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