drivers/xen/gntdev.c

   1 /******************************************************************************
   2  * gntdev.c
   3  *
   4  * Device for accessing (in user-space) pages that have been granted by other
   5  * domains.
   6  *
   7  * Copyright (c) 2006-2007, D G Murray.
   8  *           (c) 2009 Gerd Hoffmann <kraxel@redhat.com>
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write to the Free Software
  17  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  18  */
  19
  20 #undef DEBUG
  21
  22 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
  23
  24 #include <linux/module.h>
  25 #include <linux/kernel.h>
  26 #include <linux/init.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/fs.h>
  29 #include <linux/mm.h>
  30 #include <linux/mman.h>
  31 #include <linux/mmu_notifier.h>
  32 #include <linux/types.h>
  33 #include <linux/uaccess.h>
  34 #include <linux/sched.h>
  35 #include <linux/spinlock.h>
  36 #include <linux/slab.h>
  37 #include <linux/highmem.h>
  38
  39 #include <xen/xen.h>
  40 #include <xen/grant_table.h>
  41 #include <xen/balloon.h>
  42 #include <xen/gntdev.h>
  43 #include <xen/events.h>
  44 #include <asm/xen/hypervisor.h>
  45 #include <asm/xen/hypercall.h>
  46 #include <asm/xen/page.h>
  47
  48 MODULE_LICENSE("GPL");
  49 MODULE_AUTHOR("Derek G. Murray <Derek.Murray@cl.cam.ac.uk>, "
  50               "Gerd Hoffmann <kraxel@redhat.com>");
  51 MODULE_DESCRIPTION("User-space granted page access driver");
  52
  53 static int limit = 1024*1024;
  54 module_param(limit, int, 0644);
  55 MODULE_PARM_DESC(limit, "Maximum number of grants that may be mapped by "
  56                 "the gntdev device");
  57
  58 static atomic_t pages_mapped = ATOMIC_INIT(0);
  59
  60 static int use_ptemod;
  61 #define populate_freeable_maps use_ptemod
  62
  63 struct gntdev_priv {
  64         /* maps with visible offsets in the file descriptor */
  65         struct list_head maps;
  66         /* maps that are not visible; will be freed on munmap.
  67          * Only populated if populate_freeable_maps == 1 */
  68         struct list_head freeable_maps;
  69         /* lock protects maps and freeable_maps */
  70         spinlock_t lock;
  71         struct mm_struct *mm;
  72         struct mmu_notifier mn;
  73 };
  74
  75 struct unmap_notify {
  76         int flags;
  77         /* Address relative to the start of the grant_map */
  78         int addr;
  79         int event;
  80 };
  81
  82 struct grant_map {
  83         struct list_head next;
  84         struct vm_area_struct *vma;
  85         int index;
  86         int count;
  87         int flags;
  88         atomic_t users;
  89         struct unmap_notify notify;
  90         struct ioctl_gntdev_grant_ref *grants;
  91         struct gnttab_map_grant_ref   *map_ops;
  92         struct gnttab_unmap_grant_ref *unmap_ops;
  93         struct gnttab_map_grant_ref   *kmap_ops;
  94         struct page **pages;
  95 };
  96
  97 static int unmap_grant_pages(struct grant_map *map, int offset, int pages);
  98
  99 /* ------------------------------------------------------------------ */
 100
 101 static void gntdev_print_maps(struct gntdev_priv *priv,
 102                               char *text, int text_index)
 103 {
 104 #ifdef DEBUG
 105         struct grant_map *map;
 106
 107         pr_debug("%s: maps list (priv %p)\n", __func__, priv);
 108         list_for_each_entry(map, &priv->maps, next)
 109                 pr_debug("  index %2d, count %2d %s\n",
 110                        map->index, map->count,
 111                        map->index == text_index && text ? text : "");
 112 #endif
 113 }
 114
 115 static void gntdev_free_map(struct grant_map *map)
 116 {
 117         if (map == NULL)
 118                 return;
 119
 120         if (map->pages)
 121                 free_xenballooned_pages(map->count, map->pages);
 122         kfree(map->pages);
 123         kfree(map->grants);
 124         kfree(map->map_ops);
 125         kfree(map->unmap_ops);
 126         kfree(map->kmap_ops);
 127         kfree(map);
 128 }
 129
 130 static struct grant_map *gntdev_alloc_map(struct gntdev_priv *priv, int count)
 131 {
 132         struct grant_map *add;
 133         int i;
 134
 135         add = kzalloc(sizeof(struct grant_map), GFP_KERNEL);
 136         if (NULL == add)
 137                 return NULL;
 138
 139         add->grants    = kcalloc(count, sizeof(add->grants[0]), GFP_KERNEL);
 140         add->map_ops   = kcalloc(count, sizeof(add->map_ops[0]), GFP_KERNEL);
 141         add->unmap_ops = kcalloc(count, sizeof(add->unmap_ops[0]), GFP_KERNEL);
 142         add->kmap_ops  = kcalloc(count, sizeof(add->kmap_ops[0]), GFP_KERNEL);
 143         add->pages     = kcalloc(count, sizeof(add->pages[0]), GFP_KERNEL);
 144         if (NULL == add->grants    ||
 145             NULL == add->map_ops   ||
 146             NULL == add->unmap_ops ||
 147             NULL == add->kmap_ops  ||
 148             NULL == add->pages)
 149                 goto err;
 150
 151         if (alloc_xenballooned_pages(count, add->pages, false /* lowmem */))
 152                 goto err;
 153
 154         for (i = 0; i < count; i++) {
 155                 add->map_ops[i].handle = -1;
 156                 add->unmap_ops[i].handle = -1;
 157                 add->kmap_ops[i].handle = -1;
 158         }
 159
 160         add->index = 0;
 161         add->count = count;
 162         atomic_set(&add->users, 1);
 163
 164         return add;
 165
 166 err:
 167         gntdev_free_map(add);
 168         return NULL;
 169 }
 170
 171 static void gntdev_add_map(struct gntdev_priv *priv, struct grant_map *add)
 172 {
 173         struct grant_map *map;
 174
 175         list_for_each_entry(map, &priv->maps, next) {
 176                 if (add->index + add->count < map->index) {
 177                         list_add_tail(&add->next, &map->next);
 178                         goto done;
 179                 }
 180                 add->index = map->index + map->count;
 181         }
 182         list_add_tail(&add->next, &priv->maps);
 183
 184 done:
 185         gntdev_print_maps(priv, "[new]", add->index);
 186 }
 187
 188 static struct grant_map *gntdev_find_map_index(struct gntdev_priv *priv,
 189                 int index, int count)
 190 {
 191         struct grant_map *map;
 192
 193         list_for_each_entry(map, &priv->maps, next) {
 194                 if (map->index != index)
 195                         continue;
 196                 if (count && map->count != count)
 197                         continue;
 198                 return map;
 199         }
 200         return NULL;
 201 }
 202
 203 static void gntdev_put_map(struct gntdev_priv *priv, struct grant_map *map)
 204 {
 205         if (!map)
 206                 return;
 207
 208         if (!atomic_dec_and_test(&map->users))
 209                 return;
 210
 211         atomic_sub(map->count, &pages_mapped);
 212
 213         if (map->notify.flags & UNMAP_NOTIFY_SEND_EVENT) {
 214                 notify_remote_via_evtchn(map->notify.event);
 215                 evtchn_put(map->notify.event);
 216         }
 217
 218         if (populate_freeable_maps && priv) {
 219                 spin_lock(&priv->lock);
 220                 list_del(&map->next);
 221                 spin_unlock(&priv->lock);
 222         }
 223
 224         if (map->pages && !use_ptemod)
 225                 unmap_grant_pages(map, 0, map->count);
 226         gntdev_free_map(map);
 227 }
 228
 229 /* ------------------------------------------------------------------ */
 230
 231 static int find_grant_ptes(pte_t *pte, pgtable_t token,
 232                 unsigned long addr, void *data)
 233 {
 234         struct grant_map *map = data;
 235         unsigned int pgnr = (addr - map->vma->vm_start) >> PAGE_SHIFT;
 236         int flags = map->flags | GNTMAP_application_map | GNTMAP_contains_pte;
 237         u64 pte_maddr;
 238
 239         BUG_ON(pgnr >= map->count);
 240         pte_maddr = arbitrary_virt_to_machine(pte).maddr;
 241
 242         gnttab_set_map_op(&map->map_ops[pgnr], pte_maddr, flags,
 243                           map->grants[pgnr].ref,
 244                           map->grants[pgnr].domid);
 245         gnttab_set_unmap_op(&map->unmap_ops[pgnr], pte_maddr, flags,
 246                             -1 /* handle */);
 247         return 0;
 248 }
 249
 250 static int map_grant_pages(struct grant_map *map)
 251 {
 252         int i, err = 0;
 253
 254         if (!use_ptemod) {
 255                 /* Note: it could already be mapped */
 256                 if (map->map_ops[0].handle != -1)
 257                         return 0;
 258                 for (i = 0; i < map->count; i++) {
 259                         unsigned long addr = (unsigned long)
 260                                 pfn_to_kaddr(page_to_pfn(map->pages[i]));
 261                         gnttab_set_map_op(&map->map_ops[i], addr, map->flags,
 262                                 map->grants[i].ref,
 263                                 map->grants[i].domid);
 264                         gnttab_set_unmap_op(&map->unmap_ops[i], addr,
 265                                 map->flags, -1 /* handle */);
 266                 }
 267         } else {
 268                 /*
 269                  * Setup the map_ops corresponding to the pte entries pointing
 270                  * to the kernel linear addresses of the struct pages.
 271                  * These ptes are completely different from the user ptes dealt
 272                  * with find_grant_ptes.
 273                  */
 274                 for (i = 0; i < map->count; i++) {
 275                         unsigned long address = (unsigned long)
 276                                 pfn_to_kaddr(page_to_pfn(map->pages[i]));
 277                         BUG_ON(PageHighMem(map->pages[i]));
 278
 279                         gnttab_set_map_op(&map->kmap_ops[i], address,
 280                                 map->flags | GNTMAP_host_map,
 281                                 map->grants[i].ref,
 282                                 map->grants[i].domid);
 283                 }
 284         }
 285
 286         pr_debug("map %d+%d\n", map->index, map->count);
 287         err = gnttab_map_refs_userspace(map->map_ops,
 288                                         use_ptemod ? map->kmap_ops : NULL,
 289                                         map->pages,
 290                                         map->count);
 291         if (err)
 292                 return err;
 293
 294         for (i = 0; i < map->count; i++) {
 295                 if (map->map_ops[i].status)
 296                         err = -EINVAL;
 297                 else {
 298                         BUG_ON(map->map_ops[i].handle == -1);
 299                         map->unmap_ops[i].handle = map->map_ops[i].handle;
 300                         pr_debug("map handle=%d\n", map->map_ops[i].handle);
 301                 }
 302         }
 303         return err;
 304 }
 305
 306 static int __unmap_grant_pages(struct grant_map *map, int offset, int pages)
 307 {
 308         int i, err = 0;
 309
 310         if (map->notify.flags & UNMAP_NOTIFY_CLEAR_BYTE) {
 311                 int pgno = (map->notify.addr >> PAGE_SHIFT);
 312                 if (pgno >= offset && pgno < offset + pages) {
 313                         /* No need for kmap, pages are in lowmem */
 314                         uint8_t *tmp = pfn_to_kaddr(page_to_pfn(map->pages[pgno]));
 315                         tmp[map->notify.addr & (PAGE_SIZE-1)] = 0;
 316                         map->notify.flags &= ~UNMAP_NOTIFY_CLEAR_BYTE;
 317                 }
 318         }
 319
 320         err = gnttab_unmap_refs_userspace(map->unmap_ops + offset,
 321                                           use_ptemod ? map->kmap_ops + offset : NULL,
 322                                           map->pages + offset,
 323                                           pages);
 324         if (err)
 325                 return err;
 326
 327         for (i = 0; i < pages; i++) {
 328                 if (map->unmap_ops[offset+i].status)
 329                         err = -EINVAL;
 330                 pr_debug("unmap handle=%d st=%d\n",
 331                         map->unmap_ops[offset+i].handle,
 332                         map->unmap_ops[offset+i].status);
 333                 map->unmap_ops[offset+i].handle = -1;
 334         }
 335         return err;
 336 }
 337
 338 static int unmap_grant_pages(struct grant_map *map, int offset, int pages)
 339 {
 340         int range, err = 0;
 341
 342         pr_debug("unmap %d+%d [%d+%d]\n", map->index, map->count, offset, pages);
 343
 344         /* It is possible the requested range will have a "hole" where we
 345          * already unmapped some of the grants. Only unmap valid ranges.
 346          */
 347         while (pages && !err) {
 348                 while (pages && map->unmap_ops[offset].handle == -1) {
 349                         offset++;
 350                         pages--;
 351                 }
 352                 range = 0;
 353                 while (range < pages) {
 354                         if (map->unmap_ops[offset+range].handle == -1) {
 355                                 range--;
 356                                 break;
 357                         }
 358                         range++;
 359                 }
 360                 err = __unmap_grant_pages(map, offset, range);
 361                 offset += range;
 362                 pages -= range;
 363         }
 364
 365         return err;
 366 }
 367
 368 /* ------------------------------------------------------------------ */
 369
 370 static void gntdev_vma_open(struct vm_area_struct *vma)
 371 {
 372         struct grant_map *map = vma->vm_private_data;
 373
 374         pr_debug("gntdev_vma_open %p\n", vma);
 375         atomic_inc(&map->users);
 376 }
 377
 378 static void gntdev_vma_close(struct vm_area_struct *vma)
 379 {
 380         struct grant_map *map = vma->vm_private_data;
 381         struct file *file = vma->vm_file;
 382         struct gntdev_priv *priv = file->private_data;
 383
 384         pr_debug("gntdev_vma_close %p\n", vma);
 385         if (use_ptemod) {
 386                 /* It is possible that an mmu notifier could be running
 387                  * concurrently, so take priv->lock to ensure that the vma won't
 388                  * vanishing during the unmap_grant_pages call, since we will
 389                  * spin here until that completes. Such a concurrent call will
 390                  * not do any unmapping, since that has been done prior to
 391                  * closing the vma, but it may still iterate the unmap_ops list.
 392                  */
 393                 spin_lock(&priv->lock);
 394                 map->vma = NULL;
 395                 spin_unlock(&priv->lock);
 396         }
 397         vma->vm_private_data = NULL;
 398         gntdev_put_map(priv, map);
 399 }
 400
 401 static struct vm_operations_struct gntdev_vmops = {
 402         .open = gntdev_vma_open,
 403         .close = gntdev_vma_close,
 404 };
 405
 406 /* ------------------------------------------------------------------ */
 407
 408 static void unmap_if_in_range(struct grant_map *map,
 409                               unsigned long start, unsigned long end)
 410 {
 411         unsigned long mstart, mend;
 412         int err;
 413
 414         if (!map->vma)
 415                 return;
 416         if (map->vma->vm_start >= end)
 417                 return;
 418         if (map->vma->vm_end <= start)
 419                 return;
 420         mstart = max(start, map->vma->vm_start);
 421         mend   = min(end,   map->vma->vm_end);
 422         pr_debug("map %d+%d (%lx %lx), range %lx %lx, mrange %lx %lx\n",
 423                         map->index, map->count,
 424                         map->vma->vm_start, map->vma->vm_end,
 425                         start, end, mstart, mend);
 426         err = unmap_grant_pages(map,
 427                                 (mstart - map->vma->vm_start) >> PAGE_SHIFT,
 428                                 (mend - mstart) >> PAGE_SHIFT);
 429         WARN_ON(err);
 430 }
 431
 432 static void mn_invl_range_start(struct mmu_notifier *mn,
 433                                 struct mm_struct *mm,
 434                                 unsigned long start, unsigned long end)
 435 {
 436         struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
 437         struct grant_map *map;
 438
 439         spin_lock(&priv->lock);
 440         list_for_each_entry(map, &priv->maps, next) {
 441                 unmap_if_in_range(map, start, end);
 442         }
 443         list_for_each_entry(map, &priv->freeable_maps, next) {
 444                 unmap_if_in_range(map, start, end);
 445         }
 446         spin_unlock(&priv->lock);
 447 }
 448
 449 static void mn_invl_page(struct mmu_notifier *mn,
 450                          struct mm_struct *mm,
 451                          unsigned long address)
 452 {
 453         mn_invl_range_start(mn, mm, address, address + PAGE_SIZE);
 454 }
 455
 456 static void mn_release(struct mmu_notifier *mn,
 457                        struct mm_struct *mm)
 458 {
 459         struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
 460         struct grant_map *map;
 461         int err;
 462
 463         spin_lock(&priv->lock);
 464         list_for_each_entry(map, &priv->maps, next) {
 465                 if (!map->vma)
 466                         continue;
 467                 pr_debug("map %d+%d (%lx %lx)\n",
 468                                 map->index, map->count,
 469                                 map->vma->vm_start, map->vma->vm_end);
 470                 err = unmap_grant_pages(map, /* offset */ 0, map->count);
 471                 WARN_ON(err);
 472         }
 473         list_for_each_entry(map, &priv->freeable_maps, next) {
 474                 if (!map->vma)
 475                         continue;
 476                 pr_debug("map %d+%d (%lx %lx)\n",
 477                                 map->index, map->count,
 478                                 map->vma->vm_start, map->vma->vm_end);
 479                 err = unmap_grant_pages(map, /* offset */ 0, map->count);
 480                 WARN_ON(err);
 481         }
 482         spin_unlock(&priv->lock);
 483 }
 484
 485 static struct mmu_notifier_ops gntdev_mmu_ops = {
 486         .release                = mn_release,
 487         .invalidate_page        = mn_invl_page,
 488         .invalidate_range_start = mn_invl_range_start,
 489 };
 490
 491 /* ------------------------------------------------------------------ */
 492
 493 static int gntdev_open(struct inode *inode, struct file *flip)
 494 {
 495         struct gntdev_priv *priv;
 496         int ret = 0;
 497
 498         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 499         if (!priv)
 500                 return -ENOMEM;
 501
 502         INIT_LIST_HEAD(&priv->maps);
 503         INIT_LIST_HEAD(&priv->freeable_maps);
 504         spin_lock_init(&priv->lock);
 505
 506         if (use_ptemod) {
 507                 priv->mm = get_task_mm(current);
 508                 if (!priv->mm) {
 509                         kfree(priv);
 510                         return -ENOMEM;
 511                 }
 512                 priv->mn.ops = &gntdev_mmu_ops;
 513                 ret = mmu_notifier_register(&priv->mn, priv->mm);
 514                 mmput(priv->mm);
 515         }
 516
 517         if (ret) {
 518                 kfree(priv);
 519                 return ret;
 520         }
 521
 522         flip->private_data = priv;
 523         pr_debug("priv %p\n", priv);
 524
 525         return 0;
 526 }
 527
 528 static int gntdev_release(struct inode *inode, struct file *flip)
 529 {
 530         struct gntdev_priv *priv = flip->private_data;
 531         struct grant_map *map;
 532
 533         pr_debug("priv %p\n", priv);
 534
 535         while (!list_empty(&priv->maps)) {
 536                 map = list_entry(priv->maps.next, struct grant_map, next);
 537                 list_del(&map->next);
 538                 gntdev_put_map(NULL /* already removed */, map);
 539         }
 540         WARN_ON(!list_empty(&priv->freeable_maps));
 541
 542         if (use_ptemod)
 543                 mmu_notifier_unregister(&priv->mn, priv->mm);
 544         kfree(priv);
 545         return 0;
 546 }
 547
 548 static long gntdev_ioctl_map_grant_ref(struct gntdev_priv *priv,
 549                                        struct ioctl_gntdev_map_grant_ref __user *u)
 550 {
 551         struct ioctl_gntdev_map_grant_ref op;
 552         struct grant_map *map;
 553         int err;
 554
 555         if (copy_from_user(&op, u, sizeof(op)) != 0)
 556                 return -EFAULT;
 557         pr_debug("priv %p, add %d\n", priv, op.count);
 558         if (unlikely(op.count <= 0))
 559                 return -EINVAL;
 560
 561         err = -ENOMEM;
 562         map = gntdev_alloc_map(priv, op.count);
 563         if (!map)
 564                 return err;
 565
 566         if (unlikely(atomic_add_return(op.count, &pages_mapped) > limit)) {
 567                 pr_debug("can't map: over limit\n");
 568                 gntdev_put_map(NULL, map);
 569                 return err;
 570         }
 571
 572         if (copy_from_user(map->grants, &u->refs,
 573                            sizeof(map->grants[0]) * op.count) != 0) {
 574                 gntdev_put_map(NULL, map);
 575                 return -EFAULT;
 576         }
 577
 578         spin_lock(&priv->lock);
 579         gntdev_add_map(priv, map);
 580         op.index = map->index << PAGE_SHIFT;
 581         spin_unlock(&priv->lock);
 582
 583         if (copy_to_user(u, &op, sizeof(op)) != 0)
 584                 return -EFAULT;
 585
 586         return 0;
 587 }
 588
 589 static long gntdev_ioctl_unmap_grant_ref(struct gntdev_priv *priv,
 590                                          struct ioctl_gntdev_unmap_grant_ref __user *u)
 591 {
 592         struct ioctl_gntdev_unmap_grant_ref op;
 593         struct grant_map *map;
 594         int err = -ENOENT;
 595
 596         if (copy_from_user(&op, u, sizeof(op)) != 0)
 597                 return -EFAULT;
 598         pr_debug("priv %p, del %d+%d\n", priv, (int)op.index, (int)op.count);
 599
 600         spin_lock(&priv->lock);
 601         map = gntdev_find_map_index(priv, op.index >> PAGE_SHIFT, op.count);
 602         if (map) {
 603                 list_del(&map->next);
 604                 if (populate_freeable_maps)
 605                         list_add_tail(&map->next, &priv->freeable_maps);
 606                 err = 0;
 607         }
 608         spin_unlock(&priv->lock);
 609         if (map)
 610                 gntdev_put_map(priv, map);
 611         return err;
 612 }
 613
 614 static long gntdev_ioctl_get_offset_for_vaddr(struct gntdev_priv *priv,
 615                                               struct ioctl_gntdev_get_offset_for_vaddr __user *u)
 616 {
 617         struct ioctl_gntdev_get_offset_for_vaddr op;
 618         struct vm_area_struct *vma;
 619         struct grant_map *map;
 620         int rv = -EINVAL;
 621
 622         if (copy_from_user(&op, u, sizeof(op)) != 0)
 623                 return -EFAULT;
 624         pr_debug("priv %p, offset for vaddr %lx\n", priv, (unsigned long)op.vaddr);
 625
 626         down_read(&current->mm->mmap_sem);
 627         vma = find_vma(current->mm, op.vaddr);
 628         if (!vma || vma->vm_ops != &gntdev_vmops)
 629                 goto out_unlock;
 630
 631         map = vma->vm_private_data;
 632         if (!map)
 633                 goto out_unlock;
 634
 635         op.offset = map->index << PAGE_SHIFT;
 636         op.count = map->count;
 637         rv = 0;
 638
 639  out_unlock:
 640         up_read(&current->mm->mmap_sem);
 641
 642         if (rv == 0 && copy_to_user(u, &op, sizeof(op)) != 0)
 643                 return -EFAULT;
 644         return rv;
 645 }
 646
 647 static long gntdev_ioctl_notify(struct gntdev_priv *priv, void __user *u)
 648 {
 649         struct ioctl_gntdev_unmap_notify op;
 650         struct grant_map *map;
 651         int rc;
 652         int out_flags;
 653         unsigned int out_event;
 654
 655         if (copy_from_user(&op, u, sizeof(op)))
 656                 return -EFAULT;
 657
 658         if (op.action & ~(UNMAP_NOTIFY_CLEAR_BYTE|UNMAP_NOTIFY_SEND_EVENT))
 659                 return -EINVAL;
 660
 661         /* We need to grab a reference to the event channel we are going to use
 662          * to send the notify before releasing the reference we may already have
 663          * (if someone has called this ioctl twice). This is required so that
 664          * it is possible to change the clear_byte part of the notification
 665          * without disturbing the event channel part, which may now be the last
 666          * reference to that event channel.
 667          */
 668         if (op.action & UNMAP_NOTIFY_SEND_EVENT) {
 669                 if (evtchn_get(op.event_channel_port))
 670                         return -EINVAL;
 671         }
 672
 673         out_flags = op.action;
 674         out_event = op.event_channel_port;
 675
 676         spin_lock(&priv->lock);
 677
 678         list_for_each_entry(map, &priv->maps, next) {
 679                 uint64_t begin = map->index << PAGE_SHIFT;
 680                 uint64_t end = (map->index + map->count) << PAGE_SHIFT;
 681                 if (op.index >= begin && op.index < end)
 682                         goto found;
 683         }
 684         rc = -ENOENT;
 685         goto unlock_out;
 686
 687  found:
 688         if ((op.action & UNMAP_NOTIFY_CLEAR_BYTE) &&
 689                         (map->flags & GNTMAP_readonly)) {
 690                 rc = -EINVAL;
 691                 goto unlock_out;
 692         }
 693
 694         out_flags = map->notify.flags;
 695         out_event = map->notify.event;
 696
 697         map->notify.flags = op.action;
 698         map->notify.addr = op.index - (map->index << PAGE_SHIFT);
 699         map->notify.event = op.event_channel_port;
 700
 701         rc = 0;
 702
 703  unlock_out:
 704         spin_unlock(&priv->lock);
 705
 706         /* Drop the reference to the event channel we did not save in the map */
 707         if (out_flags & UNMAP_NOTIFY_SEND_EVENT)
 708                 evtchn_put(out_event);
 709
 710         return rc;
 711 }
 712
 713 static long gntdev_ioctl(struct file *flip,
 714                          unsigned int cmd, unsigned long arg)
 715 {
 716         struct gntdev_priv *priv = flip->private_data;
 717         void __user *ptr = (void __user *)arg;
 718
 719         switch (cmd) {
 720         case IOCTL_GNTDEV_MAP_GRANT_REF:
 721                 return gntdev_ioctl_map_grant_ref(priv, ptr);
 722
 723         case IOCTL_GNTDEV_UNMAP_GRANT_REF:
 724                 return gntdev_ioctl_unmap_grant_ref(priv, ptr);
 725
 726         case IOCTL_GNTDEV_GET_OFFSET_FOR_VADDR:
 727                 return gntdev_ioctl_get_offset_for_vaddr(priv, ptr);
 728
 729         case IOCTL_GNTDEV_SET_UNMAP_NOTIFY:
 730                 return gntdev_ioctl_notify(priv, ptr);
 731
 732         default:
 733                 pr_debug("priv %p, unknown cmd %x\n", priv, cmd);
 734                 return -ENOIOCTLCMD;
 735         }
 736
 737         return 0;
 738 }
 739
 740 static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma)
 741 {
 742         struct gntdev_priv *priv = flip->private_data;
 743         int index = vma->vm_pgoff;
 744         int count = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
 745         struct grant_map *map;
 746         int i, err = -EINVAL;
 747
 748         if ((vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_SHARED))
 749                 return -EINVAL;
 750
 751         pr_debug("map %d+%d at %lx (pgoff %lx)\n",
 752                         index, count, vma->vm_start, vma->vm_pgoff);
 753
 754         spin_lock(&priv->lock);
 755         map = gntdev_find_map_index(priv, index, count);
 756         if (!map)
 757                 goto unlock_out;
 758         if (use_ptemod && map->vma)
 759                 goto unlock_out;
 760         if (use_ptemod && priv->mm != vma->vm_mm) {
 761                 pr_warn("Huh? Other mm?\n");
 762                 goto unlock_out;
 763         }
 764
 765         atomic_inc(&map->users);
 766
 767         vma->vm_ops = &gntdev_vmops;
 768
 769         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
 770
 771         if (use_ptemod)
 772                 vma->vm_flags |= VM_DONTCOPY;
 773
 774         vma->vm_private_data = map;
 775
 776         if (use_ptemod)
 777                 map->vma = vma;
 778
 779         if (map->flags) {
 780                 if ((vma->vm_flags & VM_WRITE) &&
 781                                 (map->flags & GNTMAP_readonly))
 782                         goto out_unlock_put;
 783         } else {
 784                 map->flags = GNTMAP_host_map;
 785                 if (!(vma->vm_flags & VM_WRITE))
 786                         map->flags |= GNTMAP_readonly;
 787         }
 788
 789         spin_unlock(&priv->lock);
 790
 791         if (use_ptemod) {
 792                 err = apply_to_page_range(vma->vm_mm, vma->vm_start,
 793                                           vma->vm_end - vma->vm_start,
 794                                           find_grant_ptes, map);
 795                 if (err) {
 796                         pr_warn("find_grant_ptes() failure.\n");
 797                         goto out_put_map;
 798                 }
 799         }
 800
 801         err = map_grant_pages(map);
 802         if (err)
 803                 goto out_put_map;
 804
 805         if (!use_ptemod) {
 806                 for (i = 0; i < count; i++) {
 807                         err = vm_insert_page(vma, vma->vm_start + i*PAGE_SIZE,
 808                                 map->pages[i]);
 809                         if (err)
 810                                 goto out_put_map;
 811                 }
 812         }
 813
 814         return 0;
 815
 816 unlock_out:
 817         spin_unlock(&priv->lock);
 818         return err;
 819
 820 out_unlock_put:
 821         spin_unlock(&priv->lock);
 822 out_put_map:
 823         if (use_ptemod)
 824                 map->vma = NULL;
 825         gntdev_put_map(priv, map);
 826         return err;
 827 }
 828
 829 static const struct file_operations gntdev_fops = {
 830         .owner = THIS_MODULE,
 831         .open = gntdev_open,
 832         .release = gntdev_release,
 833         .mmap = gntdev_mmap,
 834         .unlocked_ioctl = gntdev_ioctl
 835 };
 836
 837 static struct miscdevice gntdev_miscdev = {
 838         .minor        = MISC_DYNAMIC_MINOR,
 839         .name         = "xen/gntdev",
 840         .fops         = &gntdev_fops,
 841 };
 842
 843 /* ------------------------------------------------------------------ */
 844
 845 static int __init gntdev_init(void)
 846 {
 847         int err;
 848
 849         if (!xen_domain())
 850                 return -ENODEV;
 851
 852         use_ptemod = !xen_feature(XENFEAT_auto_translated_physmap);
 853
 854         err = misc_register(&gntdev_miscdev);
 855         if (err != 0) {
 856                 pr_err("Could not register gntdev device\n");
 857                 return err;
 858         }
 859         return 0;
 860 }
 861
 862 static void __exit gntdev_exit(void)
 863 {
 864         misc_deregister(&gntdev_miscdev);
 865 }
 866
 867 module_init(gntdev_init);
 868 module_exit(gntdev_exit);
 869
 870 /* ------------------------------------------------------------------ */