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
  36 #include <xen/xen.h>
  37 #include <xen/grant_table.h>
  38 #include <xen/gntdev.h>
  39 #include <asm/xen/hypervisor.h>
  40 #include <asm/xen/hypercall.h>
  41 #include <asm/xen/page.h>
  42
  43 MODULE_LICENSE("GPL");
  44 MODULE_AUTHOR("Derek G. Murray <Derek.Murray@cl.cam.ac.uk>, "
  45               "Gerd Hoffmann <kraxel@redhat.com>");
  46 MODULE_DESCRIPTION("User-space granted page access driver");
  47
  48 static int limit = 1024;
  49 module_param(limit, int, 0644);
  50 MODULE_PARM_DESC(limit, "Maximum number of grants that may be mapped at "
  51                 "once by a gntdev instance");
  52
  53 struct gntdev_priv {
  54         struct list_head maps;
  55         uint32_t used;
  56         uint32_t limit;
  57         /* lock protects maps from concurrent changes */
  58         spinlock_t lock;
  59         struct mm_struct *mm;
  60         struct mmu_notifier mn;
  61 };
  62
  63 struct grant_map {
  64         struct list_head next;
  65         struct gntdev_priv *priv;
  66         struct vm_area_struct *vma;
  67         int index;
  68         int count;
  69         int flags;
  70         int is_mapped;
  71         struct ioctl_gntdev_grant_ref *grants;
  72         struct gnttab_map_grant_ref   *map_ops;
  73         struct gnttab_unmap_grant_ref *unmap_ops;
  74         struct page **pages;
  75 };
  76
  77 /* ------------------------------------------------------------------ */
  78
  79 static void gntdev_print_maps(struct gntdev_priv *priv,
  80                               char *text, int text_index)
  81 {
  82 #ifdef DEBUG
  83         struct grant_map *map;
  84
  85         pr_debug("maps list (priv %p, usage %d/%d)\n",
  86                priv, priv->used, priv->limit);
  87
  88         list_for_each_entry(map, &priv->maps, next)
  89                 pr_debug("  index %2d, count %2d %s\n",
  90                        map->index, map->count,
  91                        map->index == text_index && text ? text : "");
  92 #endif
  93 }
  94
  95 static struct grant_map *gntdev_alloc_map(struct gntdev_priv *priv, int count)
  96 {
  97         struct grant_map *add;
  98         int i;
  99
 100         add = kzalloc(sizeof(struct grant_map), GFP_KERNEL);
 101         if (NULL == add)
 102                 return NULL;
 103
 104         add->grants    = kzalloc(sizeof(add->grants[0])    * count, GFP_KERNEL);
 105         add->map_ops   = kzalloc(sizeof(add->map_ops[0])   * count, GFP_KERNEL);
 106         add->unmap_ops = kzalloc(sizeof(add->unmap_ops[0]) * count, GFP_KERNEL);
 107         add->pages     = kzalloc(sizeof(add->pages[0])     * count, GFP_KERNEL);
 108         if (NULL == add->grants    ||
 109             NULL == add->map_ops   ||
 110             NULL == add->unmap_ops ||
 111             NULL == add->pages)
 112                 goto err;
 113
 114         for (i = 0; i < count; i++) {
 115                 add->pages[i] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
 116                 if (add->pages[i] == NULL)
 117                         goto err;
 118         }
 119
 120         add->index = 0;
 121         add->count = count;
 122         add->priv  = priv;
 123
 124         if (add->count + priv->used > priv->limit)
 125                 goto err;
 126
 127         return add;
 128
 129 err:
 130         if (add->pages)
 131                 for (i = 0; i < count; i++) {
 132                         if (add->pages[i])
 133                                 __free_page(add->pages[i]);
 134                 }
 135         kfree(add->pages);
 136         kfree(add->grants);
 137         kfree(add->map_ops);
 138         kfree(add->unmap_ops);
 139         kfree(add);
 140         return NULL;
 141 }
 142
 143 static void gntdev_add_map(struct gntdev_priv *priv, struct grant_map *add)
 144 {
 145         struct grant_map *map;
 146
 147         list_for_each_entry(map, &priv->maps, next) {
 148                 if (add->index + add->count < map->index) {
 149                         list_add_tail(&add->next, &map->next);
 150                         goto done;
 151                 }
 152                 add->index = map->index + map->count;
 153         }
 154         list_add_tail(&add->next, &priv->maps);
 155
 156 done:
 157         priv->used += add->count;
 158         gntdev_print_maps(priv, "[new]", add->index);
 159 }
 160
 161 static struct grant_map *gntdev_find_map_index(struct gntdev_priv *priv,
 162                 int index, int count)
 163 {
 164         struct grant_map *map;
 165
 166         list_for_each_entry(map, &priv->maps, next) {
 167                 if (map->index != index)
 168                         continue;
 169                 if (map->count != count)
 170                         continue;
 171                 return map;
 172         }
 173         return NULL;
 174 }
 175
 176 static struct grant_map *gntdev_find_map_vaddr(struct gntdev_priv *priv,
 177                                                unsigned long vaddr)
 178 {
 179         struct grant_map *map;
 180
 181         list_for_each_entry(map, &priv->maps, next) {
 182                 if (!map->vma)
 183                         continue;
 184                 if (vaddr < map->vma->vm_start)
 185                         continue;
 186                 if (vaddr >= map->vma->vm_end)
 187                         continue;
 188                 return map;
 189         }
 190         return NULL;
 191 }
 192
 193 static int gntdev_del_map(struct grant_map *map)
 194 {
 195         int i;
 196
 197         if (map->vma)
 198                 return -EBUSY;
 199         for (i = 0; i < map->count; i++)
 200                 if (map->unmap_ops[i].handle)
 201                         return -EBUSY;
 202
 203         map->priv->used -= map->count;
 204         list_del(&map->next);
 205         return 0;
 206 }
 207
 208 static void gntdev_free_map(struct grant_map *map)
 209 {
 210         int i;
 211
 212         if (!map)
 213                 return;
 214
 215         if (map->pages)
 216                 for (i = 0; i < map->count; i++) {
 217                         if (map->pages[i])
 218                                 __free_page(map->pages[i]);
 219                 }
 220         kfree(map->pages);
 221         kfree(map->grants);
 222         kfree(map->map_ops);
 223         kfree(map->unmap_ops);
 224         kfree(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         u64 pte_maddr;
 235
 236         BUG_ON(pgnr >= map->count);
 237         pte_maddr = arbitrary_virt_to_machine(pte).maddr;
 238
 239         gnttab_set_map_op(&map->map_ops[pgnr], pte_maddr,
 240                           GNTMAP_contains_pte | map->flags,
 241                           map->grants[pgnr].ref,
 242                           map->grants[pgnr].domid);
 243         gnttab_set_unmap_op(&map->unmap_ops[pgnr], pte_maddr,
 244                             GNTMAP_contains_pte | map->flags,
 245                             0 /* handle */);
 246         return 0;
 247 }
 248
 249 static int map_grant_pages(struct grant_map *map)
 250 {
 251         int i, err = 0;
 252
 253         pr_debug("map %d+%d\n", map->index, map->count);
 254         err = gnttab_map_refs(map->map_ops, map->pages, map->count);
 255         if (err)
 256                 return err;
 257
 258         for (i = 0; i < map->count; i++) {
 259                 if (map->map_ops[i].status)
 260                         err = -EINVAL;
 261                 map->unmap_ops[i].handle = map->map_ops[i].handle;
 262         }
 263         return err;
 264 }
 265
 266 static int unmap_grant_pages(struct grant_map *map, int offset, int pages)
 267 {
 268         int i, err = 0;
 269
 270         pr_debug("map %d+%d [%d+%d]\n", map->index, map->count, offset, pages);
 271         err = gnttab_unmap_refs(map->unmap_ops + offset, map->pages, pages);
 272         if (err)
 273                 return err;
 274
 275         for (i = 0; i < pages; i++) {
 276                 if (map->unmap_ops[offset+i].status)
 277                         err = -EINVAL;
 278                 map->unmap_ops[offset+i].handle = 0;
 279         }
 280         return err;
 281 }
 282
 283 /* ------------------------------------------------------------------ */
 284
 285 static void gntdev_vma_close(struct vm_area_struct *vma)
 286 {
 287         struct grant_map *map = vma->vm_private_data;
 288
 289         pr_debug("close %p\n", vma);
 290         map->is_mapped = 0;
 291         map->vma = NULL;
 292         vma->vm_private_data = NULL;
 293 }
 294
 295 static int gntdev_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 296 {
 297         pr_debug("vaddr %p, pgoff %ld (shouldn't happen)\n",
 298                         vmf->virtual_address, vmf->pgoff);
 299         vmf->flags = VM_FAULT_ERROR;
 300         return 0;
 301 }
 302
 303 static struct vm_operations_struct gntdev_vmops = {
 304         .close = gntdev_vma_close,
 305         .fault = gntdev_vma_fault,
 306 };
 307
 308 /* ------------------------------------------------------------------ */
 309
 310 static void mn_invl_range_start(struct mmu_notifier *mn,
 311                                 struct mm_struct *mm,
 312                                 unsigned long start, unsigned long end)
 313 {
 314         struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
 315         struct grant_map *map;
 316         unsigned long mstart, mend;
 317         int err;
 318
 319         spin_lock(&priv->lock);
 320         list_for_each_entry(map, &priv->maps, next) {
 321                 if (!map->vma)
 322                         continue;
 323                 if (!map->is_mapped)
 324                         continue;
 325                 if (map->vma->vm_start >= end)
 326                         continue;
 327                 if (map->vma->vm_end <= start)
 328                         continue;
 329                 mstart = max(start, map->vma->vm_start);
 330                 mend   = min(end,   map->vma->vm_end);
 331                 pr_debug("map %d+%d (%lx %lx), range %lx %lx, mrange %lx %lx\n",
 332                                 map->index, map->count,
 333                                 map->vma->vm_start, map->vma->vm_end,
 334                                 start, end, mstart, mend);
 335                 err = unmap_grant_pages(map,
 336                                         (mstart - map->vma->vm_start) >> PAGE_SHIFT,
 337                                         (mend - mstart) >> PAGE_SHIFT);
 338                 WARN_ON(err);
 339         }
 340         spin_unlock(&priv->lock);
 341 }
 342
 343 static void mn_invl_page(struct mmu_notifier *mn,
 344                          struct mm_struct *mm,
 345                          unsigned long address)
 346 {
 347         mn_invl_range_start(mn, mm, address, address + PAGE_SIZE);
 348 }
 349
 350 static void mn_release(struct mmu_notifier *mn,
 351                        struct mm_struct *mm)
 352 {
 353         struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
 354         struct grant_map *map;
 355         int err;
 356
 357         spin_lock(&priv->lock);
 358         list_for_each_entry(map, &priv->maps, next) {
 359                 if (!map->vma)
 360                         continue;
 361                 pr_debug("map %d+%d (%lx %lx)\n",
 362                                 map->index, map->count,
 363                                 map->vma->vm_start, map->vma->vm_end);
 364                 err = unmap_grant_pages(map, /* offset */ 0, map->count);
 365                 WARN_ON(err);
 366         }
 367         spin_unlock(&priv->lock);
 368 }
 369
 370 struct mmu_notifier_ops gntdev_mmu_ops = {
 371         .release                = mn_release,
 372         .invalidate_page        = mn_invl_page,
 373         .invalidate_range_start = mn_invl_range_start,
 374 };
 375
 376 /* ------------------------------------------------------------------ */
 377
 378 static int gntdev_open(struct inode *inode, struct file *flip)
 379 {
 380         struct gntdev_priv *priv;
 381         int ret = 0;
 382
 383         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 384         if (!priv)
 385                 return -ENOMEM;
 386
 387         INIT_LIST_HEAD(&priv->maps);
 388         spin_lock_init(&priv->lock);
 389         priv->limit = limit;
 390
 391         priv->mm = get_task_mm(current);
 392         if (!priv->mm) {
 393                 kfree(priv);
 394                 return -ENOMEM;
 395         }
 396         priv->mn.ops = &gntdev_mmu_ops;
 397         ret = mmu_notifier_register(&priv->mn, priv->mm);
 398         mmput(priv->mm);
 399
 400         if (ret) {
 401                 kfree(priv);
 402                 return ret;
 403         }
 404
 405         flip->private_data = priv;
 406         pr_debug("priv %p\n", priv);
 407
 408         return 0;
 409 }
 410
 411 static int gntdev_release(struct inode *inode, struct file *flip)
 412 {
 413         struct gntdev_priv *priv = flip->private_data;
 414         struct grant_map *map;
 415         int err;
 416
 417         pr_debug("priv %p\n", priv);
 418
 419         spin_lock(&priv->lock);
 420         while (!list_empty(&priv->maps)) {
 421                 map = list_entry(priv->maps.next, struct grant_map, next);
 422                 err = gntdev_del_map(map);
 423                 if (WARN_ON(err))
 424                         gntdev_free_map(map);
 425
 426         }
 427         spin_unlock(&priv->lock);
 428
 429         mmu_notifier_unregister(&priv->mn, priv->mm);
 430         kfree(priv);
 431         return 0;
 432 }
 433
 434 static long gntdev_ioctl_map_grant_ref(struct gntdev_priv *priv,
 435                                        struct ioctl_gntdev_map_grant_ref __user *u)
 436 {
 437         struct ioctl_gntdev_map_grant_ref op;
 438         struct grant_map *map;
 439         int err;
 440
 441         if (copy_from_user(&op, u, sizeof(op)) != 0)
 442                 return -EFAULT;
 443         pr_debug("priv %p, add %d\n", priv, op.count);
 444         if (unlikely(op.count <= 0))
 445                 return -EINVAL;
 446         if (unlikely(op.count > priv->limit))
 447                 return -EINVAL;
 448
 449         err = -ENOMEM;
 450         map = gntdev_alloc_map(priv, op.count);
 451         if (!map)
 452                 return err;
 453         if (copy_from_user(map->grants, &u->refs,
 454                            sizeof(map->grants[0]) * op.count) != 0) {
 455                 gntdev_free_map(map);
 456                 return err;
 457         }
 458
 459         spin_lock(&priv->lock);
 460         gntdev_add_map(priv, map);
 461         op.index = map->index << PAGE_SHIFT;
 462         spin_unlock(&priv->lock);
 463
 464         if (copy_to_user(u, &op, sizeof(op)) != 0) {
 465                 spin_lock(&priv->lock);
 466                 gntdev_del_map(map);
 467                 spin_unlock(&priv->lock);
 468                 gntdev_free_map(map);
 469                 return err;
 470         }
 471         return 0;
 472 }
 473
 474 static long gntdev_ioctl_unmap_grant_ref(struct gntdev_priv *priv,
 475                                          struct ioctl_gntdev_unmap_grant_ref __user *u)
 476 {
 477         struct ioctl_gntdev_unmap_grant_ref op;
 478         struct grant_map *map;
 479         int err = -ENOENT;
 480
 481         if (copy_from_user(&op, u, sizeof(op)) != 0)
 482                 return -EFAULT;
 483         pr_debug("priv %p, del %d+%d\n", priv, (int)op.index, (int)op.count);
 484
 485         spin_lock(&priv->lock);
 486         map = gntdev_find_map_index(priv, op.index >> PAGE_SHIFT, op.count);
 487         if (map)
 488                 err = gntdev_del_map(map);
 489         spin_unlock(&priv->lock);
 490         if (!err)
 491                 gntdev_free_map(map);
 492         return err;
 493 }
 494
 495 static long gntdev_ioctl_get_offset_for_vaddr(struct gntdev_priv *priv,
 496                                               struct ioctl_gntdev_get_offset_for_vaddr __user *u)
 497 {
 498         struct ioctl_gntdev_get_offset_for_vaddr op;
 499         struct grant_map *map;
 500
 501         if (copy_from_user(&op, u, sizeof(op)) != 0)
 502                 return -EFAULT;
 503         pr_debug("priv %p, offset for vaddr %lx\n", priv, (unsigned long)op.vaddr);
 504
 505         spin_lock(&priv->lock);
 506         map = gntdev_find_map_vaddr(priv, op.vaddr);
 507         if (map == NULL ||
 508             map->vma->vm_start != op.vaddr) {
 509                 spin_unlock(&priv->lock);
 510                 return -EINVAL;
 511         }
 512         op.offset = map->index << PAGE_SHIFT;
 513         op.count = map->count;
 514         spin_unlock(&priv->lock);
 515
 516         if (copy_to_user(u, &op, sizeof(op)) != 0)
 517                 return -EFAULT;
 518         return 0;
 519 }
 520
 521 static long gntdev_ioctl_set_max_grants(struct gntdev_priv *priv,
 522                                         struct ioctl_gntdev_set_max_grants __user *u)
 523 {
 524         struct ioctl_gntdev_set_max_grants op;
 525
 526         if (copy_from_user(&op, u, sizeof(op)) != 0)
 527                 return -EFAULT;
 528         pr_debug("priv %p, limit %d\n", priv, op.count);
 529         if (op.count > limit)
 530                 return -E2BIG;
 531
 532         spin_lock(&priv->lock);
 533         priv->limit = op.count;
 534         spin_unlock(&priv->lock);
 535         return 0;
 536 }
 537
 538 static long gntdev_ioctl(struct file *flip,
 539                          unsigned int cmd, unsigned long arg)
 540 {
 541         struct gntdev_priv *priv = flip->private_data;
 542         void __user *ptr = (void __user *)arg;
 543
 544         switch (cmd) {
 545         case IOCTL_GNTDEV_MAP_GRANT_REF:
 546                 return gntdev_ioctl_map_grant_ref(priv, ptr);
 547
 548         case IOCTL_GNTDEV_UNMAP_GRANT_REF:
 549                 return gntdev_ioctl_unmap_grant_ref(priv, ptr);
 550
 551         case IOCTL_GNTDEV_GET_OFFSET_FOR_VADDR:
 552                 return gntdev_ioctl_get_offset_for_vaddr(priv, ptr);
 553
 554         case IOCTL_GNTDEV_SET_MAX_GRANTS:
 555                 return gntdev_ioctl_set_max_grants(priv, ptr);
 556
 557         default:
 558                 pr_debug("priv %p, unknown cmd %x\n", priv, cmd);
 559                 return -ENOIOCTLCMD;
 560         }
 561
 562         return 0;
 563 }
 564
 565 static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma)
 566 {
 567         struct gntdev_priv *priv = flip->private_data;
 568         int index = vma->vm_pgoff;
 569         int count = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
 570         struct grant_map *map;
 571         int err = -EINVAL;
 572
 573         if ((vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_SHARED))
 574                 return -EINVAL;
 575
 576         pr_debug("map %d+%d at %lx (pgoff %lx)\n",
 577                         index, count, vma->vm_start, vma->vm_pgoff);
 578
 579         spin_lock(&priv->lock);
 580         map = gntdev_find_map_index(priv, index, count);
 581         if (!map)
 582                 goto unlock_out;
 583         if (map->vma)
 584                 goto unlock_out;
 585         if (priv->mm != vma->vm_mm) {
 586                 printk(KERN_WARNING "Huh? Other mm?\n");
 587                 goto unlock_out;
 588         }
 589
 590         vma->vm_ops = &gntdev_vmops;
 591
 592         vma->vm_flags |= VM_RESERVED|VM_DONTCOPY|VM_DONTEXPAND|VM_PFNMAP;
 593
 594         vma->vm_private_data = map;
 595         map->vma = vma;
 596
 597         map->flags = GNTMAP_host_map | GNTMAP_application_map;
 598         if (!(vma->vm_flags & VM_WRITE))
 599                 map->flags |= GNTMAP_readonly;
 600
 601         spin_unlock(&priv->lock);
 602
 603         err = apply_to_page_range(vma->vm_mm, vma->vm_start,
 604                                   vma->vm_end - vma->vm_start,
 605                                   find_grant_ptes, map);
 606         if (err) {
 607                 printk(KERN_WARNING "find_grant_ptes() failure.\n");
 608                 return err;
 609         }
 610
 611         err = map_grant_pages(map);
 612         if (err) {
 613                 printk(KERN_WARNING "map_grant_pages() failure.\n");
 614                 return err;
 615         }
 616
 617         map->is_mapped = 1;
 618
 619         return 0;
 620
 621 unlock_out:
 622         spin_unlock(&priv->lock);
 623         return err;
 624 }
 625
 626 static const struct file_operations gntdev_fops = {
 627         .owner = THIS_MODULE,
 628         .open = gntdev_open,
 629         .release = gntdev_release,
 630         .mmap = gntdev_mmap,
 631         .unlocked_ioctl = gntdev_ioctl
 632 };
 633
 634 static struct miscdevice gntdev_miscdev = {
 635         .minor        = MISC_DYNAMIC_MINOR,
 636         .name         = "xen/gntdev",
 637         .fops         = &gntdev_fops,
 638 };
 639
 640 /* ------------------------------------------------------------------ */
 641
 642 static int __init gntdev_init(void)
 643 {
 644         int err;
 645
 646         if (!xen_domain())
 647                 return -ENODEV;
 648
 649         err = misc_register(&gntdev_miscdev);
 650         if (err != 0) {
 651                 printk(KERN_ERR "Could not register gntdev device\n");
 652                 return err;
 653         }
 654         return 0;
 655 }
 656
 657 static void __exit gntdev_exit(void)
 658 {
 659         misc_deregister(&gntdev_miscdev);
 660 }
 661
 662 module_init(gntdev_init);
 663 module_exit(gntdev_exit);
 664
 665 /* ------------------------------------------------------------------ */