Commit | Line | Data |
---|---|---|
7a338472 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
3a4d5c94 MT |
2 | /* Copyright (C) 2009 Red Hat, Inc. |
3 | * Copyright (C) 2006 Rusty Russell IBM Corporation | |
4 | * | |
5 | * Author: Michael S. Tsirkin <mst@redhat.com> | |
6 | * | |
7 | * Inspiration, some code, and most witty comments come from | |
61516587 | 8 | * Documentation/virtual/lguest/lguest.c, by Rusty Russell |
3a4d5c94 | 9 | * |
3a4d5c94 MT |
10 | * Generic code for virtio server in host kernel. |
11 | */ | |
12 | ||
13 | #include <linux/eventfd.h> | |
14 | #include <linux/vhost.h> | |
35596b27 | 15 | #include <linux/uio.h> |
3a4d5c94 MT |
16 | #include <linux/mm.h> |
17 | #include <linux/miscdevice.h> | |
18 | #include <linux/mutex.h> | |
3a4d5c94 MT |
19 | #include <linux/poll.h> |
20 | #include <linux/file.h> | |
21 | #include <linux/highmem.h> | |
5a0e3ad6 | 22 | #include <linux/slab.h> |
4de7255f | 23 | #include <linux/vmalloc.h> |
c23f3445 | 24 | #include <linux/kthread.h> |
9e3d1957 | 25 | #include <linux/cgroup.h> |
6ac1afbf | 26 | #include <linux/module.h> |
bcfeacab | 27 | #include <linux/sort.h> |
6e84f315 | 28 | #include <linux/sched/mm.h> |
174cd4b1 | 29 | #include <linux/sched/signal.h> |
a9709d68 | 30 | #include <linux/interval_tree_generic.h> |
ff002269 | 31 | #include <linux/nospec.h> |
8f6a7f96 | 32 | #include <linux/kcov.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 | |
2751c988 | 52 | #ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY |
c5072037 | 53 | static void vhost_disable_cross_endian(struct vhost_virtqueue *vq) |
2751c988 GK |
54 | { |
55 | vq->user_be = !virtio_legacy_is_little_endian(); | |
56 | } | |
57 | ||
c5072037 GK |
58 | static void vhost_enable_cross_endian_big(struct vhost_virtqueue *vq) |
59 | { | |
60 | vq->user_be = true; | |
61 | } | |
62 | ||
63 | static void vhost_enable_cross_endian_little(struct vhost_virtqueue *vq) | |
64 | { | |
65 | vq->user_be = false; | |
66 | } | |
67 | ||
2751c988 GK |
68 | static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp) |
69 | { | |
70 | struct vhost_vring_state s; | |
71 | ||
72 | if (vq->private_data) | |
73 | return -EBUSY; | |
74 | ||
75 | if (copy_from_user(&s, argp, sizeof(s))) | |
76 | return -EFAULT; | |
77 | ||
78 | if (s.num != VHOST_VRING_LITTLE_ENDIAN && | |
79 | s.num != VHOST_VRING_BIG_ENDIAN) | |
80 | return -EINVAL; | |
81 | ||
c5072037 GK |
82 | if (s.num == VHOST_VRING_BIG_ENDIAN) |
83 | vhost_enable_cross_endian_big(vq); | |
84 | else | |
85 | vhost_enable_cross_endian_little(vq); | |
2751c988 GK |
86 | |
87 | return 0; | |
88 | } | |
89 | ||
90 | static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx, | |
91 | int __user *argp) | |
92 | { | |
93 | struct vhost_vring_state s = { | |
94 | .index = idx, | |
95 | .num = vq->user_be | |
96 | }; | |
97 | ||
98 | if (copy_to_user(argp, &s, sizeof(s))) | |
99 | return -EFAULT; | |
100 | ||
101 | return 0; | |
102 | } | |
103 | ||
104 | static void vhost_init_is_le(struct vhost_virtqueue *vq) | |
105 | { | |
106 | /* Note for legacy virtio: user_be is initialized at reset time | |
107 | * according to the host endianness. If userspace does not set an | |
108 | * explicit endianness, the default behavior is native endian, as | |
109 | * expected by legacy virtio. | |
110 | */ | |
111 | vq->is_le = vhost_has_feature(vq, VIRTIO_F_VERSION_1) || !vq->user_be; | |
112 | } | |
113 | #else | |
c5072037 | 114 | static void vhost_disable_cross_endian(struct vhost_virtqueue *vq) |
2751c988 GK |
115 | { |
116 | } | |
117 | ||
118 | static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp) | |
119 | { | |
120 | return -ENOIOCTLCMD; | |
121 | } | |
122 | ||
123 | static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx, | |
124 | int __user *argp) | |
125 | { | |
126 | return -ENOIOCTLCMD; | |
127 | } | |
128 | ||
129 | static void vhost_init_is_le(struct vhost_virtqueue *vq) | |
130 | { | |
cda8bba0 HP |
131 | vq->is_le = vhost_has_feature(vq, VIRTIO_F_VERSION_1) |
132 | || virtio_legacy_is_little_endian(); | |
2751c988 GK |
133 | } |
134 | #endif /* CONFIG_VHOST_CROSS_ENDIAN_LEGACY */ | |
135 | ||
c5072037 GK |
136 | static void vhost_reset_is_le(struct vhost_virtqueue *vq) |
137 | { | |
cda8bba0 | 138 | vhost_init_is_le(vq); |
c5072037 GK |
139 | } |
140 | ||
7235acdb JW |
141 | struct vhost_flush_struct { |
142 | struct vhost_work work; | |
143 | struct completion wait_event; | |
144 | }; | |
145 | ||
146 | static void vhost_flush_work(struct vhost_work *work) | |
147 | { | |
148 | struct vhost_flush_struct *s; | |
149 | ||
150 | s = container_of(work, struct vhost_flush_struct, work); | |
151 | complete(&s->wait_event); | |
152 | } | |
153 | ||
3a4d5c94 MT |
154 | static void vhost_poll_func(struct file *file, wait_queue_head_t *wqh, |
155 | poll_table *pt) | |
156 | { | |
157 | struct vhost_poll *poll; | |
3a4d5c94 | 158 | |
d47effe1 | 159 | poll = container_of(pt, struct vhost_poll, table); |
3a4d5c94 MT |
160 | poll->wqh = wqh; |
161 | add_wait_queue(wqh, &poll->wait); | |
162 | } | |
163 | ||
ac6424b9 | 164 | static int vhost_poll_wakeup(wait_queue_entry_t *wait, unsigned mode, int sync, |
3a4d5c94 MT |
165 | void *key) |
166 | { | |
c23f3445 | 167 | struct vhost_poll *poll = container_of(wait, struct vhost_poll, wait); |
01fcb1cb | 168 | struct vhost_work *work = &poll->work; |
c23f3445 | 169 | |
3ad6f93e | 170 | if (!(key_to_poll(key) & poll->mask)) |
3a4d5c94 MT |
171 | return 0; |
172 | ||
01fcb1cb JW |
173 | if (!poll->dev->use_worker) |
174 | work->fn(work); | |
175 | else | |
176 | vhost_poll_queue(poll); | |
177 | ||
3a4d5c94 MT |
178 | return 0; |
179 | } | |
180 | ||
163049ae | 181 | void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn) |
87d6a412 | 182 | { |
04b96e55 | 183 | clear_bit(VHOST_WORK_QUEUED, &work->flags); |
87d6a412 | 184 | work->fn = fn; |
87d6a412 | 185 | } |
6ac1afbf | 186 | EXPORT_SYMBOL_GPL(vhost_work_init); |
87d6a412 | 187 | |
3a4d5c94 | 188 | /* Init poll structure */ |
c23f3445 | 189 | void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn, |
58e3b602 | 190 | __poll_t mask, struct vhost_dev *dev) |
3a4d5c94 | 191 | { |
3a4d5c94 MT |
192 | init_waitqueue_func_entry(&poll->wait, vhost_poll_wakeup); |
193 | init_poll_funcptr(&poll->table, vhost_poll_func); | |
194 | poll->mask = mask; | |
c23f3445 | 195 | poll->dev = dev; |
2b8b328b | 196 | poll->wqh = NULL; |
c23f3445 | 197 | |
87d6a412 | 198 | vhost_work_init(&poll->work, fn); |
3a4d5c94 | 199 | } |
6ac1afbf | 200 | EXPORT_SYMBOL_GPL(vhost_poll_init); |
3a4d5c94 MT |
201 | |
202 | /* Start polling a file. We add ourselves to file's wait queue. The caller must | |
203 | * keep a reference to a file until after vhost_poll_stop is called. */ | |
2b8b328b | 204 | int vhost_poll_start(struct vhost_poll *poll, struct file *file) |
3a4d5c94 | 205 | { |
e6c8adca | 206 | __poll_t mask; |
d47effe1 | 207 | |
70181d51 JW |
208 | if (poll->wqh) |
209 | return 0; | |
210 | ||
9965ed17 | 211 | mask = vfs_poll(file, &poll->table); |
3a4d5c94 | 212 | if (mask) |
3ad6f93e | 213 | vhost_poll_wakeup(&poll->wait, 0, 0, poll_to_key(mask)); |
a9a08845 | 214 | if (mask & EPOLLERR) { |
dc6455a7 | 215 | vhost_poll_stop(poll); |
896fc242 | 216 | return -EINVAL; |
2b8b328b JW |
217 | } |
218 | ||
896fc242 | 219 | return 0; |
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 | |
b2ffa407 | 234 | void vhost_dev_flush(struct vhost_dev *dev) |
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 | } |
b2ffa407 | 246 | EXPORT_SYMBOL_GPL(vhost_dev_flush); |
3a4d5c94 | 247 | |
163049ae | 248 | void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work) |
3a4d5c94 | 249 | { |
04b96e55 JW |
250 | if (!dev->worker) |
251 | return; | |
c23f3445 | 252 | |
04b96e55 JW |
253 | if (!test_and_set_bit(VHOST_WORK_QUEUED, &work->flags)) { |
254 | /* We can only add the work to the list after we're | |
255 | * sure it was not in the list. | |
635abf01 | 256 | * test_and_set_bit() implies a memory barrier. |
04b96e55 | 257 | */ |
04b96e55 | 258 | llist_add(&work->node, &dev->work_list); |
c23f3445 TH |
259 | wake_up_process(dev->worker); |
260 | } | |
3a4d5c94 | 261 | } |
6ac1afbf | 262 | EXPORT_SYMBOL_GPL(vhost_work_queue); |
3a4d5c94 | 263 | |
526d3e7f JW |
264 | /* A lockless hint for busy polling code to exit the loop */ |
265 | bool vhost_has_work(struct vhost_dev *dev) | |
266 | { | |
04b96e55 | 267 | return !llist_empty(&dev->work_list); |
526d3e7f JW |
268 | } |
269 | EXPORT_SYMBOL_GPL(vhost_has_work); | |
270 | ||
87d6a412 MT |
271 | void vhost_poll_queue(struct vhost_poll *poll) |
272 | { | |
273 | vhost_work_queue(poll->dev, &poll->work); | |
274 | } | |
6ac1afbf | 275 | EXPORT_SYMBOL_GPL(vhost_poll_queue); |
87d6a412 | 276 | |
f8894913 JW |
277 | static void __vhost_vq_meta_reset(struct vhost_virtqueue *vq) |
278 | { | |
279 | int j; | |
280 | ||
281 | for (j = 0; j < VHOST_NUM_ADDRS; j++) | |
282 | vq->meta_iotlb[j] = NULL; | |
283 | } | |
284 | ||
285 | static void vhost_vq_meta_reset(struct vhost_dev *d) | |
286 | { | |
287 | int i; | |
288 | ||
86a07da3 | 289 | for (i = 0; i < d->nvqs; ++i) |
f8894913 JW |
290 | __vhost_vq_meta_reset(d->vqs[i]); |
291 | } | |
292 | ||
265a0ad8 ZL |
293 | static void vhost_vring_call_reset(struct vhost_vring_call *call_ctx) |
294 | { | |
295 | call_ctx->ctx = NULL; | |
296 | memset(&call_ctx->producer, 0x0, sizeof(struct irq_bypass_producer)); | |
265a0ad8 ZL |
297 | } |
298 | ||
6bcf3422 MC |
299 | bool vhost_vq_is_setup(struct vhost_virtqueue *vq) |
300 | { | |
301 | return vq->avail && vq->desc && vq->used && vhost_vq_access_ok(vq); | |
302 | } | |
303 | EXPORT_SYMBOL_GPL(vhost_vq_is_setup); | |
304 | ||
3a4d5c94 MT |
305 | static void vhost_vq_reset(struct vhost_dev *dev, |
306 | struct vhost_virtqueue *vq) | |
307 | { | |
308 | vq->num = 1; | |
309 | vq->desc = NULL; | |
310 | vq->avail = NULL; | |
311 | vq->used = NULL; | |
312 | vq->last_avail_idx = 0; | |
313 | vq->avail_idx = 0; | |
314 | vq->last_used_idx = 0; | |
8ea8cf89 MT |
315 | vq->signalled_used = 0; |
316 | vq->signalled_used_valid = false; | |
3a4d5c94 | 317 | vq->used_flags = 0; |
3a4d5c94 MT |
318 | vq->log_used = false; |
319 | vq->log_addr = -1ull; | |
3a4d5c94 | 320 | vq->private_data = NULL; |
ea16c514 | 321 | vq->acked_features = 0; |
429711ae | 322 | vq->acked_backend_features = 0; |
3a4d5c94 MT |
323 | vq->log_base = NULL; |
324 | vq->error_ctx = NULL; | |
3a4d5c94 | 325 | vq->kick = NULL; |
73a99f08 | 326 | vq->log_ctx = NULL; |
c5072037 | 327 | vhost_disable_cross_endian(vq); |
beb691e6 | 328 | vhost_reset_is_le(vq); |
03088137 | 329 | vq->busyloop_timeout = 0; |
a9709d68 | 330 | vq->umem = NULL; |
6b1e6cc7 | 331 | vq->iotlb = NULL; |
265a0ad8 | 332 | vhost_vring_call_reset(&vq->call_ctx); |
f8894913 | 333 | __vhost_vq_meta_reset(vq); |
3a4d5c94 MT |
334 | } |
335 | ||
c23f3445 TH |
336 | static int vhost_worker(void *data) |
337 | { | |
338 | struct vhost_dev *dev = data; | |
04b96e55 JW |
339 | struct vhost_work *work, *work_next; |
340 | struct llist_node *node; | |
c23f3445 | 341 | |
f5678e7f | 342 | kthread_use_mm(dev->mm); |
64e1c807 | 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 | 362 | __set_current_state(TASK_RUNNING); |
8f6a7f96 | 363 | kcov_remote_start_common(dev->kcov_handle); |
c23f3445 | 364 | work->fn(work); |
8f6a7f96 | 365 | kcov_remote_stop(); |
d550dda1 NHE |
366 | if (need_resched()) |
367 | schedule(); | |
04b96e55 | 368 | } |
c23f3445 | 369 | } |
f5678e7f | 370 | kthread_unuse_mm(dev->mm); |
64e1c807 | 371 | return 0; |
c23f3445 TH |
372 | } |
373 | ||
bab632d6 MT |
374 | static void vhost_vq_free_iovecs(struct vhost_virtqueue *vq) |
375 | { | |
376 | kfree(vq->indirect); | |
377 | vq->indirect = NULL; | |
378 | kfree(vq->log); | |
379 | vq->log = NULL; | |
380 | kfree(vq->heads); | |
381 | vq->heads = NULL; | |
bab632d6 MT |
382 | } |
383 | ||
e0e9b406 JW |
384 | /* Helper to allocate iovec buffers for all vqs. */ |
385 | static long vhost_dev_alloc_iovecs(struct vhost_dev *dev) | |
386 | { | |
6d5e6aa8 | 387 | struct vhost_virtqueue *vq; |
e0e9b406 | 388 | int i; |
d47effe1 | 389 | |
e0e9b406 | 390 | for (i = 0; i < dev->nvqs; ++i) { |
6d5e6aa8 | 391 | vq = dev->vqs[i]; |
6da2ec56 KC |
392 | vq->indirect = kmalloc_array(UIO_MAXIOV, |
393 | sizeof(*vq->indirect), | |
394 | GFP_KERNEL); | |
b46a0bf7 | 395 | vq->log = kmalloc_array(dev->iov_limit, sizeof(*vq->log), |
6da2ec56 | 396 | GFP_KERNEL); |
b46a0bf7 | 397 | vq->heads = kmalloc_array(dev->iov_limit, sizeof(*vq->heads), |
6da2ec56 | 398 | GFP_KERNEL); |
6d5e6aa8 | 399 | if (!vq->indirect || !vq->log || !vq->heads) |
e0e9b406 JW |
400 | goto err_nomem; |
401 | } | |
402 | return 0; | |
d47effe1 | 403 | |
e0e9b406 | 404 | err_nomem: |
bab632d6 | 405 | for (; i >= 0; --i) |
3ab2e420 | 406 | vhost_vq_free_iovecs(dev->vqs[i]); |
e0e9b406 JW |
407 | return -ENOMEM; |
408 | } | |
409 | ||
410 | static void vhost_dev_free_iovecs(struct vhost_dev *dev) | |
411 | { | |
412 | int i; | |
d47effe1 | 413 | |
bab632d6 | 414 | for (i = 0; i < dev->nvqs; ++i) |
3ab2e420 | 415 | vhost_vq_free_iovecs(dev->vqs[i]); |
e0e9b406 JW |
416 | } |
417 | ||
e82b9b07 JW |
418 | bool vhost_exceeds_weight(struct vhost_virtqueue *vq, |
419 | int pkts, int total_len) | |
420 | { | |
421 | struct vhost_dev *dev = vq->dev; | |
422 | ||
423 | if ((dev->byte_weight && total_len >= dev->byte_weight) || | |
424 | pkts >= dev->weight) { | |
425 | vhost_poll_queue(&vq->poll); | |
426 | return true; | |
427 | } | |
428 | ||
429 | return false; | |
430 | } | |
431 | EXPORT_SYMBOL_GPL(vhost_exceeds_weight); | |
432 | ||
4942e825 JW |
433 | static size_t vhost_get_avail_size(struct vhost_virtqueue *vq, |
434 | unsigned int num) | |
435 | { | |
436 | size_t event __maybe_unused = | |
437 | vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; | |
438 | ||
439 | return sizeof(*vq->avail) + | |
440 | sizeof(*vq->avail->ring) * num + event; | |
441 | } | |
442 | ||
443 | static size_t vhost_get_used_size(struct vhost_virtqueue *vq, | |
444 | unsigned int num) | |
445 | { | |
446 | size_t event __maybe_unused = | |
447 | vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; | |
448 | ||
449 | return sizeof(*vq->used) + | |
450 | sizeof(*vq->used->ring) * num + event; | |
451 | } | |
452 | ||
453 | static size_t vhost_get_desc_size(struct vhost_virtqueue *vq, | |
454 | unsigned int num) | |
455 | { | |
456 | return sizeof(*vq->desc) * num; | |
457 | } | |
458 | ||
59566b6e | 459 | void vhost_dev_init(struct vhost_dev *dev, |
e82b9b07 | 460 | struct vhost_virtqueue **vqs, int nvqs, |
792a4f2e | 461 | int iov_limit, int weight, int byte_weight, |
01fcb1cb | 462 | bool use_worker, |
91233ad7 | 463 | int (*msg_handler)(struct vhost_dev *dev, u32 asid, |
792a4f2e | 464 | struct vhost_iotlb_msg *msg)) |
3a4d5c94 | 465 | { |
6d5e6aa8 | 466 | struct vhost_virtqueue *vq; |
3a4d5c94 | 467 | int i; |
c23f3445 | 468 | |
3a4d5c94 MT |
469 | dev->vqs = vqs; |
470 | dev->nvqs = nvqs; | |
471 | mutex_init(&dev->mutex); | |
472 | dev->log_ctx = NULL; | |
a9709d68 | 473 | dev->umem = NULL; |
6b1e6cc7 | 474 | dev->iotlb = NULL; |
3a4d5c94 | 475 | dev->mm = NULL; |
c23f3445 | 476 | dev->worker = NULL; |
b46a0bf7 | 477 | dev->iov_limit = iov_limit; |
e82b9b07 JW |
478 | dev->weight = weight; |
479 | dev->byte_weight = byte_weight; | |
01fcb1cb | 480 | dev->use_worker = use_worker; |
792a4f2e | 481 | dev->msg_handler = msg_handler; |
04b96e55 | 482 | init_llist_head(&dev->work_list); |
6b1e6cc7 JW |
483 | init_waitqueue_head(&dev->wait); |
484 | INIT_LIST_HEAD(&dev->read_list); | |
485 | INIT_LIST_HEAD(&dev->pending_list); | |
486 | spin_lock_init(&dev->iotlb_lock); | |
3d2c7d37 | 487 | |
3a4d5c94 MT |
488 | |
489 | for (i = 0; i < dev->nvqs; ++i) { | |
6d5e6aa8 AH |
490 | vq = dev->vqs[i]; |
491 | vq->log = NULL; | |
492 | vq->indirect = NULL; | |
493 | vq->heads = NULL; | |
494 | vq->dev = dev; | |
495 | mutex_init(&vq->mutex); | |
496 | vhost_vq_reset(dev, vq); | |
497 | if (vq->handle_kick) | |
498 | vhost_poll_init(&vq->poll, vq->handle_kick, | |
a9a08845 | 499 | EPOLLIN, dev); |
3a4d5c94 | 500 | } |
3a4d5c94 | 501 | } |
6ac1afbf | 502 | EXPORT_SYMBOL_GPL(vhost_dev_init); |
3a4d5c94 MT |
503 | |
504 | /* Caller should have device mutex */ | |
505 | long vhost_dev_check_owner(struct vhost_dev *dev) | |
506 | { | |
507 | /* Are you the owner? If not, I don't think you mean to do that */ | |
508 | return dev->mm == current->mm ? 0 : -EPERM; | |
509 | } | |
6ac1afbf | 510 | EXPORT_SYMBOL_GPL(vhost_dev_check_owner); |
3a4d5c94 | 511 | |
87d6a412 | 512 | struct vhost_attach_cgroups_struct { |
d47effe1 KK |
513 | struct vhost_work work; |
514 | struct task_struct *owner; | |
515 | int ret; | |
87d6a412 MT |
516 | }; |
517 | ||
518 | static void vhost_attach_cgroups_work(struct vhost_work *work) | |
519 | { | |
d47effe1 KK |
520 | struct vhost_attach_cgroups_struct *s; |
521 | ||
522 | s = container_of(work, struct vhost_attach_cgroups_struct, work); | |
523 | s->ret = cgroup_attach_task_all(s->owner, current); | |
87d6a412 MT |
524 | } |
525 | ||
526 | static int vhost_attach_cgroups(struct vhost_dev *dev) | |
527 | { | |
d47effe1 KK |
528 | struct vhost_attach_cgroups_struct attach; |
529 | ||
530 | attach.owner = current; | |
531 | vhost_work_init(&attach.work, vhost_attach_cgroups_work); | |
532 | vhost_work_queue(dev, &attach.work); | |
b2ffa407 | 533 | vhost_dev_flush(dev); |
d47effe1 | 534 | return attach.ret; |
87d6a412 MT |
535 | } |
536 | ||
05c05351 MT |
537 | /* Caller should have device mutex */ |
538 | bool vhost_dev_has_owner(struct vhost_dev *dev) | |
539 | { | |
540 | return dev->mm; | |
541 | } | |
6ac1afbf | 542 | EXPORT_SYMBOL_GPL(vhost_dev_has_owner); |
05c05351 | 543 | |
5ce995f3 JW |
544 | static void vhost_attach_mm(struct vhost_dev *dev) |
545 | { | |
546 | /* No owner, become one */ | |
547 | if (dev->use_worker) { | |
548 | dev->mm = get_task_mm(current); | |
549 | } else { | |
550 | /* vDPA device does not use worker thead, so there's | |
551 | * no need to hold the address space for mm. This help | |
552 | * to avoid deadlock in the case of mmap() which may | |
553 | * held the refcnt of the file and depends on release | |
554 | * method to remove vma. | |
555 | */ | |
556 | dev->mm = current->mm; | |
557 | mmgrab(dev->mm); | |
558 | } | |
559 | } | |
560 | ||
561 | static void vhost_detach_mm(struct vhost_dev *dev) | |
562 | { | |
563 | if (!dev->mm) | |
564 | return; | |
565 | ||
566 | if (dev->use_worker) | |
567 | mmput(dev->mm); | |
568 | else | |
569 | mmdrop(dev->mm); | |
570 | ||
571 | dev->mm = NULL; | |
572 | } | |
573 | ||
3a4d5c94 | 574 | /* Caller should have device mutex */ |
54db63c2 | 575 | long vhost_dev_set_owner(struct vhost_dev *dev) |
3a4d5c94 | 576 | { |
c23f3445 TH |
577 | struct task_struct *worker; |
578 | int err; | |
d47effe1 | 579 | |
3a4d5c94 | 580 | /* Is there an owner already? */ |
05c05351 | 581 | if (vhost_dev_has_owner(dev)) { |
c23f3445 TH |
582 | err = -EBUSY; |
583 | goto err_mm; | |
584 | } | |
d47effe1 | 585 | |
5ce995f3 JW |
586 | vhost_attach_mm(dev); |
587 | ||
8f6a7f96 | 588 | dev->kcov_handle = kcov_common_handle(); |
01fcb1cb JW |
589 | if (dev->use_worker) { |
590 | worker = kthread_create(vhost_worker, dev, | |
591 | "vhost-%d", current->pid); | |
592 | if (IS_ERR(worker)) { | |
593 | err = PTR_ERR(worker); | |
594 | goto err_worker; | |
595 | } | |
c23f3445 | 596 | |
01fcb1cb JW |
597 | dev->worker = worker; |
598 | wake_up_process(worker); /* avoid contributing to loadavg */ | |
87d6a412 | 599 | |
01fcb1cb JW |
600 | err = vhost_attach_cgroups(dev); |
601 | if (err) | |
602 | goto err_cgroup; | |
603 | } | |
c23f3445 | 604 | |
e0e9b406 JW |
605 | err = vhost_dev_alloc_iovecs(dev); |
606 | if (err) | |
607 | goto err_cgroup; | |
608 | ||
3a4d5c94 | 609 | return 0; |
9e3d1957 | 610 | err_cgroup: |
01fcb1cb JW |
611 | if (dev->worker) { |
612 | kthread_stop(dev->worker); | |
613 | dev->worker = NULL; | |
614 | } | |
c23f3445 | 615 | err_worker: |
5ce995f3 | 616 | vhost_detach_mm(dev); |
8f6a7f96 | 617 | dev->kcov_handle = 0; |
c23f3445 TH |
618 | err_mm: |
619 | return err; | |
3a4d5c94 | 620 | } |
6ac1afbf | 621 | EXPORT_SYMBOL_GPL(vhost_dev_set_owner); |
3a4d5c94 | 622 | |
0bbe3066 JW |
623 | static struct vhost_iotlb *iotlb_alloc(void) |
624 | { | |
625 | return vhost_iotlb_alloc(max_iotlb_entries, | |
626 | VHOST_IOTLB_FLAG_RETIRE); | |
627 | } | |
628 | ||
629 | struct vhost_iotlb *vhost_dev_reset_owner_prepare(void) | |
a9709d68 | 630 | { |
0bbe3066 | 631 | return iotlb_alloc(); |
150b9e51 | 632 | } |
6ac1afbf | 633 | EXPORT_SYMBOL_GPL(vhost_dev_reset_owner_prepare); |
3a4d5c94 | 634 | |
150b9e51 | 635 | /* Caller should have device mutex */ |
0bbe3066 | 636 | void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_iotlb *umem) |
150b9e51 | 637 | { |
47283bef MT |
638 | int i; |
639 | ||
f6f93f75 | 640 | vhost_dev_cleanup(dev); |
3a4d5c94 | 641 | |
a9709d68 | 642 | dev->umem = umem; |
47283bef MT |
643 | /* We don't need VQ locks below since vhost_dev_cleanup makes sure |
644 | * VQs aren't running. | |
645 | */ | |
646 | for (i = 0; i < dev->nvqs; ++i) | |
a9709d68 | 647 | dev->vqs[i]->umem = umem; |
3a4d5c94 | 648 | } |
6ac1afbf | 649 | EXPORT_SYMBOL_GPL(vhost_dev_reset_owner); |
3a4d5c94 | 650 | |
b211616d | 651 | void vhost_dev_stop(struct vhost_dev *dev) |
bab632d6 MT |
652 | { |
653 | int i; | |
b211616d MT |
654 | |
655 | for (i = 0; i < dev->nvqs; ++i) { | |
6ca84326 | 656 | if (dev->vqs[i]->kick && dev->vqs[i]->handle_kick) |
3ab2e420 | 657 | vhost_poll_stop(&dev->vqs[i]->poll); |
bab632d6 | 658 | } |
6ca84326 | 659 | |
b2ffa407 | 660 | vhost_dev_flush(dev); |
bab632d6 | 661 | } |
6ac1afbf | 662 | EXPORT_SYMBOL_GPL(vhost_dev_stop); |
bab632d6 | 663 | |
6b1e6cc7 JW |
664 | static void vhost_clear_msg(struct vhost_dev *dev) |
665 | { | |
666 | struct vhost_msg_node *node, *n; | |
667 | ||
668 | spin_lock(&dev->iotlb_lock); | |
669 | ||
670 | list_for_each_entry_safe(node, n, &dev->read_list, node) { | |
671 | list_del(&node->node); | |
672 | kfree(node); | |
673 | } | |
674 | ||
675 | list_for_each_entry_safe(node, n, &dev->pending_list, node) { | |
676 | list_del(&node->node); | |
677 | kfree(node); | |
678 | } | |
679 | ||
680 | spin_unlock(&dev->iotlb_lock); | |
681 | } | |
682 | ||
f6f93f75 | 683 | void vhost_dev_cleanup(struct vhost_dev *dev) |
3a4d5c94 MT |
684 | { |
685 | int i; | |
d47effe1 | 686 | |
3a4d5c94 | 687 | for (i = 0; i < dev->nvqs; ++i) { |
3ab2e420 AH |
688 | if (dev->vqs[i]->error_ctx) |
689 | eventfd_ctx_put(dev->vqs[i]->error_ctx); | |
3ab2e420 AH |
690 | if (dev->vqs[i]->kick) |
691 | fput(dev->vqs[i]->kick); | |
265a0ad8 ZL |
692 | if (dev->vqs[i]->call_ctx.ctx) |
693 | eventfd_ctx_put(dev->vqs[i]->call_ctx.ctx); | |
3ab2e420 | 694 | vhost_vq_reset(dev, dev->vqs[i]); |
3a4d5c94 | 695 | } |
e0e9b406 | 696 | vhost_dev_free_iovecs(dev); |
3a4d5c94 MT |
697 | if (dev->log_ctx) |
698 | eventfd_ctx_put(dev->log_ctx); | |
699 | dev->log_ctx = NULL; | |
3a4d5c94 | 700 | /* No one will access memory at this point */ |
0bbe3066 | 701 | vhost_iotlb_free(dev->umem); |
a9709d68 | 702 | dev->umem = NULL; |
0bbe3066 | 703 | vhost_iotlb_free(dev->iotlb); |
6b1e6cc7 JW |
704 | dev->iotlb = NULL; |
705 | vhost_clear_msg(dev); | |
a9a08845 | 706 | wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM); |
04b96e55 | 707 | WARN_ON(!llist_empty(&dev->work_list)); |
78b620ce ED |
708 | if (dev->worker) { |
709 | kthread_stop(dev->worker); | |
710 | dev->worker = NULL; | |
8f6a7f96 | 711 | dev->kcov_handle = 0; |
78b620ce | 712 | } |
5ce995f3 | 713 | vhost_detach_mm(dev); |
3a4d5c94 | 714 | } |
6ac1afbf | 715 | EXPORT_SYMBOL_GPL(vhost_dev_cleanup); |
3a4d5c94 | 716 | |
ddd3d408 | 717 | static bool log_access_ok(void __user *log_base, u64 addr, unsigned long sz) |
3a4d5c94 MT |
718 | { |
719 | u64 a = addr / VHOST_PAGE_SIZE / 8; | |
d47effe1 | 720 | |
3a4d5c94 MT |
721 | /* Make sure 64 bit math will not overflow. */ |
722 | if (a > ULONG_MAX - (unsigned long)log_base || | |
723 | a + (unsigned long)log_base > ULONG_MAX) | |
ddd3d408 | 724 | return false; |
3a4d5c94 | 725 | |
96d4f267 | 726 | return access_ok(log_base + a, |
3a4d5c94 MT |
727 | (sz + VHOST_PAGE_SIZE * 8 - 1) / VHOST_PAGE_SIZE / 8); |
728 | } | |
729 | ||
f7ad318e | 730 | /* Make sure 64 bit math will not overflow. */ |
ec33d031 MT |
731 | static bool vhost_overflow(u64 uaddr, u64 size) |
732 | { | |
f7ad318e XY |
733 | if (uaddr > ULONG_MAX || size > ULONG_MAX) |
734 | return true; | |
735 | ||
736 | if (!size) | |
737 | return false; | |
738 | ||
739 | return uaddr > ULONG_MAX - size + 1; | |
ec33d031 MT |
740 | } |
741 | ||
3a4d5c94 | 742 | /* Caller should have vq mutex and device mutex. */ |
0bbe3066 | 743 | static bool vq_memory_access_ok(void __user *log_base, struct vhost_iotlb *umem, |
ddd3d408 | 744 | int log_all) |
3a4d5c94 | 745 | { |
0bbe3066 | 746 | struct vhost_iotlb_map *map; |
179b284e | 747 | |
a9709d68 | 748 | if (!umem) |
ddd3d408 | 749 | return false; |
179b284e | 750 | |
0bbe3066 JW |
751 | list_for_each_entry(map, &umem->list, link) { |
752 | unsigned long a = map->addr; | |
a9709d68 | 753 | |
0bbe3066 | 754 | if (vhost_overflow(map->addr, map->size)) |
ddd3d408 | 755 | return false; |
ec33d031 MT |
756 | |
757 | ||
0bbe3066 | 758 | if (!access_ok((void __user *)a, map->size)) |
ddd3d408 | 759 | return false; |
3a4d5c94 | 760 | else if (log_all && !log_access_ok(log_base, |
0bbe3066 JW |
761 | map->start, |
762 | map->size)) | |
ddd3d408 | 763 | return false; |
3a4d5c94 | 764 | } |
ddd3d408 | 765 | return true; |
3a4d5c94 MT |
766 | } |
767 | ||
f8894913 JW |
768 | static inline void __user *vhost_vq_meta_fetch(struct vhost_virtqueue *vq, |
769 | u64 addr, unsigned int size, | |
770 | int type) | |
771 | { | |
0bbe3066 | 772 | const struct vhost_iotlb_map *map = vq->meta_iotlb[type]; |
f8894913 | 773 | |
0bbe3066 | 774 | if (!map) |
f8894913 JW |
775 | return NULL; |
776 | ||
1b0be99f | 777 | return (void __user *)(uintptr_t)(map->addr + addr - map->start); |
f8894913 JW |
778 | } |
779 | ||
3a4d5c94 MT |
780 | /* Can we switch to this memory table? */ |
781 | /* Caller should have device mutex but not vq mutex */ | |
0bbe3066 | 782 | static bool memory_access_ok(struct vhost_dev *d, struct vhost_iotlb *umem, |
ddd3d408 | 783 | int log_all) |
3a4d5c94 MT |
784 | { |
785 | int i; | |
d47effe1 | 786 | |
3a4d5c94 | 787 | for (i = 0; i < d->nvqs; ++i) { |
ddd3d408 | 788 | bool ok; |
ea16c514 MT |
789 | bool log; |
790 | ||
3ab2e420 | 791 | mutex_lock(&d->vqs[i]->mutex); |
ea16c514 | 792 | log = log_all || vhost_has_feature(d->vqs[i], VHOST_F_LOG_ALL); |
3a4d5c94 | 793 | /* If ring is inactive, will check when it's enabled. */ |
3ab2e420 | 794 | if (d->vqs[i]->private_data) |
a9709d68 JW |
795 | ok = vq_memory_access_ok(d->vqs[i]->log_base, |
796 | umem, log); | |
3a4d5c94 | 797 | else |
ddd3d408 | 798 | ok = true; |
3ab2e420 | 799 | mutex_unlock(&d->vqs[i]->mutex); |
3a4d5c94 | 800 | if (!ok) |
ddd3d408 | 801 | return false; |
3a4d5c94 | 802 | } |
ddd3d408 | 803 | return true; |
3a4d5c94 MT |
804 | } |
805 | ||
6b1e6cc7 JW |
806 | static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len, |
807 | struct iovec iov[], int iov_size, int access); | |
bfe2bc51 | 808 | |
72952cc0 | 809 | static int vhost_copy_to_user(struct vhost_virtqueue *vq, void __user *to, |
bfe2bc51 JW |
810 | const void *from, unsigned size) |
811 | { | |
6b1e6cc7 | 812 | int ret; |
bfe2bc51 | 813 | |
6b1e6cc7 JW |
814 | if (!vq->iotlb) |
815 | return __copy_to_user(to, from, size); | |
816 | else { | |
817 | /* This function should be called after iotlb | |
818 | * prefetch, which means we're sure that all vq | |
819 | * could be access through iotlb. So -EAGAIN should | |
820 | * not happen in this case. | |
821 | */ | |
6b1e6cc7 | 822 | struct iov_iter t; |
f8894913 JW |
823 | void __user *uaddr = vhost_vq_meta_fetch(vq, |
824 | (u64)(uintptr_t)to, size, | |
7ced6c98 | 825 | VHOST_ADDR_USED); |
f8894913 JW |
826 | |
827 | if (uaddr) | |
828 | return __copy_to_user(uaddr, from, size); | |
829 | ||
6b1e6cc7 JW |
830 | ret = translate_desc(vq, (u64)(uintptr_t)to, size, vq->iotlb_iov, |
831 | ARRAY_SIZE(vq->iotlb_iov), | |
832 | VHOST_ACCESS_WO); | |
833 | if (ret < 0) | |
834 | goto out; | |
835 | iov_iter_init(&t, WRITE, vq->iotlb_iov, ret, size); | |
836 | ret = copy_to_iter(from, size, &t); | |
837 | if (ret == size) | |
838 | ret = 0; | |
839 | } | |
840 | out: | |
841 | return ret; | |
842 | } | |
bfe2bc51 JW |
843 | |
844 | static int vhost_copy_from_user(struct vhost_virtqueue *vq, void *to, | |
72952cc0 | 845 | void __user *from, unsigned size) |
bfe2bc51 | 846 | { |
6b1e6cc7 JW |
847 | int ret; |
848 | ||
849 | if (!vq->iotlb) | |
850 | return __copy_from_user(to, from, size); | |
851 | else { | |
852 | /* This function should be called after iotlb | |
853 | * prefetch, which means we're sure that vq | |
854 | * could be access through iotlb. So -EAGAIN should | |
855 | * not happen in this case. | |
856 | */ | |
f8894913 JW |
857 | void __user *uaddr = vhost_vq_meta_fetch(vq, |
858 | (u64)(uintptr_t)from, size, | |
859 | VHOST_ADDR_DESC); | |
6b1e6cc7 | 860 | struct iov_iter f; |
f8894913 JW |
861 | |
862 | if (uaddr) | |
863 | return __copy_from_user(to, uaddr, size); | |
864 | ||
6b1e6cc7 JW |
865 | ret = translate_desc(vq, (u64)(uintptr_t)from, size, vq->iotlb_iov, |
866 | ARRAY_SIZE(vq->iotlb_iov), | |
867 | VHOST_ACCESS_RO); | |
868 | if (ret < 0) { | |
869 | vq_err(vq, "IOTLB translation failure: uaddr " | |
870 | "%p size 0x%llx\n", from, | |
871 | (unsigned long long) size); | |
872 | goto out; | |
873 | } | |
874 | iov_iter_init(&f, READ, vq->iotlb_iov, ret, size); | |
875 | ret = copy_from_iter(to, size, &f); | |
876 | if (ret == size) | |
877 | ret = 0; | |
878 | } | |
879 | ||
880 | out: | |
881 | return ret; | |
882 | } | |
883 | ||
f8894913 JW |
884 | static void __user *__vhost_get_user_slow(struct vhost_virtqueue *vq, |
885 | void __user *addr, unsigned int size, | |
886 | int type) | |
6b1e6cc7 JW |
887 | { |
888 | int ret; | |
889 | ||
6b1e6cc7 JW |
890 | ret = translate_desc(vq, (u64)(uintptr_t)addr, size, vq->iotlb_iov, |
891 | ARRAY_SIZE(vq->iotlb_iov), | |
892 | VHOST_ACCESS_RO); | |
893 | if (ret < 0) { | |
894 | vq_err(vq, "IOTLB translation failure: uaddr " | |
895 | "%p size 0x%llx\n", addr, | |
896 | (unsigned long long) size); | |
897 | return NULL; | |
898 | } | |
899 | ||
900 | if (ret != 1 || vq->iotlb_iov[0].iov_len != size) { | |
901 | vq_err(vq, "Non atomic userspace memory access: uaddr " | |
902 | "%p size 0x%llx\n", addr, | |
903 | (unsigned long long) size); | |
904 | return NULL; | |
905 | } | |
906 | ||
907 | return vq->iotlb_iov[0].iov_base; | |
908 | } | |
909 | ||
f8894913 JW |
910 | /* This function should be called after iotlb |
911 | * prefetch, which means we're sure that vq | |
912 | * could be access through iotlb. So -EAGAIN should | |
913 | * not happen in this case. | |
914 | */ | |
915 | static inline void __user *__vhost_get_user(struct vhost_virtqueue *vq, | |
1b0be99f | 916 | void __user *addr, unsigned int size, |
f8894913 JW |
917 | int type) |
918 | { | |
919 | void __user *uaddr = vhost_vq_meta_fetch(vq, | |
920 | (u64)(uintptr_t)addr, size, type); | |
921 | if (uaddr) | |
922 | return uaddr; | |
923 | ||
924 | return __vhost_get_user_slow(vq, addr, size, type); | |
925 | } | |
926 | ||
927 | #define vhost_put_user(vq, x, ptr) \ | |
6b1e6cc7 | 928 | ({ \ |
002ef18e | 929 | int ret; \ |
6b1e6cc7 JW |
930 | if (!vq->iotlb) { \ |
931 | ret = __put_user(x, ptr); \ | |
932 | } else { \ | |
933 | __typeof__(ptr) to = \ | |
f8894913 JW |
934 | (__typeof__(ptr)) __vhost_get_user(vq, ptr, \ |
935 | sizeof(*ptr), VHOST_ADDR_USED); \ | |
6b1e6cc7 JW |
936 | if (to != NULL) \ |
937 | ret = __put_user(x, to); \ | |
938 | else \ | |
939 | ret = -EFAULT; \ | |
940 | } \ | |
941 | ret; \ | |
942 | }) | |
943 | ||
7b5d753e JW |
944 | static inline int vhost_put_avail_event(struct vhost_virtqueue *vq) |
945 | { | |
946 | return vhost_put_user(vq, cpu_to_vhost16(vq, vq->avail_idx), | |
947 | vhost_avail_event(vq)); | |
948 | } | |
949 | ||
950 | static inline int vhost_put_used(struct vhost_virtqueue *vq, | |
951 | struct vring_used_elem *head, int idx, | |
952 | int count) | |
953 | { | |
954 | return vhost_copy_to_user(vq, vq->used->ring + idx, head, | |
955 | count * sizeof(*head)); | |
956 | } | |
957 | ||
958 | static inline int vhost_put_used_flags(struct vhost_virtqueue *vq) | |
959 | ||
960 | { | |
961 | return vhost_put_user(vq, cpu_to_vhost16(vq, vq->used_flags), | |
962 | &vq->used->flags); | |
963 | } | |
964 | ||
965 | static inline int vhost_put_used_idx(struct vhost_virtqueue *vq) | |
966 | ||
967 | { | |
968 | return vhost_put_user(vq, cpu_to_vhost16(vq, vq->last_used_idx), | |
969 | &vq->used->idx); | |
970 | } | |
971 | ||
f8894913 | 972 | #define vhost_get_user(vq, x, ptr, type) \ |
6b1e6cc7 JW |
973 | ({ \ |
974 | int ret; \ | |
975 | if (!vq->iotlb) { \ | |
976 | ret = __get_user(x, ptr); \ | |
977 | } else { \ | |
978 | __typeof__(ptr) from = \ | |
f8894913 JW |
979 | (__typeof__(ptr)) __vhost_get_user(vq, ptr, \ |
980 | sizeof(*ptr), \ | |
981 | type); \ | |
6b1e6cc7 JW |
982 | if (from != NULL) \ |
983 | ret = __get_user(x, from); \ | |
984 | else \ | |
985 | ret = -EFAULT; \ | |
986 | } \ | |
987 | ret; \ | |
988 | }) | |
989 | ||
f8894913 JW |
990 | #define vhost_get_avail(vq, x, ptr) \ |
991 | vhost_get_user(vq, x, ptr, VHOST_ADDR_AVAIL) | |
992 | ||
993 | #define vhost_get_used(vq, x, ptr) \ | |
994 | vhost_get_user(vq, x, ptr, VHOST_ADDR_USED) | |
995 | ||
86a07da3 JW |
996 | static void vhost_dev_lock_vqs(struct vhost_dev *d) |
997 | { | |
998 | int i = 0; | |
999 | for (i = 0; i < d->nvqs; ++i) | |
1000 | mutex_lock_nested(&d->vqs[i]->mutex, i); | |
1001 | } | |
1002 | ||
1003 | static void vhost_dev_unlock_vqs(struct vhost_dev *d) | |
1004 | { | |
1005 | int i = 0; | |
1006 | for (i = 0; i < d->nvqs; ++i) | |
1007 | mutex_unlock(&d->vqs[i]->mutex); | |
1008 | } | |
1009 | ||
7b5d753e JW |
1010 | static inline int vhost_get_avail_idx(struct vhost_virtqueue *vq, |
1011 | __virtio16 *idx) | |
1012 | { | |
1013 | return vhost_get_avail(vq, *idx, &vq->avail->idx); | |
1014 | } | |
1015 | ||
1016 | static inline int vhost_get_avail_head(struct vhost_virtqueue *vq, | |
1017 | __virtio16 *head, int idx) | |
1018 | { | |
1019 | return vhost_get_avail(vq, *head, | |
1020 | &vq->avail->ring[idx & (vq->num - 1)]); | |
1021 | } | |
1022 | ||
1023 | static inline int vhost_get_avail_flags(struct vhost_virtqueue *vq, | |
1024 | __virtio16 *flags) | |
1025 | { | |
1026 | return vhost_get_avail(vq, *flags, &vq->avail->flags); | |
1027 | } | |
1028 | ||
1029 | static inline int vhost_get_used_event(struct vhost_virtqueue *vq, | |
1030 | __virtio16 *event) | |
1031 | { | |
1032 | return vhost_get_avail(vq, *event, vhost_used_event(vq)); | |
1033 | } | |
1034 | ||
1035 | static inline int vhost_get_used_idx(struct vhost_virtqueue *vq, | |
1036 | __virtio16 *idx) | |
1037 | { | |
1038 | return vhost_get_used(vq, *idx, &vq->used->idx); | |
1039 | } | |
1040 | ||
1041 | static inline int vhost_get_desc(struct vhost_virtqueue *vq, | |
1042 | struct vring_desc *desc, int idx) | |
1043 | { | |
1044 | return vhost_copy_from_user(vq, desc, vq->desc + idx, sizeof(*desc)); | |
1045 | } | |
1046 | ||
6b1e6cc7 JW |
1047 | static void vhost_iotlb_notify_vq(struct vhost_dev *d, |
1048 | struct vhost_iotlb_msg *msg) | |
1049 | { | |
1050 | struct vhost_msg_node *node, *n; | |
1051 | ||
1052 | spin_lock(&d->iotlb_lock); | |
1053 | ||
1054 | list_for_each_entry_safe(node, n, &d->pending_list, node) { | |
1055 | struct vhost_iotlb_msg *vq_msg = &node->msg.iotlb; | |
1056 | if (msg->iova <= vq_msg->iova && | |
2d66f997 | 1057 | msg->iova + msg->size - 1 >= vq_msg->iova && |
6b1e6cc7 JW |
1058 | vq_msg->type == VHOST_IOTLB_MISS) { |
1059 | vhost_poll_queue(&node->vq->poll); | |
1060 | list_del(&node->node); | |
1061 | kfree(node); | |
1062 | } | |
1063 | } | |
1064 | ||
1065 | spin_unlock(&d->iotlb_lock); | |
1066 | } | |
1067 | ||
ddd3d408 | 1068 | static bool umem_access_ok(u64 uaddr, u64 size, int access) |
6b1e6cc7 JW |
1069 | { |
1070 | unsigned long a = uaddr; | |
1071 | ||
ec33d031 MT |
1072 | /* Make sure 64 bit math will not overflow. */ |
1073 | if (vhost_overflow(uaddr, size)) | |
ddd3d408 | 1074 | return false; |
ec33d031 | 1075 | |
6b1e6cc7 | 1076 | if ((access & VHOST_ACCESS_RO) && |
96d4f267 | 1077 | !access_ok((void __user *)a, size)) |
ddd3d408 | 1078 | return false; |
6b1e6cc7 | 1079 | if ((access & VHOST_ACCESS_WO) && |
96d4f267 | 1080 | !access_ok((void __user *)a, size)) |
ddd3d408 SH |
1081 | return false; |
1082 | return true; | |
6b1e6cc7 JW |
1083 | } |
1084 | ||
91233ad7 | 1085 | static int vhost_process_iotlb_msg(struct vhost_dev *dev, u32 asid, |
72952cc0 | 1086 | struct vhost_iotlb_msg *msg) |
6b1e6cc7 JW |
1087 | { |
1088 | int ret = 0; | |
1089 | ||
91233ad7 GD |
1090 | if (asid != 0) |
1091 | return -EINVAL; | |
1092 | ||
1b15ad68 | 1093 | mutex_lock(&dev->mutex); |
86a07da3 | 1094 | vhost_dev_lock_vqs(dev); |
6b1e6cc7 JW |
1095 | switch (msg->type) { |
1096 | case VHOST_IOTLB_UPDATE: | |
1097 | if (!dev->iotlb) { | |
1098 | ret = -EFAULT; | |
1099 | break; | |
1100 | } | |
ddd3d408 | 1101 | if (!umem_access_ok(msg->uaddr, msg->size, msg->perm)) { |
6b1e6cc7 JW |
1102 | ret = -EFAULT; |
1103 | break; | |
1104 | } | |
f8894913 | 1105 | vhost_vq_meta_reset(dev); |
0bbe3066 JW |
1106 | if (vhost_iotlb_add_range(dev->iotlb, msg->iova, |
1107 | msg->iova + msg->size - 1, | |
1108 | msg->uaddr, msg->perm)) { | |
6b1e6cc7 JW |
1109 | ret = -ENOMEM; |
1110 | break; | |
1111 | } | |
1112 | vhost_iotlb_notify_vq(dev, msg); | |
1113 | break; | |
1114 | case VHOST_IOTLB_INVALIDATE: | |
6f3180af JW |
1115 | if (!dev->iotlb) { |
1116 | ret = -EFAULT; | |
1117 | break; | |
1118 | } | |
f8894913 | 1119 | vhost_vq_meta_reset(dev); |
0bbe3066 JW |
1120 | vhost_iotlb_del_range(dev->iotlb, msg->iova, |
1121 | msg->iova + msg->size - 1); | |
6b1e6cc7 JW |
1122 | break; |
1123 | default: | |
1124 | ret = -EINVAL; | |
1125 | break; | |
1126 | } | |
1127 | ||
86a07da3 | 1128 | vhost_dev_unlock_vqs(dev); |
1b15ad68 JW |
1129 | mutex_unlock(&dev->mutex); |
1130 | ||
6b1e6cc7 JW |
1131 | return ret; |
1132 | } | |
1133 | ssize_t vhost_chr_write_iter(struct vhost_dev *dev, | |
1134 | struct iov_iter *from) | |
1135 | { | |
429711ae JW |
1136 | struct vhost_iotlb_msg msg; |
1137 | size_t offset; | |
1138 | int type, ret; | |
91233ad7 | 1139 | u32 asid = 0; |
6b1e6cc7 | 1140 | |
429711ae | 1141 | ret = copy_from_iter(&type, sizeof(type), from); |
74ad7419 PT |
1142 | if (ret != sizeof(type)) { |
1143 | ret = -EINVAL; | |
6b1e6cc7 | 1144 | goto done; |
74ad7419 | 1145 | } |
6b1e6cc7 | 1146 | |
429711ae | 1147 | switch (type) { |
6b1e6cc7 | 1148 | case VHOST_IOTLB_MSG: |
429711ae JW |
1149 | /* There maybe a hole after type for V1 message type, |
1150 | * so skip it here. | |
1151 | */ | |
1152 | offset = offsetof(struct vhost_msg, iotlb) - sizeof(int); | |
1153 | break; | |
1154 | case VHOST_IOTLB_MSG_V2: | |
91233ad7 GD |
1155 | if (vhost_backend_has_feature(dev->vqs[0], |
1156 | VHOST_BACKEND_F_IOTLB_ASID)) { | |
1157 | ret = copy_from_iter(&asid, sizeof(asid), from); | |
1158 | if (ret != sizeof(asid)) { | |
1159 | ret = -EINVAL; | |
1160 | goto done; | |
1161 | } | |
aaca8373 | 1162 | offset = 0; |
91233ad7 GD |
1163 | } else |
1164 | offset = sizeof(__u32); | |
6b1e6cc7 JW |
1165 | break; |
1166 | default: | |
1167 | ret = -EINVAL; | |
429711ae | 1168 | goto done; |
6b1e6cc7 JW |
1169 | } |
1170 | ||
429711ae JW |
1171 | iov_iter_advance(from, offset); |
1172 | ret = copy_from_iter(&msg, sizeof(msg), from); | |
74ad7419 PT |
1173 | if (ret != sizeof(msg)) { |
1174 | ret = -EINVAL; | |
429711ae | 1175 | goto done; |
74ad7419 | 1176 | } |
792a4f2e | 1177 | |
95932ab2 JW |
1178 | if ((msg.type == VHOST_IOTLB_UPDATE || |
1179 | msg.type == VHOST_IOTLB_INVALIDATE) && | |
1180 | msg.size == 0) { | |
e2ae38cf AR |
1181 | ret = -EINVAL; |
1182 | goto done; | |
1183 | } | |
1184 | ||
792a4f2e | 1185 | if (dev->msg_handler) |
91233ad7 | 1186 | ret = dev->msg_handler(dev, asid, &msg); |
792a4f2e | 1187 | else |
91233ad7 | 1188 | ret = vhost_process_iotlb_msg(dev, asid, &msg); |
792a4f2e | 1189 | if (ret) { |
429711ae JW |
1190 | ret = -EFAULT; |
1191 | goto done; | |
1192 | } | |
1193 | ||
1194 | ret = (type == VHOST_IOTLB_MSG) ? sizeof(struct vhost_msg) : | |
1195 | sizeof(struct vhost_msg_v2); | |
6b1e6cc7 JW |
1196 | done: |
1197 | return ret; | |
1198 | } | |
1199 | EXPORT_SYMBOL(vhost_chr_write_iter); | |
1200 | ||
afc9a42b | 1201 | __poll_t vhost_chr_poll(struct file *file, struct vhost_dev *dev, |
6b1e6cc7 JW |
1202 | poll_table *wait) |
1203 | { | |
afc9a42b | 1204 | __poll_t mask = 0; |
6b1e6cc7 JW |
1205 | |
1206 | poll_wait(file, &dev->wait, wait); | |
1207 | ||
1208 | if (!list_empty(&dev->read_list)) | |
a9a08845 | 1209 | mask |= EPOLLIN | EPOLLRDNORM; |
6b1e6cc7 JW |
1210 | |
1211 | return mask; | |
1212 | } | |
1213 | EXPORT_SYMBOL(vhost_chr_poll); | |
1214 | ||
1215 | ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to, | |
1216 | int noblock) | |
1217 | { | |
1218 | DEFINE_WAIT(wait); | |
1219 | struct vhost_msg_node *node; | |
1220 | ssize_t ret = 0; | |
1221 | unsigned size = sizeof(struct vhost_msg); | |
1222 | ||
1223 | if (iov_iter_count(to) < size) | |
1224 | return 0; | |
1225 | ||
1226 | while (1) { | |
1227 | if (!noblock) | |
1228 | prepare_to_wait(&dev->wait, &wait, | |
1229 | TASK_INTERRUPTIBLE); | |
1230 | ||
1231 | node = vhost_dequeue_msg(dev, &dev->read_list); | |
1232 | if (node) | |
1233 | break; | |
1234 | if (noblock) { | |
1235 | ret = -EAGAIN; | |
1236 | break; | |
1237 | } | |
1238 | if (signal_pending(current)) { | |
1239 | ret = -ERESTARTSYS; | |
1240 | break; | |
1241 | } | |
1242 | if (!dev->iotlb) { | |
1243 | ret = -EBADFD; | |
1244 | break; | |
1245 | } | |
1246 | ||
1247 | schedule(); | |
1248 | } | |
1249 | ||
1250 | if (!noblock) | |
1251 | finish_wait(&dev->wait, &wait); | |
1252 | ||
1253 | if (node) { | |
429711ae JW |
1254 | struct vhost_iotlb_msg *msg; |
1255 | void *start = &node->msg; | |
6b1e6cc7 | 1256 | |
429711ae JW |
1257 | switch (node->msg.type) { |
1258 | case VHOST_IOTLB_MSG: | |
1259 | size = sizeof(node->msg); | |
1260 | msg = &node->msg.iotlb; | |
1261 | break; | |
1262 | case VHOST_IOTLB_MSG_V2: | |
1263 | size = sizeof(node->msg_v2); | |
1264 | msg = &node->msg_v2.iotlb; | |
1265 | break; | |
1266 | default: | |
1267 | BUG(); | |
1268 | break; | |
1269 | } | |
1270 | ||
1271 | ret = copy_to_iter(start, size, to); | |
1272 | if (ret != size || msg->type != VHOST_IOTLB_MISS) { | |
6b1e6cc7 JW |
1273 | kfree(node); |
1274 | return ret; | |
1275 | } | |
6b1e6cc7 JW |
1276 | vhost_enqueue_msg(dev, &dev->pending_list, node); |
1277 | } | |
1278 | ||
1279 | return ret; | |
1280 | } | |
1281 | EXPORT_SYMBOL_GPL(vhost_chr_read_iter); | |
1282 | ||
1283 | static int vhost_iotlb_miss(struct vhost_virtqueue *vq, u64 iova, int access) | |
1284 | { | |
1285 | struct vhost_dev *dev = vq->dev; | |
1286 | struct vhost_msg_node *node; | |
1287 | struct vhost_iotlb_msg *msg; | |
429711ae | 1288 | bool v2 = vhost_backend_has_feature(vq, VHOST_BACKEND_F_IOTLB_MSG_V2); |
6b1e6cc7 | 1289 | |
429711ae | 1290 | node = vhost_new_msg(vq, v2 ? VHOST_IOTLB_MSG_V2 : VHOST_IOTLB_MSG); |
6b1e6cc7 JW |
1291 | if (!node) |
1292 | return -ENOMEM; | |
1293 | ||
429711ae JW |
1294 | if (v2) { |
1295 | node->msg_v2.type = VHOST_IOTLB_MSG_V2; | |
1296 | msg = &node->msg_v2.iotlb; | |
1297 | } else { | |
1298 | msg = &node->msg.iotlb; | |
1299 | } | |
1300 | ||
6b1e6cc7 JW |
1301 | msg->type = VHOST_IOTLB_MISS; |
1302 | msg->iova = iova; | |
1303 | msg->perm = access; | |
1304 | ||
1305 | vhost_enqueue_msg(dev, &dev->read_list, node); | |
1306 | ||
1307 | return 0; | |
bfe2bc51 JW |
1308 | } |
1309 | ||
ddd3d408 | 1310 | static bool vq_access_ok(struct vhost_virtqueue *vq, unsigned int num, |
a865e420 MT |
1311 | vring_desc_t __user *desc, |
1312 | vring_avail_t __user *avail, | |
1313 | vring_used_t __user *used) | |
6b1e6cc7 | 1314 | |
3a4d5c94 | 1315 | { |
0210a8db GK |
1316 | /* If an IOTLB device is present, the vring addresses are |
1317 | * GIOVAs. Access validation occurs at prefetch time. */ | |
1318 | if (vq->iotlb) | |
1319 | return true; | |
1320 | ||
4942e825 JW |
1321 | return access_ok(desc, vhost_get_desc_size(vq, num)) && |
1322 | access_ok(avail, vhost_get_avail_size(vq, num)) && | |
1323 | access_ok(used, vhost_get_used_size(vq, num)); | |
3a4d5c94 MT |
1324 | } |
1325 | ||
f8894913 | 1326 | static void vhost_vq_meta_update(struct vhost_virtqueue *vq, |
0bbe3066 | 1327 | const struct vhost_iotlb_map *map, |
f8894913 JW |
1328 | int type) |
1329 | { | |
1330 | int access = (type == VHOST_ADDR_USED) ? | |
1331 | VHOST_ACCESS_WO : VHOST_ACCESS_RO; | |
1332 | ||
0bbe3066 JW |
1333 | if (likely(map->perm & access)) |
1334 | vq->meta_iotlb[type] = map; | |
f8894913 JW |
1335 | } |
1336 | ||
ddd3d408 SH |
1337 | static bool iotlb_access_ok(struct vhost_virtqueue *vq, |
1338 | int access, u64 addr, u64 len, int type) | |
6b1e6cc7 | 1339 | { |
0bbe3066 JW |
1340 | const struct vhost_iotlb_map *map; |
1341 | struct vhost_iotlb *umem = vq->iotlb; | |
ca2c5b33 | 1342 | u64 s = 0, size, orig_addr = addr, last = addr + len - 1; |
f8894913 JW |
1343 | |
1344 | if (vhost_vq_meta_fetch(vq, addr, len, type)) | |
1345 | return true; | |
6b1e6cc7 JW |
1346 | |
1347 | while (len > s) { | |
0bbe3066 JW |
1348 | map = vhost_iotlb_itree_first(umem, addr, last); |
1349 | if (map == NULL || map->start > addr) { | |
6b1e6cc7 JW |
1350 | vhost_iotlb_miss(vq, addr, access); |
1351 | return false; | |
0bbe3066 | 1352 | } else if (!(map->perm & access)) { |
6b1e6cc7 JW |
1353 | /* Report the possible access violation by |
1354 | * request another translation from userspace. | |
1355 | */ | |
1356 | return false; | |
1357 | } | |
1358 | ||
0bbe3066 | 1359 | size = map->size - addr + map->start; |
f8894913 JW |
1360 | |
1361 | if (orig_addr == addr && size >= len) | |
0bbe3066 | 1362 | vhost_vq_meta_update(vq, map, type); |
f8894913 | 1363 | |
6b1e6cc7 JW |
1364 | s += size; |
1365 | addr += size; | |
1366 | } | |
1367 | ||
1368 | return true; | |
1369 | } | |
1370 | ||
9b5e830b | 1371 | int vq_meta_prefetch(struct vhost_virtqueue *vq) |
6b1e6cc7 | 1372 | { |
6b1e6cc7 JW |
1373 | unsigned int num = vq->num; |
1374 | ||
3d2c7d37 | 1375 | if (!vq->iotlb) |
6b1e6cc7 JW |
1376 | return 1; |
1377 | ||
0bbe3066 | 1378 | return iotlb_access_ok(vq, VHOST_MAP_RO, (u64)(uintptr_t)vq->desc, |
4942e825 | 1379 | vhost_get_desc_size(vq, num), VHOST_ADDR_DESC) && |
0bbe3066 | 1380 | iotlb_access_ok(vq, VHOST_MAP_RO, (u64)(uintptr_t)vq->avail, |
4942e825 | 1381 | vhost_get_avail_size(vq, num), |
f8894913 | 1382 | VHOST_ADDR_AVAIL) && |
0bbe3066 | 1383 | iotlb_access_ok(vq, VHOST_MAP_WO, (u64)(uintptr_t)vq->used, |
4942e825 | 1384 | vhost_get_used_size(vq, num), VHOST_ADDR_USED); |
6b1e6cc7 | 1385 | } |
9b5e830b | 1386 | EXPORT_SYMBOL_GPL(vq_meta_prefetch); |
6b1e6cc7 | 1387 | |
3a4d5c94 MT |
1388 | /* Can we log writes? */ |
1389 | /* Caller should have device mutex but not vq mutex */ | |
ddd3d408 | 1390 | bool vhost_log_access_ok(struct vhost_dev *dev) |
3a4d5c94 | 1391 | { |
a9709d68 | 1392 | return memory_access_ok(dev, dev->umem, 1); |
3a4d5c94 | 1393 | } |
6ac1afbf | 1394 | EXPORT_SYMBOL_GPL(vhost_log_access_ok); |
3a4d5c94 | 1395 | |
ab512251 GK |
1396 | static bool vq_log_used_access_ok(struct vhost_virtqueue *vq, |
1397 | void __user *log_base, | |
1398 | bool log_used, | |
1399 | u64 log_addr) | |
1400 | { | |
1401 | /* If an IOTLB device is present, log_addr is a GIOVA that | |
1402 | * will never be logged by log_used(). */ | |
1403 | if (vq->iotlb) | |
1404 | return true; | |
1405 | ||
1406 | return !log_used || log_access_ok(log_base, log_addr, | |
1407 | vhost_get_used_size(vq, vq->num)); | |
1408 | } | |
1409 | ||
3a4d5c94 MT |
1410 | /* Verify access for write logging. */ |
1411 | /* Caller should have vq mutex and device mutex */ | |
ddd3d408 SH |
1412 | static bool vq_log_access_ok(struct vhost_virtqueue *vq, |
1413 | void __user *log_base) | |
3a4d5c94 | 1414 | { |
a9709d68 | 1415 | return vq_memory_access_ok(log_base, vq->umem, |
ea16c514 | 1416 | vhost_has_feature(vq, VHOST_F_LOG_ALL)) && |
ab512251 | 1417 | vq_log_used_access_ok(vq, log_base, vq->log_used, vq->log_addr); |
3a4d5c94 MT |
1418 | } |
1419 | ||
1420 | /* Can we start vq? */ | |
1421 | /* Caller should have vq mutex and device mutex */ | |
ddd3d408 | 1422 | bool vhost_vq_access_ok(struct vhost_virtqueue *vq) |
3a4d5c94 | 1423 | { |
d14d2b78 | 1424 | if (!vq_log_access_ok(vq, vq->log_base)) |
ddd3d408 | 1425 | return false; |
d65026c6 | 1426 | |
d65026c6 | 1427 | return vq_access_ok(vq, vq->num, vq->desc, vq->avail, vq->used); |
3a4d5c94 | 1428 | } |
6ac1afbf | 1429 | EXPORT_SYMBOL_GPL(vhost_vq_access_ok); |
3a4d5c94 MT |
1430 | |
1431 | static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m) | |
1432 | { | |
a9709d68 JW |
1433 | struct vhost_memory mem, *newmem; |
1434 | struct vhost_memory_region *region; | |
0bbe3066 | 1435 | struct vhost_iotlb *newumem, *oldumem; |
3a4d5c94 | 1436 | unsigned long size = offsetof(struct vhost_memory, regions); |
98f9ca0a | 1437 | int i; |
d47effe1 | 1438 | |
7ad9c9d2 TY |
1439 | if (copy_from_user(&mem, m, size)) |
1440 | return -EFAULT; | |
3a4d5c94 MT |
1441 | if (mem.padding) |
1442 | return -EOPNOTSUPP; | |
c9ce42f7 | 1443 | if (mem.nregions > max_mem_regions) |
3a4d5c94 | 1444 | return -E2BIG; |
b2303d7b MW |
1445 | newmem = kvzalloc(struct_size(newmem, regions, mem.nregions), |
1446 | GFP_KERNEL); | |
3a4d5c94 MT |
1447 | if (!newmem) |
1448 | return -ENOMEM; | |
1449 | ||
1450 | memcpy(newmem, &mem, size); | |
7ad9c9d2 | 1451 | if (copy_from_user(newmem->regions, m->regions, |
bf11d71a | 1452 | flex_array_size(newmem, regions, mem.nregions))) { |
bcfeacab | 1453 | kvfree(newmem); |
7ad9c9d2 | 1454 | return -EFAULT; |
3a4d5c94 MT |
1455 | } |
1456 | ||
0bbe3066 | 1457 | newumem = iotlb_alloc(); |
a9709d68 | 1458 | if (!newumem) { |
4de7255f | 1459 | kvfree(newmem); |
a9709d68 JW |
1460 | return -ENOMEM; |
1461 | } | |
1462 | ||
a9709d68 JW |
1463 | for (region = newmem->regions; |
1464 | region < newmem->regions + mem.nregions; | |
1465 | region++) { | |
0bbe3066 JW |
1466 | if (vhost_iotlb_add_range(newumem, |
1467 | region->guest_phys_addr, | |
1468 | region->guest_phys_addr + | |
1469 | region->memory_size - 1, | |
1470 | region->userspace_addr, | |
1471 | VHOST_MAP_RW)) | |
a9709d68 | 1472 | goto err; |
a02c3789 | 1473 | } |
a9709d68 JW |
1474 | |
1475 | if (!memory_access_ok(d, newumem, 0)) | |
1476 | goto err; | |
1477 | ||
1478 | oldumem = d->umem; | |
1479 | d->umem = newumem; | |
98f9ca0a | 1480 | |
47283bef | 1481 | /* All memory accesses are done under some VQ mutex. */ |
98f9ca0a MT |
1482 | for (i = 0; i < d->nvqs; ++i) { |
1483 | mutex_lock(&d->vqs[i]->mutex); | |
a9709d68 | 1484 | d->vqs[i]->umem = newumem; |
98f9ca0a MT |
1485 | mutex_unlock(&d->vqs[i]->mutex); |
1486 | } | |
a9709d68 JW |
1487 | |
1488 | kvfree(newmem); | |
0bbe3066 | 1489 | vhost_iotlb_free(oldumem); |
3a4d5c94 | 1490 | return 0; |
a9709d68 JW |
1491 | |
1492 | err: | |
0bbe3066 | 1493 | vhost_iotlb_free(newumem); |
a9709d68 JW |
1494 | kvfree(newmem); |
1495 | return -EFAULT; | |
3a4d5c94 MT |
1496 | } |
1497 | ||
feebcaea JW |
1498 | static long vhost_vring_set_num(struct vhost_dev *d, |
1499 | struct vhost_virtqueue *vq, | |
1500 | void __user *argp) | |
1501 | { | |
1502 | struct vhost_vring_state s; | |
1503 | ||
1504 | /* Resizing ring with an active backend? | |
1505 | * You don't want to do that. */ | |
1506 | if (vq->private_data) | |
1507 | return -EBUSY; | |
1508 | ||
1509 | if (copy_from_user(&s, argp, sizeof s)) | |
1510 | return -EFAULT; | |
1511 | ||
1512 | if (!s.num || s.num > 0xffff || (s.num & (s.num - 1))) | |
1513 | return -EINVAL; | |
1514 | vq->num = s.num; | |
1515 | ||
1516 | return 0; | |
1517 | } | |
1518 | ||
1519 | static long vhost_vring_set_addr(struct vhost_dev *d, | |
1520 | struct vhost_virtqueue *vq, | |
1521 | void __user *argp) | |
1522 | { | |
1523 | struct vhost_vring_addr a; | |
1524 | ||
1525 | if (copy_from_user(&a, argp, sizeof a)) | |
1526 | return -EFAULT; | |
1527 | if (a.flags & ~(0x1 << VHOST_VRING_F_LOG)) | |
1528 | return -EOPNOTSUPP; | |
1529 | ||
1530 | /* For 32bit, verify that the top 32bits of the user | |
1531 | data are set to zero. */ | |
1532 | if ((u64)(unsigned long)a.desc_user_addr != a.desc_user_addr || | |
1533 | (u64)(unsigned long)a.used_user_addr != a.used_user_addr || | |
1534 | (u64)(unsigned long)a.avail_user_addr != a.avail_user_addr) | |
1535 | return -EFAULT; | |
1536 | ||
1537 | /* Make sure it's safe to cast pointers to vring types. */ | |
1538 | BUILD_BUG_ON(__alignof__ *vq->avail > VRING_AVAIL_ALIGN_SIZE); | |
1539 | BUILD_BUG_ON(__alignof__ *vq->used > VRING_USED_ALIGN_SIZE); | |
1540 | if ((a.avail_user_addr & (VRING_AVAIL_ALIGN_SIZE - 1)) || | |
1541 | (a.used_user_addr & (VRING_USED_ALIGN_SIZE - 1)) || | |
1542 | (a.log_guest_addr & (VRING_USED_ALIGN_SIZE - 1))) | |
1543 | return -EINVAL; | |
1544 | ||
1545 | /* We only verify access here if backend is configured. | |
1546 | * If it is not, we don't as size might not have been setup. | |
1547 | * We will verify when backend is configured. */ | |
1548 | if (vq->private_data) { | |
1549 | if (!vq_access_ok(vq, vq->num, | |
1550 | (void __user *)(unsigned long)a.desc_user_addr, | |
1551 | (void __user *)(unsigned long)a.avail_user_addr, | |
1552 | (void __user *)(unsigned long)a.used_user_addr)) | |
1553 | return -EINVAL; | |
1554 | ||
1555 | /* Also validate log access for used ring if enabled. */ | |
ab512251 GK |
1556 | if (!vq_log_used_access_ok(vq, vq->log_base, |
1557 | a.flags & (0x1 << VHOST_VRING_F_LOG), | |
1558 | a.log_guest_addr)) | |
feebcaea JW |
1559 | return -EINVAL; |
1560 | } | |
1561 | ||
1562 | vq->log_used = !!(a.flags & (0x1 << VHOST_VRING_F_LOG)); | |
1563 | vq->desc = (void __user *)(unsigned long)a.desc_user_addr; | |
1564 | vq->avail = (void __user *)(unsigned long)a.avail_user_addr; | |
1565 | vq->log_addr = a.log_guest_addr; | |
1566 | vq->used = (void __user *)(unsigned long)a.used_user_addr; | |
1567 | ||
1568 | return 0; | |
1569 | } | |
1570 | ||
1571 | static long vhost_vring_set_num_addr(struct vhost_dev *d, | |
1572 | struct vhost_virtqueue *vq, | |
1573 | unsigned int ioctl, | |
1574 | void __user *argp) | |
1575 | { | |
1576 | long r; | |
1577 | ||
1578 | mutex_lock(&vq->mutex); | |
1579 | ||
1580 | switch (ioctl) { | |
1581 | case VHOST_SET_VRING_NUM: | |
1582 | r = vhost_vring_set_num(d, vq, argp); | |
1583 | break; | |
1584 | case VHOST_SET_VRING_ADDR: | |
1585 | r = vhost_vring_set_addr(d, vq, argp); | |
1586 | break; | |
1587 | default: | |
1588 | BUG(); | |
1589 | } | |
1590 | ||
1591 | mutex_unlock(&vq->mutex); | |
1592 | ||
1593 | return r; | |
1594 | } | |
26b36604 | 1595 | long vhost_vring_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp) |
3a4d5c94 | 1596 | { |
cecb46f1 AV |
1597 | struct file *eventfp, *filep = NULL; |
1598 | bool pollstart = false, pollstop = false; | |
3a4d5c94 MT |
1599 | struct eventfd_ctx *ctx = NULL; |
1600 | u32 __user *idxp = argp; | |
1601 | struct vhost_virtqueue *vq; | |
1602 | struct vhost_vring_state s; | |
1603 | struct vhost_vring_file f; | |
3a4d5c94 MT |
1604 | u32 idx; |
1605 | long r; | |
1606 | ||
1607 | r = get_user(idx, idxp); | |
1608 | if (r < 0) | |
1609 | return r; | |
0f3d9a17 | 1610 | if (idx >= d->nvqs) |
3a4d5c94 MT |
1611 | return -ENOBUFS; |
1612 | ||
ff002269 | 1613 | idx = array_index_nospec(idx, d->nvqs); |
3ab2e420 | 1614 | vq = d->vqs[idx]; |
3a4d5c94 | 1615 | |
feebcaea JW |
1616 | if (ioctl == VHOST_SET_VRING_NUM || |
1617 | ioctl == VHOST_SET_VRING_ADDR) { | |
1618 | return vhost_vring_set_num_addr(d, vq, ioctl, argp); | |
1619 | } | |
1620 | ||
3a4d5c94 MT |
1621 | mutex_lock(&vq->mutex); |
1622 | ||
1623 | switch (ioctl) { | |
3a4d5c94 MT |
1624 | case VHOST_SET_VRING_BASE: |
1625 | /* Moving base with an active backend? | |
1626 | * You don't want to do that. */ | |
1627 | if (vq->private_data) { | |
1628 | r = -EBUSY; | |
1629 | break; | |
1630 | } | |
7ad9c9d2 TY |
1631 | if (copy_from_user(&s, argp, sizeof s)) { |
1632 | r = -EFAULT; | |
3a4d5c94 | 1633 | break; |
7ad9c9d2 | 1634 | } |
3a4d5c94 MT |
1635 | if (s.num > 0xffff) { |
1636 | r = -EINVAL; | |
1637 | break; | |
1638 | } | |
8d65843c | 1639 | vq->last_avail_idx = s.num; |
3a4d5c94 MT |
1640 | /* Forget the cached index value. */ |
1641 | vq->avail_idx = vq->last_avail_idx; | |
1642 | break; | |
1643 | case VHOST_GET_VRING_BASE: | |
1644 | s.index = idx; | |
1645 | s.num = vq->last_avail_idx; | |
7ad9c9d2 TY |
1646 | if (copy_to_user(argp, &s, sizeof s)) |
1647 | r = -EFAULT; | |
3a4d5c94 | 1648 | break; |
3a4d5c94 | 1649 | case VHOST_SET_VRING_KICK: |
7ad9c9d2 TY |
1650 | if (copy_from_user(&f, argp, sizeof f)) { |
1651 | r = -EFAULT; | |
3a4d5c94 | 1652 | break; |
7ad9c9d2 | 1653 | } |
e0136c16 | 1654 | eventfp = f.fd == VHOST_FILE_UNBIND ? NULL : eventfd_fget(f.fd); |
535297a6 MT |
1655 | if (IS_ERR(eventfp)) { |
1656 | r = PTR_ERR(eventfp); | |
1657 | break; | |
1658 | } | |
3a4d5c94 | 1659 | if (eventfp != vq->kick) { |
cecb46f1 AV |
1660 | pollstop = (filep = vq->kick) != NULL; |
1661 | pollstart = (vq->kick = eventfp) != NULL; | |
3a4d5c94 MT |
1662 | } else |
1663 | filep = eventfp; | |
1664 | break; | |
1665 | case VHOST_SET_VRING_CALL: | |
7ad9c9d2 TY |
1666 | if (copy_from_user(&f, argp, sizeof f)) { |
1667 | r = -EFAULT; | |
3a4d5c94 | 1668 | break; |
7ad9c9d2 | 1669 | } |
e0136c16 | 1670 | ctx = f.fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(f.fd); |
e050c7d9 EB |
1671 | if (IS_ERR(ctx)) { |
1672 | r = PTR_ERR(ctx); | |
535297a6 MT |
1673 | break; |
1674 | } | |
265a0ad8 | 1675 | |
265a0ad8 | 1676 | swap(ctx, vq->call_ctx.ctx); |
3a4d5c94 MT |
1677 | break; |
1678 | case VHOST_SET_VRING_ERR: | |
7ad9c9d2 TY |
1679 | if (copy_from_user(&f, argp, sizeof f)) { |
1680 | r = -EFAULT; | |
3a4d5c94 | 1681 | break; |
7ad9c9d2 | 1682 | } |
e0136c16 | 1683 | ctx = f.fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(f.fd); |
09f332a5 EB |
1684 | if (IS_ERR(ctx)) { |
1685 | r = PTR_ERR(ctx); | |
535297a6 MT |
1686 | break; |
1687 | } | |
09f332a5 | 1688 | swap(ctx, vq->error_ctx); |
3a4d5c94 | 1689 | break; |
2751c988 GK |
1690 | case VHOST_SET_VRING_ENDIAN: |
1691 | r = vhost_set_vring_endian(vq, argp); | |
1692 | break; | |
1693 | case VHOST_GET_VRING_ENDIAN: | |
1694 | r = vhost_get_vring_endian(vq, idx, argp); | |
1695 | break; | |
03088137 JW |
1696 | case VHOST_SET_VRING_BUSYLOOP_TIMEOUT: |
1697 | if (copy_from_user(&s, argp, sizeof(s))) { | |
1698 | r = -EFAULT; | |
1699 | break; | |
1700 | } | |
1701 | vq->busyloop_timeout = s.num; | |
1702 | break; | |
1703 | case VHOST_GET_VRING_BUSYLOOP_TIMEOUT: | |
1704 | s.index = idx; | |
1705 | s.num = vq->busyloop_timeout; | |
1706 | if (copy_to_user(argp, &s, sizeof(s))) | |
1707 | r = -EFAULT; | |
1708 | break; | |
3a4d5c94 MT |
1709 | default: |
1710 | r = -ENOIOCTLCMD; | |
1711 | } | |
1712 | ||
1713 | if (pollstop && vq->handle_kick) | |
1714 | vhost_poll_stop(&vq->poll); | |
1715 | ||
e050c7d9 | 1716 | if (!IS_ERR_OR_NULL(ctx)) |
3a4d5c94 MT |
1717 | eventfd_ctx_put(ctx); |
1718 | if (filep) | |
1719 | fput(filep); | |
1720 | ||
1721 | if (pollstart && vq->handle_kick) | |
2b8b328b | 1722 | r = vhost_poll_start(&vq->poll, vq->kick); |
3a4d5c94 MT |
1723 | |
1724 | mutex_unlock(&vq->mutex); | |
1725 | ||
1726 | if (pollstop && vq->handle_kick) | |
b2ffa407 | 1727 | vhost_dev_flush(vq->poll.dev); |
3a4d5c94 MT |
1728 | return r; |
1729 | } | |
6ac1afbf | 1730 | EXPORT_SYMBOL_GPL(vhost_vring_ioctl); |
3a4d5c94 | 1731 | |
6b1e6cc7 JW |
1732 | int vhost_init_device_iotlb(struct vhost_dev *d, bool enabled) |
1733 | { | |
0bbe3066 | 1734 | struct vhost_iotlb *niotlb, *oiotlb; |
6b1e6cc7 JW |
1735 | int i; |
1736 | ||
0bbe3066 | 1737 | niotlb = iotlb_alloc(); |
6b1e6cc7 JW |
1738 | if (!niotlb) |
1739 | return -ENOMEM; | |
1740 | ||
1741 | oiotlb = d->iotlb; | |
1742 | d->iotlb = niotlb; | |
1743 | ||
1744 | for (i = 0; i < d->nvqs; ++i) { | |
b13f9c63 JW |
1745 | struct vhost_virtqueue *vq = d->vqs[i]; |
1746 | ||
1747 | mutex_lock(&vq->mutex); | |
1748 | vq->iotlb = niotlb; | |
1749 | __vhost_vq_meta_reset(vq); | |
1750 | mutex_unlock(&vq->mutex); | |
6b1e6cc7 JW |
1751 | } |
1752 | ||
0bbe3066 | 1753 | vhost_iotlb_free(oiotlb); |
6b1e6cc7 JW |
1754 | |
1755 | return 0; | |
1756 | } | |
1757 | EXPORT_SYMBOL_GPL(vhost_init_device_iotlb); | |
1758 | ||
3a4d5c94 | 1759 | /* Caller must have device mutex */ |
935cdee7 | 1760 | long vhost_dev_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp) |
3a4d5c94 | 1761 | { |
d25cc43c | 1762 | struct eventfd_ctx *ctx; |
3a4d5c94 MT |
1763 | u64 p; |
1764 | long r; | |
1765 | int i, fd; | |
1766 | ||
1767 | /* If you are not the owner, you can become one */ | |
1768 | if (ioctl == VHOST_SET_OWNER) { | |
1769 | r = vhost_dev_set_owner(d); | |
1770 | goto done; | |
1771 | } | |
1772 | ||
1773 | /* You must be the owner to do anything else */ | |
1774 | r = vhost_dev_check_owner(d); | |
1775 | if (r) | |
1776 | goto done; | |
1777 | ||
1778 | switch (ioctl) { | |
1779 | case VHOST_SET_MEM_TABLE: | |
1780 | r = vhost_set_memory(d, argp); | |
1781 | break; | |
1782 | case VHOST_SET_LOG_BASE: | |
7ad9c9d2 TY |
1783 | if (copy_from_user(&p, argp, sizeof p)) { |
1784 | r = -EFAULT; | |
3a4d5c94 | 1785 | break; |
7ad9c9d2 | 1786 | } |
3a4d5c94 MT |
1787 | if ((u64)(unsigned long)p != p) { |
1788 | r = -EFAULT; | |
1789 | break; | |
1790 | } | |
1791 | for (i = 0; i < d->nvqs; ++i) { | |
1792 | struct vhost_virtqueue *vq; | |
1793 | void __user *base = (void __user *)(unsigned long)p; | |
3ab2e420 | 1794 | vq = d->vqs[i]; |
3a4d5c94 MT |
1795 | mutex_lock(&vq->mutex); |
1796 | /* If ring is inactive, will check when it's enabled. */ | |
ea16c514 | 1797 | if (vq->private_data && !vq_log_access_ok(vq, base)) |
3a4d5c94 MT |
1798 | r = -EFAULT; |
1799 | else | |
1800 | vq->log_base = base; | |
1801 | mutex_unlock(&vq->mutex); | |
1802 | } | |
1803 | break; | |
1804 | case VHOST_SET_LOG_FD: | |
1805 | r = get_user(fd, (int __user *)argp); | |
1806 | if (r < 0) | |
1807 | break; | |
e0136c16 | 1808 | ctx = fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(fd); |
d25cc43c EB |
1809 | if (IS_ERR(ctx)) { |
1810 | r = PTR_ERR(ctx); | |
3a4d5c94 MT |
1811 | break; |
1812 | } | |
d25cc43c | 1813 | swap(ctx, d->log_ctx); |
3a4d5c94 | 1814 | for (i = 0; i < d->nvqs; ++i) { |
3ab2e420 AH |
1815 | mutex_lock(&d->vqs[i]->mutex); |
1816 | d->vqs[i]->log_ctx = d->log_ctx; | |
1817 | mutex_unlock(&d->vqs[i]->mutex); | |
3a4d5c94 MT |
1818 | } |
1819 | if (ctx) | |
1820 | eventfd_ctx_put(ctx); | |
3a4d5c94 MT |
1821 | break; |
1822 | default: | |
935cdee7 | 1823 | r = -ENOIOCTLCMD; |
3a4d5c94 MT |
1824 | break; |
1825 | } | |
1826 | done: | |
1827 | return r; | |
1828 | } | |
6ac1afbf | 1829 | EXPORT_SYMBOL_GPL(vhost_dev_ioctl); |
3a4d5c94 | 1830 | |
3a4d5c94 MT |
1831 | /* TODO: This is really inefficient. We need something like get_user() |
1832 | * (instruction directly accesses the data, with an exception table entry | |
cb1aaebe | 1833 | * returning -EFAULT). See Documentation/x86/exception-tables.rst. |
3a4d5c94 MT |
1834 | */ |
1835 | static int set_bit_to_user(int nr, void __user *addr) | |
1836 | { | |
1837 | unsigned long log = (unsigned long)addr; | |
1838 | struct page *page; | |
1839 | void *base; | |
1840 | int bit = nr + (log % PAGE_SIZE) * 8; | |
1841 | int r; | |
d47effe1 | 1842 | |
690623e1 | 1843 | r = pin_user_pages_fast(log, 1, FOLL_WRITE, &page); |
d6db3f5c | 1844 | if (r < 0) |
3a4d5c94 | 1845 | return r; |
d6db3f5c | 1846 | BUG_ON(r != 1); |
c6daa7ff | 1847 | base = kmap_atomic(page); |
3a4d5c94 | 1848 | set_bit(bit, base); |
c6daa7ff | 1849 | kunmap_atomic(base); |
690623e1 | 1850 | unpin_user_pages_dirty_lock(&page, 1, true); |
3a4d5c94 MT |
1851 | return 0; |
1852 | } | |
1853 | ||
1854 | static int log_write(void __user *log_base, | |
1855 | u64 write_address, u64 write_length) | |
1856 | { | |
28831ee6 | 1857 | u64 write_page = write_address / VHOST_PAGE_SIZE; |
3a4d5c94 | 1858 | int r; |
d47effe1 | 1859 | |
3a4d5c94 MT |
1860 | if (!write_length) |
1861 | return 0; | |
3bf9be40 | 1862 | write_length += write_address % VHOST_PAGE_SIZE; |
3a4d5c94 MT |
1863 | for (;;) { |
1864 | u64 base = (u64)(unsigned long)log_base; | |
28831ee6 MT |
1865 | u64 log = base + write_page / 8; |
1866 | int bit = write_page % 8; | |
3a4d5c94 MT |
1867 | if ((u64)(unsigned long)log != log) |
1868 | return -EFAULT; | |
1869 | r = set_bit_to_user(bit, (void __user *)(unsigned long)log); | |
1870 | if (r < 0) | |
1871 | return r; | |
1872 | if (write_length <= VHOST_PAGE_SIZE) | |
1873 | break; | |
1874 | write_length -= VHOST_PAGE_SIZE; | |
28831ee6 | 1875 | write_page += 1; |
3a4d5c94 MT |
1876 | } |
1877 | return r; | |
1878 | } | |
1879 | ||
cc5e7107 JW |
1880 | static int log_write_hva(struct vhost_virtqueue *vq, u64 hva, u64 len) |
1881 | { | |
0bbe3066 JW |
1882 | struct vhost_iotlb *umem = vq->umem; |
1883 | struct vhost_iotlb_map *u; | |
cc5e7107 JW |
1884 | u64 start, end, l, min; |
1885 | int r; | |
1886 | bool hit = false; | |
1887 | ||
1888 | while (len) { | |
1889 | min = len; | |
1890 | /* More than one GPAs can be mapped into a single HVA. So | |
1891 | * iterate all possible umems here to be safe. | |
1892 | */ | |
0bbe3066 JW |
1893 | list_for_each_entry(u, &umem->list, link) { |
1894 | if (u->addr > hva - 1 + len || | |
1895 | u->addr - 1 + u->size < hva) | |
cc5e7107 | 1896 | continue; |
0bbe3066 JW |
1897 | start = max(u->addr, hva); |
1898 | end = min(u->addr - 1 + u->size, hva - 1 + len); | |
cc5e7107 JW |
1899 | l = end - start + 1; |
1900 | r = log_write(vq->log_base, | |
0bbe3066 | 1901 | u->start + start - u->addr, |
cc5e7107 JW |
1902 | l); |
1903 | if (r < 0) | |
1904 | return r; | |
1905 | hit = true; | |
1906 | min = min(l, min); | |
1907 | } | |
1908 | ||
1909 | if (!hit) | |
1910 | return -EFAULT; | |
1911 | ||
1912 | len -= min; | |
1913 | hva += min; | |
1914 | } | |
1915 | ||
1916 | return 0; | |
1917 | } | |
1918 | ||
1919 | static int log_used(struct vhost_virtqueue *vq, u64 used_offset, u64 len) | |
1920 | { | |
5e5e8736 | 1921 | struct iovec *iov = vq->log_iov; |
cc5e7107 JW |
1922 | int i, ret; |
1923 | ||
1924 | if (!vq->iotlb) | |
1925 | return log_write(vq->log_base, vq->log_addr + used_offset, len); | |
1926 | ||
1927 | ret = translate_desc(vq, (uintptr_t)vq->used + used_offset, | |
1928 | len, iov, 64, VHOST_ACCESS_WO); | |
816db766 | 1929 | if (ret < 0) |
cc5e7107 JW |
1930 | return ret; |
1931 | ||
1932 | for (i = 0; i < ret; i++) { | |
1933 | ret = log_write_hva(vq, (uintptr_t)iov[i].iov_base, | |
1934 | iov[i].iov_len); | |
1935 | if (ret) | |
1936 | return ret; | |
1937 | } | |
1938 | ||
1939 | return 0; | |
1940 | } | |
1941 | ||
3a4d5c94 | 1942 | int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log, |
cc5e7107 | 1943 | unsigned int log_num, u64 len, struct iovec *iov, int count) |
3a4d5c94 MT |
1944 | { |
1945 | int i, r; | |
1946 | ||
1947 | /* Make sure data written is seen before log. */ | |
5659338c | 1948 | smp_wmb(); |
cc5e7107 JW |
1949 | |
1950 | if (vq->iotlb) { | |
1951 | for (i = 0; i < count; i++) { | |
1952 | r = log_write_hva(vq, (uintptr_t)iov[i].iov_base, | |
1953 | iov[i].iov_len); | |
1954 | if (r < 0) | |
1955 | return r; | |
1956 | } | |
1957 | return 0; | |
1958 | } | |
1959 | ||
3a4d5c94 MT |
1960 | for (i = 0; i < log_num; ++i) { |
1961 | u64 l = min(log[i].len, len); | |
1962 | r = log_write(vq->log_base, log[i].addr, l); | |
1963 | if (r < 0) | |
1964 | return r; | |
1965 | len -= l; | |
5786aee8 MT |
1966 | if (!len) { |
1967 | if (vq->log_ctx) | |
1968 | eventfd_signal(vq->log_ctx, 1); | |
3a4d5c94 | 1969 | return 0; |
5786aee8 | 1970 | } |
3a4d5c94 | 1971 | } |
3a4d5c94 MT |
1972 | /* Length written exceeds what we have stored. This is a bug. */ |
1973 | BUG(); | |
1974 | return 0; | |
1975 | } | |
6ac1afbf | 1976 | EXPORT_SYMBOL_GPL(vhost_log_write); |
3a4d5c94 | 1977 | |
2723feaa JW |
1978 | static int vhost_update_used_flags(struct vhost_virtqueue *vq) |
1979 | { | |
1980 | void __user *used; | |
7b5d753e | 1981 | if (vhost_put_used_flags(vq)) |
2723feaa JW |
1982 | return -EFAULT; |
1983 | if (unlikely(vq->log_used)) { | |
1984 | /* Make sure the flag is seen before log. */ | |
1985 | smp_wmb(); | |
1986 | /* Log used flag write. */ | |
1987 | used = &vq->used->flags; | |
cc5e7107 JW |
1988 | log_used(vq, (used - (void __user *)vq->used), |
1989 | sizeof vq->used->flags); | |
2723feaa JW |
1990 | if (vq->log_ctx) |
1991 | eventfd_signal(vq->log_ctx, 1); | |
1992 | } | |
1993 | return 0; | |
1994 | } | |
1995 | ||
4c809363 | 1996 | static int vhost_update_avail_event(struct vhost_virtqueue *vq) |
2723feaa | 1997 | { |
7b5d753e | 1998 | if (vhost_put_avail_event(vq)) |
2723feaa JW |
1999 | return -EFAULT; |
2000 | if (unlikely(vq->log_used)) { | |
2001 | void __user *used; | |
2002 | /* Make sure the event is seen before log. */ | |
2003 | smp_wmb(); | |
2004 | /* Log avail event write */ | |
2005 | used = vhost_avail_event(vq); | |
cc5e7107 JW |
2006 | log_used(vq, (used - (void __user *)vq->used), |
2007 | sizeof *vhost_avail_event(vq)); | |
2723feaa JW |
2008 | if (vq->log_ctx) |
2009 | eventfd_signal(vq->log_ctx, 1); | |
2010 | } | |
2011 | return 0; | |
2012 | } | |
2013 | ||
80f7d030 | 2014 | int vhost_vq_init_access(struct vhost_virtqueue *vq) |
2723feaa | 2015 | { |
3b1bbe89 | 2016 | __virtio16 last_used_idx; |
2723feaa | 2017 | int r; |
e1f33be9 GK |
2018 | bool is_le = vq->is_le; |
2019 | ||
cda8bba0 | 2020 | if (!vq->private_data) |
2723feaa | 2021 | return 0; |
2751c988 GK |
2022 | |
2023 | vhost_init_is_le(vq); | |
2723feaa JW |
2024 | |
2025 | r = vhost_update_used_flags(vq); | |
2026 | if (r) | |
e1f33be9 | 2027 | goto err; |
2723feaa | 2028 | vq->signalled_used_valid = false; |
6b1e6cc7 | 2029 | if (!vq->iotlb && |
96d4f267 | 2030 | !access_ok(&vq->used->idx, sizeof vq->used->idx)) { |
e1f33be9 GK |
2031 | r = -EFAULT; |
2032 | goto err; | |
2033 | } | |
7b5d753e | 2034 | r = vhost_get_used_idx(vq, &last_used_idx); |
6b1e6cc7 JW |
2035 | if (r) { |
2036 | vq_err(vq, "Can't access used idx at %p\n", | |
2037 | &vq->used->idx); | |
e1f33be9 | 2038 | goto err; |
6b1e6cc7 | 2039 | } |
3b1bbe89 | 2040 | vq->last_used_idx = vhost16_to_cpu(vq, last_used_idx); |
64f7f051 | 2041 | return 0; |
6b1e6cc7 | 2042 | |
e1f33be9 GK |
2043 | err: |
2044 | vq->is_le = is_le; | |
2045 | return r; | |
2723feaa | 2046 | } |
80f7d030 | 2047 | EXPORT_SYMBOL_GPL(vhost_vq_init_access); |
2723feaa | 2048 | |
47283bef | 2049 | static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len, |
6b1e6cc7 | 2050 | struct iovec iov[], int iov_size, int access) |
3a4d5c94 | 2051 | { |
0bbe3066 | 2052 | const struct vhost_iotlb_map *map; |
6b1e6cc7 | 2053 | struct vhost_dev *dev = vq->dev; |
0bbe3066 | 2054 | struct vhost_iotlb *umem = dev->iotlb ? dev->iotlb : dev->umem; |
3a4d5c94 MT |
2055 | struct iovec *_iov; |
2056 | u64 s = 0; | |
2057 | int ret = 0; | |
2058 | ||
3a4d5c94 MT |
2059 | while ((u64)len > s) { |
2060 | u64 size; | |
7b3384fc | 2061 | if (unlikely(ret >= iov_size)) { |
3a4d5c94 MT |
2062 | ret = -ENOBUFS; |
2063 | break; | |
2064 | } | |
6b1e6cc7 | 2065 | |
0bbe3066 JW |
2066 | map = vhost_iotlb_itree_first(umem, addr, addr + len - 1); |
2067 | if (map == NULL || map->start > addr) { | |
6b1e6cc7 JW |
2068 | if (umem != dev->iotlb) { |
2069 | ret = -EFAULT; | |
2070 | break; | |
2071 | } | |
2072 | ret = -EAGAIN; | |
2073 | break; | |
0bbe3066 | 2074 | } else if (!(map->perm & access)) { |
6b1e6cc7 | 2075 | ret = -EPERM; |
3a4d5c94 MT |
2076 | break; |
2077 | } | |
6b1e6cc7 | 2078 | |
3a4d5c94 | 2079 | _iov = iov + ret; |
0bbe3066 | 2080 | size = map->size - addr + map->start; |
bd97120f | 2081 | _iov->iov_len = min((u64)len - s, size); |
0d4a3f2a | 2082 | _iov->iov_base = (void __user *)(unsigned long) |
0bbe3066 | 2083 | (map->addr + addr - map->start); |
3a4d5c94 MT |
2084 | s += size; |
2085 | addr += size; | |
2086 | ++ret; | |
2087 | } | |
2088 | ||
6b1e6cc7 JW |
2089 | if (ret == -EAGAIN) |
2090 | vhost_iotlb_miss(vq, addr, access); | |
3a4d5c94 MT |
2091 | return ret; |
2092 | } | |
2093 | ||
2094 | /* Each buffer in the virtqueues is actually a chain of descriptors. This | |
2095 | * function returns the next descriptor in the chain, | |
2096 | * or -1U if we're at the end. */ | |
3b1bbe89 | 2097 | static unsigned next_desc(struct vhost_virtqueue *vq, struct vring_desc *desc) |
3a4d5c94 MT |
2098 | { |
2099 | unsigned int next; | |
2100 | ||
2101 | /* If this descriptor says it doesn't chain, we're done. */ | |
3b1bbe89 | 2102 | if (!(desc->flags & cpu_to_vhost16(vq, VRING_DESC_F_NEXT))) |
3a4d5c94 MT |
2103 | return -1U; |
2104 | ||
2105 | /* Check they're not leading us off end of descriptors. */ | |
3a5db0b1 | 2106 | next = vhost16_to_cpu(vq, READ_ONCE(desc->next)); |
3a4d5c94 MT |
2107 | return next; |
2108 | } | |
2109 | ||
47283bef | 2110 | static int get_indirect(struct vhost_virtqueue *vq, |
7b3384fc MT |
2111 | struct iovec iov[], unsigned int iov_size, |
2112 | unsigned int *out_num, unsigned int *in_num, | |
2113 | struct vhost_log *log, unsigned int *log_num, | |
2114 | struct vring_desc *indirect) | |
3a4d5c94 MT |
2115 | { |
2116 | struct vring_desc desc; | |
2117 | unsigned int i = 0, count, found = 0; | |
3b1bbe89 | 2118 | u32 len = vhost32_to_cpu(vq, indirect->len); |
aad9a1ce | 2119 | struct iov_iter from; |
6b1e6cc7 | 2120 | int ret, access; |
3a4d5c94 MT |
2121 | |
2122 | /* Sanity check */ | |
3b1bbe89 | 2123 | if (unlikely(len % sizeof desc)) { |
3a4d5c94 MT |
2124 | vq_err(vq, "Invalid length in indirect descriptor: " |
2125 | "len 0x%llx not multiple of 0x%zx\n", | |
3b1bbe89 | 2126 | (unsigned long long)len, |
3a4d5c94 MT |
2127 | sizeof desc); |
2128 | return -EINVAL; | |
2129 | } | |
2130 | ||
3b1bbe89 | 2131 | ret = translate_desc(vq, vhost64_to_cpu(vq, indirect->addr), len, vq->indirect, |
6b1e6cc7 | 2132 | UIO_MAXIOV, VHOST_ACCESS_RO); |
7b3384fc | 2133 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
2134 | if (ret != -EAGAIN) |
2135 | vq_err(vq, "Translation failure %d in indirect.\n", ret); | |
3a4d5c94 MT |
2136 | return ret; |
2137 | } | |
aad9a1ce | 2138 | iov_iter_init(&from, READ, vq->indirect, ret, len); |
3b1bbe89 | 2139 | count = len / sizeof desc; |
3a4d5c94 MT |
2140 | /* Buffers are chained via a 16 bit next field, so |
2141 | * we can have at most 2^16 of these. */ | |
7b3384fc | 2142 | if (unlikely(count > USHRT_MAX + 1)) { |
3a4d5c94 MT |
2143 | vq_err(vq, "Indirect buffer length too big: %d\n", |
2144 | indirect->len); | |
2145 | return -E2BIG; | |
2146 | } | |
2147 | ||
2148 | do { | |
2149 | unsigned iov_count = *in_num + *out_num; | |
7b3384fc | 2150 | if (unlikely(++found > count)) { |
3a4d5c94 MT |
2151 | vq_err(vq, "Loop detected: last one at %u " |
2152 | "indirect size %u\n", | |
2153 | i, count); | |
2154 | return -EINVAL; | |
2155 | } | |
cbbd26b8 | 2156 | if (unlikely(!copy_from_iter_full(&desc, sizeof(desc), &from))) { |
3a4d5c94 | 2157 | vq_err(vq, "Failed indirect descriptor: idx %d, %zx\n", |
3b1bbe89 | 2158 | i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc); |
3a4d5c94 MT |
2159 | return -EINVAL; |
2160 | } | |
3b1bbe89 | 2161 | if (unlikely(desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT))) { |
3a4d5c94 | 2162 | vq_err(vq, "Nested indirect descriptor: idx %d, %zx\n", |
3b1bbe89 | 2163 | i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc); |
3a4d5c94 MT |
2164 | return -EINVAL; |
2165 | } | |
2166 | ||
6b1e6cc7 JW |
2167 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE)) |
2168 | access = VHOST_ACCESS_WO; | |
2169 | else | |
2170 | access = VHOST_ACCESS_RO; | |
2171 | ||
3b1bbe89 MT |
2172 | ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr), |
2173 | vhost32_to_cpu(vq, desc.len), iov + iov_count, | |
6b1e6cc7 | 2174 | iov_size - iov_count, access); |
7b3384fc | 2175 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
2176 | if (ret != -EAGAIN) |
2177 | vq_err(vq, "Translation failure %d indirect idx %d\n", | |
2178 | ret, i); | |
3a4d5c94 MT |
2179 | return ret; |
2180 | } | |
2181 | /* If this is an input descriptor, increment that count. */ | |
6b1e6cc7 | 2182 | if (access == VHOST_ACCESS_WO) { |
3a4d5c94 | 2183 | *in_num += ret; |
060423bf | 2184 | if (unlikely(log && ret)) { |
3b1bbe89 MT |
2185 | log[*log_num].addr = vhost64_to_cpu(vq, desc.addr); |
2186 | log[*log_num].len = vhost32_to_cpu(vq, desc.len); | |
3a4d5c94 MT |
2187 | ++*log_num; |
2188 | } | |
2189 | } else { | |
2190 | /* If it's an output descriptor, they're all supposed | |
2191 | * to come before any input descriptors. */ | |
7b3384fc | 2192 | if (unlikely(*in_num)) { |
3a4d5c94 MT |
2193 | vq_err(vq, "Indirect descriptor " |
2194 | "has out after in: idx %d\n", i); | |
2195 | return -EINVAL; | |
2196 | } | |
2197 | *out_num += ret; | |
2198 | } | |
3b1bbe89 | 2199 | } while ((i = next_desc(vq, &desc)) != -1); |
3a4d5c94 MT |
2200 | return 0; |
2201 | } | |
2202 | ||
2203 | /* This looks in the virtqueue and for the first available buffer, and converts | |
2204 | * it to an iovec for convenient access. Since descriptors consist of some | |
2205 | * number of output then some number of input descriptors, it's actually two | |
2206 | * iovecs, but we pack them into one and note how many of each there were. | |
2207 | * | |
d5675bd2 MT |
2208 | * This function returns the descriptor number found, or vq->num (which is |
2209 | * never a valid descriptor number) if none was found. A negative code is | |
2210 | * returned on error. */ | |
47283bef | 2211 | int vhost_get_vq_desc(struct vhost_virtqueue *vq, |
d5675bd2 MT |
2212 | struct iovec iov[], unsigned int iov_size, |
2213 | unsigned int *out_num, unsigned int *in_num, | |
2214 | struct vhost_log *log, unsigned int *log_num) | |
3a4d5c94 MT |
2215 | { |
2216 | struct vring_desc desc; | |
2217 | unsigned int i, head, found = 0; | |
2218 | u16 last_avail_idx; | |
3b1bbe89 MT |
2219 | __virtio16 avail_idx; |
2220 | __virtio16 ring_head; | |
6b1e6cc7 | 2221 | int ret, access; |
3a4d5c94 MT |
2222 | |
2223 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
2224 | last_avail_idx = vq->last_avail_idx; | |
3a4d5c94 | 2225 | |
e3b56cdd | 2226 | if (vq->avail_idx == vq->last_avail_idx) { |
7b5d753e | 2227 | if (unlikely(vhost_get_avail_idx(vq, &avail_idx))) { |
e3b56cdd JW |
2228 | vq_err(vq, "Failed to access avail idx at %p\n", |
2229 | &vq->avail->idx); | |
2230 | return -EFAULT; | |
2231 | } | |
2232 | vq->avail_idx = vhost16_to_cpu(vq, avail_idx); | |
3a4d5c94 | 2233 | |
e3b56cdd JW |
2234 | if (unlikely((u16)(vq->avail_idx - last_avail_idx) > vq->num)) { |
2235 | vq_err(vq, "Guest moved used index from %u to %u", | |
2236 | last_avail_idx, vq->avail_idx); | |
2237 | return -EFAULT; | |
2238 | } | |
2239 | ||
2240 | /* If there's nothing new since last we looked, return | |
2241 | * invalid. | |
2242 | */ | |
2243 | if (vq->avail_idx == last_avail_idx) | |
2244 | return vq->num; | |
3a4d5c94 | 2245 | |
e3b56cdd JW |
2246 | /* Only get avail ring entries after they have been |
2247 | * exposed by guest. | |
2248 | */ | |
2249 | smp_rmb(); | |
2250 | } | |
3a4d5c94 MT |
2251 | |
2252 | /* Grab the next descriptor number they're advertising, and increment | |
2253 | * the index we've seen. */ | |
7b5d753e | 2254 | if (unlikely(vhost_get_avail_head(vq, &ring_head, last_avail_idx))) { |
3a4d5c94 MT |
2255 | vq_err(vq, "Failed to read head: idx %d address %p\n", |
2256 | last_avail_idx, | |
2257 | &vq->avail->ring[last_avail_idx % vq->num]); | |
d5675bd2 | 2258 | return -EFAULT; |
3a4d5c94 MT |
2259 | } |
2260 | ||
3b1bbe89 MT |
2261 | head = vhost16_to_cpu(vq, ring_head); |
2262 | ||
3a4d5c94 | 2263 | /* If their number is silly, that's an error. */ |
7b3384fc | 2264 | if (unlikely(head >= vq->num)) { |
3a4d5c94 MT |
2265 | vq_err(vq, "Guest says index %u > %u is available", |
2266 | head, vq->num); | |
d5675bd2 | 2267 | return -EINVAL; |
3a4d5c94 MT |
2268 | } |
2269 | ||
2270 | /* When we start there are none of either input nor output. */ | |
2271 | *out_num = *in_num = 0; | |
2272 | if (unlikely(log)) | |
2273 | *log_num = 0; | |
2274 | ||
2275 | i = head; | |
2276 | do { | |
2277 | unsigned iov_count = *in_num + *out_num; | |
7b3384fc | 2278 | if (unlikely(i >= vq->num)) { |
3a4d5c94 MT |
2279 | vq_err(vq, "Desc index is %u > %u, head = %u", |
2280 | i, vq->num, head); | |
d5675bd2 | 2281 | return -EINVAL; |
3a4d5c94 | 2282 | } |
7b3384fc | 2283 | if (unlikely(++found > vq->num)) { |
3a4d5c94 MT |
2284 | vq_err(vq, "Loop detected: last one at %u " |
2285 | "vq size %u head %u\n", | |
2286 | i, vq->num, head); | |
d5675bd2 | 2287 | return -EINVAL; |
3a4d5c94 | 2288 | } |
7b5d753e | 2289 | ret = vhost_get_desc(vq, &desc, i); |
7b3384fc | 2290 | if (unlikely(ret)) { |
3a4d5c94 MT |
2291 | vq_err(vq, "Failed to get descriptor: idx %d addr %p\n", |
2292 | i, vq->desc + i); | |
d5675bd2 | 2293 | return -EFAULT; |
3a4d5c94 | 2294 | } |
3b1bbe89 | 2295 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT)) { |
47283bef | 2296 | ret = get_indirect(vq, iov, iov_size, |
3a4d5c94 MT |
2297 | out_num, in_num, |
2298 | log, log_num, &desc); | |
7b3384fc | 2299 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
2300 | if (ret != -EAGAIN) |
2301 | vq_err(vq, "Failure detected " | |
2302 | "in indirect descriptor at idx %d\n", i); | |
d5675bd2 | 2303 | return ret; |
3a4d5c94 MT |
2304 | } |
2305 | continue; | |
2306 | } | |
2307 | ||
6b1e6cc7 JW |
2308 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE)) |
2309 | access = VHOST_ACCESS_WO; | |
2310 | else | |
2311 | access = VHOST_ACCESS_RO; | |
3b1bbe89 MT |
2312 | ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr), |
2313 | vhost32_to_cpu(vq, desc.len), iov + iov_count, | |
6b1e6cc7 | 2314 | iov_size - iov_count, access); |
7b3384fc | 2315 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
2316 | if (ret != -EAGAIN) |
2317 | vq_err(vq, "Translation failure %d descriptor idx %d\n", | |
2318 | ret, i); | |
d5675bd2 | 2319 | return ret; |
3a4d5c94 | 2320 | } |
6b1e6cc7 | 2321 | if (access == VHOST_ACCESS_WO) { |
3a4d5c94 MT |
2322 | /* If this is an input descriptor, |
2323 | * increment that count. */ | |
2324 | *in_num += ret; | |
060423bf | 2325 | if (unlikely(log && ret)) { |
3b1bbe89 MT |
2326 | log[*log_num].addr = vhost64_to_cpu(vq, desc.addr); |
2327 | log[*log_num].len = vhost32_to_cpu(vq, desc.len); | |
3a4d5c94 MT |
2328 | ++*log_num; |
2329 | } | |
2330 | } else { | |
2331 | /* If it's an output descriptor, they're all supposed | |
2332 | * to come before any input descriptors. */ | |
7b3384fc | 2333 | if (unlikely(*in_num)) { |
3a4d5c94 MT |
2334 | vq_err(vq, "Descriptor has out after in: " |
2335 | "idx %d\n", i); | |
d5675bd2 | 2336 | return -EINVAL; |
3a4d5c94 MT |
2337 | } |
2338 | *out_num += ret; | |
2339 | } | |
3b1bbe89 | 2340 | } while ((i = next_desc(vq, &desc)) != -1); |
3a4d5c94 MT |
2341 | |
2342 | /* On success, increment avail index. */ | |
2343 | vq->last_avail_idx++; | |
8ea8cf89 MT |
2344 | |
2345 | /* Assume notifications from guest are disabled at this point, | |
2346 | * if they aren't we would need to update avail_event index. */ | |
2347 | BUG_ON(!(vq->used_flags & VRING_USED_F_NO_NOTIFY)); | |
3a4d5c94 MT |
2348 | return head; |
2349 | } | |
6ac1afbf | 2350 | EXPORT_SYMBOL_GPL(vhost_get_vq_desc); |
3a4d5c94 MT |
2351 | |
2352 | /* Reverse the effect of vhost_get_vq_desc. Useful for error handling. */ | |
8dd014ad | 2353 | void vhost_discard_vq_desc(struct vhost_virtqueue *vq, int n) |
3a4d5c94 | 2354 | { |
8dd014ad | 2355 | vq->last_avail_idx -= n; |
3a4d5c94 | 2356 | } |
6ac1afbf | 2357 | EXPORT_SYMBOL_GPL(vhost_discard_vq_desc); |
3a4d5c94 MT |
2358 | |
2359 | /* After we've used one of their buffers, we tell them about it. We'll then | |
2360 | * want to notify the guest, using eventfd. */ | |
2361 | int vhost_add_used(struct vhost_virtqueue *vq, unsigned int head, int len) | |
2362 | { | |
3b1bbe89 MT |
2363 | struct vring_used_elem heads = { |
2364 | cpu_to_vhost32(vq, head), | |
2365 | cpu_to_vhost32(vq, len) | |
2366 | }; | |
3a4d5c94 | 2367 | |
c49e4e57 | 2368 | return vhost_add_used_n(vq, &heads, 1); |
3a4d5c94 | 2369 | } |
6ac1afbf | 2370 | EXPORT_SYMBOL_GPL(vhost_add_used); |
3a4d5c94 | 2371 | |
8dd014ad DS |
2372 | static int __vhost_add_used_n(struct vhost_virtqueue *vq, |
2373 | struct vring_used_elem *heads, | |
2374 | unsigned count) | |
2375 | { | |
a865e420 | 2376 | vring_used_elem_t __user *used; |
8ea8cf89 | 2377 | u16 old, new; |
8dd014ad DS |
2378 | int start; |
2379 | ||
5fba13b5 | 2380 | start = vq->last_used_idx & (vq->num - 1); |
8dd014ad | 2381 | used = vq->used->ring + start; |
7b5d753e | 2382 | if (vhost_put_used(vq, heads, start, count)) { |
8dd014ad DS |
2383 | vq_err(vq, "Failed to write used"); |
2384 | return -EFAULT; | |
2385 | } | |
2386 | if (unlikely(vq->log_used)) { | |
2387 | /* Make sure data is seen before log. */ | |
2388 | smp_wmb(); | |
2389 | /* Log used ring entry write. */ | |
cc5e7107 JW |
2390 | log_used(vq, ((void __user *)used - (void __user *)vq->used), |
2391 | count * sizeof *used); | |
8dd014ad | 2392 | } |
8ea8cf89 MT |
2393 | old = vq->last_used_idx; |
2394 | new = (vq->last_used_idx += count); | |
2395 | /* If the driver never bothers to signal in a very long while, | |
2396 | * used index might wrap around. If that happens, invalidate | |
2397 | * signalled_used index we stored. TODO: make sure driver | |
2398 | * signals at least once in 2^16 and remove this. */ | |
2399 | if (unlikely((u16)(new - vq->signalled_used) < (u16)(new - old))) | |
2400 | vq->signalled_used_valid = false; | |
8dd014ad DS |
2401 | return 0; |
2402 | } | |
2403 | ||
2404 | /* After we've used one of their buffers, we tell them about it. We'll then | |
2405 | * want to notify the guest, using eventfd. */ | |
2406 | int vhost_add_used_n(struct vhost_virtqueue *vq, struct vring_used_elem *heads, | |
2407 | unsigned count) | |
2408 | { | |
2409 | int start, n, r; | |
2410 | ||
5fba13b5 | 2411 | start = vq->last_used_idx & (vq->num - 1); |
8dd014ad DS |
2412 | n = vq->num - start; |
2413 | if (n < count) { | |
2414 | r = __vhost_add_used_n(vq, heads, n); | |
2415 | if (r < 0) | |
2416 | return r; | |
2417 | heads += n; | |
2418 | count -= n; | |
2419 | } | |
2420 | r = __vhost_add_used_n(vq, heads, count); | |
2421 | ||
2422 | /* Make sure buffer is written before we update index. */ | |
2423 | smp_wmb(); | |
7b5d753e | 2424 | if (vhost_put_used_idx(vq)) { |
8dd014ad DS |
2425 | vq_err(vq, "Failed to increment used idx"); |
2426 | return -EFAULT; | |
2427 | } | |
2428 | if (unlikely(vq->log_used)) { | |
841df922 JW |
2429 | /* Make sure used idx is seen before log. */ |
2430 | smp_wmb(); | |
8dd014ad | 2431 | /* Log used index update. */ |
cc5e7107 JW |
2432 | log_used(vq, offsetof(struct vring_used, idx), |
2433 | sizeof vq->used->idx); | |
8dd014ad DS |
2434 | if (vq->log_ctx) |
2435 | eventfd_signal(vq->log_ctx, 1); | |
2436 | } | |
2437 | return r; | |
2438 | } | |
6ac1afbf | 2439 | EXPORT_SYMBOL_GPL(vhost_add_used_n); |
8dd014ad | 2440 | |
8ea8cf89 | 2441 | static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 | 2442 | { |
3b1bbe89 MT |
2443 | __u16 old, new; |
2444 | __virtio16 event; | |
8ea8cf89 | 2445 | bool v; |
8d65843c JW |
2446 | /* Flush out used index updates. This is paired |
2447 | * with the barrier that the Guest executes when enabling | |
2448 | * interrupts. */ | |
2449 | smp_mb(); | |
0d499356 | 2450 | |
ea16c514 | 2451 | if (vhost_has_feature(vq, VIRTIO_F_NOTIFY_ON_EMPTY) && |
8ea8cf89 MT |
2452 | unlikely(vq->avail_idx == vq->last_avail_idx)) |
2453 | return true; | |
2454 | ||
ea16c514 | 2455 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
3b1bbe89 | 2456 | __virtio16 flags; |
7b5d753e | 2457 | if (vhost_get_avail_flags(vq, &flags)) { |
8ea8cf89 MT |
2458 | vq_err(vq, "Failed to get flags"); |
2459 | return true; | |
2460 | } | |
3b1bbe89 | 2461 | return !(flags & cpu_to_vhost16(vq, VRING_AVAIL_F_NO_INTERRUPT)); |
3a4d5c94 | 2462 | } |
8ea8cf89 MT |
2463 | old = vq->signalled_used; |
2464 | v = vq->signalled_used_valid; | |
2465 | new = vq->signalled_used = vq->last_used_idx; | |
2466 | vq->signalled_used_valid = true; | |
3a4d5c94 | 2467 | |
8ea8cf89 MT |
2468 | if (unlikely(!v)) |
2469 | return true; | |
3a4d5c94 | 2470 | |
7b5d753e | 2471 | if (vhost_get_used_event(vq, &event)) { |
8ea8cf89 MT |
2472 | vq_err(vq, "Failed to get used event idx"); |
2473 | return true; | |
2474 | } | |
8d65843c | 2475 | return vring_need_event(vhost16_to_cpu(vq, event), new, old); |
8ea8cf89 MT |
2476 | } |
2477 | ||
2478 | /* This actually signals the guest, using eventfd. */ | |
2479 | void vhost_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq) | |
2480 | { | |
3a4d5c94 | 2481 | /* Signal the Guest tell them we used something up. */ |
265a0ad8 ZL |
2482 | if (vq->call_ctx.ctx && vhost_notify(dev, vq)) |
2483 | eventfd_signal(vq->call_ctx.ctx, 1); | |
3a4d5c94 | 2484 | } |
6ac1afbf | 2485 | EXPORT_SYMBOL_GPL(vhost_signal); |
3a4d5c94 MT |
2486 | |
2487 | /* And here's the combo meal deal. Supersize me! */ | |
2488 | void vhost_add_used_and_signal(struct vhost_dev *dev, | |
2489 | struct vhost_virtqueue *vq, | |
2490 | unsigned int head, int len) | |
2491 | { | |
2492 | vhost_add_used(vq, head, len); | |
2493 | vhost_signal(dev, vq); | |
2494 | } | |
6ac1afbf | 2495 | EXPORT_SYMBOL_GPL(vhost_add_used_and_signal); |
3a4d5c94 | 2496 | |
8dd014ad DS |
2497 | /* multi-buffer version of vhost_add_used_and_signal */ |
2498 | void vhost_add_used_and_signal_n(struct vhost_dev *dev, | |
2499 | struct vhost_virtqueue *vq, | |
2500 | struct vring_used_elem *heads, unsigned count) | |
2501 | { | |
2502 | vhost_add_used_n(vq, heads, count); | |
2503 | vhost_signal(dev, vq); | |
2504 | } | |
6ac1afbf | 2505 | EXPORT_SYMBOL_GPL(vhost_add_used_and_signal_n); |
8dd014ad | 2506 | |
d4a60603 JW |
2507 | /* return true if we're sure that avaiable ring is empty */ |
2508 | bool vhost_vq_avail_empty(struct vhost_dev *dev, struct vhost_virtqueue *vq) | |
2509 | { | |
2510 | __virtio16 avail_idx; | |
2511 | int r; | |
2512 | ||
275bf960 JW |
2513 | if (vq->avail_idx != vq->last_avail_idx) |
2514 | return false; | |
2515 | ||
7b5d753e | 2516 | r = vhost_get_avail_idx(vq, &avail_idx); |
275bf960 | 2517 | if (unlikely(r)) |
d4a60603 | 2518 | return false; |
275bf960 | 2519 | vq->avail_idx = vhost16_to_cpu(vq, avail_idx); |
d4a60603 | 2520 | |
275bf960 | 2521 | return vq->avail_idx == vq->last_avail_idx; |
d4a60603 JW |
2522 | } |
2523 | EXPORT_SYMBOL_GPL(vhost_vq_avail_empty); | |
2524 | ||
3a4d5c94 | 2525 | /* OK, now we need to know about added descriptors. */ |
8ea8cf89 | 2526 | bool vhost_enable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 | 2527 | { |
3b1bbe89 | 2528 | __virtio16 avail_idx; |
3a4d5c94 | 2529 | int r; |
d47effe1 | 2530 | |
3a4d5c94 MT |
2531 | if (!(vq->used_flags & VRING_USED_F_NO_NOTIFY)) |
2532 | return false; | |
2533 | vq->used_flags &= ~VRING_USED_F_NO_NOTIFY; | |
ea16c514 | 2534 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
2723feaa | 2535 | r = vhost_update_used_flags(vq); |
8ea8cf89 MT |
2536 | if (r) { |
2537 | vq_err(vq, "Failed to enable notification at %p: %d\n", | |
2538 | &vq->used->flags, r); | |
2539 | return false; | |
2540 | } | |
2541 | } else { | |
4c809363 | 2542 | r = vhost_update_avail_event(vq); |
8ea8cf89 MT |
2543 | if (r) { |
2544 | vq_err(vq, "Failed to update avail event index at %p: %d\n", | |
2545 | vhost_avail_event(vq), r); | |
2546 | return false; | |
2547 | } | |
2548 | } | |
3a4d5c94 MT |
2549 | /* They could have slipped one in as we were doing that: make |
2550 | * sure it's written, then check again. */ | |
5659338c | 2551 | smp_mb(); |
7b5d753e | 2552 | r = vhost_get_avail_idx(vq, &avail_idx); |
3a4d5c94 MT |
2553 | if (r) { |
2554 | vq_err(vq, "Failed to check avail idx at %p: %d\n", | |
2555 | &vq->avail->idx, r); | |
2556 | return false; | |
2557 | } | |
d3bb267b | 2558 | vq->avail_idx = vhost16_to_cpu(vq, avail_idx); |
3a4d5c94 | 2559 | |
d3bb267b | 2560 | return vq->avail_idx != vq->last_avail_idx; |
3a4d5c94 | 2561 | } |
6ac1afbf | 2562 | EXPORT_SYMBOL_GPL(vhost_enable_notify); |
3a4d5c94 MT |
2563 | |
2564 | /* We don't need to be notified again. */ | |
8ea8cf89 | 2565 | void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 MT |
2566 | { |
2567 | int r; | |
d47effe1 | 2568 | |
3a4d5c94 MT |
2569 | if (vq->used_flags & VRING_USED_F_NO_NOTIFY) |
2570 | return; | |
2571 | vq->used_flags |= VRING_USED_F_NO_NOTIFY; | |
ea16c514 | 2572 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
2723feaa | 2573 | r = vhost_update_used_flags(vq); |
8ea8cf89 | 2574 | if (r) |
ae6961de | 2575 | vq_err(vq, "Failed to disable notification at %p: %d\n", |
8ea8cf89 MT |
2576 | &vq->used->flags, r); |
2577 | } | |
3a4d5c94 | 2578 | } |
6ac1afbf AH |
2579 | EXPORT_SYMBOL_GPL(vhost_disable_notify); |
2580 | ||
6b1e6cc7 JW |
2581 | /* Create a new message. */ |
2582 | struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type) | |
2583 | { | |
2584 | struct vhost_msg_node *node = kmalloc(sizeof *node, GFP_KERNEL); | |
2585 | if (!node) | |
2586 | return NULL; | |
670ae9ca MT |
2587 | |
2588 | /* Make sure all padding within the structure is initialized. */ | |
2589 | memset(&node->msg, 0, sizeof node->msg); | |
6b1e6cc7 JW |
2590 | node->vq = vq; |
2591 | node->msg.type = type; | |
2592 | return node; | |
2593 | } | |
2594 | EXPORT_SYMBOL_GPL(vhost_new_msg); | |
2595 | ||
2596 | void vhost_enqueue_msg(struct vhost_dev *dev, struct list_head *head, | |
2597 | struct vhost_msg_node *node) | |
2598 | { | |
2599 | spin_lock(&dev->iotlb_lock); | |
2600 | list_add_tail(&node->node, head); | |
2601 | spin_unlock(&dev->iotlb_lock); | |
2602 | ||
a9a08845 | 2603 | wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM); |
6b1e6cc7 JW |
2604 | } |
2605 | EXPORT_SYMBOL_GPL(vhost_enqueue_msg); | |
2606 | ||
2607 | struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev, | |
2608 | struct list_head *head) | |
2609 | { | |
2610 | struct vhost_msg_node *node = NULL; | |
2611 | ||
2612 | spin_lock(&dev->iotlb_lock); | |
2613 | if (!list_empty(head)) { | |
2614 | node = list_first_entry(head, struct vhost_msg_node, | |
2615 | node); | |
2616 | list_del(&node->node); | |
2617 | } | |
2618 | spin_unlock(&dev->iotlb_lock); | |
2619 | ||
2620 | return node; | |
2621 | } | |
2622 | EXPORT_SYMBOL_GPL(vhost_dequeue_msg); | |
2623 | ||
460f7ce1 JW |
2624 | void vhost_set_backend_features(struct vhost_dev *dev, u64 features) |
2625 | { | |
2626 | struct vhost_virtqueue *vq; | |
2627 | int i; | |
2628 | ||
2629 | mutex_lock(&dev->mutex); | |
2630 | for (i = 0; i < dev->nvqs; ++i) { | |
2631 | vq = dev->vqs[i]; | |
2632 | mutex_lock(&vq->mutex); | |
2633 | vq->acked_backend_features = features; | |
2634 | mutex_unlock(&vq->mutex); | |
2635 | } | |
2636 | mutex_unlock(&dev->mutex); | |
2637 | } | |
2638 | EXPORT_SYMBOL_GPL(vhost_set_backend_features); | |
6b1e6cc7 | 2639 | |
6ac1afbf AH |
2640 | static int __init vhost_init(void) |
2641 | { | |
2642 | return 0; | |
2643 | } | |
2644 | ||
2645 | static void __exit vhost_exit(void) | |
2646 | { | |
2647 | } | |
2648 | ||
2649 | module_init(vhost_init); | |
2650 | module_exit(vhost_exit); | |
2651 | ||
2652 | MODULE_VERSION("0.0.1"); | |
2653 | MODULE_LICENSE("GPL v2"); | |
2654 | MODULE_AUTHOR("Michael S. Tsirkin"); | |
2655 | MODULE_DESCRIPTION("Host kernel accelerator for virtio"); |