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