blob: e41c79c986ea1c43f21675ab5579475ba4e0dc93 [file] [log] [blame]
Gerd Hoffmannab315232010-12-14 18:40:46 +00001/******************************************************************************
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
Joe Perches283c0972013-06-28 03:21:41 -070022#define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
23
Gerd Hoffmannab315232010-12-14 18:40:46 +000024#include <linux/module.h>
25#include <linux/kernel.h>
26#include <linux/init.h>
27#include <linux/miscdevice.h>
28#include <linux/fs.h>
29#include <linux/mm.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/spinlock.h>
36#include <linux/slab.h>
Daniel De Graafaab8f112011-02-03 12:19:02 -050037#include <linux/highmem.h>
Gerd Hoffmannab315232010-12-14 18:40:46 +000038
39#include <xen/xen.h>
40#include <xen/grant_table.h>
Daniel De Graafca47cea2011-03-09 18:07:34 -050041#include <xen/balloon.h>
Gerd Hoffmannab315232010-12-14 18:40:46 +000042#include <xen/gntdev.h>
Daniel De Graafbdc612d2011-02-03 12:19:04 -050043#include <xen/events.h>
Gerd Hoffmannab315232010-12-14 18:40:46 +000044#include <asm/xen/hypervisor.h>
45#include <asm/xen/hypercall.h>
46#include <asm/xen/page.h>
47
48MODULE_LICENSE("GPL");
49MODULE_AUTHOR("Derek G. Murray <Derek.Murray@cl.cam.ac.uk>, "
50 "Gerd Hoffmann <kraxel@redhat.com>");
51MODULE_DESCRIPTION("User-space granted page access driver");
52
Daniel De Graafef910822011-02-03 12:18:59 -050053static int limit = 1024*1024;
Gerd Hoffmannab315232010-12-14 18:40:46 +000054module_param(limit, int, 0644);
Daniel De Graafef910822011-02-03 12:18:59 -050055MODULE_PARM_DESC(limit, "Maximum number of grants that may be mapped by "
56 "the gntdev device");
57
58static atomic_t pages_mapped = ATOMIC_INIT(0);
Gerd Hoffmannab315232010-12-14 18:40:46 +000059
Daniel De Graafaab8f112011-02-03 12:19:02 -050060static int use_ptemod;
Daniel De Graaf16a1d022013-01-02 22:57:12 +000061#define populate_freeable_maps use_ptemod
Daniel De Graafaab8f112011-02-03 12:19:02 -050062
Gerd Hoffmannab315232010-12-14 18:40:46 +000063struct gntdev_priv {
Daniel De Graaf16a1d022013-01-02 22:57:12 +000064 /* maps with visible offsets in the file descriptor */
Gerd Hoffmannab315232010-12-14 18:40:46 +000065 struct list_head maps;
Daniel De Graaf16a1d022013-01-02 22:57:12 +000066 /* maps that are not visible; will be freed on munmap.
67 * Only populated if populate_freeable_maps == 1 */
68 struct list_head freeable_maps;
69 /* lock protects maps and freeable_maps */
Gerd Hoffmannab315232010-12-14 18:40:46 +000070 spinlock_t lock;
71 struct mm_struct *mm;
72 struct mmu_notifier mn;
73};
74
Daniel De Graafbdc612d2011-02-03 12:19:04 -050075struct unmap_notify {
76 int flags;
77 /* Address relative to the start of the grant_map */
78 int addr;
79 int event;
80};
81
Gerd Hoffmannab315232010-12-14 18:40:46 +000082struct grant_map {
83 struct list_head next;
Gerd Hoffmannab315232010-12-14 18:40:46 +000084 struct vm_area_struct *vma;
85 int index;
86 int count;
87 int flags;
Daniel De Graaf68b025c2011-02-03 12:19:01 -050088 atomic_t users;
Daniel De Graafbdc612d2011-02-03 12:19:04 -050089 struct unmap_notify notify;
Gerd Hoffmannab315232010-12-14 18:40:46 +000090 struct ioctl_gntdev_grant_ref *grants;
91 struct gnttab_map_grant_ref *map_ops;
92 struct gnttab_unmap_grant_ref *unmap_ops;
Stefano Stabellini0930bba2011-09-29 11:57:56 +010093 struct gnttab_map_grant_ref *kmap_ops;
Stefano Stabellinia12b4eb2010-12-10 14:56:42 +000094 struct page **pages;
Gerd Hoffmannab315232010-12-14 18:40:46 +000095};
96
Daniel De Graafaab8f112011-02-03 12:19:02 -050097static int unmap_grant_pages(struct grant_map *map, int offset, int pages);
98
Gerd Hoffmannab315232010-12-14 18:40:46 +000099/* ------------------------------------------------------------------ */
100
101static void gntdev_print_maps(struct gntdev_priv *priv,
102 char *text, int text_index)
103{
104#ifdef DEBUG
105 struct grant_map *map;
106
Daniel De Graafef910822011-02-03 12:18:59 -0500107 pr_debug("%s: maps list (priv %p)\n", __func__, priv);
Gerd Hoffmannab315232010-12-14 18:40:46 +0000108 list_for_each_entry(map, &priv->maps, next)
109 pr_debug(" index %2d, count %2d %s\n",
110 map->index, map->count,
111 map->index == text_index && text ? text : "");
112#endif
113}
114
David Vrabela67baeb2012-10-24 12:39:02 +0100115static void gntdev_free_map(struct grant_map *map)
116{
117 if (map == NULL)
118 return;
119
120 if (map->pages)
121 free_xenballooned_pages(map->count, map->pages);
122 kfree(map->pages);
123 kfree(map->grants);
124 kfree(map->map_ops);
125 kfree(map->unmap_ops);
126 kfree(map->kmap_ops);
127 kfree(map);
128}
129
Gerd Hoffmannab315232010-12-14 18:40:46 +0000130static struct grant_map *gntdev_alloc_map(struct gntdev_priv *priv, int count)
131{
132 struct grant_map *add;
Stefano Stabellinia12b4eb2010-12-10 14:56:42 +0000133 int i;
Gerd Hoffmannab315232010-12-14 18:40:46 +0000134
135 add = kzalloc(sizeof(struct grant_map), GFP_KERNEL);
136 if (NULL == add)
137 return NULL;
138
Dan Carpenterfc6e0c32011-11-04 21:23:32 +0300139 add->grants = kcalloc(count, sizeof(add->grants[0]), GFP_KERNEL);
140 add->map_ops = kcalloc(count, sizeof(add->map_ops[0]), GFP_KERNEL);
141 add->unmap_ops = kcalloc(count, sizeof(add->unmap_ops[0]), GFP_KERNEL);
142 add->kmap_ops = kcalloc(count, sizeof(add->kmap_ops[0]), GFP_KERNEL);
143 add->pages = kcalloc(count, sizeof(add->pages[0]), GFP_KERNEL);
Stefano Stabellinia12b4eb2010-12-10 14:56:42 +0000144 if (NULL == add->grants ||
145 NULL == add->map_ops ||
146 NULL == add->unmap_ops ||
Stefano Stabellini0930bba2011-09-29 11:57:56 +0100147 NULL == add->kmap_ops ||
Stefano Stabellinia12b4eb2010-12-10 14:56:42 +0000148 NULL == add->pages)
Gerd Hoffmannab315232010-12-14 18:40:46 +0000149 goto err;
150
Stefano Stabellini693394b2011-09-29 11:57:55 +0100151 if (alloc_xenballooned_pages(count, add->pages, false /* lowmem */))
Daniel De Graafca47cea2011-03-09 18:07:34 -0500152 goto err;
153
Stefano Stabellinia12b4eb2010-12-10 14:56:42 +0000154 for (i = 0; i < count; i++) {
Daniel De Graaf77c35ac2011-02-23 08:11:35 -0500155 add->map_ops[i].handle = -1;
156 add->unmap_ops[i].handle = -1;
Stefano Stabellini0930bba2011-09-29 11:57:56 +0100157 add->kmap_ops[i].handle = -1;
Stefano Stabellinia12b4eb2010-12-10 14:56:42 +0000158 }
159
Gerd Hoffmannab315232010-12-14 18:40:46 +0000160 add->index = 0;
161 add->count = count;
Daniel De Graaf68b025c2011-02-03 12:19:01 -0500162 atomic_set(&add->users, 1);
Gerd Hoffmannab315232010-12-14 18:40:46 +0000163
Gerd Hoffmannab315232010-12-14 18:40:46 +0000164 return add;
165
166err:
David Vrabela67baeb2012-10-24 12:39:02 +0100167 gntdev_free_map(add);
Gerd Hoffmannab315232010-12-14 18:40:46 +0000168 return NULL;
169}
170
171static void gntdev_add_map(struct gntdev_priv *priv, struct grant_map *add)
172{
173 struct grant_map *map;
174
175 list_for_each_entry(map, &priv->maps, next) {
176 if (add->index + add->count < map->index) {
177 list_add_tail(&add->next, &map->next);
178 goto done;
179 }
180 add->index = map->index + map->count;
181 }
182 list_add_tail(&add->next, &priv->maps);
183
184done:
Gerd Hoffmannab315232010-12-14 18:40:46 +0000185 gntdev_print_maps(priv, "[new]", add->index);
186}
187
188static struct grant_map *gntdev_find_map_index(struct gntdev_priv *priv,
189 int index, int count)
190{
191 struct grant_map *map;
192
193 list_for_each_entry(map, &priv->maps, next) {
194 if (map->index != index)
195 continue;
Daniel De Graafbdc612d2011-02-03 12:19:04 -0500196 if (count && map->count != count)
Gerd Hoffmannab315232010-12-14 18:40:46 +0000197 continue;
198 return map;
199 }
200 return NULL;
201}
202
Daniel De Graaf16a1d022013-01-02 22:57:12 +0000203static void gntdev_put_map(struct gntdev_priv *priv, struct grant_map *map)
Gerd Hoffmannab315232010-12-14 18:40:46 +0000204{
205 if (!map)
206 return;
Stefano Stabellinia12b4eb2010-12-10 14:56:42 +0000207
Daniel De Graaf68b025c2011-02-03 12:19:01 -0500208 if (!atomic_dec_and_test(&map->users))
209 return;
210
211 atomic_sub(map->count, &pages_mapped);
212
Daniel De Graaf0cc678f2011-10-27 17:58:49 -0400213 if (map->notify.flags & UNMAP_NOTIFY_SEND_EVENT) {
Daniel De Graafbdc612d2011-02-03 12:19:04 -0500214 notify_remote_via_evtchn(map->notify.event);
Daniel De Graaf0cc678f2011-10-27 17:58:49 -0400215 evtchn_put(map->notify.event);
216 }
Daniel De Graafbdc612d2011-02-03 12:19:04 -0500217
Daniel De Graaf16a1d022013-01-02 22:57:12 +0000218 if (populate_freeable_maps && priv) {
219 spin_lock(&priv->lock);
220 list_del(&map->next);
221 spin_unlock(&priv->lock);
222 }
223
David Vrabela67baeb2012-10-24 12:39:02 +0100224 if (map->pages && !use_ptemod)
225 unmap_grant_pages(map, 0, map->count);
226 gntdev_free_map(map);
Gerd Hoffmannab315232010-12-14 18:40:46 +0000227}
228
229/* ------------------------------------------------------------------ */
230
231static int find_grant_ptes(pte_t *pte, pgtable_t token,
232 unsigned long addr, void *data)
233{
234 struct grant_map *map = data;
235 unsigned int pgnr = (addr - map->vma->vm_start) >> PAGE_SHIFT;
Daniel De Graafaab8f112011-02-03 12:19:02 -0500236 int flags = map->flags | GNTMAP_application_map | GNTMAP_contains_pte;
Gerd Hoffmannab315232010-12-14 18:40:46 +0000237 u64 pte_maddr;
238
239 BUG_ON(pgnr >= map->count);
Jeremy Fitzhardingeba5d1012010-12-08 10:54:32 -0800240 pte_maddr = arbitrary_virt_to_machine(pte).maddr;
241
Daniel De Graafaab8f112011-02-03 12:19:02 -0500242 gnttab_set_map_op(&map->map_ops[pgnr], pte_maddr, flags,
Gerd Hoffmannab315232010-12-14 18:40:46 +0000243 map->grants[pgnr].ref,
244 map->grants[pgnr].domid);
Daniel De Graafaab8f112011-02-03 12:19:02 -0500245 gnttab_set_unmap_op(&map->unmap_ops[pgnr], pte_maddr, flags,
Daniel De Graaf77c35ac2011-02-23 08:11:35 -0500246 -1 /* handle */);
Gerd Hoffmannab315232010-12-14 18:40:46 +0000247 return 0;
248}
249
250static int map_grant_pages(struct grant_map *map)
251{
252 int i, err = 0;
Daniel De Graafaab8f112011-02-03 12:19:02 -0500253
254 if (!use_ptemod) {
Daniel De Graaf12996fc2011-02-09 16:11:32 -0500255 /* Note: it could already be mapped */
Daniel De Graaf77c35ac2011-02-23 08:11:35 -0500256 if (map->map_ops[0].handle != -1)
Daniel De Graaf12996fc2011-02-09 16:11:32 -0500257 return 0;
Daniel De Graafaab8f112011-02-03 12:19:02 -0500258 for (i = 0; i < map->count; i++) {
Ian Campbell38eaeb02011-03-08 16:56:43 +0000259 unsigned long addr = (unsigned long)
Daniel De Graafaab8f112011-02-03 12:19:02 -0500260 pfn_to_kaddr(page_to_pfn(map->pages[i]));
261 gnttab_set_map_op(&map->map_ops[i], addr, map->flags,
262 map->grants[i].ref,
263 map->grants[i].domid);
264 gnttab_set_unmap_op(&map->unmap_ops[i], addr,
Daniel De Graaf77c35ac2011-02-23 08:11:35 -0500265 map->flags, -1 /* handle */);
Daniel De Graafaab8f112011-02-03 12:19:02 -0500266 }
Stefano Stabellini0930bba2011-09-29 11:57:56 +0100267 } else {
268 /*
269 * Setup the map_ops corresponding to the pte entries pointing
270 * to the kernel linear addresses of the struct pages.
271 * These ptes are completely different from the user ptes dealt
272 * with find_grant_ptes.
273 */
274 for (i = 0; i < map->count; i++) {
Stefano Stabellini0930bba2011-09-29 11:57:56 +0100275 unsigned long address = (unsigned long)
276 pfn_to_kaddr(page_to_pfn(map->pages[i]));
Stefano Stabellini0930bba2011-09-29 11:57:56 +0100277 BUG_ON(PageHighMem(map->pages[i]));
278
Stefano Stabelliniee072642013-07-23 17:23:54 +0000279 gnttab_set_map_op(&map->kmap_ops[i], address,
280 map->flags | GNTMAP_host_map,
Stefano Stabellini0930bba2011-09-29 11:57:56 +0100281 map->grants[i].ref,
282 map->grants[i].domid);
283 }
Daniel De Graafaab8f112011-02-03 12:19:02 -0500284 }
Gerd Hoffmannab315232010-12-14 18:40:46 +0000285
286 pr_debug("map %d+%d\n", map->index, map->count);
Stefano Stabellini0930bba2011-09-29 11:57:56 +0100287 err = gnttab_map_refs(map->map_ops, use_ptemod ? map->kmap_ops : NULL,
288 map->pages, map->count);
Gerd Hoffmannab315232010-12-14 18:40:46 +0000289 if (err)
290 return err;
291
292 for (i = 0; i < map->count; i++) {
293 if (map->map_ops[i].status)
294 err = -EINVAL;
Daniel De Graaf77c35ac2011-02-23 08:11:35 -0500295 else {
296 BUG_ON(map->map_ops[i].handle == -1);
297 map->unmap_ops[i].handle = map->map_ops[i].handle;
298 pr_debug("map handle=%d\n", map->map_ops[i].handle);
299 }
Gerd Hoffmannab315232010-12-14 18:40:46 +0000300 }
301 return err;
302}
303
Daniel De Graafb57c1862011-02-09 15:12:00 -0500304static int __unmap_grant_pages(struct grant_map *map, int offset, int pages)
Gerd Hoffmannab315232010-12-14 18:40:46 +0000305{
306 int i, err = 0;
307
Daniel De Graafbdc612d2011-02-03 12:19:04 -0500308 if (map->notify.flags & UNMAP_NOTIFY_CLEAR_BYTE) {
309 int pgno = (map->notify.addr >> PAGE_SHIFT);
Daniel De Graaf1affa982013-01-02 17:57:13 -0500310 if (pgno >= offset && pgno < offset + pages) {
311 /* No need for kmap, pages are in lowmem */
312 uint8_t *tmp = pfn_to_kaddr(page_to_pfn(map->pages[pgno]));
Daniel De Graafbdc612d2011-02-03 12:19:04 -0500313 tmp[map->notify.addr & (PAGE_SIZE-1)] = 0;
Daniel De Graafbdc612d2011-02-03 12:19:04 -0500314 map->notify.flags &= ~UNMAP_NOTIFY_CLEAR_BYTE;
315 }
316 }
317
Stefano Stabellini2fc136e2012-09-12 12:44:30 +0100318 err = gnttab_unmap_refs(map->unmap_ops + offset,
319 use_ptemod ? map->kmap_ops + offset : NULL, map->pages + offset,
320 pages);
Gerd Hoffmannab315232010-12-14 18:40:46 +0000321 if (err)
322 return err;
323
324 for (i = 0; i < pages; i++) {
325 if (map->unmap_ops[offset+i].status)
326 err = -EINVAL;
Daniel De Graaf77c35ac2011-02-23 08:11:35 -0500327 pr_debug("unmap handle=%d st=%d\n",
328 map->unmap_ops[offset+i].handle,
329 map->unmap_ops[offset+i].status);
330 map->unmap_ops[offset+i].handle = -1;
Gerd Hoffmannab315232010-12-14 18:40:46 +0000331 }
332 return err;
333}
334
Daniel De Graafb57c1862011-02-09 15:12:00 -0500335static int unmap_grant_pages(struct grant_map *map, int offset, int pages)
336{
337 int range, err = 0;
338
339 pr_debug("unmap %d+%d [%d+%d]\n", map->index, map->count, offset, pages);
340
341 /* It is possible the requested range will have a "hole" where we
342 * already unmapped some of the grants. Only unmap valid ranges.
343 */
344 while (pages && !err) {
Daniel De Graaf77c35ac2011-02-23 08:11:35 -0500345 while (pages && map->unmap_ops[offset].handle == -1) {
Daniel De Graafb57c1862011-02-09 15:12:00 -0500346 offset++;
347 pages--;
348 }
349 range = 0;
350 while (range < pages) {
Daniel De Graaf77c35ac2011-02-23 08:11:35 -0500351 if (map->unmap_ops[offset+range].handle == -1) {
Daniel De Graafb57c1862011-02-09 15:12:00 -0500352 range--;
353 break;
354 }
355 range++;
356 }
357 err = __unmap_grant_pages(map, offset, range);
358 offset += range;
359 pages -= range;
360 }
361
362 return err;
363}
364
Gerd Hoffmannab315232010-12-14 18:40:46 +0000365/* ------------------------------------------------------------------ */
366
Daniel De Graafd79647a2011-03-07 15:18:57 -0500367static void gntdev_vma_open(struct vm_area_struct *vma)
368{
369 struct grant_map *map = vma->vm_private_data;
370
371 pr_debug("gntdev_vma_open %p\n", vma);
372 atomic_inc(&map->users);
373}
374
Gerd Hoffmannab315232010-12-14 18:40:46 +0000375static void gntdev_vma_close(struct vm_area_struct *vma)
376{
377 struct grant_map *map = vma->vm_private_data;
Daniel De Graaf16a1d022013-01-02 22:57:12 +0000378 struct file *file = vma->vm_file;
379 struct gntdev_priv *priv = file->private_data;
Gerd Hoffmannab315232010-12-14 18:40:46 +0000380
Daniel De Graafd79647a2011-03-07 15:18:57 -0500381 pr_debug("gntdev_vma_close %p\n", vma);
Daniel De Graaf2512f292013-01-02 22:57:11 +0000382 if (use_ptemod) {
Daniel De Graaf2512f292013-01-02 22:57:11 +0000383 /* It is possible that an mmu notifier could be running
384 * concurrently, so take priv->lock to ensure that the vma won't
385 * vanishing during the unmap_grant_pages call, since we will
386 * spin here until that completes. Such a concurrent call will
387 * not do any unmapping, since that has been done prior to
388 * closing the vma, but it may still iterate the unmap_ops list.
389 */
390 spin_lock(&priv->lock);
391 map->vma = NULL;
392 spin_unlock(&priv->lock);
393 }
Gerd Hoffmannab315232010-12-14 18:40:46 +0000394 vma->vm_private_data = NULL;
Daniel De Graaf16a1d022013-01-02 22:57:12 +0000395 gntdev_put_map(priv, map);
Gerd Hoffmannab315232010-12-14 18:40:46 +0000396}
397
Gerd Hoffmannab315232010-12-14 18:40:46 +0000398static struct vm_operations_struct gntdev_vmops = {
Daniel De Graafd79647a2011-03-07 15:18:57 -0500399 .open = gntdev_vma_open,
Gerd Hoffmannab315232010-12-14 18:40:46 +0000400 .close = gntdev_vma_close,
Gerd Hoffmannab315232010-12-14 18:40:46 +0000401};
402
403/* ------------------------------------------------------------------ */
404
Daniel De Graaf16a1d022013-01-02 22:57:12 +0000405static void unmap_if_in_range(struct grant_map *map,
406 unsigned long start, unsigned long end)
407{
408 unsigned long mstart, mend;
409 int err;
410
411 if (!map->vma)
412 return;
413 if (map->vma->vm_start >= end)
414 return;
415 if (map->vma->vm_end <= start)
416 return;
417 mstart = max(start, map->vma->vm_start);
418 mend = min(end, map->vma->vm_end);
419 pr_debug("map %d+%d (%lx %lx), range %lx %lx, mrange %lx %lx\n",
420 map->index, map->count,
421 map->vma->vm_start, map->vma->vm_end,
422 start, end, mstart, mend);
423 err = unmap_grant_pages(map,
424 (mstart - map->vma->vm_start) >> PAGE_SHIFT,
425 (mend - mstart) >> PAGE_SHIFT);
426 WARN_ON(err);
427}
428
Gerd Hoffmannab315232010-12-14 18:40:46 +0000429static void mn_invl_range_start(struct mmu_notifier *mn,
430 struct mm_struct *mm,
431 unsigned long start, unsigned long end)
432{
433 struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
434 struct grant_map *map;
Gerd Hoffmannab315232010-12-14 18:40:46 +0000435
436 spin_lock(&priv->lock);
437 list_for_each_entry(map, &priv->maps, next) {
Daniel De Graaf16a1d022013-01-02 22:57:12 +0000438 unmap_if_in_range(map, start, end);
439 }
440 list_for_each_entry(map, &priv->freeable_maps, next) {
441 unmap_if_in_range(map, start, end);
Gerd Hoffmannab315232010-12-14 18:40:46 +0000442 }
443 spin_unlock(&priv->lock);
444}
445
446static void mn_invl_page(struct mmu_notifier *mn,
447 struct mm_struct *mm,
448 unsigned long address)
449{
450 mn_invl_range_start(mn, mm, address, address + PAGE_SIZE);
451}
452
453static void mn_release(struct mmu_notifier *mn,
454 struct mm_struct *mm)
455{
456 struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn);
457 struct grant_map *map;
458 int err;
459
460 spin_lock(&priv->lock);
461 list_for_each_entry(map, &priv->maps, next) {
462 if (!map->vma)
463 continue;
464 pr_debug("map %d+%d (%lx %lx)\n",
465 map->index, map->count,
466 map->vma->vm_start, map->vma->vm_end);
467 err = unmap_grant_pages(map, /* offset */ 0, map->count);
468 WARN_ON(err);
469 }
Daniel De Graaf16a1d022013-01-02 22:57:12 +0000470 list_for_each_entry(map, &priv->freeable_maps, next) {
471 if (!map->vma)
472 continue;
473 pr_debug("map %d+%d (%lx %lx)\n",
474 map->index, map->count,
475 map->vma->vm_start, map->vma->vm_end);
476 err = unmap_grant_pages(map, /* offset */ 0, map->count);
477 WARN_ON(err);
478 }
Gerd Hoffmannab315232010-12-14 18:40:46 +0000479 spin_unlock(&priv->lock);
480}
481
Konrad Rzeszutek Wilkb8b0f552012-08-21 14:49:34 -0400482static struct mmu_notifier_ops gntdev_mmu_ops = {
Gerd Hoffmannab315232010-12-14 18:40:46 +0000483 .release = mn_release,
484 .invalidate_page = mn_invl_page,
485 .invalidate_range_start = mn_invl_range_start,
486};
487
488/* ------------------------------------------------------------------ */
489
490static int gntdev_open(struct inode *inode, struct file *flip)
491{
492 struct gntdev_priv *priv;
493 int ret = 0;
494
495 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
496 if (!priv)
497 return -ENOMEM;
498
499 INIT_LIST_HEAD(&priv->maps);
Daniel De Graaf16a1d022013-01-02 22:57:12 +0000500 INIT_LIST_HEAD(&priv->freeable_maps);
Gerd Hoffmannab315232010-12-14 18:40:46 +0000501 spin_lock_init(&priv->lock);
Gerd Hoffmannab315232010-12-14 18:40:46 +0000502
Daniel De Graafaab8f112011-02-03 12:19:02 -0500503 if (use_ptemod) {
504 priv->mm = get_task_mm(current);
505 if (!priv->mm) {
506 kfree(priv);
507 return -ENOMEM;
508 }
509 priv->mn.ops = &gntdev_mmu_ops;
510 ret = mmu_notifier_register(&priv->mn, priv->mm);
511 mmput(priv->mm);
Gerd Hoffmannab315232010-12-14 18:40:46 +0000512 }
Gerd Hoffmannab315232010-12-14 18:40:46 +0000513
514 if (ret) {
515 kfree(priv);
516 return ret;
517 }
518
519 flip->private_data = priv;
520 pr_debug("priv %p\n", priv);
521
522 return 0;
523}
524
525static int gntdev_release(struct inode *inode, struct file *flip)
526{
527 struct gntdev_priv *priv = flip->private_data;
528 struct grant_map *map;
Gerd Hoffmannab315232010-12-14 18:40:46 +0000529
530 pr_debug("priv %p\n", priv);
531
Gerd Hoffmannab315232010-12-14 18:40:46 +0000532 while (!list_empty(&priv->maps)) {
533 map = list_entry(priv->maps.next, struct grant_map, next);
Daniel De Graaf68b025c2011-02-03 12:19:01 -0500534 list_del(&map->next);
Daniel De Graaf16a1d022013-01-02 22:57:12 +0000535 gntdev_put_map(NULL /* already removed */, map);
Gerd Hoffmannab315232010-12-14 18:40:46 +0000536 }
Daniel De Graaf16a1d022013-01-02 22:57:12 +0000537 WARN_ON(!list_empty(&priv->freeable_maps));
Gerd Hoffmannab315232010-12-14 18:40:46 +0000538
Daniel De Graafaab8f112011-02-03 12:19:02 -0500539 if (use_ptemod)
540 mmu_notifier_unregister(&priv->mn, priv->mm);
Gerd Hoffmannab315232010-12-14 18:40:46 +0000541 kfree(priv);
542 return 0;
543}
544
545static long gntdev_ioctl_map_grant_ref(struct gntdev_priv *priv,
546 struct ioctl_gntdev_map_grant_ref __user *u)
547{
548 struct ioctl_gntdev_map_grant_ref op;
549 struct grant_map *map;
550 int err;
551
552 if (copy_from_user(&op, u, sizeof(op)) != 0)
553 return -EFAULT;
554 pr_debug("priv %p, add %d\n", priv, op.count);
555 if (unlikely(op.count <= 0))
556 return -EINVAL;
Gerd Hoffmannab315232010-12-14 18:40:46 +0000557
558 err = -ENOMEM;
559 map = gntdev_alloc_map(priv, op.count);
560 if (!map)
561 return err;
Daniel De Graafef910822011-02-03 12:18:59 -0500562
Daniel De Graaf68b025c2011-02-03 12:19:01 -0500563 if (unlikely(atomic_add_return(op.count, &pages_mapped) > limit)) {
564 pr_debug("can't map: over limit\n");
Daniel De Graaf16a1d022013-01-02 22:57:12 +0000565 gntdev_put_map(NULL, map);
Gerd Hoffmannab315232010-12-14 18:40:46 +0000566 return err;
567 }
568
Daniel De Graaf68b025c2011-02-03 12:19:01 -0500569 if (copy_from_user(map->grants, &u->refs,
570 sizeof(map->grants[0]) * op.count) != 0) {
Daniel De Graaf16a1d022013-01-02 22:57:12 +0000571 gntdev_put_map(NULL, map);
572 return -EFAULT;
Daniel De Graafef910822011-02-03 12:18:59 -0500573 }
574
Gerd Hoffmannab315232010-12-14 18:40:46 +0000575 spin_lock(&priv->lock);
576 gntdev_add_map(priv, map);
577 op.index = map->index << PAGE_SHIFT;
578 spin_unlock(&priv->lock);
579
Daniel De Graaf68b025c2011-02-03 12:19:01 -0500580 if (copy_to_user(u, &op, sizeof(op)) != 0)
581 return -EFAULT;
582
Gerd Hoffmannab315232010-12-14 18:40:46 +0000583 return 0;
584}
585
586static long gntdev_ioctl_unmap_grant_ref(struct gntdev_priv *priv,
587 struct ioctl_gntdev_unmap_grant_ref __user *u)
588{
589 struct ioctl_gntdev_unmap_grant_ref op;
590 struct grant_map *map;
591 int err = -ENOENT;
592
593 if (copy_from_user(&op, u, sizeof(op)) != 0)
594 return -EFAULT;
595 pr_debug("priv %p, del %d+%d\n", priv, (int)op.index, (int)op.count);
596
597 spin_lock(&priv->lock);
598 map = gntdev_find_map_index(priv, op.index >> PAGE_SHIFT, op.count);
Daniel De Graaf68b025c2011-02-03 12:19:01 -0500599 if (map) {
600 list_del(&map->next);
Daniel De Graaf16a1d022013-01-02 22:57:12 +0000601 if (populate_freeable_maps)
602 list_add_tail(&map->next, &priv->freeable_maps);
Daniel De Graaf68b025c2011-02-03 12:19:01 -0500603 err = 0;
604 }
Gerd Hoffmannab315232010-12-14 18:40:46 +0000605 spin_unlock(&priv->lock);
Daniel De Graaf1f1503b2011-10-11 15:16:06 -0400606 if (map)
Daniel De Graaf16a1d022013-01-02 22:57:12 +0000607 gntdev_put_map(priv, map);
Gerd Hoffmannab315232010-12-14 18:40:46 +0000608 return err;
609}
610
611static long gntdev_ioctl_get_offset_for_vaddr(struct gntdev_priv *priv,
612 struct ioctl_gntdev_get_offset_for_vaddr __user *u)
613{
614 struct ioctl_gntdev_get_offset_for_vaddr op;
Daniel De Graafa8792112011-02-03 12:19:00 -0500615 struct vm_area_struct *vma;
Gerd Hoffmannab315232010-12-14 18:40:46 +0000616 struct grant_map *map;
Daniel De Graaf2512f292013-01-02 22:57:11 +0000617 int rv = -EINVAL;
Gerd Hoffmannab315232010-12-14 18:40:46 +0000618
619 if (copy_from_user(&op, u, sizeof(op)) != 0)
620 return -EFAULT;
621 pr_debug("priv %p, offset for vaddr %lx\n", priv, (unsigned long)op.vaddr);
622
Daniel De Graaf2512f292013-01-02 22:57:11 +0000623 down_read(&current->mm->mmap_sem);
Daniel De Graafa8792112011-02-03 12:19:00 -0500624 vma = find_vma(current->mm, op.vaddr);
625 if (!vma || vma->vm_ops != &gntdev_vmops)
Daniel De Graaf2512f292013-01-02 22:57:11 +0000626 goto out_unlock;
Daniel De Graafa8792112011-02-03 12:19:00 -0500627
628 map = vma->vm_private_data;
629 if (!map)
Daniel De Graaf2512f292013-01-02 22:57:11 +0000630 goto out_unlock;
Daniel De Graafa8792112011-02-03 12:19:00 -0500631
Gerd Hoffmannab315232010-12-14 18:40:46 +0000632 op.offset = map->index << PAGE_SHIFT;
633 op.count = map->count;
Daniel De Graaf2512f292013-01-02 22:57:11 +0000634 rv = 0;
Gerd Hoffmannab315232010-12-14 18:40:46 +0000635
Daniel De Graaf2512f292013-01-02 22:57:11 +0000636 out_unlock:
637 up_read(&current->mm->mmap_sem);
638
639 if (rv == 0 && copy_to_user(u, &op, sizeof(op)) != 0)
Gerd Hoffmannab315232010-12-14 18:40:46 +0000640 return -EFAULT;
Daniel De Graaf2512f292013-01-02 22:57:11 +0000641 return rv;
Gerd Hoffmannab315232010-12-14 18:40:46 +0000642}
643
Daniel De Graafbdc612d2011-02-03 12:19:04 -0500644static long gntdev_ioctl_notify(struct gntdev_priv *priv, void __user *u)
645{
646 struct ioctl_gntdev_unmap_notify op;
647 struct grant_map *map;
648 int rc;
Daniel De Graaf0cc678f2011-10-27 17:58:49 -0400649 int out_flags;
650 unsigned int out_event;
Daniel De Graafbdc612d2011-02-03 12:19:04 -0500651
652 if (copy_from_user(&op, u, sizeof(op)))
653 return -EFAULT;
654
655 if (op.action & ~(UNMAP_NOTIFY_CLEAR_BYTE|UNMAP_NOTIFY_SEND_EVENT))
656 return -EINVAL;
657
Daniel De Graaf0cc678f2011-10-27 17:58:49 -0400658 /* We need to grab a reference to the event channel we are going to use
659 * to send the notify before releasing the reference we may already have
660 * (if someone has called this ioctl twice). This is required so that
661 * it is possible to change the clear_byte part of the notification
662 * without disturbing the event channel part, which may now be the last
663 * reference to that event channel.
664 */
665 if (op.action & UNMAP_NOTIFY_SEND_EVENT) {
666 if (evtchn_get(op.event_channel_port))
667 return -EINVAL;
668 }
669
670 out_flags = op.action;
671 out_event = op.event_channel_port;
672
Daniel De Graafbdc612d2011-02-03 12:19:04 -0500673 spin_lock(&priv->lock);
674
675 list_for_each_entry(map, &priv->maps, next) {
676 uint64_t begin = map->index << PAGE_SHIFT;
677 uint64_t end = (map->index + map->count) << PAGE_SHIFT;
678 if (op.index >= begin && op.index < end)
679 goto found;
680 }
681 rc = -ENOENT;
682 goto unlock_out;
683
684 found:
Daniel De Graaf9960be92011-02-09 18:15:50 -0500685 if ((op.action & UNMAP_NOTIFY_CLEAR_BYTE) &&
686 (map->flags & GNTMAP_readonly)) {
687 rc = -EINVAL;
688 goto unlock_out;
689 }
690
Daniel De Graaf0cc678f2011-10-27 17:58:49 -0400691 out_flags = map->notify.flags;
692 out_event = map->notify.event;
693
Daniel De Graafbdc612d2011-02-03 12:19:04 -0500694 map->notify.flags = op.action;
695 map->notify.addr = op.index - (map->index << PAGE_SHIFT);
696 map->notify.event = op.event_channel_port;
Daniel De Graaf0cc678f2011-10-27 17:58:49 -0400697
Daniel De Graafbdc612d2011-02-03 12:19:04 -0500698 rc = 0;
Daniel De Graaf0cc678f2011-10-27 17:58:49 -0400699
Daniel De Graafbdc612d2011-02-03 12:19:04 -0500700 unlock_out:
701 spin_unlock(&priv->lock);
Daniel De Graaf0cc678f2011-10-27 17:58:49 -0400702
703 /* Drop the reference to the event channel we did not save in the map */
704 if (out_flags & UNMAP_NOTIFY_SEND_EVENT)
705 evtchn_put(out_event);
706
Daniel De Graafbdc612d2011-02-03 12:19:04 -0500707 return rc;
708}
709
Gerd Hoffmannab315232010-12-14 18:40:46 +0000710static long gntdev_ioctl(struct file *flip,
711 unsigned int cmd, unsigned long arg)
712{
713 struct gntdev_priv *priv = flip->private_data;
714 void __user *ptr = (void __user *)arg;
715
716 switch (cmd) {
717 case IOCTL_GNTDEV_MAP_GRANT_REF:
718 return gntdev_ioctl_map_grant_ref(priv, ptr);
719
720 case IOCTL_GNTDEV_UNMAP_GRANT_REF:
721 return gntdev_ioctl_unmap_grant_ref(priv, ptr);
722
723 case IOCTL_GNTDEV_GET_OFFSET_FOR_VADDR:
724 return gntdev_ioctl_get_offset_for_vaddr(priv, ptr);
725
Daniel De Graafbdc612d2011-02-03 12:19:04 -0500726 case IOCTL_GNTDEV_SET_UNMAP_NOTIFY:
727 return gntdev_ioctl_notify(priv, ptr);
728
Gerd Hoffmannab315232010-12-14 18:40:46 +0000729 default:
730 pr_debug("priv %p, unknown cmd %x\n", priv, cmd);
731 return -ENOIOCTLCMD;
732 }
733
734 return 0;
735}
736
737static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma)
738{
739 struct gntdev_priv *priv = flip->private_data;
740 int index = vma->vm_pgoff;
741 int count = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
742 struct grant_map *map;
Daniel De Graafaab8f112011-02-03 12:19:02 -0500743 int i, err = -EINVAL;
Gerd Hoffmannab315232010-12-14 18:40:46 +0000744
745 if ((vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_SHARED))
746 return -EINVAL;
747
748 pr_debug("map %d+%d at %lx (pgoff %lx)\n",
749 index, count, vma->vm_start, vma->vm_pgoff);
750
751 spin_lock(&priv->lock);
752 map = gntdev_find_map_index(priv, index, count);
753 if (!map)
754 goto unlock_out;
Daniel De Graafaab8f112011-02-03 12:19:02 -0500755 if (use_ptemod && map->vma)
Gerd Hoffmannab315232010-12-14 18:40:46 +0000756 goto unlock_out;
Daniel De Graafaab8f112011-02-03 12:19:02 -0500757 if (use_ptemod && priv->mm != vma->vm_mm) {
Joe Perches283c0972013-06-28 03:21:41 -0700758 pr_warn("Huh? Other mm?\n");
Gerd Hoffmannab315232010-12-14 18:40:46 +0000759 goto unlock_out;
760 }
761
Daniel De Graaf68b025c2011-02-03 12:19:01 -0500762 atomic_inc(&map->users);
763
Gerd Hoffmannab315232010-12-14 18:40:46 +0000764 vma->vm_ops = &gntdev_vmops;
765
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -0700766 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
Daniel De Graafd79647a2011-03-07 15:18:57 -0500767
768 if (use_ptemod)
Stefano Stabellinie8e937b2012-04-03 18:05:47 +0100769 vma->vm_flags |= VM_DONTCOPY;
Gerd Hoffmannab315232010-12-14 18:40:46 +0000770
771 vma->vm_private_data = map;
Gerd Hoffmannab315232010-12-14 18:40:46 +0000772
Daniel De Graafaab8f112011-02-03 12:19:02 -0500773 if (use_ptemod)
774 map->vma = vma;
775
Daniel De Graaf12996fc2011-02-09 16:11:32 -0500776 if (map->flags) {
777 if ((vma->vm_flags & VM_WRITE) &&
778 (map->flags & GNTMAP_readonly))
Dan Carpentera93e20a2011-03-19 08:45:43 +0300779 goto out_unlock_put;
Daniel De Graaf12996fc2011-02-09 16:11:32 -0500780 } else {
781 map->flags = GNTMAP_host_map;
782 if (!(vma->vm_flags & VM_WRITE))
783 map->flags |= GNTMAP_readonly;
784 }
Gerd Hoffmannab315232010-12-14 18:40:46 +0000785
Daniel De Graaff0a70c82011-01-07 11:51:47 +0000786 spin_unlock(&priv->lock);
787
Daniel De Graafaab8f112011-02-03 12:19:02 -0500788 if (use_ptemod) {
789 err = apply_to_page_range(vma->vm_mm, vma->vm_start,
790 vma->vm_end - vma->vm_start,
791 find_grant_ptes, map);
792 if (err) {
Joe Perches283c0972013-06-28 03:21:41 -0700793 pr_warn("find_grant_ptes() failure.\n");
Daniel De Graaf90b6f302011-02-03 14:16:54 -0500794 goto out_put_map;
Daniel De Graafaab8f112011-02-03 12:19:02 -0500795 }
Gerd Hoffmannab315232010-12-14 18:40:46 +0000796 }
797
798 err = map_grant_pages(map);
Daniel De Graaf90b6f302011-02-03 14:16:54 -0500799 if (err)
800 goto out_put_map;
Daniel De Graaff0a70c82011-01-07 11:51:47 +0000801
Daniel De Graafaab8f112011-02-03 12:19:02 -0500802 if (!use_ptemod) {
803 for (i = 0; i < count; i++) {
804 err = vm_insert_page(vma, vma->vm_start + i*PAGE_SIZE,
805 map->pages[i]);
806 if (err)
Daniel De Graaf90b6f302011-02-03 14:16:54 -0500807 goto out_put_map;
Daniel De Graafaab8f112011-02-03 12:19:02 -0500808 }
809 }
810
Daniel De Graaff0a70c82011-01-07 11:51:47 +0000811 return 0;
812
Gerd Hoffmannab315232010-12-14 18:40:46 +0000813unlock_out:
814 spin_unlock(&priv->lock);
815 return err;
Daniel De Graaf90b6f302011-02-03 14:16:54 -0500816
Dan Carpentera93e20a2011-03-19 08:45:43 +0300817out_unlock_put:
818 spin_unlock(&priv->lock);
Daniel De Graaf90b6f302011-02-03 14:16:54 -0500819out_put_map:
Daniel De Graaf84e40752011-02-09 15:11:59 -0500820 if (use_ptemod)
821 map->vma = NULL;
Daniel De Graaf16a1d022013-01-02 22:57:12 +0000822 gntdev_put_map(priv, map);
Daniel De Graaf90b6f302011-02-03 14:16:54 -0500823 return err;
Gerd Hoffmannab315232010-12-14 18:40:46 +0000824}
825
826static const struct file_operations gntdev_fops = {
827 .owner = THIS_MODULE,
828 .open = gntdev_open,
829 .release = gntdev_release,
830 .mmap = gntdev_mmap,
831 .unlocked_ioctl = gntdev_ioctl
832};
833
834static struct miscdevice gntdev_miscdev = {
835 .minor = MISC_DYNAMIC_MINOR,
836 .name = "xen/gntdev",
837 .fops = &gntdev_fops,
838};
839
840/* ------------------------------------------------------------------ */
841
842static int __init gntdev_init(void)
843{
844 int err;
845
846 if (!xen_domain())
847 return -ENODEV;
848
Daniel De Graafaab8f112011-02-03 12:19:02 -0500849 use_ptemod = xen_pv_domain();
850
Gerd Hoffmannab315232010-12-14 18:40:46 +0000851 err = misc_register(&gntdev_miscdev);
852 if (err != 0) {
Joe Perches283c0972013-06-28 03:21:41 -0700853 pr_err("Could not register gntdev device\n");
Gerd Hoffmannab315232010-12-14 18:40:46 +0000854 return err;
855 }
856 return 0;
857}
858
859static void __exit gntdev_exit(void)
860{
861 misc_deregister(&gntdev_miscdev);
862}
863
864module_init(gntdev_init);
865module_exit(gntdev_exit);
866
867/* ------------------------------------------------------------------ */