staging: hv: Fix GARP not sent after Quick Migration
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / xen / gntdev.c
blobb0f9e8fb005255d0f49ddf9e2aa3584dbbbd9d7b
1 /******************************************************************************
2 * gntdev.c
4 * Device for accessing (in user-space) pages that have been granted by other
5 * domains.
7 * Copyright (c) 2006-2007, D G Murray.
8 * (c) 2009 Gerd Hoffmann <kraxel@redhat.com>
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.
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
20 #undef DEBUG
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>
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>
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");
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");
56 static atomic_t pages_mapped = ATOMIC_INIT(0);
58 static int use_ptemod;
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;
68 struct unmap_notify {
69 int flags;
70 /* Address relative to the start of the grant_map */
71 int addr;
72 int event;
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 page **pages;
89 static int unmap_grant_pages(struct grant_map *map, int offset, int pages);
91 /* ------------------------------------------------------------------ */
93 static void gntdev_print_maps(struct gntdev_priv *priv,
94 char *text, int text_index)
96 #ifdef DEBUG
97 struct grant_map *map;
99 pr_debug("%s: maps list (priv %p)\n", __func__, priv);
100 list_for_each_entry(map, &priv->maps, next)
101 pr_debug(" index %2d, count %2d %s\n",
102 map->index, map->count,
103 map->index == text_index && text ? text : "");
104 #endif
107 static struct grant_map *gntdev_alloc_map(struct gntdev_priv *priv, int count)
109 struct grant_map *add;
110 int i;
112 add = kzalloc(sizeof(struct grant_map), GFP_KERNEL);
113 if (NULL == add)
114 return NULL;
116 add->grants = kzalloc(sizeof(add->grants[0]) * count, GFP_KERNEL);
117 add->map_ops = kzalloc(sizeof(add->map_ops[0]) * count, GFP_KERNEL);
118 add->unmap_ops = kzalloc(sizeof(add->unmap_ops[0]) * count, GFP_KERNEL);
119 add->pages = kzalloc(sizeof(add->pages[0]) * count, GFP_KERNEL);
120 if (NULL == add->grants ||
121 NULL == add->map_ops ||
122 NULL == add->unmap_ops ||
123 NULL == add->pages)
124 goto err;
126 if (alloc_xenballooned_pages(count, add->pages))
127 goto err;
129 for (i = 0; i < count; i++) {
130 add->map_ops[i].handle = -1;
131 add->unmap_ops[i].handle = -1;
134 add->index = 0;
135 add->count = count;
136 atomic_set(&add->users, 1);
138 return add;
140 err:
141 kfree(add->pages);
142 kfree(add->grants);
143 kfree(add->map_ops);
144 kfree(add->unmap_ops);
145 kfree(add);
146 return NULL;
149 static void gntdev_add_map(struct gntdev_priv *priv, struct grant_map *add)
151 struct grant_map *map;
153 list_for_each_entry(map, &priv->maps, next) {
154 if (add->index + add->count < map->index) {
155 list_add_tail(&add->next, &map->next);
156 goto done;
158 add->index = map->index + map->count;
160 list_add_tail(&add->next, &priv->maps);
162 done:
163 gntdev_print_maps(priv, "[new]", add->index);
166 static struct grant_map *gntdev_find_map_index(struct gntdev_priv *priv,
167 int index, int count)
169 struct grant_map *map;
171 list_for_each_entry(map, &priv->maps, next) {
172 if (map->index != index)
173 continue;
174 if (count && map->count != count)
175 continue;
176 return map;
178 return NULL;
181 static void gntdev_put_map(struct grant_map *map)
183 if (!map)
184 return;
186 if (!atomic_dec_and_test(&map->users))
187 return;
189 atomic_sub(map->count, &pages_mapped);
191 if (map->notify.flags & UNMAP_NOTIFY_SEND_EVENT) {
192 notify_remote_via_evtchn(map->notify.event);
195 if (map->pages) {
196 if (!use_ptemod)
197 unmap_grant_pages(map, 0, map->count);
199 free_xenballooned_pages(map->count, map->pages);
201 kfree(map->pages);
202 kfree(map->grants);
203 kfree(map->map_ops);
204 kfree(map->unmap_ops);
205 kfree(map);
208 /* ------------------------------------------------------------------ */
210 static int find_grant_ptes(pte_t *pte, pgtable_t token,
211 unsigned long addr, void *data)
213 struct grant_map *map = data;
214 unsigned int pgnr = (addr - map->vma->vm_start) >> PAGE_SHIFT;
215 int flags = map->flags | GNTMAP_application_map | GNTMAP_contains_pte;
216 u64 pte_maddr;
218 BUG_ON(pgnr >= map->count);
219 pte_maddr = arbitrary_virt_to_machine(pte).maddr;
221 gnttab_set_map_op(&map->map_ops[pgnr], pte_maddr, flags,
222 map->grants[pgnr].ref,
223 map->grants[pgnr].domid);
224 gnttab_set_unmap_op(&map->unmap_ops[pgnr], pte_maddr, flags,
225 -1 /* handle */);
226 return 0;
229 static int map_grant_pages(struct grant_map *map)
231 int i, err = 0;
233 if (!use_ptemod) {
234 /* Note: it could already be mapped */
235 if (map->map_ops[0].handle != -1)
236 return 0;
237 for (i = 0; i < map->count; i++) {
238 unsigned long addr = (unsigned long)
239 pfn_to_kaddr(page_to_pfn(map->pages[i]));
240 gnttab_set_map_op(&map->map_ops[i], addr, map->flags,
241 map->grants[i].ref,
242 map->grants[i].domid);
243 gnttab_set_unmap_op(&map->unmap_ops[i], addr,
244 map->flags, -1 /* handle */);
248 pr_debug("map %d+%d\n", map->index, map->count);
249 err = gnttab_map_refs(map->map_ops, map->pages, map->count);
250 if (err)
251 return err;
253 for (i = 0; i < map->count; i++) {
254 if (map->map_ops[i].status)
255 err = -EINVAL;
256 else {
257 BUG_ON(map->map_ops[i].handle == -1);
258 map->unmap_ops[i].handle = map->map_ops[i].handle;
259 pr_debug("map handle=%d\n", map->map_ops[i].handle);
262 return err;
265 static int __unmap_grant_pages(struct grant_map *map, int offset, int pages)
267 int i, err = 0;
269 if (map->notify.flags & UNMAP_NOTIFY_CLEAR_BYTE) {
270 int pgno = (map->notify.addr >> PAGE_SHIFT);
271 if (pgno >= offset && pgno < offset + pages && use_ptemod) {
272 void __user *tmp = (void __user *)
273 map->vma->vm_start + map->notify.addr;
274 err = copy_to_user(tmp, &err, 1);
275 if (err)
276 return -EFAULT;
277 map->notify.flags &= ~UNMAP_NOTIFY_CLEAR_BYTE;
278 } else if (pgno >= offset && pgno < offset + pages) {
279 uint8_t *tmp = kmap(map->pages[pgno]);
280 tmp[map->notify.addr & (PAGE_SIZE-1)] = 0;
281 kunmap(map->pages[pgno]);
282 map->notify.flags &= ~UNMAP_NOTIFY_CLEAR_BYTE;
286 err = gnttab_unmap_refs(map->unmap_ops + offset, map->pages + offset, pages);
287 if (err)
288 return err;
290 for (i = 0; i < pages; i++) {
291 if (map->unmap_ops[offset+i].status)
292 err = -EINVAL;
293 pr_debug("unmap handle=%d st=%d\n",
294 map->unmap_ops[offset+i].handle,
295 map->unmap_ops[offset+i].status);
296 map->unmap_ops[offset+i].handle = -1;
298 return err;
301 static int unmap_grant_pages(struct grant_map *map, int offset, int pages)
303 int range, err = 0;
305 pr_debug("unmap %d+%d [%d+%d]\n", map->index, map->count, offset, pages);
307 /* It is possible the requested range will have a "hole" where we
308 * already unmapped some of the grants. Only unmap valid ranges.
310 while (pages && !err) {
311 while (pages && map->unmap_ops[offset].handle == -1) {
312 offset++;
313 pages--;
315 range = 0;
316 while (range < pages) {
317 if (map->unmap_ops[offset+range].handle == -1) {
318 range--;
319 break;
321 range++;
323 err = __unmap_grant_pages(map, offset, range);
324 offset += range;
325 pages -= range;
328 return err;
331 /* ------------------------------------------------------------------ */
333 static void gntdev_vma_close(struct vm_area_struct *vma)
335 struct grant_map *map = vma->vm_private_data;
337 pr_debug("close %p\n", vma);
338 map->vma = NULL;
339 vma->vm_private_data = NULL;
340 gntdev_put_map(map);
343 static struct vm_operations_struct gntdev_vmops = {
344 .close = gntdev_vma_close,
347 /* ------------------------------------------------------------------ */
349 static void mn_invl_range_start(struct mmu_notifier *mn,
350 struct mm_struct *mm,
351 unsigned long start, unsigned long end)
353 struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
354 struct grant_map *map;
355 unsigned long mstart, mend;
356 int err;
358 spin_lock(&priv->lock);
359 list_for_each_entry(map, &priv->maps, next) {
360 if (!map->vma)
361 continue;
362 if (map->vma->vm_start >= end)
363 continue;
364 if (map->vma->vm_end <= start)
365 continue;
366 mstart = max(start, map->vma->vm_start);
367 mend = min(end, map->vma->vm_end);
368 pr_debug("map %d+%d (%lx %lx), range %lx %lx, mrange %lx %lx\n",
369 map->index, map->count,
370 map->vma->vm_start, map->vma->vm_end,
371 start, end, mstart, mend);
372 err = unmap_grant_pages(map,
373 (mstart - map->vma->vm_start) >> PAGE_SHIFT,
374 (mend - mstart) >> PAGE_SHIFT);
375 WARN_ON(err);
377 spin_unlock(&priv->lock);
380 static void mn_invl_page(struct mmu_notifier *mn,
381 struct mm_struct *mm,
382 unsigned long address)
384 mn_invl_range_start(mn, mm, address, address + PAGE_SIZE);
387 static void mn_release(struct mmu_notifier *mn,
388 struct mm_struct *mm)
390 struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
391 struct grant_map *map;
392 int err;
394 spin_lock(&priv->lock);
395 list_for_each_entry(map, &priv->maps, next) {
396 if (!map->vma)
397 continue;
398 pr_debug("map %d+%d (%lx %lx)\n",
399 map->index, map->count,
400 map->vma->vm_start, map->vma->vm_end);
401 err = unmap_grant_pages(map, /* offset */ 0, map->count);
402 WARN_ON(err);
404 spin_unlock(&priv->lock);
407 struct mmu_notifier_ops gntdev_mmu_ops = {
408 .release = mn_release,
409 .invalidate_page = mn_invl_page,
410 .invalidate_range_start = mn_invl_range_start,
413 /* ------------------------------------------------------------------ */
415 static int gntdev_open(struct inode *inode, struct file *flip)
417 struct gntdev_priv *priv;
418 int ret = 0;
420 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
421 if (!priv)
422 return -ENOMEM;
424 INIT_LIST_HEAD(&priv->maps);
425 spin_lock_init(&priv->lock);
427 if (use_ptemod) {
428 priv->mm = get_task_mm(current);
429 if (!priv->mm) {
430 kfree(priv);
431 return -ENOMEM;
433 priv->mn.ops = &gntdev_mmu_ops;
434 ret = mmu_notifier_register(&priv->mn, priv->mm);
435 mmput(priv->mm);
438 if (ret) {
439 kfree(priv);
440 return ret;
443 flip->private_data = priv;
444 pr_debug("priv %p\n", priv);
446 return 0;
449 static int gntdev_release(struct inode *inode, struct file *flip)
451 struct gntdev_priv *priv = flip->private_data;
452 struct grant_map *map;
454 pr_debug("priv %p\n", priv);
456 spin_lock(&priv->lock);
457 while (!list_empty(&priv->maps)) {
458 map = list_entry(priv->maps.next, struct grant_map, next);
459 list_del(&map->next);
460 gntdev_put_map(map);
462 spin_unlock(&priv->lock);
464 if (use_ptemod)
465 mmu_notifier_unregister(&priv->mn, priv->mm);
466 kfree(priv);
467 return 0;
470 static long gntdev_ioctl_map_grant_ref(struct gntdev_priv *priv,
471 struct ioctl_gntdev_map_grant_ref __user *u)
473 struct ioctl_gntdev_map_grant_ref op;
474 struct grant_map *map;
475 int err;
477 if (copy_from_user(&op, u, sizeof(op)) != 0)
478 return -EFAULT;
479 pr_debug("priv %p, add %d\n", priv, op.count);
480 if (unlikely(op.count <= 0))
481 return -EINVAL;
483 err = -ENOMEM;
484 map = gntdev_alloc_map(priv, op.count);
485 if (!map)
486 return err;
488 if (unlikely(atomic_add_return(op.count, &pages_mapped) > limit)) {
489 pr_debug("can't map: over limit\n");
490 gntdev_put_map(map);
491 return err;
494 if (copy_from_user(map->grants, &u->refs,
495 sizeof(map->grants[0]) * op.count) != 0) {
496 gntdev_put_map(map);
497 return err;
500 spin_lock(&priv->lock);
501 gntdev_add_map(priv, map);
502 op.index = map->index << PAGE_SHIFT;
503 spin_unlock(&priv->lock);
505 if (copy_to_user(u, &op, sizeof(op)) != 0)
506 return -EFAULT;
508 return 0;
511 static long gntdev_ioctl_unmap_grant_ref(struct gntdev_priv *priv,
512 struct ioctl_gntdev_unmap_grant_ref __user *u)
514 struct ioctl_gntdev_unmap_grant_ref op;
515 struct grant_map *map;
516 int err = -ENOENT;
518 if (copy_from_user(&op, u, sizeof(op)) != 0)
519 return -EFAULT;
520 pr_debug("priv %p, del %d+%d\n", priv, (int)op.index, (int)op.count);
522 spin_lock(&priv->lock);
523 map = gntdev_find_map_index(priv, op.index >> PAGE_SHIFT, op.count);
524 if (map) {
525 list_del(&map->next);
526 gntdev_put_map(map);
527 err = 0;
529 spin_unlock(&priv->lock);
530 return err;
533 static long gntdev_ioctl_get_offset_for_vaddr(struct gntdev_priv *priv,
534 struct ioctl_gntdev_get_offset_for_vaddr __user *u)
536 struct ioctl_gntdev_get_offset_for_vaddr op;
537 struct vm_area_struct *vma;
538 struct grant_map *map;
540 if (copy_from_user(&op, u, sizeof(op)) != 0)
541 return -EFAULT;
542 pr_debug("priv %p, offset for vaddr %lx\n", priv, (unsigned long)op.vaddr);
544 vma = find_vma(current->mm, op.vaddr);
545 if (!vma || vma->vm_ops != &gntdev_vmops)
546 return -EINVAL;
548 map = vma->vm_private_data;
549 if (!map)
550 return -EINVAL;
552 op.offset = map->index << PAGE_SHIFT;
553 op.count = map->count;
555 if (copy_to_user(u, &op, sizeof(op)) != 0)
556 return -EFAULT;
557 return 0;
560 static long gntdev_ioctl_notify(struct gntdev_priv *priv, void __user *u)
562 struct ioctl_gntdev_unmap_notify op;
563 struct grant_map *map;
564 int rc;
566 if (copy_from_user(&op, u, sizeof(op)))
567 return -EFAULT;
569 if (op.action & ~(UNMAP_NOTIFY_CLEAR_BYTE|UNMAP_NOTIFY_SEND_EVENT))
570 return -EINVAL;
572 spin_lock(&priv->lock);
574 list_for_each_entry(map, &priv->maps, next) {
575 uint64_t begin = map->index << PAGE_SHIFT;
576 uint64_t end = (map->index + map->count) << PAGE_SHIFT;
577 if (op.index >= begin && op.index < end)
578 goto found;
580 rc = -ENOENT;
581 goto unlock_out;
583 found:
584 if ((op.action & UNMAP_NOTIFY_CLEAR_BYTE) &&
585 (map->flags & GNTMAP_readonly)) {
586 rc = -EINVAL;
587 goto unlock_out;
590 map->notify.flags = op.action;
591 map->notify.addr = op.index - (map->index << PAGE_SHIFT);
592 map->notify.event = op.event_channel_port;
593 rc = 0;
594 unlock_out:
595 spin_unlock(&priv->lock);
596 return rc;
599 static long gntdev_ioctl(struct file *flip,
600 unsigned int cmd, unsigned long arg)
602 struct gntdev_priv *priv = flip->private_data;
603 void __user *ptr = (void __user *)arg;
605 switch (cmd) {
606 case IOCTL_GNTDEV_MAP_GRANT_REF:
607 return gntdev_ioctl_map_grant_ref(priv, ptr);
609 case IOCTL_GNTDEV_UNMAP_GRANT_REF:
610 return gntdev_ioctl_unmap_grant_ref(priv, ptr);
612 case IOCTL_GNTDEV_GET_OFFSET_FOR_VADDR:
613 return gntdev_ioctl_get_offset_for_vaddr(priv, ptr);
615 case IOCTL_GNTDEV_SET_UNMAP_NOTIFY:
616 return gntdev_ioctl_notify(priv, ptr);
618 default:
619 pr_debug("priv %p, unknown cmd %x\n", priv, cmd);
620 return -ENOIOCTLCMD;
623 return 0;
626 static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma)
628 struct gntdev_priv *priv = flip->private_data;
629 int index = vma->vm_pgoff;
630 int count = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
631 struct grant_map *map;
632 int i, err = -EINVAL;
634 if ((vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_SHARED))
635 return -EINVAL;
637 pr_debug("map %d+%d at %lx (pgoff %lx)\n",
638 index, count, vma->vm_start, vma->vm_pgoff);
640 spin_lock(&priv->lock);
641 map = gntdev_find_map_index(priv, index, count);
642 if (!map)
643 goto unlock_out;
644 if (use_ptemod && map->vma)
645 goto unlock_out;
646 if (use_ptemod && priv->mm != vma->vm_mm) {
647 printk(KERN_WARNING "Huh? Other mm?\n");
648 goto unlock_out;
651 atomic_inc(&map->users);
653 vma->vm_ops = &gntdev_vmops;
655 vma->vm_flags |= VM_RESERVED|VM_DONTCOPY|VM_DONTEXPAND|VM_PFNMAP;
657 vma->vm_private_data = map;
659 if (use_ptemod)
660 map->vma = vma;
662 if (map->flags) {
663 if ((vma->vm_flags & VM_WRITE) &&
664 (map->flags & GNTMAP_readonly))
665 goto out_unlock_put;
666 } else {
667 map->flags = GNTMAP_host_map;
668 if (!(vma->vm_flags & VM_WRITE))
669 map->flags |= GNTMAP_readonly;
672 spin_unlock(&priv->lock);
674 if (use_ptemod) {
675 err = apply_to_page_range(vma->vm_mm, vma->vm_start,
676 vma->vm_end - vma->vm_start,
677 find_grant_ptes, map);
678 if (err) {
679 printk(KERN_WARNING "find_grant_ptes() failure.\n");
680 goto out_put_map;
684 err = map_grant_pages(map);
685 if (err)
686 goto out_put_map;
688 if (!use_ptemod) {
689 for (i = 0; i < count; i++) {
690 err = vm_insert_page(vma, vma->vm_start + i*PAGE_SIZE,
691 map->pages[i]);
692 if (err)
693 goto out_put_map;
697 return 0;
699 unlock_out:
700 spin_unlock(&priv->lock);
701 return err;
703 out_unlock_put:
704 spin_unlock(&priv->lock);
705 out_put_map:
706 if (use_ptemod)
707 map->vma = NULL;
708 gntdev_put_map(map);
709 return err;
712 static const struct file_operations gntdev_fops = {
713 .owner = THIS_MODULE,
714 .open = gntdev_open,
715 .release = gntdev_release,
716 .mmap = gntdev_mmap,
717 .unlocked_ioctl = gntdev_ioctl
720 static struct miscdevice gntdev_miscdev = {
721 .minor = MISC_DYNAMIC_MINOR,
722 .name = "xen/gntdev",
723 .fops = &gntdev_fops,
726 /* ------------------------------------------------------------------ */
728 static int __init gntdev_init(void)
730 int err;
732 if (!xen_domain())
733 return -ENODEV;
735 use_ptemod = xen_pv_domain();
737 err = misc_register(&gntdev_miscdev);
738 if (err != 0) {
739 printk(KERN_ERR "Could not register gntdev device\n");
740 return err;
742 return 0;
745 static void __exit gntdev_exit(void)
747 misc_deregister(&gntdev_miscdev);
750 module_init(gntdev_init);
751 module_exit(gntdev_exit);
753 /* ------------------------------------------------------------------ */