1 /******************************************************************************
4 * Device for accessing (in user-space) pages that have been granted by other
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
22 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
24 #include <linux/module.h>
25 #include <linux/kernel.h>
26 #include <linux/init.h>
27 #include <linux/miscdevice.h>
30 #include <linux/mman.h>
31 #include <linux/mmu_notifier.h>
32 #include <linux/types.h>
33 #include <linux/uaccess.h>
34 #include <linux/sched.h>
35 #include <linux/sched/mm.h>
36 #include <linux/spinlock.h>
37 #include <linux/slab.h>
38 #include <linux/highmem.h>
39 #include <linux/refcount.h>
42 #include <xen/grant_table.h>
43 #include <xen/balloon.h>
44 #include <xen/gntdev.h>
45 #include <xen/events.h>
47 #include <asm/xen/hypervisor.h>
48 #include <asm/xen/hypercall.h>
50 MODULE_LICENSE("GPL");
51 MODULE_AUTHOR("Derek G. Murray <Derek.Murray@cl.cam.ac.uk>, "
52 "Gerd Hoffmann <kraxel@redhat.com>");
53 MODULE_DESCRIPTION("User-space granted page access driver");
55 static int limit = 1024*1024;
56 module_param(limit, int, 0644);
57 MODULE_PARM_DESC(limit, "Maximum number of grants that may be mapped by "
60 static atomic_t pages_mapped = ATOMIC_INIT(0);
62 static int use_ptemod;
63 #define populate_freeable_maps use_ptemod
66 /* maps with visible offsets in the file descriptor */
67 struct list_head maps;
68 /* maps that are not visible; will be freed on munmap.
69 * Only populated if populate_freeable_maps == 1 */
70 struct list_head freeable_maps;
71 /* lock protects maps and freeable_maps */
74 struct mmu_notifier mn;
79 /* Address relative to the start of the grant_map */
85 struct list_head next;
86 struct vm_area_struct *vma;
91 struct unmap_notify notify;
92 struct ioctl_gntdev_grant_ref *grants;
93 struct gnttab_map_grant_ref *map_ops;
94 struct gnttab_unmap_grant_ref *unmap_ops;
95 struct gnttab_map_grant_ref *kmap_ops;
96 struct gnttab_unmap_grant_ref *kunmap_ops;
98 unsigned long pages_vm_start;
101 static int unmap_grant_pages(struct grant_map *map, int offset, int pages);
103 /* ------------------------------------------------------------------ */
105 static void gntdev_print_maps(struct gntdev_priv *priv,
106 char *text, int text_index)
109 struct grant_map *map;
111 pr_debug("%s: maps list (priv %p)\n", __func__, priv);
112 list_for_each_entry(map, &priv->maps, next)
113 pr_debug(" index %2d, count %2d %s\n",
114 map->index, map->count,
115 map->index == text_index && text ? text : "");
119 static void gntdev_free_map(struct grant_map *map)
125 gnttab_free_pages(map->count, map->pages);
129 kfree(map->unmap_ops);
130 kfree(map->kmap_ops);
131 kfree(map->kunmap_ops);
135 static struct grant_map *gntdev_alloc_map(struct gntdev_priv *priv, int count)
137 struct grant_map *add;
140 add = kzalloc(sizeof(struct grant_map), GFP_KERNEL);
144 add->grants = kcalloc(count, sizeof(add->grants[0]), GFP_KERNEL);
145 add->map_ops = kcalloc(count, sizeof(add->map_ops[0]), GFP_KERNEL);
146 add->unmap_ops = kcalloc(count, sizeof(add->unmap_ops[0]), GFP_KERNEL);
147 add->kmap_ops = kcalloc(count, sizeof(add->kmap_ops[0]), GFP_KERNEL);
148 add->kunmap_ops = kcalloc(count, sizeof(add->kunmap_ops[0]), GFP_KERNEL);
149 add->pages = kcalloc(count, sizeof(add->pages[0]), GFP_KERNEL);
150 if (NULL == add->grants ||
151 NULL == add->map_ops ||
152 NULL == add->unmap_ops ||
153 NULL == add->kmap_ops ||
154 NULL == add->kunmap_ops ||
158 if (gnttab_alloc_pages(count, add->pages))
161 for (i = 0; i < count; i++) {
162 add->map_ops[i].handle = -1;
163 add->unmap_ops[i].handle = -1;
164 add->kmap_ops[i].handle = -1;
165 add->kunmap_ops[i].handle = -1;
170 refcount_set(&add->users, 1);
175 gntdev_free_map(add);
179 static void gntdev_add_map(struct gntdev_priv *priv, struct grant_map *add)
181 struct grant_map *map;
183 list_for_each_entry(map, &priv->maps, next) {
184 if (add->index + add->count < map->index) {
185 list_add_tail(&add->next, &map->next);
188 add->index = map->index + map->count;
190 list_add_tail(&add->next, &priv->maps);
193 gntdev_print_maps(priv, "[new]", add->index);
196 static struct grant_map *gntdev_find_map_index(struct gntdev_priv *priv,
197 int index, int count)
199 struct grant_map *map;
201 list_for_each_entry(map, &priv->maps, next) {
202 if (map->index != index)
204 if (count && map->count != count)
211 static void gntdev_put_map(struct gntdev_priv *priv, struct grant_map *map)
216 if (!refcount_dec_and_test(&map->users))
219 atomic_sub(map->count, &pages_mapped);
221 if (map->notify.flags & UNMAP_NOTIFY_SEND_EVENT) {
222 notify_remote_via_evtchn(map->notify.event);
223 evtchn_put(map->notify.event);
226 if (populate_freeable_maps && priv) {
227 mutex_lock(&priv->lock);
228 list_del(&map->next);
229 mutex_unlock(&priv->lock);
232 if (map->pages && !use_ptemod)
233 unmap_grant_pages(map, 0, map->count);
234 gntdev_free_map(map);
237 /* ------------------------------------------------------------------ */
239 static int find_grant_ptes(pte_t *pte, pgtable_t token,
240 unsigned long addr, void *data)
242 struct grant_map *map = data;
243 unsigned int pgnr = (addr - map->vma->vm_start) >> PAGE_SHIFT;
244 int flags = map->flags | GNTMAP_application_map | GNTMAP_contains_pte;
247 BUG_ON(pgnr >= map->count);
248 pte_maddr = arbitrary_virt_to_machine(pte).maddr;
251 * Set the PTE as special to force get_user_pages_fast() fall
252 * back to the slow path. If this is not supported as part of
253 * the grant map, it will be done afterwards.
255 if (xen_feature(XENFEAT_gnttab_map_avail_bits))
256 flags |= (1 << _GNTMAP_guest_avail0);
258 gnttab_set_map_op(&map->map_ops[pgnr], pte_maddr, flags,
259 map->grants[pgnr].ref,
260 map->grants[pgnr].domid);
261 gnttab_set_unmap_op(&map->unmap_ops[pgnr], pte_maddr, flags,
267 static int set_grant_ptes_as_special(pte_t *pte, pgtable_t token,
268 unsigned long addr, void *data)
270 set_pte_at(current->mm, addr, pte, pte_mkspecial(*pte));
275 static int map_grant_pages(struct grant_map *map)
280 /* Note: it could already be mapped */
281 if (map->map_ops[0].handle != -1)
283 for (i = 0; i < map->count; i++) {
284 unsigned long addr = (unsigned long)
285 pfn_to_kaddr(page_to_pfn(map->pages[i]));
286 gnttab_set_map_op(&map->map_ops[i], addr, map->flags,
288 map->grants[i].domid);
289 gnttab_set_unmap_op(&map->unmap_ops[i], addr,
290 map->flags, -1 /* handle */);
294 * Setup the map_ops corresponding to the pte entries pointing
295 * to the kernel linear addresses of the struct pages.
296 * These ptes are completely different from the user ptes dealt
297 * with find_grant_ptes.
298 * Note that GNTMAP_device_map isn't needed here: The
299 * dev_bus_addr output field gets consumed only from ->map_ops,
300 * and by not requesting it when mapping we also avoid needing
301 * to mirror dev_bus_addr into ->unmap_ops (and holding an extra
302 * reference to the page in the hypervisor).
304 unsigned int flags = (map->flags & ~GNTMAP_device_map) |
307 for (i = 0; i < map->count; i++) {
308 unsigned long address = (unsigned long)
309 pfn_to_kaddr(page_to_pfn(map->pages[i]));
310 BUG_ON(PageHighMem(map->pages[i]));
312 gnttab_set_map_op(&map->kmap_ops[i], address, flags,
314 map->grants[i].domid);
315 gnttab_set_unmap_op(&map->kunmap_ops[i], address,
320 pr_debug("map %d+%d\n", map->index, map->count);
321 err = gnttab_map_refs(map->map_ops, use_ptemod ? map->kmap_ops : NULL,
322 map->pages, map->count);
324 for (i = 0; i < map->count; i++) {
325 if (map->map_ops[i].status == GNTST_okay)
326 map->unmap_ops[i].handle = map->map_ops[i].handle;
330 if (map->flags & GNTMAP_device_map)
331 map->unmap_ops[i].dev_bus_addr = map->map_ops[i].dev_bus_addr;
334 if (map->kmap_ops[i].status == GNTST_okay)
335 map->kunmap_ops[i].handle = map->kmap_ops[i].handle;
343 static int __unmap_grant_pages(struct grant_map *map, int offset, int pages)
346 struct gntab_unmap_queue_data unmap_data;
348 if (map->notify.flags & UNMAP_NOTIFY_CLEAR_BYTE) {
349 int pgno = (map->notify.addr >> PAGE_SHIFT);
350 if (pgno >= offset && pgno < offset + pages) {
351 /* No need for kmap, pages are in lowmem */
352 uint8_t *tmp = pfn_to_kaddr(page_to_pfn(map->pages[pgno]));
353 tmp[map->notify.addr & (PAGE_SIZE-1)] = 0;
354 map->notify.flags &= ~UNMAP_NOTIFY_CLEAR_BYTE;
358 unmap_data.unmap_ops = map->unmap_ops + offset;
359 unmap_data.kunmap_ops = use_ptemod ? map->kunmap_ops + offset : NULL;
360 unmap_data.pages = map->pages + offset;
361 unmap_data.count = pages;
363 err = gnttab_unmap_refs_sync(&unmap_data);
367 for (i = 0; i < pages; i++) {
368 if (map->unmap_ops[offset+i].status)
370 pr_debug("unmap handle=%d st=%d\n",
371 map->unmap_ops[offset+i].handle,
372 map->unmap_ops[offset+i].status);
373 map->unmap_ops[offset+i].handle = -1;
378 static int unmap_grant_pages(struct grant_map *map, int offset, int pages)
382 pr_debug("unmap %d+%d [%d+%d]\n", map->index, map->count, offset, pages);
384 /* It is possible the requested range will have a "hole" where we
385 * already unmapped some of the grants. Only unmap valid ranges.
387 while (pages && !err) {
388 while (pages && map->unmap_ops[offset].handle == -1) {
393 while (range < pages) {
394 if (map->unmap_ops[offset+range].handle == -1)
398 err = __unmap_grant_pages(map, offset, range);
406 /* ------------------------------------------------------------------ */
408 static void gntdev_vma_open(struct vm_area_struct *vma)
410 struct grant_map *map = vma->vm_private_data;
412 pr_debug("gntdev_vma_open %p\n", vma);
413 refcount_inc(&map->users);
416 static void gntdev_vma_close(struct vm_area_struct *vma)
418 struct grant_map *map = vma->vm_private_data;
419 struct file *file = vma->vm_file;
420 struct gntdev_priv *priv = file->private_data;
422 pr_debug("gntdev_vma_close %p\n", vma);
424 /* It is possible that an mmu notifier could be running
425 * concurrently, so take priv->lock to ensure that the vma won't
426 * vanishing during the unmap_grant_pages call, since we will
427 * spin here until that completes. Such a concurrent call will
428 * not do any unmapping, since that has been done prior to
429 * closing the vma, but it may still iterate the unmap_ops list.
431 mutex_lock(&priv->lock);
433 mutex_unlock(&priv->lock);
435 vma->vm_private_data = NULL;
436 gntdev_put_map(priv, map);
439 static struct page *gntdev_vma_find_special_page(struct vm_area_struct *vma,
442 struct grant_map *map = vma->vm_private_data;
444 return map->pages[(addr - map->pages_vm_start) >> PAGE_SHIFT];
447 static const struct vm_operations_struct gntdev_vmops = {
448 .open = gntdev_vma_open,
449 .close = gntdev_vma_close,
450 .find_special_page = gntdev_vma_find_special_page,
453 /* ------------------------------------------------------------------ */
455 static void unmap_if_in_range(struct grant_map *map,
456 unsigned long start, unsigned long end)
458 unsigned long mstart, mend;
463 if (map->vma->vm_start >= end)
465 if (map->vma->vm_end <= start)
467 mstart = max(start, map->vma->vm_start);
468 mend = min(end, map->vma->vm_end);
469 pr_debug("map %d+%d (%lx %lx), range %lx %lx, mrange %lx %lx\n",
470 map->index, map->count,
471 map->vma->vm_start, map->vma->vm_end,
472 start, end, mstart, mend);
473 err = unmap_grant_pages(map,
474 (mstart - map->vma->vm_start) >> PAGE_SHIFT,
475 (mend - mstart) >> PAGE_SHIFT);
479 static void mn_invl_range_start(struct mmu_notifier *mn,
480 struct mm_struct *mm,
481 unsigned long start, unsigned long end)
483 struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
484 struct grant_map *map;
486 mutex_lock(&priv->lock);
487 list_for_each_entry(map, &priv->maps, next) {
488 unmap_if_in_range(map, start, end);
490 list_for_each_entry(map, &priv->freeable_maps, next) {
491 unmap_if_in_range(map, start, end);
493 mutex_unlock(&priv->lock);
496 static void mn_release(struct mmu_notifier *mn,
497 struct mm_struct *mm)
499 struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
500 struct grant_map *map;
503 mutex_lock(&priv->lock);
504 list_for_each_entry(map, &priv->maps, next) {
507 pr_debug("map %d+%d (%lx %lx)\n",
508 map->index, map->count,
509 map->vma->vm_start, map->vma->vm_end);
510 err = unmap_grant_pages(map, /* offset */ 0, map->count);
513 list_for_each_entry(map, &priv->freeable_maps, next) {
516 pr_debug("map %d+%d (%lx %lx)\n",
517 map->index, map->count,
518 map->vma->vm_start, map->vma->vm_end);
519 err = unmap_grant_pages(map, /* offset */ 0, map->count);
522 mutex_unlock(&priv->lock);
525 static const struct mmu_notifier_ops gntdev_mmu_ops = {
526 .release = mn_release,
527 .invalidate_range_start = mn_invl_range_start,
530 /* ------------------------------------------------------------------ */
532 static int gntdev_open(struct inode *inode, struct file *flip)
534 struct gntdev_priv *priv;
537 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
541 INIT_LIST_HEAD(&priv->maps);
542 INIT_LIST_HEAD(&priv->freeable_maps);
543 mutex_init(&priv->lock);
546 priv->mm = get_task_mm(current);
551 priv->mn.ops = &gntdev_mmu_ops;
552 ret = mmu_notifier_register(&priv->mn, priv->mm);
561 flip->private_data = priv;
562 pr_debug("priv %p\n", priv);
567 static int gntdev_release(struct inode *inode, struct file *flip)
569 struct gntdev_priv *priv = flip->private_data;
570 struct grant_map *map;
572 pr_debug("priv %p\n", priv);
574 mutex_lock(&priv->lock);
575 while (!list_empty(&priv->maps)) {
576 map = list_entry(priv->maps.next, struct grant_map, next);
577 list_del(&map->next);
578 gntdev_put_map(NULL /* already removed */, map);
580 WARN_ON(!list_empty(&priv->freeable_maps));
581 mutex_unlock(&priv->lock);
584 mmu_notifier_unregister(&priv->mn, priv->mm);
589 static long gntdev_ioctl_map_grant_ref(struct gntdev_priv *priv,
590 struct ioctl_gntdev_map_grant_ref __user *u)
592 struct ioctl_gntdev_map_grant_ref op;
593 struct grant_map *map;
596 if (copy_from_user(&op, u, sizeof(op)) != 0)
598 pr_debug("priv %p, add %d\n", priv, op.count);
599 if (unlikely(op.count <= 0))
603 map = gntdev_alloc_map(priv, op.count);
607 if (unlikely(atomic_add_return(op.count, &pages_mapped) > limit)) {
608 pr_debug("can't map: over limit\n");
609 gntdev_put_map(NULL, map);
613 if (copy_from_user(map->grants, &u->refs,
614 sizeof(map->grants[0]) * op.count) != 0) {
615 gntdev_put_map(NULL, map);
619 mutex_lock(&priv->lock);
620 gntdev_add_map(priv, map);
621 op.index = map->index << PAGE_SHIFT;
622 mutex_unlock(&priv->lock);
624 if (copy_to_user(u, &op, sizeof(op)) != 0)
630 static long gntdev_ioctl_unmap_grant_ref(struct gntdev_priv *priv,
631 struct ioctl_gntdev_unmap_grant_ref __user *u)
633 struct ioctl_gntdev_unmap_grant_ref op;
634 struct grant_map *map;
637 if (copy_from_user(&op, u, sizeof(op)) != 0)
639 pr_debug("priv %p, del %d+%d\n", priv, (int)op.index, (int)op.count);
641 mutex_lock(&priv->lock);
642 map = gntdev_find_map_index(priv, op.index >> PAGE_SHIFT, op.count);
644 list_del(&map->next);
645 if (populate_freeable_maps)
646 list_add_tail(&map->next, &priv->freeable_maps);
649 mutex_unlock(&priv->lock);
651 gntdev_put_map(priv, map);
655 static long gntdev_ioctl_get_offset_for_vaddr(struct gntdev_priv *priv,
656 struct ioctl_gntdev_get_offset_for_vaddr __user *u)
658 struct ioctl_gntdev_get_offset_for_vaddr op;
659 struct vm_area_struct *vma;
660 struct grant_map *map;
663 if (copy_from_user(&op, u, sizeof(op)) != 0)
665 pr_debug("priv %p, offset for vaddr %lx\n", priv, (unsigned long)op.vaddr);
667 down_read(¤t->mm->mmap_sem);
668 vma = find_vma(current->mm, op.vaddr);
669 if (!vma || vma->vm_ops != &gntdev_vmops)
672 map = vma->vm_private_data;
676 op.offset = map->index << PAGE_SHIFT;
677 op.count = map->count;
681 up_read(¤t->mm->mmap_sem);
683 if (rv == 0 && copy_to_user(u, &op, sizeof(op)) != 0)
688 static long gntdev_ioctl_notify(struct gntdev_priv *priv, void __user *u)
690 struct ioctl_gntdev_unmap_notify op;
691 struct grant_map *map;
694 unsigned int out_event;
696 if (copy_from_user(&op, u, sizeof(op)))
699 if (op.action & ~(UNMAP_NOTIFY_CLEAR_BYTE|UNMAP_NOTIFY_SEND_EVENT))
702 /* We need to grab a reference to the event channel we are going to use
703 * to send the notify before releasing the reference we may already have
704 * (if someone has called this ioctl twice). This is required so that
705 * it is possible to change the clear_byte part of the notification
706 * without disturbing the event channel part, which may now be the last
707 * reference to that event channel.
709 if (op.action & UNMAP_NOTIFY_SEND_EVENT) {
710 if (evtchn_get(op.event_channel_port))
714 out_flags = op.action;
715 out_event = op.event_channel_port;
717 mutex_lock(&priv->lock);
719 list_for_each_entry(map, &priv->maps, next) {
720 uint64_t begin = map->index << PAGE_SHIFT;
721 uint64_t end = (map->index + map->count) << PAGE_SHIFT;
722 if (op.index >= begin && op.index < end)
729 if ((op.action & UNMAP_NOTIFY_CLEAR_BYTE) &&
730 (map->flags & GNTMAP_readonly)) {
735 out_flags = map->notify.flags;
736 out_event = map->notify.event;
738 map->notify.flags = op.action;
739 map->notify.addr = op.index - (map->index << PAGE_SHIFT);
740 map->notify.event = op.event_channel_port;
745 mutex_unlock(&priv->lock);
747 /* Drop the reference to the event channel we did not save in the map */
748 if (out_flags & UNMAP_NOTIFY_SEND_EVENT)
749 evtchn_put(out_event);
754 #define GNTDEV_COPY_BATCH 16
756 struct gntdev_copy_batch {
757 struct gnttab_copy ops[GNTDEV_COPY_BATCH];
758 struct page *pages[GNTDEV_COPY_BATCH];
759 s16 __user *status[GNTDEV_COPY_BATCH];
761 unsigned int nr_pages;
764 static int gntdev_get_page(struct gntdev_copy_batch *batch, void __user *virt,
765 bool writeable, unsigned long *gfn)
767 unsigned long addr = (unsigned long)virt;
769 unsigned long xen_pfn;
772 ret = get_user_pages_fast(addr, 1, writeable, &page);
776 batch->pages[batch->nr_pages++] = page;
778 xen_pfn = page_to_xen_pfn(page) + XEN_PFN_DOWN(addr & ~PAGE_MASK);
779 *gfn = pfn_to_gfn(xen_pfn);
784 static void gntdev_put_pages(struct gntdev_copy_batch *batch)
788 for (i = 0; i < batch->nr_pages; i++)
789 put_page(batch->pages[i]);
793 static int gntdev_copy(struct gntdev_copy_batch *batch)
797 gnttab_batch_copy(batch->ops, batch->nr_ops);
798 gntdev_put_pages(batch);
801 * For each completed op, update the status if the op failed
802 * and all previous ops for the segment were successful.
804 for (i = 0; i < batch->nr_ops; i++) {
805 s16 status = batch->ops[i].status;
808 if (status == GNTST_okay)
811 if (__get_user(old_status, batch->status[i]))
814 if (old_status != GNTST_okay)
817 if (__put_user(status, batch->status[i]))
825 static int gntdev_grant_copy_seg(struct gntdev_copy_batch *batch,
826 struct gntdev_grant_copy_segment *seg,
832 * Disallow local -> local copies since there is only space in
833 * batch->pages for one page per-op and this would be a very
834 * expensive memcpy().
836 if (!(seg->flags & (GNTCOPY_source_gref | GNTCOPY_dest_gref)))
839 /* Can't cross page if source/dest is a grant ref. */
840 if (seg->flags & GNTCOPY_source_gref) {
841 if (seg->source.foreign.offset + seg->len > XEN_PAGE_SIZE)
844 if (seg->flags & GNTCOPY_dest_gref) {
845 if (seg->dest.foreign.offset + seg->len > XEN_PAGE_SIZE)
849 if (put_user(GNTST_okay, status))
852 while (copied < seg->len) {
853 struct gnttab_copy *op;
859 if (batch->nr_ops >= GNTDEV_COPY_BATCH) {
860 ret = gntdev_copy(batch);
865 len = seg->len - copied;
867 op = &batch->ops[batch->nr_ops];
870 if (seg->flags & GNTCOPY_source_gref) {
871 op->source.u.ref = seg->source.foreign.ref;
872 op->source.domid = seg->source.foreign.domid;
873 op->source.offset = seg->source.foreign.offset + copied;
874 op->flags |= GNTCOPY_source_gref;
876 virt = seg->source.virt + copied;
877 off = (unsigned long)virt & ~XEN_PAGE_MASK;
878 len = min(len, (size_t)XEN_PAGE_SIZE - off);
880 ret = gntdev_get_page(batch, virt, false, &gfn);
884 op->source.u.gmfn = gfn;
885 op->source.domid = DOMID_SELF;
886 op->source.offset = off;
889 if (seg->flags & GNTCOPY_dest_gref) {
890 op->dest.u.ref = seg->dest.foreign.ref;
891 op->dest.domid = seg->dest.foreign.domid;
892 op->dest.offset = seg->dest.foreign.offset + copied;
893 op->flags |= GNTCOPY_dest_gref;
895 virt = seg->dest.virt + copied;
896 off = (unsigned long)virt & ~XEN_PAGE_MASK;
897 len = min(len, (size_t)XEN_PAGE_SIZE - off);
899 ret = gntdev_get_page(batch, virt, true, &gfn);
903 op->dest.u.gmfn = gfn;
904 op->dest.domid = DOMID_SELF;
905 op->dest.offset = off;
911 batch->status[batch->nr_ops] = status;
918 static long gntdev_ioctl_grant_copy(struct gntdev_priv *priv, void __user *u)
920 struct ioctl_gntdev_grant_copy copy;
921 struct gntdev_copy_batch batch;
925 if (copy_from_user(©, u, sizeof(copy)))
931 for (i = 0; i < copy.count; i++) {
932 struct gntdev_grant_copy_segment seg;
934 if (copy_from_user(&seg, ©.segments[i], sizeof(seg))) {
939 ret = gntdev_grant_copy_seg(&batch, &seg, ©.segments[i].status);
946 ret = gntdev_copy(&batch);
950 gntdev_put_pages(&batch);
954 static long gntdev_ioctl(struct file *flip,
955 unsigned int cmd, unsigned long arg)
957 struct gntdev_priv *priv = flip->private_data;
958 void __user *ptr = (void __user *)arg;
961 case IOCTL_GNTDEV_MAP_GRANT_REF:
962 return gntdev_ioctl_map_grant_ref(priv, ptr);
964 case IOCTL_GNTDEV_UNMAP_GRANT_REF:
965 return gntdev_ioctl_unmap_grant_ref(priv, ptr);
967 case IOCTL_GNTDEV_GET_OFFSET_FOR_VADDR:
968 return gntdev_ioctl_get_offset_for_vaddr(priv, ptr);
970 case IOCTL_GNTDEV_SET_UNMAP_NOTIFY:
971 return gntdev_ioctl_notify(priv, ptr);
973 case IOCTL_GNTDEV_GRANT_COPY:
974 return gntdev_ioctl_grant_copy(priv, ptr);
977 pr_debug("priv %p, unknown cmd %x\n", priv, cmd);
984 static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma)
986 struct gntdev_priv *priv = flip->private_data;
987 int index = vma->vm_pgoff;
988 int count = vma_pages(vma);
989 struct grant_map *map;
990 int i, err = -EINVAL;
992 if ((vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_SHARED))
995 pr_debug("map %d+%d at %lx (pgoff %lx)\n",
996 index, count, vma->vm_start, vma->vm_pgoff);
998 mutex_lock(&priv->lock);
999 map = gntdev_find_map_index(priv, index, count);
1002 if (use_ptemod && map->vma)
1004 if (use_ptemod && priv->mm != vma->vm_mm) {
1005 pr_warn("Huh? Other mm?\n");
1009 refcount_inc(&map->users);
1011 vma->vm_ops = &gntdev_vmops;
1013 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP | VM_MIXEDMAP;
1016 vma->vm_flags |= VM_DONTCOPY;
1018 vma->vm_private_data = map;
1024 if ((vma->vm_flags & VM_WRITE) &&
1025 (map->flags & GNTMAP_readonly))
1026 goto out_unlock_put;
1028 map->flags = GNTMAP_host_map;
1029 if (!(vma->vm_flags & VM_WRITE))
1030 map->flags |= GNTMAP_readonly;
1033 mutex_unlock(&priv->lock);
1036 map->pages_vm_start = vma->vm_start;
1037 err = apply_to_page_range(vma->vm_mm, vma->vm_start,
1038 vma->vm_end - vma->vm_start,
1039 find_grant_ptes, map);
1041 pr_warn("find_grant_ptes() failure.\n");
1046 err = map_grant_pages(map);
1051 for (i = 0; i < count; i++) {
1052 err = vm_insert_page(vma, vma->vm_start + i*PAGE_SIZE,
1060 * If the PTEs were not made special by the grant map
1061 * hypercall, do so here.
1063 * This is racy since the mapping is already visible
1064 * to userspace but userspace should be well-behaved
1065 * enough to not touch it until the mmap() call
1068 if (!xen_feature(XENFEAT_gnttab_map_avail_bits)) {
1069 apply_to_page_range(vma->vm_mm, vma->vm_start,
1070 vma->vm_end - vma->vm_start,
1071 set_grant_ptes_as_special, NULL);
1079 mutex_unlock(&priv->lock);
1083 mutex_unlock(&priv->lock);
1087 unmap_grant_pages(map, 0, map->count);
1089 gntdev_put_map(priv, map);
1093 static const struct file_operations gntdev_fops = {
1094 .owner = THIS_MODULE,
1095 .open = gntdev_open,
1096 .release = gntdev_release,
1097 .mmap = gntdev_mmap,
1098 .unlocked_ioctl = gntdev_ioctl
1101 static struct miscdevice gntdev_miscdev = {
1102 .minor = MISC_DYNAMIC_MINOR,
1103 .name = "xen/gntdev",
1104 .fops = &gntdev_fops,
1107 /* ------------------------------------------------------------------ */
1109 static int __init gntdev_init(void)
1116 use_ptemod = !xen_feature(XENFEAT_auto_translated_physmap);
1118 err = misc_register(&gntdev_miscdev);
1120 pr_err("Could not register gntdev device\n");
1126 static void __exit gntdev_exit(void)
1128 misc_deregister(&gntdev_miscdev);
1131 module_init(gntdev_init);
1132 module_exit(gntdev_exit);
1134 /* ------------------------------------------------------------------ */