutex_init(&dev->mutex);
 624         dev->log_ctx = NULL;
 625         dev->umem = NULL;
 626         dev->iotlb = NULL;
 627         dev->mm = NULL;
 628         dev->worker = NULL;
 629         dev->iov_limit = iov_limit;
 630         dev->weight = weight;
 631         dev->byte_weight = byte_weight;
 632         init_llist_head(&dev->work_list);
 633         init_waitqueue_head(&dev->wait);
 634         INIT_LIST_HEAD(&dev->read_list);
 635         INIT_LIST_HEAD(&dev->pending_list);
 636         spin_lock_init(&dev->iotlb_lock);
 637 #if VHOST_ARCH_CAN_ACCEL_UACCESS
 638         vhost_init_maps(dev);
 639 #endif
 640
 641         for (i = 0; i < dev->nvqs; ++i) {
 642                 vq = dev->vqs[i];
 643                 vq->log = NULL;
 644                 vq->indirect = NULL;
 645                 vq->heads = NULL;
 646                 vq->dev = dev;
 647                 mutex_init(&vq->mutex);
 648                 spin_lock_init(&vq->mmu_lock);
 649                 vhost_vq_reset(dev, vq);
 650                 if (vq->handle_kick)
 651                         vhost_poll_init(&vq->poll, vq->handle_kick,
 652                                         EPOLLIN, dev);
 653         }
 654 }
 655 EXPORT_SYMBOL_GPL(vhost_dev_init);
 656
 657 /* Caller should have device mutex */
 658 long vhost_dev_check_owner(struct vhost_dev *dev)
 659 {
 660         /* Are you the owner? If not, I don't think you mean to do that */
 661         return dev->mm == current->mm ? 0 : -EPERM;
 662 }
 663 EXPORT_SYMBOL_GPL(vhost_dev_check_owner);
 664
 665 struct vhost_attach_cgroups_struct {
 666         struct vhost_work work;
 667         struct task_struct *owner;
 668         int ret;
 669 };
 670
 671 static void vhost_attach_cgroups_work(struct vhost_work *work)
 672 {
 673         struct vhost_attach_cgroups_struct *s;
 674
 675         s = container_of(work, struct vhost_attach_cgroups_struct, work);
 676         s->ret = cgroup_attach_task_all(s->owner, current);
 677 }
 678
 679 static int vhost_attach_cgroups(struct vhost_dev *dev)
 680 {
 681         struct vhost_attach_cgroups_struct attach;
 682
 683         attach.owner = current;
 684         vhost_work_init(&attach.work, vhost_attach_cgroups_work);
 685         vhost_work_queue(dev, &attach.work);
 686         vhost_work_flush(dev, &attach.work);
 687         return attach.ret;
 688 }
 689
 690 /* Caller should have device mutex */
 691 bool vhost_dev_has_owner(struct vhost_dev *dev)
 692 {
 693         return dev->mm;
 694 }
 695 EXPORT_SYMBOL_GPL(vhost_dev_has_owner);
 696
 697 /* Caller should have device mutex */
 698 long vhost_dev_set_owner(struct vhost_dev *dev)
 699 {
 700         struct task_struct *worker;
 701         int err;
 702
 703         /* Is there an owner already? */
 704         if (vhost_dev_has_owner(dev)) {
 705                 err = -EBUSY;
 706                 goto err_mm;
 707         }
 708
 709         /* No owner, become one */
 710         dev->mm = get_task_mm(current);
 711         worker = kthread_create(vhost_worker, dev, "vhost-%d", current->pid);
 712         if (IS_ERR(worker)) {
 713                 err = PTR_ERR(worker);
 714                 goto err_worker;
 715         }
 716
 717         dev->worker = worker;
 718         wake_up_process(worker);        /* avoid contributing to loadavg */
 719
 720         err = vhost_attach_cgroups(dev);
 721         if (err)
 722                 goto err_cgroup;
 723
 724         err = vhost_dev_alloc_iovecs(dev);
 725         if (err)
 726                 goto err_cgroup;
 727
 728 #if VHOST_ARCH_CAN_ACCEL_UACCESS
 729         err = mmu_notifier_register(&dev->mmu_notifier, dev->mm);
 730         if (err)
 731                 goto err_mmu_notifier;
 732 #endif
 733
 734         return 0;
 735
 736 #if VHOST_ARCH_CAN_ACCEL_UACCESS
 737 err_mmu_notifier:
 738         vhost_dev_free_iovecs(dev);
 739 #endif
 740 err_cgroup:
 741         kthread_stop(worker);
 742         dev->worker = NULL;
 743 err_worker:
 744         if (dev->mm)
 745                 mmput(dev->mm);
 746         dev->mm = NULL;
 747 err_mm:
 748         return err;
 749 }
 750 EXPORT_SYMBOL_GPL(vhost_dev_set_owner);
 751
 752 struct vhost_umem *vhost_dev_reset_owner_prepare(void)
 753 {
 754         return kvzalloc(sizeof(struct vhost_umem), GFP_KERNEL);
 755 }
 756 EXPORT_SYMBOL_GPL(vhost_dev_reset_owner_prepare);
 757
 758 /* Caller should have device mutex */
 759 void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_umem *umem)
 760 {
 761         int i;
 762
 763         vhost_dev_cleanup(dev);
 764
 765         /* Restore memory to default empty mapping. */
 766         INIT_LIST_HEAD(&umem->umem_list);
 767         dev->umem = umem;
 768         /* We don't need VQ locks below since vhost_dev_cleanup makes sure
 769          * VQs aren't running.
 770          */
 771         for (i = 0; i < dev->nvqs; ++i)
 772                 dev->vqs[i]->umem = umem;
 773 }
 774 EXPORT_SYMBOL_GPL(vhost_dev_reset_owner);
 775
 776 void vhost_dev_stop(struct vhost_dev *dev)
 777 {
 778         int i;
 779
 780         for (i = 0; i < dev->nvqs; ++i) {
 781                 if (dev->vqs[i]->kick && dev->vqs[i]->handle_kick) {
 782                         vhost_poll_stop(&dev->vqs[i]->poll);
 783                         vhost_poll_flush(&dev->vqs[i]->poll);
 784                 }
 785         }
 786 }
 787 EXPORT_SYMBOL_GPL(vhost_dev_stop);
 788
 789 static void vhost_umem_free(struct vhost_umem *umem,
 790                             struct vhost_umem_node *node)
 791 {
 792         vhost_umem_interval_tree_remove(node, &umem->umem_tree);
 793         list_del(&node->link);
 794         kfree(node);
 795         umem->numem--;
 796 }
 797
 798 static void vhost_umem_clean(struct vhost_umem *umem)
 799 {
 800         struct vhost_umem_node *node, *tmp;
 801
 802         if (!umem)
 803                 return;
 804
 805         list_for_each_entry_safe(node, tmp, &umem->umem_list, link)
 806                 vhost_umem_free(umem, node);
 807
 808         kvfree(umem);
 809 }
 810
 811 static void vhost_clear_msg(struct vhost_dev *dev)
 812 {
 813         struct vhost_msg_node *node, *n;
 814
 815         spin_lock(&dev->iotlb_lock);
 816
 817         list_for_each_entry_safe(node, n, &dev->read_list, node) {
 818                 list_del(&node->node);
 819                 kfree(node);
 820         }
 821
 822         list_for_each_entry_safe(node, n, &dev->pending_list, node) {
 823                 list_del(&node->node);
 824                 kfree(node);
 825         }
 826
 827         spin_unlock(&dev->iotlb_lock);
 828 }
 829
 830 #if VHOST_ARCH_CAN_ACCEL_UACCESS
 831 static void vhost_setup_uaddr(struct vhost_virtqueue *vq,
 832                               int index, unsigned long uaddr,
 833                               size_t size, bool write)
 834 {
 835         struct vhost_uaddr *addr = &vq->uaddrs[index];
 836
 837         addr->uaddr = uaddr;
 838         addr->size = size;
 839         addr->write = write;
 840 }
 841
 842 static void vhost_setup_vq_uaddr(struct vhost_virtqueue *vq)
 843 {
 844         vhost_setup_uaddr(vq, VHOST_ADDR_DESC,
 845                           (unsigned long)vq->desc,
 846                           vhost_get_desc_size(vq, vq->num),
 847                           false);
 848         vhost_setup_uaddr(vq, VHOST_ADDR_AVAIL,
 849                           (unsigned long)vq->avail,
 850                           vhost_get_avail_size(vq, vq->num),
 851                           false);
 852         vhost_setup_uaddr(vq, VHOST_ADDR_USED,
 853                           (unsigned long)vq->used,
 854                           vhost_get_used_size(vq, vq->num),
 855                           true);
 856 }
 857
 858 static int vhost_map_prefetch(struct vhost_virtqueue *vq,
 859                                int index)
 860 {
 861         struct vhost_map *map;
 862         struct vhost_uaddr *uaddr = &vq->uaddrs[index];
 863         struct page **pages;
 864         int npages = DIV_ROUND_UP(uaddr->size, PAGE_SIZE);
 865         int npinned;
 866         void *vaddr, *v;
 867         int err;
 868         int i;
 869
 870         spin_lock(&vq->mmu_lock);
 871
 872         err = -EFAULT;
 873         if (vq->invalidate_count)
 874                 goto err;
 875
 876         err = -ENOMEM;
 877         map = kmalloc(sizeof(*map), GFP_ATOMIC);
 878         if (!map)
 879                 goto err;
 880
 881         pages = kmalloc_array(npages, sizeof(struct page *), GFP_ATOMIC);
 882         if (!pages)
 883                 goto err_pages;
 884
 885         err = EFAULT;
 886         npinned = __get_user_pages_fast(uaddr->uaddr, npages,
 887                                         uaddr->write, pages);
 888         if (npinned > 0)
 889                 release_pages(pages, npinned);
 890         if (npinned != npages)
 891                 goto err_gup;
 892
 893         for (i = 0; i < npinned; i++)
 894                 if (PageHighMem(pages[i]))
 895                         goto err_gup;
 896
 897         vaddr = v = page_address(pages[0]);
 898
 899         /* For simplicity, fallback to userspace address if VA is not
 900          * contigious.
 901          */
 902         for (i = 1; i < npinned; i++) {
 903                 v += PAGE_SIZE;
 904                 if (v != page_address(pages[i]))
 905                         goto err_gup;
 906         }
 907
 908         map->addr = vaddr + (uaddr->uaddr & (PAGE_SIZE - 1));
 909         map->npages = npages;
 910         map->pages = pages;
 911
 912         rcu_assign_pointer(vq->maps[index], map);
 913         /* No need for a synchronize_rcu(). This function should be
 914          * called by dev->worker so we are serialized with all
 915          * readers.
 916          */
 917         spin_unlock(&vq->mmu_lock);
 918
 919         return 0;
 920
 921 err_gup:
 922         kfree(pages);
 923 err_pages:
 924         kfree(map);
 925 err:
 926         spin_unlock(&vq->mmu_lock);
 927         return err;
 928 }
 929 #endif
 930
 931 void vhost_dev_cleanup(struct vhost_dev *dev)
 932 {
 933         int i;
 934
 935         for (i = 0; i < dev->nvqs; ++i) {
 936                 if (dev->vqs[i]->error_ctx)
 937                         eventfd_ctx_put(dev->vqs[i]->error_ctx);
 938                 if (dev->vqs[i]->kick)
 939                         fput(dev->vqs[i]->kick);
 940                 if (dev->vqs[i]->call_ctx)
 941                         eventfd_ctx_put(dev->vqs[i]->call_ctx);
 942                 vhost_vq_reset(dev, dev->vqs[i]);
 943         }
 944         vhost_dev_free_iovecs(dev);
 945         if (dev->log_ctx)
 946                 eventfd_ctx_put(dev->log_ctx);
 947         dev->log_ctx = NULL;
 948         /* No one will access memory at this point */
 949         vhost_umem_clean(dev->umem);
 950         dev->umem = NULL;
 951         vhost_umem_clean(dev->iotlb);
 952         dev->iotlb = NULL;
 953         vhost_clear_msg(dev);
 954         wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM);
 955         WARN_ON(!llist_empty(&dev->work_list));
 956         if (dev->worker) {
 957                 kthread_stop(dev->worker);
 958                 dev->worker = NULL;
 959         }
 960         if (dev->mm) {
 961 #if VHOST_ARCH_CAN_ACCEL_UACCESS
 962                 mmu_notifier_unregister(&dev->mmu_notifier, dev->mm);
 963 #endif
 964                 mmput(dev->mm);
 965         }
 966 #if VHOST_ARCH_CAN_ACCEL_UACCESS
 967         for (i = 0; i < dev->nvqs; i++)
 968                 vhost_uninit_vq_maps(dev->vqs[i]);
 969 #endif
 970         dev->mm = NULL;
 971 }
 972 EXPORT_SYMBOL_GPL(vhost_dev_cleanup);
 973
 974 static bool log_access_ok(void __user *log_base, u64 addr, unsigned long sz)
 975 {
 976         u64 a = addr / VHOST_PAGE_SIZE / 8;
 977
 978         /* Make sure 64 bit math will not overflow. */
 979         if (a > ULONG_MAX - (unsigned long)log_base ||
 980             a + (unsigned long)log_base > ULONG_MAX)
 981                 return false;
 982
 983         return access_ok(log_base + a,
 984                          (sz + VHOST_PAGE_SIZE * 8 - 1) / VHOST_PAGE_SIZE / 8);
 985 }
 986
 987 static bool vhost_overflow(u64 uaddr, u64 size)
 988 {
 989         /* Make sure 64 bit math will not overflow. */
 990         return uaddr > ULONG_MAX || size > ULONG_MAX || uaddr > ULONG_MAX - size;
 991 }
 992
 993 /* Caller should have vq mutex and device mutex. */
 994 static bool vq_memory_access_ok(void __user *log_base, struct vhost_umem *umem,
 995                                 int log_all)
 996 {
 997         struct vhost_umem_node *node;
 998
 999         if (!umem)
1000                 return false;
1001
1002         list_for_each_entry(node, &umem->umem_list, link) {
1003                 unsigned long a = node->userspace_addr;
1004
1005                 if (vhost_overflow(node->userspace_addr, node->size))
1006                         return false;
1007
1008
1009                 if (!access_ok((void __user *)a,
1010                                     node->size))
1011                         return false;
1012                 else if (log_all && !log_access_ok(log_base,
1013                                                    node->start,
1014                                                    node->size))
1015                         return false;
1016         }
1017         return true;
1018 }
1019
1020 static inline void __user *vhost_vq_meta_fetch(struct vhost_virtqueue *vq,
1021                                                u64 addr, unsigned int size,
1022                                                int type)
1023 {
1024         const struct vhost_umem_node *node = vq->meta_iotlb[type];
1025
1026         if (!node)
1027                 return NULL;
1028
1029         return (void *)(uintptr_t)(node->userspace_addr + addr - node->start);
1030 }
1031
1032 /* Can we switch to this memory table? */
1033 /* Caller should have device mutex but not vq mutex */
1034 static bool memory_access_ok(struct vhost_dev *d, struct vhost_umem *umem,
1035                              int log_all)
1036 {
1037         int i;
1038
1039         for (i = 0; i < d->nvqs; ++i) {
1040                 bool ok;
1041                 bool log;
1042
1043                 mutex_lock(&d->vqs[i]->mutex);
1044                 log = log_all || vhost_has_feature(d->vqs[i], VHOST_F_LOG_ALL);
1045                 /* If ring is inactive, will check when it's enabled. */
1046                 if (d->vqs[i]->private_data)
1047                         ok = vq_memory_access_ok(d->vqs[i]->log_base,
1048                                                  umem, log);
1049                 else
1050                         ok = true;
1051                 mutex_unlock(&d->vqs[i]->mutex);
1052                 if (!ok)
1053                         return false;
1054         }
1055         return true;
1056 }
1057
1058 static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len,
1059                           struct iovec iov[], int iov_size, int access);
1060
1061 static int vhost_copy_to_user(struct vhost_virtqueue *vq, void __user *to,
1062                               const void *from, unsigned size)
1063 {
1064         int ret;
1065
1066         if (!vq->iotlb)
1067                 return __copy_to_user(to, from, size);
1068         else {
1069                 /* This function should be called after iotlb
1070                  * prefetch, which means we're sure that all vq
1071                  * could be access through iotlb. So -EAGAIN should
1072                  * not happen in this case.
1073                  */
1074                 struct iov_iter t;
1075                 void __user *uaddr = vhost_vq_meta_fetch(vq,
1076                                      (u64)(uintptr_t)to, size,
1077                                      VHOST_ADDR_USED);
1078
1079                 if (uaddr)
1080                         return __copy_to_user(uaddr, from, size);
1081
1082                 ret = translate_desc(vq, (u64)(uintptr_t)to, size, vq->iotlb_iov,
1083                                      ARRAY_SIZE(vq->iotlb_iov),
1084                                      VHOST_ACCESS_WO);
1085                 if (ret < 0)
1086                         goto out;
1087                 iov_iter_init(&t, WRITE, vq->iotlb_iov, ret, size);
1088                 ret = copy_to_iter(from, size, &t);
1089                 if (ret == size)
1090                         ret = 0;
1091         }
1092 out:
1093         return ret;
1094 }
1095
1096 static int vhost_copy_from_user(struct vhost_virtqueue *vq, void *to,
1097                                 void __user *from, unsigned size)
1098 {
1099         int ret;
1100
1101         if (!vq->iotlb)
1102                 return __copy_from_user(to, from, size);
1103         else {
1104                 /* This function should be called after iotlb
1105                  * prefetch, which means we're sure that vq
1106                  * could be access through iotlb. So -EAGAIN should
1107                  * not happen in this case.
1108                  */
1109                 void __user *uaddr = vhost_vq_meta_fetch(vq,
1110                                      (u64)(uintptr_t)from, size,
1111                                      VHOST_ADDR_DESC);
1112                 struct iov_iter f;
1113
1114                 if (uaddr)
1115                         return __copy_from_user(to, uaddr, size);
1116
1117                 ret = translate_desc(vq, (u64)(uintptr_t)from, size, vq->iotlb_iov,
1118                                      ARRAY_SIZE(vq->iotlb_iov),
1119                                      VHOST_ACCESS_RO);
1120                 if (ret < 0) {
1121                         vq_err(vq, "IOTLB translation failure: uaddr "
1122                                "%p size 0x%llx\n", from,
1123                                (unsigned long long) size);
1124                         goto out;
1125                 }
1126                 iov_iter_init(&f, READ, vq->iotlb_iov, ret, size);
1127                 ret = copy_from_iter(to, size, &f);
1128                 if (ret == size)
1129                         ret = 0;
1130         }
1131
1132 out:
1133         return ret;
1134 }
1135
1136 static void __user *__vhost_get_user_slow(struct vhost_virtqueue *vq,
1137                                           void __user *addr, unsigned int size,
1138                                           int type)
1139 {
1140         int ret;
1141
1142         ret = translate_desc(vq, (u64)(uintptr_t)addr, size, vq->iotlb_iov,
1143                              ARRAY_SIZE(vq->iotlb_iov),
1144                              VHOST_ACCESS_RO);
1145         if (ret < 0) {
1146                 vq_err(vq, "IOTLB translation failure: uaddr "
1147                         "%p size 0x%llx\n", addr,
1148                         (unsigned long long) size);
1149                 return NULL;
1150         }
1151
1152         if (ret != 1 || vq->iotlb_iov[0].iov_len != size) {
1153                 vq_err(vq, "Non atomic userspace memory access: uaddr "
1154                         "%p size 0x%llx\n", addr,
1155                         (unsigned long long) size);
1156                 return NULL;
1157         }
1158
1159         return vq->iotlb_iov[0].iov_base;
1160 }
1161
1162 /* This function should be called after iotlb
1163  * prefetch, which means we're sure that vq
1164  * could be access through iotlb. So -EAGAIN should
1165  * not happen in this case.
1166  */
1167 static inline void __user *__vhost_get_user(struct vhost_virtqueue *vq,
1168                                             void *addr, unsigned int size,
1169                                             int type)
1170 {
1171         void __user *uaddr = vhost_vq_meta_fetch(vq,
1172                              (u64)(uintptr_t)addr, size, type);
1173         if (uaddr)
1174                 return uaddr;
1175
1176         return __vhost_get_user_slow(vq, addr, size, type);
1177 }
1178
1179 #define vhost_put_user(vq, x, ptr)              \
1180 ({ \
1181         int ret = -EFAULT; \
1182         if (!vq->iotlb) { \
1183                 ret = __put_user(x, ptr); \
1184         } else { \
1185                 __typeof__(ptr) to = \
1186                         (__typeof__(ptr)) __vhost_get_user(vq, ptr,     \
1187                                           sizeof(*ptr), VHOST_ADDR_USED); \
1188                 if (to != NULL) \
1189                         ret = __put_user(x, to); \
1190                 else \
1191                         ret = -EFAULT;  \
1192         } \
1193         ret; \
1194 })
1195
1196 static inline int vhost_put_avail_event(struct vhost_virtqueue *vq)
1197 {
1198 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1199         struct vhost_map *map;
1200         struct vring_used *used;
1201
1202         if (!vq->iotlb) {
1203                 rcu_read_lock();
1204
1205                 map = rcu_dereference(vq->maps[VHOST_ADDR_USED]);
1206                 if (likely(map)) {
1207                         used = map->addr;
1208                         *((__virtio16 *)&used->ring[vq->num]) =
1209                                 cpu_to_vhost16(vq, vq->avail_idx);
1210                         rcu_read_unlock();
1211                         return 0;
1212                 }
1213
1214                 rcu_read_unlock();
1215         }
1216 #endif
1217
1218         return vhost_put_user(vq, cpu_to_vhost16(vq, vq->avail_idx),
1219                               vhost_avail_event(vq));
1220 }
1221
1222 static inline int vhost_put_used(struct vhost_virtqueue *vq,
1223                                  struct vring_used_elem *head, int idx,
1224                                  int count)
1225 {
1226 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1227         struct vhost_map *map;
1228         struct vring_used *used;
1229         size_t size;
1230
1231         if (!vq->iotlb) {
1232                 rcu_read_lock();
1233
1234                 map = rcu_dereference(vq->maps[VHOST_ADDR_USED]);
1235                 if (likely(map)) {
1236                         used = map->addr;
1237                         size = count * sizeof(*head);
1238                         memcpy(used->ring + idx, head, size);
1239                         rcu_read_unlock();
1240                         return 0;
1241                 }
1242
1243                 rcu_read_unlock();
1244         }
1245 #endif
1246
1247         return vhost_copy_to_user(vq, vq->used->ring + idx, head,
1248                                   count * sizeof(*head));
1249 }
1250
1251 static inline int vhost_put_used_flags(struct vhost_virtqueue *vq)
1252
1253 {
1254 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1255         struct vhost_map *map;
1256         struct vring_used *used;
1257
1258         if (!vq->iotlb) {
1259                 rcu_read_lock();
1260
1261                 map = rcu_dereference(vq->maps[VHOST_ADDR_USED]);
1262                 if (likely(map)) {
1263                         used = map->addr;
1264                         used->flags = cpu_to_vhost16(vq, vq->used_flags);
1265                         rcu_read_unlock();
1266                         return 0;
1267                 }
1268
1269                 rcu_read_unlock();
1270         }
1271 #endif
1272
1273         return vhost_put_user(vq, cpu_to_vhost16(vq, vq->used_flags),
1274                               &vq->used->flags);
1275 }
1276
1277 static inline int vhost_put_used_idx(struct vhost_virtqueue *vq)
1278
1279 {
1280 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1281         struct vhost_map *map;
1282         struct vring_used *used;
1283
1284         if (!vq->iotlb) {
1285                 rcu_read_lock();
1286
1287                 map = rcu_dereference(vq->maps[VHOST_ADDR_USED]);
1288                 if (likely(map)) {
1289                         used = map->addr;
1290                         used->idx = cpu_to_vhost16(vq, vq->last_used_idx);
1291                         rcu_read_unlock();
1292                         return 0;
1293                 }
1294
1295                 rcu_read_unlock();
1296         }
1297 #endif
1298
1299         return vhost_put_user(vq, cpu_to_vhost16(vq, vq->last_used_idx),
1300                               &vq->used->idx);
1301 }
1302
1303 #define vhost_get_user(vq, x, ptr, type)                \
1304 ({ \
1305         int ret; \
1306         if (!vq->iotlb) { \
1307                 ret = __get_user(x, ptr); \
1308         } else { \
1309                 __typeof__(ptr) from = \
1310                         (__typeof__(ptr)) __vhost_get_user(vq, ptr, \
1311                                                            sizeof(*ptr), \
1312                                                            type); \
1313                 if (from != NULL) \
1314                         ret = __get_user(x, from); \
1315                 else \
1316                         ret = -EFAULT; \
1317         } \
1318         ret; \
1319 })
1320
1321 #define vhost_get_avail(vq, x, ptr) \
1322         vhost_get_user(vq, x, ptr, VHOST_ADDR_AVAIL)
1323
1324 #define vhost_get_used(vq, x, ptr) \
1325         vhost_get_user(vq, x, ptr, VHOST_ADDR_USED)
1326
1327 static void vhost_dev_lock_vqs(struct vhost_dev *d)
1328 {
1329         int i = 0;
1330         for (i = 0; i < d->nvqs; ++i)
1331                 mutex_lock_nested(&d->vqs[i]->mutex, i);
1332 }
1333
1334 static void vhost_dev_unlock_vqs(struct vhost_dev *d)
1335 {
1336         int i = 0;
1337         for (i = 0; i < d->nvqs; ++i)
1338                 mutex_unlock(&d->vqs[i]->mutex);
1339 }
1340
1341 static inline int vhost_get_avail_idx(struct vhost_virtqueue *vq,
1342                                       __virtio16 *idx)
1343 {
1344 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1345         struct vhost_map *map;
1346         struct vring_avail *avail;
1347
1348         if (!vq->iotlb) {
1349                 rcu_read_lock();
1350
1351                 map = rcu_dereference(vq->maps[VHOST_ADDR_AVAIL]);
1352                 if (likely(map)) {
1353                         avail = map->addr;
1354                         *idx = avail->idx;
1355                         rcu_read_unlock();
1356                         return 0;
1357                 }
1358
1359                 rcu_read_unlock();
1360         }
1361 #endif
1362
1363         return vhost_get_avail(vq, *idx, &vq->avail->idx);
1364 }
1365
1366 static inline int vhost_get_avail_head(struct vhost_virtqueue *vq,
1367                                        __virtio16 *head, int idx)
1368 {
1369 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1370         struct vhost_map *map;
1371         struct vring_avail *avail;
1372
1373         if (!vq->iotlb) {
1374                 rcu_read_lock();
1375
1376                 map = rcu_dereference(vq->maps[VHOST_ADDR_AVAIL]);
1377                 if (likely(map)) {
1378                         avail = map->addr;
1379                         *head = avail->ring[idx & (vq->num - 1)];
1380                         rcu_read_unlock();
1381                         return 0;
1382                 }
1383
1384                 rcu_read_unlock();
1385         }
1386 #endif
1387
1388         return vhost_get_avail(vq, *head,
1389                                &vq->avail->ring[idx & (vq->num - 1)]);
1390 }
1391
1392 static inline int vhost_get_avail_flags(struct vhost_virtqueue *vq,
1393                                         __virtio16 *flags)
1394 {
1395 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1396         struct vhost_map *map;
1397         struct vring_avail *avail;
1398
1399         if (!vq->iotlb) {
1400                 rcu_read_lock();
1401
1402                 map = rcu_dereference(vq->maps[VHOST_ADDR_AVAIL]);
1403                 if (likely(map)) {
1404                         avail = map->addr;
1405                         *flags = avail->flags;
1406                         rcu_read_unlock();
1407                         return 0;
1408                 }
1409
1410                 rcu_read_unlock();
1411         }
1412 #endif
1413
1414         return vhost_get_avail(vq, *flags, &vq->avail->flags);
1415 }
1416
1417 static inline int vhost_get_used_event(struct vhost_virtqueue *vq,
1418                                        __virtio16 *event)
1419 {
1420 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1421         struct vhost_map *map;
1422         struct vring_avail *avail;
1423
1424         if (!vq->iotlb) {
1425                 rcu_read_lock();
1426                 map = rcu_dereference(vq->maps[VHOST_ADDR_AVAIL]);
1427                 if (likely(map)) {
1428                         avail = map->addr;
1429                         *event = (__virtio16)avail->ring[vq->num];
1430                         rcu_read_unlock();
1431                         return 0;
1432                 }
1433                 rcu_read_unlock();
1434         }
1435 #endif
1436
1437         return vhost_get_avail(vq, *event, vhost_used_event(vq));
1438 }
1439
1440 static inline int vhost_get_used_idx(struct vhost_virtqueue *vq,
1441                                      __virtio16 *idx)
1442 {
1443 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1444         struct vhost_map *map;
1445         struct vring_used *used;
1446
1447         if (!vq->iotlb) {
1448                 rcu_read_lock();
1449
1450                 map = rcu_dereference(vq->maps[VHOST_ADDR_USED]);
1451                 if (likely(map)) {
1452                         used = map->addr;
1453                         *idx = used->idx;
1454                         rcu_read_unlock();
1455                         return 0;
1456                 }
1457
1458                 rcu_read_unlock();
1459         }
1460 #endif
1461
1462         return vhost_get_used(vq, *idx, &vq->used->idx);
1463 }
1464
1465 static inline int vhost_get_desc(struct vhost_virtqueue *vq,
1466                                  struct vring_desc *desc, int idx)
1467 {
1468 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1469         struct vhost_map *map;
1470         struct vring_desc *d;
1471
1472         if (!vq->iotlb) {
1473                 rcu_read_lock();
1474
1475                 map = rcu_dereference(vq->maps[VHOST_ADDR_DESC]);
1476                 if (likely(map)) {
1477                         d = map->addr;
1478                         *desc = *(d + idx);
1479                         rcu_read_unlock();
1480                         return 0;
1481                 }
1482
1483                 rcu_read_unlock();
1484         }
1485 #endif
1486
1487         return vhost_copy_from_user(vq, desc, vq->desc + idx, sizeof(*desc));
1488 }
1489
1490 static int vhost_new_umem_range(struct vhost_umem *umem,
1491                                 u64 start, u64 size, u64 end,
1492                                 u64 userspace_addr, int perm)
1493 {
1494         struct vhost_umem_node *tmp, *node;
1495
1496         if (!size)
1497                 return -EFAULT;
1498
1499         node = kmalloc(sizeof(*node), GFP_ATOMIC);
1500         if (!node)
1501                 return -ENOMEM;
1502
1503         if (umem->numem == max_iotlb_entries) {
1504                 tmp = list_first_entry(&umem->umem_list, typeof(*tmp), link);
1505                 vhost_umem_free(umem, tmp);
1506         }
1507
1508         node->start = start;
1509         node->size = size;
1510         node->last = end;
1511         node->userspace_addr = userspace_addr;
1512         node->perm = perm;
1513         INIT_LIST_HEAD(&node->link);
1514         list_add_tail(&node->link, &umem->umem_list);
1515         vhost_umem_interval_tree_insert(node, &umem->umem_tree);
1516         umem->numem++;
1517
1518         return 0;
1519 }
1520
1521 static void vhost_del_umem_range(struct vhost_umem *umem,
1522                                  u64 start, u64 end)
1523 {
1524         struct vhost_umem_node *node;
1525
1526         while ((node = vhost_umem_interval_tree_iter_first(&umem->umem_tree,
1527                                                            start, end)))
1528                 vhost_umem_free(umem, node);
1529 }
1530
1531 static void vhost_iotlb_notify_vq(struct vhost_dev *d,
1532                                   struct vhost_iotlb_msg *msg)
1533 {
1534         struct vhost_msg_node *node, *n;
1535
1536         spin_lock(&d->iotlb_lock);
1537
1538         list_for_each_entry_safe(node, n, &d->pending_list, node) {
1539                 struct vhost_iotlb_msg *vq_msg = &node->msg.iotlb;
1540                 if (msg->iova <= vq_msg->iova &&
1541                     msg->iova + msg->size - 1 >= vq_msg->iova &&
1542                     vq_msg->type == VHOST_IOTLB_MISS) {
1543                         vhost_poll_queue(&node->vq->poll);
1544                         list_del(&node->node);
1545                         kfree(node);
1546                 }
1547         }
1548
1549         spin_unlock(&d->iotlb_lock);
1550 }
1551
1552 static bool umem_access_ok(u64 uaddr, u64 size, int access)
1553 {
1554         unsigned long a = uaddr;
1555
1556         /* Make sure 64 bit math will not overflow. */
1557         if (vhost_overflow(uaddr, size))
1558                 return false;
1559
1560         if ((access & VHOST_ACCESS_RO) &&
1561             !access_ok((void __user *)a, size))
1562                 return false;
1563         if ((access & VHOST_ACCESS_WO) &&
1564             !access_ok((void __user *)a, size))
1565                 return false;
1566         return true;
1567 }
1568
1569 static int vhost_process_iotlb_msg(struct vhost_dev *dev,
1570                                    struct vhost_iotlb_msg *msg)
1571 {
1572         int ret = 0;
1573
1574         mutex_lock(&dev->mutex);
1575         vhost_dev_lock_vqs(dev);
1576         switch (msg->type) {
1577         case VHOST_IOTLB_UPDATE:
1578                 if (!dev->iotlb) {
1579                         ret = -EFAULT;
1580                         break;
1581                 }
1582                 if (!umem_access_ok(msg->uaddr, msg->size, msg->perm)) {
1583                         ret = -EFAULT;
1584                         break;
1585                 }
1586                 vhost_vq_meta_reset(dev);
1587                 if (vhost_new_umem_range(dev->iotlb, msg->iova, msg->size,
1588                                          msg->iova + msg->size - 1,
1589                                          msg->uaddr, msg->perm)) {
1590                         ret = -ENOMEM;
1591                         break;
1592                 }
1593                 vhost_iotlb_notify_vq(dev, msg);
1594                 break;
1595         case VHOST_IOTLB_INVALIDATE:
1596                 if (!dev->iotlb) {
1597                         ret = -EFAULT;
1598                         break;
1599                 }
1600                 vhost_vq_meta_reset(dev);
1601                 vhost_del_umem_range(dev->iotlb, msg->iova,
1602                                      msg->iova + msg->size - 1);
1603                 break;
1604         default:
1605                 ret = -EINVAL;
1606                 break;
1607         }
1608
1609         vhost_dev_unlock_vqs(dev);
1610         mutex_unlock(&dev->mutex);
1611
1612         return ret;
1613 }
1614 ssize_t vhost_chr_write_iter(struct vhost_dev *dev,
1615                              struct iov_iter *from)
1616 {
1617         struct vhost_iotlb_msg msg;
1618         size_t offset;
1619         int type, ret;
1620
1621         ret = copy_from_iter(&type, sizeof(type), from);
1622         if (ret != sizeof(type)) {
1623                 ret = -EINVAL;
1624                 goto done;
1625         }
1626
1627         switch (type) {
1628         case VHOST_IOTLB_MSG:
1629                 /* There maybe a hole after type for V1 message type,
1630                  * so skip it here.
1631                  */
1632                 offset = offsetof(struct vhost_msg, iotlb) - sizeof(int);
1633                 break;
1634         case VHOST_IOTLB_MSG_V2:
1635                 offset = sizeof(__u32);
1636                 break;
1637         default:
1638                 ret = -EINVAL;
1639                 goto done;
1640         }
1641
1642         iov_iter_advance(from, offset);
1643         ret = copy_from_iter(&msg, sizeof(msg), from);
1644         if (ret != sizeof(msg)) {
1645                 ret = -EINVAL;
1646                 goto done;
1647         }
1648         if (vhost_process_iotlb_msg(dev, &msg)) {
1649                 ret = -EFAULT;
1650                 goto done;
1651         }
1652
1653         ret = (type == VHOST_IOTLB_MSG) ? sizeof(struct vhost_msg) :
1654               sizeof(struct vhost_msg_v2);
1655 done:
1656         return ret;
1657 }
1658 EXPORT_SYMBOL(vhost_chr_write_iter);
1659
1660 __poll_t vhost_chr_poll(struct file *file, struct vhost_dev *dev,
1661                             poll_table *wait)
1662 {
1663         __poll_t mask = 0;
1664
1665         poll_wait(file, &dev->wait, wait);
1666
1667         if (!list_empty(&dev->read_list))
1668                 mask |= EPOLLIN | EPOLLRDNORM;
1669
1670         return mask;
1671 }
1672 EXPORT_SYMBOL(vhost_chr_poll);
1673
1674 ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to,
1675                             int noblock)
1676 {
1677         DEFINE_WAIT(wait);
1678         struct vhost_msg_node *node;
1679         ssize_t ret = 0;
1680         unsigned size = sizeof(struct vhost_msg);
1681
1682         if (iov_iter_count(to) < size)
1683                 return 0;
1684
1685         while (1) {
1686                 if (!noblock)
1687                         prepare_to_wait(&dev->wait, &wait,
1688                                         TASK_INTERRUPTIBLE);
1689
1690                 node = vhost_dequeue_msg(dev, &dev->read_list);
1691                 if (node)
1692                         break;
1693                 if (noblock) {
1694                         ret = -EAGAIN;
1695                         break;
1696                 }
1697                 if (signal_pending(current)) {
1698                         ret = -ERESTARTSYS;
1699                         break;
1700                 }
1701                 if (!dev->iotlb) {
1702                         ret = -EBADFD;
1703                         break;
1704                 }
1705
1706                 schedule();
1707         }
1708
1709         if (!noblock)
1710                 finish_wait(&dev->wait, &wait);
1711
1712         if (node) {
1713                 struct vhost_iotlb_msg *msg;
1714                 void *start = &node->msg;
1715
1716                 switch (node->msg.type) {
1717                 case VHOST_IOTLB_MSG:
1718                         size = sizeof(node->msg);
1719                         msg = &node->msg.iotlb;
1720                         break;
1721                 case VHOST_IOTLB_MSG_V2:
1722                         size = sizeof(node->msg_v2);
1723                         msg = &node->msg_v2.iotlb;
1724                         break;
1725                 default:
1726                         BUG();
1727                         break;
1728                 }
1729
1730                 ret = copy_to_iter(start, size, to);
1731                 if (ret != size || msg->type != VHOST_IOTLB_MISS) {
1732                         kfree(node);
1733                         return ret;
1734                 }
1735                 vhost_enqueue_msg(dev, &dev->pending_list, node);
1736         }
1737
1738         return ret;
1739 }
1740 EXPORT_SYMBOL_GPL(vhost_chr_read_iter);
1741
1742 static int vhost_iotlb_miss(struct vhost_virtqueue *vq, u64 iova, int access)
1743 {
1744         struct vhost_dev *dev = vq->dev;
1745         struct vhost_msg_node *node;
1746         struct vhost_iotlb_msg *msg;
1747         bool v2 = vhost_backend_has_feature(vq, VHOST_BACKEND_F_IOTLB_MSG_V2);
1748
1749         node = vhost_new_msg(vq, v2 ? VHOST_IOTLB_MSG_V2 : VHOST_IOTLB_MSG);
1750         if (!node)
1751                 return -ENOMEM;
1752
1753         if (v2) {
1754                 node->msg_v2.type = VHOST_IOTLB_MSG_V2;
1755                 msg = &node->msg_v2.iotlb;
1756         } else {
1757                 msg = &node->msg.iotlb;
1758         }
1759
1760         msg->type = VHOST_IOTLB_MISS;
1761         msg->iova = iova;
1762         msg->perm = access;
1763
1764         vhost_enqueue_msg(dev, &dev->read_list, node);
1765
1766         return 0;
1767 }
1768
1769 static bool vq_access_ok(struct vhost_virtqueue *vq, unsigned int num,
1770                          struct vring_desc __user *desc,
1771                          struct vring_avail __user *avail,
1772                          struct vring_used __user *used)
1773
1774 {
1775         return access_ok(desc, vhost_get_desc_size(vq, num)) &&
1776                access_ok(avail, vhost_get_avail_size(vq, num)) &&
1777                access_ok(used, vhost_get_used_size(vq, num));
1778 }
1779
1780 static void vhost_vq_meta_update(struct vhost_virtqueue *vq,
1781                                  const struct vhost_umem_node *node,
1782                                  int type)
1783 {
1784         int access = (type == VHOST_ADDR_USED) ?
1785                      VHOST_ACCESS_WO : VHOST_ACCESS_RO;
1786
1787         if (likely(node->perm & access))
1788                 vq->meta_iotlb[type] = node;
1789 }
1790
1791 static bool iotlb_access_ok(struct vhost_virtqueue *vq,
1792                             int access, u64 addr, u64 len, int type)
1793 {
1794         const struct vhost_umem_node *node;
1795         struct vhost_umem *umem = vq->iotlb;
1796         u64 s = 0, size, orig_addr = addr, last = addr + len - 1;
1797
1798         if (vhost_vq_meta_fetch(vq, addr, len, type))
1799                 return true;
1800
1801         while (len > s) {
1802                 node = vhost_umem_interval_tree_iter_first(&umem->umem_tree,
1803                                                            addr,
1804                                                            last);
1805                 if (node == NULL || node->start > addr) {
1806                         vhost_iotlb_miss(vq, addr, access);
1807                         return false;
1808                 } else if (!(node->perm & access)) {
1809                         /* Report the possible access violation by
1810                          * request another translation from userspace.
1811                          */
1812                         return false;
1813                 }
1814
1815                 size = node->size - addr + node->start;
1816
1817                 if (orig_addr == addr && size >= len)
1818                         vhost_vq_meta_update(vq, node, type);
1819
1820                 s += size;
1821                 addr += size;
1822         }
1823
1824         return true;
1825 }
1826
1827 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1828 static void vhost_vq_map_prefetch(struct vhost_virtqueue *vq)
1829 {
1830         struct vhost_map __rcu *map;
1831         int i;
1832
1833         for (i = 0; i < VHOST_NUM_ADDRS; i++) {
1834                 rcu_read_lock();
1835                 map = rcu_dereference(vq->maps[i]);
1836                 rcu_read_unlock();
1837                 if (unlikely(!map))
1838                         vhost_map_prefetch(vq, i);
1839         }
1840 }
1841 #endif
1842
1843 int vq_meta_prefetch(struct vhost_virtqueue *vq)
1844 {
1845         unsigned int num = vq->num;
1846
1847         if (!vq->iotlb) {
1848 #if VHOST_ARCH_CAN_ACCEL_UACCESS
1849                 vhost_vq_map_prefetch(vq);
1850 #endif
1851                 return 1;
1852         }
1853
1854         return iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->desc,
1855                                vhost_get_desc_size(vq, num), VHOST_ADDR_DESC) &&
1856                iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->avail,
1857                                vhost_get_avail_size(vq, num),
1858                                VHOST_ADDR_AVAIL) &&
1859                iotlb_access_ok(vq, VHOST_ACCESS_WO, (u64)(uintptr_t)vq->used,
1860                                vhost_get_used_size(vq, num), VHOST_ADDR_USED);
1861 }
1862 EXPORT_SYMBOL_GPL(vq_meta_prefetch);
1863
1864 /* Can we log writes? */
1865 /* Caller should have device mutex but not vq mutex */
1866 bool vhost_log_access_ok(struct vhost_dev *dev)
1867 {
1868         return memory_access_ok(dev, dev->umem, 1);
1869 }
1870 EXPORT_SYMBOL_GPL(vhost_log_access_ok);
1871
1872 /* Verify access for write logging. */
1873 /* Caller should have vq mutex and device mutex */
1874 static bool vq_log_access_ok(struct vhost_virtqueue *vq,
1875                              void __user *log_base)
1876 {
1877         return vq_memory_access_ok(log_base, vq->umem,
1878                                    vhost_has_feature(vq, VHOST_F_LOG_ALL)) &&
1879                 (!vq->log_used || log_access_ok(log_base, vq->log_addr,
1880                                   vhost_get_used_size(vq, vq->num)));
1881 }
1882
1883 /* Can we start vq? */
1884 /* Caller should have vq mutex and device mutex */
1885 bool vhost_vq_access_ok(struct vhost_virtqueue *vq)
1886 {
1887         if (!vq_log_access_ok(vq, vq->log_base))
1888                 return false;
1889
1890         /* Access validation occurs at prefetch time with IOTLB */
1891         if (vq->iotlb)
1892                 return true;
1893
1894         return vq_access_ok(vq, vq->num, vq->desc, vq->avail, vq->used);
1895 }
1896 EXPORT_SYMBOL_GPL(vhost_vq_access_ok);
1897
1898 static struct vhost_umem *vhost_umem_alloc(void)
1899 {
1900         struct vhost_umem *umem = kvzalloc(sizeof(*umem), GFP_KERNEL);
1901
1902         if (!umem)
1903                 return NULL;
1904
1905         umem->umem_tree = RB_ROOT_CACHED;
1906         umem->numem = 0;
1907         INIT_LIST_HEAD(&umem->umem_list);
1908
1909         return umem;
1910 }
1911
1912 static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m)
1913 {
1914         struct vhost_memory mem, *newmem;
1915         struct vhost_memory_region *region;
1916         struct vhost_umem *newumem, *oldumem;
1917         unsigned long size = offsetof(struct vhost_memory, regions);
1918         int i;
1919
1920         if (copy_from_user(&mem, m, size))
1921                 return -EFAULT;
1922         if (mem.padding)
1923                 return -EOPNOTSUPP;
1924         if (mem.nregions > max_mem_regions)
1925                 return -E2BIG;
1926         newmem = kvzalloc(struct_size(newmem, regions, mem.nregions),
1927                         GFP_KERNEL);
1928         if (!newmem)
1929                 return -ENOMEM;
1930
1931         memcpy(newmem, &mem, size);
1932         if (copy_from_user(newmem->regions, m->regions,
1933                            mem.nregions * sizeof *m->regions)) {
1934                 kvfree(newmem);
1935                 return -EFAULT;
1936         }
1937
1938         newumem = vhost_umem_alloc();
1939         if (!newumem) {
1940                 kvfree(newmem);
1941                 return -ENOMEM;
1942         }
1943
1944         for (region = newmem->regions;
1945              region < newmem->regions + mem.nregions;
1946              region++) {
1947                 if (vhost_new_umem_range(newumem,
1948                                          region->guest_phys_addr,
1949                                          region->memory_size,
1950                                          region->guest_phys_addr +
1951                                          region->memory_size - 1,
1952                                          region->userspace_addr,
1953                                          VHOST_ACCESS_RW))
1954                         goto err;
1955         }
1956
1957         if (!memory_access_ok(d, newumem, 0))
1958                 goto err;
1959
1960         oldumem = d->umem;
1961         d->umem = newumem;
1962
1963         /* All memory accesses are done under some VQ mutex. */
1964         for (i = 0; i < d->nvqs; ++i) {
1965                 mutex_lock(&d->vqs[i]->mutex);
1966                 d->vqs[i]->umem = newumem;
1967                 mutex_unlock(&d->vqs[i]->mutex);
1968         }
1969
1970         kvfree(newmem);
1971         vhost_umem_clean(oldumem);
1972         return 0;
1973
1974 err:
1975         vhost_umem_clean(newumem);
1976         kvfree(newmem);
1977         return -EFAULT;
1978 }
1979
1980 static long vhost_vring_set_num(struct vhost_dev *d,
1981                                 struct vhost_virtqueue *vq,
1982                                 void __user *argp)
1983 {
1984         struct vhost_vring_state s;
1985
1986         /* Resizing ring with an active backend?
1987          * You don't want to do that. */
1988         if (vq->private_data)
1989                 return -EBUSY;
1990
1991         if (copy_from_user(&s, argp, sizeof s))
1992                 return -EFAULT;
1993
1994         if (!s.num || s.num > 0xffff || (s.num & (s.num - 1)))
1995                 return -EINVAL;
1996         vq->num = s.num;
1997
1998         return 0;
1999 }
2000
2001 static long vhost_vring_set_addr(struct vhost_dev *d,
2002                                  struct vhost_virtqueue *vq,
2003                                  void __user *argp)
2004 {
2005         struct vhost_vring_addr a;
2006
2007         if (copy_from_user(&a, argp, sizeof a))
2008                 return -EFAULT;
2009         if (a.flags & ~(0x1 << VHOST_VRING_F_LOG))
2010                 return -EOPNOTSUPP;
2011
2012         /* For 32bit, verify that the top 32bits of the user
2013            data are set to zero. */
2014         if ((u64)(unsigned long)a.desc_user_addr != a.desc_user_addr ||
2015             (u64)(unsigned long)a.used_user_addr != a.used_user_addr ||
2016             (u64)(unsigned long)a.avail_user_addr != a.avail_user_addr)
2017                 return -EFAULT;
2018
2019         /* Make sure it's safe to cast pointers to vring types. */
2020         BUILD_BUG_ON(__alignof__ *vq->avail > VRING_AVAIL_ALIGN_SIZE);
2021         BUILD_BUG_ON(__alignof__ *vq->used > VRING_USED_ALIGN_SIZE);
2022         if ((a.avail_user_addr & (VRING_AVAIL_ALIGN_SIZE - 1)) ||
2023             (a.used_user_addr & (VRING_USED_ALIGN_SIZE - 1)) ||
2024             (a.log_guest_addr & (VRING_USED_ALIGN_SIZE - 1)))
2025                 return -EINVAL;
2026
2027         /* We only verify access here if backend is configured.
2028          * If it is not, we don't as size might not have been setup.
2029          * We will verify when backend is configured. */
2030         if (vq->private_data) {
2031                 if (!vq_access_ok(vq, vq->num,
2032                         (void __user *)(unsigned long)a.desc_user_addr,
2033                         (void __user *)(unsigned long)a.avail_user_addr,
2034                         (void __user *)(unsigned long)a.used_user_addr))
2035                         return -EINVAL;
2036
2037                 /* Also validate log access for used ring if enabled. */
2038                 if ((a.flags & (0x1 << VHOST_VRING_F_LOG)) &&
2039                         !log_access_ok(vq->log_base, a.log_guest_addr,
2040                                 sizeof *vq->used +
2041                                 vq->num * sizeof *vq->used->ring))
2042                         return -EINVAL;
2043         }
2044
2045         vq->log_used = !!(a.flags & (0x1 << VHOST_VRING_F_LOG));
2046         vq->desc = (void __user *)(unsigned long)a.desc_user_addr;
2047         vq->avail = (void __user *)(unsigned long)a.avail_user_addr;
2048         vq->log_addr = a.log_guest_addr;
2049         vq->used = (void __user *)(unsigned long)a.used_user_addr;
2050
2051         return 0;
2052 }
2053
2054 static long vhost_vring_set_num_addr(struct vhost_dev *d,
2055                                      struct vhost_virtqueue *vq,
2056                                      unsigned int ioctl,
2057                                      void __user *argp)
2058 {
2059         long r;
2060
2061         mutex_lock(&vq->mutex);
2062
2063 #if VHOST_ARCH_CAN_ACCEL_UACCESS
2064         /* Unregister MMU notifer to allow invalidation callback
2065          * can access vq->uaddrs[] without holding a lock.
2066          */
2067         if (d->mm)
2068                 mmu_notifier_unregister(&d->mmu_notifier, d->mm);
2069
2070         vhost_uninit_vq_maps(vq);
2071 #endif
2072
2073         switch (ioctl) {
2074         case VHOST_SET_VRING_NUM:
2075                 r = vhost_vring_set_num(d, vq, argp);
2076                 break;
2077         case VHOST_SET_VRING_ADDR:
2078                 r = vhost_vring_set_addr(d, vq, argp);
2079                 break;
2080         default:
2081                 BUG();
2082         }
2083
2084 #if VHOST_ARCH_CAN_ACCEL_UACCESS
2085         vhost_setup_vq_uaddr(vq);
2086
2087         if (d->mm)
2088                 mmu_notifier_register(&d->mmu_notifier, d->mm);
2089 #endif
2090
2091         mutex_unlock(&vq->mutex);
2092
2093         return r;
2094 }
2095 long vhost_vring_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp)
2096 {
2097         struct file *eventfp, *filep = NULL;
2098         bool pollstart = false, pollstop = false;
2099         struct eventfd_ctx *ctx = NULL;
2100         u32 __user *idxp = argp;
2101         struct vhost_virtqueue *vq;
2102         struct vhost_vring_state s;
2103         struct vhost_vring_file f;
2104         u32 idx;
2105         long r;
2106
2107         r = get_user(idx, idxp);
2108         if (r < 0)
2109                 return r;
2110         if (idx >= d->nvqs)
2111                 return -ENOBUFS;
2112
2113         idx = array_index_nospec(idx, d->nvqs);
2114         vq = d->vqs[idx];
2115
2116         if (ioctl == VHOST_SET_VRING_NUM ||
2117             ioctl == VHOST_SET_VRING_ADDR) {
2118                 return vhost_vring_set_num_addr(d, vq, ioctl, argp);
2119         }
2120
2121         mutex_lock(&vq->mutex);
2122
2123         switch (ioctl) {
2124         case VHOST_SET_VRING_BASE:
2125                 /* Moving base with an active backend?
2126                  * You don't want to do that. */
2127                 if (vq->private_data) {
2128                         r = -EBUSY;
2129                         break;
2130                 }
2131                 if (copy_from_user(&s, argp, sizeof s)) {
2132                         r = -EFAULT;
2133                         break;
2134                 }
2135                 if (s.num > 0xffff) {
2136                         r = -EINVAL;
2137                         break;
2138                 }
2139                 vq->last_avail_idx = s.num;
2140                 /* Forget the cached index value. */
2141                 vq->avail_idx = vq->last_avail_idx;
2142                 break;
2143         case VHOST_GET_VRING_BASE:
2144                 s.index = idx;
2145                 s.num = vq->last_avail_idx;
2146                 if (copy_to_user(argp, &s, sizeof s))
2147                         r = -EFAULT;
2148                 break;
2149         case VHOST_SET_VRING_KICK:
2150                 if (copy_from_user(&f, argp, sizeof f)) {
2151                         r = -EFAULT;
2152                         break;
2153                 }
2154                 eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd);
2155                 if (IS_ERR(eventfp)) {
2156                         r = PTR_ERR(eventfp);
2157                         break;
2158                 }
2159                 if (eventfp != vq->kick) {
2160                         pollstop = (filep = vq->kick) != NULL;
2161                         pollstart = (vq->kick = eventfp) != NULL;
2162                 } else
2163                         filep = eventfp;
2164                 break;
2165         case VHOST_SET_VRING_CALL:
2166                 if (copy_from_user(&f, argp, sizeof f)) {
2167                         r = -EFAULT;
2168                         break;
2169                 }
2170                 ctx = f.fd == -1 ? NULL : eventfd_ctx_fdget(f.fd);
2171                 if (IS_ERR(ctx)) {
2172                         r = PTR_ERR(ctx);
2173                         break;
2174                 }
2175                 swap(ctx, vq->call_ctx);
2176                 break;
2177         case VHOST_SET_VRING_ERR:
2178                 if (copy_from_user(&f, argp, sizeof f)) {
2179                         r = -EFAULT;
2180                         break;
2181                 }
2182                 ctx = f.fd == -1 ? NULL : eventfd_ctx_fdget(f.fd);
2183                 if (IS_ERR(ctx)) {
2184                         r = PTR_ERR(ctx);
2185                         break;
2186                 }
2187                 swap(ctx, vq->error_ctx);
2188                 break;
2189         case VHOST_SET_VRING_ENDIAN:
2190                 r = vhost_set_vring_endian(vq, argp);
2191                 break;
2192         case VHOST_GET_VRING_ENDIAN:
2193                 r = vhost_get_vring_endian(vq, idx, argp);
2194                 break;
2195         case VHOST_SET_VRING_BUSYLOOP_TIMEOUT:
2196                 if (copy_from_user(&s, argp, sizeof(s))) {
2197                         r = -EFAULT;
2198                         break;
2199                 }
2200                 vq->busyloop_timeout = s.num;
2201                 break;
2202         case VHOST_GET_VRING_BUSYLOOP_TIMEOUT:
2203                 s.index = idx;
2204                 s.num = vq->busyloop_timeout;
2205                 if (copy_to_user(argp, &s, sizeof(s)))
2206                         r = -EFAULT;
2207                 break;
2208         default:
2209                 r = -ENOIOCTLCMD;
2210         }
2211
2212         if (pollstop && vq->handle_kick)
2213                 vhost_poll_stop(&vq->poll);
2214
2215         if (!IS_ERR_OR_NULL(ctx))
2216                 eventfd_ctx_put(ctx);
2217         if (filep)
2218                 fput(filep);
2219
2220         if (pollstart && vq->handle_kick)
2221                 r = vhost_poll_start(&vq->poll, vq->kick);
2222
2223         mutex_unlock(&vq->mutex);
2224
2225         if (pollstop && vq->handle_kick)
2226                 vhost_poll_flush(&vq->poll);
2227         return r;
2228 }
2229 EXPORT_SYMBOL_GPL(vhost_vring_ioctl);
2230
2231 int vhost_init_device_iotlb(struct vhost_dev *d, bool enabled)
2232 {
2233         struct vhost_umem *niotlb, *oiotlb;
2234         int i;
2235
2236         niotlb = vhost_umem_alloc();
2237         if (!niotlb)
2238                 return -ENOMEM;
2239
2240         oiotlb = d->iotlb;
2241         d->iotlb = niotlb;
2242
2243         for (i = 0; i < d->nvqs; ++i) {
2244                 struct vhost_virtqueue *vq = d->vqs[i];
2245
2246                 mutex_lock(&vq->mutex);
2247                 vq->iotlb = niotlb;
2248                 __vhost_vq_meta_reset(vq);
2249                 mutex_unlock(&vq->mutex);
2250         }
2251
2252         vhost_umem_clean(oiotlb);
2253
2254         return 0;
2255 }
2256 EXPORT_SYMBOL_GPL(vhost_init_device_iotlb);
2257
2258 /* Caller must have device mutex */
2259 long vhost_dev_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp)
2260 {
2261         struct eventfd_ctx *ctx;
2262         u64 p;
2263         long r;
2264         int i, fd;
2265
2266         /* If you are not the owner, you can become one */
2267         if (ioctl == VHOST_SET_OWNER) {
2268                 r = vhost_dev_set_owner(d);
2269                 goto done;
2270         }
2271
2272         /* You must be the owner to do anything else */
2273         r = vhost_dev_check_owner(d);
2274         if (r)
2275                 goto done;
2276
2277         switch (ioctl) {
2278         case VHOST_SET_MEM_TABLE:
2279                 r = vhost_set_memory(d, argp);
2280                 break;
2281         case VHOST_SET_LOG_BASE:
2282                 if (copy_from_user(&p, argp, sizeof p)) {
2283                         r = -EFAULT;
2284                         break;
2285                 }
2286                 if ((u64)(unsigned long)p != p) {
2287                         r = -EFAULT;
2288                         break;
2289                 }
2290                 for (i = 0; i < d->nvqs; ++i) {
2291                         struct vhost_virtqueue *vq;
2292                         void __user *base = (void __user *)(unsigned long)p;
2293                         vq = d->vqs[i];
2294                         mutex_lock(&vq->mutex);
2295                         /* If ring is inactive, will check when it's enabled. */
2296                         if (vq->private_data && !vq_log_access_ok(vq, base))
2297                                 r = -EFAULT;
2298                         else
2299                                 vq->log_base = base;
2300                         mutex_unlock(&vq->mutex);
2301                 }
2302                 break;
2303         case VHOST_SET_LOG_FD:
2304                 r = get_user(fd, (int __user *)argp);
2305                 if (r < 0)
2306                         break;
2307                 ctx = fd == -1 ? NULL : eventfd_ctx_fdget(fd);
2308                 if (IS_ERR(ctx)) {
2309                         r = PTR_ERR(ctx);
2310                         break;
2311                 }
2312                 swap(ctx, d->log_ctx);
2313                 for (i = 0; i < d->nvqs; ++i) {
2314                         mutex_lock(&d->vqs[i]->mutex);
2315                         d->vqs[i]->log_ctx = d->log_ctx;
2316                         mutex_unlock(&d->vqs[i]->mutex);
2317                 }
2318                 if (ctx)
2319                         eventfd_ctx_put(ctx);
2320                 break;
2321         default:
2322                 r = -ENOIOCTLCMD;
2323                 break;
2324         }
2325 done:
2326         return r;
2327 }
2328 EXPORT_SYMBOL_GPL(vhost_dev_ioctl);
2329
2330 /* TODO: This is really inefficient.  We need something like get_user()
2331  * (instruction directly accesses the data, with an exception table entry
2332  * returning -EFAULT). See Documentation/x86/exception-tables.rst.
2333  */
2334 static int set_bit_to_user(int nr, void __user *addr)
2335 {
2336         unsigned long log = (unsigned long)addr;
2337         struct page *page;
2338         void *base;
2339         int bit = nr + (log % PAGE_SIZE) * 8;
2340         int r;
2341
2342         r = get_user_pages_fast(log, 1, FOLL_WRITE, &page);
2343         if (r < 0)
2344                 return r;
2345         BUG_ON(r != 1);
2346         base = kmap_atomic(page);
2347         set_bit(bit, base);
2348         kunmap_atomic(base);
2349         set_page_dirty_lock(page);
2350         put_page(page);
2351         return 0;
2352 }
2353
2354 static int log_write(void __user *log_base,
2355                      u64 write_address, u64 write_length)
2356 {
2357         u64 write_page = write_address / VHOST_PAGE_SIZE;
2358         int r;
2359
2360         if (!write_length)
2361                 return 0;
2362         write_length += write_address % VHOST_PAGE_SIZE;
2363         for (;;) {
2364                 u64 base = (u64)(unsigned long)log_base;
2365                 u64 log = base + write_page / 8;
2366                 int bit = write_page % 8;
2367                 if ((u64)(unsigned long)log != log)
2368                         return -EFAULT;
2369                 r = set_bit_to_user(bit, (void __user *)(unsigned long)log);
2370                 if (r < 0)
2371                         return r;
2372                 if (write_length <= VHOST_PAGE_SIZE)
2373                         break;
2374                 write_length -= VHOST_PAGE_SIZE;
2375                 write_page += 1;
2376         }
2377         return r;
2378 }
2379
2380 static int log_write_hva(struct vhost_virtqueue *vq, u64 hva, u64 len)
2381 {
2382         struct vhost_umem *umem = vq->umem;
2383         struct vhost_umem_node *u;
2384         u64 start, end, l, min;
2385         int r;
2386         bool hit = false;
2387
2388         while (len) {
2389                 min = len;
2390                 /* More than one GPAs can be mapped into a single HVA. So
2391                  * iterate all possible umems here to be safe.
2392                  */
2393                 list_for_each_entry(u, &umem->umem_list, link) {
2394                         if (u->userspace_addr > hva - 1 + len ||
2395                             u->userspace_addr - 1 + u->size < hva)
2396                                 continue;
2397                         start = max(u->userspace_addr, hva);
2398                         end = min(u->userspace_addr - 1 + u->size,
2399                                   hva - 1 + len);
2400                         l = end - start + 1;
2401                         r = log_write(vq->log_base,
2402                                       u->start + start - u->userspace_addr,
2403                                       l);
2404                         if (r < 0)
2405                                 return r;
2406                         hit = true;
2407                         min = min(l, min);
2408                 }
2409
2410                 if (!hit)
2411                         return -EFAULT;
2412
2413                 len -= min;
2414                 hva += min;
2415         }
2416
2417         return 0;
2418 }
2419
2420 static int log_used(struct vhost_virtqueue *vq, u64 used_offset, u64 len)
2421 {
2422         struct iovec iov[64];
2423         int i, ret;
2424
2425         if (!vq->iotlb)
2426                 return log_write(vq->log_base, vq->log_addr + used_offset, len);
2427
2428         ret = translate_desc(vq, (uintptr_t)vq->used + used_offset,
2429                              len, iov, 64, VHOST_ACCESS_WO);
2430         if (ret < 0)
2431                 return ret;
2432
2433         for (i = 0; i < ret; i++) {
2434                 ret = log_write_hva(vq, (uintptr_t)iov[i].iov_base,
2435                                     iov[i].iov_len);
2436                 if (ret)
2437                         return ret;
2438         }
2439
2440         return 0;
2441 }
2442
2443 int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
2444                     unsigned int log_num, u64 len, struct iovec *iov, int count)
2445 {
2446         int i, r;
2447
2448         /* Make sure data written is seen before log. */
2449         smp_wmb();
2450
2451         if (vq->iotlb) {
2452                 for (i = 0; i < count; i++) {
2453                         r = log_write_hva(vq, (uintptr_t)iov[i].iov_base,
2454                                           iov[i].iov_len);
2455                         if (r < 0)
2456                                 return r;
2457                 }
2458                 return 0;
2459         }
2460
2461         for (i = 0; i < log_num; ++i) {
2462                 u64 l = min(log[i].len, len);
2463                 r = log_write(vq->log_base, log[i].addr, l);
2464                 if (r < 0)
2465                         return r;
2466                 len -= l;
2467                 if (!len) {
2468                         if (vq->log_ctx)
2469                                 eventfd_signal(vq->log_ctx, 1);
2470                         return 0;
2471                 }
2472         }
2473         /* Length written exceeds what we have stored. This is a bug. */
2474         BUG();
2475         return 0;
2476 }
2477 EXPORT_SYMBOL_GPL(vhost_log_write);
2478
2479 static int vhost_update_used_flags(struct vhost_virtqueue *vq)
2480 {
2481         void __user *used;
2482         if (vhost_put_used_flags(vq))
2483                 return -EFAULT;
2484         if (unlikely(vq->log_used)) {
2485                 /* Make sure the flag is seen before log. */
2486                 smp_wmb();
2487                 /* Log used flag write. */
2488                 used = &vq->used->flags;
2489                 log_used(vq, (used - (void __user *)vq->used),
2490                          sizeof vq->used->flags);
2491                 if (vq->log_ctx)
2492                         eventfd_signal(vq->log_ctx, 1);
2493         }
2494         return 0;
2495 }
2496
2497 static int vhost_update_avail_event(struct vhost_virtqueue *vq, u16 avail_event)
2498 {
2499         if (vhost_put_avail_event(vq))
2500                 return -EFAULT;
2501         if (unlikely(vq->log_used)) {
2502                 void __user *used;
2503                 /* Make sure the event is seen before log. */
2504                 smp_wmb();
2505                 /* Log avail event write */
2506                 used = vhost_avail_event(vq);
2507                 log_used(vq, (used - (void __user *)vq->used),
2508                          sizeof *vhost_avail_event(vq));
2509                 if (vq->log_ctx)
2510                         eventfd_signal(vq->log_ctx, 1);
2511         }
2512         return 0;
2513 }
2514
2515 int vhost_vq_init_access(struct vhost_virtqueue *vq)
2516 {
2517         __virtio16 last_used_idx;
2518         int r;
2519         bool is_le = vq->is_le;
2520
2521         if (!vq->private_data)
2522                 return 0;
2523
2524         vhost_init_is_le(vq);
2525
2526         r = vhost_update_used_flags(vq);
2527         if (r)
2528                 goto err;
2529         vq->signalled_used_valid = false;
2530         if (!vq->iotlb &&
2531             !access_ok(&vq->used->idx, sizeof vq->used->idx)) {
2532                 r = -EFAULT;
2533                 goto err;
2534         }
2535         r = vhost_get_used_idx(vq, &last_used_idx);
2536         if (r) {
2537                 vq_err(vq, "Can't access used idx at %p\n",
2538                        &vq->used->idx);
2539                 goto err;
2540         }
2541         vq->last_used_idx = vhost16_to_cpu(vq, last_used_idx);
2542         return 0;
2543
2544 err:
2545         vq->is_le = is_le;
2546         return r;
2547 }
2548 EXPORT_SYMBOL_GPL(vhost_vq_init_access);
2549
2550 static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len,
2551                           struct iovec iov[], int iov_size, int access)
2552 {
2553         const struct vhost_umem_node *node;
2554         struct vhost_dev *dev = vq->dev;
2555         struct vhost_umem *umem = dev->iotlb ? dev->iotlb : dev->umem;
2556         struct iovec *_iov;
2557         u64 s = 0;
2558         int ret = 0;
2559
2560         while ((u64)len > s) {
2561                 u64 size;
2562                 if (unlikely(ret >= iov_size)) {
2563                         ret = -ENOBUFS;
2564                         break;
2565                 }
2566
2567                 node = vhost_umem_interval_tree_iter_first(&umem->umem_tree,
2568                                                         addr, addr + len - 1);
2569                 if (node == NULL || node->start > addr) {
2570                         if (umem != dev->iotlb) {
2571                                 ret = -EFAULT;
2572                                 break;
2573                         }
2574                         ret = -EAGAIN;
2575                         break;
2576                 } else if (!(node->perm & access)) {
2577                         ret = -EPERM;
2578                         break;
2579                 }
2580
2581                 _iov = iov + ret;
2582                 size = node->size - addr + node->start;
2583                 _iov->iov_len = min((u64)len - s, size);
2584                 _iov->iov_base = (void __user *)(unsigned long)
2585                         (node->userspace_addr + addr - node->start);
2586                 s += size;
2587                 addr += size;
2588                 ++ret;
2589         }
2590
2591         if (ret == -EAGAIN)
2592                 vhost_iotlb_miss(vq, addr, access);
2593         return ret;
2594 }
2595
2596 /* Each buffer in the virtqueues is actually a chain of descriptors.  This
2597  * function returns the next descriptor in the chain,
2598  * or -1U if we're at the end. */
2599 static unsigned next_desc(struct vhost_virtqueue *vq, struct vring_desc *desc)
2600 {
2601         unsigned int next;
2602
2603         /* If this descriptor says it doesn't chain, we're done. */
2604         if (!(desc->flags & cpu_to_vhost16(vq, VRING_DESC_F_NEXT)))
2605                 return -1U;
2606
2607         /* Check they're not leading us off end of descriptors. */
2608         next = vhost16_to_cpu(vq, READ_ONCE(desc->next));
2609         return next;
2610 }
2611
2612 static int get_indirect(struct vhost_virtqueue *vq,
2613                         struct iovec iov[], unsigned int iov_size,
2614                         unsigned int *out_num, unsigned int *in_num,
2615                         struct vhost_log *log, unsigned int *log_num,
2616                         struct vring_desc *indirect)
2617 {
2618         struct vring_desc desc;
2619         unsigned int i = 0, count, found = 0;
2620         u32 len = vhost32_to_cpu(vq, indirect->len);
2621         struct iov_iter from;
2622         int ret, access;
2623
2624         /* Sanity check */
2625         if (unlikely(len % sizeof desc)) {
2626                 vq_err(vq, "Invalid length in indirect descriptor: "
2627                        "len 0x%llx not multiple of 0x%zx\n",
2628                        (unsigned long long)len,
2629                        sizeof desc);
2630                 return -EINVAL;
2631         }
2632
2633         ret = translate_desc(vq, vhost64_to_cpu(vq, indirect->addr), len, vq->indirect,
2634                              UIO_MAXIOV, VHOST_ACCESS_RO);
2635         if (unlikely(ret < 0)) {
2636                 if (ret != -EAGAIN)
2637                         vq_err(vq, "Translation failure %d in indirect.\n", ret);
2638                 return ret;
2639         }
2640         iov_iter_init(&from, READ, vq->indirect, ret, len);
2641
2642         /* We will use the result as an address to read from, so most
2643          * architectures only need a compiler barrier here. */
2644         read_barrier_depends();
2645
2646         count = len / sizeof desc;
2647         /* Buffers are chained via a 16 bit next field, so
2648          * we can have at most 2^16 of these. */
2649         if (unlikely(count > USHRT_MAX + 1)) {
2650                 vq_err(vq, "Indirect buffer length too big: %d\n",
2651                        indirect->len);
2652                 return -E2BIG;
2653         }
2654
2655         do {
2656                 unsigned iov_count = *in_num + *out_num;
2657                 if (unlikely(++found > count)) {
2658                         vq_err(vq, "Loop detected: last one at %u "
2659                                "indirect size %u\n",
2660                                i, count);
2661                         return -EINVAL;
2662                 }
2663                 if (unlikely(!copy_from_iter_full(&desc, sizeof(desc), &from))) {
2664                         vq_err(vq, "Failed indirect descriptor: idx %d, %zx\n",
2665                                i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc);
2666                         return -EINVAL;
2667                 }
2668                 if (unlikely(desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT))) {
2669                         vq_err(vq, "Nested indirect descriptor: idx %d, %zx\n",
2670                                i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc);
2671                         return -EINVAL;
2672                 }
2673
2674                 if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE))
2675                         access = VHOST_ACCESS_WO;
2676                 else
2677                         access = VHOST_ACCESS_RO;
2678
2679                 ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr),
2680                                      vhost32_to_cpu(vq, desc.len), iov + iov_count,
2681                                      iov_size - iov_count, access);
2682                 if (unlikely(ret < 0)) {
2683                         if (ret != -EAGAIN)
2684                                 vq_err(vq, "Translation failure %d indirect idx %d\n",
2685                                         ret, i);
2686                         return ret;
2687                 }
2688                 /* If this is an input descriptor, increment that count. */
2689                 if (access == VHOST_ACCESS_WO) {
2690                         *in_num += ret;
2691                         if (unlikely(log)) {
2692                                 log[*log_num].addr = vhost64_to_cpu(vq, desc.addr);
2693                                 log[*log_num].len = vhost32_to_cpu(vq, desc.len);
2694                                 ++*log_num;
2695                         }
2696                 } else {
2697                         /* If it's an output descriptor, they're all supposed
2698                          * to come before any input descriptors. */
2699                         if (unlikely(*in_num)) {
2700                                 vq_err(vq, "Indirect descriptor "
2701                                        "has out after in: idx %d\n", i);
2702                                 return -EINVAL;
2703                         }
2704                         *out_num += ret;
2705                 }
2706         } while ((i = next_desc(vq, &desc)) != -1);
2707         return 0;
2708 }
2709
2710 /* This looks in the virtqueue and for the first available buffer, and converts
2711  * it to an iovec for convenient access.  Since descriptors consist of some
2712  * number of output then some number of input descriptors, it's actually two
2713  * iovecs, but we pack them into one and note how many of each there were.
2714  *
2715  * This function returns the descriptor number found, or vq->num (which is
2716  * never a valid descriptor number) if none was found.  A negative code is
2717  * returned on error. */
2718 int vhost_get_vq_desc(struct vhost_virtqueue *vq,
2719                       struct iovec iov[], unsigned int iov_size,
2720                       unsigned int *out_num, unsigned int *in_num,
2721                       struct vhost_log *log, unsigned int *log_num)
2722 {
2723         struct vring_desc desc;
2724         unsigned int i, head, found = 0;
2725         u16 last_avail_idx;
2726         __virtio16 avail_idx;
2727         __virtio16 ring_head;
2728         int ret, access;
2729
2730         /* Check it isn't doing very strange things with descriptor numbers. */
2731         last_avail_idx = vq->last_avail_idx;
2732
2733         if (vq->avail_idx == vq->last_avail_idx) {
2734                 if (unlikely(vhost_get_avail_idx(vq, &avail_idx))) {
2735                         vq_err(vq, "Failed to access avail idx at %p\n",
2736                                 &vq->avail->idx);
2737                         return -EFAULT;
2738                 }
2739                 vq->avail_idx = vhost16_to_cpu(vq, avail_idx);
2740
2741                 if (unlikely((u16)(vq->avail_idx - last_avail_idx) > vq->num)) {
2742                         vq_err(vq, "Guest moved used index from %u to %u",
2743                                 last_avail_idx, vq->avail_idx);
2744                         return -EFAULT;
2745                 }
2746
2747                 /* If there's nothing new since last we looked, return
2748                  * invalid.
2749                  */
2750                 if (vq->avail_idx == last_avail_idx)
2751                         return vq->num;
2752
2753                 /* Only get avail ring entries after they have been
2754                  * exposed by guest.
2755                  */
2756                 smp_rmb();
2757         }
2758
2759         /* Grab the next descriptor number they're advertising, and increment
2760          * the index we've seen. */
2761         if (unlikely(vhost_get_avail_head(vq, &ring_head, last_avail_idx))) {
2762                 vq_err(vq, "Failed to read head: idx %d address %p\n",
2763                        last_avail_idx,
2764                        &vq->avail->ring[last_avail_idx % vq->num]);
2765                 return -EFAULT;
2766         }
2767
2768         head = vhost16_to_cpu(vq, ring_head);
2769
2770         /* If their number is silly, that's an error. */
2771         if (unlikely(head >= vq->num)) {
2772                 vq_err(vq, "Guest says index %u > %u is available",
2773                        head, vq->num);
2774                 return -EINVAL;
2775         }
2776
2777         /* When we start there are none of either input nor output. */
2778         *out_num = *in_num = 0;
2779         if (unlikely(log))
2780                 *log_num = 0;
2781
2782         i = head;
2783         do {
2784                 unsigned iov_count = *in_num + *out_num;
2785                 if (unlikely(i >= vq->num)) {
2786                         vq_err(vq, "Desc index is %u > %u, head = %u",
2787                                i, vq->num, head);
2788                         return -EINVAL;
2789                 }
2790                 if (unlikely(++found > vq->num)) {
2791                         vq_err(vq, "Loop detected: last one at %u "
2792                                "vq size %u head %u\n",
2793                                i, vq->num, head);
2794                         return -EINVAL;
2795                 }
2796                 ret = vhost_get_desc(vq, &desc, i);
2797                 if (unlikely(ret)) {
2798                         vq_err(vq, "Failed to get descriptor: idx %d addr %p\n",
2799                                i, vq->desc + i);
2800                         return -EFAULT;
2801                 }
2802                 if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT)) {
2803                         ret = get_indirect(vq, iov, iov_size,
2804                                            out_num, in_num,
2805                                            log, log_num, &desc);
2806                         if (unlikely(ret < 0)) {
2807                                 if (ret != -EAGAIN)
2808                                         vq_err(vq, "Failure detected "
2809                                                 "in indirect descriptor at idx %d\n", i);
2810                                 return ret;
2811                         }
2812                         continue;
2813                 }
2814
2815                 if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE))
2816                         access = VHOST_ACCESS_WO;
2817                 else
2818                         access = VHOST_ACCESS_RO;
2819                 ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr),
2820                                      vhost32_to_cpu(vq, desc.len), iov + iov_count,
2821                                      iov_size - iov_count, access);
2822                 if (unlikely(ret < 0)) {
2823                         if (ret != -EAGAIN)
2824                                 vq_err(vq, "Translation failure %d descriptor idx %d\n",
2825                                         ret, i);
2826                         return ret;
2827                 }
2828                 if (access == VHOST_ACCESS_WO) {
2829                         /* If this is an input descriptor,
2830                          * increment that count. */
2831                         *in_num += ret;
2832                         if (unlikely(log)) {
2833                                 log[*log_num].addr = vhost64_to_cpu(vq, desc.addr);
2834                                 log[*log_num].len = vhost32_to_cpu(vq, desc.len);
2835                                 ++*log_num;
2836                         }
2837                 } else {
2838                         /* If it's an output descriptor, they're all supposed
2839                          * to come before any input descriptors. */
2840                         if (unlikely(*in_num)) {
2841                                 vq_err(vq, "Descriptor has out after in: "
2842                                        "idx %d\n", i);
2843                                 return -EINVAL;
2844                         }
2845                         *out_num += ret;
2846                 }
2847         } while ((i = next_desc(vq, &desc)) != -1);
2848
2849         /* On success, increment avail index. */
2850         vq->last_avail_idx++;
2851
2852         /* Assume notifications from guest are disabled at this point,
2853          * if they aren't we would need to update avail_event index. */
2854         BUG_ON(!(vq->used_flags & VRING_USED_F_NO_NOTIFY));
2855         return head;
2856 }
2857 EXPORT_SYMBOL_GPL(vhost_get_vq_desc);
2858
2859 /* Reverse the effect of vhost_get_vq_desc. Useful for error handling. */
2860 void vhost_discard_vq_desc(struct vhost_virtqueue *vq, int n)
2861 {
2862         vq->last_avail_idx -= n;
2863 }
2864 EXPORT_SYMBOL_GPL(vhost_discard_vq_desc);
2865
2866 /* After we've used one of their buffers, we tell them about it.  We'll then
2867  * want to notify the guest, using eventfd. */
2868 int vhost_add_used(struct vhost_virtqueue *vq, unsigned int head, int len)
2869 {
2870         struct vring_used_elem heads = {
2871                 cpu_to_vhost32(vq, head),
2872                 cpu_to_vhost32(vq, len)
2873         };
2874
2875         return vhost_add_used_n(vq, &heads, 1);
2876 }
2877 EXPORT_SYMBOL_GPL(vhost_add_used);
2878
2879 static int __vhost_add_used_n(struct vhost_virtqueue *vq,
2880                             struct vring_used_elem *heads,
2881                             unsigned count)
2882 {
2883         struct vring_used_elem __user *used;
2884         u16 old, new;
2885         int start;
2886
2887         start = vq->last_used_idx & (vq->num - 1);
2888         used = vq->used->ring + start;
2889         if (vhost_put_used(vq, heads, start, count)) {
2890                 vq_err(vq, "Failed to write used");
2891                 return -EFAULT;
2892         }
2893         if (unlikely(vq->log_used)) {
2894                 /* Make sure data is seen before log. */
2895                 smp_wmb();
2896                 /* Log used ring entry write. */
2897                 log_used(vq, ((void __user *)used - (void __user *)vq->used),
2898                          count * sizeof *used);
2899         }
2900         old = vq->last_used_idx;
2901         new = (vq->last_used_idx += count);
2902         /* If the driver never bothers to signal in a very long while,
2903          * used index might wrap around. If that happens, invalidate
2904          * signalled_used index we stored. TODO: make sure driver
2905          * signals at least once in 2^16 and remove this. */
2906         if (unlikely((u16)(new - vq->signalled_used) < (u16)(new - old)))
2907                 vq->signalled_used_valid = false;
2908         return 0;
2909 }
2910
2911 /* After we've used one of their buffers, we tell them about it.  We'll then
2912  * want to notify the guest, using eventfd. */
2913 int vhost_add_used_n(struct vhost_virtqueue *vq, struct vring_used_elem *heads,
2914                      unsigned count)
2915 {
2916         int start, n, r;
2917
2918         start = vq->last_used_idx & (vq->num - 1);
2919         n = vq->num - start;
2920         if (n < count) {
2921                 r = __vhost_add_used_n(vq, heads, n);
2922                 if (r < 0)
2923                         return r;
2924                 heads += n;
2925                 count -= n;
2926         }
2927         r = __vhost_add_used_n(vq, heads, count);
2928
2929         /* Make sure buffer is written before we update index. */
2930         smp_wmb();
2931         if (vhost_put_used_idx(vq)) {
2932                 vq_err(vq, "Failed to increment used idx");
2933                 return -EFAULT;
2934         }
2935         if (unlikely(vq->log_used)) {
2936                 /* Make sure used idx is seen before log. */
2937                 smp_wmb();
2938                 /* Log used index update. */
2939                 log_used(vq, offsetof(struct vring_used, idx),
2940                          sizeof vq->used->idx);
2941                 if (vq->log_ctx)
2942                         eventfd_signal(vq->log_ctx, 1);
2943         }
2944         return r;
2945 }
2946 EXPORT_SYMBOL_GPL(vhost_add_used_n);
2947
2948 static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2949 {
2950         __u16 old, new;
2951         __virtio16 event;
2952         bool v;
2953         /* Flush out used index updates. This is paired
2954          * with the barrier that the Guest executes when enabling
2955          * interrupts. */
2956         smp_mb();
2957
2958         if (vhost_has_feature(vq, VIRTIO_F_NOTIFY_ON_EMPTY) &&
2959             unlikely(vq->avail_idx == vq->last_avail_idx))
2960                 return true;
2961
2962         if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
2963                 __virtio16 flags;
2964                 if (vhost_get_avail_flags(vq, &flags)) {
2965                         vq_err(vq, "Failed to get flags");
2966                         return true;
2967                 }
2968                 return !(flags & cpu_to_vhost16(vq, VRING_AVAIL_F_NO_INTERRUPT));
2969         }
2970         old = vq->signalled_used;
2971         v = vq->signalled_used_valid;
2972         new = vq->signalled_used = vq->last_used_idx;
2973         vq->signalled_used_valid = true;
2974
2975         if (unlikely(!v))
2976                 return true;
2977
2978         if (vhost_get_used_event(vq, &event)) {
2979                 vq_err(vq, "Failed to get used event idx");
2980                 return true;
2981         }
2982         return vring_need_event(vhost16_to_cpu(vq, event), new, old);
2983 }
2984
2985 /* This actually signals the guest, using eventfd. */
2986 void vhost_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq)
2987 {
2988         /* Signal the Guest tell them we used something up. */
2989         if (vq->call_ctx && vhost_notify(dev, vq))
2990                 eventfd_signal(vq->call_ctx, 1);
2991 }
2992 EXPORT_SYMBOL_GPL(vhost_signal);
2993
2994 /* And here's the combo meal deal.  Supersize me! */
2995 void vhost_add_used_and_signal(struct vhost_dev *dev,
2996                                struct vhost_virtqueue *vq,
2997                                unsigned int head, int len)
2998 {
2999         vhost_add_used(vq, head, len);
3000         vhost_signal(dev, vq);
3001 }
3002 EXPORT_SYMBOL_GPL(vhost_add_used_and_signal);
3003
3004 /* multi-buffer version of vhost_add_used_and_signal */
3005 void vhost_add_used_and_signal_n(struct vhost_dev *dev,
3006                                  struct vhost_virtqueue *vq,
3007                                  struct vring_used_elem *heads, unsigned count)
3008 {
3009         vhost_add_used_n(vq, heads, count);
3010         vhost_signal(dev, vq);
3011 }
3012 EXPORT_SYMBOL_GPL(vhost_add_used_and_signal_n);
3013
3014 /* return true if we're sure that avaiable ring is empty */
3015 bool vhost_vq_avail_empty(struct vhost_dev *dev, struct vhost_virtqueue *vq)
3016 {
3017         __virtio16 avail_idx;
3018         int r;
3019
3020         if (vq->avail_idx != vq->last_avail_idx)
3021                 return false;
3022
3023         r = vhost_get_avail_idx(vq, &avail_idx);
3024         if (unlikely(r))
3025                 return false;
3026         vq->avail_idx = vhost16_to_cpu(vq, avail_idx);
3027
3028         return vq->avail_idx == vq->last_avail_idx;
3029 }
3030 EXPORT_SYMBOL_GPL(vhost_vq_avail_empty);
3031
3032 /* OK, now we need to know about added descriptors. */
3033 bool vhost_enable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
3034 {
3035         __virtio16 avail_idx;
3036         int r;
3037
3038         if (!(vq->used_flags & VRING_USED_F_NO_NOTIFY))
3039                 return false;
3040         vq->used_flags &= ~VRING_USED_F_NO_NOTIFY;
3041         if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
3042                 r = vhost_update_used_flags(vq);
3043                 if (r) {
3044                         vq_err(vq, "Failed to enable notification at %p: %d\n",
3045                                &vq->used->flags, r);
3046                         return false;
3047                 }
3048         } else {
3049                 r = vhost_update_avail_event(vq, vq->avail_idx);
3050                 if (r) {
3051                         vq_err(vq, "Failed to update avail event index at %p: %d\n",
3052                                vhost_avail_event(vq), r);
3053                         return false;
3054                 }
3055         }
3056         /* They could have slipped one in as we were doing that: make
3057          * sure it's written, then check again. */
3058         smp_mb();
3059         r = vhost_get_avail_idx(vq, &avail_idx);
3060         if (r) {
3061                 vq_err(vq, "Failed to check avail idx at %p: %d\n",
3062                        &vq->avail->idx, r);
3063                 return false;
3064         }
3065
3066         return vhost16_to_cpu(vq, avail_idx) != vq->avail_idx;
3067 }
3068 EXPORT_SYMBOL_GPL(vhost_enable_notify);
3069
3070 /* We don't need to be notified again. */
3071 void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
3072 {
3073         int r;
3074
3075         if (vq->used_flags & VRING_USED_F_NO_NOTIFY)
3076                 return;
3077         vq->used_flags |= VRING_USED_F_NO_NOTIFY;
3078         if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) {
3079                 r = vhost_update_used_flags(vq);
3080                 if (r)
3081                         vq_err(vq, "Failed to enable notification at %p: %d\n",
3082                                &vq->used->flags, r);
3083         }
3084 }
3085 EXPORT_SYMBOL_GPL(vhost_disable_notify);
3086
3087 /* Create a new message. */
3088 struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type)
3089 {
3090         struct vhost_msg_node *node = kmalloc(sizeof *node, GFP_KERNEL);
3091         if (!node)
3092                 return NULL;
3093
3094         /* Make sure all padding within the structure is initialized. */
3095         memset(&node->msg, 0, sizeof node->msg);
3096         node->vq = vq;
3097         node->msg.type = type;
3098         return node;
3099 }
3100 EXPORT_SYMBOL_GPL(vhost_new_msg);
3101
3102 void vhost_enqueue_msg(struct vhost_dev *dev, struct list_head *head,
3103                        struct vhost_msg_node *node)
3104 {
3105         spin_lock(&dev->iotlb_lock);
3106         list_add_tail(&node->node, head);
3107         spin_unlock(&dev->iotlb_lock);
3108
3109         wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM);
3110 }
3111 EXPORT_SYMBOL_GPL(vhost_enqueue_msg);
3112
3113 struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev,
3114                                          struct list_head *head)
3115 {
3116         struct vhost_msg_node *node = NULL;
3117
3118         spin_lock(&dev->iotlb_lock);
3119         if (!list_empty(head)) {
3120                 node = list_first_entry(head, struct vhost_msg_node,
3121                                         node);
3122                 list_del(&node->node);
3123         }
3124         spin_unlock(&dev->iotlb_lock);
3125
3126         return node;
3127 }
3128 EXPORT_SYMBOL_GPL(vhost_dequeue_msg);
3129
3130
3131 static int __init vhost_init(void)
3132 {
3133         return 0;
3134 }
3135
3136 static void __exit vhost_exit(void)
3137 {
3138 }
3139
3140 module_init(vhost_init);
3141 module_exit(vhost_exit);
3142
3143 MODULE_VERSION("0.0.1");
3144 MODULE_LICENSE("GPL v2");
3145 MODULE_AUTHOR("Michael S. Tsirkin");
3146 MODULE_DESCRIPTION("Host kernel accelerator for virtio");