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> |
6e84f315 | 30 | #include <linux/sched/mm.h> |
174cd4b1 | 31 | #include <linux/sched/signal.h> |
a9709d68 | 32 | #include <linux/interval_tree_generic.h> |
3a4d5c94 | 33 | |
3a4d5c94 MT |
34 | #include "vhost.h" |
35 | ||
c9ce42f7 IM |
36 | static ushort max_mem_regions = 64; |
37 | module_param(max_mem_regions, ushort, 0444); | |
38 | MODULE_PARM_DESC(max_mem_regions, | |
39 | "Maximum number of memory regions in memory map. (default: 64)"); | |
6b1e6cc7 JW |
40 | static int max_iotlb_entries = 2048; |
41 | module_param(max_iotlb_entries, int, 0444); | |
42 | MODULE_PARM_DESC(max_iotlb_entries, | |
43 | "Maximum number of iotlb entries. (default: 2048)"); | |
c9ce42f7 | 44 | |
3a4d5c94 | 45 | enum { |
3a4d5c94 MT |
46 | VHOST_MEMORY_F_LOG = 0x1, |
47 | }; | |
48 | ||
3b1bbe89 MT |
49 | #define vhost_used_event(vq) ((__virtio16 __user *)&vq->avail->ring[vq->num]) |
50 | #define vhost_avail_event(vq) ((__virtio16 __user *)&vq->used->ring[vq->num]) | |
8ea8cf89 | 51 | |
a9709d68 JW |
52 | INTERVAL_TREE_DEFINE(struct vhost_umem_node, |
53 | rb, __u64, __subtree_last, | |
2f952c01 | 54 | START, LAST, static inline, vhost_umem_interval_tree); |
a9709d68 | 55 | |
2751c988 | 56 | #ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY |
c5072037 | 57 | static void vhost_disable_cross_endian(struct vhost_virtqueue *vq) |
2751c988 GK |
58 | { |
59 | vq->user_be = !virtio_legacy_is_little_endian(); | |
60 | } | |
61 | ||
c5072037 GK |
62 | static void vhost_enable_cross_endian_big(struct vhost_virtqueue *vq) |
63 | { | |
64 | vq->user_be = true; | |
65 | } | |
66 | ||
67 | static void vhost_enable_cross_endian_little(struct vhost_virtqueue *vq) | |
68 | { | |
69 | vq->user_be = false; | |
70 | } | |
71 | ||
2751c988 GK |
72 | static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp) |
73 | { | |
74 | struct vhost_vring_state s; | |
75 | ||
76 | if (vq->private_data) | |
77 | return -EBUSY; | |
78 | ||
79 | if (copy_from_user(&s, argp, sizeof(s))) | |
80 | return -EFAULT; | |
81 | ||
82 | if (s.num != VHOST_VRING_LITTLE_ENDIAN && | |
83 | s.num != VHOST_VRING_BIG_ENDIAN) | |
84 | return -EINVAL; | |
85 | ||
c5072037 GK |
86 | if (s.num == VHOST_VRING_BIG_ENDIAN) |
87 | vhost_enable_cross_endian_big(vq); | |
88 | else | |
89 | vhost_enable_cross_endian_little(vq); | |
2751c988 GK |
90 | |
91 | return 0; | |
92 | } | |
93 | ||
94 | static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx, | |
95 | int __user *argp) | |
96 | { | |
97 | struct vhost_vring_state s = { | |
98 | .index = idx, | |
99 | .num = vq->user_be | |
100 | }; | |
101 | ||
102 | if (copy_to_user(argp, &s, sizeof(s))) | |
103 | return -EFAULT; | |
104 | ||
105 | return 0; | |
106 | } | |
107 | ||
108 | static void vhost_init_is_le(struct vhost_virtqueue *vq) | |
109 | { | |
110 | /* Note for legacy virtio: user_be is initialized at reset time | |
111 | * according to the host endianness. If userspace does not set an | |
112 | * explicit endianness, the default behavior is native endian, as | |
113 | * expected by legacy virtio. | |
114 | */ | |
115 | vq->is_le = vhost_has_feature(vq, VIRTIO_F_VERSION_1) || !vq->user_be; | |
116 | } | |
117 | #else | |
c5072037 | 118 | static void vhost_disable_cross_endian(struct vhost_virtqueue *vq) |
2751c988 GK |
119 | { |
120 | } | |
121 | ||
122 | static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp) | |
123 | { | |
124 | return -ENOIOCTLCMD; | |
125 | } | |
126 | ||
127 | static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx, | |
128 | int __user *argp) | |
129 | { | |
130 | return -ENOIOCTLCMD; | |
131 | } | |
132 | ||
133 | static void vhost_init_is_le(struct vhost_virtqueue *vq) | |
134 | { | |
cda8bba0 HP |
135 | vq->is_le = vhost_has_feature(vq, VIRTIO_F_VERSION_1) |
136 | || virtio_legacy_is_little_endian(); | |
2751c988 GK |
137 | } |
138 | #endif /* CONFIG_VHOST_CROSS_ENDIAN_LEGACY */ | |
139 | ||
c5072037 GK |
140 | static void vhost_reset_is_le(struct vhost_virtqueue *vq) |
141 | { | |
cda8bba0 | 142 | vhost_init_is_le(vq); |
c5072037 GK |
143 | } |
144 | ||
7235acdb JW |
145 | struct vhost_flush_struct { |
146 | struct vhost_work work; | |
147 | struct completion wait_event; | |
148 | }; | |
149 | ||
150 | static void vhost_flush_work(struct vhost_work *work) | |
151 | { | |
152 | struct vhost_flush_struct *s; | |
153 | ||
154 | s = container_of(work, struct vhost_flush_struct, work); | |
155 | complete(&s->wait_event); | |
156 | } | |
157 | ||
3a4d5c94 MT |
158 | static void vhost_poll_func(struct file *file, wait_queue_head_t *wqh, |
159 | poll_table *pt) | |
160 | { | |
161 | struct vhost_poll *poll; | |
3a4d5c94 | 162 | |
d47effe1 | 163 | poll = container_of(pt, struct vhost_poll, table); |
3a4d5c94 MT |
164 | poll->wqh = wqh; |
165 | add_wait_queue(wqh, &poll->wait); | |
166 | } | |
167 | ||
ac6424b9 | 168 | static int vhost_poll_wakeup(wait_queue_entry_t *wait, unsigned mode, int sync, |
3a4d5c94 MT |
169 | void *key) |
170 | { | |
c23f3445 TH |
171 | struct vhost_poll *poll = container_of(wait, struct vhost_poll, wait); |
172 | ||
3a4d5c94 MT |
173 | if (!((unsigned long)key & poll->mask)) |
174 | return 0; | |
175 | ||
c23f3445 | 176 | vhost_poll_queue(poll); |
3a4d5c94 MT |
177 | return 0; |
178 | } | |
179 | ||
163049ae | 180 | void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn) |
87d6a412 | 181 | { |
04b96e55 | 182 | clear_bit(VHOST_WORK_QUEUED, &work->flags); |
87d6a412 | 183 | work->fn = fn; |
87d6a412 | 184 | } |
6ac1afbf | 185 | EXPORT_SYMBOL_GPL(vhost_work_init); |
87d6a412 | 186 | |
3a4d5c94 | 187 | /* Init poll structure */ |
c23f3445 TH |
188 | void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn, |
189 | unsigned long mask, struct vhost_dev *dev) | |
3a4d5c94 | 190 | { |
3a4d5c94 MT |
191 | init_waitqueue_func_entry(&poll->wait, vhost_poll_wakeup); |
192 | init_poll_funcptr(&poll->table, vhost_poll_func); | |
193 | poll->mask = mask; | |
c23f3445 | 194 | poll->dev = dev; |
2b8b328b | 195 | poll->wqh = NULL; |
c23f3445 | 196 | |
87d6a412 | 197 | vhost_work_init(&poll->work, fn); |
3a4d5c94 | 198 | } |
6ac1afbf | 199 | EXPORT_SYMBOL_GPL(vhost_poll_init); |
3a4d5c94 MT |
200 | |
201 | /* Start polling a file. We add ourselves to file's wait queue. The caller must | |
202 | * keep a reference to a file until after vhost_poll_stop is called. */ | |
2b8b328b | 203 | int vhost_poll_start(struct vhost_poll *poll, struct file *file) |
3a4d5c94 MT |
204 | { |
205 | unsigned long mask; | |
2b8b328b | 206 | int ret = 0; |
d47effe1 | 207 | |
70181d51 JW |
208 | if (poll->wqh) |
209 | return 0; | |
210 | ||
3a4d5c94 MT |
211 | mask = file->f_op->poll(file, &poll->table); |
212 | if (mask) | |
213 | vhost_poll_wakeup(&poll->wait, 0, 0, (void *)mask); | |
2b8b328b JW |
214 | if (mask & POLLERR) { |
215 | if (poll->wqh) | |
216 | remove_wait_queue(poll->wqh, &poll->wait); | |
217 | ret = -EINVAL; | |
218 | } | |
219 | ||
220 | return ret; | |
3a4d5c94 | 221 | } |
6ac1afbf | 222 | EXPORT_SYMBOL_GPL(vhost_poll_start); |
3a4d5c94 MT |
223 | |
224 | /* Stop polling a file. After this function returns, it becomes safe to drop the | |
225 | * file reference. You must also flush afterwards. */ | |
226 | void vhost_poll_stop(struct vhost_poll *poll) | |
227 | { | |
2b8b328b JW |
228 | if (poll->wqh) { |
229 | remove_wait_queue(poll->wqh, &poll->wait); | |
230 | poll->wqh = NULL; | |
231 | } | |
3a4d5c94 | 232 | } |
6ac1afbf | 233 | EXPORT_SYMBOL_GPL(vhost_poll_stop); |
3a4d5c94 | 234 | |
7235acdb | 235 | void vhost_work_flush(struct vhost_dev *dev, struct vhost_work *work) |
0174b0c3 | 236 | { |
7235acdb | 237 | struct vhost_flush_struct flush; |
d47effe1 | 238 | |
7235acdb JW |
239 | if (dev->worker) { |
240 | init_completion(&flush.wait_event); | |
241 | vhost_work_init(&flush.work, vhost_flush_work); | |
0174b0c3 | 242 | |
7235acdb JW |
243 | vhost_work_queue(dev, &flush.work); |
244 | wait_for_completion(&flush.wait_event); | |
245 | } | |
3a4d5c94 | 246 | } |
6ac1afbf | 247 | EXPORT_SYMBOL_GPL(vhost_work_flush); |
3a4d5c94 | 248 | |
87d6a412 MT |
249 | /* Flush any work that has been scheduled. When calling this, don't hold any |
250 | * locks that are also used by the callback. */ | |
251 | void vhost_poll_flush(struct vhost_poll *poll) | |
252 | { | |
253 | vhost_work_flush(poll->dev, &poll->work); | |
254 | } | |
6ac1afbf | 255 | EXPORT_SYMBOL_GPL(vhost_poll_flush); |
87d6a412 | 256 | |
163049ae | 257 | void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work) |
3a4d5c94 | 258 | { |
04b96e55 JW |
259 | if (!dev->worker) |
260 | return; | |
c23f3445 | 261 | |
04b96e55 JW |
262 | if (!test_and_set_bit(VHOST_WORK_QUEUED, &work->flags)) { |
263 | /* We can only add the work to the list after we're | |
264 | * sure it was not in the list. | |
635abf01 | 265 | * test_and_set_bit() implies a memory barrier. |
04b96e55 | 266 | */ |
04b96e55 | 267 | llist_add(&work->node, &dev->work_list); |
c23f3445 TH |
268 | wake_up_process(dev->worker); |
269 | } | |
3a4d5c94 | 270 | } |
6ac1afbf | 271 | EXPORT_SYMBOL_GPL(vhost_work_queue); |
3a4d5c94 | 272 | |
526d3e7f JW |
273 | /* A lockless hint for busy polling code to exit the loop */ |
274 | bool vhost_has_work(struct vhost_dev *dev) | |
275 | { | |
04b96e55 | 276 | return !llist_empty(&dev->work_list); |
526d3e7f JW |
277 | } |
278 | EXPORT_SYMBOL_GPL(vhost_has_work); | |
279 | ||
87d6a412 MT |
280 | void vhost_poll_queue(struct vhost_poll *poll) |
281 | { | |
282 | vhost_work_queue(poll->dev, &poll->work); | |
283 | } | |
6ac1afbf | 284 | EXPORT_SYMBOL_GPL(vhost_poll_queue); |
87d6a412 | 285 | |
f8894913 JW |
286 | static void __vhost_vq_meta_reset(struct vhost_virtqueue *vq) |
287 | { | |
288 | int j; | |
289 | ||
290 | for (j = 0; j < VHOST_NUM_ADDRS; j++) | |
291 | vq->meta_iotlb[j] = NULL; | |
292 | } | |
293 | ||
294 | static void vhost_vq_meta_reset(struct vhost_dev *d) | |
295 | { | |
296 | int i; | |
297 | ||
298 | for (i = 0; i < d->nvqs; ++i) | |
299 | __vhost_vq_meta_reset(d->vqs[i]); | |
300 | } | |
301 | ||
3a4d5c94 MT |
302 | static void vhost_vq_reset(struct vhost_dev *dev, |
303 | struct vhost_virtqueue *vq) | |
304 | { | |
305 | vq->num = 1; | |
306 | vq->desc = NULL; | |
307 | vq->avail = NULL; | |
308 | vq->used = NULL; | |
309 | vq->last_avail_idx = 0; | |
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; | |
3a4d5c94 MT |
321 | vq->kick = NULL; |
322 | vq->call_ctx = NULL; | |
73a99f08 | 323 | vq->log_ctx = NULL; |
c5072037 GK |
324 | vhost_reset_is_le(vq); |
325 | vhost_disable_cross_endian(vq); | |
03088137 | 326 | vq->busyloop_timeout = 0; |
a9709d68 | 327 | vq->umem = NULL; |
6b1e6cc7 | 328 | vq->iotlb = NULL; |
f8894913 | 329 | __vhost_vq_meta_reset(vq); |
3a4d5c94 MT |
330 | } |
331 | ||
c23f3445 TH |
332 | static int vhost_worker(void *data) |
333 | { | |
334 | struct vhost_dev *dev = data; | |
04b96e55 JW |
335 | struct vhost_work *work, *work_next; |
336 | struct llist_node *node; | |
d7ffde35 | 337 | mm_segment_t oldfs = get_fs(); |
c23f3445 | 338 | |
d7ffde35 | 339 | set_fs(USER_DS); |
64e1c807 MT |
340 | use_mm(dev->mm); |
341 | ||
c23f3445 TH |
342 | for (;;) { |
343 | /* mb paired w/ kthread_stop */ | |
344 | set_current_state(TASK_INTERRUPTIBLE); | |
345 | ||
c23f3445 | 346 | if (kthread_should_stop()) { |
c23f3445 | 347 | __set_current_state(TASK_RUNNING); |
64e1c807 | 348 | break; |
c23f3445 | 349 | } |
04b96e55 JW |
350 | |
351 | node = llist_del_all(&dev->work_list); | |
352 | if (!node) | |
353 | schedule(); | |
354 | ||
355 | node = llist_reverse_order(node); | |
356 | /* make sure flag is seen after deletion */ | |
357 | smp_wmb(); | |
358 | llist_for_each_entry_safe(work, work_next, node, node) { | |
359 | clear_bit(VHOST_WORK_QUEUED, &work->flags); | |
c23f3445 TH |
360 | __set_current_state(TASK_RUNNING); |
361 | work->fn(work); | |
d550dda1 NHE |
362 | if (need_resched()) |
363 | schedule(); | |
04b96e55 | 364 | } |
c23f3445 | 365 | } |
64e1c807 | 366 | unuse_mm(dev->mm); |
d7ffde35 | 367 | set_fs(oldfs); |
64e1c807 | 368 | return 0; |
c23f3445 TH |
369 | } |
370 | ||
bab632d6 MT |
371 | static void vhost_vq_free_iovecs(struct vhost_virtqueue *vq) |
372 | { | |
373 | kfree(vq->indirect); | |
374 | vq->indirect = NULL; | |
375 | kfree(vq->log); | |
376 | vq->log = NULL; | |
377 | kfree(vq->heads); | |
378 | vq->heads = NULL; | |
bab632d6 MT |
379 | } |
380 | ||
e0e9b406 JW |
381 | /* Helper to allocate iovec buffers for all vqs. */ |
382 | static long vhost_dev_alloc_iovecs(struct vhost_dev *dev) | |
383 | { | |
6d5e6aa8 | 384 | struct vhost_virtqueue *vq; |
e0e9b406 | 385 | int i; |
d47effe1 | 386 | |
e0e9b406 | 387 | for (i = 0; i < dev->nvqs; ++i) { |
6d5e6aa8 AH |
388 | vq = dev->vqs[i]; |
389 | vq->indirect = kmalloc(sizeof *vq->indirect * UIO_MAXIOV, | |
390 | GFP_KERNEL); | |
391 | vq->log = kmalloc(sizeof *vq->log * UIO_MAXIOV, GFP_KERNEL); | |
392 | vq->heads = kmalloc(sizeof *vq->heads * UIO_MAXIOV, GFP_KERNEL); | |
393 | if (!vq->indirect || !vq->log || !vq->heads) | |
e0e9b406 JW |
394 | goto err_nomem; |
395 | } | |
396 | return 0; | |
d47effe1 | 397 | |
e0e9b406 | 398 | err_nomem: |
bab632d6 | 399 | for (; i >= 0; --i) |
3ab2e420 | 400 | vhost_vq_free_iovecs(dev->vqs[i]); |
e0e9b406 JW |
401 | return -ENOMEM; |
402 | } | |
403 | ||
404 | static void vhost_dev_free_iovecs(struct vhost_dev *dev) | |
405 | { | |
406 | int i; | |
d47effe1 | 407 | |
bab632d6 | 408 | for (i = 0; i < dev->nvqs; ++i) |
3ab2e420 | 409 | vhost_vq_free_iovecs(dev->vqs[i]); |
e0e9b406 JW |
410 | } |
411 | ||
59566b6e | 412 | void vhost_dev_init(struct vhost_dev *dev, |
3ab2e420 | 413 | struct vhost_virtqueue **vqs, int nvqs) |
3a4d5c94 | 414 | { |
6d5e6aa8 | 415 | struct vhost_virtqueue *vq; |
3a4d5c94 | 416 | int i; |
c23f3445 | 417 | |
3a4d5c94 MT |
418 | dev->vqs = vqs; |
419 | dev->nvqs = nvqs; | |
420 | mutex_init(&dev->mutex); | |
421 | dev->log_ctx = NULL; | |
422 | dev->log_file = NULL; | |
a9709d68 | 423 | dev->umem = NULL; |
6b1e6cc7 | 424 | dev->iotlb = NULL; |
3a4d5c94 | 425 | dev->mm = NULL; |
c23f3445 | 426 | dev->worker = NULL; |
04b96e55 | 427 | init_llist_head(&dev->work_list); |
6b1e6cc7 JW |
428 | init_waitqueue_head(&dev->wait); |
429 | INIT_LIST_HEAD(&dev->read_list); | |
430 | INIT_LIST_HEAD(&dev->pending_list); | |
431 | spin_lock_init(&dev->iotlb_lock); | |
04b96e55 | 432 | |
3a4d5c94 MT |
433 | |
434 | for (i = 0; i < dev->nvqs; ++i) { | |
6d5e6aa8 AH |
435 | vq = dev->vqs[i]; |
436 | vq->log = NULL; | |
437 | vq->indirect = NULL; | |
438 | vq->heads = NULL; | |
439 | vq->dev = dev; | |
440 | mutex_init(&vq->mutex); | |
441 | vhost_vq_reset(dev, vq); | |
442 | if (vq->handle_kick) | |
443 | vhost_poll_init(&vq->poll, vq->handle_kick, | |
444 | POLLIN, dev); | |
3a4d5c94 | 445 | } |
3a4d5c94 | 446 | } |
6ac1afbf | 447 | EXPORT_SYMBOL_GPL(vhost_dev_init); |
3a4d5c94 MT |
448 | |
449 | /* Caller should have device mutex */ | |
450 | long vhost_dev_check_owner(struct vhost_dev *dev) | |
451 | { | |
452 | /* Are you the owner? If not, I don't think you mean to do that */ | |
453 | return dev->mm == current->mm ? 0 : -EPERM; | |
454 | } | |
6ac1afbf | 455 | EXPORT_SYMBOL_GPL(vhost_dev_check_owner); |
3a4d5c94 | 456 | |
87d6a412 | 457 | struct vhost_attach_cgroups_struct { |
d47effe1 KK |
458 | struct vhost_work work; |
459 | struct task_struct *owner; | |
460 | int ret; | |
87d6a412 MT |
461 | }; |
462 | ||
463 | static void vhost_attach_cgroups_work(struct vhost_work *work) | |
464 | { | |
d47effe1 KK |
465 | struct vhost_attach_cgroups_struct *s; |
466 | ||
467 | s = container_of(work, struct vhost_attach_cgroups_struct, work); | |
468 | s->ret = cgroup_attach_task_all(s->owner, current); | |
87d6a412 MT |
469 | } |
470 | ||
471 | static int vhost_attach_cgroups(struct vhost_dev *dev) | |
472 | { | |
d47effe1 KK |
473 | struct vhost_attach_cgroups_struct attach; |
474 | ||
475 | attach.owner = current; | |
476 | vhost_work_init(&attach.work, vhost_attach_cgroups_work); | |
477 | vhost_work_queue(dev, &attach.work); | |
478 | vhost_work_flush(dev, &attach.work); | |
479 | return attach.ret; | |
87d6a412 MT |
480 | } |
481 | ||
05c05351 MT |
482 | /* Caller should have device mutex */ |
483 | bool vhost_dev_has_owner(struct vhost_dev *dev) | |
484 | { | |
485 | return dev->mm; | |
486 | } | |
6ac1afbf | 487 | EXPORT_SYMBOL_GPL(vhost_dev_has_owner); |
05c05351 | 488 | |
3a4d5c94 | 489 | /* Caller should have device mutex */ |
54db63c2 | 490 | long vhost_dev_set_owner(struct vhost_dev *dev) |
3a4d5c94 | 491 | { |
c23f3445 TH |
492 | struct task_struct *worker; |
493 | int err; | |
d47effe1 | 494 | |
3a4d5c94 | 495 | /* Is there an owner already? */ |
05c05351 | 496 | if (vhost_dev_has_owner(dev)) { |
c23f3445 TH |
497 | err = -EBUSY; |
498 | goto err_mm; | |
499 | } | |
d47effe1 | 500 | |
3a4d5c94 MT |
501 | /* No owner, become one */ |
502 | dev->mm = get_task_mm(current); | |
c23f3445 TH |
503 | worker = kthread_create(vhost_worker, dev, "vhost-%d", current->pid); |
504 | if (IS_ERR(worker)) { | |
505 | err = PTR_ERR(worker); | |
506 | goto err_worker; | |
507 | } | |
508 | ||
509 | dev->worker = worker; | |
87d6a412 MT |
510 | wake_up_process(worker); /* avoid contributing to loadavg */ |
511 | ||
512 | err = vhost_attach_cgroups(dev); | |
9e3d1957 MT |
513 | if (err) |
514 | goto err_cgroup; | |
c23f3445 | 515 | |
e0e9b406 JW |
516 | err = vhost_dev_alloc_iovecs(dev); |
517 | if (err) | |
518 | goto err_cgroup; | |
519 | ||
3a4d5c94 | 520 | return 0; |
9e3d1957 MT |
521 | err_cgroup: |
522 | kthread_stop(worker); | |
615cc221 | 523 | dev->worker = NULL; |
c23f3445 TH |
524 | err_worker: |
525 | if (dev->mm) | |
526 | mmput(dev->mm); | |
527 | dev->mm = NULL; | |
528 | err_mm: | |
529 | return err; | |
3a4d5c94 | 530 | } |
6ac1afbf | 531 | EXPORT_SYMBOL_GPL(vhost_dev_set_owner); |
3a4d5c94 | 532 | |
a9709d68 JW |
533 | struct vhost_umem *vhost_dev_reset_owner_prepare(void) |
534 | { | |
6c5ab651 | 535 | return kvzalloc(sizeof(struct vhost_umem), GFP_KERNEL); |
150b9e51 | 536 | } |
6ac1afbf | 537 | EXPORT_SYMBOL_GPL(vhost_dev_reset_owner_prepare); |
3a4d5c94 | 538 | |
150b9e51 | 539 | /* Caller should have device mutex */ |
a9709d68 | 540 | void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_umem *umem) |
150b9e51 | 541 | { |
47283bef MT |
542 | int i; |
543 | ||
f6f93f75 | 544 | vhost_dev_cleanup(dev); |
3a4d5c94 | 545 | |
150b9e51 | 546 | /* Restore memory to default empty mapping. */ |
a9709d68 JW |
547 | INIT_LIST_HEAD(&umem->umem_list); |
548 | dev->umem = umem; | |
47283bef MT |
549 | /* We don't need VQ locks below since vhost_dev_cleanup makes sure |
550 | * VQs aren't running. | |
551 | */ | |
552 | for (i = 0; i < dev->nvqs; ++i) | |
a9709d68 | 553 | dev->vqs[i]->umem = umem; |
3a4d5c94 | 554 | } |
6ac1afbf | 555 | EXPORT_SYMBOL_GPL(vhost_dev_reset_owner); |
3a4d5c94 | 556 | |
b211616d | 557 | void vhost_dev_stop(struct vhost_dev *dev) |
bab632d6 MT |
558 | { |
559 | int i; | |
b211616d MT |
560 | |
561 | for (i = 0; i < dev->nvqs; ++i) { | |
3ab2e420 AH |
562 | if (dev->vqs[i]->kick && dev->vqs[i]->handle_kick) { |
563 | vhost_poll_stop(&dev->vqs[i]->poll); | |
564 | vhost_poll_flush(&dev->vqs[i]->poll); | |
b211616d | 565 | } |
bab632d6 | 566 | } |
bab632d6 | 567 | } |
6ac1afbf | 568 | EXPORT_SYMBOL_GPL(vhost_dev_stop); |
bab632d6 | 569 | |
6b1e6cc7 JW |
570 | static void vhost_umem_free(struct vhost_umem *umem, |
571 | struct vhost_umem_node *node) | |
572 | { | |
573 | vhost_umem_interval_tree_remove(node, &umem->umem_tree); | |
574 | list_del(&node->link); | |
575 | kfree(node); | |
576 | umem->numem--; | |
577 | } | |
578 | ||
a9709d68 JW |
579 | static void vhost_umem_clean(struct vhost_umem *umem) |
580 | { | |
581 | struct vhost_umem_node *node, *tmp; | |
582 | ||
583 | if (!umem) | |
584 | return; | |
585 | ||
6b1e6cc7 JW |
586 | list_for_each_entry_safe(node, tmp, &umem->umem_list, link) |
587 | vhost_umem_free(umem, node); | |
588 | ||
a9709d68 JW |
589 | kvfree(umem); |
590 | } | |
591 | ||
6b1e6cc7 JW |
592 | static void vhost_clear_msg(struct vhost_dev *dev) |
593 | { | |
594 | struct vhost_msg_node *node, *n; | |
595 | ||
596 | spin_lock(&dev->iotlb_lock); | |
597 | ||
598 | list_for_each_entry_safe(node, n, &dev->read_list, node) { | |
599 | list_del(&node->node); | |
600 | kfree(node); | |
601 | } | |
602 | ||
603 | list_for_each_entry_safe(node, n, &dev->pending_list, node) { | |
604 | list_del(&node->node); | |
605 | kfree(node); | |
606 | } | |
607 | ||
608 | spin_unlock(&dev->iotlb_lock); | |
609 | } | |
610 | ||
f6f93f75 | 611 | void vhost_dev_cleanup(struct vhost_dev *dev) |
3a4d5c94 MT |
612 | { |
613 | int i; | |
d47effe1 | 614 | |
3a4d5c94 | 615 | for (i = 0; i < dev->nvqs; ++i) { |
3ab2e420 AH |
616 | if (dev->vqs[i]->error_ctx) |
617 | eventfd_ctx_put(dev->vqs[i]->error_ctx); | |
3ab2e420 AH |
618 | if (dev->vqs[i]->kick) |
619 | fput(dev->vqs[i]->kick); | |
620 | if (dev->vqs[i]->call_ctx) | |
621 | eventfd_ctx_put(dev->vqs[i]->call_ctx); | |
3ab2e420 | 622 | vhost_vq_reset(dev, dev->vqs[i]); |
3a4d5c94 | 623 | } |
e0e9b406 | 624 | vhost_dev_free_iovecs(dev); |
3a4d5c94 MT |
625 | if (dev->log_ctx) |
626 | eventfd_ctx_put(dev->log_ctx); | |
627 | dev->log_ctx = NULL; | |
628 | if (dev->log_file) | |
629 | fput(dev->log_file); | |
630 | dev->log_file = NULL; | |
631 | /* No one will access memory at this point */ | |
a9709d68 JW |
632 | vhost_umem_clean(dev->umem); |
633 | dev->umem = NULL; | |
6b1e6cc7 JW |
634 | vhost_umem_clean(dev->iotlb); |
635 | dev->iotlb = NULL; | |
636 | vhost_clear_msg(dev); | |
637 | wake_up_interruptible_poll(&dev->wait, POLLIN | POLLRDNORM); | |
04b96e55 | 638 | WARN_ON(!llist_empty(&dev->work_list)); |
78b620ce ED |
639 | if (dev->worker) { |
640 | kthread_stop(dev->worker); | |
641 | dev->worker = NULL; | |
642 | } | |
533a19b4 MT |
643 | if (dev->mm) |
644 | mmput(dev->mm); | |
645 | dev->mm = NULL; | |
3a4d5c94 | 646 | } |
6ac1afbf | 647 | EXPORT_SYMBOL_GPL(vhost_dev_cleanup); |
3a4d5c94 MT |
648 | |
649 | static int log_access_ok(void __user *log_base, u64 addr, unsigned long sz) | |
650 | { | |
651 | u64 a = addr / VHOST_PAGE_SIZE / 8; | |
d47effe1 | 652 | |
3a4d5c94 MT |
653 | /* Make sure 64 bit math will not overflow. */ |
654 | if (a > ULONG_MAX - (unsigned long)log_base || | |
655 | a + (unsigned long)log_base > ULONG_MAX) | |
6d97e55f | 656 | return 0; |
3a4d5c94 MT |
657 | |
658 | return access_ok(VERIFY_WRITE, log_base + a, | |
659 | (sz + VHOST_PAGE_SIZE * 8 - 1) / VHOST_PAGE_SIZE / 8); | |
660 | } | |
661 | ||
ec33d031 MT |
662 | static bool vhost_overflow(u64 uaddr, u64 size) |
663 | { | |
664 | /* Make sure 64 bit math will not overflow. */ | |
665 | return uaddr > ULONG_MAX || size > ULONG_MAX || uaddr > ULONG_MAX - size; | |
666 | } | |
667 | ||
3a4d5c94 | 668 | /* Caller should have vq mutex and device mutex. */ |
a9709d68 | 669 | static int vq_memory_access_ok(void __user *log_base, struct vhost_umem *umem, |
3a4d5c94 MT |
670 | int log_all) |
671 | { | |
a9709d68 | 672 | struct vhost_umem_node *node; |
179b284e | 673 | |
a9709d68 | 674 | if (!umem) |
f8322fbe | 675 | return 0; |
179b284e | 676 | |
a9709d68 JW |
677 | list_for_each_entry(node, &umem->umem_list, link) { |
678 | unsigned long a = node->userspace_addr; | |
679 | ||
ec33d031 | 680 | if (vhost_overflow(node->userspace_addr, node->size)) |
3a4d5c94 | 681 | return 0; |
ec33d031 MT |
682 | |
683 | ||
684 | if (!access_ok(VERIFY_WRITE, (void __user *)a, | |
a9709d68 | 685 | node->size)) |
3a4d5c94 MT |
686 | return 0; |
687 | else if (log_all && !log_access_ok(log_base, | |
a9709d68 JW |
688 | node->start, |
689 | node->size)) | |
3a4d5c94 MT |
690 | return 0; |
691 | } | |
692 | return 1; | |
693 | } | |
694 | ||
f8894913 JW |
695 | static inline void __user *vhost_vq_meta_fetch(struct vhost_virtqueue *vq, |
696 | u64 addr, unsigned int size, | |
697 | int type) | |
698 | { | |
699 | const struct vhost_umem_node *node = vq->meta_iotlb[type]; | |
700 | ||
701 | if (!node) | |
702 | return NULL; | |
703 | ||
704 | return (void *)(uintptr_t)(node->userspace_addr + addr - node->start); | |
705 | } | |
706 | ||
3a4d5c94 MT |
707 | /* Can we switch to this memory table? */ |
708 | /* Caller should have device mutex but not vq mutex */ | |
a9709d68 | 709 | static int memory_access_ok(struct vhost_dev *d, struct vhost_umem *umem, |
3a4d5c94 MT |
710 | int log_all) |
711 | { | |
712 | int i; | |
d47effe1 | 713 | |
3a4d5c94 MT |
714 | for (i = 0; i < d->nvqs; ++i) { |
715 | int ok; | |
ea16c514 MT |
716 | bool log; |
717 | ||
3ab2e420 | 718 | mutex_lock(&d->vqs[i]->mutex); |
ea16c514 | 719 | log = log_all || vhost_has_feature(d->vqs[i], VHOST_F_LOG_ALL); |
3a4d5c94 | 720 | /* If ring is inactive, will check when it's enabled. */ |
3ab2e420 | 721 | if (d->vqs[i]->private_data) |
a9709d68 JW |
722 | ok = vq_memory_access_ok(d->vqs[i]->log_base, |
723 | umem, log); | |
3a4d5c94 MT |
724 | else |
725 | ok = 1; | |
3ab2e420 | 726 | mutex_unlock(&d->vqs[i]->mutex); |
3a4d5c94 MT |
727 | if (!ok) |
728 | return 0; | |
729 | } | |
730 | return 1; | |
731 | } | |
732 | ||
6b1e6cc7 JW |
733 | static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len, |
734 | struct iovec iov[], int iov_size, int access); | |
bfe2bc51 | 735 | |
72952cc0 | 736 | static int vhost_copy_to_user(struct vhost_virtqueue *vq, void __user *to, |
bfe2bc51 JW |
737 | const void *from, unsigned size) |
738 | { | |
6b1e6cc7 | 739 | int ret; |
bfe2bc51 | 740 | |
6b1e6cc7 JW |
741 | if (!vq->iotlb) |
742 | return __copy_to_user(to, from, size); | |
743 | else { | |
744 | /* This function should be called after iotlb | |
745 | * prefetch, which means we're sure that all vq | |
746 | * could be access through iotlb. So -EAGAIN should | |
747 | * not happen in this case. | |
748 | */ | |
6b1e6cc7 | 749 | struct iov_iter t; |
f8894913 JW |
750 | void __user *uaddr = vhost_vq_meta_fetch(vq, |
751 | (u64)(uintptr_t)to, size, | |
752 | VHOST_ADDR_DESC); | |
753 | ||
754 | if (uaddr) | |
755 | return __copy_to_user(uaddr, from, size); | |
756 | ||
6b1e6cc7 JW |
757 | ret = translate_desc(vq, (u64)(uintptr_t)to, size, vq->iotlb_iov, |
758 | ARRAY_SIZE(vq->iotlb_iov), | |
759 | VHOST_ACCESS_WO); | |
760 | if (ret < 0) | |
761 | goto out; | |
762 | iov_iter_init(&t, WRITE, vq->iotlb_iov, ret, size); | |
763 | ret = copy_to_iter(from, size, &t); | |
764 | if (ret == size) | |
765 | ret = 0; | |
766 | } | |
767 | out: | |
768 | return ret; | |
769 | } | |
bfe2bc51 JW |
770 | |
771 | static int vhost_copy_from_user(struct vhost_virtqueue *vq, void *to, | |
72952cc0 | 772 | void __user *from, unsigned size) |
bfe2bc51 | 773 | { |
6b1e6cc7 JW |
774 | int ret; |
775 | ||
776 | if (!vq->iotlb) | |
777 | return __copy_from_user(to, from, size); | |
778 | else { | |
779 | /* This function should be called after iotlb | |
780 | * prefetch, which means we're sure that vq | |
781 | * could be access through iotlb. So -EAGAIN should | |
782 | * not happen in this case. | |
783 | */ | |
f8894913 JW |
784 | void __user *uaddr = vhost_vq_meta_fetch(vq, |
785 | (u64)(uintptr_t)from, size, | |
786 | VHOST_ADDR_DESC); | |
6b1e6cc7 | 787 | struct iov_iter f; |
f8894913 JW |
788 | |
789 | if (uaddr) | |
790 | return __copy_from_user(to, uaddr, size); | |
791 | ||
6b1e6cc7 JW |
792 | ret = translate_desc(vq, (u64)(uintptr_t)from, size, vq->iotlb_iov, |
793 | ARRAY_SIZE(vq->iotlb_iov), | |
794 | VHOST_ACCESS_RO); | |
795 | if (ret < 0) { | |
796 | vq_err(vq, "IOTLB translation failure: uaddr " | |
797 | "%p size 0x%llx\n", from, | |
798 | (unsigned long long) size); | |
799 | goto out; | |
800 | } | |
801 | iov_iter_init(&f, READ, vq->iotlb_iov, ret, size); | |
802 | ret = copy_from_iter(to, size, &f); | |
803 | if (ret == size) | |
804 | ret = 0; | |
805 | } | |
806 | ||
807 | out: | |
808 | return ret; | |
809 | } | |
810 | ||
f8894913 JW |
811 | static void __user *__vhost_get_user_slow(struct vhost_virtqueue *vq, |
812 | void __user *addr, unsigned int size, | |
813 | int type) | |
6b1e6cc7 JW |
814 | { |
815 | int ret; | |
816 | ||
6b1e6cc7 JW |
817 | ret = translate_desc(vq, (u64)(uintptr_t)addr, size, vq->iotlb_iov, |
818 | ARRAY_SIZE(vq->iotlb_iov), | |
819 | VHOST_ACCESS_RO); | |
820 | if (ret < 0) { | |
821 | vq_err(vq, "IOTLB translation failure: uaddr " | |
822 | "%p size 0x%llx\n", addr, | |
823 | (unsigned long long) size); | |
824 | return NULL; | |
825 | } | |
826 | ||
827 | if (ret != 1 || vq->iotlb_iov[0].iov_len != size) { | |
828 | vq_err(vq, "Non atomic userspace memory access: uaddr " | |
829 | "%p size 0x%llx\n", addr, | |
830 | (unsigned long long) size); | |
831 | return NULL; | |
832 | } | |
833 | ||
834 | return vq->iotlb_iov[0].iov_base; | |
835 | } | |
836 | ||
f8894913 JW |
837 | /* This function should be called after iotlb |
838 | * prefetch, which means we're sure that vq | |
839 | * could be access through iotlb. So -EAGAIN should | |
840 | * not happen in this case. | |
841 | */ | |
842 | static inline void __user *__vhost_get_user(struct vhost_virtqueue *vq, | |
843 | void *addr, unsigned int size, | |
844 | int type) | |
845 | { | |
846 | void __user *uaddr = vhost_vq_meta_fetch(vq, | |
847 | (u64)(uintptr_t)addr, size, type); | |
848 | if (uaddr) | |
849 | return uaddr; | |
850 | ||
851 | return __vhost_get_user_slow(vq, addr, size, type); | |
852 | } | |
853 | ||
854 | #define vhost_put_user(vq, x, ptr) \ | |
6b1e6cc7 JW |
855 | ({ \ |
856 | int ret = -EFAULT; \ | |
857 | if (!vq->iotlb) { \ | |
858 | ret = __put_user(x, ptr); \ | |
859 | } else { \ | |
860 | __typeof__(ptr) to = \ | |
f8894913 JW |
861 | (__typeof__(ptr)) __vhost_get_user(vq, ptr, \ |
862 | sizeof(*ptr), VHOST_ADDR_USED); \ | |
6b1e6cc7 JW |
863 | if (to != NULL) \ |
864 | ret = __put_user(x, to); \ | |
865 | else \ | |
866 | ret = -EFAULT; \ | |
867 | } \ | |
868 | ret; \ | |
869 | }) | |
870 | ||
f8894913 | 871 | #define vhost_get_user(vq, x, ptr, type) \ |
6b1e6cc7 JW |
872 | ({ \ |
873 | int ret; \ | |
874 | if (!vq->iotlb) { \ | |
875 | ret = __get_user(x, ptr); \ | |
876 | } else { \ | |
877 | __typeof__(ptr) from = \ | |
f8894913 JW |
878 | (__typeof__(ptr)) __vhost_get_user(vq, ptr, \ |
879 | sizeof(*ptr), \ | |
880 | type); \ | |
6b1e6cc7 JW |
881 | if (from != NULL) \ |
882 | ret = __get_user(x, from); \ | |
883 | else \ | |
884 | ret = -EFAULT; \ | |
885 | } \ | |
886 | ret; \ | |
887 | }) | |
888 | ||
f8894913 JW |
889 | #define vhost_get_avail(vq, x, ptr) \ |
890 | vhost_get_user(vq, x, ptr, VHOST_ADDR_AVAIL) | |
891 | ||
892 | #define vhost_get_used(vq, x, ptr) \ | |
893 | vhost_get_user(vq, x, ptr, VHOST_ADDR_USED) | |
894 | ||
6b1e6cc7 JW |
895 | static void vhost_dev_lock_vqs(struct vhost_dev *d) |
896 | { | |
897 | int i = 0; | |
898 | for (i = 0; i < d->nvqs; ++i) | |
e9cb4239 | 899 | mutex_lock_nested(&d->vqs[i]->mutex, i); |
6b1e6cc7 JW |
900 | } |
901 | ||
902 | static void vhost_dev_unlock_vqs(struct vhost_dev *d) | |
903 | { | |
904 | int i = 0; | |
905 | for (i = 0; i < d->nvqs; ++i) | |
906 | mutex_unlock(&d->vqs[i]->mutex); | |
907 | } | |
908 | ||
909 | static int vhost_new_umem_range(struct vhost_umem *umem, | |
910 | u64 start, u64 size, u64 end, | |
911 | u64 userspace_addr, int perm) | |
912 | { | |
913 | struct vhost_umem_node *tmp, *node = kmalloc(sizeof(*node), GFP_ATOMIC); | |
914 | ||
915 | if (!node) | |
916 | return -ENOMEM; | |
917 | ||
918 | if (umem->numem == max_iotlb_entries) { | |
919 | tmp = list_first_entry(&umem->umem_list, typeof(*tmp), link); | |
920 | vhost_umem_free(umem, tmp); | |
921 | } | |
922 | ||
923 | node->start = start; | |
924 | node->size = size; | |
925 | node->last = end; | |
926 | node->userspace_addr = userspace_addr; | |
927 | node->perm = perm; | |
928 | INIT_LIST_HEAD(&node->link); | |
929 | list_add_tail(&node->link, &umem->umem_list); | |
930 | vhost_umem_interval_tree_insert(node, &umem->umem_tree); | |
931 | umem->numem++; | |
932 | ||
933 | return 0; | |
934 | } | |
935 | ||
936 | static void vhost_del_umem_range(struct vhost_umem *umem, | |
937 | u64 start, u64 end) | |
938 | { | |
939 | struct vhost_umem_node *node; | |
940 | ||
941 | while ((node = vhost_umem_interval_tree_iter_first(&umem->umem_tree, | |
942 | start, end))) | |
943 | vhost_umem_free(umem, node); | |
944 | } | |
945 | ||
946 | static void vhost_iotlb_notify_vq(struct vhost_dev *d, | |
947 | struct vhost_iotlb_msg *msg) | |
948 | { | |
949 | struct vhost_msg_node *node, *n; | |
950 | ||
951 | spin_lock(&d->iotlb_lock); | |
952 | ||
953 | list_for_each_entry_safe(node, n, &d->pending_list, node) { | |
954 | struct vhost_iotlb_msg *vq_msg = &node->msg.iotlb; | |
955 | if (msg->iova <= vq_msg->iova && | |
956 | msg->iova + msg->size - 1 > vq_msg->iova && | |
957 | vq_msg->type == VHOST_IOTLB_MISS) { | |
958 | vhost_poll_queue(&node->vq->poll); | |
959 | list_del(&node->node); | |
960 | kfree(node); | |
961 | } | |
962 | } | |
963 | ||
964 | spin_unlock(&d->iotlb_lock); | |
965 | } | |
966 | ||
967 | static int umem_access_ok(u64 uaddr, u64 size, int access) | |
968 | { | |
969 | unsigned long a = uaddr; | |
970 | ||
ec33d031 MT |
971 | /* Make sure 64 bit math will not overflow. */ |
972 | if (vhost_overflow(uaddr, size)) | |
973 | return -EFAULT; | |
974 | ||
6b1e6cc7 JW |
975 | if ((access & VHOST_ACCESS_RO) && |
976 | !access_ok(VERIFY_READ, (void __user *)a, size)) | |
977 | return -EFAULT; | |
978 | if ((access & VHOST_ACCESS_WO) && | |
979 | !access_ok(VERIFY_WRITE, (void __user *)a, size)) | |
980 | return -EFAULT; | |
981 | return 0; | |
982 | } | |
983 | ||
72952cc0 MT |
984 | static int vhost_process_iotlb_msg(struct vhost_dev *dev, |
985 | struct vhost_iotlb_msg *msg) | |
6b1e6cc7 JW |
986 | { |
987 | int ret = 0; | |
988 | ||
989 | vhost_dev_lock_vqs(dev); | |
990 | switch (msg->type) { | |
991 | case VHOST_IOTLB_UPDATE: | |
992 | if (!dev->iotlb) { | |
993 | ret = -EFAULT; | |
994 | break; | |
995 | } | |
996 | if (umem_access_ok(msg->uaddr, msg->size, msg->perm)) { | |
997 | ret = -EFAULT; | |
998 | break; | |
999 | } | |
f8894913 | 1000 | vhost_vq_meta_reset(dev); |
6b1e6cc7 JW |
1001 | if (vhost_new_umem_range(dev->iotlb, msg->iova, msg->size, |
1002 | msg->iova + msg->size - 1, | |
1003 | msg->uaddr, msg->perm)) { | |
1004 | ret = -ENOMEM; | |
1005 | break; | |
1006 | } | |
1007 | vhost_iotlb_notify_vq(dev, msg); | |
1008 | break; | |
1009 | case VHOST_IOTLB_INVALIDATE: | |
6f3180af JW |
1010 | if (!dev->iotlb) { |
1011 | ret = -EFAULT; | |
1012 | break; | |
1013 | } | |
f8894913 | 1014 | vhost_vq_meta_reset(dev); |
6b1e6cc7 JW |
1015 | vhost_del_umem_range(dev->iotlb, msg->iova, |
1016 | msg->iova + msg->size - 1); | |
1017 | break; | |
1018 | default: | |
1019 | ret = -EINVAL; | |
1020 | break; | |
1021 | } | |
1022 | ||
1023 | vhost_dev_unlock_vqs(dev); | |
1024 | return ret; | |
1025 | } | |
1026 | ssize_t vhost_chr_write_iter(struct vhost_dev *dev, | |
1027 | struct iov_iter *from) | |
1028 | { | |
1029 | struct vhost_msg_node node; | |
1030 | unsigned size = sizeof(struct vhost_msg); | |
1031 | size_t ret; | |
1032 | int err; | |
1033 | ||
1034 | if (iov_iter_count(from) < size) | |
1035 | return 0; | |
1036 | ret = copy_from_iter(&node.msg, size, from); | |
1037 | if (ret != size) | |
1038 | goto done; | |
1039 | ||
1040 | switch (node.msg.type) { | |
1041 | case VHOST_IOTLB_MSG: | |
1042 | err = vhost_process_iotlb_msg(dev, &node.msg.iotlb); | |
1043 | if (err) | |
1044 | ret = err; | |
1045 | break; | |
1046 | default: | |
1047 | ret = -EINVAL; | |
1048 | break; | |
1049 | } | |
1050 | ||
1051 | done: | |
1052 | return ret; | |
1053 | } | |
1054 | EXPORT_SYMBOL(vhost_chr_write_iter); | |
1055 | ||
1056 | unsigned int vhost_chr_poll(struct file *file, struct vhost_dev *dev, | |
1057 | poll_table *wait) | |
1058 | { | |
1059 | unsigned int mask = 0; | |
1060 | ||
1061 | poll_wait(file, &dev->wait, wait); | |
1062 | ||
1063 | if (!list_empty(&dev->read_list)) | |
1064 | mask |= POLLIN | POLLRDNORM; | |
1065 | ||
1066 | return mask; | |
1067 | } | |
1068 | EXPORT_SYMBOL(vhost_chr_poll); | |
1069 | ||
1070 | ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to, | |
1071 | int noblock) | |
1072 | { | |
1073 | DEFINE_WAIT(wait); | |
1074 | struct vhost_msg_node *node; | |
1075 | ssize_t ret = 0; | |
1076 | unsigned size = sizeof(struct vhost_msg); | |
1077 | ||
1078 | if (iov_iter_count(to) < size) | |
1079 | return 0; | |
1080 | ||
1081 | while (1) { | |
1082 | if (!noblock) | |
1083 | prepare_to_wait(&dev->wait, &wait, | |
1084 | TASK_INTERRUPTIBLE); | |
1085 | ||
1086 | node = vhost_dequeue_msg(dev, &dev->read_list); | |
1087 | if (node) | |
1088 | break; | |
1089 | if (noblock) { | |
1090 | ret = -EAGAIN; | |
1091 | break; | |
1092 | } | |
1093 | if (signal_pending(current)) { | |
1094 | ret = -ERESTARTSYS; | |
1095 | break; | |
1096 | } | |
1097 | if (!dev->iotlb) { | |
1098 | ret = -EBADFD; | |
1099 | break; | |
1100 | } | |
1101 | ||
1102 | schedule(); | |
1103 | } | |
1104 | ||
1105 | if (!noblock) | |
1106 | finish_wait(&dev->wait, &wait); | |
1107 | ||
1108 | if (node) { | |
1109 | ret = copy_to_iter(&node->msg, size, to); | |
1110 | ||
1111 | if (ret != size || node->msg.type != VHOST_IOTLB_MISS) { | |
1112 | kfree(node); | |
1113 | return ret; | |
1114 | } | |
1115 | ||
1116 | vhost_enqueue_msg(dev, &dev->pending_list, node); | |
1117 | } | |
1118 | ||
1119 | return ret; | |
1120 | } | |
1121 | EXPORT_SYMBOL_GPL(vhost_chr_read_iter); | |
1122 | ||
1123 | static int vhost_iotlb_miss(struct vhost_virtqueue *vq, u64 iova, int access) | |
1124 | { | |
1125 | struct vhost_dev *dev = vq->dev; | |
1126 | struct vhost_msg_node *node; | |
1127 | struct vhost_iotlb_msg *msg; | |
1128 | ||
1129 | node = vhost_new_msg(vq, VHOST_IOTLB_MISS); | |
1130 | if (!node) | |
1131 | return -ENOMEM; | |
1132 | ||
1133 | msg = &node->msg.iotlb; | |
1134 | msg->type = VHOST_IOTLB_MISS; | |
1135 | msg->iova = iova; | |
1136 | msg->perm = access; | |
1137 | ||
1138 | vhost_enqueue_msg(dev, &dev->read_list, node); | |
1139 | ||
1140 | return 0; | |
bfe2bc51 JW |
1141 | } |
1142 | ||
ea16c514 | 1143 | static int vq_access_ok(struct vhost_virtqueue *vq, unsigned int num, |
3a4d5c94 MT |
1144 | struct vring_desc __user *desc, |
1145 | struct vring_avail __user *avail, | |
1146 | struct vring_used __user *used) | |
6b1e6cc7 | 1147 | |
3a4d5c94 | 1148 | { |
ea16c514 | 1149 | size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; |
6b1e6cc7 | 1150 | |
3a4d5c94 MT |
1151 | return access_ok(VERIFY_READ, desc, num * sizeof *desc) && |
1152 | access_ok(VERIFY_READ, avail, | |
8ea8cf89 | 1153 | sizeof *avail + num * sizeof *avail->ring + s) && |
3a4d5c94 | 1154 | access_ok(VERIFY_WRITE, used, |
8ea8cf89 | 1155 | sizeof *used + num * sizeof *used->ring + s); |
3a4d5c94 MT |
1156 | } |
1157 | ||
f8894913 JW |
1158 | static void vhost_vq_meta_update(struct vhost_virtqueue *vq, |
1159 | const struct vhost_umem_node *node, | |
1160 | int type) | |
1161 | { | |
1162 | int access = (type == VHOST_ADDR_USED) ? | |
1163 | VHOST_ACCESS_WO : VHOST_ACCESS_RO; | |
1164 | ||
1165 | if (likely(node->perm & access)) | |
1166 | vq->meta_iotlb[type] = node; | |
1167 | } | |
1168 | ||
6b1e6cc7 | 1169 | static int iotlb_access_ok(struct vhost_virtqueue *vq, |
f8894913 | 1170 | int access, u64 addr, u64 len, int type) |
6b1e6cc7 JW |
1171 | { |
1172 | const struct vhost_umem_node *node; | |
1173 | struct vhost_umem *umem = vq->iotlb; | |
ca2c5b33 | 1174 | u64 s = 0, size, orig_addr = addr, last = addr + len - 1; |
f8894913 JW |
1175 | |
1176 | if (vhost_vq_meta_fetch(vq, addr, len, type)) | |
1177 | return true; | |
6b1e6cc7 JW |
1178 | |
1179 | while (len > s) { | |
1180 | node = vhost_umem_interval_tree_iter_first(&umem->umem_tree, | |
1181 | addr, | |
ca2c5b33 | 1182 | last); |
6b1e6cc7 JW |
1183 | if (node == NULL || node->start > addr) { |
1184 | vhost_iotlb_miss(vq, addr, access); | |
1185 | return false; | |
1186 | } else if (!(node->perm & access)) { | |
1187 | /* Report the possible access violation by | |
1188 | * request another translation from userspace. | |
1189 | */ | |
1190 | return false; | |
1191 | } | |
1192 | ||
1193 | size = node->size - addr + node->start; | |
f8894913 JW |
1194 | |
1195 | if (orig_addr == addr && size >= len) | |
1196 | vhost_vq_meta_update(vq, node, type); | |
1197 | ||
6b1e6cc7 JW |
1198 | s += size; |
1199 | addr += size; | |
1200 | } | |
1201 | ||
1202 | return true; | |
1203 | } | |
1204 | ||
1205 | int vq_iotlb_prefetch(struct vhost_virtqueue *vq) | |
1206 | { | |
1207 | size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; | |
1208 | unsigned int num = vq->num; | |
1209 | ||
1210 | if (!vq->iotlb) | |
1211 | return 1; | |
1212 | ||
1213 | return iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->desc, | |
f8894913 | 1214 | num * sizeof(*vq->desc), VHOST_ADDR_DESC) && |
6b1e6cc7 JW |
1215 | iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->avail, |
1216 | sizeof *vq->avail + | |
f8894913 JW |
1217 | num * sizeof(*vq->avail->ring) + s, |
1218 | VHOST_ADDR_AVAIL) && | |
6b1e6cc7 JW |
1219 | iotlb_access_ok(vq, VHOST_ACCESS_WO, (u64)(uintptr_t)vq->used, |
1220 | sizeof *vq->used + | |
f8894913 JW |
1221 | num * sizeof(*vq->used->ring) + s, |
1222 | VHOST_ADDR_USED); | |
6b1e6cc7 JW |
1223 | } |
1224 | EXPORT_SYMBOL_GPL(vq_iotlb_prefetch); | |
1225 | ||
3a4d5c94 MT |
1226 | /* Can we log writes? */ |
1227 | /* Caller should have device mutex but not vq mutex */ | |
1228 | int vhost_log_access_ok(struct vhost_dev *dev) | |
1229 | { | |
a9709d68 | 1230 | return memory_access_ok(dev, dev->umem, 1); |
3a4d5c94 | 1231 | } |
6ac1afbf | 1232 | EXPORT_SYMBOL_GPL(vhost_log_access_ok); |
3a4d5c94 MT |
1233 | |
1234 | /* Verify access for write logging. */ | |
1235 | /* Caller should have vq mutex and device mutex */ | |
ea16c514 | 1236 | static int vq_log_access_ok(struct vhost_virtqueue *vq, |
8ea8cf89 | 1237 | void __user *log_base) |
3a4d5c94 | 1238 | { |
ea16c514 | 1239 | size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; |
28457ee6 | 1240 | |
a9709d68 | 1241 | return vq_memory_access_ok(log_base, vq->umem, |
ea16c514 | 1242 | vhost_has_feature(vq, VHOST_F_LOG_ALL)) && |
3a4d5c94 MT |
1243 | (!vq->log_used || log_access_ok(log_base, vq->log_addr, |
1244 | sizeof *vq->used + | |
8ea8cf89 | 1245 | vq->num * sizeof *vq->used->ring + s)); |
3a4d5c94 MT |
1246 | } |
1247 | ||
1248 | /* Can we start vq? */ | |
1249 | /* Caller should have vq mutex and device mutex */ | |
1250 | int vhost_vq_access_ok(struct vhost_virtqueue *vq) | |
1251 | { | |
6b1e6cc7 JW |
1252 | if (vq->iotlb) { |
1253 | /* When device IOTLB was used, the access validation | |
1254 | * will be validated during prefetching. | |
1255 | */ | |
1256 | return 1; | |
1257 | } | |
ea16c514 MT |
1258 | return vq_access_ok(vq, vq->num, vq->desc, vq->avail, vq->used) && |
1259 | vq_log_access_ok(vq, vq->log_base); | |
3a4d5c94 | 1260 | } |
6ac1afbf | 1261 | EXPORT_SYMBOL_GPL(vhost_vq_access_ok); |
3a4d5c94 | 1262 | |
6b1e6cc7 JW |
1263 | static struct vhost_umem *vhost_umem_alloc(void) |
1264 | { | |
6c5ab651 | 1265 | struct vhost_umem *umem = kvzalloc(sizeof(*umem), GFP_KERNEL); |
6b1e6cc7 JW |
1266 | |
1267 | if (!umem) | |
1268 | return NULL; | |
1269 | ||
f808c13f | 1270 | umem->umem_tree = RB_ROOT_CACHED; |
6b1e6cc7 JW |
1271 | umem->numem = 0; |
1272 | INIT_LIST_HEAD(&umem->umem_list); | |
1273 | ||
1274 | return umem; | |
1275 | } | |
1276 | ||
3a4d5c94 MT |
1277 | static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m) |
1278 | { | |
a9709d68 JW |
1279 | struct vhost_memory mem, *newmem; |
1280 | struct vhost_memory_region *region; | |
a9709d68 | 1281 | struct vhost_umem *newumem, *oldumem; |
3a4d5c94 | 1282 | unsigned long size = offsetof(struct vhost_memory, regions); |
98f9ca0a | 1283 | int i; |
d47effe1 | 1284 | |
7ad9c9d2 TY |
1285 | if (copy_from_user(&mem, m, size)) |
1286 | return -EFAULT; | |
3a4d5c94 MT |
1287 | if (mem.padding) |
1288 | return -EOPNOTSUPP; | |
c9ce42f7 | 1289 | if (mem.nregions > max_mem_regions) |
3a4d5c94 | 1290 | return -E2BIG; |
6c5ab651 | 1291 | newmem = kvzalloc(size + mem.nregions * sizeof(*m->regions), GFP_KERNEL); |
3a4d5c94 MT |
1292 | if (!newmem) |
1293 | return -ENOMEM; | |
1294 | ||
1295 | memcpy(newmem, &mem, size); | |
7ad9c9d2 TY |
1296 | if (copy_from_user(newmem->regions, m->regions, |
1297 | mem.nregions * sizeof *m->regions)) { | |
bcfeacab | 1298 | kvfree(newmem); |
7ad9c9d2 | 1299 | return -EFAULT; |
3a4d5c94 MT |
1300 | } |
1301 | ||
6b1e6cc7 | 1302 | newumem = vhost_umem_alloc(); |
a9709d68 | 1303 | if (!newumem) { |
4de7255f | 1304 | kvfree(newmem); |
a9709d68 JW |
1305 | return -ENOMEM; |
1306 | } | |
1307 | ||
a9709d68 JW |
1308 | for (region = newmem->regions; |
1309 | region < newmem->regions + mem.nregions; | |
1310 | region++) { | |
6b1e6cc7 JW |
1311 | if (vhost_new_umem_range(newumem, |
1312 | region->guest_phys_addr, | |
1313 | region->memory_size, | |
1314 | region->guest_phys_addr + | |
1315 | region->memory_size - 1, | |
1316 | region->userspace_addr, | |
1317 | VHOST_ACCESS_RW)) | |
a9709d68 | 1318 | goto err; |
a02c3789 | 1319 | } |
a9709d68 JW |
1320 | |
1321 | if (!memory_access_ok(d, newumem, 0)) | |
1322 | goto err; | |
1323 | ||
1324 | oldumem = d->umem; | |
1325 | d->umem = newumem; | |
98f9ca0a | 1326 | |
47283bef | 1327 | /* All memory accesses are done under some VQ mutex. */ |
98f9ca0a MT |
1328 | for (i = 0; i < d->nvqs; ++i) { |
1329 | mutex_lock(&d->vqs[i]->mutex); | |
a9709d68 | 1330 | d->vqs[i]->umem = newumem; |
98f9ca0a MT |
1331 | mutex_unlock(&d->vqs[i]->mutex); |
1332 | } | |
a9709d68 JW |
1333 | |
1334 | kvfree(newmem); | |
1335 | vhost_umem_clean(oldumem); | |
3a4d5c94 | 1336 | return 0; |
a9709d68 JW |
1337 | |
1338 | err: | |
1339 | vhost_umem_clean(newumem); | |
1340 | kvfree(newmem); | |
1341 | return -EFAULT; | |
3a4d5c94 MT |
1342 | } |
1343 | ||
935cdee7 | 1344 | long vhost_vring_ioctl(struct vhost_dev *d, int ioctl, void __user *argp) |
3a4d5c94 | 1345 | { |
cecb46f1 AV |
1346 | struct file *eventfp, *filep = NULL; |
1347 | bool pollstart = false, pollstop = false; | |
3a4d5c94 MT |
1348 | struct eventfd_ctx *ctx = NULL; |
1349 | u32 __user *idxp = argp; | |
1350 | struct vhost_virtqueue *vq; | |
1351 | struct vhost_vring_state s; | |
1352 | struct vhost_vring_file f; | |
1353 | struct vhost_vring_addr a; | |
1354 | u32 idx; | |
1355 | long r; | |
1356 | ||
1357 | r = get_user(idx, idxp); | |
1358 | if (r < 0) | |
1359 | return r; | |
0f3d9a17 | 1360 | if (idx >= d->nvqs) |
3a4d5c94 MT |
1361 | return -ENOBUFS; |
1362 | ||
3ab2e420 | 1363 | vq = d->vqs[idx]; |
3a4d5c94 MT |
1364 | |
1365 | mutex_lock(&vq->mutex); | |
1366 | ||
1367 | switch (ioctl) { | |
1368 | case VHOST_SET_VRING_NUM: | |
1369 | /* Resizing ring with an active backend? | |
1370 | * You don't want to do that. */ | |
1371 | if (vq->private_data) { | |
1372 | r = -EBUSY; | |
1373 | break; | |
1374 | } | |
7ad9c9d2 TY |
1375 | if (copy_from_user(&s, argp, sizeof s)) { |
1376 | r = -EFAULT; | |
3a4d5c94 | 1377 | break; |
7ad9c9d2 | 1378 | } |
3a4d5c94 MT |
1379 | if (!s.num || s.num > 0xffff || (s.num & (s.num - 1))) { |
1380 | r = -EINVAL; | |
1381 | break; | |
1382 | } | |
1383 | vq->num = s.num; | |
1384 | break; | |
1385 | case VHOST_SET_VRING_BASE: | |
1386 | /* Moving base with an active backend? | |
1387 | * You don't want to do that. */ | |
1388 | if (vq->private_data) { | |
1389 | r = -EBUSY; | |
1390 | break; | |
1391 | } | |
7ad9c9d2 TY |
1392 | if (copy_from_user(&s, argp, sizeof s)) { |
1393 | r = -EFAULT; | |
3a4d5c94 | 1394 | break; |
7ad9c9d2 | 1395 | } |
3a4d5c94 MT |
1396 | if (s.num > 0xffff) { |
1397 | r = -EINVAL; | |
1398 | break; | |
1399 | } | |
8d65843c | 1400 | vq->last_avail_idx = s.num; |
3a4d5c94 MT |
1401 | /* Forget the cached index value. */ |
1402 | vq->avail_idx = vq->last_avail_idx; | |
1403 | break; | |
1404 | case VHOST_GET_VRING_BASE: | |
1405 | s.index = idx; | |
1406 | s.num = vq->last_avail_idx; | |
7ad9c9d2 TY |
1407 | if (copy_to_user(argp, &s, sizeof s)) |
1408 | r = -EFAULT; | |
3a4d5c94 MT |
1409 | break; |
1410 | case VHOST_SET_VRING_ADDR: | |
7ad9c9d2 TY |
1411 | if (copy_from_user(&a, argp, sizeof a)) { |
1412 | r = -EFAULT; | |
3a4d5c94 | 1413 | break; |
7ad9c9d2 | 1414 | } |
3a4d5c94 MT |
1415 | if (a.flags & ~(0x1 << VHOST_VRING_F_LOG)) { |
1416 | r = -EOPNOTSUPP; | |
1417 | break; | |
1418 | } | |
1419 | /* For 32bit, verify that the top 32bits of the user | |
1420 | data are set to zero. */ | |
1421 | if ((u64)(unsigned long)a.desc_user_addr != a.desc_user_addr || | |
1422 | (u64)(unsigned long)a.used_user_addr != a.used_user_addr || | |
1423 | (u64)(unsigned long)a.avail_user_addr != a.avail_user_addr) { | |
1424 | r = -EFAULT; | |
1425 | break; | |
1426 | } | |
5d9a07b0 MT |
1427 | |
1428 | /* Make sure it's safe to cast pointers to vring types. */ | |
1429 | BUILD_BUG_ON(__alignof__ *vq->avail > VRING_AVAIL_ALIGN_SIZE); | |
1430 | BUILD_BUG_ON(__alignof__ *vq->used > VRING_USED_ALIGN_SIZE); | |
1431 | if ((a.avail_user_addr & (VRING_AVAIL_ALIGN_SIZE - 1)) || | |
1432 | (a.used_user_addr & (VRING_USED_ALIGN_SIZE - 1)) || | |
d5424838 | 1433 | (a.log_guest_addr & (VRING_USED_ALIGN_SIZE - 1))) { |
3a4d5c94 MT |
1434 | r = -EINVAL; |
1435 | break; | |
1436 | } | |
1437 | ||
1438 | /* We only verify access here if backend is configured. | |
1439 | * If it is not, we don't as size might not have been setup. | |
1440 | * We will verify when backend is configured. */ | |
1441 | if (vq->private_data) { | |
ea16c514 | 1442 | if (!vq_access_ok(vq, vq->num, |
3a4d5c94 MT |
1443 | (void __user *)(unsigned long)a.desc_user_addr, |
1444 | (void __user *)(unsigned long)a.avail_user_addr, | |
1445 | (void __user *)(unsigned long)a.used_user_addr)) { | |
1446 | r = -EINVAL; | |
1447 | break; | |
1448 | } | |
1449 | ||
1450 | /* Also validate log access for used ring if enabled. */ | |
1451 | if ((a.flags & (0x1 << VHOST_VRING_F_LOG)) && | |
1452 | !log_access_ok(vq->log_base, a.log_guest_addr, | |
1453 | sizeof *vq->used + | |
1454 | vq->num * sizeof *vq->used->ring)) { | |
1455 | r = -EINVAL; | |
1456 | break; | |
1457 | } | |
1458 | } | |
1459 | ||
3a4d5c94 MT |
1460 | vq->log_used = !!(a.flags & (0x1 << VHOST_VRING_F_LOG)); |
1461 | vq->desc = (void __user *)(unsigned long)a.desc_user_addr; | |
1462 | vq->avail = (void __user *)(unsigned long)a.avail_user_addr; | |
1463 | vq->log_addr = a.log_guest_addr; | |
1464 | vq->used = (void __user *)(unsigned long)a.used_user_addr; | |
1465 | break; | |
1466 | case VHOST_SET_VRING_KICK: | |
7ad9c9d2 TY |
1467 | if (copy_from_user(&f, argp, sizeof f)) { |
1468 | r = -EFAULT; | |
3a4d5c94 | 1469 | break; |
7ad9c9d2 | 1470 | } |
3a4d5c94 | 1471 | eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd); |
535297a6 MT |
1472 | if (IS_ERR(eventfp)) { |
1473 | r = PTR_ERR(eventfp); | |
1474 | break; | |
1475 | } | |
3a4d5c94 | 1476 | if (eventfp != vq->kick) { |
cecb46f1 AV |
1477 | pollstop = (filep = vq->kick) != NULL; |
1478 | pollstart = (vq->kick = eventfp) != NULL; | |
3a4d5c94 MT |
1479 | } else |
1480 | filep = eventfp; | |
1481 | break; | |
1482 | case VHOST_SET_VRING_CALL: | |
7ad9c9d2 TY |
1483 | if (copy_from_user(&f, argp, sizeof f)) { |
1484 | r = -EFAULT; | |
3a4d5c94 | 1485 | break; |
7ad9c9d2 | 1486 | } |
e050c7d9 EB |
1487 | ctx = f.fd == -1 ? NULL : eventfd_ctx_fdget(f.fd); |
1488 | if (IS_ERR(ctx)) { | |
1489 | r = PTR_ERR(ctx); | |
535297a6 MT |
1490 | break; |
1491 | } | |
e050c7d9 | 1492 | swap(ctx, vq->call_ctx); |
3a4d5c94 MT |
1493 | break; |
1494 | case VHOST_SET_VRING_ERR: | |
7ad9c9d2 TY |
1495 | if (copy_from_user(&f, argp, sizeof f)) { |
1496 | r = -EFAULT; | |
3a4d5c94 | 1497 | break; |
7ad9c9d2 | 1498 | } |
09f332a5 EB |
1499 | ctx = f.fd == -1 ? NULL : eventfd_ctx_fdget(f.fd); |
1500 | if (IS_ERR(ctx)) { | |
1501 | r = PTR_ERR(ctx); | |
535297a6 MT |
1502 | break; |
1503 | } | |
09f332a5 | 1504 | swap(ctx, vq->error_ctx); |
3a4d5c94 | 1505 | break; |
2751c988 GK |
1506 | case VHOST_SET_VRING_ENDIAN: |
1507 | r = vhost_set_vring_endian(vq, argp); | |
1508 | break; | |
1509 | case VHOST_GET_VRING_ENDIAN: | |
1510 | r = vhost_get_vring_endian(vq, idx, argp); | |
1511 | break; | |
03088137 JW |
1512 | case VHOST_SET_VRING_BUSYLOOP_TIMEOUT: |
1513 | if (copy_from_user(&s, argp, sizeof(s))) { | |
1514 | r = -EFAULT; | |
1515 | break; | |
1516 | } | |
1517 | vq->busyloop_timeout = s.num; | |
1518 | break; | |
1519 | case VHOST_GET_VRING_BUSYLOOP_TIMEOUT: | |
1520 | s.index = idx; | |
1521 | s.num = vq->busyloop_timeout; | |
1522 | if (copy_to_user(argp, &s, sizeof(s))) | |
1523 | r = -EFAULT; | |
1524 | break; | |
3a4d5c94 MT |
1525 | default: |
1526 | r = -ENOIOCTLCMD; | |
1527 | } | |
1528 | ||
1529 | if (pollstop && vq->handle_kick) | |
1530 | vhost_poll_stop(&vq->poll); | |
1531 | ||
e050c7d9 | 1532 | if (!IS_ERR_OR_NULL(ctx)) |
3a4d5c94 MT |
1533 | eventfd_ctx_put(ctx); |
1534 | if (filep) | |
1535 | fput(filep); | |
1536 | ||
1537 | if (pollstart && vq->handle_kick) | |
2b8b328b | 1538 | r = vhost_poll_start(&vq->poll, vq->kick); |
3a4d5c94 MT |
1539 | |
1540 | mutex_unlock(&vq->mutex); | |
1541 | ||
1542 | if (pollstop && vq->handle_kick) | |
1543 | vhost_poll_flush(&vq->poll); | |
1544 | return r; | |
1545 | } | |
6ac1afbf | 1546 | EXPORT_SYMBOL_GPL(vhost_vring_ioctl); |
3a4d5c94 | 1547 | |
6b1e6cc7 JW |
1548 | int vhost_init_device_iotlb(struct vhost_dev *d, bool enabled) |
1549 | { | |
1550 | struct vhost_umem *niotlb, *oiotlb; | |
1551 | int i; | |
1552 | ||
1553 | niotlb = vhost_umem_alloc(); | |
1554 | if (!niotlb) | |
1555 | return -ENOMEM; | |
1556 | ||
1557 | oiotlb = d->iotlb; | |
1558 | d->iotlb = niotlb; | |
1559 | ||
1560 | for (i = 0; i < d->nvqs; ++i) { | |
1561 | mutex_lock(&d->vqs[i]->mutex); | |
1562 | d->vqs[i]->iotlb = niotlb; | |
1563 | mutex_unlock(&d->vqs[i]->mutex); | |
1564 | } | |
1565 | ||
1566 | vhost_umem_clean(oiotlb); | |
1567 | ||
1568 | return 0; | |
1569 | } | |
1570 | EXPORT_SYMBOL_GPL(vhost_init_device_iotlb); | |
1571 | ||
3a4d5c94 | 1572 | /* Caller must have device mutex */ |
935cdee7 | 1573 | long vhost_dev_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp) |
3a4d5c94 | 1574 | { |
3a4d5c94 MT |
1575 | struct file *eventfp, *filep = NULL; |
1576 | struct eventfd_ctx *ctx = NULL; | |
1577 | u64 p; | |
1578 | long r; | |
1579 | int i, fd; | |
1580 | ||
1581 | /* If you are not the owner, you can become one */ | |
1582 | if (ioctl == VHOST_SET_OWNER) { | |
1583 | r = vhost_dev_set_owner(d); | |
1584 | goto done; | |
1585 | } | |
1586 | ||
1587 | /* You must be the owner to do anything else */ | |
1588 | r = vhost_dev_check_owner(d); | |
1589 | if (r) | |
1590 | goto done; | |
1591 | ||
1592 | switch (ioctl) { | |
1593 | case VHOST_SET_MEM_TABLE: | |
1594 | r = vhost_set_memory(d, argp); | |
1595 | break; | |
1596 | case VHOST_SET_LOG_BASE: | |
7ad9c9d2 TY |
1597 | if (copy_from_user(&p, argp, sizeof p)) { |
1598 | r = -EFAULT; | |
3a4d5c94 | 1599 | break; |
7ad9c9d2 | 1600 | } |
3a4d5c94 MT |
1601 | if ((u64)(unsigned long)p != p) { |
1602 | r = -EFAULT; | |
1603 | break; | |
1604 | } | |
1605 | for (i = 0; i < d->nvqs; ++i) { | |
1606 | struct vhost_virtqueue *vq; | |
1607 | void __user *base = (void __user *)(unsigned long)p; | |
3ab2e420 | 1608 | vq = d->vqs[i]; |
3a4d5c94 MT |
1609 | mutex_lock(&vq->mutex); |
1610 | /* If ring is inactive, will check when it's enabled. */ | |
ea16c514 | 1611 | if (vq->private_data && !vq_log_access_ok(vq, base)) |
3a4d5c94 MT |
1612 | r = -EFAULT; |
1613 | else | |
1614 | vq->log_base = base; | |
1615 | mutex_unlock(&vq->mutex); | |
1616 | } | |
1617 | break; | |
1618 | case VHOST_SET_LOG_FD: | |
1619 | r = get_user(fd, (int __user *)argp); | |
1620 | if (r < 0) | |
1621 | break; | |
1622 | eventfp = fd == -1 ? NULL : eventfd_fget(fd); | |
1623 | if (IS_ERR(eventfp)) { | |
1624 | r = PTR_ERR(eventfp); | |
1625 | break; | |
1626 | } | |
1627 | if (eventfp != d->log_file) { | |
1628 | filep = d->log_file; | |
7932c0bd | 1629 | d->log_file = eventfp; |
3a4d5c94 MT |
1630 | ctx = d->log_ctx; |
1631 | d->log_ctx = eventfp ? | |
1632 | eventfd_ctx_fileget(eventfp) : NULL; | |
1633 | } else | |
1634 | filep = eventfp; | |
1635 | for (i = 0; i < d->nvqs; ++i) { | |
3ab2e420 AH |
1636 | mutex_lock(&d->vqs[i]->mutex); |
1637 | d->vqs[i]->log_ctx = d->log_ctx; | |
1638 | mutex_unlock(&d->vqs[i]->mutex); | |
3a4d5c94 MT |
1639 | } |
1640 | if (ctx) | |
1641 | eventfd_ctx_put(ctx); | |
1642 | if (filep) | |
1643 | fput(filep); | |
1644 | break; | |
1645 | default: | |
935cdee7 | 1646 | r = -ENOIOCTLCMD; |
3a4d5c94 MT |
1647 | break; |
1648 | } | |
1649 | done: | |
1650 | return r; | |
1651 | } | |
6ac1afbf | 1652 | EXPORT_SYMBOL_GPL(vhost_dev_ioctl); |
3a4d5c94 | 1653 | |
3a4d5c94 MT |
1654 | /* TODO: This is really inefficient. We need something like get_user() |
1655 | * (instruction directly accesses the data, with an exception table entry | |
1656 | * returning -EFAULT). See Documentation/x86/exception-tables.txt. | |
1657 | */ | |
1658 | static int set_bit_to_user(int nr, void __user *addr) | |
1659 | { | |
1660 | unsigned long log = (unsigned long)addr; | |
1661 | struct page *page; | |
1662 | void *base; | |
1663 | int bit = nr + (log % PAGE_SIZE) * 8; | |
1664 | int r; | |
d47effe1 | 1665 | |
3a4d5c94 | 1666 | r = get_user_pages_fast(log, 1, 1, &page); |
d6db3f5c | 1667 | if (r < 0) |
3a4d5c94 | 1668 | return r; |
d6db3f5c | 1669 | BUG_ON(r != 1); |
c6daa7ff | 1670 | base = kmap_atomic(page); |
3a4d5c94 | 1671 | set_bit(bit, base); |
c6daa7ff | 1672 | kunmap_atomic(base); |
3a4d5c94 MT |
1673 | set_page_dirty_lock(page); |
1674 | put_page(page); | |
1675 | return 0; | |
1676 | } | |
1677 | ||
1678 | static int log_write(void __user *log_base, | |
1679 | u64 write_address, u64 write_length) | |
1680 | { | |
28831ee6 | 1681 | u64 write_page = write_address / VHOST_PAGE_SIZE; |
3a4d5c94 | 1682 | int r; |
d47effe1 | 1683 | |
3a4d5c94 MT |
1684 | if (!write_length) |
1685 | return 0; | |
3bf9be40 | 1686 | write_length += write_address % VHOST_PAGE_SIZE; |
3a4d5c94 MT |
1687 | for (;;) { |
1688 | u64 base = (u64)(unsigned long)log_base; | |
28831ee6 MT |
1689 | u64 log = base + write_page / 8; |
1690 | int bit = write_page % 8; | |
3a4d5c94 MT |
1691 | if ((u64)(unsigned long)log != log) |
1692 | return -EFAULT; | |
1693 | r = set_bit_to_user(bit, (void __user *)(unsigned long)log); | |
1694 | if (r < 0) | |
1695 | return r; | |
1696 | if (write_length <= VHOST_PAGE_SIZE) | |
1697 | break; | |
1698 | write_length -= VHOST_PAGE_SIZE; | |
28831ee6 | 1699 | write_page += 1; |
3a4d5c94 MT |
1700 | } |
1701 | return r; | |
1702 | } | |
1703 | ||
1704 | int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log, | |
1705 | unsigned int log_num, u64 len) | |
1706 | { | |
1707 | int i, r; | |
1708 | ||
1709 | /* Make sure data written is seen before log. */ | |
5659338c | 1710 | smp_wmb(); |
3a4d5c94 MT |
1711 | for (i = 0; i < log_num; ++i) { |
1712 | u64 l = min(log[i].len, len); | |
1713 | r = log_write(vq->log_base, log[i].addr, l); | |
1714 | if (r < 0) | |
1715 | return r; | |
1716 | len -= l; | |
5786aee8 MT |
1717 | if (!len) { |
1718 | if (vq->log_ctx) | |
1719 | eventfd_signal(vq->log_ctx, 1); | |
3a4d5c94 | 1720 | return 0; |
5786aee8 | 1721 | } |
3a4d5c94 | 1722 | } |
3a4d5c94 MT |
1723 | /* Length written exceeds what we have stored. This is a bug. */ |
1724 | BUG(); | |
1725 | return 0; | |
1726 | } | |
6ac1afbf | 1727 | EXPORT_SYMBOL_GPL(vhost_log_write); |
3a4d5c94 | 1728 | |
2723feaa JW |
1729 | static int vhost_update_used_flags(struct vhost_virtqueue *vq) |
1730 | { | |
1731 | void __user *used; | |
bfe2bc51 JW |
1732 | if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->used_flags), |
1733 | &vq->used->flags) < 0) | |
2723feaa JW |
1734 | return -EFAULT; |
1735 | if (unlikely(vq->log_used)) { | |
1736 | /* Make sure the flag is seen before log. */ | |
1737 | smp_wmb(); | |
1738 | /* Log used flag write. */ | |
1739 | used = &vq->used->flags; | |
1740 | log_write(vq->log_base, vq->log_addr + | |
1741 | (used - (void __user *)vq->used), | |
1742 | sizeof vq->used->flags); | |
1743 | if (vq->log_ctx) | |
1744 | eventfd_signal(vq->log_ctx, 1); | |
1745 | } | |
1746 | return 0; | |
1747 | } | |
1748 | ||
1749 | static int vhost_update_avail_event(struct vhost_virtqueue *vq, u16 avail_event) | |
1750 | { | |
bfe2bc51 JW |
1751 | if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->avail_idx), |
1752 | vhost_avail_event(vq))) | |
2723feaa JW |
1753 | return -EFAULT; |
1754 | if (unlikely(vq->log_used)) { | |
1755 | void __user *used; | |
1756 | /* Make sure the event is seen before log. */ | |
1757 | smp_wmb(); | |
1758 | /* Log avail event write */ | |
1759 | used = vhost_avail_event(vq); | |
1760 | log_write(vq->log_base, vq->log_addr + | |
1761 | (used - (void __user *)vq->used), | |
1762 | sizeof *vhost_avail_event(vq)); | |
1763 | if (vq->log_ctx) | |
1764 | eventfd_signal(vq->log_ctx, 1); | |
1765 | } | |
1766 | return 0; | |
1767 | } | |
1768 | ||
80f7d030 | 1769 | int vhost_vq_init_access(struct vhost_virtqueue *vq) |
2723feaa | 1770 | { |
3b1bbe89 | 1771 | __virtio16 last_used_idx; |
2723feaa | 1772 | int r; |
e1f33be9 GK |
1773 | bool is_le = vq->is_le; |
1774 | ||
cda8bba0 | 1775 | if (!vq->private_data) |
2723feaa | 1776 | return 0; |
2751c988 GK |
1777 | |
1778 | vhost_init_is_le(vq); | |
2723feaa JW |
1779 | |
1780 | r = vhost_update_used_flags(vq); | |
1781 | if (r) | |
e1f33be9 | 1782 | goto err; |
2723feaa | 1783 | vq->signalled_used_valid = false; |
6b1e6cc7 JW |
1784 | if (!vq->iotlb && |
1785 | !access_ok(VERIFY_READ, &vq->used->idx, sizeof vq->used->idx)) { | |
e1f33be9 GK |
1786 | r = -EFAULT; |
1787 | goto err; | |
1788 | } | |
f8894913 | 1789 | r = vhost_get_used(vq, last_used_idx, &vq->used->idx); |
6b1e6cc7 JW |
1790 | if (r) { |
1791 | vq_err(vq, "Can't access used idx at %p\n", | |
1792 | &vq->used->idx); | |
e1f33be9 | 1793 | goto err; |
6b1e6cc7 | 1794 | } |
3b1bbe89 | 1795 | vq->last_used_idx = vhost16_to_cpu(vq, last_used_idx); |
64f7f051 | 1796 | return 0; |
6b1e6cc7 | 1797 | |
e1f33be9 GK |
1798 | err: |
1799 | vq->is_le = is_le; | |
1800 | return r; | |
2723feaa | 1801 | } |
80f7d030 | 1802 | EXPORT_SYMBOL_GPL(vhost_vq_init_access); |
2723feaa | 1803 | |
47283bef | 1804 | static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len, |
6b1e6cc7 | 1805 | struct iovec iov[], int iov_size, int access) |
3a4d5c94 | 1806 | { |
a9709d68 | 1807 | const struct vhost_umem_node *node; |
6b1e6cc7 JW |
1808 | struct vhost_dev *dev = vq->dev; |
1809 | struct vhost_umem *umem = dev->iotlb ? dev->iotlb : dev->umem; | |
3a4d5c94 MT |
1810 | struct iovec *_iov; |
1811 | u64 s = 0; | |
1812 | int ret = 0; | |
1813 | ||
3a4d5c94 MT |
1814 | while ((u64)len > s) { |
1815 | u64 size; | |
7b3384fc | 1816 | if (unlikely(ret >= iov_size)) { |
3a4d5c94 MT |
1817 | ret = -ENOBUFS; |
1818 | break; | |
1819 | } | |
6b1e6cc7 | 1820 | |
a9709d68 JW |
1821 | node = vhost_umem_interval_tree_iter_first(&umem->umem_tree, |
1822 | addr, addr + len - 1); | |
1823 | if (node == NULL || node->start > addr) { | |
6b1e6cc7 JW |
1824 | if (umem != dev->iotlb) { |
1825 | ret = -EFAULT; | |
1826 | break; | |
1827 | } | |
1828 | ret = -EAGAIN; | |
1829 | break; | |
1830 | } else if (!(node->perm & access)) { | |
1831 | ret = -EPERM; | |
3a4d5c94 MT |
1832 | break; |
1833 | } | |
6b1e6cc7 | 1834 | |
3a4d5c94 | 1835 | _iov = iov + ret; |
a9709d68 | 1836 | size = node->size - addr + node->start; |
bd97120f | 1837 | _iov->iov_len = min((u64)len - s, size); |
a8d3782f | 1838 | _iov->iov_base = (void __user *)(unsigned long) |
a9709d68 | 1839 | (node->userspace_addr + addr - node->start); |
3a4d5c94 MT |
1840 | s += size; |
1841 | addr += size; | |
1842 | ++ret; | |
1843 | } | |
1844 | ||
6b1e6cc7 JW |
1845 | if (ret == -EAGAIN) |
1846 | vhost_iotlb_miss(vq, addr, access); | |
3a4d5c94 MT |
1847 | return ret; |
1848 | } | |
1849 | ||
1850 | /* Each buffer in the virtqueues is actually a chain of descriptors. This | |
1851 | * function returns the next descriptor in the chain, | |
1852 | * or -1U if we're at the end. */ | |
3b1bbe89 | 1853 | static unsigned next_desc(struct vhost_virtqueue *vq, struct vring_desc *desc) |
3a4d5c94 MT |
1854 | { |
1855 | unsigned int next; | |
1856 | ||
1857 | /* If this descriptor says it doesn't chain, we're done. */ | |
3b1bbe89 | 1858 | if (!(desc->flags & cpu_to_vhost16(vq, VRING_DESC_F_NEXT))) |
3a4d5c94 MT |
1859 | return -1U; |
1860 | ||
1861 | /* Check they're not leading us off end of descriptors. */ | |
3b1bbe89 | 1862 | next = vhost16_to_cpu(vq, desc->next); |
3a4d5c94 MT |
1863 | /* Make sure compiler knows to grab that: we don't want it changing! */ |
1864 | /* We will use the result as an index in an array, so most | |
1865 | * architectures only need a compiler barrier here. */ | |
1866 | read_barrier_depends(); | |
1867 | ||
1868 | return next; | |
1869 | } | |
1870 | ||
47283bef | 1871 | static int get_indirect(struct vhost_virtqueue *vq, |
7b3384fc MT |
1872 | struct iovec iov[], unsigned int iov_size, |
1873 | unsigned int *out_num, unsigned int *in_num, | |
1874 | struct vhost_log *log, unsigned int *log_num, | |
1875 | struct vring_desc *indirect) | |
3a4d5c94 MT |
1876 | { |
1877 | struct vring_desc desc; | |
1878 | unsigned int i = 0, count, found = 0; | |
3b1bbe89 | 1879 | u32 len = vhost32_to_cpu(vq, indirect->len); |
aad9a1ce | 1880 | struct iov_iter from; |
6b1e6cc7 | 1881 | int ret, access; |
3a4d5c94 MT |
1882 | |
1883 | /* Sanity check */ | |
3b1bbe89 | 1884 | if (unlikely(len % sizeof desc)) { |
3a4d5c94 MT |
1885 | vq_err(vq, "Invalid length in indirect descriptor: " |
1886 | "len 0x%llx not multiple of 0x%zx\n", | |
3b1bbe89 | 1887 | (unsigned long long)len, |
3a4d5c94 MT |
1888 | sizeof desc); |
1889 | return -EINVAL; | |
1890 | } | |
1891 | ||
3b1bbe89 | 1892 | ret = translate_desc(vq, vhost64_to_cpu(vq, indirect->addr), len, vq->indirect, |
6b1e6cc7 | 1893 | UIO_MAXIOV, VHOST_ACCESS_RO); |
7b3384fc | 1894 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
1895 | if (ret != -EAGAIN) |
1896 | vq_err(vq, "Translation failure %d in indirect.\n", ret); | |
3a4d5c94 MT |
1897 | return ret; |
1898 | } | |
aad9a1ce | 1899 | iov_iter_init(&from, READ, vq->indirect, ret, len); |
3a4d5c94 MT |
1900 | |
1901 | /* We will use the result as an address to read from, so most | |
1902 | * architectures only need a compiler barrier here. */ | |
1903 | read_barrier_depends(); | |
1904 | ||
3b1bbe89 | 1905 | count = len / sizeof desc; |
3a4d5c94 MT |
1906 | /* Buffers are chained via a 16 bit next field, so |
1907 | * we can have at most 2^16 of these. */ | |
7b3384fc | 1908 | if (unlikely(count > USHRT_MAX + 1)) { |
3a4d5c94 MT |
1909 | vq_err(vq, "Indirect buffer length too big: %d\n", |
1910 | indirect->len); | |
1911 | return -E2BIG; | |
1912 | } | |
1913 | ||
1914 | do { | |
1915 | unsigned iov_count = *in_num + *out_num; | |
7b3384fc | 1916 | if (unlikely(++found > count)) { |
3a4d5c94 MT |
1917 | vq_err(vq, "Loop detected: last one at %u " |
1918 | "indirect size %u\n", | |
1919 | i, count); | |
1920 | return -EINVAL; | |
1921 | } | |
cbbd26b8 | 1922 | if (unlikely(!copy_from_iter_full(&desc, sizeof(desc), &from))) { |
3a4d5c94 | 1923 | vq_err(vq, "Failed indirect descriptor: idx %d, %zx\n", |
3b1bbe89 | 1924 | i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc); |
3a4d5c94 MT |
1925 | return -EINVAL; |
1926 | } | |
3b1bbe89 | 1927 | if (unlikely(desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT))) { |
3a4d5c94 | 1928 | vq_err(vq, "Nested indirect descriptor: idx %d, %zx\n", |
3b1bbe89 | 1929 | i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc); |
3a4d5c94 MT |
1930 | return -EINVAL; |
1931 | } | |
1932 | ||
6b1e6cc7 JW |
1933 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE)) |
1934 | access = VHOST_ACCESS_WO; | |
1935 | else | |
1936 | access = VHOST_ACCESS_RO; | |
1937 | ||
3b1bbe89 MT |
1938 | ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr), |
1939 | vhost32_to_cpu(vq, desc.len), iov + iov_count, | |
6b1e6cc7 | 1940 | iov_size - iov_count, access); |
7b3384fc | 1941 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
1942 | if (ret != -EAGAIN) |
1943 | vq_err(vq, "Translation failure %d indirect idx %d\n", | |
1944 | ret, i); | |
3a4d5c94 MT |
1945 | return ret; |
1946 | } | |
1947 | /* If this is an input descriptor, increment that count. */ | |
6b1e6cc7 | 1948 | if (access == VHOST_ACCESS_WO) { |
3a4d5c94 MT |
1949 | *in_num += ret; |
1950 | if (unlikely(log)) { | |
3b1bbe89 MT |
1951 | log[*log_num].addr = vhost64_to_cpu(vq, desc.addr); |
1952 | log[*log_num].len = vhost32_to_cpu(vq, desc.len); | |
3a4d5c94 MT |
1953 | ++*log_num; |
1954 | } | |
1955 | } else { | |
1956 | /* If it's an output descriptor, they're all supposed | |
1957 | * to come before any input descriptors. */ | |
7b3384fc | 1958 | if (unlikely(*in_num)) { |
3a4d5c94 MT |
1959 | vq_err(vq, "Indirect descriptor " |
1960 | "has out after in: idx %d\n", i); | |
1961 | return -EINVAL; | |
1962 | } | |
1963 | *out_num += ret; | |
1964 | } | |
3b1bbe89 | 1965 | } while ((i = next_desc(vq, &desc)) != -1); |
3a4d5c94 MT |
1966 | return 0; |
1967 | } | |
1968 | ||
1969 | /* This looks in the virtqueue and for the first available buffer, and converts | |
1970 | * it to an iovec for convenient access. Since descriptors consist of some | |
1971 | * number of output then some number of input descriptors, it's actually two | |
1972 | * iovecs, but we pack them into one and note how many of each there were. | |
1973 | * | |
d5675bd2 MT |
1974 | * This function returns the descriptor number found, or vq->num (which is |
1975 | * never a valid descriptor number) if none was found. A negative code is | |
1976 | * returned on error. */ | |
47283bef | 1977 | int vhost_get_vq_desc(struct vhost_virtqueue *vq, |
d5675bd2 MT |
1978 | struct iovec iov[], unsigned int iov_size, |
1979 | unsigned int *out_num, unsigned int *in_num, | |
1980 | struct vhost_log *log, unsigned int *log_num) | |
3a4d5c94 MT |
1981 | { |
1982 | struct vring_desc desc; | |
1983 | unsigned int i, head, found = 0; | |
1984 | u16 last_avail_idx; | |
3b1bbe89 MT |
1985 | __virtio16 avail_idx; |
1986 | __virtio16 ring_head; | |
6b1e6cc7 | 1987 | int ret, access; |
3a4d5c94 MT |
1988 | |
1989 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
1990 | last_avail_idx = vq->last_avail_idx; | |
3a4d5c94 | 1991 | |
e3b56cdd | 1992 | if (vq->avail_idx == vq->last_avail_idx) { |
f8894913 | 1993 | if (unlikely(vhost_get_avail(vq, avail_idx, &vq->avail->idx))) { |
e3b56cdd JW |
1994 | vq_err(vq, "Failed to access avail idx at %p\n", |
1995 | &vq->avail->idx); | |
1996 | return -EFAULT; | |
1997 | } | |
1998 | vq->avail_idx = vhost16_to_cpu(vq, avail_idx); | |
3a4d5c94 | 1999 | |
e3b56cdd JW |
2000 | if (unlikely((u16)(vq->avail_idx - last_avail_idx) > vq->num)) { |
2001 | vq_err(vq, "Guest moved used index from %u to %u", | |
2002 | last_avail_idx, vq->avail_idx); | |
2003 | return -EFAULT; | |
2004 | } | |
2005 | ||
2006 | /* If there's nothing new since last we looked, return | |
2007 | * invalid. | |
2008 | */ | |
2009 | if (vq->avail_idx == last_avail_idx) | |
2010 | return vq->num; | |
3a4d5c94 | 2011 | |
e3b56cdd JW |
2012 | /* Only get avail ring entries after they have been |
2013 | * exposed by guest. | |
2014 | */ | |
2015 | smp_rmb(); | |
2016 | } | |
3a4d5c94 MT |
2017 | |
2018 | /* Grab the next descriptor number they're advertising, and increment | |
2019 | * the index we've seen. */ | |
f8894913 | 2020 | if (unlikely(vhost_get_avail(vq, ring_head, |
bfe2bc51 | 2021 | &vq->avail->ring[last_avail_idx & (vq->num - 1)]))) { |
3a4d5c94 MT |
2022 | vq_err(vq, "Failed to read head: idx %d address %p\n", |
2023 | last_avail_idx, | |
2024 | &vq->avail->ring[last_avail_idx % vq->num]); | |
d5675bd2 | 2025 | return -EFAULT; |
3a4d5c94 MT |
2026 | } |
2027 | ||
3b1bbe89 MT |
2028 | head = vhost16_to_cpu(vq, ring_head); |
2029 | ||
3a4d5c94 | 2030 | /* If their number is silly, that's an error. */ |
7b3384fc | 2031 | if (unlikely(head >= vq->num)) { |
3a4d5c94 MT |
2032 | vq_err(vq, "Guest says index %u > %u is available", |
2033 | head, vq->num); | |
d5675bd2 | 2034 | return -EINVAL; |
3a4d5c94 MT |
2035 | } |
2036 | ||
2037 | /* When we start there are none of either input nor output. */ | |
2038 | *out_num = *in_num = 0; | |
2039 | if (unlikely(log)) | |
2040 | *log_num = 0; | |
2041 | ||
2042 | i = head; | |
2043 | do { | |
2044 | unsigned iov_count = *in_num + *out_num; | |
7b3384fc | 2045 | if (unlikely(i >= vq->num)) { |
3a4d5c94 MT |
2046 | vq_err(vq, "Desc index is %u > %u, head = %u", |
2047 | i, vq->num, head); | |
d5675bd2 | 2048 | return -EINVAL; |
3a4d5c94 | 2049 | } |
7b3384fc | 2050 | if (unlikely(++found > vq->num)) { |
3a4d5c94 MT |
2051 | vq_err(vq, "Loop detected: last one at %u " |
2052 | "vq size %u head %u\n", | |
2053 | i, vq->num, head); | |
d5675bd2 | 2054 | return -EINVAL; |
3a4d5c94 | 2055 | } |
bfe2bc51 JW |
2056 | ret = vhost_copy_from_user(vq, &desc, vq->desc + i, |
2057 | sizeof desc); | |
7b3384fc | 2058 | if (unlikely(ret)) { |
3a4d5c94 MT |
2059 | vq_err(vq, "Failed to get descriptor: idx %d addr %p\n", |
2060 | i, vq->desc + i); | |
d5675bd2 | 2061 | return -EFAULT; |
3a4d5c94 | 2062 | } |
3b1bbe89 | 2063 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT)) { |
47283bef | 2064 | ret = get_indirect(vq, iov, iov_size, |
3a4d5c94 MT |
2065 | out_num, in_num, |
2066 | log, log_num, &desc); | |
7b3384fc | 2067 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
2068 | if (ret != -EAGAIN) |
2069 | vq_err(vq, "Failure detected " | |
2070 | "in indirect descriptor at idx %d\n", i); | |
d5675bd2 | 2071 | return ret; |
3a4d5c94 MT |
2072 | } |
2073 | continue; | |
2074 | } | |
2075 | ||
6b1e6cc7 JW |
2076 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE)) |
2077 | access = VHOST_ACCESS_WO; | |
2078 | else | |
2079 | access = VHOST_ACCESS_RO; | |
3b1bbe89 MT |
2080 | ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr), |
2081 | vhost32_to_cpu(vq, desc.len), iov + iov_count, | |
6b1e6cc7 | 2082 | iov_size - iov_count, access); |
7b3384fc | 2083 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
2084 | if (ret != -EAGAIN) |
2085 | vq_err(vq, "Translation failure %d descriptor idx %d\n", | |
2086 | ret, i); | |
d5675bd2 | 2087 | return ret; |
3a4d5c94 | 2088 | } |
6b1e6cc7 | 2089 | if (access == VHOST_ACCESS_WO) { |
3a4d5c94 MT |
2090 | /* If this is an input descriptor, |
2091 | * increment that count. */ | |
2092 | *in_num += ret; | |
2093 | if (unlikely(log)) { | |
3b1bbe89 MT |
2094 | log[*log_num].addr = vhost64_to_cpu(vq, desc.addr); |
2095 | log[*log_num].len = vhost32_to_cpu(vq, desc.len); | |
3a4d5c94 MT |
2096 | ++*log_num; |
2097 | } | |
2098 | } else { | |
2099 | /* If it's an output descriptor, they're all supposed | |
2100 | * to come before any input descriptors. */ | |
7b3384fc | 2101 | if (unlikely(*in_num)) { |
3a4d5c94 MT |
2102 | vq_err(vq, "Descriptor has out after in: " |
2103 | "idx %d\n", i); | |
d5675bd2 | 2104 | return -EINVAL; |
3a4d5c94 MT |
2105 | } |
2106 | *out_num += ret; | |
2107 | } | |
3b1bbe89 | 2108 | } while ((i = next_desc(vq, &desc)) != -1); |
3a4d5c94 MT |
2109 | |
2110 | /* On success, increment avail index. */ | |
2111 | vq->last_avail_idx++; | |
8ea8cf89 MT |
2112 | |
2113 | /* Assume notifications from guest are disabled at this point, | |
2114 | * if they aren't we would need to update avail_event index. */ | |
2115 | BUG_ON(!(vq->used_flags & VRING_USED_F_NO_NOTIFY)); | |
3a4d5c94 MT |
2116 | return head; |
2117 | } | |
6ac1afbf | 2118 | EXPORT_SYMBOL_GPL(vhost_get_vq_desc); |
3a4d5c94 MT |
2119 | |
2120 | /* Reverse the effect of vhost_get_vq_desc. Useful for error handling. */ | |
8dd014ad | 2121 | void vhost_discard_vq_desc(struct vhost_virtqueue *vq, int n) |
3a4d5c94 | 2122 | { |
8dd014ad | 2123 | vq->last_avail_idx -= n; |
3a4d5c94 | 2124 | } |
6ac1afbf | 2125 | EXPORT_SYMBOL_GPL(vhost_discard_vq_desc); |
3a4d5c94 MT |
2126 | |
2127 | /* After we've used one of their buffers, we tell them about it. We'll then | |
2128 | * want to notify the guest, using eventfd. */ | |
2129 | int vhost_add_used(struct vhost_virtqueue *vq, unsigned int head, int len) | |
2130 | { | |
3b1bbe89 MT |
2131 | struct vring_used_elem heads = { |
2132 | cpu_to_vhost32(vq, head), | |
2133 | cpu_to_vhost32(vq, len) | |
2134 | }; | |
3a4d5c94 | 2135 | |
c49e4e57 | 2136 | return vhost_add_used_n(vq, &heads, 1); |
3a4d5c94 | 2137 | } |
6ac1afbf | 2138 | EXPORT_SYMBOL_GPL(vhost_add_used); |
3a4d5c94 | 2139 | |
8dd014ad DS |
2140 | static int __vhost_add_used_n(struct vhost_virtqueue *vq, |
2141 | struct vring_used_elem *heads, | |
2142 | unsigned count) | |
2143 | { | |
2144 | struct vring_used_elem __user *used; | |
8ea8cf89 | 2145 | u16 old, new; |
8dd014ad DS |
2146 | int start; |
2147 | ||
5fba13b5 | 2148 | start = vq->last_used_idx & (vq->num - 1); |
8dd014ad | 2149 | used = vq->used->ring + start; |
c49e4e57 | 2150 | if (count == 1) { |
bfe2bc51 | 2151 | if (vhost_put_user(vq, heads[0].id, &used->id)) { |
c49e4e57 JW |
2152 | vq_err(vq, "Failed to write used id"); |
2153 | return -EFAULT; | |
2154 | } | |
bfe2bc51 | 2155 | if (vhost_put_user(vq, heads[0].len, &used->len)) { |
c49e4e57 JW |
2156 | vq_err(vq, "Failed to write used len"); |
2157 | return -EFAULT; | |
2158 | } | |
bfe2bc51 | 2159 | } else if (vhost_copy_to_user(vq, used, heads, count * sizeof *used)) { |
8dd014ad DS |
2160 | vq_err(vq, "Failed to write used"); |
2161 | return -EFAULT; | |
2162 | } | |
2163 | if (unlikely(vq->log_used)) { | |
2164 | /* Make sure data is seen before log. */ | |
2165 | smp_wmb(); | |
2166 | /* Log used ring entry write. */ | |
2167 | log_write(vq->log_base, | |
2168 | vq->log_addr + | |
2169 | ((void __user *)used - (void __user *)vq->used), | |
2170 | count * sizeof *used); | |
2171 | } | |
8ea8cf89 MT |
2172 | old = vq->last_used_idx; |
2173 | new = (vq->last_used_idx += count); | |
2174 | /* If the driver never bothers to signal in a very long while, | |
2175 | * used index might wrap around. If that happens, invalidate | |
2176 | * signalled_used index we stored. TODO: make sure driver | |
2177 | * signals at least once in 2^16 and remove this. */ | |
2178 | if (unlikely((u16)(new - vq->signalled_used) < (u16)(new - old))) | |
2179 | vq->signalled_used_valid = false; | |
8dd014ad DS |
2180 | return 0; |
2181 | } | |
2182 | ||
2183 | /* After we've used one of their buffers, we tell them about it. We'll then | |
2184 | * want to notify the guest, using eventfd. */ | |
2185 | int vhost_add_used_n(struct vhost_virtqueue *vq, struct vring_used_elem *heads, | |
2186 | unsigned count) | |
2187 | { | |
2188 | int start, n, r; | |
2189 | ||
5fba13b5 | 2190 | start = vq->last_used_idx & (vq->num - 1); |
8dd014ad DS |
2191 | n = vq->num - start; |
2192 | if (n < count) { | |
2193 | r = __vhost_add_used_n(vq, heads, n); | |
2194 | if (r < 0) | |
2195 | return r; | |
2196 | heads += n; | |
2197 | count -= n; | |
2198 | } | |
2199 | r = __vhost_add_used_n(vq, heads, count); | |
2200 | ||
2201 | /* Make sure buffer is written before we update index. */ | |
2202 | smp_wmb(); | |
bfe2bc51 JW |
2203 | if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->last_used_idx), |
2204 | &vq->used->idx)) { | |
8dd014ad DS |
2205 | vq_err(vq, "Failed to increment used idx"); |
2206 | return -EFAULT; | |
2207 | } | |
2208 | if (unlikely(vq->log_used)) { | |
2209 | /* Log used index update. */ | |
2210 | log_write(vq->log_base, | |
2211 | vq->log_addr + offsetof(struct vring_used, idx), | |
2212 | sizeof vq->used->idx); | |
2213 | if (vq->log_ctx) | |
2214 | eventfd_signal(vq->log_ctx, 1); | |
2215 | } | |
2216 | return r; | |
2217 | } | |
6ac1afbf | 2218 | EXPORT_SYMBOL_GPL(vhost_add_used_n); |
8dd014ad | 2219 | |
8ea8cf89 | 2220 | static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 | 2221 | { |
3b1bbe89 MT |
2222 | __u16 old, new; |
2223 | __virtio16 event; | |
8ea8cf89 | 2224 | bool v; |
8d65843c JW |
2225 | /* Flush out used index updates. This is paired |
2226 | * with the barrier that the Guest executes when enabling | |
2227 | * interrupts. */ | |
2228 | smp_mb(); | |
0d499356 | 2229 | |
ea16c514 | 2230 | if (vhost_has_feature(vq, VIRTIO_F_NOTIFY_ON_EMPTY) && |
8ea8cf89 MT |
2231 | unlikely(vq->avail_idx == vq->last_avail_idx)) |
2232 | return true; | |
2233 | ||
ea16c514 | 2234 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
3b1bbe89 | 2235 | __virtio16 flags; |
f8894913 | 2236 | if (vhost_get_avail(vq, flags, &vq->avail->flags)) { |
8ea8cf89 MT |
2237 | vq_err(vq, "Failed to get flags"); |
2238 | return true; | |
2239 | } | |
3b1bbe89 | 2240 | return !(flags & cpu_to_vhost16(vq, VRING_AVAIL_F_NO_INTERRUPT)); |
3a4d5c94 | 2241 | } |
8ea8cf89 MT |
2242 | old = vq->signalled_used; |
2243 | v = vq->signalled_used_valid; | |
2244 | new = vq->signalled_used = vq->last_used_idx; | |
2245 | vq->signalled_used_valid = true; | |
3a4d5c94 | 2246 | |
8ea8cf89 MT |
2247 | if (unlikely(!v)) |
2248 | return true; | |
3a4d5c94 | 2249 | |
f8894913 | 2250 | if (vhost_get_avail(vq, event, vhost_used_event(vq))) { |
8ea8cf89 MT |
2251 | vq_err(vq, "Failed to get used event idx"); |
2252 | return true; | |
2253 | } | |
8d65843c | 2254 | return vring_need_event(vhost16_to_cpu(vq, event), new, old); |
8ea8cf89 MT |
2255 | } |
2256 | ||
2257 | /* This actually signals the guest, using eventfd. */ | |
2258 | void vhost_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq) | |
2259 | { | |
3a4d5c94 | 2260 | /* Signal the Guest tell them we used something up. */ |
8ea8cf89 | 2261 | if (vq->call_ctx && vhost_notify(dev, vq)) |
3a4d5c94 MT |
2262 | eventfd_signal(vq->call_ctx, 1); |
2263 | } | |
6ac1afbf | 2264 | EXPORT_SYMBOL_GPL(vhost_signal); |
3a4d5c94 MT |
2265 | |
2266 | /* And here's the combo meal deal. Supersize me! */ | |
2267 | void vhost_add_used_and_signal(struct vhost_dev *dev, | |
2268 | struct vhost_virtqueue *vq, | |
2269 | unsigned int head, int len) | |
2270 | { | |
2271 | vhost_add_used(vq, head, len); | |
2272 | vhost_signal(dev, vq); | |
2273 | } | |
6ac1afbf | 2274 | EXPORT_SYMBOL_GPL(vhost_add_used_and_signal); |
3a4d5c94 | 2275 | |
8dd014ad DS |
2276 | /* multi-buffer version of vhost_add_used_and_signal */ |
2277 | void vhost_add_used_and_signal_n(struct vhost_dev *dev, | |
2278 | struct vhost_virtqueue *vq, | |
2279 | struct vring_used_elem *heads, unsigned count) | |
2280 | { | |
2281 | vhost_add_used_n(vq, heads, count); | |
2282 | vhost_signal(dev, vq); | |
2283 | } | |
6ac1afbf | 2284 | EXPORT_SYMBOL_GPL(vhost_add_used_and_signal_n); |
8dd014ad | 2285 | |
d4a60603 JW |
2286 | /* return true if we're sure that avaiable ring is empty */ |
2287 | bool vhost_vq_avail_empty(struct vhost_dev *dev, struct vhost_virtqueue *vq) | |
2288 | { | |
2289 | __virtio16 avail_idx; | |
2290 | int r; | |
2291 | ||
275bf960 JW |
2292 | if (vq->avail_idx != vq->last_avail_idx) |
2293 | return false; | |
2294 | ||
f8894913 | 2295 | r = vhost_get_avail(vq, avail_idx, &vq->avail->idx); |
275bf960 | 2296 | if (unlikely(r)) |
d4a60603 | 2297 | return false; |
275bf960 | 2298 | vq->avail_idx = vhost16_to_cpu(vq, avail_idx); |
d4a60603 | 2299 | |
275bf960 | 2300 | return vq->avail_idx == vq->last_avail_idx; |
d4a60603 JW |
2301 | } |
2302 | EXPORT_SYMBOL_GPL(vhost_vq_avail_empty); | |
2303 | ||
3a4d5c94 | 2304 | /* OK, now we need to know about added descriptors. */ |
8ea8cf89 | 2305 | bool vhost_enable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 | 2306 | { |
3b1bbe89 | 2307 | __virtio16 avail_idx; |
3a4d5c94 | 2308 | int r; |
d47effe1 | 2309 | |
3a4d5c94 MT |
2310 | if (!(vq->used_flags & VRING_USED_F_NO_NOTIFY)) |
2311 | return false; | |
2312 | vq->used_flags &= ~VRING_USED_F_NO_NOTIFY; | |
ea16c514 | 2313 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
2723feaa | 2314 | r = vhost_update_used_flags(vq); |
8ea8cf89 MT |
2315 | if (r) { |
2316 | vq_err(vq, "Failed to enable notification at %p: %d\n", | |
2317 | &vq->used->flags, r); | |
2318 | return false; | |
2319 | } | |
2320 | } else { | |
2723feaa | 2321 | r = vhost_update_avail_event(vq, vq->avail_idx); |
8ea8cf89 MT |
2322 | if (r) { |
2323 | vq_err(vq, "Failed to update avail event index at %p: %d\n", | |
2324 | vhost_avail_event(vq), r); | |
2325 | return false; | |
2326 | } | |
2327 | } | |
3a4d5c94 MT |
2328 | /* They could have slipped one in as we were doing that: make |
2329 | * sure it's written, then check again. */ | |
5659338c | 2330 | smp_mb(); |
f8894913 | 2331 | r = vhost_get_avail(vq, avail_idx, &vq->avail->idx); |
3a4d5c94 MT |
2332 | if (r) { |
2333 | vq_err(vq, "Failed to check avail idx at %p: %d\n", | |
2334 | &vq->avail->idx, r); | |
2335 | return false; | |
2336 | } | |
2337 | ||
3b1bbe89 | 2338 | return vhost16_to_cpu(vq, avail_idx) != vq->avail_idx; |
3a4d5c94 | 2339 | } |
6ac1afbf | 2340 | EXPORT_SYMBOL_GPL(vhost_enable_notify); |
3a4d5c94 MT |
2341 | |
2342 | /* We don't need to be notified again. */ | |
8ea8cf89 | 2343 | void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 MT |
2344 | { |
2345 | int r; | |
d47effe1 | 2346 | |
3a4d5c94 MT |
2347 | if (vq->used_flags & VRING_USED_F_NO_NOTIFY) |
2348 | return; | |
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 | } | |
3a4d5c94 | 2356 | } |
6ac1afbf AH |
2357 | EXPORT_SYMBOL_GPL(vhost_disable_notify); |
2358 | ||
6b1e6cc7 JW |
2359 | /* Create a new message. */ |
2360 | struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type) | |
2361 | { | |
2362 | struct vhost_msg_node *node = kmalloc(sizeof *node, GFP_KERNEL); | |
2363 | if (!node) | |
2364 | return NULL; | |
2365 | node->vq = vq; | |
2366 | node->msg.type = type; | |
2367 | return node; | |
2368 | } | |
2369 | EXPORT_SYMBOL_GPL(vhost_new_msg); | |
2370 | ||
2371 | void vhost_enqueue_msg(struct vhost_dev *dev, struct list_head *head, | |
2372 | struct vhost_msg_node *node) | |
2373 | { | |
2374 | spin_lock(&dev->iotlb_lock); | |
2375 | list_add_tail(&node->node, head); | |
2376 | spin_unlock(&dev->iotlb_lock); | |
2377 | ||
2378 | wake_up_interruptible_poll(&dev->wait, POLLIN | POLLRDNORM); | |
2379 | } | |
2380 | EXPORT_SYMBOL_GPL(vhost_enqueue_msg); | |
2381 | ||
2382 | struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev, | |
2383 | struct list_head *head) | |
2384 | { | |
2385 | struct vhost_msg_node *node = NULL; | |
2386 | ||
2387 | spin_lock(&dev->iotlb_lock); | |
2388 | if (!list_empty(head)) { | |
2389 | node = list_first_entry(head, struct vhost_msg_node, | |
2390 | node); | |
2391 | list_del(&node->node); | |
2392 | } | |
2393 | spin_unlock(&dev->iotlb_lock); | |
2394 | ||
2395 | return node; | |
2396 | } | |
2397 | EXPORT_SYMBOL_GPL(vhost_dequeue_msg); | |
2398 | ||
2399 | ||
6ac1afbf AH |
2400 | static int __init vhost_init(void) |
2401 | { | |
2402 | return 0; | |
2403 | } | |
2404 | ||
2405 | static void __exit vhost_exit(void) | |
2406 | { | |
2407 | } | |
2408 | ||
2409 | module_init(vhost_init); | |
2410 | module_exit(vhost_exit); | |
2411 | ||
2412 | MODULE_VERSION("0.0.1"); | |
2413 | MODULE_LICENSE("GPL v2"); | |
2414 | MODULE_AUTHOR("Michael S. Tsirkin"); | |
2415 | MODULE_DESCRIPTION("Host kernel accelerator for virtio"); |