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(); |
d550dda1 NHE |
364 | if (need_resched()) |
365 | schedule(); | |
04b96e55 | 366 | } |
c23f3445 | 367 | } |
6e890c5d | 368 | |
64e1c807 | 369 | return 0; |
c23f3445 TH |
370 | } |
371 | ||
bab632d6 MT |
372 | static void vhost_vq_free_iovecs(struct vhost_virtqueue *vq) |
373 | { | |
374 | kfree(vq->indirect); | |
375 | vq->indirect = NULL; | |
376 | kfree(vq->log); | |
377 | vq->log = NULL; | |
378 | kfree(vq->heads); | |
379 | vq->heads = NULL; | |
bab632d6 MT |
380 | } |
381 | ||
e0e9b406 JW |
382 | /* Helper to allocate iovec buffers for all vqs. */ |
383 | static long vhost_dev_alloc_iovecs(struct vhost_dev *dev) | |
384 | { | |
6d5e6aa8 | 385 | struct vhost_virtqueue *vq; |
e0e9b406 | 386 | int i; |
d47effe1 | 387 | |
e0e9b406 | 388 | for (i = 0; i < dev->nvqs; ++i) { |
6d5e6aa8 | 389 | vq = dev->vqs[i]; |
6da2ec56 KC |
390 | vq->indirect = kmalloc_array(UIO_MAXIOV, |
391 | sizeof(*vq->indirect), | |
392 | GFP_KERNEL); | |
b46a0bf7 | 393 | vq->log = kmalloc_array(dev->iov_limit, sizeof(*vq->log), |
6da2ec56 | 394 | GFP_KERNEL); |
b46a0bf7 | 395 | vq->heads = kmalloc_array(dev->iov_limit, sizeof(*vq->heads), |
6da2ec56 | 396 | GFP_KERNEL); |
6d5e6aa8 | 397 | if (!vq->indirect || !vq->log || !vq->heads) |
e0e9b406 JW |
398 | goto err_nomem; |
399 | } | |
400 | return 0; | |
d47effe1 | 401 | |
e0e9b406 | 402 | err_nomem: |
bab632d6 | 403 | for (; i >= 0; --i) |
3ab2e420 | 404 | vhost_vq_free_iovecs(dev->vqs[i]); |
e0e9b406 JW |
405 | return -ENOMEM; |
406 | } | |
407 | ||
408 | static void vhost_dev_free_iovecs(struct vhost_dev *dev) | |
409 | { | |
410 | int i; | |
d47effe1 | 411 | |
bab632d6 | 412 | for (i = 0; i < dev->nvqs; ++i) |
3ab2e420 | 413 | vhost_vq_free_iovecs(dev->vqs[i]); |
e0e9b406 JW |
414 | } |
415 | ||
e82b9b07 JW |
416 | bool vhost_exceeds_weight(struct vhost_virtqueue *vq, |
417 | int pkts, int total_len) | |
418 | { | |
419 | struct vhost_dev *dev = vq->dev; | |
420 | ||
421 | if ((dev->byte_weight && total_len >= dev->byte_weight) || | |
422 | pkts >= dev->weight) { | |
423 | vhost_poll_queue(&vq->poll); | |
424 | return true; | |
425 | } | |
426 | ||
427 | return false; | |
428 | } | |
429 | EXPORT_SYMBOL_GPL(vhost_exceeds_weight); | |
430 | ||
4942e825 JW |
431 | static size_t vhost_get_avail_size(struct vhost_virtqueue *vq, |
432 | unsigned int num) | |
433 | { | |
434 | size_t event __maybe_unused = | |
435 | vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; | |
436 | ||
e4be66e5 | 437 | return size_add(struct_size(vq->avail, ring, num), event); |
4942e825 JW |
438 | } |
439 | ||
440 | static size_t vhost_get_used_size(struct vhost_virtqueue *vq, | |
441 | unsigned int num) | |
442 | { | |
443 | size_t event __maybe_unused = | |
444 | vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; | |
445 | ||
e4be66e5 | 446 | return size_add(struct_size(vq->used, ring, num), event); |
4942e825 JW |
447 | } |
448 | ||
449 | static size_t vhost_get_desc_size(struct vhost_virtqueue *vq, | |
450 | unsigned int num) | |
451 | { | |
452 | return sizeof(*vq->desc) * num; | |
453 | } | |
454 | ||
59566b6e | 455 | void vhost_dev_init(struct vhost_dev *dev, |
e82b9b07 | 456 | struct vhost_virtqueue **vqs, int nvqs, |
792a4f2e | 457 | int iov_limit, int weight, int byte_weight, |
01fcb1cb | 458 | bool use_worker, |
91233ad7 | 459 | int (*msg_handler)(struct vhost_dev *dev, u32 asid, |
792a4f2e | 460 | struct vhost_iotlb_msg *msg)) |
3a4d5c94 | 461 | { |
6d5e6aa8 | 462 | struct vhost_virtqueue *vq; |
3a4d5c94 | 463 | int i; |
c23f3445 | 464 | |
3a4d5c94 MT |
465 | dev->vqs = vqs; |
466 | dev->nvqs = nvqs; | |
467 | mutex_init(&dev->mutex); | |
468 | dev->log_ctx = NULL; | |
a9709d68 | 469 | dev->umem = NULL; |
6b1e6cc7 | 470 | dev->iotlb = NULL; |
3a4d5c94 | 471 | dev->mm = NULL; |
c23f3445 | 472 | dev->worker = NULL; |
b46a0bf7 | 473 | dev->iov_limit = iov_limit; |
e82b9b07 JW |
474 | dev->weight = weight; |
475 | dev->byte_weight = byte_weight; | |
01fcb1cb | 476 | dev->use_worker = use_worker; |
792a4f2e | 477 | dev->msg_handler = msg_handler; |
6b1e6cc7 JW |
478 | init_waitqueue_head(&dev->wait); |
479 | INIT_LIST_HEAD(&dev->read_list); | |
480 | INIT_LIST_HEAD(&dev->pending_list); | |
481 | spin_lock_init(&dev->iotlb_lock); | |
3d2c7d37 | 482 | |
3a4d5c94 MT |
483 | |
484 | for (i = 0; i < dev->nvqs; ++i) { | |
6d5e6aa8 AH |
485 | vq = dev->vqs[i]; |
486 | vq->log = NULL; | |
487 | vq->indirect = NULL; | |
488 | vq->heads = NULL; | |
489 | vq->dev = dev; | |
490 | mutex_init(&vq->mutex); | |
491 | vhost_vq_reset(dev, vq); | |
492 | if (vq->handle_kick) | |
493 | vhost_poll_init(&vq->poll, vq->handle_kick, | |
a9a08845 | 494 | EPOLLIN, dev); |
3a4d5c94 | 495 | } |
3a4d5c94 | 496 | } |
6ac1afbf | 497 | EXPORT_SYMBOL_GPL(vhost_dev_init); |
3a4d5c94 MT |
498 | |
499 | /* Caller should have device mutex */ | |
500 | long vhost_dev_check_owner(struct vhost_dev *dev) | |
501 | { | |
502 | /* Are you the owner? If not, I don't think you mean to do that */ | |
503 | return dev->mm == current->mm ? 0 : -EPERM; | |
504 | } | |
6ac1afbf | 505 | EXPORT_SYMBOL_GPL(vhost_dev_check_owner); |
3a4d5c94 | 506 | |
05c05351 MT |
507 | /* Caller should have device mutex */ |
508 | bool vhost_dev_has_owner(struct vhost_dev *dev) | |
509 | { | |
510 | return dev->mm; | |
511 | } | |
6ac1afbf | 512 | EXPORT_SYMBOL_GPL(vhost_dev_has_owner); |
05c05351 | 513 | |
5ce995f3 JW |
514 | static void vhost_attach_mm(struct vhost_dev *dev) |
515 | { | |
516 | /* No owner, become one */ | |
517 | if (dev->use_worker) { | |
518 | dev->mm = get_task_mm(current); | |
519 | } else { | |
520 | /* vDPA device does not use worker thead, so there's | |
521 | * no need to hold the address space for mm. This help | |
522 | * to avoid deadlock in the case of mmap() which may | |
523 | * held the refcnt of the file and depends on release | |
524 | * method to remove vma. | |
525 | */ | |
526 | dev->mm = current->mm; | |
527 | mmgrab(dev->mm); | |
528 | } | |
529 | } | |
530 | ||
531 | static void vhost_detach_mm(struct vhost_dev *dev) | |
532 | { | |
533 | if (!dev->mm) | |
534 | return; | |
535 | ||
536 | if (dev->use_worker) | |
537 | mmput(dev->mm); | |
538 | else | |
539 | mmdrop(dev->mm); | |
540 | ||
541 | dev->mm = NULL; | |
542 | } | |
543 | ||
1a5f8090 MC |
544 | static void vhost_worker_free(struct vhost_dev *dev) |
545 | { | |
546 | struct vhost_worker *worker = dev->worker; | |
547 | ||
548 | if (!worker) | |
549 | return; | |
550 | ||
551 | dev->worker = NULL; | |
552 | WARN_ON(!llist_empty(&worker->work_list)); | |
6e890c5d | 553 | vhost_task_stop(worker->vtsk); |
1a5f8090 MC |
554 | kfree(worker); |
555 | } | |
556 | ||
557 | static int vhost_worker_create(struct vhost_dev *dev) | |
558 | { | |
559 | struct vhost_worker *worker; | |
6e890c5d MC |
560 | struct vhost_task *vtsk; |
561 | char name[TASK_COMM_LEN]; | |
1a5f8090 MC |
562 | int ret; |
563 | ||
564 | worker = kzalloc(sizeof(*worker), GFP_KERNEL_ACCOUNT); | |
565 | if (!worker) | |
566 | return -ENOMEM; | |
567 | ||
568 | dev->worker = worker; | |
1a5f8090 MC |
569 | worker->kcov_handle = kcov_common_handle(); |
570 | init_llist_head(&worker->work_list); | |
6e890c5d | 571 | snprintf(name, sizeof(name), "vhost-%d", current->pid); |
1a5f8090 | 572 | |
6e890c5d MC |
573 | vtsk = vhost_task_create(vhost_worker, worker, name); |
574 | if (!vtsk) { | |
575 | ret = -ENOMEM; | |
1a5f8090 MC |
576 | goto free_worker; |
577 | } | |
578 | ||
6e890c5d MC |
579 | worker->vtsk = vtsk; |
580 | vhost_task_start(vtsk); | |
1a5f8090 MC |
581 | return 0; |
582 | ||
1a5f8090 MC |
583 | free_worker: |
584 | kfree(worker); | |
585 | dev->worker = NULL; | |
586 | return ret; | |
587 | } | |
588 | ||
3a4d5c94 | 589 | /* Caller should have device mutex */ |
54db63c2 | 590 | long vhost_dev_set_owner(struct vhost_dev *dev) |
3a4d5c94 | 591 | { |
c23f3445 | 592 | int err; |
d47effe1 | 593 | |
3a4d5c94 | 594 | /* Is there an owner already? */ |
05c05351 | 595 | if (vhost_dev_has_owner(dev)) { |
c23f3445 TH |
596 | err = -EBUSY; |
597 | goto err_mm; | |
598 | } | |
d47effe1 | 599 | |
5ce995f3 JW |
600 | vhost_attach_mm(dev); |
601 | ||
01fcb1cb | 602 | if (dev->use_worker) { |
1a5f8090 | 603 | err = vhost_worker_create(dev); |
01fcb1cb | 604 | if (err) |
1a5f8090 | 605 | goto err_worker; |
01fcb1cb | 606 | } |
c23f3445 | 607 | |
e0e9b406 JW |
608 | err = vhost_dev_alloc_iovecs(dev); |
609 | if (err) | |
1a5f8090 | 610 | goto err_iovecs; |
e0e9b406 | 611 | |
3a4d5c94 | 612 | return 0; |
1a5f8090 MC |
613 | err_iovecs: |
614 | vhost_worker_free(dev); | |
c23f3445 | 615 | err_worker: |
5ce995f3 | 616 | vhost_detach_mm(dev); |
c23f3445 TH |
617 | err_mm: |
618 | return err; | |
3a4d5c94 | 619 | } |
6ac1afbf | 620 | EXPORT_SYMBOL_GPL(vhost_dev_set_owner); |
3a4d5c94 | 621 | |
0bbe3066 JW |
622 | static struct vhost_iotlb *iotlb_alloc(void) |
623 | { | |
624 | return vhost_iotlb_alloc(max_iotlb_entries, | |
625 | VHOST_IOTLB_FLAG_RETIRE); | |
626 | } | |
627 | ||
628 | struct vhost_iotlb *vhost_dev_reset_owner_prepare(void) | |
a9709d68 | 629 | { |
0bbe3066 | 630 | return iotlb_alloc(); |
150b9e51 | 631 | } |
6ac1afbf | 632 | EXPORT_SYMBOL_GPL(vhost_dev_reset_owner_prepare); |
3a4d5c94 | 633 | |
150b9e51 | 634 | /* Caller should have device mutex */ |
0bbe3066 | 635 | void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_iotlb *umem) |
150b9e51 | 636 | { |
47283bef MT |
637 | int i; |
638 | ||
f6f93f75 | 639 | vhost_dev_cleanup(dev); |
3a4d5c94 | 640 | |
a9709d68 | 641 | dev->umem = umem; |
47283bef MT |
642 | /* We don't need VQ locks below since vhost_dev_cleanup makes sure |
643 | * VQs aren't running. | |
644 | */ | |
645 | for (i = 0; i < dev->nvqs; ++i) | |
a9709d68 | 646 | dev->vqs[i]->umem = umem; |
3a4d5c94 | 647 | } |
6ac1afbf | 648 | EXPORT_SYMBOL_GPL(vhost_dev_reset_owner); |
3a4d5c94 | 649 | |
b211616d | 650 | void vhost_dev_stop(struct vhost_dev *dev) |
bab632d6 MT |
651 | { |
652 | int i; | |
b211616d MT |
653 | |
654 | for (i = 0; i < dev->nvqs; ++i) { | |
6ca84326 | 655 | if (dev->vqs[i]->kick && dev->vqs[i]->handle_kick) |
3ab2e420 | 656 | vhost_poll_stop(&dev->vqs[i]->poll); |
bab632d6 | 657 | } |
6ca84326 | 658 | |
b2ffa407 | 659 | vhost_dev_flush(dev); |
bab632d6 | 660 | } |
6ac1afbf | 661 | EXPORT_SYMBOL_GPL(vhost_dev_stop); |
bab632d6 | 662 | |
9526f9a2 | 663 | void vhost_clear_msg(struct vhost_dev *dev) |
6b1e6cc7 JW |
664 | { |
665 | struct vhost_msg_node *node, *n; | |
666 | ||
667 | spin_lock(&dev->iotlb_lock); | |
668 | ||
669 | list_for_each_entry_safe(node, n, &dev->read_list, node) { | |
670 | list_del(&node->node); | |
671 | kfree(node); | |
672 | } | |
673 | ||
674 | list_for_each_entry_safe(node, n, &dev->pending_list, node) { | |
675 | list_del(&node->node); | |
676 | kfree(node); | |
677 | } | |
678 | ||
679 | spin_unlock(&dev->iotlb_lock); | |
680 | } | |
9526f9a2 | 681 | EXPORT_SYMBOL_GPL(vhost_clear_msg); |
6b1e6cc7 | 682 | |
f6f93f75 | 683 | void vhost_dev_cleanup(struct vhost_dev *dev) |
3a4d5c94 MT |
684 | { |
685 | int i; | |
d47effe1 | 686 | |
3a4d5c94 | 687 | for (i = 0; i < dev->nvqs; ++i) { |
3ab2e420 AH |
688 | if (dev->vqs[i]->error_ctx) |
689 | eventfd_ctx_put(dev->vqs[i]->error_ctx); | |
3ab2e420 AH |
690 | if (dev->vqs[i]->kick) |
691 | fput(dev->vqs[i]->kick); | |
265a0ad8 ZL |
692 | if (dev->vqs[i]->call_ctx.ctx) |
693 | eventfd_ctx_put(dev->vqs[i]->call_ctx.ctx); | |
3ab2e420 | 694 | vhost_vq_reset(dev, dev->vqs[i]); |
3a4d5c94 | 695 | } |
e0e9b406 | 696 | vhost_dev_free_iovecs(dev); |
3a4d5c94 MT |
697 | if (dev->log_ctx) |
698 | eventfd_ctx_put(dev->log_ctx); | |
699 | dev->log_ctx = NULL; | |
3a4d5c94 | 700 | /* No one will access memory at this point */ |
0bbe3066 | 701 | vhost_iotlb_free(dev->umem); |
a9709d68 | 702 | dev->umem = NULL; |
0bbe3066 | 703 | vhost_iotlb_free(dev->iotlb); |
6b1e6cc7 JW |
704 | dev->iotlb = NULL; |
705 | vhost_clear_msg(dev); | |
a9a08845 | 706 | wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM); |
1a5f8090 | 707 | vhost_worker_free(dev); |
5ce995f3 | 708 | vhost_detach_mm(dev); |
3a4d5c94 | 709 | } |
6ac1afbf | 710 | EXPORT_SYMBOL_GPL(vhost_dev_cleanup); |
3a4d5c94 | 711 | |
ddd3d408 | 712 | static bool log_access_ok(void __user *log_base, u64 addr, unsigned long sz) |
3a4d5c94 MT |
713 | { |
714 | u64 a = addr / VHOST_PAGE_SIZE / 8; | |
d47effe1 | 715 | |
3a4d5c94 MT |
716 | /* Make sure 64 bit math will not overflow. */ |
717 | if (a > ULONG_MAX - (unsigned long)log_base || | |
718 | a + (unsigned long)log_base > ULONG_MAX) | |
ddd3d408 | 719 | return false; |
3a4d5c94 | 720 | |
96d4f267 | 721 | return access_ok(log_base + a, |
3a4d5c94 MT |
722 | (sz + VHOST_PAGE_SIZE * 8 - 1) / VHOST_PAGE_SIZE / 8); |
723 | } | |
724 | ||
f7ad318e | 725 | /* Make sure 64 bit math will not overflow. */ |
ec33d031 MT |
726 | static bool vhost_overflow(u64 uaddr, u64 size) |
727 | { | |
f7ad318e XY |
728 | if (uaddr > ULONG_MAX || size > ULONG_MAX) |
729 | return true; | |
730 | ||
731 | if (!size) | |
732 | return false; | |
733 | ||
734 | return uaddr > ULONG_MAX - size + 1; | |
ec33d031 MT |
735 | } |
736 | ||
3a4d5c94 | 737 | /* Caller should have vq mutex and device mutex. */ |
0bbe3066 | 738 | static bool vq_memory_access_ok(void __user *log_base, struct vhost_iotlb *umem, |
ddd3d408 | 739 | int log_all) |
3a4d5c94 | 740 | { |
0bbe3066 | 741 | struct vhost_iotlb_map *map; |
179b284e | 742 | |
a9709d68 | 743 | if (!umem) |
ddd3d408 | 744 | return false; |
179b284e | 745 | |
0bbe3066 JW |
746 | list_for_each_entry(map, &umem->list, link) { |
747 | unsigned long a = map->addr; | |
a9709d68 | 748 | |
0bbe3066 | 749 | if (vhost_overflow(map->addr, map->size)) |
ddd3d408 | 750 | return false; |
ec33d031 MT |
751 | |
752 | ||
0bbe3066 | 753 | if (!access_ok((void __user *)a, map->size)) |
ddd3d408 | 754 | return false; |
3a4d5c94 | 755 | else if (log_all && !log_access_ok(log_base, |
0bbe3066 JW |
756 | map->start, |
757 | map->size)) | |
ddd3d408 | 758 | return false; |
3a4d5c94 | 759 | } |
ddd3d408 | 760 | return true; |
3a4d5c94 MT |
761 | } |
762 | ||
f8894913 JW |
763 | static inline void __user *vhost_vq_meta_fetch(struct vhost_virtqueue *vq, |
764 | u64 addr, unsigned int size, | |
765 | int type) | |
766 | { | |
0bbe3066 | 767 | const struct vhost_iotlb_map *map = vq->meta_iotlb[type]; |
f8894913 | 768 | |
0bbe3066 | 769 | if (!map) |
f8894913 JW |
770 | return NULL; |
771 | ||
1b0be99f | 772 | return (void __user *)(uintptr_t)(map->addr + addr - map->start); |
f8894913 JW |
773 | } |
774 | ||
3a4d5c94 MT |
775 | /* Can we switch to this memory table? */ |
776 | /* Caller should have device mutex but not vq mutex */ | |
0bbe3066 | 777 | static bool memory_access_ok(struct vhost_dev *d, struct vhost_iotlb *umem, |
ddd3d408 | 778 | int log_all) |
3a4d5c94 MT |
779 | { |
780 | int i; | |
d47effe1 | 781 | |
3a4d5c94 | 782 | for (i = 0; i < d->nvqs; ++i) { |
ddd3d408 | 783 | bool ok; |
ea16c514 MT |
784 | bool log; |
785 | ||
3ab2e420 | 786 | mutex_lock(&d->vqs[i]->mutex); |
ea16c514 | 787 | log = log_all || vhost_has_feature(d->vqs[i], VHOST_F_LOG_ALL); |
3a4d5c94 | 788 | /* If ring is inactive, will check when it's enabled. */ |
3ab2e420 | 789 | if (d->vqs[i]->private_data) |
a9709d68 JW |
790 | ok = vq_memory_access_ok(d->vqs[i]->log_base, |
791 | umem, log); | |
3a4d5c94 | 792 | else |
ddd3d408 | 793 | ok = true; |
3ab2e420 | 794 | mutex_unlock(&d->vqs[i]->mutex); |
3a4d5c94 | 795 | if (!ok) |
ddd3d408 | 796 | return false; |
3a4d5c94 | 797 | } |
ddd3d408 | 798 | return true; |
3a4d5c94 MT |
799 | } |
800 | ||
6b1e6cc7 JW |
801 | static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len, |
802 | struct iovec iov[], int iov_size, int access); | |
bfe2bc51 | 803 | |
72952cc0 | 804 | static int vhost_copy_to_user(struct vhost_virtqueue *vq, void __user *to, |
bfe2bc51 JW |
805 | const void *from, unsigned size) |
806 | { | |
6b1e6cc7 | 807 | int ret; |
bfe2bc51 | 808 | |
6b1e6cc7 JW |
809 | if (!vq->iotlb) |
810 | return __copy_to_user(to, from, size); | |
811 | else { | |
812 | /* This function should be called after iotlb | |
813 | * prefetch, which means we're sure that all vq | |
814 | * could be access through iotlb. So -EAGAIN should | |
815 | * not happen in this case. | |
816 | */ | |
6b1e6cc7 | 817 | struct iov_iter t; |
f8894913 JW |
818 | void __user *uaddr = vhost_vq_meta_fetch(vq, |
819 | (u64)(uintptr_t)to, size, | |
7ced6c98 | 820 | VHOST_ADDR_USED); |
f8894913 JW |
821 | |
822 | if (uaddr) | |
823 | return __copy_to_user(uaddr, from, size); | |
824 | ||
6b1e6cc7 JW |
825 | ret = translate_desc(vq, (u64)(uintptr_t)to, size, vq->iotlb_iov, |
826 | ARRAY_SIZE(vq->iotlb_iov), | |
827 | VHOST_ACCESS_WO); | |
828 | if (ret < 0) | |
829 | goto out; | |
de4eda9d | 830 | iov_iter_init(&t, ITER_DEST, vq->iotlb_iov, ret, size); |
6b1e6cc7 JW |
831 | ret = copy_to_iter(from, size, &t); |
832 | if (ret == size) | |
833 | ret = 0; | |
834 | } | |
835 | out: | |
836 | return ret; | |
837 | } | |
bfe2bc51 JW |
838 | |
839 | static int vhost_copy_from_user(struct vhost_virtqueue *vq, void *to, | |
72952cc0 | 840 | void __user *from, unsigned size) |
bfe2bc51 | 841 | { |
6b1e6cc7 JW |
842 | int ret; |
843 | ||
844 | if (!vq->iotlb) | |
845 | return __copy_from_user(to, from, size); | |
846 | else { | |
847 | /* This function should be called after iotlb | |
848 | * prefetch, which means we're sure that vq | |
849 | * could be access through iotlb. So -EAGAIN should | |
850 | * not happen in this case. | |
851 | */ | |
f8894913 JW |
852 | void __user *uaddr = vhost_vq_meta_fetch(vq, |
853 | (u64)(uintptr_t)from, size, | |
854 | VHOST_ADDR_DESC); | |
6b1e6cc7 | 855 | struct iov_iter f; |
f8894913 JW |
856 | |
857 | if (uaddr) | |
858 | return __copy_from_user(to, uaddr, size); | |
859 | ||
6b1e6cc7 JW |
860 | ret = translate_desc(vq, (u64)(uintptr_t)from, size, vq->iotlb_iov, |
861 | ARRAY_SIZE(vq->iotlb_iov), | |
862 | VHOST_ACCESS_RO); | |
863 | if (ret < 0) { | |
864 | vq_err(vq, "IOTLB translation failure: uaddr " | |
865 | "%p size 0x%llx\n", from, | |
866 | (unsigned long long) size); | |
867 | goto out; | |
868 | } | |
de4eda9d | 869 | iov_iter_init(&f, ITER_SOURCE, vq->iotlb_iov, ret, size); |
6b1e6cc7 JW |
870 | ret = copy_from_iter(to, size, &f); |
871 | if (ret == size) | |
872 | ret = 0; | |
873 | } | |
874 | ||
875 | out: | |
876 | return ret; | |
877 | } | |
878 | ||
f8894913 JW |
879 | static void __user *__vhost_get_user_slow(struct vhost_virtqueue *vq, |
880 | void __user *addr, unsigned int size, | |
881 | int type) | |
6b1e6cc7 JW |
882 | { |
883 | int ret; | |
884 | ||
6b1e6cc7 JW |
885 | ret = translate_desc(vq, (u64)(uintptr_t)addr, size, vq->iotlb_iov, |
886 | ARRAY_SIZE(vq->iotlb_iov), | |
887 | VHOST_ACCESS_RO); | |
888 | if (ret < 0) { | |
889 | vq_err(vq, "IOTLB translation failure: uaddr " | |
890 | "%p size 0x%llx\n", addr, | |
891 | (unsigned long long) size); | |
892 | return NULL; | |
893 | } | |
894 | ||
895 | if (ret != 1 || vq->iotlb_iov[0].iov_len != size) { | |
896 | vq_err(vq, "Non atomic userspace memory access: uaddr " | |
897 | "%p size 0x%llx\n", addr, | |
898 | (unsigned long long) size); | |
899 | return NULL; | |
900 | } | |
901 | ||
902 | return vq->iotlb_iov[0].iov_base; | |
903 | } | |
904 | ||
f8894913 JW |
905 | /* This function should be called after iotlb |
906 | * prefetch, which means we're sure that vq | |
907 | * could be access through iotlb. So -EAGAIN should | |
908 | * not happen in this case. | |
909 | */ | |
910 | static inline void __user *__vhost_get_user(struct vhost_virtqueue *vq, | |
1b0be99f | 911 | void __user *addr, unsigned int size, |
f8894913 JW |
912 | int type) |
913 | { | |
914 | void __user *uaddr = vhost_vq_meta_fetch(vq, | |
915 | (u64)(uintptr_t)addr, size, type); | |
916 | if (uaddr) | |
917 | return uaddr; | |
918 | ||
919 | return __vhost_get_user_slow(vq, addr, size, type); | |
920 | } | |
921 | ||
922 | #define vhost_put_user(vq, x, ptr) \ | |
6b1e6cc7 | 923 | ({ \ |
002ef18e | 924 | int ret; \ |
6b1e6cc7 JW |
925 | if (!vq->iotlb) { \ |
926 | ret = __put_user(x, ptr); \ | |
927 | } else { \ | |
928 | __typeof__(ptr) to = \ | |
f8894913 JW |
929 | (__typeof__(ptr)) __vhost_get_user(vq, ptr, \ |
930 | sizeof(*ptr), VHOST_ADDR_USED); \ | |
6b1e6cc7 JW |
931 | if (to != NULL) \ |
932 | ret = __put_user(x, to); \ | |
933 | else \ | |
934 | ret = -EFAULT; \ | |
935 | } \ | |
936 | ret; \ | |
937 | }) | |
938 | ||
7b5d753e JW |
939 | static inline int vhost_put_avail_event(struct vhost_virtqueue *vq) |
940 | { | |
941 | return vhost_put_user(vq, cpu_to_vhost16(vq, vq->avail_idx), | |
942 | vhost_avail_event(vq)); | |
943 | } | |
944 | ||
945 | static inline int vhost_put_used(struct vhost_virtqueue *vq, | |
946 | struct vring_used_elem *head, int idx, | |
947 | int count) | |
948 | { | |
949 | return vhost_copy_to_user(vq, vq->used->ring + idx, head, | |
950 | count * sizeof(*head)); | |
951 | } | |
952 | ||
953 | static inline int vhost_put_used_flags(struct vhost_virtqueue *vq) | |
954 | ||
955 | { | |
956 | return vhost_put_user(vq, cpu_to_vhost16(vq, vq->used_flags), | |
957 | &vq->used->flags); | |
958 | } | |
959 | ||
960 | static inline int vhost_put_used_idx(struct vhost_virtqueue *vq) | |
961 | ||
962 | { | |
963 | return vhost_put_user(vq, cpu_to_vhost16(vq, vq->last_used_idx), | |
964 | &vq->used->idx); | |
965 | } | |
966 | ||
f8894913 | 967 | #define vhost_get_user(vq, x, ptr, type) \ |
6b1e6cc7 JW |
968 | ({ \ |
969 | int ret; \ | |
970 | if (!vq->iotlb) { \ | |
971 | ret = __get_user(x, ptr); \ | |
972 | } else { \ | |
973 | __typeof__(ptr) from = \ | |
f8894913 JW |
974 | (__typeof__(ptr)) __vhost_get_user(vq, ptr, \ |
975 | sizeof(*ptr), \ | |
976 | type); \ | |
6b1e6cc7 JW |
977 | if (from != NULL) \ |
978 | ret = __get_user(x, from); \ | |
979 | else \ | |
980 | ret = -EFAULT; \ | |
981 | } \ | |
982 | ret; \ | |
983 | }) | |
984 | ||
f8894913 JW |
985 | #define vhost_get_avail(vq, x, ptr) \ |
986 | vhost_get_user(vq, x, ptr, VHOST_ADDR_AVAIL) | |
987 | ||
988 | #define vhost_get_used(vq, x, ptr) \ | |
989 | vhost_get_user(vq, x, ptr, VHOST_ADDR_USED) | |
990 | ||
86a07da3 JW |
991 | static void vhost_dev_lock_vqs(struct vhost_dev *d) |
992 | { | |
993 | int i = 0; | |
994 | for (i = 0; i < d->nvqs; ++i) | |
995 | mutex_lock_nested(&d->vqs[i]->mutex, i); | |
996 | } | |
997 | ||
998 | static void vhost_dev_unlock_vqs(struct vhost_dev *d) | |
999 | { | |
1000 | int i = 0; | |
1001 | for (i = 0; i < d->nvqs; ++i) | |
1002 | mutex_unlock(&d->vqs[i]->mutex); | |
1003 | } | |
1004 | ||
7b5d753e JW |
1005 | static inline int vhost_get_avail_idx(struct vhost_virtqueue *vq, |
1006 | __virtio16 *idx) | |
1007 | { | |
1008 | return vhost_get_avail(vq, *idx, &vq->avail->idx); | |
1009 | } | |
1010 | ||
1011 | static inline int vhost_get_avail_head(struct vhost_virtqueue *vq, | |
1012 | __virtio16 *head, int idx) | |
1013 | { | |
1014 | return vhost_get_avail(vq, *head, | |
1015 | &vq->avail->ring[idx & (vq->num - 1)]); | |
1016 | } | |
1017 | ||
1018 | static inline int vhost_get_avail_flags(struct vhost_virtqueue *vq, | |
1019 | __virtio16 *flags) | |
1020 | { | |
1021 | return vhost_get_avail(vq, *flags, &vq->avail->flags); | |
1022 | } | |
1023 | ||
1024 | static inline int vhost_get_used_event(struct vhost_virtqueue *vq, | |
1025 | __virtio16 *event) | |
1026 | { | |
1027 | return vhost_get_avail(vq, *event, vhost_used_event(vq)); | |
1028 | } | |
1029 | ||
1030 | static inline int vhost_get_used_idx(struct vhost_virtqueue *vq, | |
1031 | __virtio16 *idx) | |
1032 | { | |
1033 | return vhost_get_used(vq, *idx, &vq->used->idx); | |
1034 | } | |
1035 | ||
1036 | static inline int vhost_get_desc(struct vhost_virtqueue *vq, | |
1037 | struct vring_desc *desc, int idx) | |
1038 | { | |
1039 | return vhost_copy_from_user(vq, desc, vq->desc + idx, sizeof(*desc)); | |
1040 | } | |
1041 | ||
6b1e6cc7 JW |
1042 | static void vhost_iotlb_notify_vq(struct vhost_dev *d, |
1043 | struct vhost_iotlb_msg *msg) | |
1044 | { | |
1045 | struct vhost_msg_node *node, *n; | |
1046 | ||
1047 | spin_lock(&d->iotlb_lock); | |
1048 | ||
1049 | list_for_each_entry_safe(node, n, &d->pending_list, node) { | |
1050 | struct vhost_iotlb_msg *vq_msg = &node->msg.iotlb; | |
1051 | if (msg->iova <= vq_msg->iova && | |
2d66f997 | 1052 | msg->iova + msg->size - 1 >= vq_msg->iova && |
6b1e6cc7 JW |
1053 | vq_msg->type == VHOST_IOTLB_MISS) { |
1054 | vhost_poll_queue(&node->vq->poll); | |
1055 | list_del(&node->node); | |
1056 | kfree(node); | |
1057 | } | |
1058 | } | |
1059 | ||
1060 | spin_unlock(&d->iotlb_lock); | |
1061 | } | |
1062 | ||
ddd3d408 | 1063 | static bool umem_access_ok(u64 uaddr, u64 size, int access) |
6b1e6cc7 JW |
1064 | { |
1065 | unsigned long a = uaddr; | |
1066 | ||
ec33d031 MT |
1067 | /* Make sure 64 bit math will not overflow. */ |
1068 | if (vhost_overflow(uaddr, size)) | |
ddd3d408 | 1069 | return false; |
ec33d031 | 1070 | |
6b1e6cc7 | 1071 | if ((access & VHOST_ACCESS_RO) && |
96d4f267 | 1072 | !access_ok((void __user *)a, size)) |
ddd3d408 | 1073 | return false; |
6b1e6cc7 | 1074 | if ((access & VHOST_ACCESS_WO) && |
96d4f267 | 1075 | !access_ok((void __user *)a, size)) |
ddd3d408 SH |
1076 | return false; |
1077 | return true; | |
6b1e6cc7 JW |
1078 | } |
1079 | ||
91233ad7 | 1080 | static int vhost_process_iotlb_msg(struct vhost_dev *dev, u32 asid, |
72952cc0 | 1081 | struct vhost_iotlb_msg *msg) |
6b1e6cc7 JW |
1082 | { |
1083 | int ret = 0; | |
1084 | ||
91233ad7 GD |
1085 | if (asid != 0) |
1086 | return -EINVAL; | |
1087 | ||
1b15ad68 | 1088 | mutex_lock(&dev->mutex); |
86a07da3 | 1089 | vhost_dev_lock_vqs(dev); |
6b1e6cc7 JW |
1090 | switch (msg->type) { |
1091 | case VHOST_IOTLB_UPDATE: | |
1092 | if (!dev->iotlb) { | |
1093 | ret = -EFAULT; | |
1094 | break; | |
1095 | } | |
ddd3d408 | 1096 | if (!umem_access_ok(msg->uaddr, msg->size, msg->perm)) { |
6b1e6cc7 JW |
1097 | ret = -EFAULT; |
1098 | break; | |
1099 | } | |
f8894913 | 1100 | vhost_vq_meta_reset(dev); |
0bbe3066 JW |
1101 | if (vhost_iotlb_add_range(dev->iotlb, msg->iova, |
1102 | msg->iova + msg->size - 1, | |
1103 | msg->uaddr, msg->perm)) { | |
6b1e6cc7 JW |
1104 | ret = -ENOMEM; |
1105 | break; | |
1106 | } | |
1107 | vhost_iotlb_notify_vq(dev, msg); | |
1108 | break; | |
1109 | case VHOST_IOTLB_INVALIDATE: | |
6f3180af JW |
1110 | if (!dev->iotlb) { |
1111 | ret = -EFAULT; | |
1112 | break; | |
1113 | } | |
f8894913 | 1114 | vhost_vq_meta_reset(dev); |
0bbe3066 JW |
1115 | vhost_iotlb_del_range(dev->iotlb, msg->iova, |
1116 | msg->iova + msg->size - 1); | |
6b1e6cc7 JW |
1117 | break; |
1118 | default: | |
1119 | ret = -EINVAL; | |
1120 | break; | |
1121 | } | |
1122 | ||
86a07da3 | 1123 | vhost_dev_unlock_vqs(dev); |
1b15ad68 JW |
1124 | mutex_unlock(&dev->mutex); |
1125 | ||
6b1e6cc7 JW |
1126 | return ret; |
1127 | } | |
1128 | ssize_t vhost_chr_write_iter(struct vhost_dev *dev, | |
1129 | struct iov_iter *from) | |
1130 | { | |
429711ae JW |
1131 | struct vhost_iotlb_msg msg; |
1132 | size_t offset; | |
1133 | int type, ret; | |
91233ad7 | 1134 | u32 asid = 0; |
6b1e6cc7 | 1135 | |
429711ae | 1136 | ret = copy_from_iter(&type, sizeof(type), from); |
74ad7419 PT |
1137 | if (ret != sizeof(type)) { |
1138 | ret = -EINVAL; | |
6b1e6cc7 | 1139 | goto done; |
74ad7419 | 1140 | } |
6b1e6cc7 | 1141 | |
429711ae | 1142 | switch (type) { |
6b1e6cc7 | 1143 | case VHOST_IOTLB_MSG: |
429711ae JW |
1144 | /* There maybe a hole after type for V1 message type, |
1145 | * so skip it here. | |
1146 | */ | |
1147 | offset = offsetof(struct vhost_msg, iotlb) - sizeof(int); | |
1148 | break; | |
1149 | case VHOST_IOTLB_MSG_V2: | |
91233ad7 GD |
1150 | if (vhost_backend_has_feature(dev->vqs[0], |
1151 | VHOST_BACKEND_F_IOTLB_ASID)) { | |
1152 | ret = copy_from_iter(&asid, sizeof(asid), from); | |
1153 | if (ret != sizeof(asid)) { | |
1154 | ret = -EINVAL; | |
1155 | goto done; | |
1156 | } | |
aaca8373 | 1157 | offset = 0; |
91233ad7 GD |
1158 | } else |
1159 | offset = sizeof(__u32); | |
6b1e6cc7 JW |
1160 | break; |
1161 | default: | |
1162 | ret = -EINVAL; | |
429711ae | 1163 | goto done; |
6b1e6cc7 JW |
1164 | } |
1165 | ||
429711ae JW |
1166 | iov_iter_advance(from, offset); |
1167 | ret = copy_from_iter(&msg, sizeof(msg), from); | |
74ad7419 PT |
1168 | if (ret != sizeof(msg)) { |
1169 | ret = -EINVAL; | |
429711ae | 1170 | goto done; |
74ad7419 | 1171 | } |
792a4f2e | 1172 | |
95932ab2 JW |
1173 | if ((msg.type == VHOST_IOTLB_UPDATE || |
1174 | msg.type == VHOST_IOTLB_INVALIDATE) && | |
1175 | msg.size == 0) { | |
e2ae38cf AR |
1176 | ret = -EINVAL; |
1177 | goto done; | |
1178 | } | |
1179 | ||
792a4f2e | 1180 | if (dev->msg_handler) |
91233ad7 | 1181 | ret = dev->msg_handler(dev, asid, &msg); |
792a4f2e | 1182 | else |
91233ad7 | 1183 | ret = vhost_process_iotlb_msg(dev, asid, &msg); |
792a4f2e | 1184 | if (ret) { |
429711ae JW |
1185 | ret = -EFAULT; |
1186 | goto done; | |
1187 | } | |
1188 | ||
1189 | ret = (type == VHOST_IOTLB_MSG) ? sizeof(struct vhost_msg) : | |
1190 | sizeof(struct vhost_msg_v2); | |
6b1e6cc7 JW |
1191 | done: |
1192 | return ret; | |
1193 | } | |
1194 | EXPORT_SYMBOL(vhost_chr_write_iter); | |
1195 | ||
afc9a42b | 1196 | __poll_t vhost_chr_poll(struct file *file, struct vhost_dev *dev, |
6b1e6cc7 JW |
1197 | poll_table *wait) |
1198 | { | |
afc9a42b | 1199 | __poll_t mask = 0; |
6b1e6cc7 JW |
1200 | |
1201 | poll_wait(file, &dev->wait, wait); | |
1202 | ||
1203 | if (!list_empty(&dev->read_list)) | |
a9a08845 | 1204 | mask |= EPOLLIN | EPOLLRDNORM; |
6b1e6cc7 JW |
1205 | |
1206 | return mask; | |
1207 | } | |
1208 | EXPORT_SYMBOL(vhost_chr_poll); | |
1209 | ||
1210 | ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to, | |
1211 | int noblock) | |
1212 | { | |
1213 | DEFINE_WAIT(wait); | |
1214 | struct vhost_msg_node *node; | |
1215 | ssize_t ret = 0; | |
1216 | unsigned size = sizeof(struct vhost_msg); | |
1217 | ||
1218 | if (iov_iter_count(to) < size) | |
1219 | return 0; | |
1220 | ||
1221 | while (1) { | |
1222 | if (!noblock) | |
1223 | prepare_to_wait(&dev->wait, &wait, | |
1224 | TASK_INTERRUPTIBLE); | |
1225 | ||
1226 | node = vhost_dequeue_msg(dev, &dev->read_list); | |
1227 | if (node) | |
1228 | break; | |
1229 | if (noblock) { | |
1230 | ret = -EAGAIN; | |
1231 | break; | |
1232 | } | |
1233 | if (signal_pending(current)) { | |
1234 | ret = -ERESTARTSYS; | |
1235 | break; | |
1236 | } | |
1237 | if (!dev->iotlb) { | |
1238 | ret = -EBADFD; | |
1239 | break; | |
1240 | } | |
1241 | ||
1242 | schedule(); | |
1243 | } | |
1244 | ||
1245 | if (!noblock) | |
1246 | finish_wait(&dev->wait, &wait); | |
1247 | ||
1248 | if (node) { | |
429711ae JW |
1249 | struct vhost_iotlb_msg *msg; |
1250 | void *start = &node->msg; | |
6b1e6cc7 | 1251 | |
429711ae JW |
1252 | switch (node->msg.type) { |
1253 | case VHOST_IOTLB_MSG: | |
1254 | size = sizeof(node->msg); | |
1255 | msg = &node->msg.iotlb; | |
1256 | break; | |
1257 | case VHOST_IOTLB_MSG_V2: | |
1258 | size = sizeof(node->msg_v2); | |
1259 | msg = &node->msg_v2.iotlb; | |
1260 | break; | |
1261 | default: | |
1262 | BUG(); | |
1263 | break; | |
1264 | } | |
1265 | ||
1266 | ret = copy_to_iter(start, size, to); | |
1267 | if (ret != size || msg->type != VHOST_IOTLB_MISS) { | |
6b1e6cc7 JW |
1268 | kfree(node); |
1269 | return ret; | |
1270 | } | |
6b1e6cc7 JW |
1271 | vhost_enqueue_msg(dev, &dev->pending_list, node); |
1272 | } | |
1273 | ||
1274 | return ret; | |
1275 | } | |
1276 | EXPORT_SYMBOL_GPL(vhost_chr_read_iter); | |
1277 | ||
1278 | static int vhost_iotlb_miss(struct vhost_virtqueue *vq, u64 iova, int access) | |
1279 | { | |
1280 | struct vhost_dev *dev = vq->dev; | |
1281 | struct vhost_msg_node *node; | |
1282 | struct vhost_iotlb_msg *msg; | |
429711ae | 1283 | bool v2 = vhost_backend_has_feature(vq, VHOST_BACKEND_F_IOTLB_MSG_V2); |
6b1e6cc7 | 1284 | |
429711ae | 1285 | node = vhost_new_msg(vq, v2 ? VHOST_IOTLB_MSG_V2 : VHOST_IOTLB_MSG); |
6b1e6cc7 JW |
1286 | if (!node) |
1287 | return -ENOMEM; | |
1288 | ||
429711ae JW |
1289 | if (v2) { |
1290 | node->msg_v2.type = VHOST_IOTLB_MSG_V2; | |
1291 | msg = &node->msg_v2.iotlb; | |
1292 | } else { | |
1293 | msg = &node->msg.iotlb; | |
1294 | } | |
1295 | ||
6b1e6cc7 JW |
1296 | msg->type = VHOST_IOTLB_MISS; |
1297 | msg->iova = iova; | |
1298 | msg->perm = access; | |
1299 | ||
1300 | vhost_enqueue_msg(dev, &dev->read_list, node); | |
1301 | ||
1302 | return 0; | |
bfe2bc51 JW |
1303 | } |
1304 | ||
ddd3d408 | 1305 | static bool vq_access_ok(struct vhost_virtqueue *vq, unsigned int num, |
a865e420 MT |
1306 | vring_desc_t __user *desc, |
1307 | vring_avail_t __user *avail, | |
1308 | vring_used_t __user *used) | |
6b1e6cc7 | 1309 | |
3a4d5c94 | 1310 | { |
0210a8db GK |
1311 | /* If an IOTLB device is present, the vring addresses are |
1312 | * GIOVAs. Access validation occurs at prefetch time. */ | |
1313 | if (vq->iotlb) | |
1314 | return true; | |
1315 | ||
4942e825 JW |
1316 | return access_ok(desc, vhost_get_desc_size(vq, num)) && |
1317 | access_ok(avail, vhost_get_avail_size(vq, num)) && | |
1318 | access_ok(used, vhost_get_used_size(vq, num)); | |
3a4d5c94 MT |
1319 | } |
1320 | ||
f8894913 | 1321 | static void vhost_vq_meta_update(struct vhost_virtqueue *vq, |
0bbe3066 | 1322 | const struct vhost_iotlb_map *map, |
f8894913 JW |
1323 | int type) |
1324 | { | |
1325 | int access = (type == VHOST_ADDR_USED) ? | |
1326 | VHOST_ACCESS_WO : VHOST_ACCESS_RO; | |
1327 | ||
0bbe3066 JW |
1328 | if (likely(map->perm & access)) |
1329 | vq->meta_iotlb[type] = map; | |
f8894913 JW |
1330 | } |
1331 | ||
ddd3d408 SH |
1332 | static bool iotlb_access_ok(struct vhost_virtqueue *vq, |
1333 | int access, u64 addr, u64 len, int type) | |
6b1e6cc7 | 1334 | { |
0bbe3066 JW |
1335 | const struct vhost_iotlb_map *map; |
1336 | struct vhost_iotlb *umem = vq->iotlb; | |
ca2c5b33 | 1337 | u64 s = 0, size, orig_addr = addr, last = addr + len - 1; |
f8894913 JW |
1338 | |
1339 | if (vhost_vq_meta_fetch(vq, addr, len, type)) | |
1340 | return true; | |
6b1e6cc7 JW |
1341 | |
1342 | while (len > s) { | |
0bbe3066 JW |
1343 | map = vhost_iotlb_itree_first(umem, addr, last); |
1344 | if (map == NULL || map->start > addr) { | |
6b1e6cc7 JW |
1345 | vhost_iotlb_miss(vq, addr, access); |
1346 | return false; | |
0bbe3066 | 1347 | } else if (!(map->perm & access)) { |
6b1e6cc7 JW |
1348 | /* Report the possible access violation by |
1349 | * request another translation from userspace. | |
1350 | */ | |
1351 | return false; | |
1352 | } | |
1353 | ||
0bbe3066 | 1354 | size = map->size - addr + map->start; |
f8894913 JW |
1355 | |
1356 | if (orig_addr == addr && size >= len) | |
0bbe3066 | 1357 | vhost_vq_meta_update(vq, map, type); |
f8894913 | 1358 | |
6b1e6cc7 JW |
1359 | s += size; |
1360 | addr += size; | |
1361 | } | |
1362 | ||
1363 | return true; | |
1364 | } | |
1365 | ||
9b5e830b | 1366 | int vq_meta_prefetch(struct vhost_virtqueue *vq) |
6b1e6cc7 | 1367 | { |
6b1e6cc7 JW |
1368 | unsigned int num = vq->num; |
1369 | ||
3d2c7d37 | 1370 | if (!vq->iotlb) |
6b1e6cc7 JW |
1371 | return 1; |
1372 | ||
0bbe3066 | 1373 | return iotlb_access_ok(vq, VHOST_MAP_RO, (u64)(uintptr_t)vq->desc, |
4942e825 | 1374 | vhost_get_desc_size(vq, num), VHOST_ADDR_DESC) && |
0bbe3066 | 1375 | iotlb_access_ok(vq, VHOST_MAP_RO, (u64)(uintptr_t)vq->avail, |
4942e825 | 1376 | vhost_get_avail_size(vq, num), |
f8894913 | 1377 | VHOST_ADDR_AVAIL) && |
0bbe3066 | 1378 | iotlb_access_ok(vq, VHOST_MAP_WO, (u64)(uintptr_t)vq->used, |
4942e825 | 1379 | vhost_get_used_size(vq, num), VHOST_ADDR_USED); |
6b1e6cc7 | 1380 | } |
9b5e830b | 1381 | EXPORT_SYMBOL_GPL(vq_meta_prefetch); |
6b1e6cc7 | 1382 | |
3a4d5c94 MT |
1383 | /* Can we log writes? */ |
1384 | /* Caller should have device mutex but not vq mutex */ | |
ddd3d408 | 1385 | bool vhost_log_access_ok(struct vhost_dev *dev) |
3a4d5c94 | 1386 | { |
a9709d68 | 1387 | return memory_access_ok(dev, dev->umem, 1); |
3a4d5c94 | 1388 | } |
6ac1afbf | 1389 | EXPORT_SYMBOL_GPL(vhost_log_access_ok); |
3a4d5c94 | 1390 | |
ab512251 GK |
1391 | static bool vq_log_used_access_ok(struct vhost_virtqueue *vq, |
1392 | void __user *log_base, | |
1393 | bool log_used, | |
1394 | u64 log_addr) | |
1395 | { | |
1396 | /* If an IOTLB device is present, log_addr is a GIOVA that | |
1397 | * will never be logged by log_used(). */ | |
1398 | if (vq->iotlb) | |
1399 | return true; | |
1400 | ||
1401 | return !log_used || log_access_ok(log_base, log_addr, | |
1402 | vhost_get_used_size(vq, vq->num)); | |
1403 | } | |
1404 | ||
3a4d5c94 MT |
1405 | /* Verify access for write logging. */ |
1406 | /* Caller should have vq mutex and device mutex */ | |
ddd3d408 SH |
1407 | static bool vq_log_access_ok(struct vhost_virtqueue *vq, |
1408 | void __user *log_base) | |
3a4d5c94 | 1409 | { |
a9709d68 | 1410 | return vq_memory_access_ok(log_base, vq->umem, |
ea16c514 | 1411 | vhost_has_feature(vq, VHOST_F_LOG_ALL)) && |
ab512251 | 1412 | vq_log_used_access_ok(vq, log_base, vq->log_used, vq->log_addr); |
3a4d5c94 MT |
1413 | } |
1414 | ||
1415 | /* Can we start vq? */ | |
1416 | /* Caller should have vq mutex and device mutex */ | |
ddd3d408 | 1417 | bool vhost_vq_access_ok(struct vhost_virtqueue *vq) |
3a4d5c94 | 1418 | { |
d14d2b78 | 1419 | if (!vq_log_access_ok(vq, vq->log_base)) |
ddd3d408 | 1420 | return false; |
d65026c6 | 1421 | |
d65026c6 | 1422 | return vq_access_ok(vq, vq->num, vq->desc, vq->avail, vq->used); |
3a4d5c94 | 1423 | } |
6ac1afbf | 1424 | EXPORT_SYMBOL_GPL(vhost_vq_access_ok); |
3a4d5c94 MT |
1425 | |
1426 | static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m) | |
1427 | { | |
a9709d68 JW |
1428 | struct vhost_memory mem, *newmem; |
1429 | struct vhost_memory_region *region; | |
0bbe3066 | 1430 | struct vhost_iotlb *newumem, *oldumem; |
3a4d5c94 | 1431 | unsigned long size = offsetof(struct vhost_memory, regions); |
98f9ca0a | 1432 | int i; |
d47effe1 | 1433 | |
7ad9c9d2 TY |
1434 | if (copy_from_user(&mem, m, size)) |
1435 | return -EFAULT; | |
3a4d5c94 MT |
1436 | if (mem.padding) |
1437 | return -EOPNOTSUPP; | |
c9ce42f7 | 1438 | if (mem.nregions > max_mem_regions) |
3a4d5c94 | 1439 | return -E2BIG; |
b2303d7b MW |
1440 | newmem = kvzalloc(struct_size(newmem, regions, mem.nregions), |
1441 | GFP_KERNEL); | |
3a4d5c94 MT |
1442 | if (!newmem) |
1443 | return -ENOMEM; | |
1444 | ||
1445 | memcpy(newmem, &mem, size); | |
7ad9c9d2 | 1446 | if (copy_from_user(newmem->regions, m->regions, |
bf11d71a | 1447 | flex_array_size(newmem, regions, mem.nregions))) { |
bcfeacab | 1448 | kvfree(newmem); |
7ad9c9d2 | 1449 | return -EFAULT; |
3a4d5c94 MT |
1450 | } |
1451 | ||
0bbe3066 | 1452 | newumem = iotlb_alloc(); |
a9709d68 | 1453 | if (!newumem) { |
4de7255f | 1454 | kvfree(newmem); |
a9709d68 JW |
1455 | return -ENOMEM; |
1456 | } | |
1457 | ||
a9709d68 JW |
1458 | for (region = newmem->regions; |
1459 | region < newmem->regions + mem.nregions; | |
1460 | region++) { | |
0bbe3066 JW |
1461 | if (vhost_iotlb_add_range(newumem, |
1462 | region->guest_phys_addr, | |
1463 | region->guest_phys_addr + | |
1464 | region->memory_size - 1, | |
1465 | region->userspace_addr, | |
1466 | VHOST_MAP_RW)) | |
a9709d68 | 1467 | goto err; |
a02c3789 | 1468 | } |
a9709d68 JW |
1469 | |
1470 | if (!memory_access_ok(d, newumem, 0)) | |
1471 | goto err; | |
1472 | ||
1473 | oldumem = d->umem; | |
1474 | d->umem = newumem; | |
98f9ca0a | 1475 | |
47283bef | 1476 | /* All memory accesses are done under some VQ mutex. */ |
98f9ca0a MT |
1477 | for (i = 0; i < d->nvqs; ++i) { |
1478 | mutex_lock(&d->vqs[i]->mutex); | |
a9709d68 | 1479 | d->vqs[i]->umem = newumem; |
98f9ca0a MT |
1480 | mutex_unlock(&d->vqs[i]->mutex); |
1481 | } | |
a9709d68 JW |
1482 | |
1483 | kvfree(newmem); | |
0bbe3066 | 1484 | vhost_iotlb_free(oldumem); |
3a4d5c94 | 1485 | return 0; |
a9709d68 JW |
1486 | |
1487 | err: | |
0bbe3066 | 1488 | vhost_iotlb_free(newumem); |
a9709d68 JW |
1489 | kvfree(newmem); |
1490 | return -EFAULT; | |
3a4d5c94 MT |
1491 | } |
1492 | ||
feebcaea JW |
1493 | static long vhost_vring_set_num(struct vhost_dev *d, |
1494 | struct vhost_virtqueue *vq, | |
1495 | void __user *argp) | |
1496 | { | |
1497 | struct vhost_vring_state s; | |
1498 | ||
1499 | /* Resizing ring with an active backend? | |
1500 | * You don't want to do that. */ | |
1501 | if (vq->private_data) | |
1502 | return -EBUSY; | |
1503 | ||
1504 | if (copy_from_user(&s, argp, sizeof s)) | |
1505 | return -EFAULT; | |
1506 | ||
1507 | if (!s.num || s.num > 0xffff || (s.num & (s.num - 1))) | |
1508 | return -EINVAL; | |
1509 | vq->num = s.num; | |
1510 | ||
1511 | return 0; | |
1512 | } | |
1513 | ||
1514 | static long vhost_vring_set_addr(struct vhost_dev *d, | |
1515 | struct vhost_virtqueue *vq, | |
1516 | void __user *argp) | |
1517 | { | |
1518 | struct vhost_vring_addr a; | |
1519 | ||
1520 | if (copy_from_user(&a, argp, sizeof a)) | |
1521 | return -EFAULT; | |
1522 | if (a.flags & ~(0x1 << VHOST_VRING_F_LOG)) | |
1523 | return -EOPNOTSUPP; | |
1524 | ||
1525 | /* For 32bit, verify that the top 32bits of the user | |
1526 | data are set to zero. */ | |
1527 | if ((u64)(unsigned long)a.desc_user_addr != a.desc_user_addr || | |
1528 | (u64)(unsigned long)a.used_user_addr != a.used_user_addr || | |
1529 | (u64)(unsigned long)a.avail_user_addr != a.avail_user_addr) | |
1530 | return -EFAULT; | |
1531 | ||
1532 | /* Make sure it's safe to cast pointers to vring types. */ | |
1533 | BUILD_BUG_ON(__alignof__ *vq->avail > VRING_AVAIL_ALIGN_SIZE); | |
1534 | BUILD_BUG_ON(__alignof__ *vq->used > VRING_USED_ALIGN_SIZE); | |
1535 | if ((a.avail_user_addr & (VRING_AVAIL_ALIGN_SIZE - 1)) || | |
1536 | (a.used_user_addr & (VRING_USED_ALIGN_SIZE - 1)) || | |
1537 | (a.log_guest_addr & (VRING_USED_ALIGN_SIZE - 1))) | |
1538 | return -EINVAL; | |
1539 | ||
1540 | /* We only verify access here if backend is configured. | |
1541 | * If it is not, we don't as size might not have been setup. | |
1542 | * We will verify when backend is configured. */ | |
1543 | if (vq->private_data) { | |
1544 | if (!vq_access_ok(vq, vq->num, | |
1545 | (void __user *)(unsigned long)a.desc_user_addr, | |
1546 | (void __user *)(unsigned long)a.avail_user_addr, | |
1547 | (void __user *)(unsigned long)a.used_user_addr)) | |
1548 | return -EINVAL; | |
1549 | ||
1550 | /* Also validate log access for used ring if enabled. */ | |
ab512251 GK |
1551 | if (!vq_log_used_access_ok(vq, vq->log_base, |
1552 | a.flags & (0x1 << VHOST_VRING_F_LOG), | |
1553 | a.log_guest_addr)) | |
feebcaea JW |
1554 | return -EINVAL; |
1555 | } | |
1556 | ||
1557 | vq->log_used = !!(a.flags & (0x1 << VHOST_VRING_F_LOG)); | |
1558 | vq->desc = (void __user *)(unsigned long)a.desc_user_addr; | |
1559 | vq->avail = (void __user *)(unsigned long)a.avail_user_addr; | |
1560 | vq->log_addr = a.log_guest_addr; | |
1561 | vq->used = (void __user *)(unsigned long)a.used_user_addr; | |
1562 | ||
1563 | return 0; | |
1564 | } | |
1565 | ||
1566 | static long vhost_vring_set_num_addr(struct vhost_dev *d, | |
1567 | struct vhost_virtqueue *vq, | |
1568 | unsigned int ioctl, | |
1569 | void __user *argp) | |
1570 | { | |
1571 | long r; | |
1572 | ||
1573 | mutex_lock(&vq->mutex); | |
1574 | ||
1575 | switch (ioctl) { | |
1576 | case VHOST_SET_VRING_NUM: | |
1577 | r = vhost_vring_set_num(d, vq, argp); | |
1578 | break; | |
1579 | case VHOST_SET_VRING_ADDR: | |
1580 | r = vhost_vring_set_addr(d, vq, argp); | |
1581 | break; | |
1582 | default: | |
1583 | BUG(); | |
1584 | } | |
1585 | ||
1586 | mutex_unlock(&vq->mutex); | |
1587 | ||
1588 | return r; | |
1589 | } | |
26b36604 | 1590 | long vhost_vring_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp) |
3a4d5c94 | 1591 | { |
cecb46f1 AV |
1592 | struct file *eventfp, *filep = NULL; |
1593 | bool pollstart = false, pollstop = false; | |
3a4d5c94 MT |
1594 | struct eventfd_ctx *ctx = NULL; |
1595 | u32 __user *idxp = argp; | |
1596 | struct vhost_virtqueue *vq; | |
1597 | struct vhost_vring_state s; | |
1598 | struct vhost_vring_file f; | |
3a4d5c94 MT |
1599 | u32 idx; |
1600 | long r; | |
1601 | ||
1602 | r = get_user(idx, idxp); | |
1603 | if (r < 0) | |
1604 | return r; | |
0f3d9a17 | 1605 | if (idx >= d->nvqs) |
3a4d5c94 MT |
1606 | return -ENOBUFS; |
1607 | ||
ff002269 | 1608 | idx = array_index_nospec(idx, d->nvqs); |
3ab2e420 | 1609 | vq = d->vqs[idx]; |
3a4d5c94 | 1610 | |
feebcaea JW |
1611 | if (ioctl == VHOST_SET_VRING_NUM || |
1612 | ioctl == VHOST_SET_VRING_ADDR) { | |
1613 | return vhost_vring_set_num_addr(d, vq, ioctl, argp); | |
1614 | } | |
1615 | ||
3a4d5c94 MT |
1616 | mutex_lock(&vq->mutex); |
1617 | ||
1618 | switch (ioctl) { | |
3a4d5c94 MT |
1619 | case VHOST_SET_VRING_BASE: |
1620 | /* Moving base with an active backend? | |
1621 | * You don't want to do that. */ | |
1622 | if (vq->private_data) { | |
1623 | r = -EBUSY; | |
1624 | break; | |
1625 | } | |
7ad9c9d2 TY |
1626 | if (copy_from_user(&s, argp, sizeof s)) { |
1627 | r = -EFAULT; | |
3a4d5c94 | 1628 | break; |
7ad9c9d2 | 1629 | } |
3a4d5c94 MT |
1630 | if (s.num > 0xffff) { |
1631 | r = -EINVAL; | |
1632 | break; | |
1633 | } | |
8d65843c | 1634 | vq->last_avail_idx = s.num; |
3a4d5c94 MT |
1635 | /* Forget the cached index value. */ |
1636 | vq->avail_idx = vq->last_avail_idx; | |
1637 | break; | |
1638 | case VHOST_GET_VRING_BASE: | |
1639 | s.index = idx; | |
1640 | s.num = vq->last_avail_idx; | |
7ad9c9d2 TY |
1641 | if (copy_to_user(argp, &s, sizeof s)) |
1642 | r = -EFAULT; | |
3a4d5c94 | 1643 | break; |
3a4d5c94 | 1644 | case VHOST_SET_VRING_KICK: |
7ad9c9d2 TY |
1645 | if (copy_from_user(&f, argp, sizeof f)) { |
1646 | r = -EFAULT; | |
3a4d5c94 | 1647 | break; |
7ad9c9d2 | 1648 | } |
e0136c16 | 1649 | eventfp = f.fd == VHOST_FILE_UNBIND ? NULL : eventfd_fget(f.fd); |
535297a6 MT |
1650 | if (IS_ERR(eventfp)) { |
1651 | r = PTR_ERR(eventfp); | |
1652 | break; | |
1653 | } | |
3a4d5c94 | 1654 | if (eventfp != vq->kick) { |
cecb46f1 AV |
1655 | pollstop = (filep = vq->kick) != NULL; |
1656 | pollstart = (vq->kick = eventfp) != NULL; | |
3a4d5c94 MT |
1657 | } else |
1658 | filep = eventfp; | |
1659 | break; | |
1660 | case VHOST_SET_VRING_CALL: | |
7ad9c9d2 TY |
1661 | if (copy_from_user(&f, argp, sizeof f)) { |
1662 | r = -EFAULT; | |
3a4d5c94 | 1663 | break; |
7ad9c9d2 | 1664 | } |
e0136c16 | 1665 | ctx = f.fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(f.fd); |
e050c7d9 EB |
1666 | if (IS_ERR(ctx)) { |
1667 | r = PTR_ERR(ctx); | |
535297a6 MT |
1668 | break; |
1669 | } | |
265a0ad8 | 1670 | |
265a0ad8 | 1671 | swap(ctx, vq->call_ctx.ctx); |
3a4d5c94 MT |
1672 | break; |
1673 | case VHOST_SET_VRING_ERR: | |
7ad9c9d2 TY |
1674 | if (copy_from_user(&f, argp, sizeof f)) { |
1675 | r = -EFAULT; | |
3a4d5c94 | 1676 | break; |
7ad9c9d2 | 1677 | } |
e0136c16 | 1678 | ctx = f.fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(f.fd); |
09f332a5 EB |
1679 | if (IS_ERR(ctx)) { |
1680 | r = PTR_ERR(ctx); | |
535297a6 MT |
1681 | break; |
1682 | } | |
09f332a5 | 1683 | swap(ctx, vq->error_ctx); |
3a4d5c94 | 1684 | break; |
2751c988 GK |
1685 | case VHOST_SET_VRING_ENDIAN: |
1686 | r = vhost_set_vring_endian(vq, argp); | |
1687 | break; | |
1688 | case VHOST_GET_VRING_ENDIAN: | |
1689 | r = vhost_get_vring_endian(vq, idx, argp); | |
1690 | break; | |
03088137 JW |
1691 | case VHOST_SET_VRING_BUSYLOOP_TIMEOUT: |
1692 | if (copy_from_user(&s, argp, sizeof(s))) { | |
1693 | r = -EFAULT; | |
1694 | break; | |
1695 | } | |
1696 | vq->busyloop_timeout = s.num; | |
1697 | break; | |
1698 | case VHOST_GET_VRING_BUSYLOOP_TIMEOUT: | |
1699 | s.index = idx; | |
1700 | s.num = vq->busyloop_timeout; | |
1701 | if (copy_to_user(argp, &s, sizeof(s))) | |
1702 | r = -EFAULT; | |
1703 | break; | |
3a4d5c94 MT |
1704 | default: |
1705 | r = -ENOIOCTLCMD; | |
1706 | } | |
1707 | ||
1708 | if (pollstop && vq->handle_kick) | |
1709 | vhost_poll_stop(&vq->poll); | |
1710 | ||
e050c7d9 | 1711 | if (!IS_ERR_OR_NULL(ctx)) |
3a4d5c94 MT |
1712 | eventfd_ctx_put(ctx); |
1713 | if (filep) | |
1714 | fput(filep); | |
1715 | ||
1716 | if (pollstart && vq->handle_kick) | |
2b8b328b | 1717 | r = vhost_poll_start(&vq->poll, vq->kick); |
3a4d5c94 MT |
1718 | |
1719 | mutex_unlock(&vq->mutex); | |
1720 | ||
1721 | if (pollstop && vq->handle_kick) | |
b2ffa407 | 1722 | vhost_dev_flush(vq->poll.dev); |
3a4d5c94 MT |
1723 | return r; |
1724 | } | |
6ac1afbf | 1725 | EXPORT_SYMBOL_GPL(vhost_vring_ioctl); |
3a4d5c94 | 1726 | |
759aba1e | 1727 | int vhost_init_device_iotlb(struct vhost_dev *d) |
6b1e6cc7 | 1728 | { |
0bbe3066 | 1729 | struct vhost_iotlb *niotlb, *oiotlb; |
6b1e6cc7 JW |
1730 | int i; |
1731 | ||
0bbe3066 | 1732 | niotlb = iotlb_alloc(); |
6b1e6cc7 JW |
1733 | if (!niotlb) |
1734 | return -ENOMEM; | |
1735 | ||
1736 | oiotlb = d->iotlb; | |
1737 | d->iotlb = niotlb; | |
1738 | ||
1739 | for (i = 0; i < d->nvqs; ++i) { | |
b13f9c63 JW |
1740 | struct vhost_virtqueue *vq = d->vqs[i]; |
1741 | ||
1742 | mutex_lock(&vq->mutex); | |
1743 | vq->iotlb = niotlb; | |
1744 | __vhost_vq_meta_reset(vq); | |
1745 | mutex_unlock(&vq->mutex); | |
6b1e6cc7 JW |
1746 | } |
1747 | ||
0bbe3066 | 1748 | vhost_iotlb_free(oiotlb); |
6b1e6cc7 JW |
1749 | |
1750 | return 0; | |
1751 | } | |
1752 | EXPORT_SYMBOL_GPL(vhost_init_device_iotlb); | |
1753 | ||
3a4d5c94 | 1754 | /* Caller must have device mutex */ |
935cdee7 | 1755 | long vhost_dev_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp) |
3a4d5c94 | 1756 | { |
d25cc43c | 1757 | struct eventfd_ctx *ctx; |
3a4d5c94 MT |
1758 | u64 p; |
1759 | long r; | |
1760 | int i, fd; | |
1761 | ||
1762 | /* If you are not the owner, you can become one */ | |
1763 | if (ioctl == VHOST_SET_OWNER) { | |
1764 | r = vhost_dev_set_owner(d); | |
1765 | goto done; | |
1766 | } | |
1767 | ||
1768 | /* You must be the owner to do anything else */ | |
1769 | r = vhost_dev_check_owner(d); | |
1770 | if (r) | |
1771 | goto done; | |
1772 | ||
1773 | switch (ioctl) { | |
1774 | case VHOST_SET_MEM_TABLE: | |
1775 | r = vhost_set_memory(d, argp); | |
1776 | break; | |
1777 | case VHOST_SET_LOG_BASE: | |
7ad9c9d2 TY |
1778 | if (copy_from_user(&p, argp, sizeof p)) { |
1779 | r = -EFAULT; | |
3a4d5c94 | 1780 | break; |
7ad9c9d2 | 1781 | } |
3a4d5c94 MT |
1782 | if ((u64)(unsigned long)p != p) { |
1783 | r = -EFAULT; | |
1784 | break; | |
1785 | } | |
1786 | for (i = 0; i < d->nvqs; ++i) { | |
1787 | struct vhost_virtqueue *vq; | |
1788 | void __user *base = (void __user *)(unsigned long)p; | |
3ab2e420 | 1789 | vq = d->vqs[i]; |
3a4d5c94 MT |
1790 | mutex_lock(&vq->mutex); |
1791 | /* If ring is inactive, will check when it's enabled. */ | |
ea16c514 | 1792 | if (vq->private_data && !vq_log_access_ok(vq, base)) |
3a4d5c94 MT |
1793 | r = -EFAULT; |
1794 | else | |
1795 | vq->log_base = base; | |
1796 | mutex_unlock(&vq->mutex); | |
1797 | } | |
1798 | break; | |
1799 | case VHOST_SET_LOG_FD: | |
1800 | r = get_user(fd, (int __user *)argp); | |
1801 | if (r < 0) | |
1802 | break; | |
e0136c16 | 1803 | ctx = fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(fd); |
d25cc43c EB |
1804 | if (IS_ERR(ctx)) { |
1805 | r = PTR_ERR(ctx); | |
3a4d5c94 MT |
1806 | break; |
1807 | } | |
d25cc43c | 1808 | swap(ctx, d->log_ctx); |
3a4d5c94 | 1809 | for (i = 0; i < d->nvqs; ++i) { |
3ab2e420 AH |
1810 | mutex_lock(&d->vqs[i]->mutex); |
1811 | d->vqs[i]->log_ctx = d->log_ctx; | |
1812 | mutex_unlock(&d->vqs[i]->mutex); | |
3a4d5c94 MT |
1813 | } |
1814 | if (ctx) | |
1815 | eventfd_ctx_put(ctx); | |
3a4d5c94 MT |
1816 | break; |
1817 | default: | |
935cdee7 | 1818 | r = -ENOIOCTLCMD; |
3a4d5c94 MT |
1819 | break; |
1820 | } | |
1821 | done: | |
1822 | return r; | |
1823 | } | |
6ac1afbf | 1824 | EXPORT_SYMBOL_GPL(vhost_dev_ioctl); |
3a4d5c94 | 1825 | |
3a4d5c94 MT |
1826 | /* TODO: This is really inefficient. We need something like get_user() |
1827 | * (instruction directly accesses the data, with an exception table entry | |
ff61f079 | 1828 | * returning -EFAULT). See Documentation/arch/x86/exception-tables.rst. |
3a4d5c94 MT |
1829 | */ |
1830 | static int set_bit_to_user(int nr, void __user *addr) | |
1831 | { | |
1832 | unsigned long log = (unsigned long)addr; | |
1833 | struct page *page; | |
1834 | void *base; | |
1835 | int bit = nr + (log % PAGE_SIZE) * 8; | |
1836 | int r; | |
d47effe1 | 1837 | |
690623e1 | 1838 | r = pin_user_pages_fast(log, 1, FOLL_WRITE, &page); |
d6db3f5c | 1839 | if (r < 0) |
3a4d5c94 | 1840 | return r; |
d6db3f5c | 1841 | BUG_ON(r != 1); |
c6daa7ff | 1842 | base = kmap_atomic(page); |
3a4d5c94 | 1843 | set_bit(bit, base); |
c6daa7ff | 1844 | kunmap_atomic(base); |
690623e1 | 1845 | unpin_user_pages_dirty_lock(&page, 1, true); |
3a4d5c94 MT |
1846 | return 0; |
1847 | } | |
1848 | ||
1849 | static int log_write(void __user *log_base, | |
1850 | u64 write_address, u64 write_length) | |
1851 | { | |
28831ee6 | 1852 | u64 write_page = write_address / VHOST_PAGE_SIZE; |
3a4d5c94 | 1853 | int r; |
d47effe1 | 1854 | |
3a4d5c94 MT |
1855 | if (!write_length) |
1856 | return 0; | |
3bf9be40 | 1857 | write_length += write_address % VHOST_PAGE_SIZE; |
3a4d5c94 MT |
1858 | for (;;) { |
1859 | u64 base = (u64)(unsigned long)log_base; | |
28831ee6 MT |
1860 | u64 log = base + write_page / 8; |
1861 | int bit = write_page % 8; | |
3a4d5c94 MT |
1862 | if ((u64)(unsigned long)log != log) |
1863 | return -EFAULT; | |
1864 | r = set_bit_to_user(bit, (void __user *)(unsigned long)log); | |
1865 | if (r < 0) | |
1866 | return r; | |
1867 | if (write_length <= VHOST_PAGE_SIZE) | |
1868 | break; | |
1869 | write_length -= VHOST_PAGE_SIZE; | |
28831ee6 | 1870 | write_page += 1; |
3a4d5c94 MT |
1871 | } |
1872 | return r; | |
1873 | } | |
1874 | ||
cc5e7107 JW |
1875 | static int log_write_hva(struct vhost_virtqueue *vq, u64 hva, u64 len) |
1876 | { | |
0bbe3066 JW |
1877 | struct vhost_iotlb *umem = vq->umem; |
1878 | struct vhost_iotlb_map *u; | |
cc5e7107 JW |
1879 | u64 start, end, l, min; |
1880 | int r; | |
1881 | bool hit = false; | |
1882 | ||
1883 | while (len) { | |
1884 | min = len; | |
1885 | /* More than one GPAs can be mapped into a single HVA. So | |
1886 | * iterate all possible umems here to be safe. | |
1887 | */ | |
0bbe3066 JW |
1888 | list_for_each_entry(u, &umem->list, link) { |
1889 | if (u->addr > hva - 1 + len || | |
1890 | u->addr - 1 + u->size < hva) | |
cc5e7107 | 1891 | continue; |
0bbe3066 JW |
1892 | start = max(u->addr, hva); |
1893 | end = min(u->addr - 1 + u->size, hva - 1 + len); | |
cc5e7107 JW |
1894 | l = end - start + 1; |
1895 | r = log_write(vq->log_base, | |
0bbe3066 | 1896 | u->start + start - u->addr, |
cc5e7107 JW |
1897 | l); |
1898 | if (r < 0) | |
1899 | return r; | |
1900 | hit = true; | |
1901 | min = min(l, min); | |
1902 | } | |
1903 | ||
1904 | if (!hit) | |
1905 | return -EFAULT; | |
1906 | ||
1907 | len -= min; | |
1908 | hva += min; | |
1909 | } | |
1910 | ||
1911 | return 0; | |
1912 | } | |
1913 | ||
1914 | static int log_used(struct vhost_virtqueue *vq, u64 used_offset, u64 len) | |
1915 | { | |
5e5e8736 | 1916 | struct iovec *iov = vq->log_iov; |
cc5e7107 JW |
1917 | int i, ret; |
1918 | ||
1919 | if (!vq->iotlb) | |
1920 | return log_write(vq->log_base, vq->log_addr + used_offset, len); | |
1921 | ||
1922 | ret = translate_desc(vq, (uintptr_t)vq->used + used_offset, | |
1923 | len, iov, 64, VHOST_ACCESS_WO); | |
816db766 | 1924 | if (ret < 0) |
cc5e7107 JW |
1925 | return ret; |
1926 | ||
1927 | for (i = 0; i < ret; i++) { | |
1928 | ret = log_write_hva(vq, (uintptr_t)iov[i].iov_base, | |
1929 | iov[i].iov_len); | |
1930 | if (ret) | |
1931 | return ret; | |
1932 | } | |
1933 | ||
1934 | return 0; | |
1935 | } | |
1936 | ||
3a4d5c94 | 1937 | int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log, |
cc5e7107 | 1938 | unsigned int log_num, u64 len, struct iovec *iov, int count) |
3a4d5c94 MT |
1939 | { |
1940 | int i, r; | |
1941 | ||
1942 | /* Make sure data written is seen before log. */ | |
5659338c | 1943 | smp_wmb(); |
cc5e7107 JW |
1944 | |
1945 | if (vq->iotlb) { | |
1946 | for (i = 0; i < count; i++) { | |
1947 | r = log_write_hva(vq, (uintptr_t)iov[i].iov_base, | |
1948 | iov[i].iov_len); | |
1949 | if (r < 0) | |
1950 | return r; | |
1951 | } | |
1952 | return 0; | |
1953 | } | |
1954 | ||
3a4d5c94 MT |
1955 | for (i = 0; i < log_num; ++i) { |
1956 | u64 l = min(log[i].len, len); | |
1957 | r = log_write(vq->log_base, log[i].addr, l); | |
1958 | if (r < 0) | |
1959 | return r; | |
1960 | len -= l; | |
5786aee8 MT |
1961 | if (!len) { |
1962 | if (vq->log_ctx) | |
1963 | eventfd_signal(vq->log_ctx, 1); | |
3a4d5c94 | 1964 | return 0; |
5786aee8 | 1965 | } |
3a4d5c94 | 1966 | } |
3a4d5c94 MT |
1967 | /* Length written exceeds what we have stored. This is a bug. */ |
1968 | BUG(); | |
1969 | return 0; | |
1970 | } | |
6ac1afbf | 1971 | EXPORT_SYMBOL_GPL(vhost_log_write); |
3a4d5c94 | 1972 | |
2723feaa JW |
1973 | static int vhost_update_used_flags(struct vhost_virtqueue *vq) |
1974 | { | |
1975 | void __user *used; | |
7b5d753e | 1976 | if (vhost_put_used_flags(vq)) |
2723feaa JW |
1977 | return -EFAULT; |
1978 | if (unlikely(vq->log_used)) { | |
1979 | /* Make sure the flag is seen before log. */ | |
1980 | smp_wmb(); | |
1981 | /* Log used flag write. */ | |
1982 | used = &vq->used->flags; | |
cc5e7107 JW |
1983 | log_used(vq, (used - (void __user *)vq->used), |
1984 | sizeof vq->used->flags); | |
2723feaa JW |
1985 | if (vq->log_ctx) |
1986 | eventfd_signal(vq->log_ctx, 1); | |
1987 | } | |
1988 | return 0; | |
1989 | } | |
1990 | ||
4c809363 | 1991 | static int vhost_update_avail_event(struct vhost_virtqueue *vq) |
2723feaa | 1992 | { |
7b5d753e | 1993 | if (vhost_put_avail_event(vq)) |
2723feaa JW |
1994 | return -EFAULT; |
1995 | if (unlikely(vq->log_used)) { | |
1996 | void __user *used; | |
1997 | /* Make sure the event is seen before log. */ | |
1998 | smp_wmb(); | |
1999 | /* Log avail event write */ | |
2000 | used = vhost_avail_event(vq); | |
cc5e7107 JW |
2001 | log_used(vq, (used - (void __user *)vq->used), |
2002 | sizeof *vhost_avail_event(vq)); | |
2723feaa JW |
2003 | if (vq->log_ctx) |
2004 | eventfd_signal(vq->log_ctx, 1); | |
2005 | } | |
2006 | return 0; | |
2007 | } | |
2008 | ||
80f7d030 | 2009 | int vhost_vq_init_access(struct vhost_virtqueue *vq) |
2723feaa | 2010 | { |
3b1bbe89 | 2011 | __virtio16 last_used_idx; |
2723feaa | 2012 | int r; |
e1f33be9 GK |
2013 | bool is_le = vq->is_le; |
2014 | ||
cda8bba0 | 2015 | if (!vq->private_data) |
2723feaa | 2016 | return 0; |
2751c988 GK |
2017 | |
2018 | vhost_init_is_le(vq); | |
2723feaa JW |
2019 | |
2020 | r = vhost_update_used_flags(vq); | |
2021 | if (r) | |
e1f33be9 | 2022 | goto err; |
2723feaa | 2023 | vq->signalled_used_valid = false; |
6b1e6cc7 | 2024 | if (!vq->iotlb && |
96d4f267 | 2025 | !access_ok(&vq->used->idx, sizeof vq->used->idx)) { |
e1f33be9 GK |
2026 | r = -EFAULT; |
2027 | goto err; | |
2028 | } | |
7b5d753e | 2029 | r = vhost_get_used_idx(vq, &last_used_idx); |
6b1e6cc7 JW |
2030 | if (r) { |
2031 | vq_err(vq, "Can't access used idx at %p\n", | |
2032 | &vq->used->idx); | |
e1f33be9 | 2033 | goto err; |
6b1e6cc7 | 2034 | } |
3b1bbe89 | 2035 | vq->last_used_idx = vhost16_to_cpu(vq, last_used_idx); |
64f7f051 | 2036 | return 0; |
6b1e6cc7 | 2037 | |
e1f33be9 GK |
2038 | err: |
2039 | vq->is_le = is_le; | |
2040 | return r; | |
2723feaa | 2041 | } |
80f7d030 | 2042 | EXPORT_SYMBOL_GPL(vhost_vq_init_access); |
2723feaa | 2043 | |
47283bef | 2044 | static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len, |
6b1e6cc7 | 2045 | struct iovec iov[], int iov_size, int access) |
3a4d5c94 | 2046 | { |
0bbe3066 | 2047 | const struct vhost_iotlb_map *map; |
6b1e6cc7 | 2048 | struct vhost_dev *dev = vq->dev; |
0bbe3066 | 2049 | struct vhost_iotlb *umem = dev->iotlb ? dev->iotlb : dev->umem; |
3a4d5c94 | 2050 | struct iovec *_iov; |
98047313 | 2051 | u64 s = 0, last = addr + len - 1; |
3a4d5c94 MT |
2052 | int ret = 0; |
2053 | ||
3a4d5c94 MT |
2054 | while ((u64)len > s) { |
2055 | u64 size; | |
7b3384fc | 2056 | if (unlikely(ret >= iov_size)) { |
3a4d5c94 MT |
2057 | ret = -ENOBUFS; |
2058 | break; | |
2059 | } | |
6b1e6cc7 | 2060 | |
98047313 | 2061 | map = vhost_iotlb_itree_first(umem, addr, last); |
0bbe3066 | 2062 | if (map == NULL || map->start > addr) { |
6b1e6cc7 JW |
2063 | if (umem != dev->iotlb) { |
2064 | ret = -EFAULT; | |
2065 | break; | |
2066 | } | |
2067 | ret = -EAGAIN; | |
2068 | break; | |
0bbe3066 | 2069 | } else if (!(map->perm & access)) { |
6b1e6cc7 | 2070 | ret = -EPERM; |
3a4d5c94 MT |
2071 | break; |
2072 | } | |
6b1e6cc7 | 2073 | |
3a4d5c94 | 2074 | _iov = iov + ret; |
0bbe3066 | 2075 | size = map->size - addr + map->start; |
bd97120f | 2076 | _iov->iov_len = min((u64)len - s, size); |
0d4a3f2a | 2077 | _iov->iov_base = (void __user *)(unsigned long) |
0bbe3066 | 2078 | (map->addr + addr - map->start); |
3a4d5c94 MT |
2079 | s += size; |
2080 | addr += size; | |
2081 | ++ret; | |
2082 | } | |
2083 | ||
6b1e6cc7 JW |
2084 | if (ret == -EAGAIN) |
2085 | vhost_iotlb_miss(vq, addr, access); | |
3a4d5c94 MT |
2086 | return ret; |
2087 | } | |
2088 | ||
2089 | /* Each buffer in the virtqueues is actually a chain of descriptors. This | |
2090 | * function returns the next descriptor in the chain, | |
2091 | * or -1U if we're at the end. */ | |
3b1bbe89 | 2092 | static unsigned next_desc(struct vhost_virtqueue *vq, struct vring_desc *desc) |
3a4d5c94 MT |
2093 | { |
2094 | unsigned int next; | |
2095 | ||
2096 | /* If this descriptor says it doesn't chain, we're done. */ | |
3b1bbe89 | 2097 | if (!(desc->flags & cpu_to_vhost16(vq, VRING_DESC_F_NEXT))) |
3a4d5c94 MT |
2098 | return -1U; |
2099 | ||
2100 | /* Check they're not leading us off end of descriptors. */ | |
3a5db0b1 | 2101 | next = vhost16_to_cpu(vq, READ_ONCE(desc->next)); |
3a4d5c94 MT |
2102 | return next; |
2103 | } | |
2104 | ||
47283bef | 2105 | static int get_indirect(struct vhost_virtqueue *vq, |
7b3384fc MT |
2106 | struct iovec iov[], unsigned int iov_size, |
2107 | unsigned int *out_num, unsigned int *in_num, | |
2108 | struct vhost_log *log, unsigned int *log_num, | |
2109 | struct vring_desc *indirect) | |
3a4d5c94 MT |
2110 | { |
2111 | struct vring_desc desc; | |
2112 | unsigned int i = 0, count, found = 0; | |
3b1bbe89 | 2113 | u32 len = vhost32_to_cpu(vq, indirect->len); |
aad9a1ce | 2114 | struct iov_iter from; |
6b1e6cc7 | 2115 | int ret, access; |
3a4d5c94 MT |
2116 | |
2117 | /* Sanity check */ | |
3b1bbe89 | 2118 | if (unlikely(len % sizeof desc)) { |
3a4d5c94 MT |
2119 | vq_err(vq, "Invalid length in indirect descriptor: " |
2120 | "len 0x%llx not multiple of 0x%zx\n", | |
3b1bbe89 | 2121 | (unsigned long long)len, |
3a4d5c94 MT |
2122 | sizeof desc); |
2123 | return -EINVAL; | |
2124 | } | |
2125 | ||
3b1bbe89 | 2126 | ret = translate_desc(vq, vhost64_to_cpu(vq, indirect->addr), len, vq->indirect, |
6b1e6cc7 | 2127 | UIO_MAXIOV, VHOST_ACCESS_RO); |
7b3384fc | 2128 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
2129 | if (ret != -EAGAIN) |
2130 | vq_err(vq, "Translation failure %d in indirect.\n", ret); | |
3a4d5c94 MT |
2131 | return ret; |
2132 | } | |
de4eda9d | 2133 | iov_iter_init(&from, ITER_SOURCE, vq->indirect, ret, len); |
3b1bbe89 | 2134 | count = len / sizeof desc; |
3a4d5c94 MT |
2135 | /* Buffers are chained via a 16 bit next field, so |
2136 | * we can have at most 2^16 of these. */ | |
7b3384fc | 2137 | if (unlikely(count > USHRT_MAX + 1)) { |
3a4d5c94 MT |
2138 | vq_err(vq, "Indirect buffer length too big: %d\n", |
2139 | indirect->len); | |
2140 | return -E2BIG; | |
2141 | } | |
2142 | ||
2143 | do { | |
2144 | unsigned iov_count = *in_num + *out_num; | |
7b3384fc | 2145 | if (unlikely(++found > count)) { |
3a4d5c94 MT |
2146 | vq_err(vq, "Loop detected: last one at %u " |
2147 | "indirect size %u\n", | |
2148 | i, count); | |
2149 | return -EINVAL; | |
2150 | } | |
cbbd26b8 | 2151 | if (unlikely(!copy_from_iter_full(&desc, sizeof(desc), &from))) { |
3a4d5c94 | 2152 | vq_err(vq, "Failed indirect descriptor: idx %d, %zx\n", |
3b1bbe89 | 2153 | i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc); |
3a4d5c94 MT |
2154 | return -EINVAL; |
2155 | } | |
3b1bbe89 | 2156 | if (unlikely(desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT))) { |
3a4d5c94 | 2157 | vq_err(vq, "Nested indirect descriptor: idx %d, %zx\n", |
3b1bbe89 | 2158 | i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc); |
3a4d5c94 MT |
2159 | return -EINVAL; |
2160 | } | |
2161 | ||
6b1e6cc7 JW |
2162 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE)) |
2163 | access = VHOST_ACCESS_WO; | |
2164 | else | |
2165 | access = VHOST_ACCESS_RO; | |
2166 | ||
3b1bbe89 MT |
2167 | ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr), |
2168 | vhost32_to_cpu(vq, desc.len), iov + iov_count, | |
6b1e6cc7 | 2169 | iov_size - iov_count, access); |
7b3384fc | 2170 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
2171 | if (ret != -EAGAIN) |
2172 | vq_err(vq, "Translation failure %d indirect idx %d\n", | |
2173 | ret, i); | |
3a4d5c94 MT |
2174 | return ret; |
2175 | } | |
2176 | /* If this is an input descriptor, increment that count. */ | |
6b1e6cc7 | 2177 | if (access == VHOST_ACCESS_WO) { |
3a4d5c94 | 2178 | *in_num += ret; |
060423bf | 2179 | if (unlikely(log && ret)) { |
3b1bbe89 MT |
2180 | log[*log_num].addr = vhost64_to_cpu(vq, desc.addr); |
2181 | log[*log_num].len = vhost32_to_cpu(vq, desc.len); | |
3a4d5c94 MT |
2182 | ++*log_num; |
2183 | } | |
2184 | } else { | |
2185 | /* If it's an output descriptor, they're all supposed | |
2186 | * to come before any input descriptors. */ | |
7b3384fc | 2187 | if (unlikely(*in_num)) { |
3a4d5c94 MT |
2188 | vq_err(vq, "Indirect descriptor " |
2189 | "has out after in: idx %d\n", i); | |
2190 | return -EINVAL; | |
2191 | } | |
2192 | *out_num += ret; | |
2193 | } | |
3b1bbe89 | 2194 | } while ((i = next_desc(vq, &desc)) != -1); |
3a4d5c94 MT |
2195 | return 0; |
2196 | } | |
2197 | ||
2198 | /* This looks in the virtqueue and for the first available buffer, and converts | |
2199 | * it to an iovec for convenient access. Since descriptors consist of some | |
2200 | * number of output then some number of input descriptors, it's actually two | |
2201 | * iovecs, but we pack them into one and note how many of each there were. | |
2202 | * | |
d5675bd2 MT |
2203 | * This function returns the descriptor number found, or vq->num (which is |
2204 | * never a valid descriptor number) if none was found. A negative code is | |
2205 | * returned on error. */ | |
47283bef | 2206 | int vhost_get_vq_desc(struct vhost_virtqueue *vq, |
d5675bd2 MT |
2207 | struct iovec iov[], unsigned int iov_size, |
2208 | unsigned int *out_num, unsigned int *in_num, | |
2209 | struct vhost_log *log, unsigned int *log_num) | |
3a4d5c94 MT |
2210 | { |
2211 | struct vring_desc desc; | |
2212 | unsigned int i, head, found = 0; | |
2213 | u16 last_avail_idx; | |
3b1bbe89 MT |
2214 | __virtio16 avail_idx; |
2215 | __virtio16 ring_head; | |
6b1e6cc7 | 2216 | int ret, access; |
3a4d5c94 MT |
2217 | |
2218 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
2219 | last_avail_idx = vq->last_avail_idx; | |
3a4d5c94 | 2220 | |
e3b56cdd | 2221 | if (vq->avail_idx == vq->last_avail_idx) { |
7b5d753e | 2222 | if (unlikely(vhost_get_avail_idx(vq, &avail_idx))) { |
e3b56cdd JW |
2223 | vq_err(vq, "Failed to access avail idx at %p\n", |
2224 | &vq->avail->idx); | |
2225 | return -EFAULT; | |
2226 | } | |
2227 | vq->avail_idx = vhost16_to_cpu(vq, avail_idx); | |
3a4d5c94 | 2228 | |
e3b56cdd JW |
2229 | if (unlikely((u16)(vq->avail_idx - last_avail_idx) > vq->num)) { |
2230 | vq_err(vq, "Guest moved used index from %u to %u", | |
2231 | last_avail_idx, vq->avail_idx); | |
2232 | return -EFAULT; | |
2233 | } | |
2234 | ||
2235 | /* If there's nothing new since last we looked, return | |
2236 | * invalid. | |
2237 | */ | |
2238 | if (vq->avail_idx == last_avail_idx) | |
2239 | return vq->num; | |
3a4d5c94 | 2240 | |
e3b56cdd JW |
2241 | /* Only get avail ring entries after they have been |
2242 | * exposed by guest. | |
2243 | */ | |
2244 | smp_rmb(); | |
2245 | } | |
3a4d5c94 MT |
2246 | |
2247 | /* Grab the next descriptor number they're advertising, and increment | |
2248 | * the index we've seen. */ | |
7b5d753e | 2249 | if (unlikely(vhost_get_avail_head(vq, &ring_head, last_avail_idx))) { |
3a4d5c94 MT |
2250 | vq_err(vq, "Failed to read head: idx %d address %p\n", |
2251 | last_avail_idx, | |
2252 | &vq->avail->ring[last_avail_idx % vq->num]); | |
d5675bd2 | 2253 | return -EFAULT; |
3a4d5c94 MT |
2254 | } |
2255 | ||
3b1bbe89 MT |
2256 | head = vhost16_to_cpu(vq, ring_head); |
2257 | ||
3a4d5c94 | 2258 | /* If their number is silly, that's an error. */ |
7b3384fc | 2259 | if (unlikely(head >= vq->num)) { |
3a4d5c94 MT |
2260 | vq_err(vq, "Guest says index %u > %u is available", |
2261 | head, vq->num); | |
d5675bd2 | 2262 | return -EINVAL; |
3a4d5c94 MT |
2263 | } |
2264 | ||
2265 | /* When we start there are none of either input nor output. */ | |
2266 | *out_num = *in_num = 0; | |
2267 | if (unlikely(log)) | |
2268 | *log_num = 0; | |
2269 | ||
2270 | i = head; | |
2271 | do { | |
2272 | unsigned iov_count = *in_num + *out_num; | |
7b3384fc | 2273 | if (unlikely(i >= vq->num)) { |
3a4d5c94 MT |
2274 | vq_err(vq, "Desc index is %u > %u, head = %u", |
2275 | i, vq->num, head); | |
d5675bd2 | 2276 | return -EINVAL; |
3a4d5c94 | 2277 | } |
7b3384fc | 2278 | if (unlikely(++found > vq->num)) { |
3a4d5c94 MT |
2279 | vq_err(vq, "Loop detected: last one at %u " |
2280 | "vq size %u head %u\n", | |
2281 | i, vq->num, head); | |
d5675bd2 | 2282 | return -EINVAL; |
3a4d5c94 | 2283 | } |
7b5d753e | 2284 | ret = vhost_get_desc(vq, &desc, i); |
7b3384fc | 2285 | if (unlikely(ret)) { |
3a4d5c94 MT |
2286 | vq_err(vq, "Failed to get descriptor: idx %d addr %p\n", |
2287 | i, vq->desc + i); | |
d5675bd2 | 2288 | return -EFAULT; |
3a4d5c94 | 2289 | } |
3b1bbe89 | 2290 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT)) { |
47283bef | 2291 | ret = get_indirect(vq, iov, iov_size, |
3a4d5c94 MT |
2292 | out_num, in_num, |
2293 | log, log_num, &desc); | |
7b3384fc | 2294 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
2295 | if (ret != -EAGAIN) |
2296 | vq_err(vq, "Failure detected " | |
2297 | "in indirect descriptor at idx %d\n", i); | |
d5675bd2 | 2298 | return ret; |
3a4d5c94 MT |
2299 | } |
2300 | continue; | |
2301 | } | |
2302 | ||
6b1e6cc7 JW |
2303 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE)) |
2304 | access = VHOST_ACCESS_WO; | |
2305 | else | |
2306 | access = VHOST_ACCESS_RO; | |
3b1bbe89 MT |
2307 | ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr), |
2308 | vhost32_to_cpu(vq, desc.len), iov + iov_count, | |
6b1e6cc7 | 2309 | iov_size - iov_count, access); |
7b3384fc | 2310 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
2311 | if (ret != -EAGAIN) |
2312 | vq_err(vq, "Translation failure %d descriptor idx %d\n", | |
2313 | ret, i); | |
d5675bd2 | 2314 | return ret; |
3a4d5c94 | 2315 | } |
6b1e6cc7 | 2316 | if (access == VHOST_ACCESS_WO) { |
3a4d5c94 MT |
2317 | /* If this is an input descriptor, |
2318 | * increment that count. */ | |
2319 | *in_num += ret; | |
060423bf | 2320 | if (unlikely(log && ret)) { |
3b1bbe89 MT |
2321 | log[*log_num].addr = vhost64_to_cpu(vq, desc.addr); |
2322 | log[*log_num].len = vhost32_to_cpu(vq, desc.len); | |
3a4d5c94 MT |
2323 | ++*log_num; |
2324 | } | |
2325 | } else { | |
2326 | /* If it's an output descriptor, they're all supposed | |
2327 | * to come before any input descriptors. */ | |
7b3384fc | 2328 | if (unlikely(*in_num)) { |
3a4d5c94 MT |
2329 | vq_err(vq, "Descriptor has out after in: " |
2330 | "idx %d\n", i); | |
d5675bd2 | 2331 | return -EINVAL; |
3a4d5c94 MT |
2332 | } |
2333 | *out_num += ret; | |
2334 | } | |
3b1bbe89 | 2335 | } while ((i = next_desc(vq, &desc)) != -1); |
3a4d5c94 MT |
2336 | |
2337 | /* On success, increment avail index. */ | |
2338 | vq->last_avail_idx++; | |
8ea8cf89 MT |
2339 | |
2340 | /* Assume notifications from guest are disabled at this point, | |
2341 | * if they aren't we would need to update avail_event index. */ | |
2342 | BUG_ON(!(vq->used_flags & VRING_USED_F_NO_NOTIFY)); | |
3a4d5c94 MT |
2343 | return head; |
2344 | } | |
6ac1afbf | 2345 | EXPORT_SYMBOL_GPL(vhost_get_vq_desc); |
3a4d5c94 MT |
2346 | |
2347 | /* Reverse the effect of vhost_get_vq_desc. Useful for error handling. */ | |
8dd014ad | 2348 | void vhost_discard_vq_desc(struct vhost_virtqueue *vq, int n) |
3a4d5c94 | 2349 | { |
8dd014ad | 2350 | vq->last_avail_idx -= n; |
3a4d5c94 | 2351 | } |
6ac1afbf | 2352 | EXPORT_SYMBOL_GPL(vhost_discard_vq_desc); |
3a4d5c94 MT |
2353 | |
2354 | /* After we've used one of their buffers, we tell them about it. We'll then | |
2355 | * want to notify the guest, using eventfd. */ | |
2356 | int vhost_add_used(struct vhost_virtqueue *vq, unsigned int head, int len) | |
2357 | { | |
3b1bbe89 MT |
2358 | struct vring_used_elem heads = { |
2359 | cpu_to_vhost32(vq, head), | |
2360 | cpu_to_vhost32(vq, len) | |
2361 | }; | |
3a4d5c94 | 2362 | |
c49e4e57 | 2363 | return vhost_add_used_n(vq, &heads, 1); |
3a4d5c94 | 2364 | } |
6ac1afbf | 2365 | EXPORT_SYMBOL_GPL(vhost_add_used); |
3a4d5c94 | 2366 | |
8dd014ad DS |
2367 | static int __vhost_add_used_n(struct vhost_virtqueue *vq, |
2368 | struct vring_used_elem *heads, | |
2369 | unsigned count) | |
2370 | { | |
a865e420 | 2371 | vring_used_elem_t __user *used; |
8ea8cf89 | 2372 | u16 old, new; |
8dd014ad DS |
2373 | int start; |
2374 | ||
5fba13b5 | 2375 | start = vq->last_used_idx & (vq->num - 1); |
8dd014ad | 2376 | used = vq->used->ring + start; |
7b5d753e | 2377 | if (vhost_put_used(vq, heads, start, count)) { |
8dd014ad DS |
2378 | vq_err(vq, "Failed to write used"); |
2379 | return -EFAULT; | |
2380 | } | |
2381 | if (unlikely(vq->log_used)) { | |
2382 | /* Make sure data is seen before log. */ | |
2383 | smp_wmb(); | |
2384 | /* Log used ring entry write. */ | |
cc5e7107 JW |
2385 | log_used(vq, ((void __user *)used - (void __user *)vq->used), |
2386 | count * sizeof *used); | |
8dd014ad | 2387 | } |
8ea8cf89 MT |
2388 | old = vq->last_used_idx; |
2389 | new = (vq->last_used_idx += count); | |
2390 | /* If the driver never bothers to signal in a very long while, | |
2391 | * used index might wrap around. If that happens, invalidate | |
2392 | * signalled_used index we stored. TODO: make sure driver | |
2393 | * signals at least once in 2^16 and remove this. */ | |
2394 | if (unlikely((u16)(new - vq->signalled_used) < (u16)(new - old))) | |
2395 | vq->signalled_used_valid = false; | |
8dd014ad DS |
2396 | return 0; |
2397 | } | |
2398 | ||
2399 | /* After we've used one of their buffers, we tell them about it. We'll then | |
2400 | * want to notify the guest, using eventfd. */ | |
2401 | int vhost_add_used_n(struct vhost_virtqueue *vq, struct vring_used_elem *heads, | |
2402 | unsigned count) | |
2403 | { | |
2404 | int start, n, r; | |
2405 | ||
5fba13b5 | 2406 | start = vq->last_used_idx & (vq->num - 1); |
8dd014ad DS |
2407 | n = vq->num - start; |
2408 | if (n < count) { | |
2409 | r = __vhost_add_used_n(vq, heads, n); | |
2410 | if (r < 0) | |
2411 | return r; | |
2412 | heads += n; | |
2413 | count -= n; | |
2414 | } | |
2415 | r = __vhost_add_used_n(vq, heads, count); | |
2416 | ||
2417 | /* Make sure buffer is written before we update index. */ | |
2418 | smp_wmb(); | |
7b5d753e | 2419 | if (vhost_put_used_idx(vq)) { |
8dd014ad DS |
2420 | vq_err(vq, "Failed to increment used idx"); |
2421 | return -EFAULT; | |
2422 | } | |
2423 | if (unlikely(vq->log_used)) { | |
841df922 JW |
2424 | /* Make sure used idx is seen before log. */ |
2425 | smp_wmb(); | |
8dd014ad | 2426 | /* Log used index update. */ |
cc5e7107 JW |
2427 | log_used(vq, offsetof(struct vring_used, idx), |
2428 | sizeof vq->used->idx); | |
8dd014ad DS |
2429 | if (vq->log_ctx) |
2430 | eventfd_signal(vq->log_ctx, 1); | |
2431 | } | |
2432 | return r; | |
2433 | } | |
6ac1afbf | 2434 | EXPORT_SYMBOL_GPL(vhost_add_used_n); |
8dd014ad | 2435 | |
8ea8cf89 | 2436 | static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 | 2437 | { |
3b1bbe89 MT |
2438 | __u16 old, new; |
2439 | __virtio16 event; | |
8ea8cf89 | 2440 | bool v; |
8d65843c JW |
2441 | /* Flush out used index updates. This is paired |
2442 | * with the barrier that the Guest executes when enabling | |
2443 | * interrupts. */ | |
2444 | smp_mb(); | |
0d499356 | 2445 | |
ea16c514 | 2446 | if (vhost_has_feature(vq, VIRTIO_F_NOTIFY_ON_EMPTY) && |
8ea8cf89 MT |
2447 | unlikely(vq->avail_idx == vq->last_avail_idx)) |
2448 | return true; | |
2449 | ||
ea16c514 | 2450 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
3b1bbe89 | 2451 | __virtio16 flags; |
7b5d753e | 2452 | if (vhost_get_avail_flags(vq, &flags)) { |
8ea8cf89 MT |
2453 | vq_err(vq, "Failed to get flags"); |
2454 | return true; | |
2455 | } | |
3b1bbe89 | 2456 | return !(flags & cpu_to_vhost16(vq, VRING_AVAIL_F_NO_INTERRUPT)); |
3a4d5c94 | 2457 | } |
8ea8cf89 MT |
2458 | old = vq->signalled_used; |
2459 | v = vq->signalled_used_valid; | |
2460 | new = vq->signalled_used = vq->last_used_idx; | |
2461 | vq->signalled_used_valid = true; | |
3a4d5c94 | 2462 | |
8ea8cf89 MT |
2463 | if (unlikely(!v)) |
2464 | return true; | |
3a4d5c94 | 2465 | |
7b5d753e | 2466 | if (vhost_get_used_event(vq, &event)) { |
8ea8cf89 MT |
2467 | vq_err(vq, "Failed to get used event idx"); |
2468 | return true; | |
2469 | } | |
8d65843c | 2470 | return vring_need_event(vhost16_to_cpu(vq, event), new, old); |
8ea8cf89 MT |
2471 | } |
2472 | ||
2473 | /* This actually signals the guest, using eventfd. */ | |
2474 | void vhost_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq) | |
2475 | { | |
3a4d5c94 | 2476 | /* Signal the Guest tell them we used something up. */ |
265a0ad8 ZL |
2477 | if (vq->call_ctx.ctx && vhost_notify(dev, vq)) |
2478 | eventfd_signal(vq->call_ctx.ctx, 1); | |
3a4d5c94 | 2479 | } |
6ac1afbf | 2480 | EXPORT_SYMBOL_GPL(vhost_signal); |
3a4d5c94 MT |
2481 | |
2482 | /* And here's the combo meal deal. Supersize me! */ | |
2483 | void vhost_add_used_and_signal(struct vhost_dev *dev, | |
2484 | struct vhost_virtqueue *vq, | |
2485 | unsigned int head, int len) | |
2486 | { | |
2487 | vhost_add_used(vq, head, len); | |
2488 | vhost_signal(dev, vq); | |
2489 | } | |
6ac1afbf | 2490 | EXPORT_SYMBOL_GPL(vhost_add_used_and_signal); |
3a4d5c94 | 2491 | |
8dd014ad DS |
2492 | /* multi-buffer version of vhost_add_used_and_signal */ |
2493 | void vhost_add_used_and_signal_n(struct vhost_dev *dev, | |
2494 | struct vhost_virtqueue *vq, | |
2495 | struct vring_used_elem *heads, unsigned count) | |
2496 | { | |
2497 | vhost_add_used_n(vq, heads, count); | |
2498 | vhost_signal(dev, vq); | |
2499 | } | |
6ac1afbf | 2500 | EXPORT_SYMBOL_GPL(vhost_add_used_and_signal_n); |
8dd014ad | 2501 | |
d4a60603 JW |
2502 | /* return true if we're sure that avaiable ring is empty */ |
2503 | bool vhost_vq_avail_empty(struct vhost_dev *dev, struct vhost_virtqueue *vq) | |
2504 | { | |
2505 | __virtio16 avail_idx; | |
2506 | int r; | |
2507 | ||
275bf960 JW |
2508 | if (vq->avail_idx != vq->last_avail_idx) |
2509 | return false; | |
2510 | ||
7b5d753e | 2511 | r = vhost_get_avail_idx(vq, &avail_idx); |
275bf960 | 2512 | if (unlikely(r)) |
d4a60603 | 2513 | return false; |
275bf960 | 2514 | vq->avail_idx = vhost16_to_cpu(vq, avail_idx); |
d4a60603 | 2515 | |
275bf960 | 2516 | return vq->avail_idx == vq->last_avail_idx; |
d4a60603 JW |
2517 | } |
2518 | EXPORT_SYMBOL_GPL(vhost_vq_avail_empty); | |
2519 | ||
3a4d5c94 | 2520 | /* OK, now we need to know about added descriptors. */ |
8ea8cf89 | 2521 | bool vhost_enable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 | 2522 | { |
3b1bbe89 | 2523 | __virtio16 avail_idx; |
3a4d5c94 | 2524 | int r; |
d47effe1 | 2525 | |
3a4d5c94 MT |
2526 | if (!(vq->used_flags & VRING_USED_F_NO_NOTIFY)) |
2527 | return false; | |
2528 | vq->used_flags &= ~VRING_USED_F_NO_NOTIFY; | |
ea16c514 | 2529 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
2723feaa | 2530 | r = vhost_update_used_flags(vq); |
8ea8cf89 MT |
2531 | if (r) { |
2532 | vq_err(vq, "Failed to enable notification at %p: %d\n", | |
2533 | &vq->used->flags, r); | |
2534 | return false; | |
2535 | } | |
2536 | } else { | |
4c809363 | 2537 | r = vhost_update_avail_event(vq); |
8ea8cf89 MT |
2538 | if (r) { |
2539 | vq_err(vq, "Failed to update avail event index at %p: %d\n", | |
2540 | vhost_avail_event(vq), r); | |
2541 | return false; | |
2542 | } | |
2543 | } | |
3a4d5c94 MT |
2544 | /* They could have slipped one in as we were doing that: make |
2545 | * sure it's written, then check again. */ | |
5659338c | 2546 | smp_mb(); |
7b5d753e | 2547 | r = vhost_get_avail_idx(vq, &avail_idx); |
3a4d5c94 MT |
2548 | if (r) { |
2549 | vq_err(vq, "Failed to check avail idx at %p: %d\n", | |
2550 | &vq->avail->idx, r); | |
2551 | return false; | |
2552 | } | |
d3bb267b | 2553 | vq->avail_idx = vhost16_to_cpu(vq, avail_idx); |
3a4d5c94 | 2554 | |
d3bb267b | 2555 | return vq->avail_idx != vq->last_avail_idx; |
3a4d5c94 | 2556 | } |
6ac1afbf | 2557 | EXPORT_SYMBOL_GPL(vhost_enable_notify); |
3a4d5c94 MT |
2558 | |
2559 | /* We don't need to be notified again. */ | |
8ea8cf89 | 2560 | void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 MT |
2561 | { |
2562 | int r; | |
d47effe1 | 2563 | |
3a4d5c94 MT |
2564 | if (vq->used_flags & VRING_USED_F_NO_NOTIFY) |
2565 | return; | |
2566 | vq->used_flags |= VRING_USED_F_NO_NOTIFY; | |
ea16c514 | 2567 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
2723feaa | 2568 | r = vhost_update_used_flags(vq); |
8ea8cf89 | 2569 | if (r) |
ae6961de | 2570 | vq_err(vq, "Failed to disable notification at %p: %d\n", |
8ea8cf89 MT |
2571 | &vq->used->flags, r); |
2572 | } | |
3a4d5c94 | 2573 | } |
6ac1afbf AH |
2574 | EXPORT_SYMBOL_GPL(vhost_disable_notify); |
2575 | ||
6b1e6cc7 JW |
2576 | /* Create a new message. */ |
2577 | struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type) | |
2578 | { | |
2579 | struct vhost_msg_node *node = kmalloc(sizeof *node, GFP_KERNEL); | |
2580 | if (!node) | |
2581 | return NULL; | |
670ae9ca MT |
2582 | |
2583 | /* Make sure all padding within the structure is initialized. */ | |
2584 | memset(&node->msg, 0, sizeof node->msg); | |
6b1e6cc7 JW |
2585 | node->vq = vq; |
2586 | node->msg.type = type; | |
2587 | return node; | |
2588 | } | |
2589 | EXPORT_SYMBOL_GPL(vhost_new_msg); | |
2590 | ||
2591 | void vhost_enqueue_msg(struct vhost_dev *dev, struct list_head *head, | |
2592 | struct vhost_msg_node *node) | |
2593 | { | |
2594 | spin_lock(&dev->iotlb_lock); | |
2595 | list_add_tail(&node->node, head); | |
2596 | spin_unlock(&dev->iotlb_lock); | |
2597 | ||
a9a08845 | 2598 | wake_up_interruptible_poll(&dev->wait, EPOLLIN | EPOLLRDNORM); |
6b1e6cc7 JW |
2599 | } |
2600 | EXPORT_SYMBOL_GPL(vhost_enqueue_msg); | |
2601 | ||
2602 | struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev, | |
2603 | struct list_head *head) | |
2604 | { | |
2605 | struct vhost_msg_node *node = NULL; | |
2606 | ||
2607 | spin_lock(&dev->iotlb_lock); | |
2608 | if (!list_empty(head)) { | |
2609 | node = list_first_entry(head, struct vhost_msg_node, | |
2610 | node); | |
2611 | list_del(&node->node); | |
2612 | } | |
2613 | spin_unlock(&dev->iotlb_lock); | |
2614 | ||
2615 | return node; | |
2616 | } | |
2617 | EXPORT_SYMBOL_GPL(vhost_dequeue_msg); | |
2618 | ||
460f7ce1 JW |
2619 | void vhost_set_backend_features(struct vhost_dev *dev, u64 features) |
2620 | { | |
2621 | struct vhost_virtqueue *vq; | |
2622 | int i; | |
2623 | ||
2624 | mutex_lock(&dev->mutex); | |
2625 | for (i = 0; i < dev->nvqs; ++i) { | |
2626 | vq = dev->vqs[i]; | |
2627 | mutex_lock(&vq->mutex); | |
2628 | vq->acked_backend_features = features; | |
2629 | mutex_unlock(&vq->mutex); | |
2630 | } | |
2631 | mutex_unlock(&dev->mutex); | |
2632 | } | |
2633 | EXPORT_SYMBOL_GPL(vhost_set_backend_features); | |
6b1e6cc7 | 2634 | |
6ac1afbf AH |
2635 | static int __init vhost_init(void) |
2636 | { | |
2637 | return 0; | |
2638 | } | |
2639 | ||
2640 | static void __exit vhost_exit(void) | |
2641 | { | |
2642 | } | |
2643 | ||
2644 | module_init(vhost_init); | |
2645 | module_exit(vhost_exit); | |
2646 | ||
2647 | MODULE_VERSION("0.0.1"); | |
2648 | MODULE_LICENSE("GPL v2"); | |
2649 | MODULE_AUTHOR("Michael S. Tsirkin"); | |
2650 | MODULE_DESCRIPTION("Host kernel accelerator for virtio"); |