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>
9 * (c) 2018 Oleksandr Andrushchenko, EPAM Systems Inc.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
25 #include <linux/module.h>
26 #include <linux/kernel.h>
27 #include <linux/init.h>
28 #include <linux/miscdevice.h>
30 #include <linux/uaccess.h>
31 #include <linux/sched.h>
32 #include <linux/sched/mm.h>
33 #include <linux/spinlock.h>
34 #include <linux/slab.h>
35 #include <linux/highmem.h>
36 #include <linux/refcount.h>
37 #ifdef CONFIG_XEN_GRANT_DMA_ALLOC
38 #include <linux/of_device.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 #include "gntdev-common.h"
51 #ifdef CONFIG_XEN_GNTDEV_DMABUF
52 #include "gntdev-dmabuf.h"
55 MODULE_LICENSE("GPL");
56 MODULE_AUTHOR("Derek G. Murray <Derek.Murray@cl.cam.ac.uk>, "
57 "Gerd Hoffmann <kraxel@redhat.com>");
58 MODULE_DESCRIPTION("User-space granted page access driver");
60 static int limit = 1024*1024;
61 module_param(limit, int, 0644);
62 MODULE_PARM_DESC(limit, "Maximum number of grants that may be mapped by "
65 static atomic_t pages_mapped = ATOMIC_INIT(0);
67 static int use_ptemod;
68 #define populate_freeable_maps use_ptemod
70 static int unmap_grant_pages(struct gntdev_grant_map *map,
71 int offset, int pages);
73 static struct miscdevice gntdev_miscdev;
75 /* ------------------------------------------------------------------ */
77 bool gntdev_account_mapped_pages(int count)
79 return atomic_add_return(count, &pages_mapped) > limit;
82 static void gntdev_print_maps(struct gntdev_priv *priv,
83 char *text, int text_index)
86 struct gntdev_grant_map *map;
88 pr_debug("%s: maps list (priv %p)\n", __func__, priv);
89 list_for_each_entry(map, &priv->maps, next)
90 pr_debug(" index %2d, count %2d %s\n",
91 map->index, map->count,
92 map->index == text_index && text ? text : "");
96 static void gntdev_free_map(struct gntdev_grant_map *map)
101 #ifdef CONFIG_XEN_GRANT_DMA_ALLOC
102 if (map->dma_vaddr) {
103 struct gnttab_dma_alloc_args args;
105 args.dev = map->dma_dev;
106 args.coherent = !!(map->dma_flags & GNTDEV_DMA_FLAG_COHERENT);
107 args.nr_pages = map->count;
108 args.pages = map->pages;
109 args.frames = map->frames;
110 args.vaddr = map->dma_vaddr;
111 args.dev_bus_addr = map->dma_bus_addr;
113 gnttab_dma_free_pages(&args);
117 gnttab_free_pages(map->count, map->pages);
119 #ifdef CONFIG_XEN_GRANT_DMA_ALLOC
125 kfree(map->unmap_ops);
126 kfree(map->kmap_ops);
127 kfree(map->kunmap_ops);
131 struct gntdev_grant_map *gntdev_alloc_map(struct gntdev_priv *priv, int count,
134 struct gntdev_grant_map *add;
137 add = kzalloc(sizeof(*add), GFP_KERNEL);
141 add->grants = kcalloc(count, sizeof(add->grants[0]), GFP_KERNEL);
142 add->map_ops = kcalloc(count, sizeof(add->map_ops[0]), GFP_KERNEL);
143 add->unmap_ops = kcalloc(count, sizeof(add->unmap_ops[0]), GFP_KERNEL);
144 add->kmap_ops = kcalloc(count, sizeof(add->kmap_ops[0]), GFP_KERNEL);
145 add->kunmap_ops = kcalloc(count, sizeof(add->kunmap_ops[0]), GFP_KERNEL);
146 add->pages = kcalloc(count, sizeof(add->pages[0]), GFP_KERNEL);
147 if (NULL == add->grants ||
148 NULL == add->map_ops ||
149 NULL == add->unmap_ops ||
150 NULL == add->kmap_ops ||
151 NULL == add->kunmap_ops ||
155 #ifdef CONFIG_XEN_GRANT_DMA_ALLOC
156 add->dma_flags = dma_flags;
159 * Check if this mapping is requested to be backed
162 if (dma_flags & (GNTDEV_DMA_FLAG_WC | GNTDEV_DMA_FLAG_COHERENT)) {
163 struct gnttab_dma_alloc_args args;
165 add->frames = kcalloc(count, sizeof(add->frames[0]),
170 /* Remember the device, so we can free DMA memory. */
171 add->dma_dev = priv->dma_dev;
173 args.dev = priv->dma_dev;
174 args.coherent = !!(dma_flags & GNTDEV_DMA_FLAG_COHERENT);
175 args.nr_pages = count;
176 args.pages = add->pages;
177 args.frames = add->frames;
179 if (gnttab_dma_alloc_pages(&args))
182 add->dma_vaddr = args.vaddr;
183 add->dma_bus_addr = args.dev_bus_addr;
186 if (gnttab_alloc_pages(count, add->pages))
189 for (i = 0; i < count; i++) {
190 add->map_ops[i].handle = -1;
191 add->unmap_ops[i].handle = -1;
192 add->kmap_ops[i].handle = -1;
193 add->kunmap_ops[i].handle = -1;
198 refcount_set(&add->users, 1);
203 gntdev_free_map(add);
207 void gntdev_add_map(struct gntdev_priv *priv, struct gntdev_grant_map *add)
209 struct gntdev_grant_map *map;
211 list_for_each_entry(map, &priv->maps, next) {
212 if (add->index + add->count < map->index) {
213 list_add_tail(&add->next, &map->next);
216 add->index = map->index + map->count;
218 list_add_tail(&add->next, &priv->maps);
221 gntdev_print_maps(priv, "[new]", add->index);
224 static struct gntdev_grant_map *gntdev_find_map_index(struct gntdev_priv *priv,
225 int index, int count)
227 struct gntdev_grant_map *map;
229 list_for_each_entry(map, &priv->maps, next) {
230 if (map->index != index)
232 if (count && map->count != count)
239 void gntdev_put_map(struct gntdev_priv *priv, struct gntdev_grant_map *map)
244 if (!refcount_dec_and_test(&map->users))
247 atomic_sub(map->count, &pages_mapped);
249 if (map->notify.flags & UNMAP_NOTIFY_SEND_EVENT) {
250 notify_remote_via_evtchn(map->notify.event);
251 evtchn_put(map->notify.event);
254 if (populate_freeable_maps && priv) {
255 mutex_lock(&priv->lock);
256 list_del(&map->next);
257 mutex_unlock(&priv->lock);
260 if (map->pages && !use_ptemod)
261 unmap_grant_pages(map, 0, map->count);
262 gntdev_free_map(map);
265 /* ------------------------------------------------------------------ */
267 static int find_grant_ptes(pte_t *pte, pgtable_t token,
268 unsigned long addr, void *data)
270 struct gntdev_grant_map *map = data;
271 unsigned int pgnr = (addr - map->vma->vm_start) >> PAGE_SHIFT;
272 int flags = map->flags | GNTMAP_application_map | GNTMAP_contains_pte;
275 BUG_ON(pgnr >= map->count);
276 pte_maddr = arbitrary_virt_to_machine(pte).maddr;
279 * Set the PTE as special to force get_user_pages_fast() fall
280 * back to the slow path. If this is not supported as part of
281 * the grant map, it will be done afterwards.
283 if (xen_feature(XENFEAT_gnttab_map_avail_bits))
284 flags |= (1 << _GNTMAP_guest_avail0);
286 gnttab_set_map_op(&map->map_ops[pgnr], pte_maddr, flags,
287 map->grants[pgnr].ref,
288 map->grants[pgnr].domid);
289 gnttab_set_unmap_op(&map->unmap_ops[pgnr], pte_maddr, flags,
295 static int set_grant_ptes_as_special(pte_t *pte, pgtable_t token,
296 unsigned long addr, void *data)
298 set_pte_at(current->mm, addr, pte, pte_mkspecial(*pte));
303 int gntdev_map_grant_pages(struct gntdev_grant_map *map)
308 /* Note: it could already be mapped */
309 if (map->map_ops[0].handle != -1)
311 for (i = 0; i < map->count; i++) {
312 unsigned long addr = (unsigned long)
313 pfn_to_kaddr(page_to_pfn(map->pages[i]));
314 gnttab_set_map_op(&map->map_ops[i], addr, map->flags,
316 map->grants[i].domid);
317 gnttab_set_unmap_op(&map->unmap_ops[i], addr,
318 map->flags, -1 /* handle */);
322 * Setup the map_ops corresponding to the pte entries pointing
323 * to the kernel linear addresses of the struct pages.
324 * These ptes are completely different from the user ptes dealt
325 * with find_grant_ptes.
326 * Note that GNTMAP_device_map isn't needed here: The
327 * dev_bus_addr output field gets consumed only from ->map_ops,
328 * and by not requesting it when mapping we also avoid needing
329 * to mirror dev_bus_addr into ->unmap_ops (and holding an extra
330 * reference to the page in the hypervisor).
332 unsigned int flags = (map->flags & ~GNTMAP_device_map) |
335 for (i = 0; i < map->count; i++) {
336 unsigned long address = (unsigned long)
337 pfn_to_kaddr(page_to_pfn(map->pages[i]));
338 BUG_ON(PageHighMem(map->pages[i]));
340 gnttab_set_map_op(&map->kmap_ops[i], address, flags,
342 map->grants[i].domid);
343 gnttab_set_unmap_op(&map->kunmap_ops[i], address,
348 pr_debug("map %d+%d\n", map->index, map->count);
349 err = gnttab_map_refs(map->map_ops, use_ptemod ? map->kmap_ops : NULL,
350 map->pages, map->count);
352 for (i = 0; i < map->count; i++) {
353 if (map->map_ops[i].status == GNTST_okay)
354 map->unmap_ops[i].handle = map->map_ops[i].handle;
358 if (map->flags & GNTMAP_device_map)
359 map->unmap_ops[i].dev_bus_addr = map->map_ops[i].dev_bus_addr;
362 if (map->kmap_ops[i].status == GNTST_okay)
363 map->kunmap_ops[i].handle = map->kmap_ops[i].handle;
371 static int __unmap_grant_pages(struct gntdev_grant_map *map, int offset,
375 struct gntab_unmap_queue_data unmap_data;
377 if (map->notify.flags & UNMAP_NOTIFY_CLEAR_BYTE) {
378 int pgno = (map->notify.addr >> PAGE_SHIFT);
379 if (pgno >= offset && pgno < offset + pages) {
380 /* No need for kmap, pages are in lowmem */
381 uint8_t *tmp = pfn_to_kaddr(page_to_pfn(map->pages[pgno]));
382 tmp[map->notify.addr & (PAGE_SIZE-1)] = 0;
383 map->notify.flags &= ~UNMAP_NOTIFY_CLEAR_BYTE;
387 unmap_data.unmap_ops = map->unmap_ops + offset;
388 unmap_data.kunmap_ops = use_ptemod ? map->kunmap_ops + offset : NULL;
389 unmap_data.pages = map->pages + offset;
390 unmap_data.count = pages;
392 err = gnttab_unmap_refs_sync(&unmap_data);
396 for (i = 0; i < pages; i++) {
397 if (map->unmap_ops[offset+i].status)
399 pr_debug("unmap handle=%d st=%d\n",
400 map->unmap_ops[offset+i].handle,
401 map->unmap_ops[offset+i].status);
402 map->unmap_ops[offset+i].handle = -1;
407 static int unmap_grant_pages(struct gntdev_grant_map *map, int offset,
412 pr_debug("unmap %d+%d [%d+%d]\n", map->index, map->count, offset, pages);
414 /* It is possible the requested range will have a "hole" where we
415 * already unmapped some of the grants. Only unmap valid ranges.
417 while (pages && !err) {
418 while (pages && map->unmap_ops[offset].handle == -1) {
423 while (range < pages) {
424 if (map->unmap_ops[offset+range].handle == -1)
428 err = __unmap_grant_pages(map, offset, range);
436 /* ------------------------------------------------------------------ */
438 static void gntdev_vma_open(struct vm_area_struct *vma)
440 struct gntdev_grant_map *map = vma->vm_private_data;
442 pr_debug("gntdev_vma_open %p\n", vma);
443 refcount_inc(&map->users);
446 static void gntdev_vma_close(struct vm_area_struct *vma)
448 struct gntdev_grant_map *map = vma->vm_private_data;
449 struct file *file = vma->vm_file;
450 struct gntdev_priv *priv = file->private_data;
452 pr_debug("gntdev_vma_close %p\n", vma);
454 /* It is possible that an mmu notifier could be running
455 * concurrently, so take priv->lock to ensure that the vma won't
456 * vanishing during the unmap_grant_pages call, since we will
457 * spin here until that completes. Such a concurrent call will
458 * not do any unmapping, since that has been done prior to
459 * closing the vma, but it may still iterate the unmap_ops list.
461 mutex_lock(&priv->lock);
463 mutex_unlock(&priv->lock);
465 vma->vm_private_data = NULL;
466 gntdev_put_map(priv, map);
469 static struct page *gntdev_vma_find_special_page(struct vm_area_struct *vma,
472 struct gntdev_grant_map *map = vma->vm_private_data;
474 return map->pages[(addr - map->pages_vm_start) >> PAGE_SHIFT];
477 static const struct vm_operations_struct gntdev_vmops = {
478 .open = gntdev_vma_open,
479 .close = gntdev_vma_close,
480 .find_special_page = gntdev_vma_find_special_page,
483 /* ------------------------------------------------------------------ */
485 static bool in_range(struct gntdev_grant_map *map,
486 unsigned long start, unsigned long end)
490 if (map->vma->vm_start >= end)
492 if (map->vma->vm_end <= start)
498 static int unmap_if_in_range(struct gntdev_grant_map *map,
499 unsigned long start, unsigned long end,
502 unsigned long mstart, mend;
505 if (!in_range(map, start, end))
511 mstart = max(start, map->vma->vm_start);
512 mend = min(end, map->vma->vm_end);
513 pr_debug("map %d+%d (%lx %lx), range %lx %lx, mrange %lx %lx\n",
514 map->index, map->count,
515 map->vma->vm_start, map->vma->vm_end,
516 start, end, mstart, mend);
517 err = unmap_grant_pages(map,
518 (mstart - map->vma->vm_start) >> PAGE_SHIFT,
519 (mend - mstart) >> PAGE_SHIFT);
525 static int mn_invl_range_start(struct mmu_notifier *mn,
526 struct mm_struct *mm,
527 unsigned long start, unsigned long end,
530 struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
531 struct gntdev_grant_map *map;
535 mutex_lock(&priv->lock);
536 else if (!mutex_trylock(&priv->lock))
539 list_for_each_entry(map, &priv->maps, next) {
540 ret = unmap_if_in_range(map, start, end, blockable);
544 list_for_each_entry(map, &priv->freeable_maps, next) {
545 ret = unmap_if_in_range(map, start, end, blockable);
551 mutex_unlock(&priv->lock);
556 static void mn_release(struct mmu_notifier *mn,
557 struct mm_struct *mm)
559 struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
560 struct gntdev_grant_map *map;
563 mutex_lock(&priv->lock);
564 list_for_each_entry(map, &priv->maps, next) {
567 pr_debug("map %d+%d (%lx %lx)\n",
568 map->index, map->count,
569 map->vma->vm_start, map->vma->vm_end);
570 err = unmap_grant_pages(map, /* offset */ 0, map->count);
573 list_for_each_entry(map, &priv->freeable_maps, next) {
576 pr_debug("map %d+%d (%lx %lx)\n",
577 map->index, map->count,
578 map->vma->vm_start, map->vma->vm_end);
579 err = unmap_grant_pages(map, /* offset */ 0, map->count);
582 mutex_unlock(&priv->lock);
585 static const struct mmu_notifier_ops gntdev_mmu_ops = {
586 .release = mn_release,
587 .invalidate_range_start = mn_invl_range_start,
590 /* ------------------------------------------------------------------ */
592 static int gntdev_open(struct inode *inode, struct file *flip)
594 struct gntdev_priv *priv;
597 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
601 INIT_LIST_HEAD(&priv->maps);
602 INIT_LIST_HEAD(&priv->freeable_maps);
603 mutex_init(&priv->lock);
605 #ifdef CONFIG_XEN_GNTDEV_DMABUF
606 priv->dmabuf_priv = gntdev_dmabuf_init(flip);
607 if (IS_ERR(priv->dmabuf_priv)) {
608 ret = PTR_ERR(priv->dmabuf_priv);
615 priv->mm = get_task_mm(current);
620 priv->mn.ops = &gntdev_mmu_ops;
621 ret = mmu_notifier_register(&priv->mn, priv->mm);
630 flip->private_data = priv;
631 #ifdef CONFIG_XEN_GRANT_DMA_ALLOC
632 priv->dma_dev = gntdev_miscdev.this_device;
635 * The device is not spawn from a device tree, so arch_setup_dma_ops
636 * is not called, thus leaving the device with dummy DMA ops.
637 * Fix this by calling of_dma_configure() with a NULL node to set
640 of_dma_configure(priv->dma_dev, NULL, true);
642 pr_debug("priv %p\n", priv);
647 static int gntdev_release(struct inode *inode, struct file *flip)
649 struct gntdev_priv *priv = flip->private_data;
650 struct gntdev_grant_map *map;
652 pr_debug("priv %p\n", priv);
654 mutex_lock(&priv->lock);
655 while (!list_empty(&priv->maps)) {
656 map = list_entry(priv->maps.next,
657 struct gntdev_grant_map, next);
658 list_del(&map->next);
659 gntdev_put_map(NULL /* already removed */, map);
661 WARN_ON(!list_empty(&priv->freeable_maps));
662 mutex_unlock(&priv->lock);
664 #ifdef CONFIG_XEN_GNTDEV_DMABUF
665 gntdev_dmabuf_fini(priv->dmabuf_priv);
669 mmu_notifier_unregister(&priv->mn, priv->mm);
675 static long gntdev_ioctl_map_grant_ref(struct gntdev_priv *priv,
676 struct ioctl_gntdev_map_grant_ref __user *u)
678 struct ioctl_gntdev_map_grant_ref op;
679 struct gntdev_grant_map *map;
682 if (copy_from_user(&op, u, sizeof(op)) != 0)
684 pr_debug("priv %p, add %d\n", priv, op.count);
685 if (unlikely(op.count <= 0))
689 map = gntdev_alloc_map(priv, op.count, 0 /* This is not a dma-buf. */);
693 if (unlikely(gntdev_account_mapped_pages(op.count))) {
694 pr_debug("can't map: over limit\n");
695 gntdev_put_map(NULL, map);
699 if (copy_from_user(map->grants, &u->refs,
700 sizeof(map->grants[0]) * op.count) != 0) {
701 gntdev_put_map(NULL, map);
705 mutex_lock(&priv->lock);
706 gntdev_add_map(priv, map);
707 op.index = map->index << PAGE_SHIFT;
708 mutex_unlock(&priv->lock);
710 if (copy_to_user(u, &op, sizeof(op)) != 0)
716 static long gntdev_ioctl_unmap_grant_ref(struct gntdev_priv *priv,
717 struct ioctl_gntdev_unmap_grant_ref __user *u)
719 struct ioctl_gntdev_unmap_grant_ref op;
720 struct gntdev_grant_map *map;
723 if (copy_from_user(&op, u, sizeof(op)) != 0)
725 pr_debug("priv %p, del %d+%d\n", priv, (int)op.index, (int)op.count);
727 mutex_lock(&priv->lock);
728 map = gntdev_find_map_index(priv, op.index >> PAGE_SHIFT, op.count);
730 list_del(&map->next);
731 if (populate_freeable_maps)
732 list_add_tail(&map->next, &priv->freeable_maps);
735 mutex_unlock(&priv->lock);
737 gntdev_put_map(priv, map);
741 static long gntdev_ioctl_get_offset_for_vaddr(struct gntdev_priv *priv,
742 struct ioctl_gntdev_get_offset_for_vaddr __user *u)
744 struct ioctl_gntdev_get_offset_for_vaddr op;
745 struct vm_area_struct *vma;
746 struct gntdev_grant_map *map;
749 if (copy_from_user(&op, u, sizeof(op)) != 0)
751 pr_debug("priv %p, offset for vaddr %lx\n", priv, (unsigned long)op.vaddr);
753 down_read(¤t->mm->mmap_sem);
754 vma = find_vma(current->mm, op.vaddr);
755 if (!vma || vma->vm_ops != &gntdev_vmops)
758 map = vma->vm_private_data;
762 op.offset = map->index << PAGE_SHIFT;
763 op.count = map->count;
767 up_read(¤t->mm->mmap_sem);
769 if (rv == 0 && copy_to_user(u, &op, sizeof(op)) != 0)
774 static long gntdev_ioctl_notify(struct gntdev_priv *priv, void __user *u)
776 struct ioctl_gntdev_unmap_notify op;
777 struct gntdev_grant_map *map;
780 unsigned int out_event;
782 if (copy_from_user(&op, u, sizeof(op)))
785 if (op.action & ~(UNMAP_NOTIFY_CLEAR_BYTE|UNMAP_NOTIFY_SEND_EVENT))
788 /* We need to grab a reference to the event channel we are going to use
789 * to send the notify before releasing the reference we may already have
790 * (if someone has called this ioctl twice). This is required so that
791 * it is possible to change the clear_byte part of the notification
792 * without disturbing the event channel part, which may now be the last
793 * reference to that event channel.
795 if (op.action & UNMAP_NOTIFY_SEND_EVENT) {
796 if (evtchn_get(op.event_channel_port))
800 out_flags = op.action;
801 out_event = op.event_channel_port;
803 mutex_lock(&priv->lock);
805 list_for_each_entry(map, &priv->maps, next) {
806 uint64_t begin = map->index << PAGE_SHIFT;
807 uint64_t end = (map->index + map->count) << PAGE_SHIFT;
808 if (op.index >= begin && op.index < end)
815 if ((op.action & UNMAP_NOTIFY_CLEAR_BYTE) &&
816 (map->flags & GNTMAP_readonly)) {
821 out_flags = map->notify.flags;
822 out_event = map->notify.event;
824 map->notify.flags = op.action;
825 map->notify.addr = op.index - (map->index << PAGE_SHIFT);
826 map->notify.event = op.event_channel_port;
831 mutex_unlock(&priv->lock);
833 /* Drop the reference to the event channel we did not save in the map */
834 if (out_flags & UNMAP_NOTIFY_SEND_EVENT)
835 evtchn_put(out_event);
840 #define GNTDEV_COPY_BATCH 16
842 struct gntdev_copy_batch {
843 struct gnttab_copy ops[GNTDEV_COPY_BATCH];
844 struct page *pages[GNTDEV_COPY_BATCH];
845 s16 __user *status[GNTDEV_COPY_BATCH];
847 unsigned int nr_pages;
851 static int gntdev_get_page(struct gntdev_copy_batch *batch, void __user *virt,
854 unsigned long addr = (unsigned long)virt;
856 unsigned long xen_pfn;
859 ret = get_user_pages_fast(addr, 1, batch->writeable, &page);
863 batch->pages[batch->nr_pages++] = page;
865 xen_pfn = page_to_xen_pfn(page) + XEN_PFN_DOWN(addr & ~PAGE_MASK);
866 *gfn = pfn_to_gfn(xen_pfn);
871 static void gntdev_put_pages(struct gntdev_copy_batch *batch)
875 for (i = 0; i < batch->nr_pages; i++) {
876 if (batch->writeable && !PageDirty(batch->pages[i]))
877 set_page_dirty_lock(batch->pages[i]);
878 put_page(batch->pages[i]);
881 batch->writeable = false;
884 static int gntdev_copy(struct gntdev_copy_batch *batch)
888 gnttab_batch_copy(batch->ops, batch->nr_ops);
889 gntdev_put_pages(batch);
892 * For each completed op, update the status if the op failed
893 * and all previous ops for the segment were successful.
895 for (i = 0; i < batch->nr_ops; i++) {
896 s16 status = batch->ops[i].status;
899 if (status == GNTST_okay)
902 if (__get_user(old_status, batch->status[i]))
905 if (old_status != GNTST_okay)
908 if (__put_user(status, batch->status[i]))
916 static int gntdev_grant_copy_seg(struct gntdev_copy_batch *batch,
917 struct gntdev_grant_copy_segment *seg,
923 * Disallow local -> local copies since there is only space in
924 * batch->pages for one page per-op and this would be a very
925 * expensive memcpy().
927 if (!(seg->flags & (GNTCOPY_source_gref | GNTCOPY_dest_gref)))
930 /* Can't cross page if source/dest is a grant ref. */
931 if (seg->flags & GNTCOPY_source_gref) {
932 if (seg->source.foreign.offset + seg->len > XEN_PAGE_SIZE)
935 if (seg->flags & GNTCOPY_dest_gref) {
936 if (seg->dest.foreign.offset + seg->len > XEN_PAGE_SIZE)
940 if (put_user(GNTST_okay, status))
943 while (copied < seg->len) {
944 struct gnttab_copy *op;
950 if (batch->nr_ops >= GNTDEV_COPY_BATCH) {
951 ret = gntdev_copy(batch);
956 len = seg->len - copied;
958 op = &batch->ops[batch->nr_ops];
961 if (seg->flags & GNTCOPY_source_gref) {
962 op->source.u.ref = seg->source.foreign.ref;
963 op->source.domid = seg->source.foreign.domid;
964 op->source.offset = seg->source.foreign.offset + copied;
965 op->flags |= GNTCOPY_source_gref;
967 virt = seg->source.virt + copied;
968 off = (unsigned long)virt & ~XEN_PAGE_MASK;
969 len = min(len, (size_t)XEN_PAGE_SIZE - off);
970 batch->writeable = false;
972 ret = gntdev_get_page(batch, virt, &gfn);
976 op->source.u.gmfn = gfn;
977 op->source.domid = DOMID_SELF;
978 op->source.offset = off;
981 if (seg->flags & GNTCOPY_dest_gref) {
982 op->dest.u.ref = seg->dest.foreign.ref;
983 op->dest.domid = seg->dest.foreign.domid;
984 op->dest.offset = seg->dest.foreign.offset + copied;
985 op->flags |= GNTCOPY_dest_gref;
987 virt = seg->dest.virt + copied;
988 off = (unsigned long)virt & ~XEN_PAGE_MASK;
989 len = min(len, (size_t)XEN_PAGE_SIZE - off);
990 batch->writeable = true;
992 ret = gntdev_get_page(batch, virt, &gfn);
996 op->dest.u.gmfn = gfn;
997 op->dest.domid = DOMID_SELF;
998 op->dest.offset = off;
1004 batch->status[batch->nr_ops] = status;
1011 static long gntdev_ioctl_grant_copy(struct gntdev_priv *priv, void __user *u)
1013 struct ioctl_gntdev_grant_copy copy;
1014 struct gntdev_copy_batch batch;
1018 if (copy_from_user(©, u, sizeof(copy)))
1024 for (i = 0; i < copy.count; i++) {
1025 struct gntdev_grant_copy_segment seg;
1027 if (copy_from_user(&seg, ©.segments[i], sizeof(seg))) {
1032 ret = gntdev_grant_copy_seg(&batch, &seg, ©.segments[i].status);
1039 ret = gntdev_copy(&batch);
1043 gntdev_put_pages(&batch);
1047 static long gntdev_ioctl(struct file *flip,
1048 unsigned int cmd, unsigned long arg)
1050 struct gntdev_priv *priv = flip->private_data;
1051 void __user *ptr = (void __user *)arg;
1054 case IOCTL_GNTDEV_MAP_GRANT_REF:
1055 return gntdev_ioctl_map_grant_ref(priv, ptr);
1057 case IOCTL_GNTDEV_UNMAP_GRANT_REF:
1058 return gntdev_ioctl_unmap_grant_ref(priv, ptr);
1060 case IOCTL_GNTDEV_GET_OFFSET_FOR_VADDR:
1061 return gntdev_ioctl_get_offset_for_vaddr(priv, ptr);
1063 case IOCTL_GNTDEV_SET_UNMAP_NOTIFY:
1064 return gntdev_ioctl_notify(priv, ptr);
1066 case IOCTL_GNTDEV_GRANT_COPY:
1067 return gntdev_ioctl_grant_copy(priv, ptr);
1069 #ifdef CONFIG_XEN_GNTDEV_DMABUF
1070 case IOCTL_GNTDEV_DMABUF_EXP_FROM_REFS:
1071 return gntdev_ioctl_dmabuf_exp_from_refs(priv, use_ptemod, ptr);
1073 case IOCTL_GNTDEV_DMABUF_EXP_WAIT_RELEASED:
1074 return gntdev_ioctl_dmabuf_exp_wait_released(priv, ptr);
1076 case IOCTL_GNTDEV_DMABUF_IMP_TO_REFS:
1077 return gntdev_ioctl_dmabuf_imp_to_refs(priv, ptr);
1079 case IOCTL_GNTDEV_DMABUF_IMP_RELEASE:
1080 return gntdev_ioctl_dmabuf_imp_release(priv, ptr);
1084 pr_debug("priv %p, unknown cmd %x\n", priv, cmd);
1085 return -ENOIOCTLCMD;
1091 static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma)
1093 struct gntdev_priv *priv = flip->private_data;
1094 int index = vma->vm_pgoff;
1095 int count = vma_pages(vma);
1096 struct gntdev_grant_map *map;
1097 int i, err = -EINVAL;
1099 if ((vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_SHARED))
1102 pr_debug("map %d+%d at %lx (pgoff %lx)\n",
1103 index, count, vma->vm_start, vma->vm_pgoff);
1105 mutex_lock(&priv->lock);
1106 map = gntdev_find_map_index(priv, index, count);
1109 if (use_ptemod && map->vma)
1111 if (use_ptemod && priv->mm != vma->vm_mm) {
1112 pr_warn("Huh? Other mm?\n");
1116 refcount_inc(&map->users);
1118 vma->vm_ops = &gntdev_vmops;
1120 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP | VM_MIXEDMAP;
1123 vma->vm_flags |= VM_DONTCOPY;
1125 vma->vm_private_data = map;
1131 if ((vma->vm_flags & VM_WRITE) &&
1132 (map->flags & GNTMAP_readonly))
1133 goto out_unlock_put;
1135 map->flags = GNTMAP_host_map;
1136 if (!(vma->vm_flags & VM_WRITE))
1137 map->flags |= GNTMAP_readonly;
1140 mutex_unlock(&priv->lock);
1143 map->pages_vm_start = vma->vm_start;
1144 err = apply_to_page_range(vma->vm_mm, vma->vm_start,
1145 vma->vm_end - vma->vm_start,
1146 find_grant_ptes, map);
1148 pr_warn("find_grant_ptes() failure.\n");
1153 err = gntdev_map_grant_pages(map);
1158 for (i = 0; i < count; i++) {
1159 err = vm_insert_page(vma, vma->vm_start + i*PAGE_SIZE,
1167 * If the PTEs were not made special by the grant map
1168 * hypercall, do so here.
1170 * This is racy since the mapping is already visible
1171 * to userspace but userspace should be well-behaved
1172 * enough to not touch it until the mmap() call
1175 if (!xen_feature(XENFEAT_gnttab_map_avail_bits)) {
1176 apply_to_page_range(vma->vm_mm, vma->vm_start,
1177 vma->vm_end - vma->vm_start,
1178 set_grant_ptes_as_special, NULL);
1186 mutex_unlock(&priv->lock);
1190 mutex_unlock(&priv->lock);
1194 unmap_grant_pages(map, 0, map->count);
1196 gntdev_put_map(priv, map);
1200 static const struct file_operations gntdev_fops = {
1201 .owner = THIS_MODULE,
1202 .open = gntdev_open,
1203 .release = gntdev_release,
1204 .mmap = gntdev_mmap,
1205 .unlocked_ioctl = gntdev_ioctl
1208 static struct miscdevice gntdev_miscdev = {
1209 .minor = MISC_DYNAMIC_MINOR,
1210 .name = "xen/gntdev",
1211 .fops = &gntdev_fops,
1214 /* ------------------------------------------------------------------ */
1216 static int __init gntdev_init(void)
1223 use_ptemod = !xen_feature(XENFEAT_auto_translated_physmap);
1225 err = misc_register(&gntdev_miscdev);
1227 pr_err("Could not register gntdev device\n");
1233 static void __exit gntdev_exit(void)
1235 misc_deregister(&gntdev_miscdev);
1238 module_init(gntdev_init);
1239 module_exit(gntdev_exit);
1241 /* ------------------------------------------------------------------ */