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> |
6ac1afbf | 25 | #include <linux/module.h> |
bcfeacab | 26 | #include <linux/sort.h> |
6e84f315 | 27 | #include <linux/sched/mm.h> |
174cd4b1 | 28 | #include <linux/sched/signal.h> |
6e890c5d | 29 | #include <linux/sched/vhost_task.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 | */ |
1a5f8090 | 258 | llist_add(&work->node, &dev->worker->work_list); |
6e890c5d | 259 | wake_up_process(dev->worker->vtsk->task); |
c23f3445 | 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 | { | |
1a5f8090 | 267 | return dev->worker && !llist_empty(&dev->worker->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 | { | |
1a5f8090 | 338 | struct vhost_worker *worker = data; |
04b96e55 JW |
339 | struct vhost_work *work, *work_next; |
340 | struct llist_node *node; | |
c23f3445 TH |
341 | |
342 | for (;;) { | |
343 | /* mb paired w/ kthread_stop */ | |
344 | set_current_state(TASK_INTERRUPTIBLE); | |
345 | ||
6e890c5d | 346 | if (vhost_task_should_stop(worker->vtsk)) { |
c23f3445 | 347 | __set_current_state(TASK_RUNNING); |
64e1c807 | 348 | break; |
c23f3445 | 349 | } |
04b96e55 | 350 | |
1a5f8090 | 351 | node = llist_del_all(&worker->work_list); |
04b96e55 JW |
352 | if (!node) |
353 | schedule(); | |
354 | ||
355 | node = llist_reverse_order(node); | |
356 | /* make sure flag is seen after deletion */ | |
357 | smp_wmb(); | |
358 | llist_for_each_entry_safe(work, work_next, node, node) { | |
359 | clear_bit(VHOST_WORK_QUEUED, &work->flags); | |
c23f3445 | 360 | __set_current_state(TASK_RUNNING); |
1a5f8090 | 361 | kcov_remote_start_common(worker->kcov_handle); |
c23f3445 | 362 | work->fn(work); |
8f6a7f96 | 363 | kcov_remote_stop(); |
05bfb338 | 364 | cond_resched(); |
04b96e55 | 365 | } |
c23f3445 | 366 | } |
6e890c5d | 367 | |
64e1c807 | 368 | return 0; |
c23f3445 TH |
369 | } |
370 | ||
bab632d6 MT |
371 | static void vhost_vq_free_iovecs(struct vhost_virtqueue *vq) |
372 | { | |
373 | kfree(vq->indirect); | |
374 | vq->indirect = NULL; | |
375 | kfree(vq->log); | |
376 | vq->log = NULL; | |
377 | kfree(vq->heads); | |
378 | vq->heads = NULL; | |
bab632d6 MT |
379 | } |
380 | ||
e0e9b406 JW |
381 | /* Helper to allocate iovec buffers for all vqs. */ |
382 | static long vhost_dev_alloc_iovecs(struct vhost_dev *dev) | |
383 | { | |
6d5e6aa8 | 384 | struct vhost_virtqueue *vq; |
e0e9b406 | 385 | int i; |
d47effe1 | 386 | |
e0e9b406 | 387 | for (i = 0; i < dev->nvqs; ++i) { |
6d5e6aa8 | 388 | vq = dev->vqs[i]; |
6da2ec56 KC |
389 | vq->indirect = kmalloc_array(UIO_MAXIOV, |
390 | sizeof(*vq->indirect), | |
391 | GFP_KERNEL); | |
b46a0bf7 | 392 | vq->log = kmalloc_array(dev->iov_limit, sizeof(*vq->log), |
6da2ec56 | 393 | GFP_KERNEL); |
b46a0bf7 | 394 | vq->heads = kmalloc_array(dev->iov_limit, sizeof(*vq->heads), |
6da2ec56 | 395 | GFP_KERNEL); |
6d5e6aa8 | 396 | if (!vq->indirect || !vq->log || !vq->heads) |
e0e9b406 JW |
397 | goto err_nomem; |
398 | } | |
399 | return 0; | |
d47effe1 | 400 | |
e0e9b406 | 401 | err_nomem: |
bab632d6 | 402 | for (; i >= 0; --i) |
3ab2e420 | 403 | vhost_vq_free_iovecs(dev->vqs[i]); |
e0e9b406 JW |
404 | return -ENOMEM; |
405 | } | |
406 | ||
407 | static void vhost_dev_free_iovecs(struct vhost_dev *dev) | |
408 | { | |
409 | int i; | |
d47effe1 | 410 | |
bab632d6 | 411 | for (i = 0; i < dev->nvqs; ++i) |
3ab2e420 | 412 | vhost_vq_free_iovecs(dev->vqs[i]); |
e0e9b406 JW |
413 | } |
414 | ||
e82b9b07 JW |
415 | bool vhost_exceeds_weight(struct vhost_virtqueue *vq, |
416 | int pkts, int total_len) | |
417 | { | |
418 | struct vhost_dev *dev = vq->dev; | |
419 | ||
420 | if ((dev->byte_weight && total_len >= dev->byte_weight) || | |
421 | pkts >= dev->weight) { | |
422 | vhost_poll_queue(&vq->poll); | |
423 | return true; | |
424 | } | |
425 | ||
426 | return false; | |
427 | } | |
428 | EXPORT_SYMBOL_GPL(vhost_exceeds_weight); | |
429 | ||
4942e825 JW |
430 | static size_t vhost_get_avail_size(struct vhost_virtqueue *vq, |
431 | unsigned int num) | |
432 | { | |
433 | size_t event __maybe_unused = | |
434 | vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; | |
435 | ||
e4be66e5 | 436 | return size_add(struct_size(vq->avail, ring, num), event); |
4942e825 JW |
437 | } |
438 | ||
439 | static size_t vhost_get_used_size(struct vhost_virtqueue *vq, | |
440 | unsigned int num) | |
441 | { | |
442 | size_t event __maybe_unused = | |
443 | vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; | |
444 | ||
e4be66e5 | 445 | return size_add(struct_size(vq->used, ring, num), event); |
4942e825 JW |
446 | } |
447 | ||
448 | static size_t vhost_get_desc_size(struct vhost_virtqueue *vq, | |
449 | unsigned int num) | |
450 | { | |
451 | return sizeof(*vq->desc) * num; | |
452 | } | |
453 | ||
59566b6e | 454 | void vhost_dev_init(struct vhost_dev *dev, |
e82b9b07 | 455 | struct vhost_virtqueue **vqs, int nvqs, |
792a4f2e | 456 | int iov_limit, int weight, int byte_weight, |
01fcb1cb | 457 | bool use_worker, |
91233ad7 | 458 | int (*msg_handler)(struct vhost_dev *dev, u32 asid, |
792a4f2e | 459 | struct vhost_iotlb_msg *msg)) |
3a4d5c94 | 460 | { |
6d5e6aa8 | 461 | struct vhost_virtqueue *vq; |
3a4d5c94 | 462 | int i; |
c23f3445 | 463 | |
3a4d5c94 MT |
464 | dev->vqs = vqs; |
465 | dev->nvqs = nvqs; | |
466 | mutex_init(&dev->mutex); | |
467 | dev->log_ctx = NULL; | |
a9709d68 | 468 | dev->umem = NULL; |
6b1e6cc7 | 469 | dev->iotlb = NULL; |
3a4d5c94 | 470 | dev->mm = NULL; |
c23f3445 | 471 | dev->worker = NULL; |
b46a0bf7 | 472 | dev->iov_limit = iov_limit; |
e82b9b07 JW |
473 | dev->weight = weight; |
474 | dev->byte_weight = byte_weight; | |
01fcb1cb | 475 | dev->use_worker = use_worker; |
792a4f2e | 476 | dev->msg_handler = msg_handler; |
6b1e6cc7 JW |
477 | init_waitqueue_head(&dev->wait); |
478 | INIT_LIST_HEAD(&dev->read_list); | |
479 | INIT_LIST_HEAD(&dev->pending_list); | |
480 | spin_lock_init(&dev->iotlb_lock); | |
3d2c7d37 | 481 | |
3a4d5c94 MT |
482 | |
483 | for (i = 0; i < dev->nvqs; ++i) { | |
6d5e6aa8 AH |
484 | vq = dev->vqs[i]; |
485 | vq->log = NULL; | |
486 | vq->indirect = NULL; | |
487 | vq->heads = NULL; | |
488 | vq->dev = dev; | |
489 | mutex_init(&vq->mutex); | |
490 | vhost_vq_reset(dev, vq); | |
491 | if (vq->handle_kick) | |
492 | vhost_poll_init(&vq->poll, vq->handle_kick, | |
a9a08845 | 493 | EPOLLIN, dev); |
3a4d5c94 | 494 | } |
3a4d5c94 | 495 | } |
6ac1afbf | 496 | EXPORT_SYMBOL_GPL(vhost_dev_init); |
3a4d5c94 MT |
497 | |
498 | /* Caller should have device mutex */ | |
499 | long vhost_dev_check_owner(struct vhost_dev *dev) | |
500 | { | |
501 | /* Are you the owner? If not, I don't think you mean to do that */ | |
502 | return dev->mm == current->mm ? 0 : -EPERM; | |
503 | } | |
6ac1afbf | 504 | EXPORT_SYMBOL_GPL(vhost_dev_check_owner); |
3a4d5c94 | 505 | |
05c05351 MT |
506 | /* Caller should have device mutex */ |
507 | bool vhost_dev_has_owner(struct vhost_dev *dev) | |
508 | { | |
509 | return dev->mm; | |
510 | } | |
6ac1afbf | 511 | EXPORT_SYMBOL_GPL(vhost_dev_has_owner); |
05c05351 | 512 | |
5ce995f3 JW |
513 | static void vhost_attach_mm(struct vhost_dev *dev) |
514 | { | |
515 | /* No owner, become one */ | |
516 | if (dev->use_worker) { | |
517 | dev->mm = get_task_mm(current); | |
518 | } else { | |
519 | /* vDPA device does not use worker thead, so there's | |
520 | * no need to hold the address space for mm. This help | |
521 | * to avoid deadlock in the case of mmap() which may | |
522 | * held the refcnt of the file and depends on release | |
523 | * method to remove vma. | |
524 | */ | |
525 | dev->mm = current->mm; | |
526 | mmgrab(dev->mm); | |
527 | } | |
528 | } | |
529 | ||
530 | static void vhost_detach_mm(struct vhost_dev *dev) | |
531 | { | |
532 | if (!dev->mm) | |
533 | return; | |
534 | ||
535 | if (dev->use_worker) | |
536 | mmput(dev->mm); | |
537 | else | |
538 | mmdrop(dev->mm); | |
539 | ||
540 | dev->mm = NULL; | |
541 | } | |
542 | ||
1a5f8090 MC |
543 | static void vhost_worker_free(struct vhost_dev *dev) |
544 | { | |
545 | struct vhost_worker *worker = dev->worker; | |
546 | ||
547 | if (!worker) | |
548 | return; | |
549 | ||
550 | dev->worker = NULL; | |
551 | WARN_ON(!llist_empty(&worker->work_list)); | |
6e890c5d | 552 | vhost_task_stop(worker->vtsk); |
1a5f8090 MC |
553 | kfree(worker); |
554 | } | |
555 | ||
556 | static int vhost_worker_create(struct vhost_dev *dev) | |
557 | { | |
558 | struct vhost_worker *worker; | |
6e890c5d MC |
559 | struct vhost_task *vtsk; |
560 | char name[TASK_COMM_LEN]; | |
1a5f8090 MC |
561 | int ret; |
562 | ||
563 | worker = kzalloc(sizeof(*worker), GFP_KERNEL_ACCOUNT); | |
564 | if (!worker) | |
565 | return -ENOMEM; | |
566 | ||
567 | dev->worker = worker; | |
1a5f8090 MC |
568 | worker->kcov_handle = kcov_common_handle(); |
569 | init_llist_head(&worker->work_list); | |
6e890c5d | 570 | snprintf(name, sizeof(name), "vhost-%d", current->pid); |
1a5f8090 | 571 | |
6e890c5d MC |
572 | vtsk = vhost_task_create(vhost_worker, worker, name); |
573 | if (!vtsk) { | |
574 | ret = -ENOMEM; | |
1a5f8090 MC |
575 | goto free_worker; |
576 | } | |
577 | ||
6e890c5d MC |
578 | worker->vtsk = vtsk; |
579 | vhost_task_start(vtsk); | |
1a5f8090 MC |
580 | return 0; |
581 | ||
1a5f8090 MC |
582 | free_worker: |
583 | kfree(worker); | |
584 | dev->worker = NULL; | |
585 | return ret; | |
586 | } | |
587 | ||
3a4d5c94 | 588 | /* Caller should have device mutex */ |
54db63c2 | 589 | long vhost_dev_set_owner(struct vhost_dev *dev) |
3a4d5c94 | 590 | { |
c23f3445 | 591 | int err; |
d47effe1 | 592 | |
3a4d5c94 | 593 | /* Is there an owner already? */ |
05c05351 | 594 | if (vhost_dev_has_owner(dev)) { |
c23f3445 TH |
595 | err = -EBUSY; |
596 | goto err_mm; | |
597 | } | |
d47effe1 | 598 | |
5ce995f3 JW |
599 | vhost_attach_mm(dev); |
600 | ||
01fcb1cb | 601 | if (dev->use_worker) { |
1a5f8090 | 602 | err = vhost_worker_create(dev); |
01fcb1cb | 603 | if (err) |
1a5f8090 | 604 | goto err_worker; |
01fcb1cb | 605 | } |
c23f3445 | 606 | |
e0e9b406 JW |
607 | err = vhost_dev_alloc_iovecs(dev); |
608 | if (err) | |
1a5f8090 | 609 | goto err_iovecs; |
e0e9b406 | 610 | |
3a4d5c94 | 611 | return 0; |
1a5f8090 MC |
612 | err_iovecs: |
613 | vhost_worker_free(dev); | |
c23f3445 | 614 | err_worker: |
5ce995f3 | 615 | vhost_detach_mm(dev); |
c23f3445 TH |
616 | err_mm: |
617 | return err; | |
3a4d5c94 | 618 | } |
6ac1afbf | 619 | EXPORT_SYMBOL_GPL(vhost_dev_set_owner); |
3a4d5c94 | 620 | |
0bbe3066 JW |
621 | static struct vhost_iotlb *iotlb_alloc(void) |
622 | { | |
623 | return vhost_iotlb_alloc(max_iotlb_entries, | |
624 | VHOST_IOTLB_FLAG_RETIRE); | |
625 | } | |
626 | ||
627 | struct vhost_iotlb *vhost_dev_reset_owner_prepare(void) | |
a9709d68 | 628 | { |
0bbe3066 | 629 | return iotlb_alloc(); |
150b9e51 | 630 | } |
6ac1afbf | 631 | EXPORT_SYMBOL_GPL(vhost_dev_reset_owner_prepare); |
3a4d5c94 | 632 | |
150b9e51 | 633 | /* Caller should have device mutex */ |
0bbe3066 | 634 | void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_iotlb *umem) |
150b9e51 | 635 | { |
47283bef MT |
636 | int i; |
637 | ||
f6f93f75 | 638 | vhost_dev_cleanup(dev); |
3a4d5c94 | 639 | |
a9709d68 | 640 | dev->umem = umem; |
47283bef MT |
641 | /* We don't need VQ locks below since vhost_dev_cleanup makes sure |
642 | * VQs aren't running. | |
643 | */ | |
644 | for (i = 0; i < dev->nvqs; ++i) | |
a9709d68 | 645 | dev->vqs[i]->umem = umem; |
3a4d5c94 | 646 | } |
6ac1afbf | 647 | EXPORT_SYMBOL_GPL(vhost_dev_reset_owner); |
3a4d5c94 | 648 | |
b211616d | 649 | void vhost_dev_stop(struct vhost_dev *dev) |
bab632d6 MT |
650 | { |
651 | int i; | |
b211616d MT |
652 | |
653 | for (i = 0; i < dev->nvqs; ++i) { | |
6ca84326 | 654 | if (dev->vqs[i]->kick && dev->vqs[i]->handle_kick) |
3ab2e420 | 655 | vhost_poll_stop(&dev->vqs[i]->poll); |
bab632d6 | 656 | } |
6ca84326 | 657 | |
b2ffa407 | 658 | vhost_dev_flush(dev); |
bab632d6 | 659 | } |
6ac1afbf | 660 | EXPORT_SYMBOL_GPL(vhost_dev_stop); |
bab632d6 | 661 | |
9526f9a2 | 662 | void vhost_clear_msg(struct vhost_dev *dev) |
6b1e6cc7 JW |
663 | { |
664 | struct vhost_msg_node *node, *n; | |
665 | ||
666 | spin_lock(&dev->iotlb_lock); | |
667 | ||
668 | list_for_each_entry_safe(node, n, &dev->read_list, node) { | |
669 | list_del(&node->node); | |
670 | kfree(node); | |
671 | } | |
672 | ||
673 | list_for_each_entry_safe(node, n, &dev->pending_list, node) { | |
674 | list_del(&node->node); | |
675 | kfree(node); | |
676 | } | |
677 | ||
678 | spin_unlock(&dev->iotlb_lock); | |
679 | } | |
9526f9a2 | 680 | EXPORT_SYMBOL_GPL(vhost_clear_msg); |
6b1e6cc7 | 681 | |
f6f93f75 | 682 | void vhost_dev_cleanup(struct vhost_dev *dev) |
3a4d5c94 MT |
683 | { |
684 | int i; | |
d47effe1 | 685 | |
3a4d5c94 | 686 | for (i = 0; i < dev->nvqs; ++i) { |
3ab2e420 AH |
687 | if (dev->vqs[i]->error_ctx) |
688 | eventfd_ctx_put(dev->vqs[i]->error_ctx); | |
3ab2e420 AH |
689 | if (dev->vqs[i]->kick) |
690 | fput(dev->vqs[i]->kick); | |
265a0ad8 ZL |
691 | if (dev->vqs[i]->call_ctx.ctx) |
692 | eventfd_ctx_put(dev->vqs[i]->call_ctx.ctx); | |
3ab2e420 | 693 | vhost_vq_reset(dev, dev->vqs[i]); |
3a4d5c94 | 694 | } |
e0e9b406 | 695 | vhost_dev_free_iovecs(dev); |
3a4d5c94 MT |
696 | if (dev->log_ctx) |
697 | eventfd_ctx_put(dev->log_ctx); | |
698 | dev->log_ctx = NULL; | |
3a4d5c94 | 699 | /* No one will access memory at this point */ |
0bbe3066 | 700 | vhost_iotlb_free(dev->umem); |
a9709d68 | 701 | dev->umem = NULL; |
0bbe3066 | 702 | vhost_iotlb_free(dev->iotlb); |
6b1e6cc7 JW |
703 | dev->iotlb = NULL; |
704 | vhost_clear_msg(dev); | |
a9a08845 | 705 | wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM); |
1a5f8090 | 706 | vhost_worker_free(dev); |
5ce995f3 | 707 | vhost_detach_mm(dev); |
3a4d5c94 | 708 | } |
6ac1afbf | 709 | EXPORT_SYMBOL_GPL(vhost_dev_cleanup); |
3a4d5c94 | 710 | |
ddd3d408 | 711 | static bool log_access_ok(void __user *log_base, u64 addr, unsigned long sz) |
3a4d5c94 MT |
712 | { |
713 | u64 a = addr / VHOST_PAGE_SIZE / 8; | |
d47effe1 | 714 | |
3a4d5c94 MT |
715 | /* Make sure 64 bit math will not overflow. */ |
716 | if (a > ULONG_MAX - (unsigned long)log_base || | |
717 | a + (unsigned long)log_base > ULONG_MAX) | |
ddd3d408 | 718 | return false; |
3a4d5c94 | 719 | |
96d4f267 | 720 | return access_ok(log_base + a, |
3a4d5c94 MT |
721 | (sz + VHOST_PAGE_SIZE * 8 - 1) / VHOST_PAGE_SIZE / 8); |
722 | } | |
723 | ||
f7ad318e | 724 | /* Make sure 64 bit math will not overflow. */ |
ec33d031 MT |
725 | static bool vhost_overflow(u64 uaddr, u64 size) |
726 | { | |
f7ad318e XY |
727 | if (uaddr > ULONG_MAX || size > ULONG_MAX) |
728 | return true; | |
729 | ||
730 | if (!size) | |
731 | return false; | |
732 | ||
733 | return uaddr > ULONG_MAX - size + 1; | |
ec33d031 MT |
734 | } |
735 | ||
3a4d5c94 | 736 | /* Caller should have vq mutex and device mutex. */ |
0bbe3066 | 737 | static bool vq_memory_access_ok(void __user *log_base, struct vhost_iotlb *umem, |
ddd3d408 | 738 | int log_all) |
3a4d5c94 | 739 | { |
0bbe3066 | 740 | struct vhost_iotlb_map *map; |
179b284e | 741 | |
a9709d68 | 742 | if (!umem) |
ddd3d408 | 743 | return false; |
179b284e | 744 | |
0bbe3066 JW |
745 | list_for_each_entry(map, &umem->list, link) { |
746 | unsigned long a = map->addr; | |
a9709d68 | 747 | |
0bbe3066 | 748 | if (vhost_overflow(map->addr, map->size)) |
ddd3d408 | 749 | return false; |
ec33d031 MT |
750 | |
751 | ||
0bbe3066 | 752 | if (!access_ok((void __user *)a, map->size)) |
ddd3d408 | 753 | return false; |
3a4d5c94 | 754 | else if (log_all && !log_access_ok(log_base, |
0bbe3066 JW |
755 | map->start, |
756 | map->size)) | |
ddd3d408 | 757 | return false; |
3a4d5c94 | 758 | } |
ddd3d408 | 759 | return true; |
3a4d5c94 MT |
760 | } |
761 | ||
f8894913 JW |
762 | static inline void __user *vhost_vq_meta_fetch(struct vhost_virtqueue *vq, |
763 | u64 addr, unsigned int size, | |
764 | int type) | |
765 | { | |
0bbe3066 | 766 | const struct vhost_iotlb_map *map = vq->meta_iotlb[type]; |
f8894913 | 767 | |
0bbe3066 | 768 | if (!map) |
f8894913 JW |
769 | return NULL; |
770 | ||
1b0be99f | 771 | return (void __user *)(uintptr_t)(map->addr + addr - map->start); |
f8894913 JW |
772 | } |
773 | ||
3a4d5c94 MT |
774 | /* Can we switch to this memory table? */ |
775 | /* Caller should have device mutex but not vq mutex */ | |
0bbe3066 | 776 | static bool memory_access_ok(struct vhost_dev *d, struct vhost_iotlb *umem, |
ddd3d408 | 777 | int log_all) |
3a4d5c94 MT |
778 | { |
779 | int i; | |
d47effe1 | 780 | |
3a4d5c94 | 781 | for (i = 0; i < d->nvqs; ++i) { |
ddd3d408 | 782 | bool ok; |
ea16c514 MT |
783 | bool log; |
784 | ||
3ab2e420 | 785 | mutex_lock(&d->vqs[i]->mutex); |
ea16c514 | 786 | log = log_all || vhost_has_feature(d->vqs[i], VHOST_F_LOG_ALL); |
3a4d5c94 | 787 | /* If ring is inactive, will check when it's enabled. */ |
3ab2e420 | 788 | if (d->vqs[i]->private_data) |
a9709d68 JW |
789 | ok = vq_memory_access_ok(d->vqs[i]->log_base, |
790 | umem, log); | |
3a4d5c94 | 791 | else |
ddd3d408 | 792 | ok = true; |
3ab2e420 | 793 | mutex_unlock(&d->vqs[i]->mutex); |
3a4d5c94 | 794 | if (!ok) |
ddd3d408 | 795 | return false; |
3a4d5c94 | 796 | } |
ddd3d408 | 797 | return true; |
3a4d5c94 MT |
798 | } |
799 | ||
6b1e6cc7 JW |
800 | static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len, |
801 | struct iovec iov[], int iov_size, int access); | |
bfe2bc51 | 802 | |
72952cc0 | 803 | static int vhost_copy_to_user(struct vhost_virtqueue *vq, void __user *to, |
bfe2bc51 JW |
804 | const void *from, unsigned size) |
805 | { | |
6b1e6cc7 | 806 | int ret; |
bfe2bc51 | 807 | |
6b1e6cc7 JW |
808 | if (!vq->iotlb) |
809 | return __copy_to_user(to, from, size); | |
810 | else { | |
811 | /* This function should be called after iotlb | |
812 | * prefetch, which means we're sure that all vq | |
813 | * could be access through iotlb. So -EAGAIN should | |
814 | * not happen in this case. | |
815 | */ | |
6b1e6cc7 | 816 | struct iov_iter t; |
f8894913 JW |
817 | void __user *uaddr = vhost_vq_meta_fetch(vq, |
818 | (u64)(uintptr_t)to, size, | |
7ced6c98 | 819 | VHOST_ADDR_USED); |
f8894913 JW |
820 | |
821 | if (uaddr) | |
822 | return __copy_to_user(uaddr, from, size); | |
823 | ||
6b1e6cc7 JW |
824 | ret = translate_desc(vq, (u64)(uintptr_t)to, size, vq->iotlb_iov, |
825 | ARRAY_SIZE(vq->iotlb_iov), | |
826 | VHOST_ACCESS_WO); | |
827 | if (ret < 0) | |
828 | goto out; | |
de4eda9d | 829 | iov_iter_init(&t, ITER_DEST, vq->iotlb_iov, ret, size); |
6b1e6cc7 JW |
830 | ret = copy_to_iter(from, size, &t); |
831 | if (ret == size) | |
832 | ret = 0; | |
833 | } | |
834 | out: | |
835 | return ret; | |
836 | } | |
bfe2bc51 JW |
837 | |
838 | static int vhost_copy_from_user(struct vhost_virtqueue *vq, void *to, | |
72952cc0 | 839 | void __user *from, unsigned size) |
bfe2bc51 | 840 | { |
6b1e6cc7 JW |
841 | int ret; |
842 | ||
843 | if (!vq->iotlb) | |
844 | return __copy_from_user(to, from, size); | |
845 | else { | |
846 | /* This function should be called after iotlb | |
847 | * prefetch, which means we're sure that vq | |
848 | * could be access through iotlb. So -EAGAIN should | |
849 | * not happen in this case. | |
850 | */ | |
f8894913 JW |
851 | void __user *uaddr = vhost_vq_meta_fetch(vq, |
852 | (u64)(uintptr_t)from, size, | |
853 | VHOST_ADDR_DESC); | |
6b1e6cc7 | 854 | struct iov_iter f; |
f8894913 JW |
855 | |
856 | if (uaddr) | |
857 | return __copy_from_user(to, uaddr, size); | |
858 | ||
6b1e6cc7 JW |
859 | ret = translate_desc(vq, (u64)(uintptr_t)from, size, vq->iotlb_iov, |
860 | ARRAY_SIZE(vq->iotlb_iov), | |
861 | VHOST_ACCESS_RO); | |
862 | if (ret < 0) { | |
863 | vq_err(vq, "IOTLB translation failure: uaddr " | |
864 | "%p size 0x%llx\n", from, | |
865 | (unsigned long long) size); | |
866 | goto out; | |
867 | } | |
de4eda9d | 868 | iov_iter_init(&f, ITER_SOURCE, vq->iotlb_iov, ret, size); |
6b1e6cc7 JW |
869 | ret = copy_from_iter(to, size, &f); |
870 | if (ret == size) | |
871 | ret = 0; | |
872 | } | |
873 | ||
874 | out: | |
875 | return ret; | |
876 | } | |
877 | ||
f8894913 JW |
878 | static void __user *__vhost_get_user_slow(struct vhost_virtqueue *vq, |
879 | void __user *addr, unsigned int size, | |
880 | int type) | |
6b1e6cc7 JW |
881 | { |
882 | int ret; | |
883 | ||
6b1e6cc7 JW |
884 | ret = translate_desc(vq, (u64)(uintptr_t)addr, size, vq->iotlb_iov, |
885 | ARRAY_SIZE(vq->iotlb_iov), | |
886 | VHOST_ACCESS_RO); | |
887 | if (ret < 0) { | |
888 | vq_err(vq, "IOTLB translation failure: uaddr " | |
889 | "%p size 0x%llx\n", addr, | |
890 | (unsigned long long) size); | |
891 | return NULL; | |
892 | } | |
893 | ||
894 | if (ret != 1 || vq->iotlb_iov[0].iov_len != size) { | |
895 | vq_err(vq, "Non atomic userspace memory access: uaddr " | |
896 | "%p size 0x%llx\n", addr, | |
897 | (unsigned long long) size); | |
898 | return NULL; | |
899 | } | |
900 | ||
901 | return vq->iotlb_iov[0].iov_base; | |
902 | } | |
903 | ||
f8894913 JW |
904 | /* This function should be called after iotlb |
905 | * prefetch, which means we're sure that vq | |
906 | * could be access through iotlb. So -EAGAIN should | |
907 | * not happen in this case. | |
908 | */ | |
909 | static inline void __user *__vhost_get_user(struct vhost_virtqueue *vq, | |
1b0be99f | 910 | void __user *addr, unsigned int size, |
f8894913 JW |
911 | int type) |
912 | { | |
913 | void __user *uaddr = vhost_vq_meta_fetch(vq, | |
914 | (u64)(uintptr_t)addr, size, type); | |
915 | if (uaddr) | |
916 | return uaddr; | |
917 | ||
918 | return __vhost_get_user_slow(vq, addr, size, type); | |
919 | } | |
920 | ||
921 | #define vhost_put_user(vq, x, ptr) \ | |
6b1e6cc7 | 922 | ({ \ |
002ef18e | 923 | int ret; \ |
6b1e6cc7 JW |
924 | if (!vq->iotlb) { \ |
925 | ret = __put_user(x, ptr); \ | |
926 | } else { \ | |
927 | __typeof__(ptr) to = \ | |
f8894913 JW |
928 | (__typeof__(ptr)) __vhost_get_user(vq, ptr, \ |
929 | sizeof(*ptr), VHOST_ADDR_USED); \ | |
6b1e6cc7 JW |
930 | if (to != NULL) \ |
931 | ret = __put_user(x, to); \ | |
932 | else \ | |
933 | ret = -EFAULT; \ | |
934 | } \ | |
935 | ret; \ | |
936 | }) | |
937 | ||
7b5d753e JW |
938 | static inline int vhost_put_avail_event(struct vhost_virtqueue *vq) |
939 | { | |
940 | return vhost_put_user(vq, cpu_to_vhost16(vq, vq->avail_idx), | |
941 | vhost_avail_event(vq)); | |
942 | } | |
943 | ||
944 | static inline int vhost_put_used(struct vhost_virtqueue *vq, | |
945 | struct vring_used_elem *head, int idx, | |
946 | int count) | |
947 | { | |
948 | return vhost_copy_to_user(vq, vq->used->ring + idx, head, | |
949 | count * sizeof(*head)); | |
950 | } | |
951 | ||
952 | static inline int vhost_put_used_flags(struct vhost_virtqueue *vq) | |
953 | ||
954 | { | |
955 | return vhost_put_user(vq, cpu_to_vhost16(vq, vq->used_flags), | |
956 | &vq->used->flags); | |
957 | } | |
958 | ||
959 | static inline int vhost_put_used_idx(struct vhost_virtqueue *vq) | |
960 | ||
961 | { | |
962 | return vhost_put_user(vq, cpu_to_vhost16(vq, vq->last_used_idx), | |
963 | &vq->used->idx); | |
964 | } | |
965 | ||
f8894913 | 966 | #define vhost_get_user(vq, x, ptr, type) \ |
6b1e6cc7 JW |
967 | ({ \ |
968 | int ret; \ | |
969 | if (!vq->iotlb) { \ | |
970 | ret = __get_user(x, ptr); \ | |
971 | } else { \ | |
972 | __typeof__(ptr) from = \ | |
f8894913 JW |
973 | (__typeof__(ptr)) __vhost_get_user(vq, ptr, \ |
974 | sizeof(*ptr), \ | |
975 | type); \ | |
6b1e6cc7 JW |
976 | if (from != NULL) \ |
977 | ret = __get_user(x, from); \ | |
978 | else \ | |
979 | ret = -EFAULT; \ | |
980 | } \ | |
981 | ret; \ | |
982 | }) | |
983 | ||
f8894913 JW |
984 | #define vhost_get_avail(vq, x, ptr) \ |
985 | vhost_get_user(vq, x, ptr, VHOST_ADDR_AVAIL) | |
986 | ||
987 | #define vhost_get_used(vq, x, ptr) \ | |
988 | vhost_get_user(vq, x, ptr, VHOST_ADDR_USED) | |
989 | ||
86a07da3 JW |
990 | static void vhost_dev_lock_vqs(struct vhost_dev *d) |
991 | { | |
992 | int i = 0; | |
993 | for (i = 0; i < d->nvqs; ++i) | |
994 | mutex_lock_nested(&d->vqs[i]->mutex, i); | |
995 | } | |
996 | ||
997 | static void vhost_dev_unlock_vqs(struct vhost_dev *d) | |
998 | { | |
999 | int i = 0; | |
1000 | for (i = 0; i < d->nvqs; ++i) | |
1001 | mutex_unlock(&d->vqs[i]->mutex); | |
1002 | } | |
1003 | ||
7b5d753e JW |
1004 | static inline int vhost_get_avail_idx(struct vhost_virtqueue *vq, |
1005 | __virtio16 *idx) | |
1006 | { | |
1007 | return vhost_get_avail(vq, *idx, &vq->avail->idx); | |
1008 | } | |
1009 | ||
1010 | static inline int vhost_get_avail_head(struct vhost_virtqueue *vq, | |
1011 | __virtio16 *head, int idx) | |
1012 | { | |
1013 | return vhost_get_avail(vq, *head, | |
1014 | &vq->avail->ring[idx & (vq->num - 1)]); | |
1015 | } | |
1016 | ||
1017 | static inline int vhost_get_avail_flags(struct vhost_virtqueue *vq, | |
1018 | __virtio16 *flags) | |
1019 | { | |
1020 | return vhost_get_avail(vq, *flags, &vq->avail->flags); | |
1021 | } | |
1022 | ||
1023 | static inline int vhost_get_used_event(struct vhost_virtqueue *vq, | |
1024 | __virtio16 *event) | |
1025 | { | |
1026 | return vhost_get_avail(vq, *event, vhost_used_event(vq)); | |
1027 | } | |
1028 | ||
1029 | static inline int vhost_get_used_idx(struct vhost_virtqueue *vq, | |
1030 | __virtio16 *idx) | |
1031 | { | |
1032 | return vhost_get_used(vq, *idx, &vq->used->idx); | |
1033 | } | |
1034 | ||
1035 | static inline int vhost_get_desc(struct vhost_virtqueue *vq, | |
1036 | struct vring_desc *desc, int idx) | |
1037 | { | |
1038 | return vhost_copy_from_user(vq, desc, vq->desc + idx, sizeof(*desc)); | |
1039 | } | |
1040 | ||
6b1e6cc7 JW |
1041 | static void vhost_iotlb_notify_vq(struct vhost_dev *d, |
1042 | struct vhost_iotlb_msg *msg) | |
1043 | { | |
1044 | struct vhost_msg_node *node, *n; | |
1045 | ||
1046 | spin_lock(&d->iotlb_lock); | |
1047 | ||
1048 | list_for_each_entry_safe(node, n, &d->pending_list, node) { | |
1049 | struct vhost_iotlb_msg *vq_msg = &node->msg.iotlb; | |
1050 | if (msg->iova <= vq_msg->iova && | |
2d66f997 | 1051 | msg->iova + msg->size - 1 >= vq_msg->iova && |
6b1e6cc7 JW |
1052 | vq_msg->type == VHOST_IOTLB_MISS) { |
1053 | vhost_poll_queue(&node->vq->poll); | |
1054 | list_del(&node->node); | |
1055 | kfree(node); | |
1056 | } | |
1057 | } | |
1058 | ||
1059 | spin_unlock(&d->iotlb_lock); | |
1060 | } | |
1061 | ||
ddd3d408 | 1062 | static bool umem_access_ok(u64 uaddr, u64 size, int access) |
6b1e6cc7 JW |
1063 | { |
1064 | unsigned long a = uaddr; | |
1065 | ||
ec33d031 MT |
1066 | /* Make sure 64 bit math will not overflow. */ |
1067 | if (vhost_overflow(uaddr, size)) | |
ddd3d408 | 1068 | return false; |
ec33d031 | 1069 | |
6b1e6cc7 | 1070 | if ((access & VHOST_ACCESS_RO) && |
96d4f267 | 1071 | !access_ok((void __user *)a, size)) |
ddd3d408 | 1072 | return false; |
6b1e6cc7 | 1073 | if ((access & VHOST_ACCESS_WO) && |
96d4f267 | 1074 | !access_ok((void __user *)a, size)) |
ddd3d408 SH |
1075 | return false; |
1076 | return true; | |
6b1e6cc7 JW |
1077 | } |
1078 | ||
91233ad7 | 1079 | static int vhost_process_iotlb_msg(struct vhost_dev *dev, u32 asid, |
72952cc0 | 1080 | struct vhost_iotlb_msg *msg) |
6b1e6cc7 JW |
1081 | { |
1082 | int ret = 0; | |
1083 | ||
91233ad7 GD |
1084 | if (asid != 0) |
1085 | return -EINVAL; | |
1086 | ||
1b15ad68 | 1087 | mutex_lock(&dev->mutex); |
86a07da3 | 1088 | vhost_dev_lock_vqs(dev); |
6b1e6cc7 JW |
1089 | switch (msg->type) { |
1090 | case VHOST_IOTLB_UPDATE: | |
1091 | if (!dev->iotlb) { | |
1092 | ret = -EFAULT; | |
1093 | break; | |
1094 | } | |
ddd3d408 | 1095 | if (!umem_access_ok(msg->uaddr, msg->size, msg->perm)) { |
6b1e6cc7 JW |
1096 | ret = -EFAULT; |
1097 | break; | |
1098 | } | |
f8894913 | 1099 | vhost_vq_meta_reset(dev); |
0bbe3066 JW |
1100 | if (vhost_iotlb_add_range(dev->iotlb, msg->iova, |
1101 | msg->iova + msg->size - 1, | |
1102 | msg->uaddr, msg->perm)) { | |
6b1e6cc7 JW |
1103 | ret = -ENOMEM; |
1104 | break; | |
1105 | } | |
1106 | vhost_iotlb_notify_vq(dev, msg); | |
1107 | break; | |
1108 | case VHOST_IOTLB_INVALIDATE: | |
6f3180af JW |
1109 | if (!dev->iotlb) { |
1110 | ret = -EFAULT; | |
1111 | break; | |
1112 | } | |
f8894913 | 1113 | vhost_vq_meta_reset(dev); |
0bbe3066 JW |
1114 | vhost_iotlb_del_range(dev->iotlb, msg->iova, |
1115 | msg->iova + msg->size - 1); | |
6b1e6cc7 JW |
1116 | break; |
1117 | default: | |
1118 | ret = -EINVAL; | |
1119 | break; | |
1120 | } | |
1121 | ||
86a07da3 | 1122 | vhost_dev_unlock_vqs(dev); |
1b15ad68 JW |
1123 | mutex_unlock(&dev->mutex); |
1124 | ||
6b1e6cc7 JW |
1125 | return ret; |
1126 | } | |
1127 | ssize_t vhost_chr_write_iter(struct vhost_dev *dev, | |
1128 | struct iov_iter *from) | |
1129 | { | |
429711ae JW |
1130 | struct vhost_iotlb_msg msg; |
1131 | size_t offset; | |
1132 | int type, ret; | |
91233ad7 | 1133 | u32 asid = 0; |
6b1e6cc7 | 1134 | |
429711ae | 1135 | ret = copy_from_iter(&type, sizeof(type), from); |
74ad7419 PT |
1136 | if (ret != sizeof(type)) { |
1137 | ret = -EINVAL; | |
6b1e6cc7 | 1138 | goto done; |
74ad7419 | 1139 | } |
6b1e6cc7 | 1140 | |
429711ae | 1141 | switch (type) { |
6b1e6cc7 | 1142 | case VHOST_IOTLB_MSG: |
429711ae JW |
1143 | /* There maybe a hole after type for V1 message type, |
1144 | * so skip it here. | |
1145 | */ | |
1146 | offset = offsetof(struct vhost_msg, iotlb) - sizeof(int); | |
1147 | break; | |
1148 | case VHOST_IOTLB_MSG_V2: | |
91233ad7 GD |
1149 | if (vhost_backend_has_feature(dev->vqs[0], |
1150 | VHOST_BACKEND_F_IOTLB_ASID)) { | |
1151 | ret = copy_from_iter(&asid, sizeof(asid), from); | |
1152 | if (ret != sizeof(asid)) { | |
1153 | ret = -EINVAL; | |
1154 | goto done; | |
1155 | } | |
aaca8373 | 1156 | offset = 0; |
91233ad7 GD |
1157 | } else |
1158 | offset = sizeof(__u32); | |
6b1e6cc7 JW |
1159 | break; |
1160 | default: | |
1161 | ret = -EINVAL; | |
429711ae | 1162 | goto done; |
6b1e6cc7 JW |
1163 | } |
1164 | ||
429711ae JW |
1165 | iov_iter_advance(from, offset); |
1166 | ret = copy_from_iter(&msg, sizeof(msg), from); | |
74ad7419 PT |
1167 | if (ret != sizeof(msg)) { |
1168 | ret = -EINVAL; | |
429711ae | 1169 | goto done; |
74ad7419 | 1170 | } |
792a4f2e | 1171 | |
95932ab2 JW |
1172 | if ((msg.type == VHOST_IOTLB_UPDATE || |
1173 | msg.type == VHOST_IOTLB_INVALIDATE) && | |
1174 | msg.size == 0) { | |
e2ae38cf AR |
1175 | ret = -EINVAL; |
1176 | goto done; | |
1177 | } | |
1178 | ||
792a4f2e | 1179 | if (dev->msg_handler) |
91233ad7 | 1180 | ret = dev->msg_handler(dev, asid, &msg); |
792a4f2e | 1181 | else |
91233ad7 | 1182 | ret = vhost_process_iotlb_msg(dev, asid, &msg); |
792a4f2e | 1183 | if (ret) { |
429711ae JW |
1184 | ret = -EFAULT; |
1185 | goto done; | |
1186 | } | |
1187 | ||
1188 | ret = (type == VHOST_IOTLB_MSG) ? sizeof(struct vhost_msg) : | |
1189 | sizeof(struct vhost_msg_v2); | |
6b1e6cc7 JW |
1190 | done: |
1191 | return ret; | |
1192 | } | |
1193 | EXPORT_SYMBOL(vhost_chr_write_iter); | |
1194 | ||
afc9a42b | 1195 | __poll_t vhost_chr_poll(struct file *file, struct vhost_dev *dev, |
6b1e6cc7 JW |
1196 | poll_table *wait) |
1197 | { | |
afc9a42b | 1198 | __poll_t mask = 0; |
6b1e6cc7 JW |
1199 | |
1200 | poll_wait(file, &dev->wait, wait); | |
1201 | ||
1202 | if (!list_empty(&dev->read_list)) | |
a9a08845 | 1203 | mask |= EPOLLIN | EPOLLRDNORM; |
6b1e6cc7 JW |
1204 | |
1205 | return mask; | |
1206 | } | |
1207 | EXPORT_SYMBOL(vhost_chr_poll); | |
1208 | ||
1209 | ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to, | |
1210 | int noblock) | |
1211 | { | |
1212 | DEFINE_WAIT(wait); | |
1213 | struct vhost_msg_node *node; | |
1214 | ssize_t ret = 0; | |
1215 | unsigned size = sizeof(struct vhost_msg); | |
1216 | ||
1217 | if (iov_iter_count(to) < size) | |
1218 | return 0; | |
1219 | ||
1220 | while (1) { | |
1221 | if (!noblock) | |
1222 | prepare_to_wait(&dev->wait, &wait, | |
1223 | TASK_INTERRUPTIBLE); | |
1224 | ||
1225 | node = vhost_dequeue_msg(dev, &dev->read_list); | |
1226 | if (node) | |
1227 | break; | |
1228 | if (noblock) { | |
1229 | ret = -EAGAIN; | |
1230 | break; | |
1231 | } | |
1232 | if (signal_pending(current)) { | |
1233 | ret = -ERESTARTSYS; | |
1234 | break; | |
1235 | } | |
1236 | if (!dev->iotlb) { | |
1237 | ret = -EBADFD; | |
1238 | break; | |
1239 | } | |
1240 | ||
1241 | schedule(); | |
1242 | } | |
1243 | ||
1244 | if (!noblock) | |
1245 | finish_wait(&dev->wait, &wait); | |
1246 | ||
1247 | if (node) { | |
429711ae JW |
1248 | struct vhost_iotlb_msg *msg; |
1249 | void *start = &node->msg; | |
6b1e6cc7 | 1250 | |
429711ae JW |
1251 | switch (node->msg.type) { |
1252 | case VHOST_IOTLB_MSG: | |
1253 | size = sizeof(node->msg); | |
1254 | msg = &node->msg.iotlb; | |
1255 | break; | |
1256 | case VHOST_IOTLB_MSG_V2: | |
1257 | size = sizeof(node->msg_v2); | |
1258 | msg = &node->msg_v2.iotlb; | |
1259 | break; | |
1260 | default: | |
1261 | BUG(); | |
1262 | break; | |
1263 | } | |
1264 | ||
1265 | ret = copy_to_iter(start, size, to); | |
1266 | if (ret != size || msg->type != VHOST_IOTLB_MISS) { | |
6b1e6cc7 JW |
1267 | kfree(node); |
1268 | return ret; | |
1269 | } | |
6b1e6cc7 JW |
1270 | vhost_enqueue_msg(dev, &dev->pending_list, node); |
1271 | } | |
1272 | ||
1273 | return ret; | |
1274 | } | |
1275 | EXPORT_SYMBOL_GPL(vhost_chr_read_iter); | |
1276 | ||
1277 | static int vhost_iotlb_miss(struct vhost_virtqueue *vq, u64 iova, int access) | |
1278 | { | |
1279 | struct vhost_dev *dev = vq->dev; | |
1280 | struct vhost_msg_node *node; | |
1281 | struct vhost_iotlb_msg *msg; | |
429711ae | 1282 | bool v2 = vhost_backend_has_feature(vq, VHOST_BACKEND_F_IOTLB_MSG_V2); |
6b1e6cc7 | 1283 | |
429711ae | 1284 | node = vhost_new_msg(vq, v2 ? VHOST_IOTLB_MSG_V2 : VHOST_IOTLB_MSG); |
6b1e6cc7 JW |
1285 | if (!node) |
1286 | return -ENOMEM; | |
1287 | ||
429711ae JW |
1288 | if (v2) { |
1289 | node->msg_v2.type = VHOST_IOTLB_MSG_V2; | |
1290 | msg = &node->msg_v2.iotlb; | |
1291 | } else { | |
1292 | msg = &node->msg.iotlb; | |
1293 | } | |
1294 | ||
6b1e6cc7 JW |
1295 | msg->type = VHOST_IOTLB_MISS; |
1296 | msg->iova = iova; | |
1297 | msg->perm = access; | |
1298 | ||
1299 | vhost_enqueue_msg(dev, &dev->read_list, node); | |
1300 | ||
1301 | return 0; | |
bfe2bc51 JW |
1302 | } |
1303 | ||
ddd3d408 | 1304 | static bool vq_access_ok(struct vhost_virtqueue *vq, unsigned int num, |
a865e420 MT |
1305 | vring_desc_t __user *desc, |
1306 | vring_avail_t __user *avail, | |
1307 | vring_used_t __user *used) | |
6b1e6cc7 | 1308 | |
3a4d5c94 | 1309 | { |
0210a8db GK |
1310 | /* If an IOTLB device is present, the vring addresses are |
1311 | * GIOVAs. Access validation occurs at prefetch time. */ | |
1312 | if (vq->iotlb) | |
1313 | return true; | |
1314 | ||
4942e825 JW |
1315 | return access_ok(desc, vhost_get_desc_size(vq, num)) && |
1316 | access_ok(avail, vhost_get_avail_size(vq, num)) && | |
1317 | access_ok(used, vhost_get_used_size(vq, num)); | |
3a4d5c94 MT |
1318 | } |
1319 | ||
f8894913 | 1320 | static void vhost_vq_meta_update(struct vhost_virtqueue *vq, |
0bbe3066 | 1321 | const struct vhost_iotlb_map *map, |
f8894913 JW |
1322 | int type) |
1323 | { | |
1324 | int access = (type == VHOST_ADDR_USED) ? | |
1325 | VHOST_ACCESS_WO : VHOST_ACCESS_RO; | |
1326 | ||
0bbe3066 JW |
1327 | if (likely(map->perm & access)) |
1328 | vq->meta_iotlb[type] = map; | |
f8894913 JW |
1329 | } |
1330 | ||
ddd3d408 SH |
1331 | static bool iotlb_access_ok(struct vhost_virtqueue *vq, |
1332 | int access, u64 addr, u64 len, int type) | |
6b1e6cc7 | 1333 | { |
0bbe3066 JW |
1334 | const struct vhost_iotlb_map *map; |
1335 | struct vhost_iotlb *umem = vq->iotlb; | |
ca2c5b33 | 1336 | u64 s = 0, size, orig_addr = addr, last = addr + len - 1; |
f8894913 JW |
1337 | |
1338 | if (vhost_vq_meta_fetch(vq, addr, len, type)) | |
1339 | return true; | |
6b1e6cc7 JW |
1340 | |
1341 | while (len > s) { | |
0bbe3066 JW |
1342 | map = vhost_iotlb_itree_first(umem, addr, last); |
1343 | if (map == NULL || map->start > addr) { | |
6b1e6cc7 JW |
1344 | vhost_iotlb_miss(vq, addr, access); |
1345 | return false; | |
0bbe3066 | 1346 | } else if (!(map->perm & access)) { |
6b1e6cc7 JW |
1347 | /* Report the possible access violation by |
1348 | * request another translation from userspace. | |
1349 | */ | |
1350 | return false; | |
1351 | } | |
1352 | ||
0bbe3066 | 1353 | size = map->size - addr + map->start; |
f8894913 JW |
1354 | |
1355 | if (orig_addr == addr && size >= len) | |
0bbe3066 | 1356 | vhost_vq_meta_update(vq, map, type); |
f8894913 | 1357 | |
6b1e6cc7 JW |
1358 | s += size; |
1359 | addr += size; | |
1360 | } | |
1361 | ||
1362 | return true; | |
1363 | } | |
1364 | ||
9b5e830b | 1365 | int vq_meta_prefetch(struct vhost_virtqueue *vq) |
6b1e6cc7 | 1366 | { |
6b1e6cc7 JW |
1367 | unsigned int num = vq->num; |
1368 | ||
3d2c7d37 | 1369 | if (!vq->iotlb) |
6b1e6cc7 JW |
1370 | return 1; |
1371 | ||
0bbe3066 | 1372 | return iotlb_access_ok(vq, VHOST_MAP_RO, (u64)(uintptr_t)vq->desc, |
4942e825 | 1373 | vhost_get_desc_size(vq, num), VHOST_ADDR_DESC) && |
0bbe3066 | 1374 | iotlb_access_ok(vq, VHOST_MAP_RO, (u64)(uintptr_t)vq->avail, |
4942e825 | 1375 | vhost_get_avail_size(vq, num), |
f8894913 | 1376 | VHOST_ADDR_AVAIL) && |
0bbe3066 | 1377 | iotlb_access_ok(vq, VHOST_MAP_WO, (u64)(uintptr_t)vq->used, |
4942e825 | 1378 | vhost_get_used_size(vq, num), VHOST_ADDR_USED); |
6b1e6cc7 | 1379 | } |
9b5e830b | 1380 | EXPORT_SYMBOL_GPL(vq_meta_prefetch); |
6b1e6cc7 | 1381 | |
3a4d5c94 MT |
1382 | /* Can we log writes? */ |
1383 | /* Caller should have device mutex but not vq mutex */ | |
ddd3d408 | 1384 | bool vhost_log_access_ok(struct vhost_dev *dev) |
3a4d5c94 | 1385 | { |
a9709d68 | 1386 | return memory_access_ok(dev, dev->umem, 1); |
3a4d5c94 | 1387 | } |
6ac1afbf | 1388 | EXPORT_SYMBOL_GPL(vhost_log_access_ok); |
3a4d5c94 | 1389 | |
ab512251 GK |
1390 | static bool vq_log_used_access_ok(struct vhost_virtqueue *vq, |
1391 | void __user *log_base, | |
1392 | bool log_used, | |
1393 | u64 log_addr) | |
1394 | { | |
1395 | /* If an IOTLB device is present, log_addr is a GIOVA that | |
1396 | * will never be logged by log_used(). */ | |
1397 | if (vq->iotlb) | |
1398 | return true; | |
1399 | ||
1400 | return !log_used || log_access_ok(log_base, log_addr, | |
1401 | vhost_get_used_size(vq, vq->num)); | |
1402 | } | |
1403 | ||
3a4d5c94 MT |
1404 | /* Verify access for write logging. */ |
1405 | /* Caller should have vq mutex and device mutex */ | |
ddd3d408 SH |
1406 | static bool vq_log_access_ok(struct vhost_virtqueue *vq, |
1407 | void __user *log_base) | |
3a4d5c94 | 1408 | { |
a9709d68 | 1409 | return vq_memory_access_ok(log_base, vq->umem, |
ea16c514 | 1410 | vhost_has_feature(vq, VHOST_F_LOG_ALL)) && |
ab512251 | 1411 | vq_log_used_access_ok(vq, log_base, vq->log_used, vq->log_addr); |
3a4d5c94 MT |
1412 | } |
1413 | ||
1414 | /* Can we start vq? */ | |
1415 | /* Caller should have vq mutex and device mutex */ | |
ddd3d408 | 1416 | bool vhost_vq_access_ok(struct vhost_virtqueue *vq) |
3a4d5c94 | 1417 | { |
d14d2b78 | 1418 | if (!vq_log_access_ok(vq, vq->log_base)) |
ddd3d408 | 1419 | return false; |
d65026c6 | 1420 | |
d65026c6 | 1421 | return vq_access_ok(vq, vq->num, vq->desc, vq->avail, vq->used); |
3a4d5c94 | 1422 | } |
6ac1afbf | 1423 | EXPORT_SYMBOL_GPL(vhost_vq_access_ok); |
3a4d5c94 MT |
1424 | |
1425 | static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m) | |
1426 | { | |
a9709d68 JW |
1427 | struct vhost_memory mem, *newmem; |
1428 | struct vhost_memory_region *region; | |
0bbe3066 | 1429 | struct vhost_iotlb *newumem, *oldumem; |
3a4d5c94 | 1430 | unsigned long size = offsetof(struct vhost_memory, regions); |
98f9ca0a | 1431 | int i; |
d47effe1 | 1432 | |
7ad9c9d2 TY |
1433 | if (copy_from_user(&mem, m, size)) |
1434 | return -EFAULT; | |
3a4d5c94 MT |
1435 | if (mem.padding) |
1436 | return -EOPNOTSUPP; | |
c9ce42f7 | 1437 | if (mem.nregions > max_mem_regions) |
3a4d5c94 | 1438 | return -E2BIG; |
b2303d7b MW |
1439 | newmem = kvzalloc(struct_size(newmem, regions, mem.nregions), |
1440 | GFP_KERNEL); | |
3a4d5c94 MT |
1441 | if (!newmem) |
1442 | return -ENOMEM; | |
1443 | ||
1444 | memcpy(newmem, &mem, size); | |
7ad9c9d2 | 1445 | if (copy_from_user(newmem->regions, m->regions, |
bf11d71a | 1446 | flex_array_size(newmem, regions, mem.nregions))) { |
bcfeacab | 1447 | kvfree(newmem); |
7ad9c9d2 | 1448 | return -EFAULT; |
3a4d5c94 MT |
1449 | } |
1450 | ||
0bbe3066 | 1451 | newumem = iotlb_alloc(); |
a9709d68 | 1452 | if (!newumem) { |
4de7255f | 1453 | kvfree(newmem); |
a9709d68 JW |
1454 | return -ENOMEM; |
1455 | } | |
1456 | ||
a9709d68 JW |
1457 | for (region = newmem->regions; |
1458 | region < newmem->regions + mem.nregions; | |
1459 | region++) { | |
0bbe3066 JW |
1460 | if (vhost_iotlb_add_range(newumem, |
1461 | region->guest_phys_addr, | |
1462 | region->guest_phys_addr + | |
1463 | region->memory_size - 1, | |
1464 | region->userspace_addr, | |
1465 | VHOST_MAP_RW)) | |
a9709d68 | 1466 | goto err; |
a02c3789 | 1467 | } |
a9709d68 JW |
1468 | |
1469 | if (!memory_access_ok(d, newumem, 0)) | |
1470 | goto err; | |
1471 | ||
1472 | oldumem = d->umem; | |
1473 | d->umem = newumem; | |
98f9ca0a | 1474 | |
47283bef | 1475 | /* All memory accesses are done under some VQ mutex. */ |
98f9ca0a MT |
1476 | for (i = 0; i < d->nvqs; ++i) { |
1477 | mutex_lock(&d->vqs[i]->mutex); | |
a9709d68 | 1478 | d->vqs[i]->umem = newumem; |
98f9ca0a MT |
1479 | mutex_unlock(&d->vqs[i]->mutex); |
1480 | } | |
a9709d68 JW |
1481 | |
1482 | kvfree(newmem); | |
0bbe3066 | 1483 | vhost_iotlb_free(oldumem); |
3a4d5c94 | 1484 | return 0; |
a9709d68 JW |
1485 | |
1486 | err: | |
0bbe3066 | 1487 | vhost_iotlb_free(newumem); |
a9709d68 JW |
1488 | kvfree(newmem); |
1489 | return -EFAULT; | |
3a4d5c94 MT |
1490 | } |
1491 | ||
feebcaea JW |
1492 | static long vhost_vring_set_num(struct vhost_dev *d, |
1493 | struct vhost_virtqueue *vq, | |
1494 | void __user *argp) | |
1495 | { | |
1496 | struct vhost_vring_state s; | |
1497 | ||
1498 | /* Resizing ring with an active backend? | |
1499 | * You don't want to do that. */ | |
1500 | if (vq->private_data) | |
1501 | return -EBUSY; | |
1502 | ||
1503 | if (copy_from_user(&s, argp, sizeof s)) | |
1504 | return -EFAULT; | |
1505 | ||
1506 | if (!s.num || s.num > 0xffff || (s.num & (s.num - 1))) | |
1507 | return -EINVAL; | |
1508 | vq->num = s.num; | |
1509 | ||
1510 | return 0; | |
1511 | } | |
1512 | ||
1513 | static long vhost_vring_set_addr(struct vhost_dev *d, | |
1514 | struct vhost_virtqueue *vq, | |
1515 | void __user *argp) | |
1516 | { | |
1517 | struct vhost_vring_addr a; | |
1518 | ||
1519 | if (copy_from_user(&a, argp, sizeof a)) | |
1520 | return -EFAULT; | |
1521 | if (a.flags & ~(0x1 << VHOST_VRING_F_LOG)) | |
1522 | return -EOPNOTSUPP; | |
1523 | ||
1524 | /* For 32bit, verify that the top 32bits of the user | |
1525 | data are set to zero. */ | |
1526 | if ((u64)(unsigned long)a.desc_user_addr != a.desc_user_addr || | |
1527 | (u64)(unsigned long)a.used_user_addr != a.used_user_addr || | |
1528 | (u64)(unsigned long)a.avail_user_addr != a.avail_user_addr) | |
1529 | return -EFAULT; | |
1530 | ||
1531 | /* Make sure it's safe to cast pointers to vring types. */ | |
1532 | BUILD_BUG_ON(__alignof__ *vq->avail > VRING_AVAIL_ALIGN_SIZE); | |
1533 | BUILD_BUG_ON(__alignof__ *vq->used > VRING_USED_ALIGN_SIZE); | |
1534 | if ((a.avail_user_addr & (VRING_AVAIL_ALIGN_SIZE - 1)) || | |
1535 | (a.used_user_addr & (VRING_USED_ALIGN_SIZE - 1)) || | |
1536 | (a.log_guest_addr & (VRING_USED_ALIGN_SIZE - 1))) | |
1537 | return -EINVAL; | |
1538 | ||
1539 | /* We only verify access here if backend is configured. | |
1540 | * If it is not, we don't as size might not have been setup. | |
1541 | * We will verify when backend is configured. */ | |
1542 | if (vq->private_data) { | |
1543 | if (!vq_access_ok(vq, vq->num, | |
1544 | (void __user *)(unsigned long)a.desc_user_addr, | |
1545 | (void __user *)(unsigned long)a.avail_user_addr, | |
1546 | (void __user *)(unsigned long)a.used_user_addr)) | |
1547 | return -EINVAL; | |
1548 | ||
1549 | /* Also validate log access for used ring if enabled. */ | |
ab512251 GK |
1550 | if (!vq_log_used_access_ok(vq, vq->log_base, |
1551 | a.flags & (0x1 << VHOST_VRING_F_LOG), | |
1552 | a.log_guest_addr)) | |
feebcaea JW |
1553 | return -EINVAL; |
1554 | } | |
1555 | ||
1556 | vq->log_used = !!(a.flags & (0x1 << VHOST_VRING_F_LOG)); | |
1557 | vq->desc = (void __user *)(unsigned long)a.desc_user_addr; | |
1558 | vq->avail = (void __user *)(unsigned long)a.avail_user_addr; | |
1559 | vq->log_addr = a.log_guest_addr; | |
1560 | vq->used = (void __user *)(unsigned long)a.used_user_addr; | |
1561 | ||
1562 | return 0; | |
1563 | } | |
1564 | ||
1565 | static long vhost_vring_set_num_addr(struct vhost_dev *d, | |
1566 | struct vhost_virtqueue *vq, | |
1567 | unsigned int ioctl, | |
1568 | void __user *argp) | |
1569 | { | |
1570 | long r; | |
1571 | ||
1572 | mutex_lock(&vq->mutex); | |
1573 | ||
1574 | switch (ioctl) { | |
1575 | case VHOST_SET_VRING_NUM: | |
1576 | r = vhost_vring_set_num(d, vq, argp); | |
1577 | break; | |
1578 | case VHOST_SET_VRING_ADDR: | |
1579 | r = vhost_vring_set_addr(d, vq, argp); | |
1580 | break; | |
1581 | default: | |
1582 | BUG(); | |
1583 | } | |
1584 | ||
1585 | mutex_unlock(&vq->mutex); | |
1586 | ||
1587 | return r; | |
1588 | } | |
26b36604 | 1589 | long vhost_vring_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp) |
3a4d5c94 | 1590 | { |
cecb46f1 AV |
1591 | struct file *eventfp, *filep = NULL; |
1592 | bool pollstart = false, pollstop = false; | |
3a4d5c94 MT |
1593 | struct eventfd_ctx *ctx = NULL; |
1594 | u32 __user *idxp = argp; | |
1595 | struct vhost_virtqueue *vq; | |
1596 | struct vhost_vring_state s; | |
1597 | struct vhost_vring_file f; | |
3a4d5c94 MT |
1598 | u32 idx; |
1599 | long r; | |
1600 | ||
1601 | r = get_user(idx, idxp); | |
1602 | if (r < 0) | |
1603 | return r; | |
0f3d9a17 | 1604 | if (idx >= d->nvqs) |
3a4d5c94 MT |
1605 | return -ENOBUFS; |
1606 | ||
ff002269 | 1607 | idx = array_index_nospec(idx, d->nvqs); |
3ab2e420 | 1608 | vq = d->vqs[idx]; |
3a4d5c94 | 1609 | |
feebcaea JW |
1610 | if (ioctl == VHOST_SET_VRING_NUM || |
1611 | ioctl == VHOST_SET_VRING_ADDR) { | |
1612 | return vhost_vring_set_num_addr(d, vq, ioctl, argp); | |
1613 | } | |
1614 | ||
3a4d5c94 MT |
1615 | mutex_lock(&vq->mutex); |
1616 | ||
1617 | switch (ioctl) { | |
3a4d5c94 MT |
1618 | case VHOST_SET_VRING_BASE: |
1619 | /* Moving base with an active backend? | |
1620 | * You don't want to do that. */ | |
1621 | if (vq->private_data) { | |
1622 | r = -EBUSY; | |
1623 | break; | |
1624 | } | |
7ad9c9d2 TY |
1625 | if (copy_from_user(&s, argp, sizeof s)) { |
1626 | r = -EFAULT; | |
3a4d5c94 | 1627 | break; |
7ad9c9d2 | 1628 | } |
3a4d5c94 MT |
1629 | if (s.num > 0xffff) { |
1630 | r = -EINVAL; | |
1631 | break; | |
1632 | } | |
8d65843c | 1633 | vq->last_avail_idx = s.num; |
3a4d5c94 MT |
1634 | /* Forget the cached index value. */ |
1635 | vq->avail_idx = vq->last_avail_idx; | |
1636 | break; | |
1637 | case VHOST_GET_VRING_BASE: | |
1638 | s.index = idx; | |
1639 | s.num = vq->last_avail_idx; | |
7ad9c9d2 TY |
1640 | if (copy_to_user(argp, &s, sizeof s)) |
1641 | r = -EFAULT; | |
3a4d5c94 | 1642 | break; |
3a4d5c94 | 1643 | case VHOST_SET_VRING_KICK: |
7ad9c9d2 TY |
1644 | if (copy_from_user(&f, argp, sizeof f)) { |
1645 | r = -EFAULT; | |
3a4d5c94 | 1646 | break; |
7ad9c9d2 | 1647 | } |
e0136c16 | 1648 | eventfp = f.fd == VHOST_FILE_UNBIND ? NULL : eventfd_fget(f.fd); |
535297a6 MT |
1649 | if (IS_ERR(eventfp)) { |
1650 | r = PTR_ERR(eventfp); | |
1651 | break; | |
1652 | } | |
3a4d5c94 | 1653 | if (eventfp != vq->kick) { |
cecb46f1 AV |
1654 | pollstop = (filep = vq->kick) != NULL; |
1655 | pollstart = (vq->kick = eventfp) != NULL; | |
3a4d5c94 MT |
1656 | } else |
1657 | filep = eventfp; | |
1658 | break; | |
1659 | case VHOST_SET_VRING_CALL: | |
7ad9c9d2 TY |
1660 | if (copy_from_user(&f, argp, sizeof f)) { |
1661 | r = -EFAULT; | |
3a4d5c94 | 1662 | break; |
7ad9c9d2 | 1663 | } |
e0136c16 | 1664 | ctx = f.fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(f.fd); |
e050c7d9 EB |
1665 | if (IS_ERR(ctx)) { |
1666 | r = PTR_ERR(ctx); | |
535297a6 MT |
1667 | break; |
1668 | } | |
265a0ad8 | 1669 | |
265a0ad8 | 1670 | swap(ctx, vq->call_ctx.ctx); |
3a4d5c94 MT |
1671 | break; |
1672 | case VHOST_SET_VRING_ERR: | |
7ad9c9d2 TY |
1673 | if (copy_from_user(&f, argp, sizeof f)) { |
1674 | r = -EFAULT; | |
3a4d5c94 | 1675 | break; |
7ad9c9d2 | 1676 | } |
e0136c16 | 1677 | ctx = f.fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(f.fd); |
09f332a5 EB |
1678 | if (IS_ERR(ctx)) { |
1679 | r = PTR_ERR(ctx); | |
535297a6 MT |
1680 | break; |
1681 | } | |
09f332a5 | 1682 | swap(ctx, vq->error_ctx); |
3a4d5c94 | 1683 | break; |
2751c988 GK |
1684 | case VHOST_SET_VRING_ENDIAN: |
1685 | r = vhost_set_vring_endian(vq, argp); | |
1686 | break; | |
1687 | case VHOST_GET_VRING_ENDIAN: | |
1688 | r = vhost_get_vring_endian(vq, idx, argp); | |
1689 | break; | |
03088137 JW |
1690 | case VHOST_SET_VRING_BUSYLOOP_TIMEOUT: |
1691 | if (copy_from_user(&s, argp, sizeof(s))) { | |
1692 | r = -EFAULT; | |
1693 | break; | |
1694 | } | |
1695 | vq->busyloop_timeout = s.num; | |
1696 | break; | |
1697 | case VHOST_GET_VRING_BUSYLOOP_TIMEOUT: | |
1698 | s.index = idx; | |
1699 | s.num = vq->busyloop_timeout; | |
1700 | if (copy_to_user(argp, &s, sizeof(s))) | |
1701 | r = -EFAULT; | |
1702 | break; | |
3a4d5c94 MT |
1703 | default: |
1704 | r = -ENOIOCTLCMD; | |
1705 | } | |
1706 | ||
1707 | if (pollstop && vq->handle_kick) | |
1708 | vhost_poll_stop(&vq->poll); | |
1709 | ||
e050c7d9 | 1710 | if (!IS_ERR_OR_NULL(ctx)) |
3a4d5c94 MT |
1711 | eventfd_ctx_put(ctx); |
1712 | if (filep) | |
1713 | fput(filep); | |
1714 | ||
1715 | if (pollstart && vq->handle_kick) | |
2b8b328b | 1716 | r = vhost_poll_start(&vq->poll, vq->kick); |
3a4d5c94 MT |
1717 | |
1718 | mutex_unlock(&vq->mutex); | |
1719 | ||
1720 | if (pollstop && vq->handle_kick) | |
b2ffa407 | 1721 | vhost_dev_flush(vq->poll.dev); |
3a4d5c94 MT |
1722 | return r; |
1723 | } | |
6ac1afbf | 1724 | EXPORT_SYMBOL_GPL(vhost_vring_ioctl); |
3a4d5c94 | 1725 | |
759aba1e | 1726 | int vhost_init_device_iotlb(struct vhost_dev *d) |
6b1e6cc7 | 1727 | { |
0bbe3066 | 1728 | struct vhost_iotlb *niotlb, *oiotlb; |
6b1e6cc7 JW |
1729 | int i; |
1730 | ||
0bbe3066 | 1731 | niotlb = iotlb_alloc(); |
6b1e6cc7 JW |
1732 | if (!niotlb) |
1733 | return -ENOMEM; | |
1734 | ||
1735 | oiotlb = d->iotlb; | |
1736 | d->iotlb = niotlb; | |
1737 | ||
1738 | for (i = 0; i < d->nvqs; ++i) { | |
b13f9c63 JW |
1739 | struct vhost_virtqueue *vq = d->vqs[i]; |
1740 | ||
1741 | mutex_lock(&vq->mutex); | |
1742 | vq->iotlb = niotlb; | |
1743 | __vhost_vq_meta_reset(vq); | |
1744 | mutex_unlock(&vq->mutex); | |
6b1e6cc7 JW |
1745 | } |
1746 | ||
0bbe3066 | 1747 | vhost_iotlb_free(oiotlb); |
6b1e6cc7 JW |
1748 | |
1749 | return 0; | |
1750 | } | |
1751 | EXPORT_SYMBOL_GPL(vhost_init_device_iotlb); | |
1752 | ||
3a4d5c94 | 1753 | /* Caller must have device mutex */ |
935cdee7 | 1754 | long vhost_dev_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp) |
3a4d5c94 | 1755 | { |
d25cc43c | 1756 | struct eventfd_ctx *ctx; |
3a4d5c94 MT |
1757 | u64 p; |
1758 | long r; | |
1759 | int i, fd; | |
1760 | ||
1761 | /* If you are not the owner, you can become one */ | |
1762 | if (ioctl == VHOST_SET_OWNER) { | |
1763 | r = vhost_dev_set_owner(d); | |
1764 | goto done; | |
1765 | } | |
1766 | ||
1767 | /* You must be the owner to do anything else */ | |
1768 | r = vhost_dev_check_owner(d); | |
1769 | if (r) | |
1770 | goto done; | |
1771 | ||
1772 | switch (ioctl) { | |
1773 | case VHOST_SET_MEM_TABLE: | |
1774 | r = vhost_set_memory(d, argp); | |
1775 | break; | |
1776 | case VHOST_SET_LOG_BASE: | |
7ad9c9d2 TY |
1777 | if (copy_from_user(&p, argp, sizeof p)) { |
1778 | r = -EFAULT; | |
3a4d5c94 | 1779 | break; |
7ad9c9d2 | 1780 | } |
3a4d5c94 MT |
1781 | if ((u64)(unsigned long)p != p) { |
1782 | r = -EFAULT; | |
1783 | break; | |
1784 | } | |
1785 | for (i = 0; i < d->nvqs; ++i) { | |
1786 | struct vhost_virtqueue *vq; | |
1787 | void __user *base = (void __user *)(unsigned long)p; | |
3ab2e420 | 1788 | vq = d->vqs[i]; |
3a4d5c94 MT |
1789 | mutex_lock(&vq->mutex); |
1790 | /* If ring is inactive, will check when it's enabled. */ | |
ea16c514 | 1791 | if (vq->private_data && !vq_log_access_ok(vq, base)) |
3a4d5c94 MT |
1792 | r = -EFAULT; |
1793 | else | |
1794 | vq->log_base = base; | |
1795 | mutex_unlock(&vq->mutex); | |
1796 | } | |
1797 | break; | |
1798 | case VHOST_SET_LOG_FD: | |
1799 | r = get_user(fd, (int __user *)argp); | |
1800 | if (r < 0) | |
1801 | break; | |
e0136c16 | 1802 | ctx = fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(fd); |
d25cc43c EB |
1803 | if (IS_ERR(ctx)) { |
1804 | r = PTR_ERR(ctx); | |
3a4d5c94 MT |
1805 | break; |
1806 | } | |
d25cc43c | 1807 | swap(ctx, d->log_ctx); |
3a4d5c94 | 1808 | for (i = 0; i < d->nvqs; ++i) { |
3ab2e420 AH |
1809 | mutex_lock(&d->vqs[i]->mutex); |
1810 | d->vqs[i]->log_ctx = d->log_ctx; | |
1811 | mutex_unlock(&d->vqs[i]->mutex); | |
3a4d5c94 MT |
1812 | } |
1813 | if (ctx) | |
1814 | eventfd_ctx_put(ctx); | |
3a4d5c94 MT |
1815 | break; |
1816 | default: | |
935cdee7 | 1817 | r = -ENOIOCTLCMD; |
3a4d5c94 MT |
1818 | break; |
1819 | } | |
1820 | done: | |
1821 | return r; | |
1822 | } | |
6ac1afbf | 1823 | EXPORT_SYMBOL_GPL(vhost_dev_ioctl); |
3a4d5c94 | 1824 | |
3a4d5c94 MT |
1825 | /* TODO: This is really inefficient. We need something like get_user() |
1826 | * (instruction directly accesses the data, with an exception table entry | |
ff61f079 | 1827 | * returning -EFAULT). See Documentation/arch/x86/exception-tables.rst. |
3a4d5c94 MT |
1828 | */ |
1829 | static int set_bit_to_user(int nr, void __user *addr) | |
1830 | { | |
1831 | unsigned long log = (unsigned long)addr; | |
1832 | struct page *page; | |
1833 | void *base; | |
1834 | int bit = nr + (log % PAGE_SIZE) * 8; | |
1835 | int r; | |
d47effe1 | 1836 | |
690623e1 | 1837 | r = pin_user_pages_fast(log, 1, FOLL_WRITE, &page); |
d6db3f5c | 1838 | if (r < 0) |
3a4d5c94 | 1839 | return r; |
d6db3f5c | 1840 | BUG_ON(r != 1); |
c6daa7ff | 1841 | base = kmap_atomic(page); |
3a4d5c94 | 1842 | set_bit(bit, base); |
c6daa7ff | 1843 | kunmap_atomic(base); |
690623e1 | 1844 | unpin_user_pages_dirty_lock(&page, 1, true); |
3a4d5c94 MT |
1845 | return 0; |
1846 | } | |
1847 | ||
1848 | static int log_write(void __user *log_base, | |
1849 | u64 write_address, u64 write_length) | |
1850 | { | |
28831ee6 | 1851 | u64 write_page = write_address / VHOST_PAGE_SIZE; |
3a4d5c94 | 1852 | int r; |
d47effe1 | 1853 | |
3a4d5c94 MT |
1854 | if (!write_length) |
1855 | return 0; | |
3bf9be40 | 1856 | write_length += write_address % VHOST_PAGE_SIZE; |
3a4d5c94 MT |
1857 | for (;;) { |
1858 | u64 base = (u64)(unsigned long)log_base; | |
28831ee6 MT |
1859 | u64 log = base + write_page / 8; |
1860 | int bit = write_page % 8; | |
3a4d5c94 MT |
1861 | if ((u64)(unsigned long)log != log) |
1862 | return -EFAULT; | |
1863 | r = set_bit_to_user(bit, (void __user *)(unsigned long)log); | |
1864 | if (r < 0) | |
1865 | return r; | |
1866 | if (write_length <= VHOST_PAGE_SIZE) | |
1867 | break; | |
1868 | write_length -= VHOST_PAGE_SIZE; | |
28831ee6 | 1869 | write_page += 1; |
3a4d5c94 MT |
1870 | } |
1871 | return r; | |
1872 | } | |
1873 | ||
cc5e7107 JW |
1874 | static int log_write_hva(struct vhost_virtqueue *vq, u64 hva, u64 len) |
1875 | { | |
0bbe3066 JW |
1876 | struct vhost_iotlb *umem = vq->umem; |
1877 | struct vhost_iotlb_map *u; | |
cc5e7107 JW |
1878 | u64 start, end, l, min; |
1879 | int r; | |
1880 | bool hit = false; | |
1881 | ||
1882 | while (len) { | |
1883 | min = len; | |
1884 | /* More than one GPAs can be mapped into a single HVA. So | |
1885 | * iterate all possible umems here to be safe. | |
1886 | */ | |
0bbe3066 JW |
1887 | list_for_each_entry(u, &umem->list, link) { |
1888 | if (u->addr > hva - 1 + len || | |
1889 | u->addr - 1 + u->size < hva) | |
cc5e7107 | 1890 | continue; |
0bbe3066 JW |
1891 | start = max(u->addr, hva); |
1892 | end = min(u->addr - 1 + u->size, hva - 1 + len); | |
cc5e7107 JW |
1893 | l = end - start + 1; |
1894 | r = log_write(vq->log_base, | |
0bbe3066 | 1895 | u->start + start - u->addr, |
cc5e7107 JW |
1896 | l); |
1897 | if (r < 0) | |
1898 | return r; | |
1899 | hit = true; | |
1900 | min = min(l, min); | |
1901 | } | |
1902 | ||
1903 | if (!hit) | |
1904 | return -EFAULT; | |
1905 | ||
1906 | len -= min; | |
1907 | hva += min; | |
1908 | } | |
1909 | ||
1910 | return 0; | |
1911 | } | |
1912 | ||
1913 | static int log_used(struct vhost_virtqueue *vq, u64 used_offset, u64 len) | |
1914 | { | |
5e5e8736 | 1915 | struct iovec *iov = vq->log_iov; |
cc5e7107 JW |
1916 | int i, ret; |
1917 | ||
1918 | if (!vq->iotlb) | |
1919 | return log_write(vq->log_base, vq->log_addr + used_offset, len); | |
1920 | ||
1921 | ret = translate_desc(vq, (uintptr_t)vq->used + used_offset, | |
1922 | len, iov, 64, VHOST_ACCESS_WO); | |
816db766 | 1923 | if (ret < 0) |
cc5e7107 JW |
1924 | return ret; |
1925 | ||
1926 | for (i = 0; i < ret; i++) { | |
1927 | ret = log_write_hva(vq, (uintptr_t)iov[i].iov_base, | |
1928 | iov[i].iov_len); | |
1929 | if (ret) | |
1930 | return ret; | |
1931 | } | |
1932 | ||
1933 | return 0; | |
1934 | } | |
1935 | ||
3a4d5c94 | 1936 | int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log, |
cc5e7107 | 1937 | unsigned int log_num, u64 len, struct iovec *iov, int count) |
3a4d5c94 MT |
1938 | { |
1939 | int i, r; | |
1940 | ||
1941 | /* Make sure data written is seen before log. */ | |
5659338c | 1942 | smp_wmb(); |
cc5e7107 JW |
1943 | |
1944 | if (vq->iotlb) { | |
1945 | for (i = 0; i < count; i++) { | |
1946 | r = log_write_hva(vq, (uintptr_t)iov[i].iov_base, | |
1947 | iov[i].iov_len); | |
1948 | if (r < 0) | |
1949 | return r; | |
1950 | } | |
1951 | return 0; | |
1952 | } | |
1953 | ||
3a4d5c94 MT |
1954 | for (i = 0; i < log_num; ++i) { |
1955 | u64 l = min(log[i].len, len); | |
1956 | r = log_write(vq->log_base, log[i].addr, l); | |
1957 | if (r < 0) | |
1958 | return r; | |
1959 | len -= l; | |
5786aee8 MT |
1960 | if (!len) { |
1961 | if (vq->log_ctx) | |
1962 | eventfd_signal(vq->log_ctx, 1); | |
3a4d5c94 | 1963 | return 0; |
5786aee8 | 1964 | } |
3a4d5c94 | 1965 | } |
3a4d5c94 MT |
1966 | /* Length written exceeds what we have stored. This is a bug. */ |
1967 | BUG(); | |
1968 | return 0; | |
1969 | } | |
6ac1afbf | 1970 | EXPORT_SYMBOL_GPL(vhost_log_write); |
3a4d5c94 | 1971 | |
2723feaa JW |
1972 | static int vhost_update_used_flags(struct vhost_virtqueue *vq) |
1973 | { | |
1974 | void __user *used; | |
7b5d753e | 1975 | if (vhost_put_used_flags(vq)) |
2723feaa JW |
1976 | return -EFAULT; |
1977 | if (unlikely(vq->log_used)) { | |
1978 | /* Make sure the flag is seen before log. */ | |
1979 | smp_wmb(); | |
1980 | /* Log used flag write. */ | |
1981 | used = &vq->used->flags; | |
cc5e7107 JW |
1982 | log_used(vq, (used - (void __user *)vq->used), |
1983 | sizeof vq->used->flags); | |
2723feaa JW |
1984 | if (vq->log_ctx) |
1985 | eventfd_signal(vq->log_ctx, 1); | |
1986 | } | |
1987 | return 0; | |
1988 | } | |
1989 | ||
4c809363 | 1990 | static int vhost_update_avail_event(struct vhost_virtqueue *vq) |
2723feaa | 1991 | { |
7b5d753e | 1992 | if (vhost_put_avail_event(vq)) |
2723feaa JW |
1993 | return -EFAULT; |
1994 | if (unlikely(vq->log_used)) { | |
1995 | void __user *used; | |
1996 | /* Make sure the event is seen before log. */ | |
1997 | smp_wmb(); | |
1998 | /* Log avail event write */ | |
1999 | used = vhost_avail_event(vq); | |
cc5e7107 JW |
2000 | log_used(vq, (used - (void __user *)vq->used), |
2001 | sizeof *vhost_avail_event(vq)); | |
2723feaa JW |
2002 | if (vq->log_ctx) |
2003 | eventfd_signal(vq->log_ctx, 1); | |
2004 | } | |
2005 | return 0; | |
2006 | } | |
2007 | ||
80f7d030 | 2008 | int vhost_vq_init_access(struct vhost_virtqueue *vq) |
2723feaa | 2009 | { |
3b1bbe89 | 2010 | __virtio16 last_used_idx; |
2723feaa | 2011 | int r; |
e1f33be9 GK |
2012 | bool is_le = vq->is_le; |
2013 | ||
cda8bba0 | 2014 | if (!vq->private_data) |
2723feaa | 2015 | return 0; |
2751c988 GK |
2016 | |
2017 | vhost_init_is_le(vq); | |
2723feaa JW |
2018 | |
2019 | r = vhost_update_used_flags(vq); | |
2020 | if (r) | |
e1f33be9 | 2021 | goto err; |
2723feaa | 2022 | vq->signalled_used_valid = false; |
6b1e6cc7 | 2023 | if (!vq->iotlb && |
96d4f267 | 2024 | !access_ok(&vq->used->idx, sizeof vq->used->idx)) { |
e1f33be9 GK |
2025 | r = -EFAULT; |
2026 | goto err; | |
2027 | } | |
7b5d753e | 2028 | r = vhost_get_used_idx(vq, &last_used_idx); |
6b1e6cc7 JW |
2029 | if (r) { |
2030 | vq_err(vq, "Can't access used idx at %p\n", | |
2031 | &vq->used->idx); | |
e1f33be9 | 2032 | goto err; |
6b1e6cc7 | 2033 | } |
3b1bbe89 | 2034 | vq->last_used_idx = vhost16_to_cpu(vq, last_used_idx); |
64f7f051 | 2035 | return 0; |
6b1e6cc7 | 2036 | |
e1f33be9 GK |
2037 | err: |
2038 | vq->is_le = is_le; | |
2039 | return r; | |
2723feaa | 2040 | } |
80f7d030 | 2041 | EXPORT_SYMBOL_GPL(vhost_vq_init_access); |
2723feaa | 2042 | |
47283bef | 2043 | static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len, |
6b1e6cc7 | 2044 | struct iovec iov[], int iov_size, int access) |
3a4d5c94 | 2045 | { |
0bbe3066 | 2046 | const struct vhost_iotlb_map *map; |
6b1e6cc7 | 2047 | struct vhost_dev *dev = vq->dev; |
0bbe3066 | 2048 | struct vhost_iotlb *umem = dev->iotlb ? dev->iotlb : dev->umem; |
3a4d5c94 | 2049 | struct iovec *_iov; |
98047313 | 2050 | u64 s = 0, last = addr + len - 1; |
3a4d5c94 MT |
2051 | int ret = 0; |
2052 | ||
3a4d5c94 MT |
2053 | while ((u64)len > s) { |
2054 | u64 size; | |
7b3384fc | 2055 | if (unlikely(ret >= iov_size)) { |
3a4d5c94 MT |
2056 | ret = -ENOBUFS; |
2057 | break; | |
2058 | } | |
6b1e6cc7 | 2059 | |
98047313 | 2060 | map = vhost_iotlb_itree_first(umem, addr, last); |
0bbe3066 | 2061 | if (map == NULL || map->start > addr) { |
6b1e6cc7 JW |
2062 | if (umem != dev->iotlb) { |
2063 | ret = -EFAULT; | |
2064 | break; | |
2065 | } | |
2066 | ret = -EAGAIN; | |
2067 | break; | |
0bbe3066 | 2068 | } else if (!(map->perm & access)) { |
6b1e6cc7 | 2069 | ret = -EPERM; |
3a4d5c94 MT |
2070 | break; |
2071 | } | |
6b1e6cc7 | 2072 | |
3a4d5c94 | 2073 | _iov = iov + ret; |
0bbe3066 | 2074 | size = map->size - addr + map->start; |
bd97120f | 2075 | _iov->iov_len = min((u64)len - s, size); |
0d4a3f2a | 2076 | _iov->iov_base = (void __user *)(unsigned long) |
0bbe3066 | 2077 | (map->addr + addr - map->start); |
3a4d5c94 MT |
2078 | s += size; |
2079 | addr += size; | |
2080 | ++ret; | |
2081 | } | |
2082 | ||
6b1e6cc7 JW |
2083 | if (ret == -EAGAIN) |
2084 | vhost_iotlb_miss(vq, addr, access); | |
3a4d5c94 MT |
2085 | return ret; |
2086 | } | |
2087 | ||
2088 | /* Each buffer in the virtqueues is actually a chain of descriptors. This | |
2089 | * function returns the next descriptor in the chain, | |
2090 | * or -1U if we're at the end. */ | |
3b1bbe89 | 2091 | static unsigned next_desc(struct vhost_virtqueue *vq, struct vring_desc *desc) |
3a4d5c94 MT |
2092 | { |
2093 | unsigned int next; | |
2094 | ||
2095 | /* If this descriptor says it doesn't chain, we're done. */ | |
3b1bbe89 | 2096 | if (!(desc->flags & cpu_to_vhost16(vq, VRING_DESC_F_NEXT))) |
3a4d5c94 MT |
2097 | return -1U; |
2098 | ||
2099 | /* Check they're not leading us off end of descriptors. */ | |
3a5db0b1 | 2100 | next = vhost16_to_cpu(vq, READ_ONCE(desc->next)); |
3a4d5c94 MT |
2101 | return next; |
2102 | } | |
2103 | ||
47283bef | 2104 | static int get_indirect(struct vhost_virtqueue *vq, |
7b3384fc MT |
2105 | struct iovec iov[], unsigned int iov_size, |
2106 | unsigned int *out_num, unsigned int *in_num, | |
2107 | struct vhost_log *log, unsigned int *log_num, | |
2108 | struct vring_desc *indirect) | |
3a4d5c94 MT |
2109 | { |
2110 | struct vring_desc desc; | |
2111 | unsigned int i = 0, count, found = 0; | |
3b1bbe89 | 2112 | u32 len = vhost32_to_cpu(vq, indirect->len); |
aad9a1ce | 2113 | struct iov_iter from; |
6b1e6cc7 | 2114 | int ret, access; |
3a4d5c94 MT |
2115 | |
2116 | /* Sanity check */ | |
3b1bbe89 | 2117 | if (unlikely(len % sizeof desc)) { |
3a4d5c94 MT |
2118 | vq_err(vq, "Invalid length in indirect descriptor: " |
2119 | "len 0x%llx not multiple of 0x%zx\n", | |
3b1bbe89 | 2120 | (unsigned long long)len, |
3a4d5c94 MT |
2121 | sizeof desc); |
2122 | return -EINVAL; | |
2123 | } | |
2124 | ||
3b1bbe89 | 2125 | ret = translate_desc(vq, vhost64_to_cpu(vq, indirect->addr), len, vq->indirect, |
6b1e6cc7 | 2126 | UIO_MAXIOV, VHOST_ACCESS_RO); |
7b3384fc | 2127 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
2128 | if (ret != -EAGAIN) |
2129 | vq_err(vq, "Translation failure %d in indirect.\n", ret); | |
3a4d5c94 MT |
2130 | return ret; |
2131 | } | |
de4eda9d | 2132 | iov_iter_init(&from, ITER_SOURCE, vq->indirect, ret, len); |
3b1bbe89 | 2133 | count = len / sizeof desc; |
3a4d5c94 MT |
2134 | /* Buffers are chained via a 16 bit next field, so |
2135 | * we can have at most 2^16 of these. */ | |
7b3384fc | 2136 | if (unlikely(count > USHRT_MAX + 1)) { |
3a4d5c94 MT |
2137 | vq_err(vq, "Indirect buffer length too big: %d\n", |
2138 | indirect->len); | |
2139 | return -E2BIG; | |
2140 | } | |
2141 | ||
2142 | do { | |
2143 | unsigned iov_count = *in_num + *out_num; | |
7b3384fc | 2144 | if (unlikely(++found > count)) { |
3a4d5c94 MT |
2145 | vq_err(vq, "Loop detected: last one at %u " |
2146 | "indirect size %u\n", | |
2147 | i, count); | |
2148 | return -EINVAL; | |
2149 | } | |
cbbd26b8 | 2150 | if (unlikely(!copy_from_iter_full(&desc, sizeof(desc), &from))) { |
3a4d5c94 | 2151 | vq_err(vq, "Failed indirect descriptor: idx %d, %zx\n", |
3b1bbe89 | 2152 | i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc); |
3a4d5c94 MT |
2153 | return -EINVAL; |
2154 | } | |
3b1bbe89 | 2155 | if (unlikely(desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT))) { |
3a4d5c94 | 2156 | vq_err(vq, "Nested indirect descriptor: idx %d, %zx\n", |
3b1bbe89 | 2157 | i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc); |
3a4d5c94 MT |
2158 | return -EINVAL; |
2159 | } | |
2160 | ||
6b1e6cc7 JW |
2161 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE)) |
2162 | access = VHOST_ACCESS_WO; | |
2163 | else | |
2164 | access = VHOST_ACCESS_RO; | |
2165 | ||
3b1bbe89 MT |
2166 | ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr), |
2167 | vhost32_to_cpu(vq, desc.len), iov + iov_count, | |
6b1e6cc7 | 2168 | iov_size - iov_count, access); |
7b3384fc | 2169 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
2170 | if (ret != -EAGAIN) |
2171 | vq_err(vq, "Translation failure %d indirect idx %d\n", | |
2172 | ret, i); | |
3a4d5c94 MT |
2173 | return ret; |
2174 | } | |
2175 | /* If this is an input descriptor, increment that count. */ | |
6b1e6cc7 | 2176 | if (access == VHOST_ACCESS_WO) { |
3a4d5c94 | 2177 | *in_num += ret; |
060423bf | 2178 | if (unlikely(log && ret)) { |
3b1bbe89 MT |
2179 | log[*log_num].addr = vhost64_to_cpu(vq, desc.addr); |
2180 | log[*log_num].len = vhost32_to_cpu(vq, desc.len); | |
3a4d5c94 MT |
2181 | ++*log_num; |
2182 | } | |
2183 | } else { | |
2184 | /* If it's an output descriptor, they're all supposed | |
2185 | * to come before any input descriptors. */ | |
7b3384fc | 2186 | if (unlikely(*in_num)) { |
3a4d5c94 MT |
2187 | vq_err(vq, "Indirect descriptor " |
2188 | "has out after in: idx %d\n", i); | |
2189 | return -EINVAL; | |
2190 | } | |
2191 | *out_num += ret; | |
2192 | } | |
3b1bbe89 | 2193 | } while ((i = next_desc(vq, &desc)) != -1); |
3a4d5c94 MT |
2194 | return 0; |
2195 | } | |
2196 | ||
2197 | /* This looks in the virtqueue and for the first available buffer, and converts | |
2198 | * it to an iovec for convenient access. Since descriptors consist of some | |
2199 | * number of output then some number of input descriptors, it's actually two | |
2200 | * iovecs, but we pack them into one and note how many of each there were. | |
2201 | * | |
d5675bd2 MT |
2202 | * This function returns the descriptor number found, or vq->num (which is |
2203 | * never a valid descriptor number) if none was found. A negative code is | |
2204 | * returned on error. */ | |
47283bef | 2205 | int vhost_get_vq_desc(struct vhost_virtqueue *vq, |
d5675bd2 MT |
2206 | struct iovec iov[], unsigned int iov_size, |
2207 | unsigned int *out_num, unsigned int *in_num, | |
2208 | struct vhost_log *log, unsigned int *log_num) | |
3a4d5c94 MT |
2209 | { |
2210 | struct vring_desc desc; | |
2211 | unsigned int i, head, found = 0; | |
2212 | u16 last_avail_idx; | |
3b1bbe89 MT |
2213 | __virtio16 avail_idx; |
2214 | __virtio16 ring_head; | |
6b1e6cc7 | 2215 | int ret, access; |
3a4d5c94 MT |
2216 | |
2217 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
2218 | last_avail_idx = vq->last_avail_idx; | |
3a4d5c94 | 2219 | |
e3b56cdd | 2220 | if (vq->avail_idx == vq->last_avail_idx) { |
7b5d753e | 2221 | if (unlikely(vhost_get_avail_idx(vq, &avail_idx))) { |
e3b56cdd JW |
2222 | vq_err(vq, "Failed to access avail idx at %p\n", |
2223 | &vq->avail->idx); | |
2224 | return -EFAULT; | |
2225 | } | |
2226 | vq->avail_idx = vhost16_to_cpu(vq, avail_idx); | |
3a4d5c94 | 2227 | |
e3b56cdd JW |
2228 | if (unlikely((u16)(vq->avail_idx - last_avail_idx) > vq->num)) { |
2229 | vq_err(vq, "Guest moved used index from %u to %u", | |
2230 | last_avail_idx, vq->avail_idx); | |
2231 | return -EFAULT; | |
2232 | } | |
2233 | ||
2234 | /* If there's nothing new since last we looked, return | |
2235 | * invalid. | |
2236 | */ | |
2237 | if (vq->avail_idx == last_avail_idx) | |
2238 | return vq->num; | |
3a4d5c94 | 2239 | |
e3b56cdd JW |
2240 | /* Only get avail ring entries after they have been |
2241 | * exposed by guest. | |
2242 | */ | |
2243 | smp_rmb(); | |
2244 | } | |
3a4d5c94 MT |
2245 | |
2246 | /* Grab the next descriptor number they're advertising, and increment | |
2247 | * the index we've seen. */ | |
7b5d753e | 2248 | if (unlikely(vhost_get_avail_head(vq, &ring_head, last_avail_idx))) { |
3a4d5c94 MT |
2249 | vq_err(vq, "Failed to read head: idx %d address %p\n", |
2250 | last_avail_idx, | |
2251 | &vq->avail->ring[last_avail_idx % vq->num]); | |
d5675bd2 | 2252 | return -EFAULT; |
3a4d5c94 MT |
2253 | } |
2254 | ||
3b1bbe89 MT |
2255 | head = vhost16_to_cpu(vq, ring_head); |
2256 | ||
3a4d5c94 | 2257 | /* If their number is silly, that's an error. */ |
7b3384fc | 2258 | if (unlikely(head >= vq->num)) { |
3a4d5c94 MT |
2259 | vq_err(vq, "Guest says index %u > %u is available", |
2260 | head, vq->num); | |
d5675bd2 | 2261 | return -EINVAL; |
3a4d5c94 MT |
2262 | } |
2263 | ||
2264 | /* When we start there are none of either input nor output. */ | |
2265 | *out_num = *in_num = 0; | |
2266 | if (unlikely(log)) | |
2267 | *log_num = 0; | |
2268 | ||
2269 | i = head; | |
2270 | do { | |
2271 | unsigned iov_count = *in_num + *out_num; | |
7b3384fc | 2272 | if (unlikely(i >= vq->num)) { |
3a4d5c94 MT |
2273 | vq_err(vq, "Desc index is %u > %u, head = %u", |
2274 | i, vq->num, head); | |
d5675bd2 | 2275 | return -EINVAL; |
3a4d5c94 | 2276 | } |
7b3384fc | 2277 | if (unlikely(++found > vq->num)) { |
3a4d5c94 MT |
2278 | vq_err(vq, "Loop detected: last one at %u " |
2279 | "vq size %u head %u\n", | |
2280 | i, vq->num, head); | |
d5675bd2 | 2281 | return -EINVAL; |
3a4d5c94 | 2282 | } |
7b5d753e | 2283 | ret = vhost_get_desc(vq, &desc, i); |
7b3384fc | 2284 | if (unlikely(ret)) { |
3a4d5c94 MT |
2285 | vq_err(vq, "Failed to get descriptor: idx %d addr %p\n", |
2286 | i, vq->desc + i); | |
d5675bd2 | 2287 | return -EFAULT; |
3a4d5c94 | 2288 | } |
3b1bbe89 | 2289 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT)) { |
47283bef | 2290 | ret = get_indirect(vq, iov, iov_size, |
3a4d5c94 MT |
2291 | out_num, in_num, |
2292 | log, log_num, &desc); | |
7b3384fc | 2293 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
2294 | if (ret != -EAGAIN) |
2295 | vq_err(vq, "Failure detected " | |
2296 | "in indirect descriptor at idx %d\n", i); | |
d5675bd2 | 2297 | return ret; |
3a4d5c94 MT |
2298 | } |
2299 | continue; | |
2300 | } | |
2301 | ||
6b1e6cc7 JW |
2302 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE)) |
2303 | access = VHOST_ACCESS_WO; | |
2304 | else | |
2305 | access = VHOST_ACCESS_RO; | |
3b1bbe89 MT |
2306 | ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr), |
2307 | vhost32_to_cpu(vq, desc.len), iov + iov_count, | |
6b1e6cc7 | 2308 | iov_size - iov_count, access); |
7b3384fc | 2309 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
2310 | if (ret != -EAGAIN) |
2311 | vq_err(vq, "Translation failure %d descriptor idx %d\n", | |
2312 | ret, i); | |
d5675bd2 | 2313 | return ret; |
3a4d5c94 | 2314 | } |
6b1e6cc7 | 2315 | if (access == VHOST_ACCESS_WO) { |
3a4d5c94 MT |
2316 | /* If this is an input descriptor, |
2317 | * increment that count. */ | |
2318 | *in_num += ret; | |
060423bf | 2319 | if (unlikely(log && ret)) { |
3b1bbe89 MT |
2320 | log[*log_num].addr = vhost64_to_cpu(vq, desc.addr); |
2321 | log[*log_num].len = vhost32_to_cpu(vq, desc.len); | |
3a4d5c94 MT |
2322 | ++*log_num; |
2323 | } | |
2324 | } else { | |
2325 | /* If it's an output descriptor, they're all supposed | |
2326 | * to come before any input descriptors. */ | |
7b3384fc | 2327 | if (unlikely(*in_num)) { |
3a4d5c94 MT |
2328 | vq_err(vq, "Descriptor has out after in: " |
2329 | "idx %d\n", i); | |
d5675bd2 | 2330 | return -EINVAL; |
3a4d5c94 MT |
2331 | } |
2332 | *out_num += ret; | |
2333 | } | |
3b1bbe89 | 2334 | } while ((i = next_desc(vq, &desc)) != -1); |
3a4d5c94 MT |
2335 | |
2336 | /* On success, increment avail index. */ | |
2337 | vq->last_avail_idx++; | |
8ea8cf89 MT |
2338 | |
2339 | /* Assume notifications from guest are disabled at this point, | |
2340 | * if they aren't we would need to update avail_event index. */ | |
2341 | BUG_ON(!(vq->used_flags & VRING_USED_F_NO_NOTIFY)); | |
3a4d5c94 MT |
2342 | return head; |
2343 | } | |
6ac1afbf | 2344 | EXPORT_SYMBOL_GPL(vhost_get_vq_desc); |
3a4d5c94 MT |
2345 | |
2346 | /* Reverse the effect of vhost_get_vq_desc. Useful for error handling. */ | |
8dd014ad | 2347 | void vhost_discard_vq_desc(struct vhost_virtqueue *vq, int n) |
3a4d5c94 | 2348 | { |
8dd014ad | 2349 | vq->last_avail_idx -= n; |
3a4d5c94 | 2350 | } |
6ac1afbf | 2351 | EXPORT_SYMBOL_GPL(vhost_discard_vq_desc); |
3a4d5c94 MT |
2352 | |
2353 | /* After we've used one of their buffers, we tell them about it. We'll then | |
2354 | * want to notify the guest, using eventfd. */ | |
2355 | int vhost_add_used(struct vhost_virtqueue *vq, unsigned int head, int len) | |
2356 | { | |
3b1bbe89 MT |
2357 | struct vring_used_elem heads = { |
2358 | cpu_to_vhost32(vq, head), | |
2359 | cpu_to_vhost32(vq, len) | |
2360 | }; | |
3a4d5c94 | 2361 | |
c49e4e57 | 2362 | return vhost_add_used_n(vq, &heads, 1); |
3a4d5c94 | 2363 | } |
6ac1afbf | 2364 | EXPORT_SYMBOL_GPL(vhost_add_used); |
3a4d5c94 | 2365 | |
8dd014ad DS |
2366 | static int __vhost_add_used_n(struct vhost_virtqueue *vq, |
2367 | struct vring_used_elem *heads, | |
2368 | unsigned count) | |
2369 | { | |
a865e420 | 2370 | vring_used_elem_t __user *used; |
8ea8cf89 | 2371 | u16 old, new; |
8dd014ad DS |
2372 | int start; |
2373 | ||
5fba13b5 | 2374 | start = vq->last_used_idx & (vq->num - 1); |
8dd014ad | 2375 | used = vq->used->ring + start; |
7b5d753e | 2376 | if (vhost_put_used(vq, heads, start, count)) { |
8dd014ad DS |
2377 | vq_err(vq, "Failed to write used"); |
2378 | return -EFAULT; | |
2379 | } | |
2380 | if (unlikely(vq->log_used)) { | |
2381 | /* Make sure data is seen before log. */ | |
2382 | smp_wmb(); | |
2383 | /* Log used ring entry write. */ | |
cc5e7107 JW |
2384 | log_used(vq, ((void __user *)used - (void __user *)vq->used), |
2385 | count * sizeof *used); | |
8dd014ad | 2386 | } |
8ea8cf89 MT |
2387 | old = vq->last_used_idx; |
2388 | new = (vq->last_used_idx += count); | |
2389 | /* If the driver never bothers to signal in a very long while, | |
2390 | * used index might wrap around. If that happens, invalidate | |
2391 | * signalled_used index we stored. TODO: make sure driver | |
2392 | * signals at least once in 2^16 and remove this. */ | |
2393 | if (unlikely((u16)(new - vq->signalled_used) < (u16)(new - old))) | |
2394 | vq->signalled_used_valid = false; | |
8dd014ad DS |
2395 | return 0; |
2396 | } | |
2397 | ||
2398 | /* After we've used one of their buffers, we tell them about it. We'll then | |
2399 | * want to notify the guest, using eventfd. */ | |
2400 | int vhost_add_used_n(struct vhost_virtqueue *vq, struct vring_used_elem *heads, | |
2401 | unsigned count) | |
2402 | { | |
2403 | int start, n, r; | |
2404 | ||
5fba13b5 | 2405 | start = vq->last_used_idx & (vq->num - 1); |
8dd014ad DS |
2406 | n = vq->num - start; |
2407 | if (n < count) { | |
2408 | r = __vhost_add_used_n(vq, heads, n); | |
2409 | if (r < 0) | |
2410 | return r; | |
2411 | heads += n; | |
2412 | count -= n; | |
2413 | } | |
2414 | r = __vhost_add_used_n(vq, heads, count); | |
2415 | ||
2416 | /* Make sure buffer is written before we update index. */ | |
2417 | smp_wmb(); | |
7b5d753e | 2418 | if (vhost_put_used_idx(vq)) { |
8dd014ad DS |
2419 | vq_err(vq, "Failed to increment used idx"); |
2420 | return -EFAULT; | |
2421 | } | |
2422 | if (unlikely(vq->log_used)) { | |
841df922 JW |
2423 | /* Make sure used idx is seen before log. */ |
2424 | smp_wmb(); | |
8dd014ad | 2425 | /* Log used index update. */ |
cc5e7107 JW |
2426 | log_used(vq, offsetof(struct vring_used, idx), |
2427 | sizeof vq->used->idx); | |
8dd014ad DS |
2428 | if (vq->log_ctx) |
2429 | eventfd_signal(vq->log_ctx, 1); | |
2430 | } | |
2431 | return r; | |
2432 | } | |
6ac1afbf | 2433 | EXPORT_SYMBOL_GPL(vhost_add_used_n); |
8dd014ad | 2434 | |
8ea8cf89 | 2435 | static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 | 2436 | { |
3b1bbe89 MT |
2437 | __u16 old, new; |
2438 | __virtio16 event; | |
8ea8cf89 | 2439 | bool v; |
8d65843c JW |
2440 | /* Flush out used index updates. This is paired |
2441 | * with the barrier that the Guest executes when enabling | |
2442 | * interrupts. */ | |
2443 | smp_mb(); | |
0d499356 | 2444 | |
ea16c514 | 2445 | if (vhost_has_feature(vq, VIRTIO_F_NOTIFY_ON_EMPTY) && |
8ea8cf89 MT |
2446 | unlikely(vq->avail_idx == vq->last_avail_idx)) |
2447 | return true; | |
2448 | ||
ea16c514 | 2449 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
3b1bbe89 | 2450 | __virtio16 flags; |
7b5d753e | 2451 | if (vhost_get_avail_flags(vq, &flags)) { |
8ea8cf89 MT |
2452 | vq_err(vq, "Failed to get flags"); |
2453 | return true; | |
2454 | } | |
3b1bbe89 | 2455 | return !(flags & cpu_to_vhost16(vq, VRING_AVAIL_F_NO_INTERRUPT)); |
3a4d5c94 | 2456 | } |
8ea8cf89 MT |
2457 | old = vq->signalled_used; |
2458 | v = vq->signalled_used_valid; | |
2459 | new = vq->signalled_used = vq->last_used_idx; | |
2460 | vq->signalled_used_valid = true; | |
3a4d5c94 | 2461 | |
8ea8cf89 MT |
2462 | if (unlikely(!v)) |
2463 | return true; | |
3a4d5c94 | 2464 | |
7b5d753e | 2465 | if (vhost_get_used_event(vq, &event)) { |
8ea8cf89 MT |
2466 | vq_err(vq, "Failed to get used event idx"); |
2467 | return true; | |
2468 | } | |
8d65843c | 2469 | return vring_need_event(vhost16_to_cpu(vq, event), new, old); |
8ea8cf89 MT |
2470 | } |
2471 | ||
2472 | /* This actually signals the guest, using eventfd. */ | |
2473 | void vhost_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq) | |
2474 | { | |
3a4d5c94 | 2475 | /* Signal the Guest tell them we used something up. */ |
265a0ad8 ZL |
2476 | if (vq->call_ctx.ctx && vhost_notify(dev, vq)) |
2477 | eventfd_signal(vq->call_ctx.ctx, 1); | |
3a4d5c94 | 2478 | } |
6ac1afbf | 2479 | EXPORT_SYMBOL_GPL(vhost_signal); |
3a4d5c94 MT |
2480 | |
2481 | /* And here's the combo meal deal. Supersize me! */ | |
2482 | void vhost_add_used_and_signal(struct vhost_dev *dev, | |
2483 | struct vhost_virtqueue *vq, | |
2484 | unsigned int head, int len) | |
2485 | { | |
2486 | vhost_add_used(vq, head, len); | |
2487 | vhost_signal(dev, vq); | |
2488 | } | |
6ac1afbf | 2489 | EXPORT_SYMBOL_GPL(vhost_add_used_and_signal); |
3a4d5c94 | 2490 | |
8dd014ad DS |
2491 | /* multi-buffer version of vhost_add_used_and_signal */ |
2492 | void vhost_add_used_and_signal_n(struct vhost_dev *dev, | |
2493 | struct vhost_virtqueue *vq, | |
2494 | struct vring_used_elem *heads, unsigned count) | |
2495 | { | |
2496 | vhost_add_used_n(vq, heads, count); | |
2497 | vhost_signal(dev, vq); | |
2498 | } | |
6ac1afbf | 2499 | EXPORT_SYMBOL_GPL(vhost_add_used_and_signal_n); |
8dd014ad | 2500 | |
d4a60603 JW |
2501 | /* return true if we're sure that avaiable ring is empty */ |
2502 | bool vhost_vq_avail_empty(struct vhost_dev *dev, struct vhost_virtqueue *vq) | |
2503 | { | |
2504 | __virtio16 avail_idx; | |
2505 | int r; | |
2506 | ||
275bf960 JW |
2507 | if (vq->avail_idx != vq->last_avail_idx) |
2508 | return false; | |
2509 | ||
7b5d753e | 2510 | r = vhost_get_avail_idx(vq, &avail_idx); |
275bf960 | 2511 | if (unlikely(r)) |
d4a60603 | 2512 | return false; |
275bf960 | 2513 | vq->avail_idx = vhost16_to_cpu(vq, avail_idx); |
d4a60603 | 2514 | |
275bf960 | 2515 | return vq->avail_idx == vq->last_avail_idx; |
d4a60603 JW |
2516 | } |
2517 | EXPORT_SYMBOL_GPL(vhost_vq_avail_empty); | |
2518 | ||
3a4d5c94 | 2519 | /* OK, now we need to know about added descriptors. */ |
8ea8cf89 | 2520 | bool vhost_enable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 | 2521 | { |
3b1bbe89 | 2522 | __virtio16 avail_idx; |
3a4d5c94 | 2523 | int r; |
d47effe1 | 2524 | |
3a4d5c94 MT |
2525 | if (!(vq->used_flags & VRING_USED_F_NO_NOTIFY)) |
2526 | return false; | |
2527 | vq->used_flags &= ~VRING_USED_F_NO_NOTIFY; | |
ea16c514 | 2528 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
2723feaa | 2529 | r = vhost_update_used_flags(vq); |
8ea8cf89 MT |
2530 | if (r) { |
2531 | vq_err(vq, "Failed to enable notification at %p: %d\n", | |
2532 | &vq->used->flags, r); | |
2533 | return false; | |
2534 | } | |
2535 | } else { | |
4c809363 | 2536 | r = vhost_update_avail_event(vq); |
8ea8cf89 MT |
2537 | if (r) { |
2538 | vq_err(vq, "Failed to update avail event index at %p: %d\n", | |
2539 | vhost_avail_event(vq), r); | |
2540 | return false; | |
2541 | } | |
2542 | } | |
3a4d5c94 MT |
2543 | /* They could have slipped one in as we were doing that: make |
2544 | * sure it's written, then check again. */ | |
5659338c | 2545 | smp_mb(); |
7b5d753e | 2546 | r = vhost_get_avail_idx(vq, &avail_idx); |
3a4d5c94 MT |
2547 | if (r) { |
2548 | vq_err(vq, "Failed to check avail idx at %p: %d\n", | |
2549 | &vq->avail->idx, r); | |
2550 | return false; | |
2551 | } | |
d3bb267b | 2552 | vq->avail_idx = vhost16_to_cpu(vq, avail_idx); |
3a4d5c94 | 2553 | |
d3bb267b | 2554 | return vq->avail_idx != vq->last_avail_idx; |
3a4d5c94 | 2555 | } |
6ac1afbf | 2556 | EXPORT_SYMBOL_GPL(vhost_enable_notify); |
3a4d5c94 MT |
2557 | |
2558 | /* We don't need to be notified again. */ | |
8ea8cf89 | 2559 | void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 MT |
2560 | { |
2561 | int r; | |
d47effe1 | 2562 | |
3a4d5c94 MT |
2563 | if (vq->used_flags & VRING_USED_F_NO_NOTIFY) |
2564 | return; | |
2565 | vq->used_flags |= VRING_USED_F_NO_NOTIFY; | |
ea16c514 | 2566 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
2723feaa | 2567 | r = vhost_update_used_flags(vq); |
8ea8cf89 | 2568 | if (r) |
ae6961de | 2569 | vq_err(vq, "Failed to disable notification at %p: %d\n", |
8ea8cf89 MT |
2570 | &vq->used->flags, r); |
2571 | } | |
3a4d5c94 | 2572 | } |
6ac1afbf AH |
2573 | EXPORT_SYMBOL_GPL(vhost_disable_notify); |
2574 | ||
6b1e6cc7 JW |
2575 | /* Create a new message. */ |
2576 | struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type) | |
2577 | { | |
2578 | struct vhost_msg_node *node = kmalloc(sizeof *node, GFP_KERNEL); | |
2579 | if (!node) | |
2580 | return NULL; | |
670ae9ca MT |
2581 | |
2582 | /* Make sure all padding within the structure is initialized. */ | |
2583 | memset(&node->msg, 0, sizeof node->msg); | |
6b1e6cc7 JW |
2584 | node->vq = vq; |
2585 | node->msg.type = type; | |
2586 | return node; | |
2587 | } | |
2588 | EXPORT_SYMBOL_GPL(vhost_new_msg); | |
2589 | ||
2590 | void vhost_enqueue_msg(struct vhost_dev *dev, struct list_head *head, | |
2591 | struct vhost_msg_node *node) | |
2592 | { | |
2593 | spin_lock(&dev->iotlb_lock); | |
2594 | list_add_tail(&node->node, head); | |
2595 | spin_unlock(&dev->iotlb_lock); | |
2596 | ||
a9a08845 | 2597 | wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM); |
6b1e6cc7 JW |
2598 | } |
2599 | EXPORT_SYMBOL_GPL(vhost_enqueue_msg); | |
2600 | ||
2601 | struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev, | |
2602 | struct list_head *head) | |
2603 | { | |
2604 | struct vhost_msg_node *node = NULL; | |
2605 | ||
2606 | spin_lock(&dev->iotlb_lock); | |
2607 | if (!list_empty(head)) { | |
2608 | node = list_first_entry(head, struct vhost_msg_node, | |
2609 | node); | |
2610 | list_del(&node->node); | |
2611 | } | |
2612 | spin_unlock(&dev->iotlb_lock); | |
2613 | ||
2614 | return node; | |
2615 | } | |
2616 | EXPORT_SYMBOL_GPL(vhost_dequeue_msg); | |
2617 | ||
460f7ce1 JW |
2618 | void vhost_set_backend_features(struct vhost_dev *dev, u64 features) |
2619 | { | |
2620 | struct vhost_virtqueue *vq; | |
2621 | int i; | |
2622 | ||
2623 | mutex_lock(&dev->mutex); | |
2624 | for (i = 0; i < dev->nvqs; ++i) { | |
2625 | vq = dev->vqs[i]; | |
2626 | mutex_lock(&vq->mutex); | |
2627 | vq->acked_backend_features = features; | |
2628 | mutex_unlock(&vq->mutex); | |
2629 | } | |
2630 | mutex_unlock(&dev->mutex); | |
2631 | } | |
2632 | EXPORT_SYMBOL_GPL(vhost_set_backend_features); | |
6b1e6cc7 | 2633 | |
6ac1afbf AH |
2634 | static int __init vhost_init(void) |
2635 | { | |
2636 | return 0; | |
2637 | } | |
2638 | ||
2639 | static void __exit vhost_exit(void) | |
2640 | { | |
2641 | } | |
2642 | ||
2643 | module_init(vhost_init); | |
2644 | module_exit(vhost_exit); | |
2645 | ||
2646 | MODULE_VERSION("0.0.1"); | |
2647 | MODULE_LICENSE("GPL v2"); | |
2648 | MODULE_AUTHOR("Michael S. Tsirkin"); | |
2649 | MODULE_DESCRIPTION("Host kernel accelerator for virtio"); |