]>
Commit | Line | Data |
---|---|---|
3a4d5c94 MT |
1 | /* Copyright (C) 2009 Red Hat, Inc. |
2 | * Copyright (C) 2006 Rusty Russell IBM Corporation | |
3 | * | |
4 | * Author: Michael S. Tsirkin <mst@redhat.com> | |
5 | * | |
6 | * Inspiration, some code, and most witty comments come from | |
61516587 | 7 | * Documentation/virtual/lguest/lguest.c, by Rusty Russell |
3a4d5c94 MT |
8 | * |
9 | * This work is licensed under the terms of the GNU GPL, version 2. | |
10 | * | |
11 | * Generic code for virtio server in host kernel. | |
12 | */ | |
13 | ||
14 | #include <linux/eventfd.h> | |
15 | #include <linux/vhost.h> | |
35596b27 | 16 | #include <linux/uio.h> |
3a4d5c94 | 17 | #include <linux/mm.h> |
64e1c807 | 18 | #include <linux/mmu_context.h> |
3a4d5c94 MT |
19 | #include <linux/miscdevice.h> |
20 | #include <linux/mutex.h> | |
3a4d5c94 MT |
21 | #include <linux/poll.h> |
22 | #include <linux/file.h> | |
23 | #include <linux/highmem.h> | |
5a0e3ad6 | 24 | #include <linux/slab.h> |
4de7255f | 25 | #include <linux/vmalloc.h> |
c23f3445 | 26 | #include <linux/kthread.h> |
9e3d1957 | 27 | #include <linux/cgroup.h> |
6ac1afbf | 28 | #include <linux/module.h> |
bcfeacab | 29 | #include <linux/sort.h> |
a9709d68 | 30 | #include <linux/interval_tree_generic.h> |
3a4d5c94 | 31 | |
3a4d5c94 MT |
32 | #include "vhost.h" |
33 | ||
c9ce42f7 IM |
34 | static ushort max_mem_regions = 64; |
35 | module_param(max_mem_regions, ushort, 0444); | |
36 | MODULE_PARM_DESC(max_mem_regions, | |
37 | "Maximum number of memory regions in memory map. (default: 64)"); | |
6b1e6cc7 JW |
38 | static int max_iotlb_entries = 2048; |
39 | module_param(max_iotlb_entries, int, 0444); | |
40 | MODULE_PARM_DESC(max_iotlb_entries, | |
41 | "Maximum number of iotlb entries. (default: 2048)"); | |
c9ce42f7 | 42 | |
3a4d5c94 | 43 | enum { |
3a4d5c94 MT |
44 | VHOST_MEMORY_F_LOG = 0x1, |
45 | }; | |
46 | ||
3b1bbe89 MT |
47 | #define vhost_used_event(vq) ((__virtio16 __user *)&vq->avail->ring[vq->num]) |
48 | #define vhost_avail_event(vq) ((__virtio16 __user *)&vq->used->ring[vq->num]) | |
8ea8cf89 | 49 | |
a9709d68 JW |
50 | INTERVAL_TREE_DEFINE(struct vhost_umem_node, |
51 | rb, __u64, __subtree_last, | |
2f952c01 | 52 | START, LAST, static inline, vhost_umem_interval_tree); |
a9709d68 | 53 | |
2751c988 | 54 | #ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY |
c5072037 | 55 | static void vhost_disable_cross_endian(struct vhost_virtqueue *vq) |
2751c988 GK |
56 | { |
57 | vq->user_be = !virtio_legacy_is_little_endian(); | |
58 | } | |
59 | ||
c5072037 GK |
60 | static void vhost_enable_cross_endian_big(struct vhost_virtqueue *vq) |
61 | { | |
62 | vq->user_be = true; | |
63 | } | |
64 | ||
65 | static void vhost_enable_cross_endian_little(struct vhost_virtqueue *vq) | |
66 | { | |
67 | vq->user_be = false; | |
68 | } | |
69 | ||
2751c988 GK |
70 | static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp) |
71 | { | |
72 | struct vhost_vring_state s; | |
73 | ||
74 | if (vq->private_data) | |
75 | return -EBUSY; | |
76 | ||
77 | if (copy_from_user(&s, argp, sizeof(s))) | |
78 | return -EFAULT; | |
79 | ||
80 | if (s.num != VHOST_VRING_LITTLE_ENDIAN && | |
81 | s.num != VHOST_VRING_BIG_ENDIAN) | |
82 | return -EINVAL; | |
83 | ||
c5072037 GK |
84 | if (s.num == VHOST_VRING_BIG_ENDIAN) |
85 | vhost_enable_cross_endian_big(vq); | |
86 | else | |
87 | vhost_enable_cross_endian_little(vq); | |
2751c988 GK |
88 | |
89 | return 0; | |
90 | } | |
91 | ||
92 | static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx, | |
93 | int __user *argp) | |
94 | { | |
95 | struct vhost_vring_state s = { | |
96 | .index = idx, | |
97 | .num = vq->user_be | |
98 | }; | |
99 | ||
100 | if (copy_to_user(argp, &s, sizeof(s))) | |
101 | return -EFAULT; | |
102 | ||
103 | return 0; | |
104 | } | |
105 | ||
106 | static void vhost_init_is_le(struct vhost_virtqueue *vq) | |
107 | { | |
108 | /* Note for legacy virtio: user_be is initialized at reset time | |
109 | * according to the host endianness. If userspace does not set an | |
110 | * explicit endianness, the default behavior is native endian, as | |
111 | * expected by legacy virtio. | |
112 | */ | |
113 | vq->is_le = vhost_has_feature(vq, VIRTIO_F_VERSION_1) || !vq->user_be; | |
114 | } | |
115 | #else | |
c5072037 | 116 | static void vhost_disable_cross_endian(struct vhost_virtqueue *vq) |
2751c988 GK |
117 | { |
118 | } | |
119 | ||
120 | static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp) | |
121 | { | |
122 | return -ENOIOCTLCMD; | |
123 | } | |
124 | ||
125 | static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx, | |
126 | int __user *argp) | |
127 | { | |
128 | return -ENOIOCTLCMD; | |
129 | } | |
130 | ||
131 | static void vhost_init_is_le(struct vhost_virtqueue *vq) | |
132 | { | |
cda8bba0 HP |
133 | vq->is_le = vhost_has_feature(vq, VIRTIO_F_VERSION_1) |
134 | || virtio_legacy_is_little_endian(); | |
2751c988 GK |
135 | } |
136 | #endif /* CONFIG_VHOST_CROSS_ENDIAN_LEGACY */ | |
137 | ||
c5072037 GK |
138 | static void vhost_reset_is_le(struct vhost_virtqueue *vq) |
139 | { | |
cda8bba0 | 140 | vhost_init_is_le(vq); |
c5072037 GK |
141 | } |
142 | ||
7235acdb JW |
143 | struct vhost_flush_struct { |
144 | struct vhost_work work; | |
145 | struct completion wait_event; | |
146 | }; | |
147 | ||
148 | static void vhost_flush_work(struct vhost_work *work) | |
149 | { | |
150 | struct vhost_flush_struct *s; | |
151 | ||
152 | s = container_of(work, struct vhost_flush_struct, work); | |
153 | complete(&s->wait_event); | |
154 | } | |
155 | ||
3a4d5c94 MT |
156 | static void vhost_poll_func(struct file *file, wait_queue_head_t *wqh, |
157 | poll_table *pt) | |
158 | { | |
159 | struct vhost_poll *poll; | |
3a4d5c94 | 160 | |
d47effe1 | 161 | poll = container_of(pt, struct vhost_poll, table); |
3a4d5c94 MT |
162 | poll->wqh = wqh; |
163 | add_wait_queue(wqh, &poll->wait); | |
164 | } | |
165 | ||
166 | static int vhost_poll_wakeup(wait_queue_t *wait, unsigned mode, int sync, | |
167 | void *key) | |
168 | { | |
c23f3445 TH |
169 | struct vhost_poll *poll = container_of(wait, struct vhost_poll, wait); |
170 | ||
3a4d5c94 MT |
171 | if (!((unsigned long)key & poll->mask)) |
172 | return 0; | |
173 | ||
c23f3445 | 174 | vhost_poll_queue(poll); |
3a4d5c94 MT |
175 | return 0; |
176 | } | |
177 | ||
163049ae | 178 | void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn) |
87d6a412 | 179 | { |
04b96e55 | 180 | clear_bit(VHOST_WORK_QUEUED, &work->flags); |
87d6a412 MT |
181 | work->fn = fn; |
182 | init_waitqueue_head(&work->done); | |
87d6a412 | 183 | } |
6ac1afbf | 184 | EXPORT_SYMBOL_GPL(vhost_work_init); |
87d6a412 | 185 | |
3a4d5c94 | 186 | /* Init poll structure */ |
c23f3445 TH |
187 | void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn, |
188 | unsigned long mask, struct vhost_dev *dev) | |
3a4d5c94 | 189 | { |
3a4d5c94 MT |
190 | init_waitqueue_func_entry(&poll->wait, vhost_poll_wakeup); |
191 | init_poll_funcptr(&poll->table, vhost_poll_func); | |
192 | poll->mask = mask; | |
c23f3445 | 193 | poll->dev = dev; |
2b8b328b | 194 | poll->wqh = NULL; |
c23f3445 | 195 | |
87d6a412 | 196 | vhost_work_init(&poll->work, fn); |
3a4d5c94 | 197 | } |
6ac1afbf | 198 | EXPORT_SYMBOL_GPL(vhost_poll_init); |
3a4d5c94 MT |
199 | |
200 | /* Start polling a file. We add ourselves to file's wait queue. The caller must | |
201 | * keep a reference to a file until after vhost_poll_stop is called. */ | |
2b8b328b | 202 | int vhost_poll_start(struct vhost_poll *poll, struct file *file) |
3a4d5c94 MT |
203 | { |
204 | unsigned long mask; | |
2b8b328b | 205 | int ret = 0; |
d47effe1 | 206 | |
70181d51 JW |
207 | if (poll->wqh) |
208 | return 0; | |
209 | ||
3a4d5c94 MT |
210 | mask = file->f_op->poll(file, &poll->table); |
211 | if (mask) | |
212 | vhost_poll_wakeup(&poll->wait, 0, 0, (void *)mask); | |
2b8b328b JW |
213 | if (mask & POLLERR) { |
214 | if (poll->wqh) | |
215 | remove_wait_queue(poll->wqh, &poll->wait); | |
216 | ret = -EINVAL; | |
217 | } | |
218 | ||
219 | return ret; | |
3a4d5c94 | 220 | } |
6ac1afbf | 221 | EXPORT_SYMBOL_GPL(vhost_poll_start); |
3a4d5c94 MT |
222 | |
223 | /* Stop polling a file. After this function returns, it becomes safe to drop the | |
224 | * file reference. You must also flush afterwards. */ | |
225 | void vhost_poll_stop(struct vhost_poll *poll) | |
226 | { | |
2b8b328b JW |
227 | if (poll->wqh) { |
228 | remove_wait_queue(poll->wqh, &poll->wait); | |
229 | poll->wqh = NULL; | |
230 | } | |
3a4d5c94 | 231 | } |
6ac1afbf | 232 | EXPORT_SYMBOL_GPL(vhost_poll_stop); |
3a4d5c94 | 233 | |
7235acdb | 234 | void vhost_work_flush(struct vhost_dev *dev, struct vhost_work *work) |
0174b0c3 | 235 | { |
7235acdb | 236 | struct vhost_flush_struct flush; |
d47effe1 | 237 | |
7235acdb JW |
238 | if (dev->worker) { |
239 | init_completion(&flush.wait_event); | |
240 | vhost_work_init(&flush.work, vhost_flush_work); | |
0174b0c3 | 241 | |
7235acdb JW |
242 | vhost_work_queue(dev, &flush.work); |
243 | wait_for_completion(&flush.wait_event); | |
244 | } | |
3a4d5c94 | 245 | } |
6ac1afbf | 246 | EXPORT_SYMBOL_GPL(vhost_work_flush); |
3a4d5c94 | 247 | |
87d6a412 MT |
248 | /* Flush any work that has been scheduled. When calling this, don't hold any |
249 | * locks that are also used by the callback. */ | |
250 | void vhost_poll_flush(struct vhost_poll *poll) | |
251 | { | |
252 | vhost_work_flush(poll->dev, &poll->work); | |
253 | } | |
6ac1afbf | 254 | EXPORT_SYMBOL_GPL(vhost_poll_flush); |
87d6a412 | 255 | |
163049ae | 256 | void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work) |
3a4d5c94 | 257 | { |
04b96e55 JW |
258 | if (!dev->worker) |
259 | return; | |
c23f3445 | 260 | |
04b96e55 JW |
261 | if (!test_and_set_bit(VHOST_WORK_QUEUED, &work->flags)) { |
262 | /* We can only add the work to the list after we're | |
263 | * sure it was not in the list. | |
635abf01 | 264 | * test_and_set_bit() implies a memory barrier. |
04b96e55 | 265 | */ |
04b96e55 | 266 | llist_add(&work->node, &dev->work_list); |
c23f3445 TH |
267 | wake_up_process(dev->worker); |
268 | } | |
3a4d5c94 | 269 | } |
6ac1afbf | 270 | EXPORT_SYMBOL_GPL(vhost_work_queue); |
3a4d5c94 | 271 | |
526d3e7f JW |
272 | /* A lockless hint for busy polling code to exit the loop */ |
273 | bool vhost_has_work(struct vhost_dev *dev) | |
274 | { | |
04b96e55 | 275 | return !llist_empty(&dev->work_list); |
526d3e7f JW |
276 | } |
277 | EXPORT_SYMBOL_GPL(vhost_has_work); | |
278 | ||
87d6a412 MT |
279 | void vhost_poll_queue(struct vhost_poll *poll) |
280 | { | |
281 | vhost_work_queue(poll->dev, &poll->work); | |
282 | } | |
6ac1afbf | 283 | EXPORT_SYMBOL_GPL(vhost_poll_queue); |
87d6a412 | 284 | |
f8894913 JW |
285 | static void __vhost_vq_meta_reset(struct vhost_virtqueue *vq) |
286 | { | |
287 | int j; | |
288 | ||
289 | for (j = 0; j < VHOST_NUM_ADDRS; j++) | |
290 | vq->meta_iotlb[j] = NULL; | |
291 | } | |
292 | ||
293 | static void vhost_vq_meta_reset(struct vhost_dev *d) | |
294 | { | |
295 | int i; | |
296 | ||
297 | for (i = 0; i < d->nvqs; ++i) | |
298 | __vhost_vq_meta_reset(d->vqs[i]); | |
299 | } | |
300 | ||
3a4d5c94 MT |
301 | static void vhost_vq_reset(struct vhost_dev *dev, |
302 | struct vhost_virtqueue *vq) | |
303 | { | |
304 | vq->num = 1; | |
305 | vq->desc = NULL; | |
306 | vq->avail = NULL; | |
307 | vq->used = NULL; | |
308 | vq->last_avail_idx = 0; | |
809ecb9b | 309 | vq->last_used_event = 0; |
3a4d5c94 MT |
310 | vq->avail_idx = 0; |
311 | vq->last_used_idx = 0; | |
8ea8cf89 MT |
312 | vq->signalled_used = 0; |
313 | vq->signalled_used_valid = false; | |
3a4d5c94 | 314 | vq->used_flags = 0; |
3a4d5c94 MT |
315 | vq->log_used = false; |
316 | vq->log_addr = -1ull; | |
3a4d5c94 | 317 | vq->private_data = NULL; |
ea16c514 | 318 | vq->acked_features = 0; |
3a4d5c94 MT |
319 | vq->log_base = NULL; |
320 | vq->error_ctx = NULL; | |
321 | vq->error = NULL; | |
322 | vq->kick = NULL; | |
323 | vq->call_ctx = NULL; | |
324 | vq->call = NULL; | |
73a99f08 | 325 | vq->log_ctx = NULL; |
c5072037 GK |
326 | vhost_reset_is_le(vq); |
327 | vhost_disable_cross_endian(vq); | |
03088137 | 328 | vq->busyloop_timeout = 0; |
a9709d68 | 329 | vq->umem = NULL; |
6b1e6cc7 | 330 | vq->iotlb = NULL; |
f8894913 | 331 | __vhost_vq_meta_reset(vq); |
3a4d5c94 MT |
332 | } |
333 | ||
c23f3445 TH |
334 | static int vhost_worker(void *data) |
335 | { | |
336 | struct vhost_dev *dev = data; | |
04b96e55 JW |
337 | struct vhost_work *work, *work_next; |
338 | struct llist_node *node; | |
d7ffde35 | 339 | mm_segment_t oldfs = get_fs(); |
c23f3445 | 340 | |
d7ffde35 | 341 | set_fs(USER_DS); |
64e1c807 MT |
342 | use_mm(dev->mm); |
343 | ||
c23f3445 TH |
344 | for (;;) { |
345 | /* mb paired w/ kthread_stop */ | |
346 | set_current_state(TASK_INTERRUPTIBLE); | |
347 | ||
c23f3445 | 348 | if (kthread_should_stop()) { |
c23f3445 | 349 | __set_current_state(TASK_RUNNING); |
64e1c807 | 350 | break; |
c23f3445 | 351 | } |
04b96e55 JW |
352 | |
353 | node = llist_del_all(&dev->work_list); | |
354 | if (!node) | |
355 | schedule(); | |
356 | ||
357 | node = llist_reverse_order(node); | |
358 | /* make sure flag is seen after deletion */ | |
359 | smp_wmb(); | |
360 | llist_for_each_entry_safe(work, work_next, node, node) { | |
361 | clear_bit(VHOST_WORK_QUEUED, &work->flags); | |
c23f3445 TH |
362 | __set_current_state(TASK_RUNNING); |
363 | work->fn(work); | |
d550dda1 NHE |
364 | if (need_resched()) |
365 | schedule(); | |
04b96e55 | 366 | } |
c23f3445 | 367 | } |
64e1c807 | 368 | unuse_mm(dev->mm); |
d7ffde35 | 369 | set_fs(oldfs); |
64e1c807 | 370 | return 0; |
c23f3445 TH |
371 | } |
372 | ||
bab632d6 MT |
373 | static void vhost_vq_free_iovecs(struct vhost_virtqueue *vq) |
374 | { | |
375 | kfree(vq->indirect); | |
376 | vq->indirect = NULL; | |
377 | kfree(vq->log); | |
378 | vq->log = NULL; | |
379 | kfree(vq->heads); | |
380 | vq->heads = NULL; | |
bab632d6 MT |
381 | } |
382 | ||
e0e9b406 JW |
383 | /* Helper to allocate iovec buffers for all vqs. */ |
384 | static long vhost_dev_alloc_iovecs(struct vhost_dev *dev) | |
385 | { | |
6d5e6aa8 | 386 | struct vhost_virtqueue *vq; |
e0e9b406 | 387 | int i; |
d47effe1 | 388 | |
e0e9b406 | 389 | for (i = 0; i < dev->nvqs; ++i) { |
6d5e6aa8 AH |
390 | vq = dev->vqs[i]; |
391 | vq->indirect = kmalloc(sizeof *vq->indirect * UIO_MAXIOV, | |
392 | GFP_KERNEL); | |
393 | vq->log = kmalloc(sizeof *vq->log * UIO_MAXIOV, GFP_KERNEL); | |
394 | vq->heads = kmalloc(sizeof *vq->heads * UIO_MAXIOV, GFP_KERNEL); | |
395 | if (!vq->indirect || !vq->log || !vq->heads) | |
e0e9b406 JW |
396 | goto err_nomem; |
397 | } | |
398 | return 0; | |
d47effe1 | 399 | |
e0e9b406 | 400 | err_nomem: |
bab632d6 | 401 | for (; i >= 0; --i) |
3ab2e420 | 402 | vhost_vq_free_iovecs(dev->vqs[i]); |
e0e9b406 JW |
403 | return -ENOMEM; |
404 | } | |
405 | ||
406 | static void vhost_dev_free_iovecs(struct vhost_dev *dev) | |
407 | { | |
408 | int i; | |
d47effe1 | 409 | |
bab632d6 | 410 | for (i = 0; i < dev->nvqs; ++i) |
3ab2e420 | 411 | vhost_vq_free_iovecs(dev->vqs[i]); |
e0e9b406 JW |
412 | } |
413 | ||
59566b6e | 414 | void vhost_dev_init(struct vhost_dev *dev, |
3ab2e420 | 415 | struct vhost_virtqueue **vqs, int nvqs) |
3a4d5c94 | 416 | { |
6d5e6aa8 | 417 | struct vhost_virtqueue *vq; |
3a4d5c94 | 418 | int i; |
c23f3445 | 419 | |
3a4d5c94 MT |
420 | dev->vqs = vqs; |
421 | dev->nvqs = nvqs; | |
422 | mutex_init(&dev->mutex); | |
423 | dev->log_ctx = NULL; | |
424 | dev->log_file = NULL; | |
a9709d68 | 425 | dev->umem = NULL; |
6b1e6cc7 | 426 | dev->iotlb = NULL; |
3a4d5c94 | 427 | dev->mm = NULL; |
c23f3445 | 428 | dev->worker = NULL; |
04b96e55 | 429 | init_llist_head(&dev->work_list); |
6b1e6cc7 JW |
430 | init_waitqueue_head(&dev->wait); |
431 | INIT_LIST_HEAD(&dev->read_list); | |
432 | INIT_LIST_HEAD(&dev->pending_list); | |
433 | spin_lock_init(&dev->iotlb_lock); | |
04b96e55 | 434 | |
3a4d5c94 MT |
435 | |
436 | for (i = 0; i < dev->nvqs; ++i) { | |
6d5e6aa8 AH |
437 | vq = dev->vqs[i]; |
438 | vq->log = NULL; | |
439 | vq->indirect = NULL; | |
440 | vq->heads = NULL; | |
441 | vq->dev = dev; | |
442 | mutex_init(&vq->mutex); | |
443 | vhost_vq_reset(dev, vq); | |
444 | if (vq->handle_kick) | |
445 | vhost_poll_init(&vq->poll, vq->handle_kick, | |
446 | POLLIN, dev); | |
3a4d5c94 | 447 | } |
3a4d5c94 | 448 | } |
6ac1afbf | 449 | EXPORT_SYMBOL_GPL(vhost_dev_init); |
3a4d5c94 MT |
450 | |
451 | /* Caller should have device mutex */ | |
452 | long vhost_dev_check_owner(struct vhost_dev *dev) | |
453 | { | |
454 | /* Are you the owner? If not, I don't think you mean to do that */ | |
455 | return dev->mm == current->mm ? 0 : -EPERM; | |
456 | } | |
6ac1afbf | 457 | EXPORT_SYMBOL_GPL(vhost_dev_check_owner); |
3a4d5c94 | 458 | |
87d6a412 | 459 | struct vhost_attach_cgroups_struct { |
d47effe1 KK |
460 | struct vhost_work work; |
461 | struct task_struct *owner; | |
462 | int ret; | |
87d6a412 MT |
463 | }; |
464 | ||
465 | static void vhost_attach_cgroups_work(struct vhost_work *work) | |
466 | { | |
d47effe1 KK |
467 | struct vhost_attach_cgroups_struct *s; |
468 | ||
469 | s = container_of(work, struct vhost_attach_cgroups_struct, work); | |
470 | s->ret = cgroup_attach_task_all(s->owner, current); | |
87d6a412 MT |
471 | } |
472 | ||
473 | static int vhost_attach_cgroups(struct vhost_dev *dev) | |
474 | { | |
d47effe1 KK |
475 | struct vhost_attach_cgroups_struct attach; |
476 | ||
477 | attach.owner = current; | |
478 | vhost_work_init(&attach.work, vhost_attach_cgroups_work); | |
479 | vhost_work_queue(dev, &attach.work); | |
480 | vhost_work_flush(dev, &attach.work); | |
481 | return attach.ret; | |
87d6a412 MT |
482 | } |
483 | ||
05c05351 MT |
484 | /* Caller should have device mutex */ |
485 | bool vhost_dev_has_owner(struct vhost_dev *dev) | |
486 | { | |
487 | return dev->mm; | |
488 | } | |
6ac1afbf | 489 | EXPORT_SYMBOL_GPL(vhost_dev_has_owner); |
05c05351 | 490 | |
3a4d5c94 | 491 | /* Caller should have device mutex */ |
54db63c2 | 492 | long vhost_dev_set_owner(struct vhost_dev *dev) |
3a4d5c94 | 493 | { |
c23f3445 TH |
494 | struct task_struct *worker; |
495 | int err; | |
d47effe1 | 496 | |
3a4d5c94 | 497 | /* Is there an owner already? */ |
05c05351 | 498 | if (vhost_dev_has_owner(dev)) { |
c23f3445 TH |
499 | err = -EBUSY; |
500 | goto err_mm; | |
501 | } | |
d47effe1 | 502 | |
3a4d5c94 MT |
503 | /* No owner, become one */ |
504 | dev->mm = get_task_mm(current); | |
c23f3445 TH |
505 | worker = kthread_create(vhost_worker, dev, "vhost-%d", current->pid); |
506 | if (IS_ERR(worker)) { | |
507 | err = PTR_ERR(worker); | |
508 | goto err_worker; | |
509 | } | |
510 | ||
511 | dev->worker = worker; | |
87d6a412 MT |
512 | wake_up_process(worker); /* avoid contributing to loadavg */ |
513 | ||
514 | err = vhost_attach_cgroups(dev); | |
9e3d1957 MT |
515 | if (err) |
516 | goto err_cgroup; | |
c23f3445 | 517 | |
e0e9b406 JW |
518 | err = vhost_dev_alloc_iovecs(dev); |
519 | if (err) | |
520 | goto err_cgroup; | |
521 | ||
3a4d5c94 | 522 | return 0; |
9e3d1957 MT |
523 | err_cgroup: |
524 | kthread_stop(worker); | |
615cc221 | 525 | dev->worker = NULL; |
c23f3445 TH |
526 | err_worker: |
527 | if (dev->mm) | |
528 | mmput(dev->mm); | |
529 | dev->mm = NULL; | |
530 | err_mm: | |
531 | return err; | |
3a4d5c94 | 532 | } |
6ac1afbf | 533 | EXPORT_SYMBOL_GPL(vhost_dev_set_owner); |
3a4d5c94 | 534 | |
a9709d68 | 535 | static void *vhost_kvzalloc(unsigned long size) |
3a4d5c94 | 536 | { |
a9709d68 JW |
537 | void *n = kzalloc(size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
538 | ||
539 | if (!n) | |
540 | n = vzalloc(size); | |
541 | return n; | |
542 | } | |
543 | ||
544 | struct vhost_umem *vhost_dev_reset_owner_prepare(void) | |
545 | { | |
546 | return vhost_kvzalloc(sizeof(struct vhost_umem)); | |
150b9e51 | 547 | } |
6ac1afbf | 548 | EXPORT_SYMBOL_GPL(vhost_dev_reset_owner_prepare); |
3a4d5c94 | 549 | |
150b9e51 | 550 | /* Caller should have device mutex */ |
a9709d68 | 551 | void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_umem *umem) |
150b9e51 | 552 | { |
47283bef MT |
553 | int i; |
554 | ||
ea5d4046 | 555 | vhost_dev_cleanup(dev, true); |
3a4d5c94 | 556 | |
150b9e51 | 557 | /* Restore memory to default empty mapping. */ |
a9709d68 JW |
558 | INIT_LIST_HEAD(&umem->umem_list); |
559 | dev->umem = umem; | |
47283bef MT |
560 | /* We don't need VQ locks below since vhost_dev_cleanup makes sure |
561 | * VQs aren't running. | |
562 | */ | |
563 | for (i = 0; i < dev->nvqs; ++i) | |
a9709d68 | 564 | dev->vqs[i]->umem = umem; |
3a4d5c94 | 565 | } |
6ac1afbf | 566 | EXPORT_SYMBOL_GPL(vhost_dev_reset_owner); |
3a4d5c94 | 567 | |
b211616d | 568 | void vhost_dev_stop(struct vhost_dev *dev) |
bab632d6 MT |
569 | { |
570 | int i; | |
b211616d MT |
571 | |
572 | for (i = 0; i < dev->nvqs; ++i) { | |
3ab2e420 AH |
573 | if (dev->vqs[i]->kick && dev->vqs[i]->handle_kick) { |
574 | vhost_poll_stop(&dev->vqs[i]->poll); | |
575 | vhost_poll_flush(&dev->vqs[i]->poll); | |
b211616d | 576 | } |
bab632d6 | 577 | } |
bab632d6 | 578 | } |
6ac1afbf | 579 | EXPORT_SYMBOL_GPL(vhost_dev_stop); |
bab632d6 | 580 | |
6b1e6cc7 JW |
581 | static void vhost_umem_free(struct vhost_umem *umem, |
582 | struct vhost_umem_node *node) | |
583 | { | |
584 | vhost_umem_interval_tree_remove(node, &umem->umem_tree); | |
585 | list_del(&node->link); | |
586 | kfree(node); | |
587 | umem->numem--; | |
588 | } | |
589 | ||
a9709d68 JW |
590 | static void vhost_umem_clean(struct vhost_umem *umem) |
591 | { | |
592 | struct vhost_umem_node *node, *tmp; | |
593 | ||
594 | if (!umem) | |
595 | return; | |
596 | ||
6b1e6cc7 JW |
597 | list_for_each_entry_safe(node, tmp, &umem->umem_list, link) |
598 | vhost_umem_free(umem, node); | |
599 | ||
a9709d68 JW |
600 | kvfree(umem); |
601 | } | |
602 | ||
6b1e6cc7 JW |
603 | static void vhost_clear_msg(struct vhost_dev *dev) |
604 | { | |
605 | struct vhost_msg_node *node, *n; | |
606 | ||
607 | spin_lock(&dev->iotlb_lock); | |
608 | ||
609 | list_for_each_entry_safe(node, n, &dev->read_list, node) { | |
610 | list_del(&node->node); | |
611 | kfree(node); | |
612 | } | |
613 | ||
614 | list_for_each_entry_safe(node, n, &dev->pending_list, node) { | |
615 | list_del(&node->node); | |
616 | kfree(node); | |
617 | } | |
618 | ||
619 | spin_unlock(&dev->iotlb_lock); | |
620 | } | |
621 | ||
ea5d4046 MT |
622 | /* Caller should have device mutex if and only if locked is set */ |
623 | void vhost_dev_cleanup(struct vhost_dev *dev, bool locked) | |
3a4d5c94 MT |
624 | { |
625 | int i; | |
d47effe1 | 626 | |
3a4d5c94 | 627 | for (i = 0; i < dev->nvqs; ++i) { |
3ab2e420 AH |
628 | if (dev->vqs[i]->error_ctx) |
629 | eventfd_ctx_put(dev->vqs[i]->error_ctx); | |
630 | if (dev->vqs[i]->error) | |
631 | fput(dev->vqs[i]->error); | |
632 | if (dev->vqs[i]->kick) | |
633 | fput(dev->vqs[i]->kick); | |
634 | if (dev->vqs[i]->call_ctx) | |
635 | eventfd_ctx_put(dev->vqs[i]->call_ctx); | |
636 | if (dev->vqs[i]->call) | |
637 | fput(dev->vqs[i]->call); | |
638 | vhost_vq_reset(dev, dev->vqs[i]); | |
3a4d5c94 | 639 | } |
e0e9b406 | 640 | vhost_dev_free_iovecs(dev); |
3a4d5c94 MT |
641 | if (dev->log_ctx) |
642 | eventfd_ctx_put(dev->log_ctx); | |
643 | dev->log_ctx = NULL; | |
644 | if (dev->log_file) | |
645 | fput(dev->log_file); | |
646 | dev->log_file = NULL; | |
647 | /* No one will access memory at this point */ | |
a9709d68 JW |
648 | vhost_umem_clean(dev->umem); |
649 | dev->umem = NULL; | |
6b1e6cc7 JW |
650 | vhost_umem_clean(dev->iotlb); |
651 | dev->iotlb = NULL; | |
652 | vhost_clear_msg(dev); | |
653 | wake_up_interruptible_poll(&dev->wait, POLLIN | POLLRDNORM); | |
04b96e55 | 654 | WARN_ON(!llist_empty(&dev->work_list)); |
78b620ce ED |
655 | if (dev->worker) { |
656 | kthread_stop(dev->worker); | |
657 | dev->worker = NULL; | |
658 | } | |
533a19b4 MT |
659 | if (dev->mm) |
660 | mmput(dev->mm); | |
661 | dev->mm = NULL; | |
3a4d5c94 | 662 | } |
6ac1afbf | 663 | EXPORT_SYMBOL_GPL(vhost_dev_cleanup); |
3a4d5c94 MT |
664 | |
665 | static int log_access_ok(void __user *log_base, u64 addr, unsigned long sz) | |
666 | { | |
667 | u64 a = addr / VHOST_PAGE_SIZE / 8; | |
d47effe1 | 668 | |
3a4d5c94 MT |
669 | /* Make sure 64 bit math will not overflow. */ |
670 | if (a > ULONG_MAX - (unsigned long)log_base || | |
671 | a + (unsigned long)log_base > ULONG_MAX) | |
6d97e55f | 672 | return 0; |
3a4d5c94 MT |
673 | |
674 | return access_ok(VERIFY_WRITE, log_base + a, | |
675 | (sz + VHOST_PAGE_SIZE * 8 - 1) / VHOST_PAGE_SIZE / 8); | |
676 | } | |
677 | ||
ec33d031 MT |
678 | static bool vhost_overflow(u64 uaddr, u64 size) |
679 | { | |
680 | /* Make sure 64 bit math will not overflow. */ | |
681 | return uaddr > ULONG_MAX || size > ULONG_MAX || uaddr > ULONG_MAX - size; | |
682 | } | |
683 | ||
3a4d5c94 | 684 | /* Caller should have vq mutex and device mutex. */ |
a9709d68 | 685 | static int vq_memory_access_ok(void __user *log_base, struct vhost_umem *umem, |
3a4d5c94 MT |
686 | int log_all) |
687 | { | |
a9709d68 | 688 | struct vhost_umem_node *node; |
179b284e | 689 | |
a9709d68 | 690 | if (!umem) |
f8322fbe | 691 | return 0; |
179b284e | 692 | |
a9709d68 JW |
693 | list_for_each_entry(node, &umem->umem_list, link) { |
694 | unsigned long a = node->userspace_addr; | |
695 | ||
ec33d031 | 696 | if (vhost_overflow(node->userspace_addr, node->size)) |
3a4d5c94 | 697 | return 0; |
ec33d031 MT |
698 | |
699 | ||
700 | if (!access_ok(VERIFY_WRITE, (void __user *)a, | |
a9709d68 | 701 | node->size)) |
3a4d5c94 MT |
702 | return 0; |
703 | else if (log_all && !log_access_ok(log_base, | |
a9709d68 JW |
704 | node->start, |
705 | node->size)) | |
3a4d5c94 MT |
706 | return 0; |
707 | } | |
708 | return 1; | |
709 | } | |
710 | ||
f8894913 JW |
711 | static inline void __user *vhost_vq_meta_fetch(struct vhost_virtqueue *vq, |
712 | u64 addr, unsigned int size, | |
713 | int type) | |
714 | { | |
715 | const struct vhost_umem_node *node = vq->meta_iotlb[type]; | |
716 | ||
717 | if (!node) | |
718 | return NULL; | |
719 | ||
720 | return (void *)(uintptr_t)(node->userspace_addr + addr - node->start); | |
721 | } | |
722 | ||
3a4d5c94 MT |
723 | /* Can we switch to this memory table? */ |
724 | /* Caller should have device mutex but not vq mutex */ | |
a9709d68 | 725 | static int memory_access_ok(struct vhost_dev *d, struct vhost_umem *umem, |
3a4d5c94 MT |
726 | int log_all) |
727 | { | |
728 | int i; | |
d47effe1 | 729 | |
3a4d5c94 MT |
730 | for (i = 0; i < d->nvqs; ++i) { |
731 | int ok; | |
ea16c514 MT |
732 | bool log; |
733 | ||
3ab2e420 | 734 | mutex_lock(&d->vqs[i]->mutex); |
ea16c514 | 735 | log = log_all || vhost_has_feature(d->vqs[i], VHOST_F_LOG_ALL); |
3a4d5c94 | 736 | /* If ring is inactive, will check when it's enabled. */ |
3ab2e420 | 737 | if (d->vqs[i]->private_data) |
a9709d68 JW |
738 | ok = vq_memory_access_ok(d->vqs[i]->log_base, |
739 | umem, log); | |
3a4d5c94 MT |
740 | else |
741 | ok = 1; | |
3ab2e420 | 742 | mutex_unlock(&d->vqs[i]->mutex); |
3a4d5c94 MT |
743 | if (!ok) |
744 | return 0; | |
745 | } | |
746 | return 1; | |
747 | } | |
748 | ||
6b1e6cc7 JW |
749 | static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len, |
750 | struct iovec iov[], int iov_size, int access); | |
bfe2bc51 | 751 | |
72952cc0 | 752 | static int vhost_copy_to_user(struct vhost_virtqueue *vq, void __user *to, |
bfe2bc51 JW |
753 | const void *from, unsigned size) |
754 | { | |
6b1e6cc7 | 755 | int ret; |
bfe2bc51 | 756 | |
6b1e6cc7 JW |
757 | if (!vq->iotlb) |
758 | return __copy_to_user(to, from, size); | |
759 | else { | |
760 | /* This function should be called after iotlb | |
761 | * prefetch, which means we're sure that all vq | |
762 | * could be access through iotlb. So -EAGAIN should | |
763 | * not happen in this case. | |
764 | */ | |
6b1e6cc7 | 765 | struct iov_iter t; |
f8894913 JW |
766 | void __user *uaddr = vhost_vq_meta_fetch(vq, |
767 | (u64)(uintptr_t)to, size, | |
768 | VHOST_ADDR_DESC); | |
769 | ||
770 | if (uaddr) | |
771 | return __copy_to_user(uaddr, from, size); | |
772 | ||
6b1e6cc7 JW |
773 | ret = translate_desc(vq, (u64)(uintptr_t)to, size, vq->iotlb_iov, |
774 | ARRAY_SIZE(vq->iotlb_iov), | |
775 | VHOST_ACCESS_WO); | |
776 | if (ret < 0) | |
777 | goto out; | |
778 | iov_iter_init(&t, WRITE, vq->iotlb_iov, ret, size); | |
779 | ret = copy_to_iter(from, size, &t); | |
780 | if (ret == size) | |
781 | ret = 0; | |
782 | } | |
783 | out: | |
784 | return ret; | |
785 | } | |
bfe2bc51 JW |
786 | |
787 | static int vhost_copy_from_user(struct vhost_virtqueue *vq, void *to, | |
72952cc0 | 788 | void __user *from, unsigned size) |
bfe2bc51 | 789 | { |
6b1e6cc7 JW |
790 | int ret; |
791 | ||
792 | if (!vq->iotlb) | |
793 | return __copy_from_user(to, from, size); | |
794 | else { | |
795 | /* This function should be called after iotlb | |
796 | * prefetch, which means we're sure that vq | |
797 | * could be access through iotlb. So -EAGAIN should | |
798 | * not happen in this case. | |
799 | */ | |
f8894913 JW |
800 | void __user *uaddr = vhost_vq_meta_fetch(vq, |
801 | (u64)(uintptr_t)from, size, | |
802 | VHOST_ADDR_DESC); | |
6b1e6cc7 | 803 | struct iov_iter f; |
f8894913 JW |
804 | |
805 | if (uaddr) | |
806 | return __copy_from_user(to, uaddr, size); | |
807 | ||
6b1e6cc7 JW |
808 | ret = translate_desc(vq, (u64)(uintptr_t)from, size, vq->iotlb_iov, |
809 | ARRAY_SIZE(vq->iotlb_iov), | |
810 | VHOST_ACCESS_RO); | |
811 | if (ret < 0) { | |
812 | vq_err(vq, "IOTLB translation failure: uaddr " | |
813 | "%p size 0x%llx\n", from, | |
814 | (unsigned long long) size); | |
815 | goto out; | |
816 | } | |
817 | iov_iter_init(&f, READ, vq->iotlb_iov, ret, size); | |
818 | ret = copy_from_iter(to, size, &f); | |
819 | if (ret == size) | |
820 | ret = 0; | |
821 | } | |
822 | ||
823 | out: | |
824 | return ret; | |
825 | } | |
826 | ||
f8894913 JW |
827 | static void __user *__vhost_get_user_slow(struct vhost_virtqueue *vq, |
828 | void __user *addr, unsigned int size, | |
829 | int type) | |
6b1e6cc7 JW |
830 | { |
831 | int ret; | |
832 | ||
6b1e6cc7 JW |
833 | ret = translate_desc(vq, (u64)(uintptr_t)addr, size, vq->iotlb_iov, |
834 | ARRAY_SIZE(vq->iotlb_iov), | |
835 | VHOST_ACCESS_RO); | |
836 | if (ret < 0) { | |
837 | vq_err(vq, "IOTLB translation failure: uaddr " | |
838 | "%p size 0x%llx\n", addr, | |
839 | (unsigned long long) size); | |
840 | return NULL; | |
841 | } | |
842 | ||
843 | if (ret != 1 || vq->iotlb_iov[0].iov_len != size) { | |
844 | vq_err(vq, "Non atomic userspace memory access: uaddr " | |
845 | "%p size 0x%llx\n", addr, | |
846 | (unsigned long long) size); | |
847 | return NULL; | |
848 | } | |
849 | ||
850 | return vq->iotlb_iov[0].iov_base; | |
851 | } | |
852 | ||
f8894913 JW |
853 | /* This function should be called after iotlb |
854 | * prefetch, which means we're sure that vq | |
855 | * could be access through iotlb. So -EAGAIN should | |
856 | * not happen in this case. | |
857 | */ | |
858 | static inline void __user *__vhost_get_user(struct vhost_virtqueue *vq, | |
859 | void *addr, unsigned int size, | |
860 | int type) | |
861 | { | |
862 | void __user *uaddr = vhost_vq_meta_fetch(vq, | |
863 | (u64)(uintptr_t)addr, size, type); | |
864 | if (uaddr) | |
865 | return uaddr; | |
866 | ||
867 | return __vhost_get_user_slow(vq, addr, size, type); | |
868 | } | |
869 | ||
870 | #define vhost_put_user(vq, x, ptr) \ | |
6b1e6cc7 JW |
871 | ({ \ |
872 | int ret = -EFAULT; \ | |
873 | if (!vq->iotlb) { \ | |
874 | ret = __put_user(x, ptr); \ | |
875 | } else { \ | |
876 | __typeof__(ptr) to = \ | |
f8894913 JW |
877 | (__typeof__(ptr)) __vhost_get_user(vq, ptr, \ |
878 | sizeof(*ptr), VHOST_ADDR_USED); \ | |
6b1e6cc7 JW |
879 | if (to != NULL) \ |
880 | ret = __put_user(x, to); \ | |
881 | else \ | |
882 | ret = -EFAULT; \ | |
883 | } \ | |
884 | ret; \ | |
885 | }) | |
886 | ||
f8894913 | 887 | #define vhost_get_user(vq, x, ptr, type) \ |
6b1e6cc7 JW |
888 | ({ \ |
889 | int ret; \ | |
890 | if (!vq->iotlb) { \ | |
891 | ret = __get_user(x, ptr); \ | |
892 | } else { \ | |
893 | __typeof__(ptr) from = \ | |
f8894913 JW |
894 | (__typeof__(ptr)) __vhost_get_user(vq, ptr, \ |
895 | sizeof(*ptr), \ | |
896 | type); \ | |
6b1e6cc7 JW |
897 | if (from != NULL) \ |
898 | ret = __get_user(x, from); \ | |
899 | else \ | |
900 | ret = -EFAULT; \ | |
901 | } \ | |
902 | ret; \ | |
903 | }) | |
904 | ||
f8894913 JW |
905 | #define vhost_get_avail(vq, x, ptr) \ |
906 | vhost_get_user(vq, x, ptr, VHOST_ADDR_AVAIL) | |
907 | ||
908 | #define vhost_get_used(vq, x, ptr) \ | |
909 | vhost_get_user(vq, x, ptr, VHOST_ADDR_USED) | |
910 | ||
6b1e6cc7 JW |
911 | static void vhost_dev_lock_vqs(struct vhost_dev *d) |
912 | { | |
913 | int i = 0; | |
914 | for (i = 0; i < d->nvqs; ++i) | |
915 | mutex_lock(&d->vqs[i]->mutex); | |
916 | } | |
917 | ||
918 | static void vhost_dev_unlock_vqs(struct vhost_dev *d) | |
919 | { | |
920 | int i = 0; | |
921 | for (i = 0; i < d->nvqs; ++i) | |
922 | mutex_unlock(&d->vqs[i]->mutex); | |
923 | } | |
924 | ||
925 | static int vhost_new_umem_range(struct vhost_umem *umem, | |
926 | u64 start, u64 size, u64 end, | |
927 | u64 userspace_addr, int perm) | |
928 | { | |
929 | struct vhost_umem_node *tmp, *node = kmalloc(sizeof(*node), GFP_ATOMIC); | |
930 | ||
931 | if (!node) | |
932 | return -ENOMEM; | |
933 | ||
934 | if (umem->numem == max_iotlb_entries) { | |
935 | tmp = list_first_entry(&umem->umem_list, typeof(*tmp), link); | |
936 | vhost_umem_free(umem, tmp); | |
937 | } | |
938 | ||
939 | node->start = start; | |
940 | node->size = size; | |
941 | node->last = end; | |
942 | node->userspace_addr = userspace_addr; | |
943 | node->perm = perm; | |
944 | INIT_LIST_HEAD(&node->link); | |
945 | list_add_tail(&node->link, &umem->umem_list); | |
946 | vhost_umem_interval_tree_insert(node, &umem->umem_tree); | |
947 | umem->numem++; | |
948 | ||
949 | return 0; | |
950 | } | |
951 | ||
952 | static void vhost_del_umem_range(struct vhost_umem *umem, | |
953 | u64 start, u64 end) | |
954 | { | |
955 | struct vhost_umem_node *node; | |
956 | ||
957 | while ((node = vhost_umem_interval_tree_iter_first(&umem->umem_tree, | |
958 | start, end))) | |
959 | vhost_umem_free(umem, node); | |
960 | } | |
961 | ||
962 | static void vhost_iotlb_notify_vq(struct vhost_dev *d, | |
963 | struct vhost_iotlb_msg *msg) | |
964 | { | |
965 | struct vhost_msg_node *node, *n; | |
966 | ||
967 | spin_lock(&d->iotlb_lock); | |
968 | ||
969 | list_for_each_entry_safe(node, n, &d->pending_list, node) { | |
970 | struct vhost_iotlb_msg *vq_msg = &node->msg.iotlb; | |
971 | if (msg->iova <= vq_msg->iova && | |
972 | msg->iova + msg->size - 1 > vq_msg->iova && | |
973 | vq_msg->type == VHOST_IOTLB_MISS) { | |
974 | vhost_poll_queue(&node->vq->poll); | |
975 | list_del(&node->node); | |
976 | kfree(node); | |
977 | } | |
978 | } | |
979 | ||
980 | spin_unlock(&d->iotlb_lock); | |
981 | } | |
982 | ||
983 | static int umem_access_ok(u64 uaddr, u64 size, int access) | |
984 | { | |
985 | unsigned long a = uaddr; | |
986 | ||
ec33d031 MT |
987 | /* Make sure 64 bit math will not overflow. */ |
988 | if (vhost_overflow(uaddr, size)) | |
989 | return -EFAULT; | |
990 | ||
6b1e6cc7 JW |
991 | if ((access & VHOST_ACCESS_RO) && |
992 | !access_ok(VERIFY_READ, (void __user *)a, size)) | |
993 | return -EFAULT; | |
994 | if ((access & VHOST_ACCESS_WO) && | |
995 | !access_ok(VERIFY_WRITE, (void __user *)a, size)) | |
996 | return -EFAULT; | |
997 | return 0; | |
998 | } | |
999 | ||
72952cc0 MT |
1000 | static int vhost_process_iotlb_msg(struct vhost_dev *dev, |
1001 | struct vhost_iotlb_msg *msg) | |
6b1e6cc7 JW |
1002 | { |
1003 | int ret = 0; | |
1004 | ||
1005 | vhost_dev_lock_vqs(dev); | |
1006 | switch (msg->type) { | |
1007 | case VHOST_IOTLB_UPDATE: | |
1008 | if (!dev->iotlb) { | |
1009 | ret = -EFAULT; | |
1010 | break; | |
1011 | } | |
1012 | if (umem_access_ok(msg->uaddr, msg->size, msg->perm)) { | |
1013 | ret = -EFAULT; | |
1014 | break; | |
1015 | } | |
f8894913 | 1016 | vhost_vq_meta_reset(dev); |
6b1e6cc7 JW |
1017 | if (vhost_new_umem_range(dev->iotlb, msg->iova, msg->size, |
1018 | msg->iova + msg->size - 1, | |
1019 | msg->uaddr, msg->perm)) { | |
1020 | ret = -ENOMEM; | |
1021 | break; | |
1022 | } | |
1023 | vhost_iotlb_notify_vq(dev, msg); | |
1024 | break; | |
1025 | case VHOST_IOTLB_INVALIDATE: | |
f8894913 | 1026 | vhost_vq_meta_reset(dev); |
6b1e6cc7 JW |
1027 | vhost_del_umem_range(dev->iotlb, msg->iova, |
1028 | msg->iova + msg->size - 1); | |
1029 | break; | |
1030 | default: | |
1031 | ret = -EINVAL; | |
1032 | break; | |
1033 | } | |
1034 | ||
1035 | vhost_dev_unlock_vqs(dev); | |
1036 | return ret; | |
1037 | } | |
1038 | ssize_t vhost_chr_write_iter(struct vhost_dev *dev, | |
1039 | struct iov_iter *from) | |
1040 | { | |
1041 | struct vhost_msg_node node; | |
1042 | unsigned size = sizeof(struct vhost_msg); | |
1043 | size_t ret; | |
1044 | int err; | |
1045 | ||
1046 | if (iov_iter_count(from) < size) | |
1047 | return 0; | |
1048 | ret = copy_from_iter(&node.msg, size, from); | |
1049 | if (ret != size) | |
1050 | goto done; | |
1051 | ||
1052 | switch (node.msg.type) { | |
1053 | case VHOST_IOTLB_MSG: | |
1054 | err = vhost_process_iotlb_msg(dev, &node.msg.iotlb); | |
1055 | if (err) | |
1056 | ret = err; | |
1057 | break; | |
1058 | default: | |
1059 | ret = -EINVAL; | |
1060 | break; | |
1061 | } | |
1062 | ||
1063 | done: | |
1064 | return ret; | |
1065 | } | |
1066 | EXPORT_SYMBOL(vhost_chr_write_iter); | |
1067 | ||
1068 | unsigned int vhost_chr_poll(struct file *file, struct vhost_dev *dev, | |
1069 | poll_table *wait) | |
1070 | { | |
1071 | unsigned int mask = 0; | |
1072 | ||
1073 | poll_wait(file, &dev->wait, wait); | |
1074 | ||
1075 | if (!list_empty(&dev->read_list)) | |
1076 | mask |= POLLIN | POLLRDNORM; | |
1077 | ||
1078 | return mask; | |
1079 | } | |
1080 | EXPORT_SYMBOL(vhost_chr_poll); | |
1081 | ||
1082 | ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to, | |
1083 | int noblock) | |
1084 | { | |
1085 | DEFINE_WAIT(wait); | |
1086 | struct vhost_msg_node *node; | |
1087 | ssize_t ret = 0; | |
1088 | unsigned size = sizeof(struct vhost_msg); | |
1089 | ||
1090 | if (iov_iter_count(to) < size) | |
1091 | return 0; | |
1092 | ||
1093 | while (1) { | |
1094 | if (!noblock) | |
1095 | prepare_to_wait(&dev->wait, &wait, | |
1096 | TASK_INTERRUPTIBLE); | |
1097 | ||
1098 | node = vhost_dequeue_msg(dev, &dev->read_list); | |
1099 | if (node) | |
1100 | break; | |
1101 | if (noblock) { | |
1102 | ret = -EAGAIN; | |
1103 | break; | |
1104 | } | |
1105 | if (signal_pending(current)) { | |
1106 | ret = -ERESTARTSYS; | |
1107 | break; | |
1108 | } | |
1109 | if (!dev->iotlb) { | |
1110 | ret = -EBADFD; | |
1111 | break; | |
1112 | } | |
1113 | ||
1114 | schedule(); | |
1115 | } | |
1116 | ||
1117 | if (!noblock) | |
1118 | finish_wait(&dev->wait, &wait); | |
1119 | ||
1120 | if (node) { | |
1121 | ret = copy_to_iter(&node->msg, size, to); | |
1122 | ||
1123 | if (ret != size || node->msg.type != VHOST_IOTLB_MISS) { | |
1124 | kfree(node); | |
1125 | return ret; | |
1126 | } | |
1127 | ||
1128 | vhost_enqueue_msg(dev, &dev->pending_list, node); | |
1129 | } | |
1130 | ||
1131 | return ret; | |
1132 | } | |
1133 | EXPORT_SYMBOL_GPL(vhost_chr_read_iter); | |
1134 | ||
1135 | static int vhost_iotlb_miss(struct vhost_virtqueue *vq, u64 iova, int access) | |
1136 | { | |
1137 | struct vhost_dev *dev = vq->dev; | |
1138 | struct vhost_msg_node *node; | |
1139 | struct vhost_iotlb_msg *msg; | |
1140 | ||
1141 | node = vhost_new_msg(vq, VHOST_IOTLB_MISS); | |
1142 | if (!node) | |
1143 | return -ENOMEM; | |
1144 | ||
1145 | msg = &node->msg.iotlb; | |
1146 | msg->type = VHOST_IOTLB_MISS; | |
1147 | msg->iova = iova; | |
1148 | msg->perm = access; | |
1149 | ||
1150 | vhost_enqueue_msg(dev, &dev->read_list, node); | |
1151 | ||
1152 | return 0; | |
bfe2bc51 JW |
1153 | } |
1154 | ||
ea16c514 | 1155 | static int vq_access_ok(struct vhost_virtqueue *vq, unsigned int num, |
3a4d5c94 MT |
1156 | struct vring_desc __user *desc, |
1157 | struct vring_avail __user *avail, | |
1158 | struct vring_used __user *used) | |
6b1e6cc7 | 1159 | |
3a4d5c94 | 1160 | { |
ea16c514 | 1161 | size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; |
6b1e6cc7 | 1162 | |
3a4d5c94 MT |
1163 | return access_ok(VERIFY_READ, desc, num * sizeof *desc) && |
1164 | access_ok(VERIFY_READ, avail, | |
8ea8cf89 | 1165 | sizeof *avail + num * sizeof *avail->ring + s) && |
3a4d5c94 | 1166 | access_ok(VERIFY_WRITE, used, |
8ea8cf89 | 1167 | sizeof *used + num * sizeof *used->ring + s); |
3a4d5c94 MT |
1168 | } |
1169 | ||
f8894913 JW |
1170 | static void vhost_vq_meta_update(struct vhost_virtqueue *vq, |
1171 | const struct vhost_umem_node *node, | |
1172 | int type) | |
1173 | { | |
1174 | int access = (type == VHOST_ADDR_USED) ? | |
1175 | VHOST_ACCESS_WO : VHOST_ACCESS_RO; | |
1176 | ||
1177 | if (likely(node->perm & access)) | |
1178 | vq->meta_iotlb[type] = node; | |
1179 | } | |
1180 | ||
6b1e6cc7 | 1181 | static int iotlb_access_ok(struct vhost_virtqueue *vq, |
f8894913 | 1182 | int access, u64 addr, u64 len, int type) |
6b1e6cc7 JW |
1183 | { |
1184 | const struct vhost_umem_node *node; | |
1185 | struct vhost_umem *umem = vq->iotlb; | |
f8894913 JW |
1186 | u64 s = 0, size, orig_addr = addr; |
1187 | ||
1188 | if (vhost_vq_meta_fetch(vq, addr, len, type)) | |
1189 | return true; | |
6b1e6cc7 JW |
1190 | |
1191 | while (len > s) { | |
1192 | node = vhost_umem_interval_tree_iter_first(&umem->umem_tree, | |
1193 | addr, | |
1194 | addr + len - 1); | |
1195 | if (node == NULL || node->start > addr) { | |
1196 | vhost_iotlb_miss(vq, addr, access); | |
1197 | return false; | |
1198 | } else if (!(node->perm & access)) { | |
1199 | /* Report the possible access violation by | |
1200 | * request another translation from userspace. | |
1201 | */ | |
1202 | return false; | |
1203 | } | |
1204 | ||
1205 | size = node->size - addr + node->start; | |
f8894913 JW |
1206 | |
1207 | if (orig_addr == addr && size >= len) | |
1208 | vhost_vq_meta_update(vq, node, type); | |
1209 | ||
6b1e6cc7 JW |
1210 | s += size; |
1211 | addr += size; | |
1212 | } | |
1213 | ||
1214 | return true; | |
1215 | } | |
1216 | ||
1217 | int vq_iotlb_prefetch(struct vhost_virtqueue *vq) | |
1218 | { | |
1219 | size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; | |
1220 | unsigned int num = vq->num; | |
1221 | ||
1222 | if (!vq->iotlb) | |
1223 | return 1; | |
1224 | ||
1225 | return iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->desc, | |
f8894913 | 1226 | num * sizeof(*vq->desc), VHOST_ADDR_DESC) && |
6b1e6cc7 JW |
1227 | iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->avail, |
1228 | sizeof *vq->avail + | |
f8894913 JW |
1229 | num * sizeof(*vq->avail->ring) + s, |
1230 | VHOST_ADDR_AVAIL) && | |
6b1e6cc7 JW |
1231 | iotlb_access_ok(vq, VHOST_ACCESS_WO, (u64)(uintptr_t)vq->used, |
1232 | sizeof *vq->used + | |
f8894913 JW |
1233 | num * sizeof(*vq->used->ring) + s, |
1234 | VHOST_ADDR_USED); | |
6b1e6cc7 JW |
1235 | } |
1236 | EXPORT_SYMBOL_GPL(vq_iotlb_prefetch); | |
1237 | ||
3a4d5c94 MT |
1238 | /* Can we log writes? */ |
1239 | /* Caller should have device mutex but not vq mutex */ | |
1240 | int vhost_log_access_ok(struct vhost_dev *dev) | |
1241 | { | |
a9709d68 | 1242 | return memory_access_ok(dev, dev->umem, 1); |
3a4d5c94 | 1243 | } |
6ac1afbf | 1244 | EXPORT_SYMBOL_GPL(vhost_log_access_ok); |
3a4d5c94 MT |
1245 | |
1246 | /* Verify access for write logging. */ | |
1247 | /* Caller should have vq mutex and device mutex */ | |
ea16c514 | 1248 | static int vq_log_access_ok(struct vhost_virtqueue *vq, |
8ea8cf89 | 1249 | void __user *log_base) |
3a4d5c94 | 1250 | { |
ea16c514 | 1251 | size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; |
28457ee6 | 1252 | |
a9709d68 | 1253 | return vq_memory_access_ok(log_base, vq->umem, |
ea16c514 | 1254 | vhost_has_feature(vq, VHOST_F_LOG_ALL)) && |
3a4d5c94 MT |
1255 | (!vq->log_used || log_access_ok(log_base, vq->log_addr, |
1256 | sizeof *vq->used + | |
8ea8cf89 | 1257 | vq->num * sizeof *vq->used->ring + s)); |
3a4d5c94 MT |
1258 | } |
1259 | ||
1260 | /* Can we start vq? */ | |
1261 | /* Caller should have vq mutex and device mutex */ | |
1262 | int vhost_vq_access_ok(struct vhost_virtqueue *vq) | |
1263 | { | |
6b1e6cc7 JW |
1264 | if (vq->iotlb) { |
1265 | /* When device IOTLB was used, the access validation | |
1266 | * will be validated during prefetching. | |
1267 | */ | |
1268 | return 1; | |
1269 | } | |
ea16c514 MT |
1270 | return vq_access_ok(vq, vq->num, vq->desc, vq->avail, vq->used) && |
1271 | vq_log_access_ok(vq, vq->log_base); | |
3a4d5c94 | 1272 | } |
6ac1afbf | 1273 | EXPORT_SYMBOL_GPL(vhost_vq_access_ok); |
3a4d5c94 | 1274 | |
6b1e6cc7 JW |
1275 | static struct vhost_umem *vhost_umem_alloc(void) |
1276 | { | |
1277 | struct vhost_umem *umem = vhost_kvzalloc(sizeof(*umem)); | |
1278 | ||
1279 | if (!umem) | |
1280 | return NULL; | |
1281 | ||
1282 | umem->umem_tree = RB_ROOT; | |
1283 | umem->numem = 0; | |
1284 | INIT_LIST_HEAD(&umem->umem_list); | |
1285 | ||
1286 | return umem; | |
1287 | } | |
1288 | ||
3a4d5c94 MT |
1289 | static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m) |
1290 | { | |
a9709d68 JW |
1291 | struct vhost_memory mem, *newmem; |
1292 | struct vhost_memory_region *region; | |
a9709d68 | 1293 | struct vhost_umem *newumem, *oldumem; |
3a4d5c94 | 1294 | unsigned long size = offsetof(struct vhost_memory, regions); |
98f9ca0a | 1295 | int i; |
d47effe1 | 1296 | |
7ad9c9d2 TY |
1297 | if (copy_from_user(&mem, m, size)) |
1298 | return -EFAULT; | |
3a4d5c94 MT |
1299 | if (mem.padding) |
1300 | return -EOPNOTSUPP; | |
c9ce42f7 | 1301 | if (mem.nregions > max_mem_regions) |
3a4d5c94 | 1302 | return -E2BIG; |
4de7255f | 1303 | newmem = vhost_kvzalloc(size + mem.nregions * sizeof(*m->regions)); |
3a4d5c94 MT |
1304 | if (!newmem) |
1305 | return -ENOMEM; | |
1306 | ||
1307 | memcpy(newmem, &mem, size); | |
7ad9c9d2 TY |
1308 | if (copy_from_user(newmem->regions, m->regions, |
1309 | mem.nregions * sizeof *m->regions)) { | |
bcfeacab | 1310 | kvfree(newmem); |
7ad9c9d2 | 1311 | return -EFAULT; |
3a4d5c94 MT |
1312 | } |
1313 | ||
6b1e6cc7 | 1314 | newumem = vhost_umem_alloc(); |
a9709d68 | 1315 | if (!newumem) { |
4de7255f | 1316 | kvfree(newmem); |
a9709d68 JW |
1317 | return -ENOMEM; |
1318 | } | |
1319 | ||
a9709d68 JW |
1320 | for (region = newmem->regions; |
1321 | region < newmem->regions + mem.nregions; | |
1322 | region++) { | |
6b1e6cc7 JW |
1323 | if (vhost_new_umem_range(newumem, |
1324 | region->guest_phys_addr, | |
1325 | region->memory_size, | |
1326 | region->guest_phys_addr + | |
1327 | region->memory_size - 1, | |
1328 | region->userspace_addr, | |
1329 | VHOST_ACCESS_RW)) | |
a9709d68 | 1330 | goto err; |
a02c3789 | 1331 | } |
a9709d68 JW |
1332 | |
1333 | if (!memory_access_ok(d, newumem, 0)) | |
1334 | goto err; | |
1335 | ||
1336 | oldumem = d->umem; | |
1337 | d->umem = newumem; | |
98f9ca0a | 1338 | |
47283bef | 1339 | /* All memory accesses are done under some VQ mutex. */ |
98f9ca0a MT |
1340 | for (i = 0; i < d->nvqs; ++i) { |
1341 | mutex_lock(&d->vqs[i]->mutex); | |
a9709d68 | 1342 | d->vqs[i]->umem = newumem; |
98f9ca0a MT |
1343 | mutex_unlock(&d->vqs[i]->mutex); |
1344 | } | |
a9709d68 JW |
1345 | |
1346 | kvfree(newmem); | |
1347 | vhost_umem_clean(oldumem); | |
3a4d5c94 | 1348 | return 0; |
a9709d68 JW |
1349 | |
1350 | err: | |
1351 | vhost_umem_clean(newumem); | |
1352 | kvfree(newmem); | |
1353 | return -EFAULT; | |
3a4d5c94 MT |
1354 | } |
1355 | ||
935cdee7 | 1356 | long vhost_vring_ioctl(struct vhost_dev *d, int ioctl, void __user *argp) |
3a4d5c94 | 1357 | { |
cecb46f1 AV |
1358 | struct file *eventfp, *filep = NULL; |
1359 | bool pollstart = false, pollstop = false; | |
3a4d5c94 MT |
1360 | struct eventfd_ctx *ctx = NULL; |
1361 | u32 __user *idxp = argp; | |
1362 | struct vhost_virtqueue *vq; | |
1363 | struct vhost_vring_state s; | |
1364 | struct vhost_vring_file f; | |
1365 | struct vhost_vring_addr a; | |
1366 | u32 idx; | |
1367 | long r; | |
1368 | ||
1369 | r = get_user(idx, idxp); | |
1370 | if (r < 0) | |
1371 | return r; | |
0f3d9a17 | 1372 | if (idx >= d->nvqs) |
3a4d5c94 MT |
1373 | return -ENOBUFS; |
1374 | ||
3ab2e420 | 1375 | vq = d->vqs[idx]; |
3a4d5c94 MT |
1376 | |
1377 | mutex_lock(&vq->mutex); | |
1378 | ||
1379 | switch (ioctl) { | |
1380 | case VHOST_SET_VRING_NUM: | |
1381 | /* Resizing ring with an active backend? | |
1382 | * You don't want to do that. */ | |
1383 | if (vq->private_data) { | |
1384 | r = -EBUSY; | |
1385 | break; | |
1386 | } | |
7ad9c9d2 TY |
1387 | if (copy_from_user(&s, argp, sizeof s)) { |
1388 | r = -EFAULT; | |
3a4d5c94 | 1389 | break; |
7ad9c9d2 | 1390 | } |
3a4d5c94 MT |
1391 | if (!s.num || s.num > 0xffff || (s.num & (s.num - 1))) { |
1392 | r = -EINVAL; | |
1393 | break; | |
1394 | } | |
1395 | vq->num = s.num; | |
1396 | break; | |
1397 | case VHOST_SET_VRING_BASE: | |
1398 | /* Moving base with an active backend? | |
1399 | * You don't want to do that. */ | |
1400 | if (vq->private_data) { | |
1401 | r = -EBUSY; | |
1402 | break; | |
1403 | } | |
7ad9c9d2 TY |
1404 | if (copy_from_user(&s, argp, sizeof s)) { |
1405 | r = -EFAULT; | |
3a4d5c94 | 1406 | break; |
7ad9c9d2 | 1407 | } |
3a4d5c94 MT |
1408 | if (s.num > 0xffff) { |
1409 | r = -EINVAL; | |
1410 | break; | |
1411 | } | |
809ecb9b | 1412 | vq->last_avail_idx = vq->last_used_event = s.num; |
3a4d5c94 MT |
1413 | /* Forget the cached index value. */ |
1414 | vq->avail_idx = vq->last_avail_idx; | |
1415 | break; | |
1416 | case VHOST_GET_VRING_BASE: | |
1417 | s.index = idx; | |
1418 | s.num = vq->last_avail_idx; | |
7ad9c9d2 TY |
1419 | if (copy_to_user(argp, &s, sizeof s)) |
1420 | r = -EFAULT; | |
3a4d5c94 MT |
1421 | break; |
1422 | case VHOST_SET_VRING_ADDR: | |
7ad9c9d2 TY |
1423 | if (copy_from_user(&a, argp, sizeof a)) { |
1424 | r = -EFAULT; | |
3a4d5c94 | 1425 | break; |
7ad9c9d2 | 1426 | } |
3a4d5c94 MT |
1427 | if (a.flags & ~(0x1 << VHOST_VRING_F_LOG)) { |
1428 | r = -EOPNOTSUPP; | |
1429 | break; | |
1430 | } | |
1431 | /* For 32bit, verify that the top 32bits of the user | |
1432 | data are set to zero. */ | |
1433 | if ((u64)(unsigned long)a.desc_user_addr != a.desc_user_addr || | |
1434 | (u64)(unsigned long)a.used_user_addr != a.used_user_addr || | |
1435 | (u64)(unsigned long)a.avail_user_addr != a.avail_user_addr) { | |
1436 | r = -EFAULT; | |
1437 | break; | |
1438 | } | |
5d9a07b0 MT |
1439 | |
1440 | /* Make sure it's safe to cast pointers to vring types. */ | |
1441 | BUILD_BUG_ON(__alignof__ *vq->avail > VRING_AVAIL_ALIGN_SIZE); | |
1442 | BUILD_BUG_ON(__alignof__ *vq->used > VRING_USED_ALIGN_SIZE); | |
1443 | if ((a.avail_user_addr & (VRING_AVAIL_ALIGN_SIZE - 1)) || | |
1444 | (a.used_user_addr & (VRING_USED_ALIGN_SIZE - 1)) || | |
d5424838 | 1445 | (a.log_guest_addr & (VRING_USED_ALIGN_SIZE - 1))) { |
3a4d5c94 MT |
1446 | r = -EINVAL; |
1447 | break; | |
1448 | } | |
1449 | ||
1450 | /* We only verify access here if backend is configured. | |
1451 | * If it is not, we don't as size might not have been setup. | |
1452 | * We will verify when backend is configured. */ | |
1453 | if (vq->private_data) { | |
ea16c514 | 1454 | if (!vq_access_ok(vq, vq->num, |
3a4d5c94 MT |
1455 | (void __user *)(unsigned long)a.desc_user_addr, |
1456 | (void __user *)(unsigned long)a.avail_user_addr, | |
1457 | (void __user *)(unsigned long)a.used_user_addr)) { | |
1458 | r = -EINVAL; | |
1459 | break; | |
1460 | } | |
1461 | ||
1462 | /* Also validate log access for used ring if enabled. */ | |
1463 | if ((a.flags & (0x1 << VHOST_VRING_F_LOG)) && | |
1464 | !log_access_ok(vq->log_base, a.log_guest_addr, | |
1465 | sizeof *vq->used + | |
1466 | vq->num * sizeof *vq->used->ring)) { | |
1467 | r = -EINVAL; | |
1468 | break; | |
1469 | } | |
1470 | } | |
1471 | ||
3a4d5c94 MT |
1472 | vq->log_used = !!(a.flags & (0x1 << VHOST_VRING_F_LOG)); |
1473 | vq->desc = (void __user *)(unsigned long)a.desc_user_addr; | |
1474 | vq->avail = (void __user *)(unsigned long)a.avail_user_addr; | |
1475 | vq->log_addr = a.log_guest_addr; | |
1476 | vq->used = (void __user *)(unsigned long)a.used_user_addr; | |
1477 | break; | |
1478 | case VHOST_SET_VRING_KICK: | |
7ad9c9d2 TY |
1479 | if (copy_from_user(&f, argp, sizeof f)) { |
1480 | r = -EFAULT; | |
3a4d5c94 | 1481 | break; |
7ad9c9d2 | 1482 | } |
3a4d5c94 | 1483 | eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd); |
535297a6 MT |
1484 | if (IS_ERR(eventfp)) { |
1485 | r = PTR_ERR(eventfp); | |
1486 | break; | |
1487 | } | |
3a4d5c94 | 1488 | if (eventfp != vq->kick) { |
cecb46f1 AV |
1489 | pollstop = (filep = vq->kick) != NULL; |
1490 | pollstart = (vq->kick = eventfp) != NULL; | |
3a4d5c94 MT |
1491 | } else |
1492 | filep = eventfp; | |
1493 | break; | |
1494 | case VHOST_SET_VRING_CALL: | |
7ad9c9d2 TY |
1495 | if (copy_from_user(&f, argp, sizeof f)) { |
1496 | r = -EFAULT; | |
3a4d5c94 | 1497 | break; |
7ad9c9d2 | 1498 | } |
3a4d5c94 | 1499 | eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd); |
535297a6 MT |
1500 | if (IS_ERR(eventfp)) { |
1501 | r = PTR_ERR(eventfp); | |
1502 | break; | |
1503 | } | |
3a4d5c94 MT |
1504 | if (eventfp != vq->call) { |
1505 | filep = vq->call; | |
1506 | ctx = vq->call_ctx; | |
1507 | vq->call = eventfp; | |
1508 | vq->call_ctx = eventfp ? | |
1509 | eventfd_ctx_fileget(eventfp) : NULL; | |
1510 | } else | |
1511 | filep = eventfp; | |
1512 | break; | |
1513 | case VHOST_SET_VRING_ERR: | |
7ad9c9d2 TY |
1514 | if (copy_from_user(&f, argp, sizeof f)) { |
1515 | r = -EFAULT; | |
3a4d5c94 | 1516 | break; |
7ad9c9d2 | 1517 | } |
3a4d5c94 | 1518 | eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd); |
535297a6 MT |
1519 | if (IS_ERR(eventfp)) { |
1520 | r = PTR_ERR(eventfp); | |
1521 | break; | |
1522 | } | |
3a4d5c94 MT |
1523 | if (eventfp != vq->error) { |
1524 | filep = vq->error; | |
1525 | vq->error = eventfp; | |
1526 | ctx = vq->error_ctx; | |
1527 | vq->error_ctx = eventfp ? | |
1528 | eventfd_ctx_fileget(eventfp) : NULL; | |
1529 | } else | |
1530 | filep = eventfp; | |
1531 | break; | |
2751c988 GK |
1532 | case VHOST_SET_VRING_ENDIAN: |
1533 | r = vhost_set_vring_endian(vq, argp); | |
1534 | break; | |
1535 | case VHOST_GET_VRING_ENDIAN: | |
1536 | r = vhost_get_vring_endian(vq, idx, argp); | |
1537 | break; | |
03088137 JW |
1538 | case VHOST_SET_VRING_BUSYLOOP_TIMEOUT: |
1539 | if (copy_from_user(&s, argp, sizeof(s))) { | |
1540 | r = -EFAULT; | |
1541 | break; | |
1542 | } | |
1543 | vq->busyloop_timeout = s.num; | |
1544 | break; | |
1545 | case VHOST_GET_VRING_BUSYLOOP_TIMEOUT: | |
1546 | s.index = idx; | |
1547 | s.num = vq->busyloop_timeout; | |
1548 | if (copy_to_user(argp, &s, sizeof(s))) | |
1549 | r = -EFAULT; | |
1550 | break; | |
3a4d5c94 MT |
1551 | default: |
1552 | r = -ENOIOCTLCMD; | |
1553 | } | |
1554 | ||
1555 | if (pollstop && vq->handle_kick) | |
1556 | vhost_poll_stop(&vq->poll); | |
1557 | ||
1558 | if (ctx) | |
1559 | eventfd_ctx_put(ctx); | |
1560 | if (filep) | |
1561 | fput(filep); | |
1562 | ||
1563 | if (pollstart && vq->handle_kick) | |
2b8b328b | 1564 | r = vhost_poll_start(&vq->poll, vq->kick); |
3a4d5c94 MT |
1565 | |
1566 | mutex_unlock(&vq->mutex); | |
1567 | ||
1568 | if (pollstop && vq->handle_kick) | |
1569 | vhost_poll_flush(&vq->poll); | |
1570 | return r; | |
1571 | } | |
6ac1afbf | 1572 | EXPORT_SYMBOL_GPL(vhost_vring_ioctl); |
3a4d5c94 | 1573 | |
6b1e6cc7 JW |
1574 | int vhost_init_device_iotlb(struct vhost_dev *d, bool enabled) |
1575 | { | |
1576 | struct vhost_umem *niotlb, *oiotlb; | |
1577 | int i; | |
1578 | ||
1579 | niotlb = vhost_umem_alloc(); | |
1580 | if (!niotlb) | |
1581 | return -ENOMEM; | |
1582 | ||
1583 | oiotlb = d->iotlb; | |
1584 | d->iotlb = niotlb; | |
1585 | ||
1586 | for (i = 0; i < d->nvqs; ++i) { | |
1587 | mutex_lock(&d->vqs[i]->mutex); | |
1588 | d->vqs[i]->iotlb = niotlb; | |
1589 | mutex_unlock(&d->vqs[i]->mutex); | |
1590 | } | |
1591 | ||
1592 | vhost_umem_clean(oiotlb); | |
1593 | ||
1594 | return 0; | |
1595 | } | |
1596 | EXPORT_SYMBOL_GPL(vhost_init_device_iotlb); | |
1597 | ||
3a4d5c94 | 1598 | /* Caller must have device mutex */ |
935cdee7 | 1599 | long vhost_dev_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp) |
3a4d5c94 | 1600 | { |
3a4d5c94 MT |
1601 | struct file *eventfp, *filep = NULL; |
1602 | struct eventfd_ctx *ctx = NULL; | |
1603 | u64 p; | |
1604 | long r; | |
1605 | int i, fd; | |
1606 | ||
1607 | /* If you are not the owner, you can become one */ | |
1608 | if (ioctl == VHOST_SET_OWNER) { | |
1609 | r = vhost_dev_set_owner(d); | |
1610 | goto done; | |
1611 | } | |
1612 | ||
1613 | /* You must be the owner to do anything else */ | |
1614 | r = vhost_dev_check_owner(d); | |
1615 | if (r) | |
1616 | goto done; | |
1617 | ||
1618 | switch (ioctl) { | |
1619 | case VHOST_SET_MEM_TABLE: | |
1620 | r = vhost_set_memory(d, argp); | |
1621 | break; | |
1622 | case VHOST_SET_LOG_BASE: | |
7ad9c9d2 TY |
1623 | if (copy_from_user(&p, argp, sizeof p)) { |
1624 | r = -EFAULT; | |
3a4d5c94 | 1625 | break; |
7ad9c9d2 | 1626 | } |
3a4d5c94 MT |
1627 | if ((u64)(unsigned long)p != p) { |
1628 | r = -EFAULT; | |
1629 | break; | |
1630 | } | |
1631 | for (i = 0; i < d->nvqs; ++i) { | |
1632 | struct vhost_virtqueue *vq; | |
1633 | void __user *base = (void __user *)(unsigned long)p; | |
3ab2e420 | 1634 | vq = d->vqs[i]; |
3a4d5c94 MT |
1635 | mutex_lock(&vq->mutex); |
1636 | /* If ring is inactive, will check when it's enabled. */ | |
ea16c514 | 1637 | if (vq->private_data && !vq_log_access_ok(vq, base)) |
3a4d5c94 MT |
1638 | r = -EFAULT; |
1639 | else | |
1640 | vq->log_base = base; | |
1641 | mutex_unlock(&vq->mutex); | |
1642 | } | |
1643 | break; | |
1644 | case VHOST_SET_LOG_FD: | |
1645 | r = get_user(fd, (int __user *)argp); | |
1646 | if (r < 0) | |
1647 | break; | |
1648 | eventfp = fd == -1 ? NULL : eventfd_fget(fd); | |
1649 | if (IS_ERR(eventfp)) { | |
1650 | r = PTR_ERR(eventfp); | |
1651 | break; | |
1652 | } | |
1653 | if (eventfp != d->log_file) { | |
1654 | filep = d->log_file; | |
7932c0bd | 1655 | d->log_file = eventfp; |
3a4d5c94 MT |
1656 | ctx = d->log_ctx; |
1657 | d->log_ctx = eventfp ? | |
1658 | eventfd_ctx_fileget(eventfp) : NULL; | |
1659 | } else | |
1660 | filep = eventfp; | |
1661 | for (i = 0; i < d->nvqs; ++i) { | |
3ab2e420 AH |
1662 | mutex_lock(&d->vqs[i]->mutex); |
1663 | d->vqs[i]->log_ctx = d->log_ctx; | |
1664 | mutex_unlock(&d->vqs[i]->mutex); | |
3a4d5c94 MT |
1665 | } |
1666 | if (ctx) | |
1667 | eventfd_ctx_put(ctx); | |
1668 | if (filep) | |
1669 | fput(filep); | |
1670 | break; | |
1671 | default: | |
935cdee7 | 1672 | r = -ENOIOCTLCMD; |
3a4d5c94 MT |
1673 | break; |
1674 | } | |
1675 | done: | |
1676 | return r; | |
1677 | } | |
6ac1afbf | 1678 | EXPORT_SYMBOL_GPL(vhost_dev_ioctl); |
3a4d5c94 | 1679 | |
3a4d5c94 MT |
1680 | /* TODO: This is really inefficient. We need something like get_user() |
1681 | * (instruction directly accesses the data, with an exception table entry | |
1682 | * returning -EFAULT). See Documentation/x86/exception-tables.txt. | |
1683 | */ | |
1684 | static int set_bit_to_user(int nr, void __user *addr) | |
1685 | { | |
1686 | unsigned long log = (unsigned long)addr; | |
1687 | struct page *page; | |
1688 | void *base; | |
1689 | int bit = nr + (log % PAGE_SIZE) * 8; | |
1690 | int r; | |
d47effe1 | 1691 | |
3a4d5c94 | 1692 | r = get_user_pages_fast(log, 1, 1, &page); |
d6db3f5c | 1693 | if (r < 0) |
3a4d5c94 | 1694 | return r; |
d6db3f5c | 1695 | BUG_ON(r != 1); |
c6daa7ff | 1696 | base = kmap_atomic(page); |
3a4d5c94 | 1697 | set_bit(bit, base); |
c6daa7ff | 1698 | kunmap_atomic(base); |
3a4d5c94 MT |
1699 | set_page_dirty_lock(page); |
1700 | put_page(page); | |
1701 | return 0; | |
1702 | } | |
1703 | ||
1704 | static int log_write(void __user *log_base, | |
1705 | u64 write_address, u64 write_length) | |
1706 | { | |
28831ee6 | 1707 | u64 write_page = write_address / VHOST_PAGE_SIZE; |
3a4d5c94 | 1708 | int r; |
d47effe1 | 1709 | |
3a4d5c94 MT |
1710 | if (!write_length) |
1711 | return 0; | |
3bf9be40 | 1712 | write_length += write_address % VHOST_PAGE_SIZE; |
3a4d5c94 MT |
1713 | for (;;) { |
1714 | u64 base = (u64)(unsigned long)log_base; | |
28831ee6 MT |
1715 | u64 log = base + write_page / 8; |
1716 | int bit = write_page % 8; | |
3a4d5c94 MT |
1717 | if ((u64)(unsigned long)log != log) |
1718 | return -EFAULT; | |
1719 | r = set_bit_to_user(bit, (void __user *)(unsigned long)log); | |
1720 | if (r < 0) | |
1721 | return r; | |
1722 | if (write_length <= VHOST_PAGE_SIZE) | |
1723 | break; | |
1724 | write_length -= VHOST_PAGE_SIZE; | |
28831ee6 | 1725 | write_page += 1; |
3a4d5c94 MT |
1726 | } |
1727 | return r; | |
1728 | } | |
1729 | ||
1730 | int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log, | |
1731 | unsigned int log_num, u64 len) | |
1732 | { | |
1733 | int i, r; | |
1734 | ||
1735 | /* Make sure data written is seen before log. */ | |
5659338c | 1736 | smp_wmb(); |
3a4d5c94 MT |
1737 | for (i = 0; i < log_num; ++i) { |
1738 | u64 l = min(log[i].len, len); | |
1739 | r = log_write(vq->log_base, log[i].addr, l); | |
1740 | if (r < 0) | |
1741 | return r; | |
1742 | len -= l; | |
5786aee8 MT |
1743 | if (!len) { |
1744 | if (vq->log_ctx) | |
1745 | eventfd_signal(vq->log_ctx, 1); | |
3a4d5c94 | 1746 | return 0; |
5786aee8 | 1747 | } |
3a4d5c94 | 1748 | } |
3a4d5c94 MT |
1749 | /* Length written exceeds what we have stored. This is a bug. */ |
1750 | BUG(); | |
1751 | return 0; | |
1752 | } | |
6ac1afbf | 1753 | EXPORT_SYMBOL_GPL(vhost_log_write); |
3a4d5c94 | 1754 | |
2723feaa JW |
1755 | static int vhost_update_used_flags(struct vhost_virtqueue *vq) |
1756 | { | |
1757 | void __user *used; | |
bfe2bc51 JW |
1758 | if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->used_flags), |
1759 | &vq->used->flags) < 0) | |
2723feaa JW |
1760 | return -EFAULT; |
1761 | if (unlikely(vq->log_used)) { | |
1762 | /* Make sure the flag is seen before log. */ | |
1763 | smp_wmb(); | |
1764 | /* Log used flag write. */ | |
1765 | used = &vq->used->flags; | |
1766 | log_write(vq->log_base, vq->log_addr + | |
1767 | (used - (void __user *)vq->used), | |
1768 | sizeof vq->used->flags); | |
1769 | if (vq->log_ctx) | |
1770 | eventfd_signal(vq->log_ctx, 1); | |
1771 | } | |
1772 | return 0; | |
1773 | } | |
1774 | ||
1775 | static int vhost_update_avail_event(struct vhost_virtqueue *vq, u16 avail_event) | |
1776 | { | |
bfe2bc51 JW |
1777 | if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->avail_idx), |
1778 | vhost_avail_event(vq))) | |
2723feaa JW |
1779 | return -EFAULT; |
1780 | if (unlikely(vq->log_used)) { | |
1781 | void __user *used; | |
1782 | /* Make sure the event is seen before log. */ | |
1783 | smp_wmb(); | |
1784 | /* Log avail event write */ | |
1785 | used = vhost_avail_event(vq); | |
1786 | log_write(vq->log_base, vq->log_addr + | |
1787 | (used - (void __user *)vq->used), | |
1788 | sizeof *vhost_avail_event(vq)); | |
1789 | if (vq->log_ctx) | |
1790 | eventfd_signal(vq->log_ctx, 1); | |
1791 | } | |
1792 | return 0; | |
1793 | } | |
1794 | ||
80f7d030 | 1795 | int vhost_vq_init_access(struct vhost_virtqueue *vq) |
2723feaa | 1796 | { |
3b1bbe89 | 1797 | __virtio16 last_used_idx; |
2723feaa | 1798 | int r; |
e1f33be9 GK |
1799 | bool is_le = vq->is_le; |
1800 | ||
cda8bba0 | 1801 | if (!vq->private_data) |
2723feaa | 1802 | return 0; |
2751c988 GK |
1803 | |
1804 | vhost_init_is_le(vq); | |
2723feaa JW |
1805 | |
1806 | r = vhost_update_used_flags(vq); | |
1807 | if (r) | |
e1f33be9 | 1808 | goto err; |
2723feaa | 1809 | vq->signalled_used_valid = false; |
6b1e6cc7 JW |
1810 | if (!vq->iotlb && |
1811 | !access_ok(VERIFY_READ, &vq->used->idx, sizeof vq->used->idx)) { | |
e1f33be9 GK |
1812 | r = -EFAULT; |
1813 | goto err; | |
1814 | } | |
f8894913 | 1815 | r = vhost_get_used(vq, last_used_idx, &vq->used->idx); |
6b1e6cc7 JW |
1816 | if (r) { |
1817 | vq_err(vq, "Can't access used idx at %p\n", | |
1818 | &vq->used->idx); | |
e1f33be9 | 1819 | goto err; |
6b1e6cc7 | 1820 | } |
3b1bbe89 | 1821 | vq->last_used_idx = vhost16_to_cpu(vq, last_used_idx); |
64f7f051 | 1822 | return 0; |
6b1e6cc7 | 1823 | |
e1f33be9 GK |
1824 | err: |
1825 | vq->is_le = is_le; | |
1826 | return r; | |
2723feaa | 1827 | } |
80f7d030 | 1828 | EXPORT_SYMBOL_GPL(vhost_vq_init_access); |
2723feaa | 1829 | |
47283bef | 1830 | static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len, |
6b1e6cc7 | 1831 | struct iovec iov[], int iov_size, int access) |
3a4d5c94 | 1832 | { |
a9709d68 | 1833 | const struct vhost_umem_node *node; |
6b1e6cc7 JW |
1834 | struct vhost_dev *dev = vq->dev; |
1835 | struct vhost_umem *umem = dev->iotlb ? dev->iotlb : dev->umem; | |
3a4d5c94 MT |
1836 | struct iovec *_iov; |
1837 | u64 s = 0; | |
1838 | int ret = 0; | |
1839 | ||
3a4d5c94 MT |
1840 | while ((u64)len > s) { |
1841 | u64 size; | |
7b3384fc | 1842 | if (unlikely(ret >= iov_size)) { |
3a4d5c94 MT |
1843 | ret = -ENOBUFS; |
1844 | break; | |
1845 | } | |
6b1e6cc7 | 1846 | |
a9709d68 JW |
1847 | node = vhost_umem_interval_tree_iter_first(&umem->umem_tree, |
1848 | addr, addr + len - 1); | |
1849 | if (node == NULL || node->start > addr) { | |
6b1e6cc7 JW |
1850 | if (umem != dev->iotlb) { |
1851 | ret = -EFAULT; | |
1852 | break; | |
1853 | } | |
1854 | ret = -EAGAIN; | |
1855 | break; | |
1856 | } else if (!(node->perm & access)) { | |
1857 | ret = -EPERM; | |
3a4d5c94 MT |
1858 | break; |
1859 | } | |
6b1e6cc7 | 1860 | |
3a4d5c94 | 1861 | _iov = iov + ret; |
a9709d68 | 1862 | size = node->size - addr + node->start; |
bd97120f | 1863 | _iov->iov_len = min((u64)len - s, size); |
a8d3782f | 1864 | _iov->iov_base = (void __user *)(unsigned long) |
a9709d68 | 1865 | (node->userspace_addr + addr - node->start); |
3a4d5c94 MT |
1866 | s += size; |
1867 | addr += size; | |
1868 | ++ret; | |
1869 | } | |
1870 | ||
6b1e6cc7 JW |
1871 | if (ret == -EAGAIN) |
1872 | vhost_iotlb_miss(vq, addr, access); | |
3a4d5c94 MT |
1873 | return ret; |
1874 | } | |
1875 | ||
1876 | /* Each buffer in the virtqueues is actually a chain of descriptors. This | |
1877 | * function returns the next descriptor in the chain, | |
1878 | * or -1U if we're at the end. */ | |
3b1bbe89 | 1879 | static unsigned next_desc(struct vhost_virtqueue *vq, struct vring_desc *desc) |
3a4d5c94 MT |
1880 | { |
1881 | unsigned int next; | |
1882 | ||
1883 | /* If this descriptor says it doesn't chain, we're done. */ | |
3b1bbe89 | 1884 | if (!(desc->flags & cpu_to_vhost16(vq, VRING_DESC_F_NEXT))) |
3a4d5c94 MT |
1885 | return -1U; |
1886 | ||
1887 | /* Check they're not leading us off end of descriptors. */ | |
3b1bbe89 | 1888 | next = vhost16_to_cpu(vq, desc->next); |
3a4d5c94 MT |
1889 | /* Make sure compiler knows to grab that: we don't want it changing! */ |
1890 | /* We will use the result as an index in an array, so most | |
1891 | * architectures only need a compiler barrier here. */ | |
1892 | read_barrier_depends(); | |
1893 | ||
1894 | return next; | |
1895 | } | |
1896 | ||
47283bef | 1897 | static int get_indirect(struct vhost_virtqueue *vq, |
7b3384fc MT |
1898 | struct iovec iov[], unsigned int iov_size, |
1899 | unsigned int *out_num, unsigned int *in_num, | |
1900 | struct vhost_log *log, unsigned int *log_num, | |
1901 | struct vring_desc *indirect) | |
3a4d5c94 MT |
1902 | { |
1903 | struct vring_desc desc; | |
1904 | unsigned int i = 0, count, found = 0; | |
3b1bbe89 | 1905 | u32 len = vhost32_to_cpu(vq, indirect->len); |
aad9a1ce | 1906 | struct iov_iter from; |
6b1e6cc7 | 1907 | int ret, access; |
3a4d5c94 MT |
1908 | |
1909 | /* Sanity check */ | |
3b1bbe89 | 1910 | if (unlikely(len % sizeof desc)) { |
3a4d5c94 MT |
1911 | vq_err(vq, "Invalid length in indirect descriptor: " |
1912 | "len 0x%llx not multiple of 0x%zx\n", | |
3b1bbe89 | 1913 | (unsigned long long)len, |
3a4d5c94 MT |
1914 | sizeof desc); |
1915 | return -EINVAL; | |
1916 | } | |
1917 | ||
3b1bbe89 | 1918 | ret = translate_desc(vq, vhost64_to_cpu(vq, indirect->addr), len, vq->indirect, |
6b1e6cc7 | 1919 | UIO_MAXIOV, VHOST_ACCESS_RO); |
7b3384fc | 1920 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
1921 | if (ret != -EAGAIN) |
1922 | vq_err(vq, "Translation failure %d in indirect.\n", ret); | |
3a4d5c94 MT |
1923 | return ret; |
1924 | } | |
aad9a1ce | 1925 | iov_iter_init(&from, READ, vq->indirect, ret, len); |
3a4d5c94 MT |
1926 | |
1927 | /* We will use the result as an address to read from, so most | |
1928 | * architectures only need a compiler barrier here. */ | |
1929 | read_barrier_depends(); | |
1930 | ||
3b1bbe89 | 1931 | count = len / sizeof desc; |
3a4d5c94 MT |
1932 | /* Buffers are chained via a 16 bit next field, so |
1933 | * we can have at most 2^16 of these. */ | |
7b3384fc | 1934 | if (unlikely(count > USHRT_MAX + 1)) { |
3a4d5c94 MT |
1935 | vq_err(vq, "Indirect buffer length too big: %d\n", |
1936 | indirect->len); | |
1937 | return -E2BIG; | |
1938 | } | |
1939 | ||
1940 | do { | |
1941 | unsigned iov_count = *in_num + *out_num; | |
7b3384fc | 1942 | if (unlikely(++found > count)) { |
3a4d5c94 MT |
1943 | vq_err(vq, "Loop detected: last one at %u " |
1944 | "indirect size %u\n", | |
1945 | i, count); | |
1946 | return -EINVAL; | |
1947 | } | |
cbbd26b8 | 1948 | if (unlikely(!copy_from_iter_full(&desc, sizeof(desc), &from))) { |
3a4d5c94 | 1949 | vq_err(vq, "Failed indirect descriptor: idx %d, %zx\n", |
3b1bbe89 | 1950 | i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc); |
3a4d5c94 MT |
1951 | return -EINVAL; |
1952 | } | |
3b1bbe89 | 1953 | if (unlikely(desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT))) { |
3a4d5c94 | 1954 | vq_err(vq, "Nested indirect descriptor: idx %d, %zx\n", |
3b1bbe89 | 1955 | i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc); |
3a4d5c94 MT |
1956 | return -EINVAL; |
1957 | } | |
1958 | ||
6b1e6cc7 JW |
1959 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE)) |
1960 | access = VHOST_ACCESS_WO; | |
1961 | else | |
1962 | access = VHOST_ACCESS_RO; | |
1963 | ||
3b1bbe89 MT |
1964 | ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr), |
1965 | vhost32_to_cpu(vq, desc.len), iov + iov_count, | |
6b1e6cc7 | 1966 | iov_size - iov_count, access); |
7b3384fc | 1967 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
1968 | if (ret != -EAGAIN) |
1969 | vq_err(vq, "Translation failure %d indirect idx %d\n", | |
1970 | ret, i); | |
3a4d5c94 MT |
1971 | return ret; |
1972 | } | |
1973 | /* If this is an input descriptor, increment that count. */ | |
6b1e6cc7 | 1974 | if (access == VHOST_ACCESS_WO) { |
3a4d5c94 MT |
1975 | *in_num += ret; |
1976 | if (unlikely(log)) { | |
3b1bbe89 MT |
1977 | log[*log_num].addr = vhost64_to_cpu(vq, desc.addr); |
1978 | log[*log_num].len = vhost32_to_cpu(vq, desc.len); | |
3a4d5c94 MT |
1979 | ++*log_num; |
1980 | } | |
1981 | } else { | |
1982 | /* If it's an output descriptor, they're all supposed | |
1983 | * to come before any input descriptors. */ | |
7b3384fc | 1984 | if (unlikely(*in_num)) { |
3a4d5c94 MT |
1985 | vq_err(vq, "Indirect descriptor " |
1986 | "has out after in: idx %d\n", i); | |
1987 | return -EINVAL; | |
1988 | } | |
1989 | *out_num += ret; | |
1990 | } | |
3b1bbe89 | 1991 | } while ((i = next_desc(vq, &desc)) != -1); |
3a4d5c94 MT |
1992 | return 0; |
1993 | } | |
1994 | ||
1995 | /* This looks in the virtqueue and for the first available buffer, and converts | |
1996 | * it to an iovec for convenient access. Since descriptors consist of some | |
1997 | * number of output then some number of input descriptors, it's actually two | |
1998 | * iovecs, but we pack them into one and note how many of each there were. | |
1999 | * | |
d5675bd2 MT |
2000 | * This function returns the descriptor number found, or vq->num (which is |
2001 | * never a valid descriptor number) if none was found. A negative code is | |
2002 | * returned on error. */ | |
47283bef | 2003 | int vhost_get_vq_desc(struct vhost_virtqueue *vq, |
d5675bd2 MT |
2004 | struct iovec iov[], unsigned int iov_size, |
2005 | unsigned int *out_num, unsigned int *in_num, | |
2006 | struct vhost_log *log, unsigned int *log_num) | |
3a4d5c94 MT |
2007 | { |
2008 | struct vring_desc desc; | |
2009 | unsigned int i, head, found = 0; | |
2010 | u16 last_avail_idx; | |
3b1bbe89 MT |
2011 | __virtio16 avail_idx; |
2012 | __virtio16 ring_head; | |
6b1e6cc7 | 2013 | int ret, access; |
3a4d5c94 MT |
2014 | |
2015 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
2016 | last_avail_idx = vq->last_avail_idx; | |
3a4d5c94 | 2017 | |
e3b56cdd | 2018 | if (vq->avail_idx == vq->last_avail_idx) { |
f8894913 | 2019 | if (unlikely(vhost_get_avail(vq, avail_idx, &vq->avail->idx))) { |
e3b56cdd JW |
2020 | vq_err(vq, "Failed to access avail idx at %p\n", |
2021 | &vq->avail->idx); | |
2022 | return -EFAULT; | |
2023 | } | |
2024 | vq->avail_idx = vhost16_to_cpu(vq, avail_idx); | |
3a4d5c94 | 2025 | |
e3b56cdd JW |
2026 | if (unlikely((u16)(vq->avail_idx - last_avail_idx) > vq->num)) { |
2027 | vq_err(vq, "Guest moved used index from %u to %u", | |
2028 | last_avail_idx, vq->avail_idx); | |
2029 | return -EFAULT; | |
2030 | } | |
2031 | ||
2032 | /* If there's nothing new since last we looked, return | |
2033 | * invalid. | |
2034 | */ | |
2035 | if (vq->avail_idx == last_avail_idx) | |
2036 | return vq->num; | |
3a4d5c94 | 2037 | |
e3b56cdd JW |
2038 | /* Only get avail ring entries after they have been |
2039 | * exposed by guest. | |
2040 | */ | |
2041 | smp_rmb(); | |
2042 | } | |
3a4d5c94 MT |
2043 | |
2044 | /* Grab the next descriptor number they're advertising, and increment | |
2045 | * the index we've seen. */ | |
f8894913 | 2046 | if (unlikely(vhost_get_avail(vq, ring_head, |
bfe2bc51 | 2047 | &vq->avail->ring[last_avail_idx & (vq->num - 1)]))) { |
3a4d5c94 MT |
2048 | vq_err(vq, "Failed to read head: idx %d address %p\n", |
2049 | last_avail_idx, | |
2050 | &vq->avail->ring[last_avail_idx % vq->num]); | |
d5675bd2 | 2051 | return -EFAULT; |
3a4d5c94 MT |
2052 | } |
2053 | ||
3b1bbe89 MT |
2054 | head = vhost16_to_cpu(vq, ring_head); |
2055 | ||
3a4d5c94 | 2056 | /* If their number is silly, that's an error. */ |
7b3384fc | 2057 | if (unlikely(head >= vq->num)) { |
3a4d5c94 MT |
2058 | vq_err(vq, "Guest says index %u > %u is available", |
2059 | head, vq->num); | |
d5675bd2 | 2060 | return -EINVAL; |
3a4d5c94 MT |
2061 | } |
2062 | ||
2063 | /* When we start there are none of either input nor output. */ | |
2064 | *out_num = *in_num = 0; | |
2065 | if (unlikely(log)) | |
2066 | *log_num = 0; | |
2067 | ||
2068 | i = head; | |
2069 | do { | |
2070 | unsigned iov_count = *in_num + *out_num; | |
7b3384fc | 2071 | if (unlikely(i >= vq->num)) { |
3a4d5c94 MT |
2072 | vq_err(vq, "Desc index is %u > %u, head = %u", |
2073 | i, vq->num, head); | |
d5675bd2 | 2074 | return -EINVAL; |
3a4d5c94 | 2075 | } |
7b3384fc | 2076 | if (unlikely(++found > vq->num)) { |
3a4d5c94 MT |
2077 | vq_err(vq, "Loop detected: last one at %u " |
2078 | "vq size %u head %u\n", | |
2079 | i, vq->num, head); | |
d5675bd2 | 2080 | return -EINVAL; |
3a4d5c94 | 2081 | } |
bfe2bc51 JW |
2082 | ret = vhost_copy_from_user(vq, &desc, vq->desc + i, |
2083 | sizeof desc); | |
7b3384fc | 2084 | if (unlikely(ret)) { |
3a4d5c94 MT |
2085 | vq_err(vq, "Failed to get descriptor: idx %d addr %p\n", |
2086 | i, vq->desc + i); | |
d5675bd2 | 2087 | return -EFAULT; |
3a4d5c94 | 2088 | } |
3b1bbe89 | 2089 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT)) { |
47283bef | 2090 | ret = get_indirect(vq, iov, iov_size, |
3a4d5c94 MT |
2091 | out_num, in_num, |
2092 | log, log_num, &desc); | |
7b3384fc | 2093 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
2094 | if (ret != -EAGAIN) |
2095 | vq_err(vq, "Failure detected " | |
2096 | "in indirect descriptor at idx %d\n", i); | |
d5675bd2 | 2097 | return ret; |
3a4d5c94 MT |
2098 | } |
2099 | continue; | |
2100 | } | |
2101 | ||
6b1e6cc7 JW |
2102 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE)) |
2103 | access = VHOST_ACCESS_WO; | |
2104 | else | |
2105 | access = VHOST_ACCESS_RO; | |
3b1bbe89 MT |
2106 | ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr), |
2107 | vhost32_to_cpu(vq, desc.len), iov + iov_count, | |
6b1e6cc7 | 2108 | iov_size - iov_count, access); |
7b3384fc | 2109 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
2110 | if (ret != -EAGAIN) |
2111 | vq_err(vq, "Translation failure %d descriptor idx %d\n", | |
2112 | ret, i); | |
d5675bd2 | 2113 | return ret; |
3a4d5c94 | 2114 | } |
6b1e6cc7 | 2115 | if (access == VHOST_ACCESS_WO) { |
3a4d5c94 MT |
2116 | /* If this is an input descriptor, |
2117 | * increment that count. */ | |
2118 | *in_num += ret; | |
2119 | if (unlikely(log)) { | |
3b1bbe89 MT |
2120 | log[*log_num].addr = vhost64_to_cpu(vq, desc.addr); |
2121 | log[*log_num].len = vhost32_to_cpu(vq, desc.len); | |
3a4d5c94 MT |
2122 | ++*log_num; |
2123 | } | |
2124 | } else { | |
2125 | /* If it's an output descriptor, they're all supposed | |
2126 | * to come before any input descriptors. */ | |
7b3384fc | 2127 | if (unlikely(*in_num)) { |
3a4d5c94 MT |
2128 | vq_err(vq, "Descriptor has out after in: " |
2129 | "idx %d\n", i); | |
d5675bd2 | 2130 | return -EINVAL; |
3a4d5c94 MT |
2131 | } |
2132 | *out_num += ret; | |
2133 | } | |
3b1bbe89 | 2134 | } while ((i = next_desc(vq, &desc)) != -1); |
3a4d5c94 MT |
2135 | |
2136 | /* On success, increment avail index. */ | |
2137 | vq->last_avail_idx++; | |
8ea8cf89 MT |
2138 | |
2139 | /* Assume notifications from guest are disabled at this point, | |
2140 | * if they aren't we would need to update avail_event index. */ | |
2141 | BUG_ON(!(vq->used_flags & VRING_USED_F_NO_NOTIFY)); | |
3a4d5c94 MT |
2142 | return head; |
2143 | } | |
6ac1afbf | 2144 | EXPORT_SYMBOL_GPL(vhost_get_vq_desc); |
3a4d5c94 MT |
2145 | |
2146 | /* Reverse the effect of vhost_get_vq_desc. Useful for error handling. */ | |
8dd014ad | 2147 | void vhost_discard_vq_desc(struct vhost_virtqueue *vq, int n) |
3a4d5c94 | 2148 | { |
8dd014ad | 2149 | vq->last_avail_idx -= n; |
3a4d5c94 | 2150 | } |
6ac1afbf | 2151 | EXPORT_SYMBOL_GPL(vhost_discard_vq_desc); |
3a4d5c94 MT |
2152 | |
2153 | /* After we've used one of their buffers, we tell them about it. We'll then | |
2154 | * want to notify the guest, using eventfd. */ | |
2155 | int vhost_add_used(struct vhost_virtqueue *vq, unsigned int head, int len) | |
2156 | { | |
3b1bbe89 MT |
2157 | struct vring_used_elem heads = { |
2158 | cpu_to_vhost32(vq, head), | |
2159 | cpu_to_vhost32(vq, len) | |
2160 | }; | |
3a4d5c94 | 2161 | |
c49e4e57 | 2162 | return vhost_add_used_n(vq, &heads, 1); |
3a4d5c94 | 2163 | } |
6ac1afbf | 2164 | EXPORT_SYMBOL_GPL(vhost_add_used); |
3a4d5c94 | 2165 | |
8dd014ad DS |
2166 | static int __vhost_add_used_n(struct vhost_virtqueue *vq, |
2167 | struct vring_used_elem *heads, | |
2168 | unsigned count) | |
2169 | { | |
2170 | struct vring_used_elem __user *used; | |
8ea8cf89 | 2171 | u16 old, new; |
8dd014ad DS |
2172 | int start; |
2173 | ||
5fba13b5 | 2174 | start = vq->last_used_idx & (vq->num - 1); |
8dd014ad | 2175 | used = vq->used->ring + start; |
c49e4e57 | 2176 | if (count == 1) { |
bfe2bc51 | 2177 | if (vhost_put_user(vq, heads[0].id, &used->id)) { |
c49e4e57 JW |
2178 | vq_err(vq, "Failed to write used id"); |
2179 | return -EFAULT; | |
2180 | } | |
bfe2bc51 | 2181 | if (vhost_put_user(vq, heads[0].len, &used->len)) { |
c49e4e57 JW |
2182 | vq_err(vq, "Failed to write used len"); |
2183 | return -EFAULT; | |
2184 | } | |
bfe2bc51 | 2185 | } else if (vhost_copy_to_user(vq, used, heads, count * sizeof *used)) { |
8dd014ad DS |
2186 | vq_err(vq, "Failed to write used"); |
2187 | return -EFAULT; | |
2188 | } | |
2189 | if (unlikely(vq->log_used)) { | |
2190 | /* Make sure data is seen before log. */ | |
2191 | smp_wmb(); | |
2192 | /* Log used ring entry write. */ | |
2193 | log_write(vq->log_base, | |
2194 | vq->log_addr + | |
2195 | ((void __user *)used - (void __user *)vq->used), | |
2196 | count * sizeof *used); | |
2197 | } | |
8ea8cf89 MT |
2198 | old = vq->last_used_idx; |
2199 | new = (vq->last_used_idx += count); | |
2200 | /* If the driver never bothers to signal in a very long while, | |
2201 | * used index might wrap around. If that happens, invalidate | |
2202 | * signalled_used index we stored. TODO: make sure driver | |
2203 | * signals at least once in 2^16 and remove this. */ | |
2204 | if (unlikely((u16)(new - vq->signalled_used) < (u16)(new - old))) | |
2205 | vq->signalled_used_valid = false; | |
8dd014ad DS |
2206 | return 0; |
2207 | } | |
2208 | ||
2209 | /* After we've used one of their buffers, we tell them about it. We'll then | |
2210 | * want to notify the guest, using eventfd. */ | |
2211 | int vhost_add_used_n(struct vhost_virtqueue *vq, struct vring_used_elem *heads, | |
2212 | unsigned count) | |
2213 | { | |
2214 | int start, n, r; | |
2215 | ||
5fba13b5 | 2216 | start = vq->last_used_idx & (vq->num - 1); |
8dd014ad DS |
2217 | n = vq->num - start; |
2218 | if (n < count) { | |
2219 | r = __vhost_add_used_n(vq, heads, n); | |
2220 | if (r < 0) | |
2221 | return r; | |
2222 | heads += n; | |
2223 | count -= n; | |
2224 | } | |
2225 | r = __vhost_add_used_n(vq, heads, count); | |
2226 | ||
2227 | /* Make sure buffer is written before we update index. */ | |
2228 | smp_wmb(); | |
bfe2bc51 JW |
2229 | if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->last_used_idx), |
2230 | &vq->used->idx)) { | |
8dd014ad DS |
2231 | vq_err(vq, "Failed to increment used idx"); |
2232 | return -EFAULT; | |
2233 | } | |
2234 | if (unlikely(vq->log_used)) { | |
2235 | /* Log used index update. */ | |
2236 | log_write(vq->log_base, | |
2237 | vq->log_addr + offsetof(struct vring_used, idx), | |
2238 | sizeof vq->used->idx); | |
2239 | if (vq->log_ctx) | |
2240 | eventfd_signal(vq->log_ctx, 1); | |
2241 | } | |
2242 | return r; | |
2243 | } | |
6ac1afbf | 2244 | EXPORT_SYMBOL_GPL(vhost_add_used_n); |
8dd014ad | 2245 | |
8ea8cf89 | 2246 | static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 | 2247 | { |
3b1bbe89 MT |
2248 | __u16 old, new; |
2249 | __virtio16 event; | |
8ea8cf89 | 2250 | bool v; |
0d499356 | 2251 | |
ea16c514 | 2252 | if (vhost_has_feature(vq, VIRTIO_F_NOTIFY_ON_EMPTY) && |
8ea8cf89 MT |
2253 | unlikely(vq->avail_idx == vq->last_avail_idx)) |
2254 | return true; | |
2255 | ||
ea16c514 | 2256 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
3b1bbe89 | 2257 | __virtio16 flags; |
809ecb9b JW |
2258 | /* Flush out used index updates. This is paired |
2259 | * with the barrier that the Guest executes when enabling | |
2260 | * interrupts. */ | |
2261 | smp_mb(); | |
f8894913 | 2262 | if (vhost_get_avail(vq, flags, &vq->avail->flags)) { |
8ea8cf89 MT |
2263 | vq_err(vq, "Failed to get flags"); |
2264 | return true; | |
2265 | } | |
3b1bbe89 | 2266 | return !(flags & cpu_to_vhost16(vq, VRING_AVAIL_F_NO_INTERRUPT)); |
3a4d5c94 | 2267 | } |
8ea8cf89 MT |
2268 | old = vq->signalled_used; |
2269 | v = vq->signalled_used_valid; | |
2270 | new = vq->signalled_used = vq->last_used_idx; | |
2271 | vq->signalled_used_valid = true; | |
3a4d5c94 | 2272 | |
8ea8cf89 MT |
2273 | if (unlikely(!v)) |
2274 | return true; | |
3a4d5c94 | 2275 | |
809ecb9b JW |
2276 | /* We're sure if the following conditions are met, there's no |
2277 | * need to notify guest: | |
2278 | * 1) cached used event is ahead of new | |
2279 | * 2) old to new updating does not cross cached used event. */ | |
2280 | if (vring_need_event(vq->last_used_event, new + vq->num, new) && | |
2281 | !vring_need_event(vq->last_used_event, new, old)) | |
2282 | return false; | |
2283 | ||
2284 | /* Flush out used index updates. This is paired | |
2285 | * with the barrier that the Guest executes when enabling | |
2286 | * interrupts. */ | |
2287 | smp_mb(); | |
2288 | ||
f8894913 | 2289 | if (vhost_get_avail(vq, event, vhost_used_event(vq))) { |
8ea8cf89 MT |
2290 | vq_err(vq, "Failed to get used event idx"); |
2291 | return true; | |
2292 | } | |
809ecb9b JW |
2293 | vq->last_used_event = vhost16_to_cpu(vq, event); |
2294 | ||
2295 | return vring_need_event(vq->last_used_event, new, old); | |
8ea8cf89 MT |
2296 | } |
2297 | ||
2298 | /* This actually signals the guest, using eventfd. */ | |
2299 | void vhost_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq) | |
2300 | { | |
3a4d5c94 | 2301 | /* Signal the Guest tell them we used something up. */ |
8ea8cf89 | 2302 | if (vq->call_ctx && vhost_notify(dev, vq)) |
3a4d5c94 MT |
2303 | eventfd_signal(vq->call_ctx, 1); |
2304 | } | |
6ac1afbf | 2305 | EXPORT_SYMBOL_GPL(vhost_signal); |
3a4d5c94 MT |
2306 | |
2307 | /* And here's the combo meal deal. Supersize me! */ | |
2308 | void vhost_add_used_and_signal(struct vhost_dev *dev, | |
2309 | struct vhost_virtqueue *vq, | |
2310 | unsigned int head, int len) | |
2311 | { | |
2312 | vhost_add_used(vq, head, len); | |
2313 | vhost_signal(dev, vq); | |
2314 | } | |
6ac1afbf | 2315 | EXPORT_SYMBOL_GPL(vhost_add_used_and_signal); |
3a4d5c94 | 2316 | |
8dd014ad DS |
2317 | /* multi-buffer version of vhost_add_used_and_signal */ |
2318 | void vhost_add_used_and_signal_n(struct vhost_dev *dev, | |
2319 | struct vhost_virtqueue *vq, | |
2320 | struct vring_used_elem *heads, unsigned count) | |
2321 | { | |
2322 | vhost_add_used_n(vq, heads, count); | |
2323 | vhost_signal(dev, vq); | |
2324 | } | |
6ac1afbf | 2325 | EXPORT_SYMBOL_GPL(vhost_add_used_and_signal_n); |
8dd014ad | 2326 | |
d4a60603 JW |
2327 | /* return true if we're sure that avaiable ring is empty */ |
2328 | bool vhost_vq_avail_empty(struct vhost_dev *dev, struct vhost_virtqueue *vq) | |
2329 | { | |
2330 | __virtio16 avail_idx; | |
2331 | int r; | |
2332 | ||
f8894913 | 2333 | r = vhost_get_avail(vq, avail_idx, &vq->avail->idx); |
d4a60603 JW |
2334 | if (r) |
2335 | return false; | |
2336 | ||
2337 | return vhost16_to_cpu(vq, avail_idx) == vq->avail_idx; | |
2338 | } | |
2339 | EXPORT_SYMBOL_GPL(vhost_vq_avail_empty); | |
2340 | ||
3a4d5c94 | 2341 | /* OK, now we need to know about added descriptors. */ |
8ea8cf89 | 2342 | bool vhost_enable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 | 2343 | { |
3b1bbe89 | 2344 | __virtio16 avail_idx; |
3a4d5c94 | 2345 | int r; |
d47effe1 | 2346 | |
3a4d5c94 MT |
2347 | if (!(vq->used_flags & VRING_USED_F_NO_NOTIFY)) |
2348 | return false; | |
2349 | vq->used_flags &= ~VRING_USED_F_NO_NOTIFY; | |
ea16c514 | 2350 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
2723feaa | 2351 | r = vhost_update_used_flags(vq); |
8ea8cf89 MT |
2352 | if (r) { |
2353 | vq_err(vq, "Failed to enable notification at %p: %d\n", | |
2354 | &vq->used->flags, r); | |
2355 | return false; | |
2356 | } | |
2357 | } else { | |
2723feaa | 2358 | r = vhost_update_avail_event(vq, vq->avail_idx); |
8ea8cf89 MT |
2359 | if (r) { |
2360 | vq_err(vq, "Failed to update avail event index at %p: %d\n", | |
2361 | vhost_avail_event(vq), r); | |
2362 | return false; | |
2363 | } | |
2364 | } | |
3a4d5c94 MT |
2365 | /* They could have slipped one in as we were doing that: make |
2366 | * sure it's written, then check again. */ | |
5659338c | 2367 | smp_mb(); |
f8894913 | 2368 | r = vhost_get_avail(vq, avail_idx, &vq->avail->idx); |
3a4d5c94 MT |
2369 | if (r) { |
2370 | vq_err(vq, "Failed to check avail idx at %p: %d\n", | |
2371 | &vq->avail->idx, r); | |
2372 | return false; | |
2373 | } | |
2374 | ||
3b1bbe89 | 2375 | return vhost16_to_cpu(vq, avail_idx) != vq->avail_idx; |
3a4d5c94 | 2376 | } |
6ac1afbf | 2377 | EXPORT_SYMBOL_GPL(vhost_enable_notify); |
3a4d5c94 MT |
2378 | |
2379 | /* We don't need to be notified again. */ | |
8ea8cf89 | 2380 | void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 MT |
2381 | { |
2382 | int r; | |
d47effe1 | 2383 | |
3a4d5c94 MT |
2384 | if (vq->used_flags & VRING_USED_F_NO_NOTIFY) |
2385 | return; | |
2386 | vq->used_flags |= VRING_USED_F_NO_NOTIFY; | |
ea16c514 | 2387 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
2723feaa | 2388 | r = vhost_update_used_flags(vq); |
8ea8cf89 MT |
2389 | if (r) |
2390 | vq_err(vq, "Failed to enable notification at %p: %d\n", | |
2391 | &vq->used->flags, r); | |
2392 | } | |
3a4d5c94 | 2393 | } |
6ac1afbf AH |
2394 | EXPORT_SYMBOL_GPL(vhost_disable_notify); |
2395 | ||
6b1e6cc7 JW |
2396 | /* Create a new message. */ |
2397 | struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type) | |
2398 | { | |
2399 | struct vhost_msg_node *node = kmalloc(sizeof *node, GFP_KERNEL); | |
2400 | if (!node) | |
2401 | return NULL; | |
2402 | node->vq = vq; | |
2403 | node->msg.type = type; | |
2404 | return node; | |
2405 | } | |
2406 | EXPORT_SYMBOL_GPL(vhost_new_msg); | |
2407 | ||
2408 | void vhost_enqueue_msg(struct vhost_dev *dev, struct list_head *head, | |
2409 | struct vhost_msg_node *node) | |
2410 | { | |
2411 | spin_lock(&dev->iotlb_lock); | |
2412 | list_add_tail(&node->node, head); | |
2413 | spin_unlock(&dev->iotlb_lock); | |
2414 | ||
2415 | wake_up_interruptible_poll(&dev->wait, POLLIN | POLLRDNORM); | |
2416 | } | |
2417 | EXPORT_SYMBOL_GPL(vhost_enqueue_msg); | |
2418 | ||
2419 | struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev, | |
2420 | struct list_head *head) | |
2421 | { | |
2422 | struct vhost_msg_node *node = NULL; | |
2423 | ||
2424 | spin_lock(&dev->iotlb_lock); | |
2425 | if (!list_empty(head)) { | |
2426 | node = list_first_entry(head, struct vhost_msg_node, | |
2427 | node); | |
2428 | list_del(&node->node); | |
2429 | } | |
2430 | spin_unlock(&dev->iotlb_lock); | |
2431 | ||
2432 | return node; | |
2433 | } | |
2434 | EXPORT_SYMBOL_GPL(vhost_dequeue_msg); | |
2435 | ||
2436 | ||
6ac1afbf AH |
2437 | static int __init vhost_init(void) |
2438 | { | |
2439 | return 0; | |
2440 | } | |
2441 | ||
2442 | static void __exit vhost_exit(void) | |
2443 | { | |
2444 | } | |
2445 | ||
2446 | module_init(vhost_init); | |
2447 | module_exit(vhost_exit); | |
2448 | ||
2449 | MODULE_VERSION("0.0.1"); | |
2450 | MODULE_LICENSE("GPL v2"); | |
2451 | MODULE_AUTHOR("Michael S. Tsirkin"); | |
2452 | MODULE_DESCRIPTION("Host kernel accelerator for virtio"); |