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 #include <linux/module.h>
  23 #include <linux/kernel.h>
  24 #include <linux/init.h>
  25 #include <linux/miscdevice.h>
  26 #include <linux/fs.h>
  27 #include <linux/mm.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/types.h>
  31 #include <linux/uaccess.h>
  32 #include <linux/sched.h>
  33 #include <linux/spinlock.h>
  34 #include <linux/slab.h>
  35 #include <linux/highmem.h>
  36
  37 #include <xen/xen.h>
  38 #include <xen/grant_table.h>
  39 #include <xen/balloon.h>
  40 #include <xen/gntdev.h>
  41 #include <xen/events.h>
  42 #include <asm/xen/hypervisor.h>
  43 #include <asm/xen/hypercall.h>
  44 #include <asm/xen/page.h>
  45
  46 MODULE_LICENSE("GPL");
  47 MODULE_AUTHOR("Derek G. Murray <Derek.Murray@cl.cam.ac.uk>, "
  48               "Gerd Hoffmann <kraxel@redhat.com>");
  49 MODULE_DESCRIPTION("User-space granted page access driver");
  50
  51 static int limit = 1024*1024;
  52 module_param(limit, int, 0644);
  53 MODULE_PARM_DESC(limit, "Maximum number of grants that may be mapped by "
  54                 "the gntdev device");
  55
  56 static atomic_t pages_mapped = ATOMIC_INIT(0);
  57
  58 static int use_ptemod;
  59 #define populate_freeable_maps use_ptemod
  60
  61 struct gntdev_priv {
  62         /* maps with visible offsets in the file descriptor */
  63         struct list_head maps;
  64         /* maps that are not visible; will be freed on munmap.
  65          * Only populated if populate_freeable_maps == 1 */
  66         struct list_head freeable_maps;
  67         /* lock protects maps and freeable_maps */
  68         spinlock_t lock;
  69         struct mm_struct *mm;
  70         struct mmu_notifier mn;
  71 };
  72
  73 struct unmap_notify {
  74         int flags;
  75         /* Address relative to the start of the grant_map */
  76         int addr;
  77         int event;
  78 };
  79
  80 struct grant_map {
  81         struct list_head next;
  82         struct vm_area_struct *vma;
  83         int index;
  84         int count;
  85         int flags;
  86         atomic_t users;
  87         struct unmap_notify notify;
  88         struct ioctl_gntdev_grant_ref *grants;
  89         struct gnttab_map_grant_ref   *map_ops;
  90         struct gnttab_unmap_grant_ref *unmap_ops;
  91         struct gnttab_map_grant_ref   *kmap_ops;
  92         struct page **pages;
  93 };
  94
  95 static int unmap_grant_pages(struct grant_map *map, int offset, int pages);
  96
  97 /* ------------------------------------------------------------------ */
  98
  99 static void gntdev_print_maps(struct gntdev_priv *priv,
 100                               char *text, int text_index)
 101 {
 102 #ifdef DEBUG
 103         struct grant_map *map;
 104
 105         pr_debug("%s: maps list (priv %p)\n", __func__, priv);
 106         list_for_each_entry(map, &priv->maps, next)
 107                 pr_debug("  index %2d, count %2d %s\n",
 108                        map->index, map->count,
 109                        map->index == text_index && text ? text : "");
 110 #endif
 111 }
 112
 113 static void gntdev_free_map(struct grant_map *map)
 114 {
 115         if (map == NULL)
 116                 return;
 117
 118         if (map->pages)
 119                 free_xenballooned_pages(map->count, map->pages);
 120         kfree(map->pages);
 121         kfree(map->grants);
 122         kfree(map->map_ops);
 123         kfree(map->unmap_ops);
 124         kfree(map->kmap_ops);
 125         kfree(map);
 126 }
 127
 128 static struct grant_map *gntdev_alloc_map(struct gntdev_priv *priv, int count)
 129 {
 130         struct grant_map *add;
 131         int i;
 132
 133         add = kzalloc(sizeof(struct grant_map), GFP_KERNEL);
 134         if (NULL == add)
 135                 return NULL;
 136
 137         add->grants    = kcalloc(count, sizeof(add->grants[0]), GFP_KERNEL);
 138         add->map_ops   = kcalloc(count, sizeof(add->map_ops[0]), GFP_KERNEL);
 139         add->unmap_ops = kcalloc(count, sizeof(add->unmap_ops[0]), GFP_KERNEL);
 140         add->kmap_ops  = kcalloc(count, sizeof(add->kmap_ops[0]), GFP_KERNEL);
 141         add->pages     = kcalloc(count, sizeof(add->pages[0]), GFP_KERNEL);
 142         if (NULL == add->grants    ||
 143             NULL == add->map_ops   ||
 144             NULL == add->unmap_ops ||
 145             NULL == add->kmap_ops  ||
 146             NULL == add->pages)
 147                 goto err;
 148
 149         if (alloc_xenballooned_pages(count, add->pages, false /* lowmem */))
 150                 goto err;
 151
 152         for (i = 0; i < count; i++) {
 153                 add->map_ops[i].handle = -1;
 154                 add->unmap_ops[i].handle = -1;
 155                 add->kmap_ops[i].handle = -1;
 156         }
 157
 158         add->index = 0;
 159         add->count = count;
 160         atomic_set(&add->users, 1);
 161
 162         return add;
 163
 164 err:
 165         gntdev_free_map(add);
 166         return NULL;
 167 }
 168
 169 static void gntdev_add_map(struct gntdev_priv *priv, struct grant_map *add)
 170 {
 171         struct grant_map *map;
 172
 173         list_for_each_entry(map, &priv->maps, next) {
 174                 if (add->index + add->count < map->index) {
 175                         list_add_tail(&add->next, &map->next);
 176                         goto done;
 177                 }
 178                 add->index = map->index + map->count;
 179         }
 180         list_add_tail(&add->next, &priv->maps);
 181
 182 done:
 183         gntdev_print_maps(priv, "[new]", add->index);
 184 }
 185
 186 static struct grant_map *gntdev_find_map_index(struct gntdev_priv *priv,
 187                 int index, int count)
 188 {
 189         struct grant_map *map;
 190
 191         list_for_each_entry(map, &priv->maps, next) {
 192                 if (map->index != index)
 193                         continue;
 194                 if (count && map->count != count)
 195                         continue;
 196                 return map;
 197         }
 198         return NULL;
 199 }
 200
 201 static void gntdev_put_map(struct gntdev_priv *priv, struct grant_map *map)
 202 {
 203         if (!map)
 204                 return;
 205
 206         if (!atomic_dec_and_test(&map->users))
 207                 return;
 208
 209         atomic_sub(map->count, &pages_mapped);
 210
 211         if (map->notify.flags & UNMAP_NOTIFY_SEND_EVENT) {
 212                 notify_remote_via_evtchn(map->notify.event);
 213                 evtchn_put(map->notify.event);
 214         }
 215
 216         if (populate_freeable_maps && priv) {
 217                 spin_lock(&priv->lock);
 218                 list_del(&map->next);
 219                 spin_unlock(&priv->lock);
 220         }
 221
 222         if (map->pages && !use_ptemod)
 223                 unmap_grant_pages(map, 0, map->count);
 224         gntdev_free_map(map);
 225 }
 226
 227 /* ------------------------------------------------------------------ */
 228
 229 static int find_grant_ptes(pte_t *pte, pgtable_t token,
 230                 unsigned long addr, void *data)
 231 {
 232         struct grant_map *map = data;
 233         unsigned int pgnr = (addr - map->vma->vm_start) >> PAGE_SHIFT;
 234         int flags = map->flags | GNTMAP_application_map | GNTMAP_contains_pte;
 235         u64 pte_maddr;
 236
 237         BUG_ON(pgnr >= map->count);
 238         pte_maddr = arbitrary_virt_to_machine(pte).maddr;
 239
 240         gnttab_set_map_op(&map->map_ops[pgnr], pte_maddr, flags,
 241                           map->grants[pgnr].ref,
 242                           map->grants[pgnr].domid);
 243         gnttab_set_unmap_op(&map->unmap_ops[pgnr], pte_maddr, flags,
 244                             -1 /* handle */);
 245         return 0;
 246 }
 247
 248 static int map_grant_pages(struct grant_map *map)
 249 {
 250         int i, err = 0;
 251
 252         if (!use_ptemod) {
 253                 /* Note: it could already be mapped */
 254                 if (map->map_ops[0].handle != -1)
 255                         return 0;
 256                 for (i = 0; i < map->count; i++) {
 257                         unsigned long addr = (unsigned long)
 258                                 pfn_to_kaddr(page_to_pfn(map->pages[i]));
 259                         gnttab_set_map_op(&map->map_ops[i], addr, map->flags,
 260                                 map->grants[i].ref,
 261                                 map->grants[i].domid);
 262                         gnttab_set_unmap_op(&map->unmap_ops[i], addr,
 263                                 map->flags, -1 /* handle */);
 264                 }
 265         } else {
 266                 /*
 267                  * Setup the map_ops corresponding to the pte entries pointing
 268                  * to the kernel linear addresses of the struct pages.
 269                  * These ptes are completely different from the user ptes dealt
 270                  * with find_grant_ptes.
 271                  */
 272                 for (i = 0; i < map->count; i++) {
 273                         unsigned level;
 274                         unsigned long address = (unsigned long)
 275                                 pfn_to_kaddr(page_to_pfn(map->pages[i]));
 276                         pte_t *ptep;
 277                         u64 pte_maddr = 0;
 278                         BUG_ON(PageHighMem(map->pages[i]));
 279
 280                         ptep = lookup_address(address, &level);
 281                         pte_maddr = arbitrary_virt_to_machine(ptep).maddr;
 282                         gnttab_set_map_op(&map->kmap_ops[i], pte_maddr,
 283                                 map->flags |
 284                                 GNTMAP_host_map |
 285                                 GNTMAP_contains_pte,
 286                                 map->grants[i].ref,
 287                                 map->grants[i].domid);
 288                 }
 289         }
 290
 291         pr_debug("map %d+%d\n", map->index, map->count);
 292         err = gnttab_map_refs(map->map_ops, use_ptemod ? map->kmap_ops : NULL,
 293                         map->pages, map->count);
 294         if (err)
 295                 return err;
 296
 297         for (i = 0; i < map->count; i++) {
 298                 if (map->map_ops[i].status)
 299                         err = -EINVAL;
 300                 else {
 301                         BUG_ON(map->map_ops[i].handle == -1);
 302                         map->unmap_ops[i].handle = map->map_ops[i].handle;
 303                         pr_debug("map handle=%d\n", map->map_ops[i].handle);
 304                 }
 305         }
 306         return err;
 307 }
 308
 309 static int __unmap_grant_pages(struct grant_map *map, int offset, int pages)
 310 {
 311         int i, err = 0;
 312
 313         if (map->notify.flags & UNMAP_NOTIFY_CLEAR_BYTE) {
 314                 int pgno = (map->notify.addr >> PAGE_SHIFT);
 315                 if (pgno >= offset && pgno < offset + pages) {
 316                         /* No need for kmap, pages are in lowmem */
 317                         uint8_t *tmp = pfn_to_kaddr(page_to_pfn(map->pages[pgno]));
 318                         tmp[map->notify.addr & (PAGE_SIZE-1)] = 0;
 319                         map->notify.flags &= ~UNMAP_NOTIFY_CLEAR_BYTE;
 320                 }
 321         }
 322
 323         err = gnttab_unmap_refs(map->unmap_ops + offset,
 324                         use_ptemod ? map->kmap_ops + offset : NULL, map->pages + offset,
 325                         pages);
 326         if (err)
 327                 return err;
 328
 329         for (i = 0; i < pages; i++) {
 330                 if (map->unmap_ops[offset+i].status)
 331                         err = -EINVAL;
 332                 pr_debug("unmap handle=%d st=%d\n",
 333                         map->unmap_ops[offset+i].handle,
 334                         map->unmap_ops[offset+i].status);
 335                 map->unmap_ops[offset+i].handle = -1;
 336         }
 337         return err;
 338 }
 339
 340 static int unmap_grant_pages(struct grant_map *map, int offset, int pages)
 341 {
 342         int range, err = 0;
 343
 344         pr_debug("unmap %d+%d [%d+%d]\n", map->index, map->count, offset, pages);
 345
 346         /* It is possible the requested range will have a "hole" where we
 347          * already unmapped some of the grants. Only unmap valid ranges.
 348          */
 349         while (pages && !err) {
 350                 while (pages && map->unmap_ops[offset].handle == -1) {
 351                         offset++;
 352                         pages--;
 353                 }
 354                 range = 0;
 355                 while (range < pages) {
 356                         if (map->unmap_ops[offset+range].handle == -1) {
 357                                 range--;
 358                                 break;
 359                         }
 360                         range++;
 361                 }
 362                 err = __unmap_grant_pages(map, offset, range);
 363                 offset += range;
 364                 pages -= range;
 365         }
 366
 367         return err;
 368 }
 369
 370 /* ------------------------------------------------------------------ */
 371
 372 static void gntdev_vma_open(struct vm_area_struct *vma)
 373 {
 374         struct grant_map *map = vma->vm_private_data;
 375
 376         pr_debug("gntdev_vma_open %p\n", vma);
 377         atomic_inc(&map->users);
 378 }
 379
 380 static void gntdev_vma_close(struct vm_area_struct *vma)
 381 {
 382         struct grant_map *map = vma->vm_private_data;
 383         struct file *file = vma->vm_file;
 384         struct gntdev_priv *priv = file->private_data;
 385
 386         pr_debug("gntdev_vma_close %p\n", vma);
 387         if (use_ptemod) {
 388                 /* It is possible that an mmu notifier could be running
 389                  * concurrently, so take priv->lock to ensure that the vma won't
 390                  * vanishing during the unmap_grant_pages call, since we will
 391                  * spin here until that completes. Such a concurrent call will
 392                  * not do any unmapping, since that has been done prior to
 393                  * closing the vma, but it may still iterate the unmap_ops list.
 394                  */
 395                 spin_lock(&priv->lock);
 396                 map->vma = NULL;
 397                 spin_unlock(&priv->lock);
 398         }
 399         vma->vm_private_data = NULL;
 400         gntdev_put_map(priv, map);
 401 }
 402
 403 static struct vm_operations_struct gntdev_vmops = {
 404         .open = gntdev_vma_open,
 405         .close = gntdev_vma_close,
 406 };
 407
 408 /* ------------------------------------------------------------------ */
 409
 410 static void unmap_if_in_range(struct grant_map *map,
 411                               unsigned long start, unsigned long end)
 412 {
 413         unsigned long mstart, mend;
 414         int err;
 415
 416         if (!map->vma)
 417                 return;
 418         if (map->vma->vm_start >= end)
 419                 return;
 420         if (map->vma->vm_end <= start)
 421                 return;
 422         mstart = max(start, map->vma->vm_start);
 423         mend   = min(end,   map->vma->vm_end);
 424         pr_debug("map %d+%d (%lx %lx), range %lx %lx, mrange %lx %lx\n",
 425                         map->index, map->count,
 426                         map->vma->vm_start, map->vma->vm_end,
 427                         start, end, mstart, mend);
 428         err = unmap_grant_pages(map,
 429                                 (mstart - map->vma->vm_start) >> PAGE_SHIFT,
 430                                 (mend - mstart) >> PAGE_SHIFT);
 431         WARN_ON(err);
 432 }
 433
 434 static void mn_invl_range_start(struct mmu_notifier *mn,
 435                                 struct mm_struct *mm,
 436                                 unsigned long start, unsigned long end)
 437 {
 438         struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
 439         struct grant_map *map;
 440
 441         spin_lock(&priv->lock);
 442         list_for_each_entry(map, &priv->maps, next) {
 443                 unmap_if_in_range(map, start, end);
 444         }
 445         list_for_each_entry(map, &priv->freeable_maps, next) {
 446                 unmap_if_in_range(map, start, end);
 447         }
 448         spin_unlock(&priv->lock);
 449 }
 450
 451 static void mn_invl_page(struct mmu_notifier *mn,
 452                          struct mm_struct *mm,
 453                          unsigned long address)
 454 {
 455         mn_invl_range_start(mn, mm, address, address + PAGE_SIZE);
 456 }
 457
 458 static void mn_release(struct mmu_notifier *mn,
 459                        struct mm_struct *mm)
 460 {
 461         struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
 462         struct grant_map *map;
 463         int err;
 464
 465         spin_lock(&priv->lock);
 466         list_for_each_entry(map, &priv->maps, next) {
 467                 if (!map->vma)
 468                         continue;
 469                 pr_debug("map %d+%d (%lx %lx)\n",
 470                                 map->index, map->count,
 471                                 map->vma->vm_start, map->vma->vm_end);
 472                 err = unmap_grant_pages(map, /* offset */ 0, map->count);
 473                 WARN_ON(err);
 474         }
 475         list_for_each_entry(map, &priv->freeable_maps, next) {
 476                 if (!map->vma)
 477                         continue;
 478                 pr_debug("map %d+%d (%lx %lx)\n",
 479                                 map->index, map->count,
 480                                 map->vma->vm_start, map->vma->vm_end);
 481                 err = unmap_grant_pages(map, /* offset */ 0, map->count);
 482                 WARN_ON(err);
 483         }
 484         spin_unlock(&priv->lock);
 485 }
 486
 487 static struct mmu_notifier_ops gntdev_mmu_ops = {
 488         .release                = mn_release,
 489         .invalidate_page        = mn_invl_page,
 490         .invalidate_range_start = mn_invl_range_start,
 491 };
 492
 493 /* ------------------------------------------------------------------ */
 494
 495 static int gntdev_open(struct inode *inode, struct file *flip)
 496 {
 497         struct gntdev_priv *priv;
 498         int ret = 0;
 499
 500         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 501         if (!priv)
 502                 return -ENOMEM;
 503
 504         INIT_LIST_HEAD(&priv->maps);
 505         INIT_LIST_HEAD(&priv->freeable_maps);
 506         spin_lock_init(&priv->lock);
 507
 508         if (use_ptemod) {
 509                 priv->mm = get_task_mm(current);
 510                 if (!priv->mm) {
 511                         kfree(priv);
 512                         return -ENOMEM;
 513                 }
 514                 priv->mn.ops = &gntdev_mmu_ops;
 515                 ret = mmu_notifier_register(&priv->mn, priv->mm);
 516                 mmput(priv->mm);
 517         }
 518
 519         if (ret) {
 520                 kfree(priv);
 521                 return ret;
 522         }
 523
 524         flip->private_data = priv;
 525         pr_debug("priv %p\n", priv);
 526
 527         return 0;
 528 }
 529
 530 static int gntdev_release(struct inode *inode, struct file *flip)
 531 {
 532         struct gntdev_priv *priv = flip->private_data;
 533         struct grant_map *map;
 534
 535         pr_debug("priv %p\n", priv);
 536
 537         while (!list_empty(&priv->maps)) {
 538                 map = list_entry(priv->maps.next, struct grant_map, next);
 539                 list_del(&map->next);
 540                 gntdev_put_map(NULL /* already removed */, map);
 541         }
 542         WARN_ON(!list_empty(&priv->freeable_maps));
 543
 544         if (use_ptemod)
 545                 mmu_notifier_unregister(&priv->mn, priv->mm);
 546         kfree(priv);
 547         return 0;
 548 }
 549
 550 static long gntdev_ioctl_map_grant_ref(struct gntdev_priv *priv,
 551                                        struct ioctl_gntdev_map_grant_ref __user *u)
 552 {
 553         struct ioctl_gntdev_map_grant_ref op;
 554         struct grant_map *map;
 555         int err;
 556
 557         if (copy_from_user(&op, u, sizeof(op)) != 0)
 558                 return -EFAULT;
 559         pr_debug("priv %p, add %d\n", priv, op.count);
 560         if (unlikely(op.count <= 0))
 561                 return -EINVAL;
 562
 563         err = -ENOMEM;
 564         map = gntdev_alloc_map(priv, op.count);
 565         if (!map)
 566                 return err;
 567
 568         if (unlikely(atomic_add_return(op.count, &pages_mapped) > limit)) {
 569                 pr_debug("can't map: over limit\n");
 570                 gntdev_put_map(NULL, map);
 571                 return err;
 572         }
 573
 574         if (copy_from_user(map->grants, &u->refs,
 575                            sizeof(map->grants[0]) * op.count) != 0) {
 576                 gntdev_put_map(NULL, map);
 577                 return -EFAULT;
 578         }
 579
 580         spin_lock(&priv->lock);
 581         gntdev_add_map(priv, map);
 582         op.index = map->index << PAGE_SHIFT;
 583         spin_unlock(&priv->lock);
 584
 585         if (copy_to_user(u, &op, sizeof(op)) != 0)
 586                 return -EFAULT;
 587
 588         return 0;
 589 }
 590
 591 static long gntdev_ioctl_unmap_grant_ref(struct gntdev_priv *priv,
 592                                          struct ioctl_gntdev_unmap_grant_ref __user *u)
 593 {
 594         struct ioctl_gntdev_unmap_grant_ref op;
 595         struct grant_map *map;
 596         int err = -ENOENT;
 597
 598         if (copy_from_user(&op, u, sizeof(op)) != 0)
 599                 return -EFAULT;
 600         pr_debug("priv %p, del %d+%d\n", priv, (int)op.index, (int)op.count);
 601
 602         spin_lock(&priv->lock);
 603         map = gntdev_find_map_index(priv, op.index >> PAGE_SHIFT, op.count);
 604         if (map) {
 605                 list_del(&map->next);
 606                 if (populate_freeable_maps)
 607                         list_add_tail(&map->next, &priv->freeable_maps);
 608                 err = 0;
 609         }
 610         spin_unlock(&priv->lock);
 611         if (map)
 612                 gntdev_put_map(priv, map);
 613         return err;
 614 }
 615
 616 static long gntdev_ioctl_get_offset_for_vaddr(struct gntdev_priv *priv,
 617                                               struct ioctl_gntdev_get_offset_for_vaddr __user *u)
 618 {
 619         struct ioctl_gntdev_get_offset_for_vaddr op;
 620         struct vm_area_struct *vma;
 621         struct grant_map *map;
 622         int rv = -EINVAL;
 623
 624         if (copy_from_user(&op, u, sizeof(op)) != 0)
 625                 return -EFAULT;
 626         pr_debug("priv %p, offset for vaddr %lx\n", priv, (unsigned long)op.vaddr);
 627
 628         down_read(&current->mm->mmap_sem);
 629         vma = find_vma(current->mm, op.vaddr);
 630         if (!vma || vma->vm_ops != &gntdev_vmops)
 631                 goto out_unlock;
 632
 633         map = vma->vm_private_data;
 634         if (!map)
 635                 goto out_unlock;
 636
 637         op.offset = map->index << PAGE_SHIFT;
 638         op.count = map->count;
 639         rv = 0;
 640
 641  out_unlock:
 642         up_read(&current->mm->mmap_sem);
 643
 644         if (rv == 0 && copy_to_user(u, &op, sizeof(op)) != 0)
 645                 return -EFAULT;
 646         return rv;
 647 }
 648
 649 static long gntdev_ioctl_notify(struct gntdev_priv *priv, void __user *u)
 650 {
 651         struct ioctl_gntdev_unmap_notify op;
 652         struct grant_map *map;
 653         int rc;
 654         int out_flags;
 655         unsigned int out_event;
 656
 657         if (copy_from_user(&op, u, sizeof(op)))
 658                 return -EFAULT;
 659
 660         if (op.action & ~(UNMAP_NOTIFY_CLEAR_BYTE|UNMAP_NOTIFY_SEND_EVENT))
 661                 return -EINVAL;
 662
 663         /* We need to grab a reference to the event channel we are going to use
 664          * to send the notify before releasing the reference we may already have
 665          * (if someone has called this ioctl twice). This is required so that
 666          * it is possible to change the clear_byte part of the notification
 667          * without disturbing the event channel part, which may now be the last
 668          * reference to that event channel.
 669          */
 670         if (op.action & UNMAP_NOTIFY_SEND_EVENT) {
 671                 if (evtchn_get(op.event_channel_port))
 672                         return -EINVAL;
 673         }
 674
 675         out_flags = op.action;
 676         out_event = op.event_channel_port;
 677
 678         spin_lock(&priv->lock);
 679
 680         list_for_each_entry(map, &priv->maps, next) {
 681                 uint64_t begin = map->index << PAGE_SHIFT;
 682                 uint64_t end = (map->index + map->count) << PAGE_SHIFT;
 683                 if (op.index >= begin && op.index < end)
 684                         goto found;
 685         }
 686         rc = -ENOENT;
 687         goto unlock_out;
 688
 689  found:
 690         if ((op.action & UNMAP_NOTIFY_CLEAR_BYTE) &&
 691                         (map->flags & GNTMAP_readonly)) {
 692                 rc = -EINVAL;
 693                 goto unlock_out;
 694         }
 695
 696         out_flags = map->notify.flags;
 697         out_event = map->notify.event;
 698
 699         map->notify.flags = op.action;
 700         map->notify.addr = op.index - (map->index << PAGE_SHIFT);
 701         map->notify.event = op.event_channel_port;
 702
 703         rc = 0;
 704
 705  unlock_out:
 706         spin_unlock(&priv->lock);
 707
 708         /* Drop the reference to the event channel we did not save in the map */
 709         if (out_flags & UNMAP_NOTIFY_SEND_EVENT)
 710                 evtchn_put(out_event);
 711
 712         return rc;
 713 }
 714
 715 static long gntdev_ioctl(struct file *flip,
 716                          unsigned int cmd, unsigned long arg)
 717 {
 718         struct gntdev_priv *priv = flip->private_data;
 719         void __user *ptr = (void __user *)arg;
 720
 721         switch (cmd) {
 722         case IOCTL_GNTDEV_MAP_GRANT_REF:
 723                 return gntdev_ioctl_map_grant_ref(priv, ptr);
 724
 725         case IOCTL_GNTDEV_UNMAP_GRANT_REF:
 726                 return gntdev_ioctl_unmap_grant_ref(priv, ptr);
 727
 728         case IOCTL_GNTDEV_GET_OFFSET_FOR_VADDR:
 729                 return gntdev_ioctl_get_offset_for_vaddr(priv, ptr);
 730
 731         case IOCTL_GNTDEV_SET_UNMAP_NOTIFY:
 732                 return gntdev_ioctl_notify(priv, ptr);
 733
 734         default:
 735                 pr_debug("priv %p, unknown cmd %x\n", priv, cmd);
 736                 return -ENOIOCTLCMD;
 737         }
 738
 739         return 0;
 740 }
 741
 742 static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma)
 743 {
 744         struct gntdev_priv *priv = flip->private_data;
 745         int index = vma->vm_pgoff;
 746         int count = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
 747         struct grant_map *map;
 748         int i, err = -EINVAL;
 749
 750         if ((vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_SHARED))
 751                 return -EINVAL;
 752
 753         pr_debug("map %d+%d at %lx (pgoff %lx)\n",
 754                         index, count, vma->vm_start, vma->vm_pgoff);
 755
 756         spin_lock(&priv->lock);
 757         map = gntdev_find_map_index(priv, index, count);
 758         if (!map)
 759                 goto unlock_out;
 760         if (use_ptemod && map->vma)
 761                 goto unlock_out;
 762         if (use_ptemod && priv->mm != vma->vm_mm) {
 763                 printk(KERN_WARNING "Huh? Other mm?\n");
 764                 goto unlock_out;
 765         }
 766
 767         atomic_inc(&map->users);
 768
 769         vma->vm_ops = &gntdev_vmops;
 770
 771         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
 772
 773         if (use_ptemod)
 774                 vma->vm_flags |= VM_DONTCOPY;
 775
 776         vma->vm_private_data = map;
 777
 778         if (use_ptemod)
 779                 map->vma = vma;
 780
 781         if (map->flags) {
 782                 if ((vma->vm_flags & VM_WRITE) &&
 783                                 (map->flags & GNTMAP_readonly))
 784                         goto out_unlock_put;
 785         } else {
 786                 map->flags = GNTMAP_host_map;
 787                 if (!(vma->vm_flags & VM_WRITE))
 788                         map->flags |= GNTMAP_readonly;
 789         }
 790
 791         spin_unlock(&priv->lock);
 792
 793         if (use_ptemod) {
 794                 err = apply_to_page_range(vma->vm_mm, vma->vm_start,
 795                                           vma->vm_end - vma->vm_start,
 796                                           find_grant_ptes, map);
 797                 if (err) {
 798                         printk(KERN_WARNING "find_grant_ptes() failure.\n");
 799                         goto out_put_map;
 800                 }
 801         }
 802
 803         err = map_grant_pages(map);
 804         if (err)
 805                 goto out_put_map;
 806
 807         if (!use_ptemod) {
 808                 for (i = 0; i < count; i++) {
 809                         err = vm_insert_page(vma, vma->vm_start + i*PAGE_SIZE,
 810                                 map->pages[i]);
 811                         if (err)
 812                                 goto out_put_map;
 813                 }
 814         }
 815
 816         return 0;
 817
 818 unlock_out:
 819         spin_unlock(&priv->lock);
 820         return err;
 821
 822 out_unlock_put:
 823         spin_unlock(&priv->lock);
 824 out_put_map:
 825         if (use_ptemod)
 826                 map->vma = NULL;
 827         gntdev_put_map(priv, map);
 828         return err;
 829 }
 830
 831 static const struct file_operations gntdev_fops = {
 832         .owner = THIS_MODULE,
 833         .open = gntdev_open,
 834         .release = gntdev_release,
 835         .mmap = gntdev_mmap,
 836         .unlocked_ioctl = gntdev_ioctl
 837 };
 838
 839 static struct miscdevice gntdev_miscdev = {
 840         .minor        = MISC_DYNAMIC_MINOR,
 841         .name         = "xen/gntdev",
 842         .fops         = &gntdev_fops,
 843 };
 844
 845 /* ------------------------------------------------------------------ */
 846
 847 static int __init gntdev_init(void)
 848 {
 849         int err;
 850
 851         if (!xen_domain())
 852                 return -ENODEV;
 853
 854         use_ptemod = xen_pv_domain();
 855
 856         err = misc_register(&gntdev_miscdev);
 857         if (err != 0) {
 858                 printk(KERN_ERR "Could not register gntdev device\n");
 859                 return err;
 860         }
 861         return 0;
 862 }
 863
 864 static void __exit gntdev_exit(void)
 865 {
 866         misc_deregister(&gntdev_miscdev);
 867 }
 868
 869 module_init(gntdev_init);
 870 module_exit(gntdev_exit);
 871
 872 /* ------------------------------------------------------------------ */