Commit | Line | Data |
---|---|---|
3a4d5c94 MT |
1 | /* Copyright (C) 2009 Red Hat, Inc. |
2 | * Copyright (C) 2006 Rusty Russell IBM Corporation | |
3 | * | |
4 | * Author: Michael S. Tsirkin <mst@redhat.com> | |
5 | * | |
6 | * Inspiration, some code, and most witty comments come from | |
61516587 | 7 | * Documentation/virtual/lguest/lguest.c, by Rusty Russell |
3a4d5c94 MT |
8 | * |
9 | * This work is licensed under the terms of the GNU GPL, version 2. | |
10 | * | |
11 | * Generic code for virtio server in host kernel. | |
12 | */ | |
13 | ||
14 | #include <linux/eventfd.h> | |
15 | #include <linux/vhost.h> | |
35596b27 | 16 | #include <linux/uio.h> |
3a4d5c94 | 17 | #include <linux/mm.h> |
64e1c807 | 18 | #include <linux/mmu_context.h> |
3a4d5c94 MT |
19 | #include <linux/miscdevice.h> |
20 | #include <linux/mutex.h> | |
3a4d5c94 MT |
21 | #include <linux/poll.h> |
22 | #include <linux/file.h> | |
23 | #include <linux/highmem.h> | |
5a0e3ad6 | 24 | #include <linux/slab.h> |
4de7255f | 25 | #include <linux/vmalloc.h> |
c23f3445 | 26 | #include <linux/kthread.h> |
9e3d1957 | 27 | #include <linux/cgroup.h> |
6ac1afbf | 28 | #include <linux/module.h> |
bcfeacab | 29 | #include <linux/sort.h> |
3a4d5c94 | 30 | |
3a4d5c94 MT |
31 | #include "vhost.h" |
32 | ||
c9ce42f7 IM |
33 | static ushort max_mem_regions = 64; |
34 | module_param(max_mem_regions, ushort, 0444); | |
35 | MODULE_PARM_DESC(max_mem_regions, | |
36 | "Maximum number of memory regions in memory map. (default: 64)"); | |
37 | ||
3a4d5c94 | 38 | enum { |
3a4d5c94 MT |
39 | VHOST_MEMORY_F_LOG = 0x1, |
40 | }; | |
41 | ||
3b1bbe89 MT |
42 | #define vhost_used_event(vq) ((__virtio16 __user *)&vq->avail->ring[vq->num]) |
43 | #define vhost_avail_event(vq) ((__virtio16 __user *)&vq->used->ring[vq->num]) | |
8ea8cf89 | 44 | |
2751c988 | 45 | #ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY |
c5072037 | 46 | static void vhost_disable_cross_endian(struct vhost_virtqueue *vq) |
2751c988 GK |
47 | { |
48 | vq->user_be = !virtio_legacy_is_little_endian(); | |
49 | } | |
50 | ||
c5072037 GK |
51 | static void vhost_enable_cross_endian_big(struct vhost_virtqueue *vq) |
52 | { | |
53 | vq->user_be = true; | |
54 | } | |
55 | ||
56 | static void vhost_enable_cross_endian_little(struct vhost_virtqueue *vq) | |
57 | { | |
58 | vq->user_be = false; | |
59 | } | |
60 | ||
2751c988 GK |
61 | static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp) |
62 | { | |
63 | struct vhost_vring_state s; | |
64 | ||
65 | if (vq->private_data) | |
66 | return -EBUSY; | |
67 | ||
68 | if (copy_from_user(&s, argp, sizeof(s))) | |
69 | return -EFAULT; | |
70 | ||
71 | if (s.num != VHOST_VRING_LITTLE_ENDIAN && | |
72 | s.num != VHOST_VRING_BIG_ENDIAN) | |
73 | return -EINVAL; | |
74 | ||
c5072037 GK |
75 | if (s.num == VHOST_VRING_BIG_ENDIAN) |
76 | vhost_enable_cross_endian_big(vq); | |
77 | else | |
78 | vhost_enable_cross_endian_little(vq); | |
2751c988 GK |
79 | |
80 | return 0; | |
81 | } | |
82 | ||
83 | static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx, | |
84 | int __user *argp) | |
85 | { | |
86 | struct vhost_vring_state s = { | |
87 | .index = idx, | |
88 | .num = vq->user_be | |
89 | }; | |
90 | ||
91 | if (copy_to_user(argp, &s, sizeof(s))) | |
92 | return -EFAULT; | |
93 | ||
94 | return 0; | |
95 | } | |
96 | ||
97 | static void vhost_init_is_le(struct vhost_virtqueue *vq) | |
98 | { | |
99 | /* Note for legacy virtio: user_be is initialized at reset time | |
100 | * according to the host endianness. If userspace does not set an | |
101 | * explicit endianness, the default behavior is native endian, as | |
102 | * expected by legacy virtio. | |
103 | */ | |
104 | vq->is_le = vhost_has_feature(vq, VIRTIO_F_VERSION_1) || !vq->user_be; | |
105 | } | |
106 | #else | |
c5072037 | 107 | static void vhost_disable_cross_endian(struct vhost_virtqueue *vq) |
2751c988 GK |
108 | { |
109 | } | |
110 | ||
111 | static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp) | |
112 | { | |
113 | return -ENOIOCTLCMD; | |
114 | } | |
115 | ||
116 | static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx, | |
117 | int __user *argp) | |
118 | { | |
119 | return -ENOIOCTLCMD; | |
120 | } | |
121 | ||
122 | static void vhost_init_is_le(struct vhost_virtqueue *vq) | |
123 | { | |
124 | if (vhost_has_feature(vq, VIRTIO_F_VERSION_1)) | |
125 | vq->is_le = true; | |
126 | } | |
127 | #endif /* CONFIG_VHOST_CROSS_ENDIAN_LEGACY */ | |
128 | ||
c5072037 GK |
129 | static void vhost_reset_is_le(struct vhost_virtqueue *vq) |
130 | { | |
131 | vq->is_le = virtio_legacy_is_little_endian(); | |
132 | } | |
133 | ||
7235acdb JW |
134 | struct vhost_flush_struct { |
135 | struct vhost_work work; | |
136 | struct completion wait_event; | |
137 | }; | |
138 | ||
139 | static void vhost_flush_work(struct vhost_work *work) | |
140 | { | |
141 | struct vhost_flush_struct *s; | |
142 | ||
143 | s = container_of(work, struct vhost_flush_struct, work); | |
144 | complete(&s->wait_event); | |
145 | } | |
146 | ||
3a4d5c94 MT |
147 | static void vhost_poll_func(struct file *file, wait_queue_head_t *wqh, |
148 | poll_table *pt) | |
149 | { | |
150 | struct vhost_poll *poll; | |
3a4d5c94 | 151 | |
d47effe1 | 152 | poll = container_of(pt, struct vhost_poll, table); |
3a4d5c94 MT |
153 | poll->wqh = wqh; |
154 | add_wait_queue(wqh, &poll->wait); | |
155 | } | |
156 | ||
157 | static int vhost_poll_wakeup(wait_queue_t *wait, unsigned mode, int sync, | |
158 | void *key) | |
159 | { | |
c23f3445 TH |
160 | struct vhost_poll *poll = container_of(wait, struct vhost_poll, wait); |
161 | ||
3a4d5c94 MT |
162 | if (!((unsigned long)key & poll->mask)) |
163 | return 0; | |
164 | ||
c23f3445 | 165 | vhost_poll_queue(poll); |
3a4d5c94 MT |
166 | return 0; |
167 | } | |
168 | ||
163049ae | 169 | void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn) |
87d6a412 MT |
170 | { |
171 | INIT_LIST_HEAD(&work->node); | |
172 | work->fn = fn; | |
173 | init_waitqueue_head(&work->done); | |
87d6a412 | 174 | } |
6ac1afbf | 175 | EXPORT_SYMBOL_GPL(vhost_work_init); |
87d6a412 | 176 | |
3a4d5c94 | 177 | /* Init poll structure */ |
c23f3445 TH |
178 | void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn, |
179 | unsigned long mask, struct vhost_dev *dev) | |
3a4d5c94 | 180 | { |
3a4d5c94 MT |
181 | init_waitqueue_func_entry(&poll->wait, vhost_poll_wakeup); |
182 | init_poll_funcptr(&poll->table, vhost_poll_func); | |
183 | poll->mask = mask; | |
c23f3445 | 184 | poll->dev = dev; |
2b8b328b | 185 | poll->wqh = NULL; |
c23f3445 | 186 | |
87d6a412 | 187 | vhost_work_init(&poll->work, fn); |
3a4d5c94 | 188 | } |
6ac1afbf | 189 | EXPORT_SYMBOL_GPL(vhost_poll_init); |
3a4d5c94 MT |
190 | |
191 | /* Start polling a file. We add ourselves to file's wait queue. The caller must | |
192 | * keep a reference to a file until after vhost_poll_stop is called. */ | |
2b8b328b | 193 | int vhost_poll_start(struct vhost_poll *poll, struct file *file) |
3a4d5c94 MT |
194 | { |
195 | unsigned long mask; | |
2b8b328b | 196 | int ret = 0; |
d47effe1 | 197 | |
70181d51 JW |
198 | if (poll->wqh) |
199 | return 0; | |
200 | ||
3a4d5c94 MT |
201 | mask = file->f_op->poll(file, &poll->table); |
202 | if (mask) | |
203 | vhost_poll_wakeup(&poll->wait, 0, 0, (void *)mask); | |
2b8b328b JW |
204 | if (mask & POLLERR) { |
205 | if (poll->wqh) | |
206 | remove_wait_queue(poll->wqh, &poll->wait); | |
207 | ret = -EINVAL; | |
208 | } | |
209 | ||
210 | return ret; | |
3a4d5c94 | 211 | } |
6ac1afbf | 212 | EXPORT_SYMBOL_GPL(vhost_poll_start); |
3a4d5c94 MT |
213 | |
214 | /* Stop polling a file. After this function returns, it becomes safe to drop the | |
215 | * file reference. You must also flush afterwards. */ | |
216 | void vhost_poll_stop(struct vhost_poll *poll) | |
217 | { | |
2b8b328b JW |
218 | if (poll->wqh) { |
219 | remove_wait_queue(poll->wqh, &poll->wait); | |
220 | poll->wqh = NULL; | |
221 | } | |
3a4d5c94 | 222 | } |
6ac1afbf | 223 | EXPORT_SYMBOL_GPL(vhost_poll_stop); |
3a4d5c94 | 224 | |
7235acdb | 225 | void vhost_work_flush(struct vhost_dev *dev, struct vhost_work *work) |
0174b0c3 | 226 | { |
7235acdb | 227 | struct vhost_flush_struct flush; |
d47effe1 | 228 | |
7235acdb JW |
229 | if (dev->worker) { |
230 | init_completion(&flush.wait_event); | |
231 | vhost_work_init(&flush.work, vhost_flush_work); | |
0174b0c3 | 232 | |
7235acdb JW |
233 | vhost_work_queue(dev, &flush.work); |
234 | wait_for_completion(&flush.wait_event); | |
235 | } | |
3a4d5c94 | 236 | } |
6ac1afbf | 237 | EXPORT_SYMBOL_GPL(vhost_work_flush); |
3a4d5c94 | 238 | |
87d6a412 MT |
239 | /* Flush any work that has been scheduled. When calling this, don't hold any |
240 | * locks that are also used by the callback. */ | |
241 | void vhost_poll_flush(struct vhost_poll *poll) | |
242 | { | |
243 | vhost_work_flush(poll->dev, &poll->work); | |
244 | } | |
6ac1afbf | 245 | EXPORT_SYMBOL_GPL(vhost_poll_flush); |
87d6a412 | 246 | |
163049ae | 247 | void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work) |
3a4d5c94 | 248 | { |
c23f3445 TH |
249 | unsigned long flags; |
250 | ||
251 | spin_lock_irqsave(&dev->work_lock, flags); | |
252 | if (list_empty(&work->node)) { | |
253 | list_add_tail(&work->node, &dev->work_list); | |
ac9fde24 | 254 | spin_unlock_irqrestore(&dev->work_lock, flags); |
c23f3445 | 255 | wake_up_process(dev->worker); |
ac9fde24 QC |
256 | } else { |
257 | spin_unlock_irqrestore(&dev->work_lock, flags); | |
c23f3445 | 258 | } |
3a4d5c94 | 259 | } |
6ac1afbf | 260 | EXPORT_SYMBOL_GPL(vhost_work_queue); |
3a4d5c94 | 261 | |
526d3e7f JW |
262 | /* A lockless hint for busy polling code to exit the loop */ |
263 | bool vhost_has_work(struct vhost_dev *dev) | |
264 | { | |
265 | return !list_empty(&dev->work_list); | |
266 | } | |
267 | EXPORT_SYMBOL_GPL(vhost_has_work); | |
268 | ||
87d6a412 MT |
269 | void vhost_poll_queue(struct vhost_poll *poll) |
270 | { | |
271 | vhost_work_queue(poll->dev, &poll->work); | |
272 | } | |
6ac1afbf | 273 | EXPORT_SYMBOL_GPL(vhost_poll_queue); |
87d6a412 | 274 | |
3a4d5c94 MT |
275 | static void vhost_vq_reset(struct vhost_dev *dev, |
276 | struct vhost_virtqueue *vq) | |
277 | { | |
278 | vq->num = 1; | |
279 | vq->desc = NULL; | |
280 | vq->avail = NULL; | |
281 | vq->used = NULL; | |
282 | vq->last_avail_idx = 0; | |
283 | vq->avail_idx = 0; | |
284 | vq->last_used_idx = 0; | |
8ea8cf89 MT |
285 | vq->signalled_used = 0; |
286 | vq->signalled_used_valid = false; | |
3a4d5c94 | 287 | vq->used_flags = 0; |
3a4d5c94 MT |
288 | vq->log_used = false; |
289 | vq->log_addr = -1ull; | |
3a4d5c94 | 290 | vq->private_data = NULL; |
ea16c514 | 291 | vq->acked_features = 0; |
3a4d5c94 MT |
292 | vq->log_base = NULL; |
293 | vq->error_ctx = NULL; | |
294 | vq->error = NULL; | |
295 | vq->kick = NULL; | |
296 | vq->call_ctx = NULL; | |
297 | vq->call = NULL; | |
73a99f08 | 298 | vq->log_ctx = NULL; |
47283bef | 299 | vq->memory = NULL; |
c5072037 GK |
300 | vhost_reset_is_le(vq); |
301 | vhost_disable_cross_endian(vq); | |
03088137 | 302 | vq->busyloop_timeout = 0; |
3a4d5c94 MT |
303 | } |
304 | ||
c23f3445 TH |
305 | static int vhost_worker(void *data) |
306 | { | |
307 | struct vhost_dev *dev = data; | |
308 | struct vhost_work *work = NULL; | |
d7ffde35 | 309 | mm_segment_t oldfs = get_fs(); |
c23f3445 | 310 | |
d7ffde35 | 311 | set_fs(USER_DS); |
64e1c807 MT |
312 | use_mm(dev->mm); |
313 | ||
c23f3445 TH |
314 | for (;;) { |
315 | /* mb paired w/ kthread_stop */ | |
316 | set_current_state(TASK_INTERRUPTIBLE); | |
317 | ||
318 | spin_lock_irq(&dev->work_lock); | |
c23f3445 TH |
319 | |
320 | if (kthread_should_stop()) { | |
321 | spin_unlock_irq(&dev->work_lock); | |
322 | __set_current_state(TASK_RUNNING); | |
64e1c807 | 323 | break; |
c23f3445 TH |
324 | } |
325 | if (!list_empty(&dev->work_list)) { | |
326 | work = list_first_entry(&dev->work_list, | |
327 | struct vhost_work, node); | |
328 | list_del_init(&work->node); | |
c23f3445 TH |
329 | } else |
330 | work = NULL; | |
331 | spin_unlock_irq(&dev->work_lock); | |
332 | ||
333 | if (work) { | |
334 | __set_current_state(TASK_RUNNING); | |
335 | work->fn(work); | |
d550dda1 NHE |
336 | if (need_resched()) |
337 | schedule(); | |
c23f3445 TH |
338 | } else |
339 | schedule(); | |
340 | ||
341 | } | |
64e1c807 | 342 | unuse_mm(dev->mm); |
d7ffde35 | 343 | set_fs(oldfs); |
64e1c807 | 344 | return 0; |
c23f3445 TH |
345 | } |
346 | ||
bab632d6 MT |
347 | static void vhost_vq_free_iovecs(struct vhost_virtqueue *vq) |
348 | { | |
349 | kfree(vq->indirect); | |
350 | vq->indirect = NULL; | |
351 | kfree(vq->log); | |
352 | vq->log = NULL; | |
353 | kfree(vq->heads); | |
354 | vq->heads = NULL; | |
bab632d6 MT |
355 | } |
356 | ||
e0e9b406 JW |
357 | /* Helper to allocate iovec buffers for all vqs. */ |
358 | static long vhost_dev_alloc_iovecs(struct vhost_dev *dev) | |
359 | { | |
6d5e6aa8 | 360 | struct vhost_virtqueue *vq; |
e0e9b406 | 361 | int i; |
d47effe1 | 362 | |
e0e9b406 | 363 | for (i = 0; i < dev->nvqs; ++i) { |
6d5e6aa8 AH |
364 | vq = dev->vqs[i]; |
365 | vq->indirect = kmalloc(sizeof *vq->indirect * UIO_MAXIOV, | |
366 | GFP_KERNEL); | |
367 | vq->log = kmalloc(sizeof *vq->log * UIO_MAXIOV, GFP_KERNEL); | |
368 | vq->heads = kmalloc(sizeof *vq->heads * UIO_MAXIOV, GFP_KERNEL); | |
369 | if (!vq->indirect || !vq->log || !vq->heads) | |
e0e9b406 JW |
370 | goto err_nomem; |
371 | } | |
372 | return 0; | |
d47effe1 | 373 | |
e0e9b406 | 374 | err_nomem: |
bab632d6 | 375 | for (; i >= 0; --i) |
3ab2e420 | 376 | vhost_vq_free_iovecs(dev->vqs[i]); |
e0e9b406 JW |
377 | return -ENOMEM; |
378 | } | |
379 | ||
380 | static void vhost_dev_free_iovecs(struct vhost_dev *dev) | |
381 | { | |
382 | int i; | |
d47effe1 | 383 | |
bab632d6 | 384 | for (i = 0; i < dev->nvqs; ++i) |
3ab2e420 | 385 | vhost_vq_free_iovecs(dev->vqs[i]); |
e0e9b406 JW |
386 | } |
387 | ||
59566b6e | 388 | void vhost_dev_init(struct vhost_dev *dev, |
3ab2e420 | 389 | struct vhost_virtqueue **vqs, int nvqs) |
3a4d5c94 | 390 | { |
6d5e6aa8 | 391 | struct vhost_virtqueue *vq; |
3a4d5c94 | 392 | int i; |
c23f3445 | 393 | |
3a4d5c94 MT |
394 | dev->vqs = vqs; |
395 | dev->nvqs = nvqs; | |
396 | mutex_init(&dev->mutex); | |
397 | dev->log_ctx = NULL; | |
398 | dev->log_file = NULL; | |
399 | dev->memory = NULL; | |
400 | dev->mm = NULL; | |
c23f3445 TH |
401 | spin_lock_init(&dev->work_lock); |
402 | INIT_LIST_HEAD(&dev->work_list); | |
403 | dev->worker = NULL; | |
3a4d5c94 MT |
404 | |
405 | for (i = 0; i < dev->nvqs; ++i) { | |
6d5e6aa8 AH |
406 | vq = dev->vqs[i]; |
407 | vq->log = NULL; | |
408 | vq->indirect = NULL; | |
409 | vq->heads = NULL; | |
410 | vq->dev = dev; | |
411 | mutex_init(&vq->mutex); | |
412 | vhost_vq_reset(dev, vq); | |
413 | if (vq->handle_kick) | |
414 | vhost_poll_init(&vq->poll, vq->handle_kick, | |
415 | POLLIN, dev); | |
3a4d5c94 | 416 | } |
3a4d5c94 | 417 | } |
6ac1afbf | 418 | EXPORT_SYMBOL_GPL(vhost_dev_init); |
3a4d5c94 MT |
419 | |
420 | /* Caller should have device mutex */ | |
421 | long vhost_dev_check_owner(struct vhost_dev *dev) | |
422 | { | |
423 | /* Are you the owner? If not, I don't think you mean to do that */ | |
424 | return dev->mm == current->mm ? 0 : -EPERM; | |
425 | } | |
6ac1afbf | 426 | EXPORT_SYMBOL_GPL(vhost_dev_check_owner); |
3a4d5c94 | 427 | |
87d6a412 | 428 | struct vhost_attach_cgroups_struct { |
d47effe1 KK |
429 | struct vhost_work work; |
430 | struct task_struct *owner; | |
431 | int ret; | |
87d6a412 MT |
432 | }; |
433 | ||
434 | static void vhost_attach_cgroups_work(struct vhost_work *work) | |
435 | { | |
d47effe1 KK |
436 | struct vhost_attach_cgroups_struct *s; |
437 | ||
438 | s = container_of(work, struct vhost_attach_cgroups_struct, work); | |
439 | s->ret = cgroup_attach_task_all(s->owner, current); | |
87d6a412 MT |
440 | } |
441 | ||
442 | static int vhost_attach_cgroups(struct vhost_dev *dev) | |
443 | { | |
d47effe1 KK |
444 | struct vhost_attach_cgroups_struct attach; |
445 | ||
446 | attach.owner = current; | |
447 | vhost_work_init(&attach.work, vhost_attach_cgroups_work); | |
448 | vhost_work_queue(dev, &attach.work); | |
449 | vhost_work_flush(dev, &attach.work); | |
450 | return attach.ret; | |
87d6a412 MT |
451 | } |
452 | ||
05c05351 MT |
453 | /* Caller should have device mutex */ |
454 | bool vhost_dev_has_owner(struct vhost_dev *dev) | |
455 | { | |
456 | return dev->mm; | |
457 | } | |
6ac1afbf | 458 | EXPORT_SYMBOL_GPL(vhost_dev_has_owner); |
05c05351 | 459 | |
3a4d5c94 | 460 | /* Caller should have device mutex */ |
54db63c2 | 461 | long vhost_dev_set_owner(struct vhost_dev *dev) |
3a4d5c94 | 462 | { |
c23f3445 TH |
463 | struct task_struct *worker; |
464 | int err; | |
d47effe1 | 465 | |
3a4d5c94 | 466 | /* Is there an owner already? */ |
05c05351 | 467 | if (vhost_dev_has_owner(dev)) { |
c23f3445 TH |
468 | err = -EBUSY; |
469 | goto err_mm; | |
470 | } | |
d47effe1 | 471 | |
3a4d5c94 MT |
472 | /* No owner, become one */ |
473 | dev->mm = get_task_mm(current); | |
c23f3445 TH |
474 | worker = kthread_create(vhost_worker, dev, "vhost-%d", current->pid); |
475 | if (IS_ERR(worker)) { | |
476 | err = PTR_ERR(worker); | |
477 | goto err_worker; | |
478 | } | |
479 | ||
480 | dev->worker = worker; | |
87d6a412 MT |
481 | wake_up_process(worker); /* avoid contributing to loadavg */ |
482 | ||
483 | err = vhost_attach_cgroups(dev); | |
9e3d1957 MT |
484 | if (err) |
485 | goto err_cgroup; | |
c23f3445 | 486 | |
e0e9b406 JW |
487 | err = vhost_dev_alloc_iovecs(dev); |
488 | if (err) | |
489 | goto err_cgroup; | |
490 | ||
3a4d5c94 | 491 | return 0; |
9e3d1957 MT |
492 | err_cgroup: |
493 | kthread_stop(worker); | |
615cc221 | 494 | dev->worker = NULL; |
c23f3445 TH |
495 | err_worker: |
496 | if (dev->mm) | |
497 | mmput(dev->mm); | |
498 | dev->mm = NULL; | |
499 | err_mm: | |
500 | return err; | |
3a4d5c94 | 501 | } |
6ac1afbf | 502 | EXPORT_SYMBOL_GPL(vhost_dev_set_owner); |
3a4d5c94 | 503 | |
150b9e51 | 504 | struct vhost_memory *vhost_dev_reset_owner_prepare(void) |
3a4d5c94 | 505 | { |
150b9e51 MT |
506 | return kmalloc(offsetof(struct vhost_memory, regions), GFP_KERNEL); |
507 | } | |
6ac1afbf | 508 | EXPORT_SYMBOL_GPL(vhost_dev_reset_owner_prepare); |
3a4d5c94 | 509 | |
150b9e51 MT |
510 | /* Caller should have device mutex */ |
511 | void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_memory *memory) | |
512 | { | |
47283bef MT |
513 | int i; |
514 | ||
ea5d4046 | 515 | vhost_dev_cleanup(dev, true); |
3a4d5c94 | 516 | |
150b9e51 | 517 | /* Restore memory to default empty mapping. */ |
3a4d5c94 | 518 | memory->nregions = 0; |
47283bef MT |
519 | dev->memory = memory; |
520 | /* We don't need VQ locks below since vhost_dev_cleanup makes sure | |
521 | * VQs aren't running. | |
522 | */ | |
523 | for (i = 0; i < dev->nvqs; ++i) | |
524 | dev->vqs[i]->memory = memory; | |
3a4d5c94 | 525 | } |
6ac1afbf | 526 | EXPORT_SYMBOL_GPL(vhost_dev_reset_owner); |
3a4d5c94 | 527 | |
b211616d | 528 | void vhost_dev_stop(struct vhost_dev *dev) |
bab632d6 MT |
529 | { |
530 | int i; | |
b211616d MT |
531 | |
532 | for (i = 0; i < dev->nvqs; ++i) { | |
3ab2e420 AH |
533 | if (dev->vqs[i]->kick && dev->vqs[i]->handle_kick) { |
534 | vhost_poll_stop(&dev->vqs[i]->poll); | |
535 | vhost_poll_flush(&dev->vqs[i]->poll); | |
b211616d | 536 | } |
bab632d6 | 537 | } |
bab632d6 | 538 | } |
6ac1afbf | 539 | EXPORT_SYMBOL_GPL(vhost_dev_stop); |
bab632d6 | 540 | |
ea5d4046 MT |
541 | /* Caller should have device mutex if and only if locked is set */ |
542 | void vhost_dev_cleanup(struct vhost_dev *dev, bool locked) | |
3a4d5c94 MT |
543 | { |
544 | int i; | |
d47effe1 | 545 | |
3a4d5c94 | 546 | for (i = 0; i < dev->nvqs; ++i) { |
3ab2e420 AH |
547 | if (dev->vqs[i]->error_ctx) |
548 | eventfd_ctx_put(dev->vqs[i]->error_ctx); | |
549 | if (dev->vqs[i]->error) | |
550 | fput(dev->vqs[i]->error); | |
551 | if (dev->vqs[i]->kick) | |
552 | fput(dev->vqs[i]->kick); | |
553 | if (dev->vqs[i]->call_ctx) | |
554 | eventfd_ctx_put(dev->vqs[i]->call_ctx); | |
555 | if (dev->vqs[i]->call) | |
556 | fput(dev->vqs[i]->call); | |
557 | vhost_vq_reset(dev, dev->vqs[i]); | |
3a4d5c94 | 558 | } |
e0e9b406 | 559 | vhost_dev_free_iovecs(dev); |
3a4d5c94 MT |
560 | if (dev->log_ctx) |
561 | eventfd_ctx_put(dev->log_ctx); | |
562 | dev->log_ctx = NULL; | |
563 | if (dev->log_file) | |
564 | fput(dev->log_file); | |
565 | dev->log_file = NULL; | |
566 | /* No one will access memory at this point */ | |
4de7255f | 567 | kvfree(dev->memory); |
47283bef | 568 | dev->memory = NULL; |
c23f3445 | 569 | WARN_ON(!list_empty(&dev->work_list)); |
78b620ce ED |
570 | if (dev->worker) { |
571 | kthread_stop(dev->worker); | |
572 | dev->worker = NULL; | |
573 | } | |
533a19b4 MT |
574 | if (dev->mm) |
575 | mmput(dev->mm); | |
576 | dev->mm = NULL; | |
3a4d5c94 | 577 | } |
6ac1afbf | 578 | EXPORT_SYMBOL_GPL(vhost_dev_cleanup); |
3a4d5c94 MT |
579 | |
580 | static int log_access_ok(void __user *log_base, u64 addr, unsigned long sz) | |
581 | { | |
582 | u64 a = addr / VHOST_PAGE_SIZE / 8; | |
d47effe1 | 583 | |
3a4d5c94 MT |
584 | /* Make sure 64 bit math will not overflow. */ |
585 | if (a > ULONG_MAX - (unsigned long)log_base || | |
586 | a + (unsigned long)log_base > ULONG_MAX) | |
6d97e55f | 587 | return 0; |
3a4d5c94 MT |
588 | |
589 | return access_ok(VERIFY_WRITE, log_base + a, | |
590 | (sz + VHOST_PAGE_SIZE * 8 - 1) / VHOST_PAGE_SIZE / 8); | |
591 | } | |
592 | ||
593 | /* Caller should have vq mutex and device mutex. */ | |
594 | static int vq_memory_access_ok(void __user *log_base, struct vhost_memory *mem, | |
595 | int log_all) | |
596 | { | |
597 | int i; | |
179b284e | 598 | |
f8322fbe MT |
599 | if (!mem) |
600 | return 0; | |
179b284e | 601 | |
3a4d5c94 MT |
602 | for (i = 0; i < mem->nregions; ++i) { |
603 | struct vhost_memory_region *m = mem->regions + i; | |
604 | unsigned long a = m->userspace_addr; | |
605 | if (m->memory_size > ULONG_MAX) | |
606 | return 0; | |
607 | else if (!access_ok(VERIFY_WRITE, (void __user *)a, | |
608 | m->memory_size)) | |
609 | return 0; | |
610 | else if (log_all && !log_access_ok(log_base, | |
611 | m->guest_phys_addr, | |
612 | m->memory_size)) | |
613 | return 0; | |
614 | } | |
615 | return 1; | |
616 | } | |
617 | ||
618 | /* Can we switch to this memory table? */ | |
619 | /* Caller should have device mutex but not vq mutex */ | |
620 | static int memory_access_ok(struct vhost_dev *d, struct vhost_memory *mem, | |
621 | int log_all) | |
622 | { | |
623 | int i; | |
d47effe1 | 624 | |
3a4d5c94 MT |
625 | for (i = 0; i < d->nvqs; ++i) { |
626 | int ok; | |
ea16c514 MT |
627 | bool log; |
628 | ||
3ab2e420 | 629 | mutex_lock(&d->vqs[i]->mutex); |
ea16c514 | 630 | log = log_all || vhost_has_feature(d->vqs[i], VHOST_F_LOG_ALL); |
3a4d5c94 | 631 | /* If ring is inactive, will check when it's enabled. */ |
3ab2e420 | 632 | if (d->vqs[i]->private_data) |
ea16c514 | 633 | ok = vq_memory_access_ok(d->vqs[i]->log_base, mem, log); |
3a4d5c94 MT |
634 | else |
635 | ok = 1; | |
3ab2e420 | 636 | mutex_unlock(&d->vqs[i]->mutex); |
3a4d5c94 MT |
637 | if (!ok) |
638 | return 0; | |
639 | } | |
640 | return 1; | |
641 | } | |
642 | ||
ea16c514 | 643 | static int vq_access_ok(struct vhost_virtqueue *vq, unsigned int num, |
3a4d5c94 MT |
644 | struct vring_desc __user *desc, |
645 | struct vring_avail __user *avail, | |
646 | struct vring_used __user *used) | |
647 | { | |
ea16c514 | 648 | size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; |
3a4d5c94 MT |
649 | return access_ok(VERIFY_READ, desc, num * sizeof *desc) && |
650 | access_ok(VERIFY_READ, avail, | |
8ea8cf89 | 651 | sizeof *avail + num * sizeof *avail->ring + s) && |
3a4d5c94 | 652 | access_ok(VERIFY_WRITE, used, |
8ea8cf89 | 653 | sizeof *used + num * sizeof *used->ring + s); |
3a4d5c94 MT |
654 | } |
655 | ||
656 | /* Can we log writes? */ | |
657 | /* Caller should have device mutex but not vq mutex */ | |
658 | int vhost_log_access_ok(struct vhost_dev *dev) | |
659 | { | |
47283bef | 660 | return memory_access_ok(dev, dev->memory, 1); |
3a4d5c94 | 661 | } |
6ac1afbf | 662 | EXPORT_SYMBOL_GPL(vhost_log_access_ok); |
3a4d5c94 MT |
663 | |
664 | /* Verify access for write logging. */ | |
665 | /* Caller should have vq mutex and device mutex */ | |
ea16c514 | 666 | static int vq_log_access_ok(struct vhost_virtqueue *vq, |
8ea8cf89 | 667 | void __user *log_base) |
3a4d5c94 | 668 | { |
ea16c514 | 669 | size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; |
28457ee6 | 670 | |
47283bef | 671 | return vq_memory_access_ok(log_base, vq->memory, |
ea16c514 | 672 | vhost_has_feature(vq, VHOST_F_LOG_ALL)) && |
3a4d5c94 MT |
673 | (!vq->log_used || log_access_ok(log_base, vq->log_addr, |
674 | sizeof *vq->used + | |
8ea8cf89 | 675 | vq->num * sizeof *vq->used->ring + s)); |
3a4d5c94 MT |
676 | } |
677 | ||
678 | /* Can we start vq? */ | |
679 | /* Caller should have vq mutex and device mutex */ | |
680 | int vhost_vq_access_ok(struct vhost_virtqueue *vq) | |
681 | { | |
ea16c514 MT |
682 | return vq_access_ok(vq, vq->num, vq->desc, vq->avail, vq->used) && |
683 | vq_log_access_ok(vq, vq->log_base); | |
3a4d5c94 | 684 | } |
6ac1afbf | 685 | EXPORT_SYMBOL_GPL(vhost_vq_access_ok); |
3a4d5c94 | 686 | |
bcfeacab IM |
687 | static int vhost_memory_reg_sort_cmp(const void *p1, const void *p2) |
688 | { | |
689 | const struct vhost_memory_region *r1 = p1, *r2 = p2; | |
690 | if (r1->guest_phys_addr < r2->guest_phys_addr) | |
691 | return 1; | |
692 | if (r1->guest_phys_addr > r2->guest_phys_addr) | |
693 | return -1; | |
694 | return 0; | |
695 | } | |
696 | ||
4de7255f IM |
697 | static void *vhost_kvzalloc(unsigned long size) |
698 | { | |
699 | void *n = kzalloc(size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); | |
700 | ||
1e099473 | 701 | if (!n) |
4de7255f | 702 | n = vzalloc(size); |
4de7255f IM |
703 | return n; |
704 | } | |
705 | ||
3a4d5c94 MT |
706 | static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m) |
707 | { | |
708 | struct vhost_memory mem, *newmem, *oldmem; | |
709 | unsigned long size = offsetof(struct vhost_memory, regions); | |
98f9ca0a | 710 | int i; |
d47effe1 | 711 | |
7ad9c9d2 TY |
712 | if (copy_from_user(&mem, m, size)) |
713 | return -EFAULT; | |
3a4d5c94 MT |
714 | if (mem.padding) |
715 | return -EOPNOTSUPP; | |
c9ce42f7 | 716 | if (mem.nregions > max_mem_regions) |
3a4d5c94 | 717 | return -E2BIG; |
4de7255f | 718 | newmem = vhost_kvzalloc(size + mem.nregions * sizeof(*m->regions)); |
3a4d5c94 MT |
719 | if (!newmem) |
720 | return -ENOMEM; | |
721 | ||
722 | memcpy(newmem, &mem, size); | |
7ad9c9d2 TY |
723 | if (copy_from_user(newmem->regions, m->regions, |
724 | mem.nregions * sizeof *m->regions)) { | |
bcfeacab | 725 | kvfree(newmem); |
7ad9c9d2 | 726 | return -EFAULT; |
3a4d5c94 | 727 | } |
bcfeacab IM |
728 | sort(newmem->regions, newmem->nregions, sizeof(*newmem->regions), |
729 | vhost_memory_reg_sort_cmp, NULL); | |
3a4d5c94 | 730 | |
ea16c514 | 731 | if (!memory_access_ok(d, newmem, 0)) { |
4de7255f | 732 | kvfree(newmem); |
3a4d5c94 | 733 | return -EFAULT; |
a02c3789 | 734 | } |
47283bef MT |
735 | oldmem = d->memory; |
736 | d->memory = newmem; | |
98f9ca0a | 737 | |
47283bef | 738 | /* All memory accesses are done under some VQ mutex. */ |
98f9ca0a MT |
739 | for (i = 0; i < d->nvqs; ++i) { |
740 | mutex_lock(&d->vqs[i]->mutex); | |
47283bef | 741 | d->vqs[i]->memory = newmem; |
98f9ca0a MT |
742 | mutex_unlock(&d->vqs[i]->mutex); |
743 | } | |
4de7255f | 744 | kvfree(oldmem); |
3a4d5c94 MT |
745 | return 0; |
746 | } | |
747 | ||
935cdee7 | 748 | long vhost_vring_ioctl(struct vhost_dev *d, int ioctl, void __user *argp) |
3a4d5c94 | 749 | { |
cecb46f1 AV |
750 | struct file *eventfp, *filep = NULL; |
751 | bool pollstart = false, pollstop = false; | |
3a4d5c94 MT |
752 | struct eventfd_ctx *ctx = NULL; |
753 | u32 __user *idxp = argp; | |
754 | struct vhost_virtqueue *vq; | |
755 | struct vhost_vring_state s; | |
756 | struct vhost_vring_file f; | |
757 | struct vhost_vring_addr a; | |
758 | u32 idx; | |
759 | long r; | |
760 | ||
761 | r = get_user(idx, idxp); | |
762 | if (r < 0) | |
763 | return r; | |
0f3d9a17 | 764 | if (idx >= d->nvqs) |
3a4d5c94 MT |
765 | return -ENOBUFS; |
766 | ||
3ab2e420 | 767 | vq = d->vqs[idx]; |
3a4d5c94 MT |
768 | |
769 | mutex_lock(&vq->mutex); | |
770 | ||
771 | switch (ioctl) { | |
772 | case VHOST_SET_VRING_NUM: | |
773 | /* Resizing ring with an active backend? | |
774 | * You don't want to do that. */ | |
775 | if (vq->private_data) { | |
776 | r = -EBUSY; | |
777 | break; | |
778 | } | |
7ad9c9d2 TY |
779 | if (copy_from_user(&s, argp, sizeof s)) { |
780 | r = -EFAULT; | |
3a4d5c94 | 781 | break; |
7ad9c9d2 | 782 | } |
3a4d5c94 MT |
783 | if (!s.num || s.num > 0xffff || (s.num & (s.num - 1))) { |
784 | r = -EINVAL; | |
785 | break; | |
786 | } | |
787 | vq->num = s.num; | |
788 | break; | |
789 | case VHOST_SET_VRING_BASE: | |
790 | /* Moving base with an active backend? | |
791 | * You don't want to do that. */ | |
792 | if (vq->private_data) { | |
793 | r = -EBUSY; | |
794 | break; | |
795 | } | |
7ad9c9d2 TY |
796 | if (copy_from_user(&s, argp, sizeof s)) { |
797 | r = -EFAULT; | |
3a4d5c94 | 798 | break; |
7ad9c9d2 | 799 | } |
3a4d5c94 MT |
800 | if (s.num > 0xffff) { |
801 | r = -EINVAL; | |
802 | break; | |
803 | } | |
804 | vq->last_avail_idx = s.num; | |
805 | /* Forget the cached index value. */ | |
806 | vq->avail_idx = vq->last_avail_idx; | |
807 | break; | |
808 | case VHOST_GET_VRING_BASE: | |
809 | s.index = idx; | |
810 | s.num = vq->last_avail_idx; | |
7ad9c9d2 TY |
811 | if (copy_to_user(argp, &s, sizeof s)) |
812 | r = -EFAULT; | |
3a4d5c94 MT |
813 | break; |
814 | case VHOST_SET_VRING_ADDR: | |
7ad9c9d2 TY |
815 | if (copy_from_user(&a, argp, sizeof a)) { |
816 | r = -EFAULT; | |
3a4d5c94 | 817 | break; |
7ad9c9d2 | 818 | } |
3a4d5c94 MT |
819 | if (a.flags & ~(0x1 << VHOST_VRING_F_LOG)) { |
820 | r = -EOPNOTSUPP; | |
821 | break; | |
822 | } | |
823 | /* For 32bit, verify that the top 32bits of the user | |
824 | data are set to zero. */ | |
825 | if ((u64)(unsigned long)a.desc_user_addr != a.desc_user_addr || | |
826 | (u64)(unsigned long)a.used_user_addr != a.used_user_addr || | |
827 | (u64)(unsigned long)a.avail_user_addr != a.avail_user_addr) { | |
828 | r = -EFAULT; | |
829 | break; | |
830 | } | |
5d9a07b0 MT |
831 | |
832 | /* Make sure it's safe to cast pointers to vring types. */ | |
833 | BUILD_BUG_ON(__alignof__ *vq->avail > VRING_AVAIL_ALIGN_SIZE); | |
834 | BUILD_BUG_ON(__alignof__ *vq->used > VRING_USED_ALIGN_SIZE); | |
835 | if ((a.avail_user_addr & (VRING_AVAIL_ALIGN_SIZE - 1)) || | |
836 | (a.used_user_addr & (VRING_USED_ALIGN_SIZE - 1)) || | |
d5424838 | 837 | (a.log_guest_addr & (VRING_USED_ALIGN_SIZE - 1))) { |
3a4d5c94 MT |
838 | r = -EINVAL; |
839 | break; | |
840 | } | |
841 | ||
842 | /* We only verify access here if backend is configured. | |
843 | * If it is not, we don't as size might not have been setup. | |
844 | * We will verify when backend is configured. */ | |
845 | if (vq->private_data) { | |
ea16c514 | 846 | if (!vq_access_ok(vq, vq->num, |
3a4d5c94 MT |
847 | (void __user *)(unsigned long)a.desc_user_addr, |
848 | (void __user *)(unsigned long)a.avail_user_addr, | |
849 | (void __user *)(unsigned long)a.used_user_addr)) { | |
850 | r = -EINVAL; | |
851 | break; | |
852 | } | |
853 | ||
854 | /* Also validate log access for used ring if enabled. */ | |
855 | if ((a.flags & (0x1 << VHOST_VRING_F_LOG)) && | |
856 | !log_access_ok(vq->log_base, a.log_guest_addr, | |
857 | sizeof *vq->used + | |
858 | vq->num * sizeof *vq->used->ring)) { | |
859 | r = -EINVAL; | |
860 | break; | |
861 | } | |
862 | } | |
863 | ||
3a4d5c94 MT |
864 | vq->log_used = !!(a.flags & (0x1 << VHOST_VRING_F_LOG)); |
865 | vq->desc = (void __user *)(unsigned long)a.desc_user_addr; | |
866 | vq->avail = (void __user *)(unsigned long)a.avail_user_addr; | |
867 | vq->log_addr = a.log_guest_addr; | |
868 | vq->used = (void __user *)(unsigned long)a.used_user_addr; | |
869 | break; | |
870 | case VHOST_SET_VRING_KICK: | |
7ad9c9d2 TY |
871 | if (copy_from_user(&f, argp, sizeof f)) { |
872 | r = -EFAULT; | |
3a4d5c94 | 873 | break; |
7ad9c9d2 | 874 | } |
3a4d5c94 | 875 | eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd); |
535297a6 MT |
876 | if (IS_ERR(eventfp)) { |
877 | r = PTR_ERR(eventfp); | |
878 | break; | |
879 | } | |
3a4d5c94 | 880 | if (eventfp != vq->kick) { |
cecb46f1 AV |
881 | pollstop = (filep = vq->kick) != NULL; |
882 | pollstart = (vq->kick = eventfp) != NULL; | |
3a4d5c94 MT |
883 | } else |
884 | filep = eventfp; | |
885 | break; | |
886 | case VHOST_SET_VRING_CALL: | |
7ad9c9d2 TY |
887 | if (copy_from_user(&f, argp, sizeof f)) { |
888 | r = -EFAULT; | |
3a4d5c94 | 889 | break; |
7ad9c9d2 | 890 | } |
3a4d5c94 | 891 | eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd); |
535297a6 MT |
892 | if (IS_ERR(eventfp)) { |
893 | r = PTR_ERR(eventfp); | |
894 | break; | |
895 | } | |
3a4d5c94 MT |
896 | if (eventfp != vq->call) { |
897 | filep = vq->call; | |
898 | ctx = vq->call_ctx; | |
899 | vq->call = eventfp; | |
900 | vq->call_ctx = eventfp ? | |
901 | eventfd_ctx_fileget(eventfp) : NULL; | |
902 | } else | |
903 | filep = eventfp; | |
904 | break; | |
905 | case VHOST_SET_VRING_ERR: | |
7ad9c9d2 TY |
906 | if (copy_from_user(&f, argp, sizeof f)) { |
907 | r = -EFAULT; | |
3a4d5c94 | 908 | break; |
7ad9c9d2 | 909 | } |
3a4d5c94 | 910 | eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd); |
535297a6 MT |
911 | if (IS_ERR(eventfp)) { |
912 | r = PTR_ERR(eventfp); | |
913 | break; | |
914 | } | |
3a4d5c94 MT |
915 | if (eventfp != vq->error) { |
916 | filep = vq->error; | |
917 | vq->error = eventfp; | |
918 | ctx = vq->error_ctx; | |
919 | vq->error_ctx = eventfp ? | |
920 | eventfd_ctx_fileget(eventfp) : NULL; | |
921 | } else | |
922 | filep = eventfp; | |
923 | break; | |
2751c988 GK |
924 | case VHOST_SET_VRING_ENDIAN: |
925 | r = vhost_set_vring_endian(vq, argp); | |
926 | break; | |
927 | case VHOST_GET_VRING_ENDIAN: | |
928 | r = vhost_get_vring_endian(vq, idx, argp); | |
929 | break; | |
03088137 JW |
930 | case VHOST_SET_VRING_BUSYLOOP_TIMEOUT: |
931 | if (copy_from_user(&s, argp, sizeof(s))) { | |
932 | r = -EFAULT; | |
933 | break; | |
934 | } | |
935 | vq->busyloop_timeout = s.num; | |
936 | break; | |
937 | case VHOST_GET_VRING_BUSYLOOP_TIMEOUT: | |
938 | s.index = idx; | |
939 | s.num = vq->busyloop_timeout; | |
940 | if (copy_to_user(argp, &s, sizeof(s))) | |
941 | r = -EFAULT; | |
942 | break; | |
3a4d5c94 MT |
943 | default: |
944 | r = -ENOIOCTLCMD; | |
945 | } | |
946 | ||
947 | if (pollstop && vq->handle_kick) | |
948 | vhost_poll_stop(&vq->poll); | |
949 | ||
950 | if (ctx) | |
951 | eventfd_ctx_put(ctx); | |
952 | if (filep) | |
953 | fput(filep); | |
954 | ||
955 | if (pollstart && vq->handle_kick) | |
2b8b328b | 956 | r = vhost_poll_start(&vq->poll, vq->kick); |
3a4d5c94 MT |
957 | |
958 | mutex_unlock(&vq->mutex); | |
959 | ||
960 | if (pollstop && vq->handle_kick) | |
961 | vhost_poll_flush(&vq->poll); | |
962 | return r; | |
963 | } | |
6ac1afbf | 964 | EXPORT_SYMBOL_GPL(vhost_vring_ioctl); |
3a4d5c94 MT |
965 | |
966 | /* Caller must have device mutex */ | |
935cdee7 | 967 | long vhost_dev_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp) |
3a4d5c94 | 968 | { |
3a4d5c94 MT |
969 | struct file *eventfp, *filep = NULL; |
970 | struct eventfd_ctx *ctx = NULL; | |
971 | u64 p; | |
972 | long r; | |
973 | int i, fd; | |
974 | ||
975 | /* If you are not the owner, you can become one */ | |
976 | if (ioctl == VHOST_SET_OWNER) { | |
977 | r = vhost_dev_set_owner(d); | |
978 | goto done; | |
979 | } | |
980 | ||
981 | /* You must be the owner to do anything else */ | |
982 | r = vhost_dev_check_owner(d); | |
983 | if (r) | |
984 | goto done; | |
985 | ||
986 | switch (ioctl) { | |
987 | case VHOST_SET_MEM_TABLE: | |
988 | r = vhost_set_memory(d, argp); | |
989 | break; | |
990 | case VHOST_SET_LOG_BASE: | |
7ad9c9d2 TY |
991 | if (copy_from_user(&p, argp, sizeof p)) { |
992 | r = -EFAULT; | |
3a4d5c94 | 993 | break; |
7ad9c9d2 | 994 | } |
3a4d5c94 MT |
995 | if ((u64)(unsigned long)p != p) { |
996 | r = -EFAULT; | |
997 | break; | |
998 | } | |
999 | for (i = 0; i < d->nvqs; ++i) { | |
1000 | struct vhost_virtqueue *vq; | |
1001 | void __user *base = (void __user *)(unsigned long)p; | |
3ab2e420 | 1002 | vq = d->vqs[i]; |
3a4d5c94 MT |
1003 | mutex_lock(&vq->mutex); |
1004 | /* If ring is inactive, will check when it's enabled. */ | |
ea16c514 | 1005 | if (vq->private_data && !vq_log_access_ok(vq, base)) |
3a4d5c94 MT |
1006 | r = -EFAULT; |
1007 | else | |
1008 | vq->log_base = base; | |
1009 | mutex_unlock(&vq->mutex); | |
1010 | } | |
1011 | break; | |
1012 | case VHOST_SET_LOG_FD: | |
1013 | r = get_user(fd, (int __user *)argp); | |
1014 | if (r < 0) | |
1015 | break; | |
1016 | eventfp = fd == -1 ? NULL : eventfd_fget(fd); | |
1017 | if (IS_ERR(eventfp)) { | |
1018 | r = PTR_ERR(eventfp); | |
1019 | break; | |
1020 | } | |
1021 | if (eventfp != d->log_file) { | |
1022 | filep = d->log_file; | |
7932c0bd | 1023 | d->log_file = eventfp; |
3a4d5c94 MT |
1024 | ctx = d->log_ctx; |
1025 | d->log_ctx = eventfp ? | |
1026 | eventfd_ctx_fileget(eventfp) : NULL; | |
1027 | } else | |
1028 | filep = eventfp; | |
1029 | for (i = 0; i < d->nvqs; ++i) { | |
3ab2e420 AH |
1030 | mutex_lock(&d->vqs[i]->mutex); |
1031 | d->vqs[i]->log_ctx = d->log_ctx; | |
1032 | mutex_unlock(&d->vqs[i]->mutex); | |
3a4d5c94 MT |
1033 | } |
1034 | if (ctx) | |
1035 | eventfd_ctx_put(ctx); | |
1036 | if (filep) | |
1037 | fput(filep); | |
1038 | break; | |
1039 | default: | |
935cdee7 | 1040 | r = -ENOIOCTLCMD; |
3a4d5c94 MT |
1041 | break; |
1042 | } | |
1043 | done: | |
1044 | return r; | |
1045 | } | |
6ac1afbf | 1046 | EXPORT_SYMBOL_GPL(vhost_dev_ioctl); |
3a4d5c94 MT |
1047 | |
1048 | static const struct vhost_memory_region *find_region(struct vhost_memory *mem, | |
1049 | __u64 addr, __u32 len) | |
1050 | { | |
bcfeacab IM |
1051 | const struct vhost_memory_region *reg; |
1052 | int start = 0, end = mem->nregions; | |
1053 | ||
1054 | while (start < end) { | |
1055 | int slot = start + (end - start) / 2; | |
1056 | reg = mem->regions + slot; | |
1057 | if (addr >= reg->guest_phys_addr) | |
1058 | end = slot; | |
1059 | else | |
1060 | start = slot + 1; | |
3a4d5c94 | 1061 | } |
bcfeacab IM |
1062 | |
1063 | reg = mem->regions + start; | |
1064 | if (addr >= reg->guest_phys_addr && | |
1065 | reg->guest_phys_addr + reg->memory_size > addr) | |
1066 | return reg; | |
3a4d5c94 MT |
1067 | return NULL; |
1068 | } | |
1069 | ||
1070 | /* TODO: This is really inefficient. We need something like get_user() | |
1071 | * (instruction directly accesses the data, with an exception table entry | |
1072 | * returning -EFAULT). See Documentation/x86/exception-tables.txt. | |
1073 | */ | |
1074 | static int set_bit_to_user(int nr, void __user *addr) | |
1075 | { | |
1076 | unsigned long log = (unsigned long)addr; | |
1077 | struct page *page; | |
1078 | void *base; | |
1079 | int bit = nr + (log % PAGE_SIZE) * 8; | |
1080 | int r; | |
d47effe1 | 1081 | |
3a4d5c94 | 1082 | r = get_user_pages_fast(log, 1, 1, &page); |
d6db3f5c | 1083 | if (r < 0) |
3a4d5c94 | 1084 | return r; |
d6db3f5c | 1085 | BUG_ON(r != 1); |
c6daa7ff | 1086 | base = kmap_atomic(page); |
3a4d5c94 | 1087 | set_bit(bit, base); |
c6daa7ff | 1088 | kunmap_atomic(base); |
3a4d5c94 MT |
1089 | set_page_dirty_lock(page); |
1090 | put_page(page); | |
1091 | return 0; | |
1092 | } | |
1093 | ||
1094 | static int log_write(void __user *log_base, | |
1095 | u64 write_address, u64 write_length) | |
1096 | { | |
28831ee6 | 1097 | u64 write_page = write_address / VHOST_PAGE_SIZE; |
3a4d5c94 | 1098 | int r; |
d47effe1 | 1099 | |
3a4d5c94 MT |
1100 | if (!write_length) |
1101 | return 0; | |
3bf9be40 | 1102 | write_length += write_address % VHOST_PAGE_SIZE; |
3a4d5c94 MT |
1103 | for (;;) { |
1104 | u64 base = (u64)(unsigned long)log_base; | |
28831ee6 MT |
1105 | u64 log = base + write_page / 8; |
1106 | int bit = write_page % 8; | |
3a4d5c94 MT |
1107 | if ((u64)(unsigned long)log != log) |
1108 | return -EFAULT; | |
1109 | r = set_bit_to_user(bit, (void __user *)(unsigned long)log); | |
1110 | if (r < 0) | |
1111 | return r; | |
1112 | if (write_length <= VHOST_PAGE_SIZE) | |
1113 | break; | |
1114 | write_length -= VHOST_PAGE_SIZE; | |
28831ee6 | 1115 | write_page += 1; |
3a4d5c94 MT |
1116 | } |
1117 | return r; | |
1118 | } | |
1119 | ||
1120 | int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log, | |
1121 | unsigned int log_num, u64 len) | |
1122 | { | |
1123 | int i, r; | |
1124 | ||
1125 | /* Make sure data written is seen before log. */ | |
5659338c | 1126 | smp_wmb(); |
3a4d5c94 MT |
1127 | for (i = 0; i < log_num; ++i) { |
1128 | u64 l = min(log[i].len, len); | |
1129 | r = log_write(vq->log_base, log[i].addr, l); | |
1130 | if (r < 0) | |
1131 | return r; | |
1132 | len -= l; | |
5786aee8 MT |
1133 | if (!len) { |
1134 | if (vq->log_ctx) | |
1135 | eventfd_signal(vq->log_ctx, 1); | |
3a4d5c94 | 1136 | return 0; |
5786aee8 | 1137 | } |
3a4d5c94 | 1138 | } |
3a4d5c94 MT |
1139 | /* Length written exceeds what we have stored. This is a bug. */ |
1140 | BUG(); | |
1141 | return 0; | |
1142 | } | |
6ac1afbf | 1143 | EXPORT_SYMBOL_GPL(vhost_log_write); |
3a4d5c94 | 1144 | |
2723feaa JW |
1145 | static int vhost_update_used_flags(struct vhost_virtqueue *vq) |
1146 | { | |
1147 | void __user *used; | |
3b1bbe89 | 1148 | if (__put_user(cpu_to_vhost16(vq, vq->used_flags), &vq->used->flags) < 0) |
2723feaa JW |
1149 | return -EFAULT; |
1150 | if (unlikely(vq->log_used)) { | |
1151 | /* Make sure the flag is seen before log. */ | |
1152 | smp_wmb(); | |
1153 | /* Log used flag write. */ | |
1154 | used = &vq->used->flags; | |
1155 | log_write(vq->log_base, vq->log_addr + | |
1156 | (used - (void __user *)vq->used), | |
1157 | sizeof vq->used->flags); | |
1158 | if (vq->log_ctx) | |
1159 | eventfd_signal(vq->log_ctx, 1); | |
1160 | } | |
1161 | return 0; | |
1162 | } | |
1163 | ||
1164 | static int vhost_update_avail_event(struct vhost_virtqueue *vq, u16 avail_event) | |
1165 | { | |
3b1bbe89 | 1166 | if (__put_user(cpu_to_vhost16(vq, vq->avail_idx), vhost_avail_event(vq))) |
2723feaa JW |
1167 | return -EFAULT; |
1168 | if (unlikely(vq->log_used)) { | |
1169 | void __user *used; | |
1170 | /* Make sure the event is seen before log. */ | |
1171 | smp_wmb(); | |
1172 | /* Log avail event write */ | |
1173 | used = vhost_avail_event(vq); | |
1174 | log_write(vq->log_base, vq->log_addr + | |
1175 | (used - (void __user *)vq->used), | |
1176 | sizeof *vhost_avail_event(vq)); | |
1177 | if (vq->log_ctx) | |
1178 | eventfd_signal(vq->log_ctx, 1); | |
1179 | } | |
1180 | return 0; | |
1181 | } | |
1182 | ||
80f7d030 | 1183 | int vhost_vq_init_access(struct vhost_virtqueue *vq) |
2723feaa | 1184 | { |
3b1bbe89 | 1185 | __virtio16 last_used_idx; |
2723feaa | 1186 | int r; |
e1f33be9 GK |
1187 | bool is_le = vq->is_le; |
1188 | ||
2751c988 | 1189 | if (!vq->private_data) { |
c5072037 | 1190 | vhost_reset_is_le(vq); |
2723feaa | 1191 | return 0; |
2751c988 GK |
1192 | } |
1193 | ||
1194 | vhost_init_is_le(vq); | |
2723feaa JW |
1195 | |
1196 | r = vhost_update_used_flags(vq); | |
1197 | if (r) | |
e1f33be9 | 1198 | goto err; |
2723feaa | 1199 | vq->signalled_used_valid = false; |
e1f33be9 GK |
1200 | if (!access_ok(VERIFY_READ, &vq->used->idx, sizeof vq->used->idx)) { |
1201 | r = -EFAULT; | |
1202 | goto err; | |
1203 | } | |
64f7f051 MT |
1204 | r = __get_user(last_used_idx, &vq->used->idx); |
1205 | if (r) | |
e1f33be9 | 1206 | goto err; |
3b1bbe89 | 1207 | vq->last_used_idx = vhost16_to_cpu(vq, last_used_idx); |
64f7f051 | 1208 | return 0; |
e1f33be9 GK |
1209 | err: |
1210 | vq->is_le = is_le; | |
1211 | return r; | |
2723feaa | 1212 | } |
80f7d030 | 1213 | EXPORT_SYMBOL_GPL(vhost_vq_init_access); |
2723feaa | 1214 | |
47283bef | 1215 | static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len, |
a8d3782f | 1216 | struct iovec iov[], int iov_size) |
3a4d5c94 MT |
1217 | { |
1218 | const struct vhost_memory_region *reg; | |
1219 | struct vhost_memory *mem; | |
1220 | struct iovec *_iov; | |
1221 | u64 s = 0; | |
1222 | int ret = 0; | |
1223 | ||
47283bef | 1224 | mem = vq->memory; |
3a4d5c94 MT |
1225 | while ((u64)len > s) { |
1226 | u64 size; | |
7b3384fc | 1227 | if (unlikely(ret >= iov_size)) { |
3a4d5c94 MT |
1228 | ret = -ENOBUFS; |
1229 | break; | |
1230 | } | |
1231 | reg = find_region(mem, addr, len); | |
7b3384fc | 1232 | if (unlikely(!reg)) { |
3a4d5c94 MT |
1233 | ret = -EFAULT; |
1234 | break; | |
1235 | } | |
1236 | _iov = iov + ret; | |
1237 | size = reg->memory_size - addr + reg->guest_phys_addr; | |
bd97120f | 1238 | _iov->iov_len = min((u64)len - s, size); |
a8d3782f | 1239 | _iov->iov_base = (void __user *)(unsigned long) |
3a4d5c94 MT |
1240 | (reg->userspace_addr + addr - reg->guest_phys_addr); |
1241 | s += size; | |
1242 | addr += size; | |
1243 | ++ret; | |
1244 | } | |
1245 | ||
3a4d5c94 MT |
1246 | return ret; |
1247 | } | |
1248 | ||
1249 | /* Each buffer in the virtqueues is actually a chain of descriptors. This | |
1250 | * function returns the next descriptor in the chain, | |
1251 | * or -1U if we're at the end. */ | |
3b1bbe89 | 1252 | static unsigned next_desc(struct vhost_virtqueue *vq, struct vring_desc *desc) |
3a4d5c94 MT |
1253 | { |
1254 | unsigned int next; | |
1255 | ||
1256 | /* If this descriptor says it doesn't chain, we're done. */ | |
3b1bbe89 | 1257 | if (!(desc->flags & cpu_to_vhost16(vq, VRING_DESC_F_NEXT))) |
3a4d5c94 MT |
1258 | return -1U; |
1259 | ||
1260 | /* Check they're not leading us off end of descriptors. */ | |
3b1bbe89 | 1261 | next = vhost16_to_cpu(vq, desc->next); |
3a4d5c94 MT |
1262 | /* Make sure compiler knows to grab that: we don't want it changing! */ |
1263 | /* We will use the result as an index in an array, so most | |
1264 | * architectures only need a compiler barrier here. */ | |
1265 | read_barrier_depends(); | |
1266 | ||
1267 | return next; | |
1268 | } | |
1269 | ||
47283bef | 1270 | static int get_indirect(struct vhost_virtqueue *vq, |
7b3384fc MT |
1271 | struct iovec iov[], unsigned int iov_size, |
1272 | unsigned int *out_num, unsigned int *in_num, | |
1273 | struct vhost_log *log, unsigned int *log_num, | |
1274 | struct vring_desc *indirect) | |
3a4d5c94 MT |
1275 | { |
1276 | struct vring_desc desc; | |
1277 | unsigned int i = 0, count, found = 0; | |
3b1bbe89 | 1278 | u32 len = vhost32_to_cpu(vq, indirect->len); |
aad9a1ce | 1279 | struct iov_iter from; |
3a4d5c94 MT |
1280 | int ret; |
1281 | ||
1282 | /* Sanity check */ | |
3b1bbe89 | 1283 | if (unlikely(len % sizeof desc)) { |
3a4d5c94 MT |
1284 | vq_err(vq, "Invalid length in indirect descriptor: " |
1285 | "len 0x%llx not multiple of 0x%zx\n", | |
3b1bbe89 | 1286 | (unsigned long long)len, |
3a4d5c94 MT |
1287 | sizeof desc); |
1288 | return -EINVAL; | |
1289 | } | |
1290 | ||
3b1bbe89 | 1291 | ret = translate_desc(vq, vhost64_to_cpu(vq, indirect->addr), len, vq->indirect, |
e0e9b406 | 1292 | UIO_MAXIOV); |
7b3384fc | 1293 | if (unlikely(ret < 0)) { |
3a4d5c94 MT |
1294 | vq_err(vq, "Translation failure %d in indirect.\n", ret); |
1295 | return ret; | |
1296 | } | |
aad9a1ce | 1297 | iov_iter_init(&from, READ, vq->indirect, ret, len); |
3a4d5c94 MT |
1298 | |
1299 | /* We will use the result as an address to read from, so most | |
1300 | * architectures only need a compiler barrier here. */ | |
1301 | read_barrier_depends(); | |
1302 | ||
3b1bbe89 | 1303 | count = len / sizeof desc; |
3a4d5c94 MT |
1304 | /* Buffers are chained via a 16 bit next field, so |
1305 | * we can have at most 2^16 of these. */ | |
7b3384fc | 1306 | if (unlikely(count > USHRT_MAX + 1)) { |
3a4d5c94 MT |
1307 | vq_err(vq, "Indirect buffer length too big: %d\n", |
1308 | indirect->len); | |
1309 | return -E2BIG; | |
1310 | } | |
1311 | ||
1312 | do { | |
1313 | unsigned iov_count = *in_num + *out_num; | |
7b3384fc | 1314 | if (unlikely(++found > count)) { |
3a4d5c94 MT |
1315 | vq_err(vq, "Loop detected: last one at %u " |
1316 | "indirect size %u\n", | |
1317 | i, count); | |
1318 | return -EINVAL; | |
1319 | } | |
aad9a1ce AV |
1320 | if (unlikely(copy_from_iter(&desc, sizeof(desc), &from) != |
1321 | sizeof(desc))) { | |
3a4d5c94 | 1322 | vq_err(vq, "Failed indirect descriptor: idx %d, %zx\n", |
3b1bbe89 | 1323 | i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc); |
3a4d5c94 MT |
1324 | return -EINVAL; |
1325 | } | |
3b1bbe89 | 1326 | if (unlikely(desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT))) { |
3a4d5c94 | 1327 | vq_err(vq, "Nested indirect descriptor: idx %d, %zx\n", |
3b1bbe89 | 1328 | i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc); |
3a4d5c94 MT |
1329 | return -EINVAL; |
1330 | } | |
1331 | ||
3b1bbe89 MT |
1332 | ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr), |
1333 | vhost32_to_cpu(vq, desc.len), iov + iov_count, | |
3a4d5c94 | 1334 | iov_size - iov_count); |
7b3384fc | 1335 | if (unlikely(ret < 0)) { |
3a4d5c94 MT |
1336 | vq_err(vq, "Translation failure %d indirect idx %d\n", |
1337 | ret, i); | |
1338 | return ret; | |
1339 | } | |
1340 | /* If this is an input descriptor, increment that count. */ | |
3b1bbe89 | 1341 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE)) { |
3a4d5c94 MT |
1342 | *in_num += ret; |
1343 | if (unlikely(log)) { | |
3b1bbe89 MT |
1344 | log[*log_num].addr = vhost64_to_cpu(vq, desc.addr); |
1345 | log[*log_num].len = vhost32_to_cpu(vq, desc.len); | |
3a4d5c94 MT |
1346 | ++*log_num; |
1347 | } | |
1348 | } else { | |
1349 | /* If it's an output descriptor, they're all supposed | |
1350 | * to come before any input descriptors. */ | |
7b3384fc | 1351 | if (unlikely(*in_num)) { |
3a4d5c94 MT |
1352 | vq_err(vq, "Indirect descriptor " |
1353 | "has out after in: idx %d\n", i); | |
1354 | return -EINVAL; | |
1355 | } | |
1356 | *out_num += ret; | |
1357 | } | |
3b1bbe89 | 1358 | } while ((i = next_desc(vq, &desc)) != -1); |
3a4d5c94 MT |
1359 | return 0; |
1360 | } | |
1361 | ||
1362 | /* This looks in the virtqueue and for the first available buffer, and converts | |
1363 | * it to an iovec for convenient access. Since descriptors consist of some | |
1364 | * number of output then some number of input descriptors, it's actually two | |
1365 | * iovecs, but we pack them into one and note how many of each there were. | |
1366 | * | |
d5675bd2 MT |
1367 | * This function returns the descriptor number found, or vq->num (which is |
1368 | * never a valid descriptor number) if none was found. A negative code is | |
1369 | * returned on error. */ | |
47283bef | 1370 | int vhost_get_vq_desc(struct vhost_virtqueue *vq, |
d5675bd2 MT |
1371 | struct iovec iov[], unsigned int iov_size, |
1372 | unsigned int *out_num, unsigned int *in_num, | |
1373 | struct vhost_log *log, unsigned int *log_num) | |
3a4d5c94 MT |
1374 | { |
1375 | struct vring_desc desc; | |
1376 | unsigned int i, head, found = 0; | |
1377 | u16 last_avail_idx; | |
3b1bbe89 MT |
1378 | __virtio16 avail_idx; |
1379 | __virtio16 ring_head; | |
3a4d5c94 MT |
1380 | int ret; |
1381 | ||
1382 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
1383 | last_avail_idx = vq->last_avail_idx; | |
3b1bbe89 | 1384 | if (unlikely(__get_user(avail_idx, &vq->avail->idx))) { |
3a4d5c94 MT |
1385 | vq_err(vq, "Failed to access avail idx at %p\n", |
1386 | &vq->avail->idx); | |
d5675bd2 | 1387 | return -EFAULT; |
3a4d5c94 | 1388 | } |
3b1bbe89 | 1389 | vq->avail_idx = vhost16_to_cpu(vq, avail_idx); |
3a4d5c94 | 1390 | |
7b3384fc | 1391 | if (unlikely((u16)(vq->avail_idx - last_avail_idx) > vq->num)) { |
3a4d5c94 MT |
1392 | vq_err(vq, "Guest moved used index from %u to %u", |
1393 | last_avail_idx, vq->avail_idx); | |
d5675bd2 | 1394 | return -EFAULT; |
3a4d5c94 MT |
1395 | } |
1396 | ||
1397 | /* If there's nothing new since last we looked, return invalid. */ | |
1398 | if (vq->avail_idx == last_avail_idx) | |
1399 | return vq->num; | |
1400 | ||
1401 | /* Only get avail ring entries after they have been exposed by guest. */ | |
5659338c | 1402 | smp_rmb(); |
3a4d5c94 MT |
1403 | |
1404 | /* Grab the next descriptor number they're advertising, and increment | |
1405 | * the index we've seen. */ | |
3b1bbe89 | 1406 | if (unlikely(__get_user(ring_head, |
5fba13b5 | 1407 | &vq->avail->ring[last_avail_idx & (vq->num - 1)]))) { |
3a4d5c94 MT |
1408 | vq_err(vq, "Failed to read head: idx %d address %p\n", |
1409 | last_avail_idx, | |
1410 | &vq->avail->ring[last_avail_idx % vq->num]); | |
d5675bd2 | 1411 | return -EFAULT; |
3a4d5c94 MT |
1412 | } |
1413 | ||
3b1bbe89 MT |
1414 | head = vhost16_to_cpu(vq, ring_head); |
1415 | ||
3a4d5c94 | 1416 | /* If their number is silly, that's an error. */ |
7b3384fc | 1417 | if (unlikely(head >= vq->num)) { |
3a4d5c94 MT |
1418 | vq_err(vq, "Guest says index %u > %u is available", |
1419 | head, vq->num); | |
d5675bd2 | 1420 | return -EINVAL; |
3a4d5c94 MT |
1421 | } |
1422 | ||
1423 | /* When we start there are none of either input nor output. */ | |
1424 | *out_num = *in_num = 0; | |
1425 | if (unlikely(log)) | |
1426 | *log_num = 0; | |
1427 | ||
1428 | i = head; | |
1429 | do { | |
1430 | unsigned iov_count = *in_num + *out_num; | |
7b3384fc | 1431 | if (unlikely(i >= vq->num)) { |
3a4d5c94 MT |
1432 | vq_err(vq, "Desc index is %u > %u, head = %u", |
1433 | i, vq->num, head); | |
d5675bd2 | 1434 | return -EINVAL; |
3a4d5c94 | 1435 | } |
7b3384fc | 1436 | if (unlikely(++found > vq->num)) { |
3a4d5c94 MT |
1437 | vq_err(vq, "Loop detected: last one at %u " |
1438 | "vq size %u head %u\n", | |
1439 | i, vq->num, head); | |
d5675bd2 | 1440 | return -EINVAL; |
3a4d5c94 | 1441 | } |
fcc042a2 | 1442 | ret = __copy_from_user(&desc, vq->desc + i, sizeof desc); |
7b3384fc | 1443 | if (unlikely(ret)) { |
3a4d5c94 MT |
1444 | vq_err(vq, "Failed to get descriptor: idx %d addr %p\n", |
1445 | i, vq->desc + i); | |
d5675bd2 | 1446 | return -EFAULT; |
3a4d5c94 | 1447 | } |
3b1bbe89 | 1448 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT)) { |
47283bef | 1449 | ret = get_indirect(vq, iov, iov_size, |
3a4d5c94 MT |
1450 | out_num, in_num, |
1451 | log, log_num, &desc); | |
7b3384fc | 1452 | if (unlikely(ret < 0)) { |
3a4d5c94 MT |
1453 | vq_err(vq, "Failure detected " |
1454 | "in indirect descriptor at idx %d\n", i); | |
d5675bd2 | 1455 | return ret; |
3a4d5c94 MT |
1456 | } |
1457 | continue; | |
1458 | } | |
1459 | ||
3b1bbe89 MT |
1460 | ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr), |
1461 | vhost32_to_cpu(vq, desc.len), iov + iov_count, | |
3a4d5c94 | 1462 | iov_size - iov_count); |
7b3384fc | 1463 | if (unlikely(ret < 0)) { |
3a4d5c94 MT |
1464 | vq_err(vq, "Translation failure %d descriptor idx %d\n", |
1465 | ret, i); | |
d5675bd2 | 1466 | return ret; |
3a4d5c94 | 1467 | } |
3b1bbe89 | 1468 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE)) { |
3a4d5c94 MT |
1469 | /* If this is an input descriptor, |
1470 | * increment that count. */ | |
1471 | *in_num += ret; | |
1472 | if (unlikely(log)) { | |
3b1bbe89 MT |
1473 | log[*log_num].addr = vhost64_to_cpu(vq, desc.addr); |
1474 | log[*log_num].len = vhost32_to_cpu(vq, desc.len); | |
3a4d5c94 MT |
1475 | ++*log_num; |
1476 | } | |
1477 | } else { | |
1478 | /* If it's an output descriptor, they're all supposed | |
1479 | * to come before any input descriptors. */ | |
7b3384fc | 1480 | if (unlikely(*in_num)) { |
3a4d5c94 MT |
1481 | vq_err(vq, "Descriptor has out after in: " |
1482 | "idx %d\n", i); | |
d5675bd2 | 1483 | return -EINVAL; |
3a4d5c94 MT |
1484 | } |
1485 | *out_num += ret; | |
1486 | } | |
3b1bbe89 | 1487 | } while ((i = next_desc(vq, &desc)) != -1); |
3a4d5c94 MT |
1488 | |
1489 | /* On success, increment avail index. */ | |
1490 | vq->last_avail_idx++; | |
8ea8cf89 MT |
1491 | |
1492 | /* Assume notifications from guest are disabled at this point, | |
1493 | * if they aren't we would need to update avail_event index. */ | |
1494 | BUG_ON(!(vq->used_flags & VRING_USED_F_NO_NOTIFY)); | |
3a4d5c94 MT |
1495 | return head; |
1496 | } | |
6ac1afbf | 1497 | EXPORT_SYMBOL_GPL(vhost_get_vq_desc); |
3a4d5c94 MT |
1498 | |
1499 | /* Reverse the effect of vhost_get_vq_desc. Useful for error handling. */ | |
8dd014ad | 1500 | void vhost_discard_vq_desc(struct vhost_virtqueue *vq, int n) |
3a4d5c94 | 1501 | { |
8dd014ad | 1502 | vq->last_avail_idx -= n; |
3a4d5c94 | 1503 | } |
6ac1afbf | 1504 | EXPORT_SYMBOL_GPL(vhost_discard_vq_desc); |
3a4d5c94 MT |
1505 | |
1506 | /* After we've used one of their buffers, we tell them about it. We'll then | |
1507 | * want to notify the guest, using eventfd. */ | |
1508 | int vhost_add_used(struct vhost_virtqueue *vq, unsigned int head, int len) | |
1509 | { | |
3b1bbe89 MT |
1510 | struct vring_used_elem heads = { |
1511 | cpu_to_vhost32(vq, head), | |
1512 | cpu_to_vhost32(vq, len) | |
1513 | }; | |
3a4d5c94 | 1514 | |
c49e4e57 | 1515 | return vhost_add_used_n(vq, &heads, 1); |
3a4d5c94 | 1516 | } |
6ac1afbf | 1517 | EXPORT_SYMBOL_GPL(vhost_add_used); |
3a4d5c94 | 1518 | |
8dd014ad DS |
1519 | static int __vhost_add_used_n(struct vhost_virtqueue *vq, |
1520 | struct vring_used_elem *heads, | |
1521 | unsigned count) | |
1522 | { | |
1523 | struct vring_used_elem __user *used; | |
8ea8cf89 | 1524 | u16 old, new; |
8dd014ad DS |
1525 | int start; |
1526 | ||
5fba13b5 | 1527 | start = vq->last_used_idx & (vq->num - 1); |
8dd014ad | 1528 | used = vq->used->ring + start; |
c49e4e57 JW |
1529 | if (count == 1) { |
1530 | if (__put_user(heads[0].id, &used->id)) { | |
1531 | vq_err(vq, "Failed to write used id"); | |
1532 | return -EFAULT; | |
1533 | } | |
1534 | if (__put_user(heads[0].len, &used->len)) { | |
1535 | vq_err(vq, "Failed to write used len"); | |
1536 | return -EFAULT; | |
1537 | } | |
1538 | } else if (__copy_to_user(used, heads, count * sizeof *used)) { | |
8dd014ad DS |
1539 | vq_err(vq, "Failed to write used"); |
1540 | return -EFAULT; | |
1541 | } | |
1542 | if (unlikely(vq->log_used)) { | |
1543 | /* Make sure data is seen before log. */ | |
1544 | smp_wmb(); | |
1545 | /* Log used ring entry write. */ | |
1546 | log_write(vq->log_base, | |
1547 | vq->log_addr + | |
1548 | ((void __user *)used - (void __user *)vq->used), | |
1549 | count * sizeof *used); | |
1550 | } | |
8ea8cf89 MT |
1551 | old = vq->last_used_idx; |
1552 | new = (vq->last_used_idx += count); | |
1553 | /* If the driver never bothers to signal in a very long while, | |
1554 | * used index might wrap around. If that happens, invalidate | |
1555 | * signalled_used index we stored. TODO: make sure driver | |
1556 | * signals at least once in 2^16 and remove this. */ | |
1557 | if (unlikely((u16)(new - vq->signalled_used) < (u16)(new - old))) | |
1558 | vq->signalled_used_valid = false; | |
8dd014ad DS |
1559 | return 0; |
1560 | } | |
1561 | ||
1562 | /* After we've used one of their buffers, we tell them about it. We'll then | |
1563 | * want to notify the guest, using eventfd. */ | |
1564 | int vhost_add_used_n(struct vhost_virtqueue *vq, struct vring_used_elem *heads, | |
1565 | unsigned count) | |
1566 | { | |
1567 | int start, n, r; | |
1568 | ||
5fba13b5 | 1569 | start = vq->last_used_idx & (vq->num - 1); |
8dd014ad DS |
1570 | n = vq->num - start; |
1571 | if (n < count) { | |
1572 | r = __vhost_add_used_n(vq, heads, n); | |
1573 | if (r < 0) | |
1574 | return r; | |
1575 | heads += n; | |
1576 | count -= n; | |
1577 | } | |
1578 | r = __vhost_add_used_n(vq, heads, count); | |
1579 | ||
1580 | /* Make sure buffer is written before we update index. */ | |
1581 | smp_wmb(); | |
3b1bbe89 | 1582 | if (__put_user(cpu_to_vhost16(vq, vq->last_used_idx), &vq->used->idx)) { |
8dd014ad DS |
1583 | vq_err(vq, "Failed to increment used idx"); |
1584 | return -EFAULT; | |
1585 | } | |
1586 | if (unlikely(vq->log_used)) { | |
1587 | /* Log used index update. */ | |
1588 | log_write(vq->log_base, | |
1589 | vq->log_addr + offsetof(struct vring_used, idx), | |
1590 | sizeof vq->used->idx); | |
1591 | if (vq->log_ctx) | |
1592 | eventfd_signal(vq->log_ctx, 1); | |
1593 | } | |
1594 | return r; | |
1595 | } | |
6ac1afbf | 1596 | EXPORT_SYMBOL_GPL(vhost_add_used_n); |
8dd014ad | 1597 | |
8ea8cf89 | 1598 | static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 | 1599 | { |
3b1bbe89 MT |
1600 | __u16 old, new; |
1601 | __virtio16 event; | |
8ea8cf89 | 1602 | bool v; |
0d499356 MT |
1603 | /* Flush out used index updates. This is paired |
1604 | * with the barrier that the Guest executes when enabling | |
1605 | * interrupts. */ | |
1606 | smp_mb(); | |
1607 | ||
ea16c514 | 1608 | if (vhost_has_feature(vq, VIRTIO_F_NOTIFY_ON_EMPTY) && |
8ea8cf89 MT |
1609 | unlikely(vq->avail_idx == vq->last_avail_idx)) |
1610 | return true; | |
1611 | ||
ea16c514 | 1612 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
3b1bbe89 | 1613 | __virtio16 flags; |
8ea8cf89 MT |
1614 | if (__get_user(flags, &vq->avail->flags)) { |
1615 | vq_err(vq, "Failed to get flags"); | |
1616 | return true; | |
1617 | } | |
3b1bbe89 | 1618 | return !(flags & cpu_to_vhost16(vq, VRING_AVAIL_F_NO_INTERRUPT)); |
3a4d5c94 | 1619 | } |
8ea8cf89 MT |
1620 | old = vq->signalled_used; |
1621 | v = vq->signalled_used_valid; | |
1622 | new = vq->signalled_used = vq->last_used_idx; | |
1623 | vq->signalled_used_valid = true; | |
3a4d5c94 | 1624 | |
8ea8cf89 MT |
1625 | if (unlikely(!v)) |
1626 | return true; | |
3a4d5c94 | 1627 | |
64f7f051 | 1628 | if (__get_user(event, vhost_used_event(vq))) { |
8ea8cf89 MT |
1629 | vq_err(vq, "Failed to get used event idx"); |
1630 | return true; | |
1631 | } | |
3b1bbe89 | 1632 | return vring_need_event(vhost16_to_cpu(vq, event), new, old); |
8ea8cf89 MT |
1633 | } |
1634 | ||
1635 | /* This actually signals the guest, using eventfd. */ | |
1636 | void vhost_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq) | |
1637 | { | |
3a4d5c94 | 1638 | /* Signal the Guest tell them we used something up. */ |
8ea8cf89 | 1639 | if (vq->call_ctx && vhost_notify(dev, vq)) |
3a4d5c94 MT |
1640 | eventfd_signal(vq->call_ctx, 1); |
1641 | } | |
6ac1afbf | 1642 | EXPORT_SYMBOL_GPL(vhost_signal); |
3a4d5c94 MT |
1643 | |
1644 | /* And here's the combo meal deal. Supersize me! */ | |
1645 | void vhost_add_used_and_signal(struct vhost_dev *dev, | |
1646 | struct vhost_virtqueue *vq, | |
1647 | unsigned int head, int len) | |
1648 | { | |
1649 | vhost_add_used(vq, head, len); | |
1650 | vhost_signal(dev, vq); | |
1651 | } | |
6ac1afbf | 1652 | EXPORT_SYMBOL_GPL(vhost_add_used_and_signal); |
3a4d5c94 | 1653 | |
8dd014ad DS |
1654 | /* multi-buffer version of vhost_add_used_and_signal */ |
1655 | void vhost_add_used_and_signal_n(struct vhost_dev *dev, | |
1656 | struct vhost_virtqueue *vq, | |
1657 | struct vring_used_elem *heads, unsigned count) | |
1658 | { | |
1659 | vhost_add_used_n(vq, heads, count); | |
1660 | vhost_signal(dev, vq); | |
1661 | } | |
6ac1afbf | 1662 | EXPORT_SYMBOL_GPL(vhost_add_used_and_signal_n); |
8dd014ad | 1663 | |
d4a60603 JW |
1664 | /* return true if we're sure that avaiable ring is empty */ |
1665 | bool vhost_vq_avail_empty(struct vhost_dev *dev, struct vhost_virtqueue *vq) | |
1666 | { | |
1667 | __virtio16 avail_idx; | |
1668 | int r; | |
1669 | ||
1670 | r = __get_user(avail_idx, &vq->avail->idx); | |
1671 | if (r) | |
1672 | return false; | |
1673 | ||
1674 | return vhost16_to_cpu(vq, avail_idx) == vq->avail_idx; | |
1675 | } | |
1676 | EXPORT_SYMBOL_GPL(vhost_vq_avail_empty); | |
1677 | ||
3a4d5c94 | 1678 | /* OK, now we need to know about added descriptors. */ |
8ea8cf89 | 1679 | bool vhost_enable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 | 1680 | { |
3b1bbe89 | 1681 | __virtio16 avail_idx; |
3a4d5c94 | 1682 | int r; |
d47effe1 | 1683 | |
3a4d5c94 MT |
1684 | if (!(vq->used_flags & VRING_USED_F_NO_NOTIFY)) |
1685 | return false; | |
1686 | vq->used_flags &= ~VRING_USED_F_NO_NOTIFY; | |
ea16c514 | 1687 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
2723feaa | 1688 | r = vhost_update_used_flags(vq); |
8ea8cf89 MT |
1689 | if (r) { |
1690 | vq_err(vq, "Failed to enable notification at %p: %d\n", | |
1691 | &vq->used->flags, r); | |
1692 | return false; | |
1693 | } | |
1694 | } else { | |
2723feaa | 1695 | r = vhost_update_avail_event(vq, vq->avail_idx); |
8ea8cf89 MT |
1696 | if (r) { |
1697 | vq_err(vq, "Failed to update avail event index at %p: %d\n", | |
1698 | vhost_avail_event(vq), r); | |
1699 | return false; | |
1700 | } | |
1701 | } | |
3a4d5c94 MT |
1702 | /* They could have slipped one in as we were doing that: make |
1703 | * sure it's written, then check again. */ | |
5659338c | 1704 | smp_mb(); |
8b7347aa | 1705 | r = __get_user(avail_idx, &vq->avail->idx); |
3a4d5c94 MT |
1706 | if (r) { |
1707 | vq_err(vq, "Failed to check avail idx at %p: %d\n", | |
1708 | &vq->avail->idx, r); | |
1709 | return false; | |
1710 | } | |
1711 | ||
3b1bbe89 | 1712 | return vhost16_to_cpu(vq, avail_idx) != vq->avail_idx; |
3a4d5c94 | 1713 | } |
6ac1afbf | 1714 | EXPORT_SYMBOL_GPL(vhost_enable_notify); |
3a4d5c94 MT |
1715 | |
1716 | /* We don't need to be notified again. */ | |
8ea8cf89 | 1717 | void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 MT |
1718 | { |
1719 | int r; | |
d47effe1 | 1720 | |
3a4d5c94 MT |
1721 | if (vq->used_flags & VRING_USED_F_NO_NOTIFY) |
1722 | return; | |
1723 | vq->used_flags |= VRING_USED_F_NO_NOTIFY; | |
ea16c514 | 1724 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
2723feaa | 1725 | r = vhost_update_used_flags(vq); |
8ea8cf89 MT |
1726 | if (r) |
1727 | vq_err(vq, "Failed to enable notification at %p: %d\n", | |
1728 | &vq->used->flags, r); | |
1729 | } | |
3a4d5c94 | 1730 | } |
6ac1afbf AH |
1731 | EXPORT_SYMBOL_GPL(vhost_disable_notify); |
1732 | ||
1733 | static int __init vhost_init(void) | |
1734 | { | |
1735 | return 0; | |
1736 | } | |
1737 | ||
1738 | static void __exit vhost_exit(void) | |
1739 | { | |
1740 | } | |
1741 | ||
1742 | module_init(vhost_init); | |
1743 | module_exit(vhost_exit); | |
1744 | ||
1745 | MODULE_VERSION("0.0.1"); | |
1746 | MODULE_LICENSE("GPL v2"); | |
1747 | MODULE_AUTHOR("Michael S. Tsirkin"); | |
1748 | MODULE_DESCRIPTION("Host kernel accelerator for virtio"); |