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 */ | |
1401c00e | 70 | struct mutex lock; |
ab31523c GH |
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) | |
ff4b156f | 122 | gnttab_free_pages(map->count, map->pages); |
a67baeb7 DV |
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 | ||
ff4b156f | 155 | if (gnttab_alloc_pages(count, add->pages)) |
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 | 223 | if (populate_freeable_maps && priv) { |
1401c00e | 224 | mutex_lock(&priv->lock); |
16a1d022 | 225 | list_del(&map->next); |
1401c00e | 226 | mutex_unlock(&priv->lock); |
16a1d022 DDG |
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 | ||
74528225 JH |
312 | struct unmap_grant_pages_callback_data |
313 | { | |
314 | struct completion completion; | |
315 | int result; | |
316 | }; | |
317 | ||
318 | static void unmap_grant_callback(int result, | |
319 | struct gntab_unmap_queue_data *data) | |
320 | { | |
321 | struct unmap_grant_pages_callback_data* d = data->data; | |
322 | ||
323 | d->result = result; | |
324 | complete(&d->completion); | |
325 | } | |
326 | ||
b57c1869 | 327 | static int __unmap_grant_pages(struct grant_map *map, int offset, int pages) |
ab31523c GH |
328 | { |
329 | int i, err = 0; | |
74528225 JH |
330 | struct gntab_unmap_queue_data unmap_data; |
331 | struct unmap_grant_pages_callback_data data; | |
332 | ||
333 | init_completion(&data.completion); | |
334 | unmap_data.data = &data; | |
335 | unmap_data.done= &unmap_grant_callback; | |
ab31523c | 336 | |
bdc612dc DDG |
337 | if (map->notify.flags & UNMAP_NOTIFY_CLEAR_BYTE) { |
338 | int pgno = (map->notify.addr >> PAGE_SHIFT); | |
1affa98d DDG |
339 | if (pgno >= offset && pgno < offset + pages) { |
340 | /* No need for kmap, pages are in lowmem */ | |
341 | uint8_t *tmp = pfn_to_kaddr(page_to_pfn(map->pages[pgno])); | |
bdc612dc | 342 | tmp[map->notify.addr & (PAGE_SIZE-1)] = 0; |
bdc612dc DDG |
343 | map->notify.flags &= ~UNMAP_NOTIFY_CLEAR_BYTE; |
344 | } | |
345 | } | |
346 | ||
74528225 JH |
347 | unmap_data.unmap_ops = map->unmap_ops + offset; |
348 | unmap_data.kunmap_ops = use_ptemod ? map->kunmap_ops + offset : NULL; | |
349 | unmap_data.pages = map->pages + offset; | |
350 | unmap_data.count = pages; | |
351 | ||
352 | gnttab_unmap_refs_async(&unmap_data); | |
353 | ||
354 | wait_for_completion(&data.completion); | |
355 | if (data.result) | |
356 | return data.result; | |
ab31523c GH |
357 | |
358 | for (i = 0; i < pages; i++) { | |
359 | if (map->unmap_ops[offset+i].status) | |
360 | err = -EINVAL; | |
77c35acb DDG |
361 | pr_debug("unmap handle=%d st=%d\n", |
362 | map->unmap_ops[offset+i].handle, | |
363 | map->unmap_ops[offset+i].status); | |
364 | map->unmap_ops[offset+i].handle = -1; | |
ab31523c GH |
365 | } |
366 | return err; | |
367 | } | |
368 | ||
b57c1869 DDG |
369 | static int unmap_grant_pages(struct grant_map *map, int offset, int pages) |
370 | { | |
371 | int range, err = 0; | |
372 | ||
373 | pr_debug("unmap %d+%d [%d+%d]\n", map->index, map->count, offset, pages); | |
374 | ||
375 | /* It is possible the requested range will have a "hole" where we | |
376 | * already unmapped some of the grants. Only unmap valid ranges. | |
377 | */ | |
378 | while (pages && !err) { | |
77c35acb | 379 | while (pages && map->unmap_ops[offset].handle == -1) { |
b57c1869 DDG |
380 | offset++; |
381 | pages--; | |
382 | } | |
383 | range = 0; | |
384 | while (range < pages) { | |
77c35acb | 385 | if (map->unmap_ops[offset+range].handle == -1) { |
b57c1869 DDG |
386 | range--; |
387 | break; | |
388 | } | |
389 | range++; | |
390 | } | |
391 | err = __unmap_grant_pages(map, offset, range); | |
392 | offset += range; | |
393 | pages -= range; | |
394 | } | |
395 | ||
396 | return err; | |
397 | } | |
398 | ||
ab31523c GH |
399 | /* ------------------------------------------------------------------ */ |
400 | ||
d79647ae DDG |
401 | static void gntdev_vma_open(struct vm_area_struct *vma) |
402 | { | |
403 | struct grant_map *map = vma->vm_private_data; | |
404 | ||
405 | pr_debug("gntdev_vma_open %p\n", vma); | |
406 | atomic_inc(&map->users); | |
407 | } | |
408 | ||
ab31523c GH |
409 | static void gntdev_vma_close(struct vm_area_struct *vma) |
410 | { | |
411 | struct grant_map *map = vma->vm_private_data; | |
16a1d022 DDG |
412 | struct file *file = vma->vm_file; |
413 | struct gntdev_priv *priv = file->private_data; | |
ab31523c | 414 | |
d79647ae | 415 | pr_debug("gntdev_vma_close %p\n", vma); |
2512f298 | 416 | if (use_ptemod) { |
2512f298 DDG |
417 | /* It is possible that an mmu notifier could be running |
418 | * concurrently, so take priv->lock to ensure that the vma won't | |
419 | * vanishing during the unmap_grant_pages call, since we will | |
420 | * spin here until that completes. Such a concurrent call will | |
421 | * not do any unmapping, since that has been done prior to | |
422 | * closing the vma, but it may still iterate the unmap_ops list. | |
423 | */ | |
1401c00e | 424 | mutex_lock(&priv->lock); |
2512f298 | 425 | map->vma = NULL; |
1401c00e | 426 | mutex_unlock(&priv->lock); |
2512f298 | 427 | } |
ab31523c | 428 | vma->vm_private_data = NULL; |
16a1d022 | 429 | gntdev_put_map(priv, map); |
ab31523c GH |
430 | } |
431 | ||
ab31523c | 432 | static struct vm_operations_struct gntdev_vmops = { |
d79647ae | 433 | .open = gntdev_vma_open, |
ab31523c | 434 | .close = gntdev_vma_close, |
ab31523c GH |
435 | }; |
436 | ||
437 | /* ------------------------------------------------------------------ */ | |
438 | ||
16a1d022 DDG |
439 | static void unmap_if_in_range(struct grant_map *map, |
440 | unsigned long start, unsigned long end) | |
441 | { | |
442 | unsigned long mstart, mend; | |
443 | int err; | |
444 | ||
445 | if (!map->vma) | |
446 | return; | |
447 | if (map->vma->vm_start >= end) | |
448 | return; | |
449 | if (map->vma->vm_end <= start) | |
450 | return; | |
451 | mstart = max(start, map->vma->vm_start); | |
452 | mend = min(end, map->vma->vm_end); | |
453 | pr_debug("map %d+%d (%lx %lx), range %lx %lx, mrange %lx %lx\n", | |
454 | map->index, map->count, | |
455 | map->vma->vm_start, map->vma->vm_end, | |
456 | start, end, mstart, mend); | |
457 | err = unmap_grant_pages(map, | |
458 | (mstart - map->vma->vm_start) >> PAGE_SHIFT, | |
459 | (mend - mstart) >> PAGE_SHIFT); | |
460 | WARN_ON(err); | |
461 | } | |
462 | ||
ab31523c GH |
463 | static void mn_invl_range_start(struct mmu_notifier *mn, |
464 | struct mm_struct *mm, | |
465 | unsigned long start, unsigned long end) | |
466 | { | |
467 | struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn); | |
468 | struct grant_map *map; | |
ab31523c | 469 | |
1401c00e | 470 | mutex_lock(&priv->lock); |
ab31523c | 471 | list_for_each_entry(map, &priv->maps, next) { |
16a1d022 DDG |
472 | unmap_if_in_range(map, start, end); |
473 | } | |
474 | list_for_each_entry(map, &priv->freeable_maps, next) { | |
475 | unmap_if_in_range(map, start, end); | |
ab31523c | 476 | } |
1401c00e | 477 | mutex_unlock(&priv->lock); |
ab31523c GH |
478 | } |
479 | ||
480 | static void mn_invl_page(struct mmu_notifier *mn, | |
481 | struct mm_struct *mm, | |
482 | unsigned long address) | |
483 | { | |
484 | mn_invl_range_start(mn, mm, address, address + PAGE_SIZE); | |
485 | } | |
486 | ||
487 | static void mn_release(struct mmu_notifier *mn, | |
488 | struct mm_struct *mm) | |
489 | { | |
490 | struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn); | |
491 | struct grant_map *map; | |
492 | int err; | |
493 | ||
1401c00e | 494 | mutex_lock(&priv->lock); |
ab31523c GH |
495 | list_for_each_entry(map, &priv->maps, next) { |
496 | if (!map->vma) | |
497 | continue; | |
498 | pr_debug("map %d+%d (%lx %lx)\n", | |
499 | map->index, map->count, | |
500 | map->vma->vm_start, map->vma->vm_end); | |
501 | err = unmap_grant_pages(map, /* offset */ 0, map->count); | |
502 | WARN_ON(err); | |
503 | } | |
16a1d022 DDG |
504 | list_for_each_entry(map, &priv->freeable_maps, next) { |
505 | if (!map->vma) | |
506 | continue; | |
507 | pr_debug("map %d+%d (%lx %lx)\n", | |
508 | map->index, map->count, | |
509 | map->vma->vm_start, map->vma->vm_end); | |
510 | err = unmap_grant_pages(map, /* offset */ 0, map->count); | |
511 | WARN_ON(err); | |
512 | } | |
1401c00e | 513 | mutex_unlock(&priv->lock); |
ab31523c GH |
514 | } |
515 | ||
b8b0f559 | 516 | static struct mmu_notifier_ops gntdev_mmu_ops = { |
ab31523c GH |
517 | .release = mn_release, |
518 | .invalidate_page = mn_invl_page, | |
519 | .invalidate_range_start = mn_invl_range_start, | |
520 | }; | |
521 | ||
522 | /* ------------------------------------------------------------------ */ | |
523 | ||
524 | static int gntdev_open(struct inode *inode, struct file *flip) | |
525 | { | |
526 | struct gntdev_priv *priv; | |
527 | int ret = 0; | |
528 | ||
529 | priv = kzalloc(sizeof(*priv), GFP_KERNEL); | |
530 | if (!priv) | |
531 | return -ENOMEM; | |
532 | ||
533 | INIT_LIST_HEAD(&priv->maps); | |
16a1d022 | 534 | INIT_LIST_HEAD(&priv->freeable_maps); |
1401c00e | 535 | mutex_init(&priv->lock); |
ab31523c | 536 | |
aab8f11a DDG |
537 | if (use_ptemod) { |
538 | priv->mm = get_task_mm(current); | |
539 | if (!priv->mm) { | |
540 | kfree(priv); | |
541 | return -ENOMEM; | |
542 | } | |
543 | priv->mn.ops = &gntdev_mmu_ops; | |
544 | ret = mmu_notifier_register(&priv->mn, priv->mm); | |
545 | mmput(priv->mm); | |
ab31523c | 546 | } |
ab31523c GH |
547 | |
548 | if (ret) { | |
549 | kfree(priv); | |
550 | return ret; | |
551 | } | |
552 | ||
553 | flip->private_data = priv; | |
554 | pr_debug("priv %p\n", priv); | |
555 | ||
556 | return 0; | |
557 | } | |
558 | ||
559 | static int gntdev_release(struct inode *inode, struct file *flip) | |
560 | { | |
561 | struct gntdev_priv *priv = flip->private_data; | |
562 | struct grant_map *map; | |
ab31523c GH |
563 | |
564 | pr_debug("priv %p\n", priv); | |
565 | ||
ab31523c GH |
566 | while (!list_empty(&priv->maps)) { |
567 | map = list_entry(priv->maps.next, struct grant_map, next); | |
68b025c8 | 568 | list_del(&map->next); |
16a1d022 | 569 | gntdev_put_map(NULL /* already removed */, map); |
ab31523c | 570 | } |
16a1d022 | 571 | WARN_ON(!list_empty(&priv->freeable_maps)); |
ab31523c | 572 | |
aab8f11a DDG |
573 | if (use_ptemod) |
574 | mmu_notifier_unregister(&priv->mn, priv->mm); | |
ab31523c GH |
575 | kfree(priv); |
576 | return 0; | |
577 | } | |
578 | ||
579 | static long gntdev_ioctl_map_grant_ref(struct gntdev_priv *priv, | |
580 | struct ioctl_gntdev_map_grant_ref __user *u) | |
581 | { | |
582 | struct ioctl_gntdev_map_grant_ref op; | |
583 | struct grant_map *map; | |
584 | int err; | |
585 | ||
586 | if (copy_from_user(&op, u, sizeof(op)) != 0) | |
587 | return -EFAULT; | |
588 | pr_debug("priv %p, add %d\n", priv, op.count); | |
589 | if (unlikely(op.count <= 0)) | |
590 | return -EINVAL; | |
ab31523c GH |
591 | |
592 | err = -ENOMEM; | |
593 | map = gntdev_alloc_map(priv, op.count); | |
594 | if (!map) | |
595 | return err; | |
ef91082e | 596 | |
68b025c8 DDG |
597 | if (unlikely(atomic_add_return(op.count, &pages_mapped) > limit)) { |
598 | pr_debug("can't map: over limit\n"); | |
16a1d022 | 599 | gntdev_put_map(NULL, map); |
ab31523c GH |
600 | return err; |
601 | } | |
602 | ||
68b025c8 DDG |
603 | if (copy_from_user(map->grants, &u->refs, |
604 | sizeof(map->grants[0]) * op.count) != 0) { | |
16a1d022 DDG |
605 | gntdev_put_map(NULL, map); |
606 | return -EFAULT; | |
ef91082e DDG |
607 | } |
608 | ||
1401c00e | 609 | mutex_lock(&priv->lock); |
ab31523c GH |
610 | gntdev_add_map(priv, map); |
611 | op.index = map->index << PAGE_SHIFT; | |
1401c00e | 612 | mutex_unlock(&priv->lock); |
ab31523c | 613 | |
68b025c8 DDG |
614 | if (copy_to_user(u, &op, sizeof(op)) != 0) |
615 | return -EFAULT; | |
616 | ||
ab31523c GH |
617 | return 0; |
618 | } | |
619 | ||
620 | static long gntdev_ioctl_unmap_grant_ref(struct gntdev_priv *priv, | |
621 | struct ioctl_gntdev_unmap_grant_ref __user *u) | |
622 | { | |
623 | struct ioctl_gntdev_unmap_grant_ref op; | |
624 | struct grant_map *map; | |
625 | int err = -ENOENT; | |
626 | ||
627 | if (copy_from_user(&op, u, sizeof(op)) != 0) | |
628 | return -EFAULT; | |
629 | pr_debug("priv %p, del %d+%d\n", priv, (int)op.index, (int)op.count); | |
630 | ||
1401c00e | 631 | mutex_lock(&priv->lock); |
ab31523c | 632 | map = gntdev_find_map_index(priv, op.index >> PAGE_SHIFT, op.count); |
68b025c8 DDG |
633 | if (map) { |
634 | list_del(&map->next); | |
16a1d022 DDG |
635 | if (populate_freeable_maps) |
636 | list_add_tail(&map->next, &priv->freeable_maps); | |
68b025c8 DDG |
637 | err = 0; |
638 | } | |
1401c00e | 639 | mutex_unlock(&priv->lock); |
1f1503ba | 640 | if (map) |
16a1d022 | 641 | gntdev_put_map(priv, map); |
ab31523c GH |
642 | return err; |
643 | } | |
644 | ||
645 | static long gntdev_ioctl_get_offset_for_vaddr(struct gntdev_priv *priv, | |
646 | struct ioctl_gntdev_get_offset_for_vaddr __user *u) | |
647 | { | |
648 | struct ioctl_gntdev_get_offset_for_vaddr op; | |
a879211b | 649 | struct vm_area_struct *vma; |
ab31523c | 650 | struct grant_map *map; |
2512f298 | 651 | int rv = -EINVAL; |
ab31523c GH |
652 | |
653 | if (copy_from_user(&op, u, sizeof(op)) != 0) | |
654 | return -EFAULT; | |
655 | pr_debug("priv %p, offset for vaddr %lx\n", priv, (unsigned long)op.vaddr); | |
656 | ||
2512f298 | 657 | down_read(¤t->mm->mmap_sem); |
a879211b DDG |
658 | vma = find_vma(current->mm, op.vaddr); |
659 | if (!vma || vma->vm_ops != &gntdev_vmops) | |
2512f298 | 660 | goto out_unlock; |
a879211b DDG |
661 | |
662 | map = vma->vm_private_data; | |
663 | if (!map) | |
2512f298 | 664 | goto out_unlock; |
a879211b | 665 | |
ab31523c GH |
666 | op.offset = map->index << PAGE_SHIFT; |
667 | op.count = map->count; | |
2512f298 | 668 | rv = 0; |
ab31523c | 669 | |
2512f298 DDG |
670 | out_unlock: |
671 | up_read(¤t->mm->mmap_sem); | |
672 | ||
673 | if (rv == 0 && copy_to_user(u, &op, sizeof(op)) != 0) | |
ab31523c | 674 | return -EFAULT; |
2512f298 | 675 | return rv; |
ab31523c GH |
676 | } |
677 | ||
bdc612dc DDG |
678 | static long gntdev_ioctl_notify(struct gntdev_priv *priv, void __user *u) |
679 | { | |
680 | struct ioctl_gntdev_unmap_notify op; | |
681 | struct grant_map *map; | |
682 | int rc; | |
0cc678f8 DDG |
683 | int out_flags; |
684 | unsigned int out_event; | |
bdc612dc DDG |
685 | |
686 | if (copy_from_user(&op, u, sizeof(op))) | |
687 | return -EFAULT; | |
688 | ||
689 | if (op.action & ~(UNMAP_NOTIFY_CLEAR_BYTE|UNMAP_NOTIFY_SEND_EVENT)) | |
690 | return -EINVAL; | |
691 | ||
0cc678f8 DDG |
692 | /* We need to grab a reference to the event channel we are going to use |
693 | * to send the notify before releasing the reference we may already have | |
694 | * (if someone has called this ioctl twice). This is required so that | |
695 | * it is possible to change the clear_byte part of the notification | |
696 | * without disturbing the event channel part, which may now be the last | |
697 | * reference to that event channel. | |
698 | */ | |
699 | if (op.action & UNMAP_NOTIFY_SEND_EVENT) { | |
700 | if (evtchn_get(op.event_channel_port)) | |
701 | return -EINVAL; | |
702 | } | |
703 | ||
704 | out_flags = op.action; | |
705 | out_event = op.event_channel_port; | |
706 | ||
1401c00e | 707 | mutex_lock(&priv->lock); |
bdc612dc DDG |
708 | |
709 | list_for_each_entry(map, &priv->maps, next) { | |
710 | uint64_t begin = map->index << PAGE_SHIFT; | |
711 | uint64_t end = (map->index + map->count) << PAGE_SHIFT; | |
712 | if (op.index >= begin && op.index < end) | |
713 | goto found; | |
714 | } | |
715 | rc = -ENOENT; | |
716 | goto unlock_out; | |
717 | ||
718 | found: | |
9960be97 DDG |
719 | if ((op.action & UNMAP_NOTIFY_CLEAR_BYTE) && |
720 | (map->flags & GNTMAP_readonly)) { | |
721 | rc = -EINVAL; | |
722 | goto unlock_out; | |
723 | } | |
724 | ||
0cc678f8 DDG |
725 | out_flags = map->notify.flags; |
726 | out_event = map->notify.event; | |
727 | ||
bdc612dc DDG |
728 | map->notify.flags = op.action; |
729 | map->notify.addr = op.index - (map->index << PAGE_SHIFT); | |
730 | map->notify.event = op.event_channel_port; | |
0cc678f8 | 731 | |
bdc612dc | 732 | rc = 0; |
0cc678f8 | 733 | |
bdc612dc | 734 | unlock_out: |
1401c00e | 735 | mutex_unlock(&priv->lock); |
0cc678f8 DDG |
736 | |
737 | /* Drop the reference to the event channel we did not save in the map */ | |
738 | if (out_flags & UNMAP_NOTIFY_SEND_EVENT) | |
739 | evtchn_put(out_event); | |
740 | ||
bdc612dc DDG |
741 | return rc; |
742 | } | |
743 | ||
ab31523c GH |
744 | static long gntdev_ioctl(struct file *flip, |
745 | unsigned int cmd, unsigned long arg) | |
746 | { | |
747 | struct gntdev_priv *priv = flip->private_data; | |
748 | void __user *ptr = (void __user *)arg; | |
749 | ||
750 | switch (cmd) { | |
751 | case IOCTL_GNTDEV_MAP_GRANT_REF: | |
752 | return gntdev_ioctl_map_grant_ref(priv, ptr); | |
753 | ||
754 | case IOCTL_GNTDEV_UNMAP_GRANT_REF: | |
755 | return gntdev_ioctl_unmap_grant_ref(priv, ptr); | |
756 | ||
757 | case IOCTL_GNTDEV_GET_OFFSET_FOR_VADDR: | |
758 | return gntdev_ioctl_get_offset_for_vaddr(priv, ptr); | |
759 | ||
bdc612dc DDG |
760 | case IOCTL_GNTDEV_SET_UNMAP_NOTIFY: |
761 | return gntdev_ioctl_notify(priv, ptr); | |
762 | ||
ab31523c GH |
763 | default: |
764 | pr_debug("priv %p, unknown cmd %x\n", priv, cmd); | |
765 | return -ENOIOCTLCMD; | |
766 | } | |
767 | ||
768 | return 0; | |
769 | } | |
770 | ||
771 | static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma) | |
772 | { | |
773 | struct gntdev_priv *priv = flip->private_data; | |
774 | int index = vma->vm_pgoff; | |
775 | int count = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; | |
776 | struct grant_map *map; | |
aab8f11a | 777 | int i, err = -EINVAL; |
ab31523c GH |
778 | |
779 | if ((vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_SHARED)) | |
780 | return -EINVAL; | |
781 | ||
782 | pr_debug("map %d+%d at %lx (pgoff %lx)\n", | |
783 | index, count, vma->vm_start, vma->vm_pgoff); | |
784 | ||
1401c00e | 785 | mutex_lock(&priv->lock); |
ab31523c GH |
786 | map = gntdev_find_map_index(priv, index, count); |
787 | if (!map) | |
788 | goto unlock_out; | |
aab8f11a | 789 | if (use_ptemod && map->vma) |
ab31523c | 790 | goto unlock_out; |
aab8f11a | 791 | if (use_ptemod && priv->mm != vma->vm_mm) { |
283c0972 | 792 | pr_warn("Huh? Other mm?\n"); |
ab31523c GH |
793 | goto unlock_out; |
794 | } | |
795 | ||
68b025c8 DDG |
796 | atomic_inc(&map->users); |
797 | ||
ab31523c GH |
798 | vma->vm_ops = &gntdev_vmops; |
799 | ||
314e51b9 | 800 | vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP; |
d79647ae DDG |
801 | |
802 | if (use_ptemod) | |
e8e937be | 803 | vma->vm_flags |= VM_DONTCOPY; |
ab31523c GH |
804 | |
805 | vma->vm_private_data = map; | |
ab31523c | 806 | |
aab8f11a DDG |
807 | if (use_ptemod) |
808 | map->vma = vma; | |
809 | ||
12996fc3 DDG |
810 | if (map->flags) { |
811 | if ((vma->vm_flags & VM_WRITE) && | |
812 | (map->flags & GNTMAP_readonly)) | |
a93e20a8 | 813 | goto out_unlock_put; |
12996fc3 DDG |
814 | } else { |
815 | map->flags = GNTMAP_host_map; | |
816 | if (!(vma->vm_flags & VM_WRITE)) | |
817 | map->flags |= GNTMAP_readonly; | |
818 | } | |
ab31523c | 819 | |
1401c00e | 820 | mutex_unlock(&priv->lock); |
f0a70c88 | 821 | |
aab8f11a DDG |
822 | if (use_ptemod) { |
823 | err = apply_to_page_range(vma->vm_mm, vma->vm_start, | |
824 | vma->vm_end - vma->vm_start, | |
825 | find_grant_ptes, map); | |
826 | if (err) { | |
283c0972 | 827 | pr_warn("find_grant_ptes() failure.\n"); |
90b6f305 | 828 | goto out_put_map; |
aab8f11a | 829 | } |
ab31523c GH |
830 | } |
831 | ||
832 | err = map_grant_pages(map); | |
90b6f305 DDG |
833 | if (err) |
834 | goto out_put_map; | |
f0a70c88 | 835 | |
aab8f11a DDG |
836 | if (!use_ptemod) { |
837 | for (i = 0; i < count; i++) { | |
838 | err = vm_insert_page(vma, vma->vm_start + i*PAGE_SIZE, | |
839 | map->pages[i]); | |
840 | if (err) | |
90b6f305 | 841 | goto out_put_map; |
aab8f11a DDG |
842 | } |
843 | } | |
844 | ||
f0a70c88 DDG |
845 | return 0; |
846 | ||
ab31523c | 847 | unlock_out: |
1401c00e | 848 | mutex_unlock(&priv->lock); |
ab31523c | 849 | return err; |
90b6f305 | 850 | |
a93e20a8 | 851 | out_unlock_put: |
1401c00e | 852 | mutex_unlock(&priv->lock); |
90b6f305 | 853 | out_put_map: |
84e4075d DDG |
854 | if (use_ptemod) |
855 | map->vma = NULL; | |
16a1d022 | 856 | gntdev_put_map(priv, map); |
90b6f305 | 857 | return err; |
ab31523c GH |
858 | } |
859 | ||
860 | static const struct file_operations gntdev_fops = { | |
861 | .owner = THIS_MODULE, | |
862 | .open = gntdev_open, | |
863 | .release = gntdev_release, | |
864 | .mmap = gntdev_mmap, | |
865 | .unlocked_ioctl = gntdev_ioctl | |
866 | }; | |
867 | ||
868 | static struct miscdevice gntdev_miscdev = { | |
869 | .minor = MISC_DYNAMIC_MINOR, | |
870 | .name = "xen/gntdev", | |
871 | .fops = &gntdev_fops, | |
872 | }; | |
873 | ||
874 | /* ------------------------------------------------------------------ */ | |
875 | ||
876 | static int __init gntdev_init(void) | |
877 | { | |
878 | int err; | |
879 | ||
880 | if (!xen_domain()) | |
881 | return -ENODEV; | |
882 | ||
6926f6d6 | 883 | use_ptemod = !xen_feature(XENFEAT_auto_translated_physmap); |
aab8f11a | 884 | |
ab31523c GH |
885 | err = misc_register(&gntdev_miscdev); |
886 | if (err != 0) { | |
283c0972 | 887 | pr_err("Could not register gntdev device\n"); |
ab31523c GH |
888 | return err; |
889 | } | |
890 | return 0; | |
891 | } | |
892 | ||
893 | static void __exit gntdev_exit(void) | |
894 | { | |
895 | misc_deregister(&gntdev_miscdev); | |
896 | } | |
897 | ||
898 | module_init(gntdev_init); | |
899 | module_exit(gntdev_exit); | |
900 | ||
901 | /* ------------------------------------------------------------------ */ |