]>
Commit | Line | Data |
---|---|---|
ab31523c GH |
1 | /****************************************************************************** |
2 | * gntdev.c | |
3 | * | |
4 | * Device for accessing (in user-space) pages that have been granted by other | |
5 | * domains. | |
6 | * | |
7 | * Copyright (c) 2006-2007, D G Murray. | |
8 | * (c) 2009 Gerd Hoffmann <kraxel@redhat.com> | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License | |
16 | * along with this program; if not, write to the Free Software | |
17 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
18 | */ | |
19 | ||
20 | #undef DEBUG | |
21 | ||
283c0972 JP |
22 | #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt |
23 | ||
ab31523c GH |
24 | #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> | |
aab8f11a | 37 | #include <linux/highmem.h> |
ab31523c GH |
38 | |
39 | #include <xen/xen.h> | |
40 | #include <xen/grant_table.h> | |
ca47ceaa | 41 | #include <xen/balloon.h> |
ab31523c | 42 | #include <xen/gntdev.h> |
bdc612dc | 43 | #include <xen/events.h> |
ab31523c GH |
44 | #include <asm/xen/hypervisor.h> |
45 | #include <asm/xen/hypercall.h> | |
46 | #include <asm/xen/page.h> | |
47 | ||
48 | MODULE_LICENSE("GPL"); | |
49 | MODULE_AUTHOR("Derek G. Murray <Derek.Murray@cl.cam.ac.uk>, " | |
50 | "Gerd Hoffmann <kraxel@redhat.com>"); | |
51 | MODULE_DESCRIPTION("User-space granted page access driver"); | |
52 | ||
ef91082e | 53 | static int limit = 1024*1024; |
ab31523c | 54 | module_param(limit, int, 0644); |
ef91082e DDG |
55 | MODULE_PARM_DESC(limit, "Maximum number of grants that may be mapped by " |
56 | "the gntdev device"); | |
57 | ||
58 | static atomic_t pages_mapped = ATOMIC_INIT(0); | |
ab31523c | 59 | |
aab8f11a | 60 | static int use_ptemod; |
16a1d022 | 61 | #define populate_freeable_maps use_ptemod |
aab8f11a | 62 | |
ab31523c | 63 | struct gntdev_priv { |
16a1d022 | 64 | /* maps with visible offsets in the file descriptor */ |
ab31523c | 65 | struct list_head maps; |
16a1d022 DDG |
66 | /* 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 */ | |
ab31523c GH |
70 | spinlock_t lock; |
71 | struct mm_struct *mm; | |
72 | struct mmu_notifier mn; | |
73 | }; | |
74 | ||
bdc612dc DDG |
75 | struct unmap_notify { |
76 | int flags; | |
77 | /* Address relative to the start of the grant_map */ | |
78 | int addr; | |
79 | int event; | |
80 | }; | |
81 | ||
ab31523c GH |
82 | struct grant_map { |
83 | struct list_head next; | |
ab31523c GH |
84 | struct vm_area_struct *vma; |
85 | int index; | |
86 | int count; | |
87 | int flags; | |
68b025c8 | 88 | atomic_t users; |
bdc612dc | 89 | struct unmap_notify notify; |
ab31523c GH |
90 | struct ioctl_gntdev_grant_ref *grants; |
91 | struct gnttab_map_grant_ref *map_ops; | |
92 | struct gnttab_unmap_grant_ref *unmap_ops; | |
0930bba6 | 93 | struct gnttab_map_grant_ref *kmap_ops; |
853d0289 | 94 | struct gnttab_unmap_grant_ref *kunmap_ops; |
a12b4eb3 | 95 | struct page **pages; |
ab31523c GH |
96 | }; |
97 | ||
aab8f11a DDG |
98 | static int unmap_grant_pages(struct grant_map *map, int offset, int pages); |
99 | ||
ab31523c GH |
100 | /* ------------------------------------------------------------------ */ |
101 | ||
102 | static void gntdev_print_maps(struct gntdev_priv *priv, | |
103 | char *text, int text_index) | |
104 | { | |
105 | #ifdef DEBUG | |
106 | struct grant_map *map; | |
107 | ||
ef91082e | 108 | pr_debug("%s: maps list (priv %p)\n", __func__, priv); |
ab31523c GH |
109 | list_for_each_entry(map, &priv->maps, next) |
110 | pr_debug(" index %2d, count %2d %s\n", | |
111 | map->index, map->count, | |
112 | map->index == text_index && text ? text : ""); | |
113 | #endif | |
114 | } | |
115 | ||
a67baeb7 DV |
116 | static void gntdev_free_map(struct grant_map *map) |
117 | { | |
118 | if (map == NULL) | |
119 | return; | |
120 | ||
121 | if (map->pages) | |
122 | free_xenballooned_pages(map->count, map->pages); | |
123 | kfree(map->pages); | |
124 | kfree(map->grants); | |
125 | kfree(map->map_ops); | |
126 | kfree(map->unmap_ops); | |
127 | kfree(map->kmap_ops); | |
853d0289 | 128 | kfree(map->kunmap_ops); |
a67baeb7 DV |
129 | kfree(map); |
130 | } | |
131 | ||
ab31523c GH |
132 | static struct grant_map *gntdev_alloc_map(struct gntdev_priv *priv, int count) |
133 | { | |
134 | struct grant_map *add; | |
a12b4eb3 | 135 | int i; |
ab31523c GH |
136 | |
137 | add = kzalloc(sizeof(struct grant_map), GFP_KERNEL); | |
138 | if (NULL == add) | |
139 | return NULL; | |
140 | ||
fc6e0c3b DC |
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); | |
853d0289 | 145 | add->kunmap_ops = kcalloc(count, sizeof(add->kunmap_ops[0]), GFP_KERNEL); |
fc6e0c3b | 146 | add->pages = kcalloc(count, sizeof(add->pages[0]), GFP_KERNEL); |
a12b4eb3 SS |
147 | if (NULL == add->grants || |
148 | NULL == add->map_ops || | |
149 | NULL == add->unmap_ops || | |
0930bba6 | 150 | NULL == add->kmap_ops || |
853d0289 | 151 | NULL == add->kunmap_ops || |
a12b4eb3 | 152 | NULL == add->pages) |
ab31523c GH |
153 | goto err; |
154 | ||
693394b8 | 155 | if (alloc_xenballooned_pages(count, add->pages, false /* lowmem */)) |
ca47ceaa DDG |
156 | goto err; |
157 | ||
a12b4eb3 | 158 | for (i = 0; i < count; i++) { |
77c35acb DDG |
159 | add->map_ops[i].handle = -1; |
160 | add->unmap_ops[i].handle = -1; | |
0930bba6 | 161 | add->kmap_ops[i].handle = -1; |
853d0289 | 162 | add->kunmap_ops[i].handle = -1; |
a12b4eb3 SS |
163 | } |
164 | ||
ab31523c GH |
165 | add->index = 0; |
166 | add->count = count; | |
68b025c8 | 167 | atomic_set(&add->users, 1); |
ab31523c | 168 | |
ab31523c GH |
169 | return add; |
170 | ||
171 | err: | |
a67baeb7 | 172 | gntdev_free_map(add); |
ab31523c GH |
173 | return NULL; |
174 | } | |
175 | ||
176 | static void gntdev_add_map(struct gntdev_priv *priv, struct grant_map *add) | |
177 | { | |
178 | struct grant_map *map; | |
179 | ||
180 | list_for_each_entry(map, &priv->maps, next) { | |
181 | if (add->index + add->count < map->index) { | |
182 | list_add_tail(&add->next, &map->next); | |
183 | goto done; | |
184 | } | |
185 | add->index = map->index + map->count; | |
186 | } | |
187 | list_add_tail(&add->next, &priv->maps); | |
188 | ||
189 | done: | |
ab31523c GH |
190 | gntdev_print_maps(priv, "[new]", add->index); |
191 | } | |
192 | ||
193 | static struct grant_map *gntdev_find_map_index(struct gntdev_priv *priv, | |
194 | int index, int count) | |
195 | { | |
196 | struct grant_map *map; | |
197 | ||
198 | list_for_each_entry(map, &priv->maps, next) { | |
199 | if (map->index != index) | |
200 | continue; | |
bdc612dc | 201 | if (count && map->count != count) |
ab31523c GH |
202 | continue; |
203 | return map; | |
204 | } | |
205 | return NULL; | |
206 | } | |
207 | ||
16a1d022 | 208 | static void gntdev_put_map(struct gntdev_priv *priv, struct grant_map *map) |
ab31523c GH |
209 | { |
210 | if (!map) | |
211 | return; | |
a12b4eb3 | 212 | |
68b025c8 DDG |
213 | if (!atomic_dec_and_test(&map->users)) |
214 | return; | |
215 | ||
216 | atomic_sub(map->count, &pages_mapped); | |
217 | ||
0cc678f8 | 218 | if (map->notify.flags & UNMAP_NOTIFY_SEND_EVENT) { |
bdc612dc | 219 | notify_remote_via_evtchn(map->notify.event); |
0cc678f8 DDG |
220 | evtchn_put(map->notify.event); |
221 | } | |
bdc612dc | 222 | |
16a1d022 DDG |
223 | if (populate_freeable_maps && priv) { |
224 | spin_lock(&priv->lock); | |
225 | list_del(&map->next); | |
226 | spin_unlock(&priv->lock); | |
227 | } | |
228 | ||
a67baeb7 DV |
229 | if (map->pages && !use_ptemod) |
230 | unmap_grant_pages(map, 0, map->count); | |
231 | gntdev_free_map(map); | |
ab31523c GH |
232 | } |
233 | ||
234 | /* ------------------------------------------------------------------ */ | |
235 | ||
236 | static int find_grant_ptes(pte_t *pte, pgtable_t token, | |
237 | unsigned long addr, void *data) | |
238 | { | |
239 | struct grant_map *map = data; | |
240 | unsigned int pgnr = (addr - map->vma->vm_start) >> PAGE_SHIFT; | |
aab8f11a | 241 | int flags = map->flags | GNTMAP_application_map | GNTMAP_contains_pte; |
ab31523c GH |
242 | u64 pte_maddr; |
243 | ||
244 | BUG_ON(pgnr >= map->count); | |
ba5d1012 JF |
245 | pte_maddr = arbitrary_virt_to_machine(pte).maddr; |
246 | ||
aab8f11a | 247 | gnttab_set_map_op(&map->map_ops[pgnr], pte_maddr, flags, |
ab31523c GH |
248 | map->grants[pgnr].ref, |
249 | map->grants[pgnr].domid); | |
aab8f11a | 250 | gnttab_set_unmap_op(&map->unmap_ops[pgnr], pte_maddr, flags, |
77c35acb | 251 | -1 /* handle */); |
ab31523c GH |
252 | return 0; |
253 | } | |
254 | ||
255 | static int map_grant_pages(struct grant_map *map) | |
256 | { | |
257 | int i, err = 0; | |
aab8f11a DDG |
258 | |
259 | if (!use_ptemod) { | |
12996fc3 | 260 | /* Note: it could already be mapped */ |
77c35acb | 261 | if (map->map_ops[0].handle != -1) |
12996fc3 | 262 | return 0; |
aab8f11a | 263 | for (i = 0; i < map->count; i++) { |
38eaeb0f | 264 | unsigned long addr = (unsigned long) |
aab8f11a DDG |
265 | pfn_to_kaddr(page_to_pfn(map->pages[i])); |
266 | gnttab_set_map_op(&map->map_ops[i], addr, map->flags, | |
267 | map->grants[i].ref, | |
268 | map->grants[i].domid); | |
269 | gnttab_set_unmap_op(&map->unmap_ops[i], addr, | |
77c35acb | 270 | map->flags, -1 /* handle */); |
aab8f11a | 271 | } |
0930bba6 SS |
272 | } else { |
273 | /* | |
274 | * Setup the map_ops corresponding to the pte entries pointing | |
275 | * to the kernel linear addresses of the struct pages. | |
276 | * These ptes are completely different from the user ptes dealt | |
277 | * with find_grant_ptes. | |
278 | */ | |
279 | for (i = 0; i < map->count; i++) { | |
0930bba6 SS |
280 | unsigned long address = (unsigned long) |
281 | pfn_to_kaddr(page_to_pfn(map->pages[i])); | |
0930bba6 SS |
282 | BUG_ON(PageHighMem(map->pages[i])); |
283 | ||
ee072640 SS |
284 | gnttab_set_map_op(&map->kmap_ops[i], address, |
285 | map->flags | GNTMAP_host_map, | |
0930bba6 SS |
286 | map->grants[i].ref, |
287 | map->grants[i].domid); | |
853d0289 DV |
288 | gnttab_set_unmap_op(&map->kunmap_ops[i], address, |
289 | map->flags | GNTMAP_host_map, -1); | |
0930bba6 | 290 | } |
aab8f11a | 291 | } |
ab31523c GH |
292 | |
293 | pr_debug("map %d+%d\n", map->index, map->count); | |
e85fc980 KRW |
294 | err = gnttab_map_refs(map->map_ops, use_ptemod ? map->kmap_ops : NULL, |
295 | map->pages, map->count); | |
ab31523c GH |
296 | if (err) |
297 | return err; | |
298 | ||
299 | for (i = 0; i < map->count; i++) { | |
853d0289 | 300 | if (map->map_ops[i].status) { |
ab31523c | 301 | err = -EINVAL; |
853d0289 | 302 | continue; |
77c35acb | 303 | } |
853d0289 DV |
304 | |
305 | map->unmap_ops[i].handle = map->map_ops[i].handle; | |
306 | if (use_ptemod) | |
307 | map->kunmap_ops[i].handle = map->kmap_ops[i].handle; | |
ab31523c GH |
308 | } |
309 | return err; | |
310 | } | |
311 | ||
b57c1869 | 312 | static int __unmap_grant_pages(struct grant_map *map, int offset, int pages) |
ab31523c GH |
313 | { |
314 | int i, err = 0; | |
315 | ||
bdc612dc DDG |
316 | if (map->notify.flags & UNMAP_NOTIFY_CLEAR_BYTE) { |
317 | int pgno = (map->notify.addr >> PAGE_SHIFT); | |
1affa98d DDG |
318 | if (pgno >= offset && pgno < offset + pages) { |
319 | /* No need for kmap, pages are in lowmem */ | |
320 | uint8_t *tmp = pfn_to_kaddr(page_to_pfn(map->pages[pgno])); | |
bdc612dc | 321 | tmp[map->notify.addr & (PAGE_SIZE-1)] = 0; |
bdc612dc DDG |
322 | map->notify.flags &= ~UNMAP_NOTIFY_CLEAR_BYTE; |
323 | } | |
324 | } | |
325 | ||
e85fc980 | 326 | err = gnttab_unmap_refs(map->unmap_ops + offset, |
853d0289 | 327 | use_ptemod ? map->kunmap_ops + offset : NULL, map->pages + offset, |
e85fc980 | 328 | pages); |
ab31523c GH |
329 | if (err) |
330 | return err; | |
331 | ||
332 | for (i = 0; i < pages; i++) { | |
333 | if (map->unmap_ops[offset+i].status) | |
334 | err = -EINVAL; | |
77c35acb DDG |
335 | pr_debug("unmap handle=%d st=%d\n", |
336 | map->unmap_ops[offset+i].handle, | |
337 | map->unmap_ops[offset+i].status); | |
338 | map->unmap_ops[offset+i].handle = -1; | |
ab31523c GH |
339 | } |
340 | return err; | |
341 | } | |
342 | ||
b57c1869 DDG |
343 | static int unmap_grant_pages(struct grant_map *map, int offset, int pages) |
344 | { | |
345 | int range, err = 0; | |
346 | ||
347 | pr_debug("unmap %d+%d [%d+%d]\n", map->index, map->count, offset, pages); | |
348 | ||
349 | /* It is possible the requested range will have a "hole" where we | |
350 | * already unmapped some of the grants. Only unmap valid ranges. | |
351 | */ | |
352 | while (pages && !err) { | |
77c35acb | 353 | while (pages && map->unmap_ops[offset].handle == -1) { |
b57c1869 DDG |
354 | offset++; |
355 | pages--; | |
356 | } | |
357 | range = 0; | |
358 | while (range < pages) { | |
77c35acb | 359 | if (map->unmap_ops[offset+range].handle == -1) { |
b57c1869 DDG |
360 | range--; |
361 | break; | |
362 | } | |
363 | range++; | |
364 | } | |
365 | err = __unmap_grant_pages(map, offset, range); | |
366 | offset += range; | |
367 | pages -= range; | |
368 | } | |
369 | ||
370 | return err; | |
371 | } | |
372 | ||
ab31523c GH |
373 | /* ------------------------------------------------------------------ */ |
374 | ||
d79647ae DDG |
375 | static void gntdev_vma_open(struct vm_area_struct *vma) |
376 | { | |
377 | struct grant_map *map = vma->vm_private_data; | |
378 | ||
379 | pr_debug("gntdev_vma_open %p\n", vma); | |
380 | atomic_inc(&map->users); | |
381 | } | |
382 | ||
ab31523c GH |
383 | static void gntdev_vma_close(struct vm_area_struct *vma) |
384 | { | |
385 | struct grant_map *map = vma->vm_private_data; | |
16a1d022 DDG |
386 | struct file *file = vma->vm_file; |
387 | struct gntdev_priv *priv = file->private_data; | |
ab31523c | 388 | |
d79647ae | 389 | pr_debug("gntdev_vma_close %p\n", vma); |
2512f298 | 390 | if (use_ptemod) { |
2512f298 DDG |
391 | /* It is possible that an mmu notifier could be running |
392 | * concurrently, so take priv->lock to ensure that the vma won't | |
393 | * vanishing during the unmap_grant_pages call, since we will | |
394 | * spin here until that completes. Such a concurrent call will | |
395 | * not do any unmapping, since that has been done prior to | |
396 | * closing the vma, but it may still iterate the unmap_ops list. | |
397 | */ | |
398 | spin_lock(&priv->lock); | |
399 | map->vma = NULL; | |
400 | spin_unlock(&priv->lock); | |
401 | } | |
ab31523c | 402 | vma->vm_private_data = NULL; |
16a1d022 | 403 | gntdev_put_map(priv, map); |
ab31523c GH |
404 | } |
405 | ||
ab31523c | 406 | static struct vm_operations_struct gntdev_vmops = { |
d79647ae | 407 | .open = gntdev_vma_open, |
ab31523c | 408 | .close = gntdev_vma_close, |
ab31523c GH |
409 | }; |
410 | ||
411 | /* ------------------------------------------------------------------ */ | |
412 | ||
16a1d022 DDG |
413 | static void unmap_if_in_range(struct grant_map *map, |
414 | unsigned long start, unsigned long end) | |
415 | { | |
416 | unsigned long mstart, mend; | |
417 | int err; | |
418 | ||
419 | if (!map->vma) | |
420 | return; | |
421 | if (map->vma->vm_start >= end) | |
422 | return; | |
423 | if (map->vma->vm_end <= start) | |
424 | return; | |
425 | mstart = max(start, map->vma->vm_start); | |
426 | mend = min(end, map->vma->vm_end); | |
427 | pr_debug("map %d+%d (%lx %lx), range %lx %lx, mrange %lx %lx\n", | |
428 | map->index, map->count, | |
429 | map->vma->vm_start, map->vma->vm_end, | |
430 | start, end, mstart, mend); | |
431 | err = unmap_grant_pages(map, | |
432 | (mstart - map->vma->vm_start) >> PAGE_SHIFT, | |
433 | (mend - mstart) >> PAGE_SHIFT); | |
434 | WARN_ON(err); | |
435 | } | |
436 | ||
ab31523c GH |
437 | static void mn_invl_range_start(struct mmu_notifier *mn, |
438 | struct mm_struct *mm, | |
439 | unsigned long start, unsigned long end) | |
440 | { | |
441 | struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn); | |
442 | struct grant_map *map; | |
ab31523c GH |
443 | |
444 | spin_lock(&priv->lock); | |
445 | list_for_each_entry(map, &priv->maps, next) { | |
16a1d022 DDG |
446 | unmap_if_in_range(map, start, end); |
447 | } | |
448 | list_for_each_entry(map, &priv->freeable_maps, next) { | |
449 | unmap_if_in_range(map, start, end); | |
ab31523c GH |
450 | } |
451 | spin_unlock(&priv->lock); | |
452 | } | |
453 | ||
454 | static void mn_invl_page(struct mmu_notifier *mn, | |
455 | struct mm_struct *mm, | |
456 | unsigned long address) | |
457 | { | |
458 | mn_invl_range_start(mn, mm, address, address + PAGE_SIZE); | |
459 | } | |
460 | ||
461 | static void mn_release(struct mmu_notifier *mn, | |
462 | struct mm_struct *mm) | |
463 | { | |
464 | struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn); | |
465 | struct grant_map *map; | |
466 | int err; | |
467 | ||
468 | spin_lock(&priv->lock); | |
469 | list_for_each_entry(map, &priv->maps, next) { | |
470 | if (!map->vma) | |
471 | continue; | |
472 | pr_debug("map %d+%d (%lx %lx)\n", | |
473 | map->index, map->count, | |
474 | map->vma->vm_start, map->vma->vm_end); | |
475 | err = unmap_grant_pages(map, /* offset */ 0, map->count); | |
476 | WARN_ON(err); | |
477 | } | |
16a1d022 DDG |
478 | list_for_each_entry(map, &priv->freeable_maps, next) { |
479 | if (!map->vma) | |
480 | continue; | |
481 | pr_debug("map %d+%d (%lx %lx)\n", | |
482 | map->index, map->count, | |
483 | map->vma->vm_start, map->vma->vm_end); | |
484 | err = unmap_grant_pages(map, /* offset */ 0, map->count); | |
485 | WARN_ON(err); | |
486 | } | |
ab31523c GH |
487 | spin_unlock(&priv->lock); |
488 | } | |
489 | ||
b8b0f559 | 490 | static struct mmu_notifier_ops gntdev_mmu_ops = { |
ab31523c GH |
491 | .release = mn_release, |
492 | .invalidate_page = mn_invl_page, | |
493 | .invalidate_range_start = mn_invl_range_start, | |
494 | }; | |
495 | ||
496 | /* ------------------------------------------------------------------ */ | |
497 | ||
498 | static int gntdev_open(struct inode *inode, struct file *flip) | |
499 | { | |
500 | struct gntdev_priv *priv; | |
501 | int ret = 0; | |
502 | ||
503 | priv = kzalloc(sizeof(*priv), GFP_KERNEL); | |
504 | if (!priv) | |
505 | return -ENOMEM; | |
506 | ||
507 | INIT_LIST_HEAD(&priv->maps); | |
16a1d022 | 508 | INIT_LIST_HEAD(&priv->freeable_maps); |
ab31523c | 509 | spin_lock_init(&priv->lock); |
ab31523c | 510 | |
aab8f11a DDG |
511 | if (use_ptemod) { |
512 | priv->mm = get_task_mm(current); | |
513 | if (!priv->mm) { | |
514 | kfree(priv); | |
515 | return -ENOMEM; | |
516 | } | |
517 | priv->mn.ops = &gntdev_mmu_ops; | |
518 | ret = mmu_notifier_register(&priv->mn, priv->mm); | |
519 | mmput(priv->mm); | |
ab31523c | 520 | } |
ab31523c GH |
521 | |
522 | if (ret) { | |
523 | kfree(priv); | |
524 | return ret; | |
525 | } | |
526 | ||
527 | flip->private_data = priv; | |
528 | pr_debug("priv %p\n", priv); | |
529 | ||
530 | return 0; | |
531 | } | |
532 | ||
533 | static int gntdev_release(struct inode *inode, struct file *flip) | |
534 | { | |
535 | struct gntdev_priv *priv = flip->private_data; | |
536 | struct grant_map *map; | |
ab31523c GH |
537 | |
538 | pr_debug("priv %p\n", priv); | |
539 | ||
ab31523c GH |
540 | while (!list_empty(&priv->maps)) { |
541 | map = list_entry(priv->maps.next, struct grant_map, next); | |
68b025c8 | 542 | list_del(&map->next); |
16a1d022 | 543 | gntdev_put_map(NULL /* already removed */, map); |
ab31523c | 544 | } |
16a1d022 | 545 | WARN_ON(!list_empty(&priv->freeable_maps)); |
ab31523c | 546 | |
aab8f11a DDG |
547 | if (use_ptemod) |
548 | mmu_notifier_unregister(&priv->mn, priv->mm); | |
ab31523c GH |
549 | kfree(priv); |
550 | return 0; | |
551 | } | |
552 | ||
553 | static long gntdev_ioctl_map_grant_ref(struct gntdev_priv *priv, | |
554 | struct ioctl_gntdev_map_grant_ref __user *u) | |
555 | { | |
556 | struct ioctl_gntdev_map_grant_ref op; | |
557 | struct grant_map *map; | |
558 | int err; | |
559 | ||
560 | if (copy_from_user(&op, u, sizeof(op)) != 0) | |
561 | return -EFAULT; | |
562 | pr_debug("priv %p, add %d\n", priv, op.count); | |
563 | if (unlikely(op.count <= 0)) | |
564 | return -EINVAL; | |
ab31523c GH |
565 | |
566 | err = -ENOMEM; | |
567 | map = gntdev_alloc_map(priv, op.count); | |
568 | if (!map) | |
569 | return err; | |
ef91082e | 570 | |
68b025c8 DDG |
571 | if (unlikely(atomic_add_return(op.count, &pages_mapped) > limit)) { |
572 | pr_debug("can't map: over limit\n"); | |
16a1d022 | 573 | gntdev_put_map(NULL, map); |
ab31523c GH |
574 | return err; |
575 | } | |
576 | ||
68b025c8 DDG |
577 | if (copy_from_user(map->grants, &u->refs, |
578 | sizeof(map->grants[0]) * op.count) != 0) { | |
16a1d022 DDG |
579 | gntdev_put_map(NULL, map); |
580 | return -EFAULT; | |
ef91082e DDG |
581 | } |
582 | ||
ab31523c GH |
583 | spin_lock(&priv->lock); |
584 | gntdev_add_map(priv, map); | |
585 | op.index = map->index << PAGE_SHIFT; | |
586 | spin_unlock(&priv->lock); | |
587 | ||
68b025c8 DDG |
588 | if (copy_to_user(u, &op, sizeof(op)) != 0) |
589 | return -EFAULT; | |
590 | ||
ab31523c GH |
591 | return 0; |
592 | } | |
593 | ||
594 | static long gntdev_ioctl_unmap_grant_ref(struct gntdev_priv *priv, | |
595 | struct ioctl_gntdev_unmap_grant_ref __user *u) | |
596 | { | |
597 | struct ioctl_gntdev_unmap_grant_ref op; | |
598 | struct grant_map *map; | |
599 | int err = -ENOENT; | |
600 | ||
601 | if (copy_from_user(&op, u, sizeof(op)) != 0) | |
602 | return -EFAULT; | |
603 | pr_debug("priv %p, del %d+%d\n", priv, (int)op.index, (int)op.count); | |
604 | ||
605 | spin_lock(&priv->lock); | |
606 | map = gntdev_find_map_index(priv, op.index >> PAGE_SHIFT, op.count); | |
68b025c8 DDG |
607 | if (map) { |
608 | list_del(&map->next); | |
16a1d022 DDG |
609 | if (populate_freeable_maps) |
610 | list_add_tail(&map->next, &priv->freeable_maps); | |
68b025c8 DDG |
611 | err = 0; |
612 | } | |
ab31523c | 613 | spin_unlock(&priv->lock); |
1f1503ba | 614 | if (map) |
16a1d022 | 615 | gntdev_put_map(priv, map); |
ab31523c GH |
616 | return err; |
617 | } | |
618 | ||
619 | static long gntdev_ioctl_get_offset_for_vaddr(struct gntdev_priv *priv, | |
620 | struct ioctl_gntdev_get_offset_for_vaddr __user *u) | |
621 | { | |
622 | struct ioctl_gntdev_get_offset_for_vaddr op; | |
a879211b | 623 | struct vm_area_struct *vma; |
ab31523c | 624 | struct grant_map *map; |
2512f298 | 625 | int rv = -EINVAL; |
ab31523c GH |
626 | |
627 | if (copy_from_user(&op, u, sizeof(op)) != 0) | |
628 | return -EFAULT; | |
629 | pr_debug("priv %p, offset for vaddr %lx\n", priv, (unsigned long)op.vaddr); | |
630 | ||
2512f298 | 631 | down_read(¤t->mm->mmap_sem); |
a879211b DDG |
632 | vma = find_vma(current->mm, op.vaddr); |
633 | if (!vma || vma->vm_ops != &gntdev_vmops) | |
2512f298 | 634 | goto out_unlock; |
a879211b DDG |
635 | |
636 | map = vma->vm_private_data; | |
637 | if (!map) | |
2512f298 | 638 | goto out_unlock; |
a879211b | 639 | |
ab31523c GH |
640 | op.offset = map->index << PAGE_SHIFT; |
641 | op.count = map->count; | |
2512f298 | 642 | rv = 0; |
ab31523c | 643 | |
2512f298 DDG |
644 | out_unlock: |
645 | up_read(¤t->mm->mmap_sem); | |
646 | ||
647 | if (rv == 0 && copy_to_user(u, &op, sizeof(op)) != 0) | |
ab31523c | 648 | return -EFAULT; |
2512f298 | 649 | return rv; |
ab31523c GH |
650 | } |
651 | ||
bdc612dc DDG |
652 | static long gntdev_ioctl_notify(struct gntdev_priv *priv, void __user *u) |
653 | { | |
654 | struct ioctl_gntdev_unmap_notify op; | |
655 | struct grant_map *map; | |
656 | int rc; | |
0cc678f8 DDG |
657 | int out_flags; |
658 | unsigned int out_event; | |
bdc612dc DDG |
659 | |
660 | if (copy_from_user(&op, u, sizeof(op))) | |
661 | return -EFAULT; | |
662 | ||
663 | if (op.action & ~(UNMAP_NOTIFY_CLEAR_BYTE|UNMAP_NOTIFY_SEND_EVENT)) | |
664 | return -EINVAL; | |
665 | ||
0cc678f8 DDG |
666 | /* We need to grab a reference to the event channel we are going to use |
667 | * to send the notify before releasing the reference we may already have | |
668 | * (if someone has called this ioctl twice). This is required so that | |
669 | * it is possible to change the clear_byte part of the notification | |
670 | * without disturbing the event channel part, which may now be the last | |
671 | * reference to that event channel. | |
672 | */ | |
673 | if (op.action & UNMAP_NOTIFY_SEND_EVENT) { | |
674 | if (evtchn_get(op.event_channel_port)) | |
675 | return -EINVAL; | |
676 | } | |
677 | ||
678 | out_flags = op.action; | |
679 | out_event = op.event_channel_port; | |
680 | ||
bdc612dc DDG |
681 | spin_lock(&priv->lock); |
682 | ||
683 | list_for_each_entry(map, &priv->maps, next) { | |
684 | uint64_t begin = map->index << PAGE_SHIFT; | |
685 | uint64_t end = (map->index + map->count) << PAGE_SHIFT; | |
686 | if (op.index >= begin && op.index < end) | |
687 | goto found; | |
688 | } | |
689 | rc = -ENOENT; | |
690 | goto unlock_out; | |
691 | ||
692 | found: | |
9960be97 DDG |
693 | if ((op.action & UNMAP_NOTIFY_CLEAR_BYTE) && |
694 | (map->flags & GNTMAP_readonly)) { | |
695 | rc = -EINVAL; | |
696 | goto unlock_out; | |
697 | } | |
698 | ||
0cc678f8 DDG |
699 | out_flags = map->notify.flags; |
700 | out_event = map->notify.event; | |
701 | ||
bdc612dc DDG |
702 | map->notify.flags = op.action; |
703 | map->notify.addr = op.index - (map->index << PAGE_SHIFT); | |
704 | map->notify.event = op.event_channel_port; | |
0cc678f8 | 705 | |
bdc612dc | 706 | rc = 0; |
0cc678f8 | 707 | |
bdc612dc DDG |
708 | unlock_out: |
709 | spin_unlock(&priv->lock); | |
0cc678f8 DDG |
710 | |
711 | /* Drop the reference to the event channel we did not save in the map */ | |
712 | if (out_flags & UNMAP_NOTIFY_SEND_EVENT) | |
713 | evtchn_put(out_event); | |
714 | ||
bdc612dc DDG |
715 | return rc; |
716 | } | |
717 | ||
ab31523c GH |
718 | static long gntdev_ioctl(struct file *flip, |
719 | unsigned int cmd, unsigned long arg) | |
720 | { | |
721 | struct gntdev_priv *priv = flip->private_data; | |
722 | void __user *ptr = (void __user *)arg; | |
723 | ||
724 | switch (cmd) { | |
725 | case IOCTL_GNTDEV_MAP_GRANT_REF: | |
726 | return gntdev_ioctl_map_grant_ref(priv, ptr); | |
727 | ||
728 | case IOCTL_GNTDEV_UNMAP_GRANT_REF: | |
729 | return gntdev_ioctl_unmap_grant_ref(priv, ptr); | |
730 | ||
731 | case IOCTL_GNTDEV_GET_OFFSET_FOR_VADDR: | |
732 | return gntdev_ioctl_get_offset_for_vaddr(priv, ptr); | |
733 | ||
bdc612dc DDG |
734 | case IOCTL_GNTDEV_SET_UNMAP_NOTIFY: |
735 | return gntdev_ioctl_notify(priv, ptr); | |
736 | ||
ab31523c GH |
737 | default: |
738 | pr_debug("priv %p, unknown cmd %x\n", priv, cmd); | |
739 | return -ENOIOCTLCMD; | |
740 | } | |
741 | ||
742 | return 0; | |
743 | } | |
744 | ||
745 | static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma) | |
746 | { | |
747 | struct gntdev_priv *priv = flip->private_data; | |
748 | int index = vma->vm_pgoff; | |
749 | int count = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; | |
750 | struct grant_map *map; | |
aab8f11a | 751 | int i, err = -EINVAL; |
ab31523c GH |
752 | |
753 | if ((vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_SHARED)) | |
754 | return -EINVAL; | |
755 | ||
756 | pr_debug("map %d+%d at %lx (pgoff %lx)\n", | |
757 | index, count, vma->vm_start, vma->vm_pgoff); | |
758 | ||
759 | spin_lock(&priv->lock); | |
760 | map = gntdev_find_map_index(priv, index, count); | |
761 | if (!map) | |
762 | goto unlock_out; | |
aab8f11a | 763 | if (use_ptemod && map->vma) |
ab31523c | 764 | goto unlock_out; |
aab8f11a | 765 | if (use_ptemod && priv->mm != vma->vm_mm) { |
283c0972 | 766 | pr_warn("Huh? Other mm?\n"); |
ab31523c GH |
767 | goto unlock_out; |
768 | } | |
769 | ||
68b025c8 DDG |
770 | atomic_inc(&map->users); |
771 | ||
ab31523c GH |
772 | vma->vm_ops = &gntdev_vmops; |
773 | ||
314e51b9 | 774 | vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP; |
d79647ae DDG |
775 | |
776 | if (use_ptemod) | |
e8e937be | 777 | vma->vm_flags |= VM_DONTCOPY; |
ab31523c GH |
778 | |
779 | vma->vm_private_data = map; | |
ab31523c | 780 | |
aab8f11a DDG |
781 | if (use_ptemod) |
782 | map->vma = vma; | |
783 | ||
12996fc3 DDG |
784 | if (map->flags) { |
785 | if ((vma->vm_flags & VM_WRITE) && | |
786 | (map->flags & GNTMAP_readonly)) | |
a93e20a8 | 787 | goto out_unlock_put; |
12996fc3 DDG |
788 | } else { |
789 | map->flags = GNTMAP_host_map; | |
790 | if (!(vma->vm_flags & VM_WRITE)) | |
791 | map->flags |= GNTMAP_readonly; | |
792 | } | |
ab31523c | 793 | |
f0a70c88 DDG |
794 | spin_unlock(&priv->lock); |
795 | ||
aab8f11a DDG |
796 | if (use_ptemod) { |
797 | err = apply_to_page_range(vma->vm_mm, vma->vm_start, | |
798 | vma->vm_end - vma->vm_start, | |
799 | find_grant_ptes, map); | |
800 | if (err) { | |
283c0972 | 801 | pr_warn("find_grant_ptes() failure.\n"); |
90b6f305 | 802 | goto out_put_map; |
aab8f11a | 803 | } |
ab31523c GH |
804 | } |
805 | ||
806 | err = map_grant_pages(map); | |
90b6f305 DDG |
807 | if (err) |
808 | goto out_put_map; | |
f0a70c88 | 809 | |
aab8f11a DDG |
810 | if (!use_ptemod) { |
811 | for (i = 0; i < count; i++) { | |
812 | err = vm_insert_page(vma, vma->vm_start + i*PAGE_SIZE, | |
813 | map->pages[i]); | |
814 | if (err) | |
90b6f305 | 815 | goto out_put_map; |
aab8f11a DDG |
816 | } |
817 | } | |
818 | ||
f0a70c88 DDG |
819 | return 0; |
820 | ||
ab31523c GH |
821 | unlock_out: |
822 | spin_unlock(&priv->lock); | |
823 | return err; | |
90b6f305 | 824 | |
a93e20a8 DC |
825 | out_unlock_put: |
826 | spin_unlock(&priv->lock); | |
90b6f305 | 827 | out_put_map: |
84e4075d DDG |
828 | if (use_ptemod) |
829 | map->vma = NULL; | |
16a1d022 | 830 | gntdev_put_map(priv, map); |
90b6f305 | 831 | return err; |
ab31523c GH |
832 | } |
833 | ||
834 | static const struct file_operations gntdev_fops = { | |
835 | .owner = THIS_MODULE, | |
836 | .open = gntdev_open, | |
837 | .release = gntdev_release, | |
838 | .mmap = gntdev_mmap, | |
839 | .unlocked_ioctl = gntdev_ioctl | |
840 | }; | |
841 | ||
842 | static struct miscdevice gntdev_miscdev = { | |
843 | .minor = MISC_DYNAMIC_MINOR, | |
844 | .name = "xen/gntdev", | |
845 | .fops = &gntdev_fops, | |
846 | }; | |
847 | ||
848 | /* ------------------------------------------------------------------ */ | |
849 | ||
850 | static int __init gntdev_init(void) | |
851 | { | |
852 | int err; | |
853 | ||
854 | if (!xen_domain()) | |
855 | return -ENODEV; | |
856 | ||
6926f6d6 | 857 | use_ptemod = !xen_feature(XENFEAT_auto_translated_physmap); |
aab8f11a | 858 | |
ab31523c GH |
859 | err = misc_register(&gntdev_miscdev); |
860 | if (err != 0) { | |
283c0972 | 861 | pr_err("Could not register gntdev device\n"); |
ab31523c GH |
862 | return err; |
863 | } | |
864 | return 0; | |
865 | } | |
866 | ||
867 | static void __exit gntdev_exit(void) | |
868 | { | |
869 | misc_deregister(&gntdev_miscdev); | |
870 | } | |
871 | ||
872 | module_init(gntdev_init); | |
873 | module_exit(gntdev_exit); | |
874 | ||
875 | /* ------------------------------------------------------------------ */ |