]>
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> | |
16 | #include <linux/virtio_net.h> | |
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/rcupdate.h> |
22 | #include <linux/poll.h> | |
23 | #include <linux/file.h> | |
24 | #include <linux/highmem.h> | |
5a0e3ad6 | 25 | #include <linux/slab.h> |
c23f3445 | 26 | #include <linux/kthread.h> |
9e3d1957 | 27 | #include <linux/cgroup.h> |
3a4d5c94 MT |
28 | |
29 | #include <linux/net.h> | |
30 | #include <linux/if_packet.h> | |
31 | #include <linux/if_arp.h> | |
32 | ||
3a4d5c94 MT |
33 | #include "vhost.h" |
34 | ||
35 | enum { | |
36 | VHOST_MEMORY_MAX_NREGIONS = 64, | |
37 | VHOST_MEMORY_F_LOG = 0x1, | |
38 | }; | |
39 | ||
bab632d6 MT |
40 | static unsigned vhost_zcopy_mask __read_mostly; |
41 | ||
8ea8cf89 MT |
42 | #define vhost_used_event(vq) ((u16 __user *)&vq->avail->ring[vq->num]) |
43 | #define vhost_avail_event(vq) ((u16 __user *)&vq->used->ring[vq->num]) | |
44 | ||
3a4d5c94 MT |
45 | static void vhost_poll_func(struct file *file, wait_queue_head_t *wqh, |
46 | poll_table *pt) | |
47 | { | |
48 | struct vhost_poll *poll; | |
3a4d5c94 | 49 | |
d47effe1 | 50 | poll = container_of(pt, struct vhost_poll, table); |
3a4d5c94 MT |
51 | poll->wqh = wqh; |
52 | add_wait_queue(wqh, &poll->wait); | |
53 | } | |
54 | ||
55 | static int vhost_poll_wakeup(wait_queue_t *wait, unsigned mode, int sync, | |
56 | void *key) | |
57 | { | |
c23f3445 TH |
58 | struct vhost_poll *poll = container_of(wait, struct vhost_poll, wait); |
59 | ||
3a4d5c94 MT |
60 | if (!((unsigned long)key & poll->mask)) |
61 | return 0; | |
62 | ||
c23f3445 | 63 | vhost_poll_queue(poll); |
3a4d5c94 MT |
64 | return 0; |
65 | } | |
66 | ||
87d6a412 MT |
67 | static void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn) |
68 | { | |
69 | INIT_LIST_HEAD(&work->node); | |
70 | work->fn = fn; | |
71 | init_waitqueue_head(&work->done); | |
72 | work->flushing = 0; | |
73 | work->queue_seq = work->done_seq = 0; | |
74 | } | |
75 | ||
3a4d5c94 | 76 | /* Init poll structure */ |
c23f3445 TH |
77 | void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn, |
78 | unsigned long mask, struct vhost_dev *dev) | |
3a4d5c94 | 79 | { |
3a4d5c94 MT |
80 | init_waitqueue_func_entry(&poll->wait, vhost_poll_wakeup); |
81 | init_poll_funcptr(&poll->table, vhost_poll_func); | |
82 | poll->mask = mask; | |
c23f3445 TH |
83 | poll->dev = dev; |
84 | ||
87d6a412 | 85 | vhost_work_init(&poll->work, fn); |
3a4d5c94 MT |
86 | } |
87 | ||
88 | /* Start polling a file. We add ourselves to file's wait queue. The caller must | |
89 | * keep a reference to a file until after vhost_poll_stop is called. */ | |
90 | void vhost_poll_start(struct vhost_poll *poll, struct file *file) | |
91 | { | |
92 | unsigned long mask; | |
d47effe1 | 93 | |
3a4d5c94 MT |
94 | mask = file->f_op->poll(file, &poll->table); |
95 | if (mask) | |
96 | vhost_poll_wakeup(&poll->wait, 0, 0, (void *)mask); | |
97 | } | |
98 | ||
99 | /* Stop polling a file. After this function returns, it becomes safe to drop the | |
100 | * file reference. You must also flush afterwards. */ | |
101 | void vhost_poll_stop(struct vhost_poll *poll) | |
102 | { | |
103 | remove_wait_queue(poll->wqh, &poll->wait); | |
104 | } | |
105 | ||
0174b0c3 MT |
106 | static bool vhost_work_seq_done(struct vhost_dev *dev, struct vhost_work *work, |
107 | unsigned seq) | |
108 | { | |
109 | int left; | |
d47effe1 | 110 | |
0174b0c3 MT |
111 | spin_lock_irq(&dev->work_lock); |
112 | left = seq - work->done_seq; | |
113 | spin_unlock_irq(&dev->work_lock); | |
114 | return left <= 0; | |
115 | } | |
116 | ||
87d6a412 | 117 | static void vhost_work_flush(struct vhost_dev *dev, struct vhost_work *work) |
3a4d5c94 | 118 | { |
c23f3445 | 119 | unsigned seq; |
c23f3445 TH |
120 | int flushing; |
121 | ||
87d6a412 | 122 | spin_lock_irq(&dev->work_lock); |
c23f3445 TH |
123 | seq = work->queue_seq; |
124 | work->flushing++; | |
87d6a412 | 125 | spin_unlock_irq(&dev->work_lock); |
0174b0c3 | 126 | wait_event(work->done, vhost_work_seq_done(dev, work, seq)); |
87d6a412 | 127 | spin_lock_irq(&dev->work_lock); |
c23f3445 | 128 | flushing = --work->flushing; |
87d6a412 | 129 | spin_unlock_irq(&dev->work_lock); |
c23f3445 | 130 | BUG_ON(flushing < 0); |
3a4d5c94 MT |
131 | } |
132 | ||
87d6a412 MT |
133 | /* Flush any work that has been scheduled. When calling this, don't hold any |
134 | * locks that are also used by the callback. */ | |
135 | void vhost_poll_flush(struct vhost_poll *poll) | |
136 | { | |
137 | vhost_work_flush(poll->dev, &poll->work); | |
138 | } | |
139 | ||
140 | static inline void vhost_work_queue(struct vhost_dev *dev, | |
141 | struct vhost_work *work) | |
3a4d5c94 | 142 | { |
c23f3445 TH |
143 | unsigned long flags; |
144 | ||
145 | spin_lock_irqsave(&dev->work_lock, flags); | |
146 | if (list_empty(&work->node)) { | |
147 | list_add_tail(&work->node, &dev->work_list); | |
148 | work->queue_seq++; | |
149 | wake_up_process(dev->worker); | |
150 | } | |
151 | spin_unlock_irqrestore(&dev->work_lock, flags); | |
3a4d5c94 MT |
152 | } |
153 | ||
87d6a412 MT |
154 | void vhost_poll_queue(struct vhost_poll *poll) |
155 | { | |
156 | vhost_work_queue(poll->dev, &poll->work); | |
157 | } | |
158 | ||
3a4d5c94 MT |
159 | static void vhost_vq_reset(struct vhost_dev *dev, |
160 | struct vhost_virtqueue *vq) | |
161 | { | |
162 | vq->num = 1; | |
163 | vq->desc = NULL; | |
164 | vq->avail = NULL; | |
165 | vq->used = NULL; | |
166 | vq->last_avail_idx = 0; | |
167 | vq->avail_idx = 0; | |
168 | vq->last_used_idx = 0; | |
8ea8cf89 MT |
169 | vq->signalled_used = 0; |
170 | vq->signalled_used_valid = false; | |
3a4d5c94 | 171 | vq->used_flags = 0; |
3a4d5c94 MT |
172 | vq->log_used = false; |
173 | vq->log_addr = -1ull; | |
8dd014ad DS |
174 | vq->vhost_hlen = 0; |
175 | vq->sock_hlen = 0; | |
3a4d5c94 MT |
176 | vq->private_data = NULL; |
177 | vq->log_base = NULL; | |
178 | vq->error_ctx = NULL; | |
179 | vq->error = NULL; | |
180 | vq->kick = NULL; | |
181 | vq->call_ctx = NULL; | |
182 | vq->call = NULL; | |
73a99f08 | 183 | vq->log_ctx = NULL; |
bab632d6 MT |
184 | vq->upend_idx = 0; |
185 | vq->done_idx = 0; | |
186 | vq->ubufs = NULL; | |
3a4d5c94 MT |
187 | } |
188 | ||
c23f3445 TH |
189 | static int vhost_worker(void *data) |
190 | { | |
191 | struct vhost_dev *dev = data; | |
192 | struct vhost_work *work = NULL; | |
193 | unsigned uninitialized_var(seq); | |
194 | ||
64e1c807 MT |
195 | use_mm(dev->mm); |
196 | ||
c23f3445 TH |
197 | for (;;) { |
198 | /* mb paired w/ kthread_stop */ | |
199 | set_current_state(TASK_INTERRUPTIBLE); | |
200 | ||
201 | spin_lock_irq(&dev->work_lock); | |
202 | if (work) { | |
203 | work->done_seq = seq; | |
204 | if (work->flushing) | |
205 | wake_up_all(&work->done); | |
206 | } | |
207 | ||
208 | if (kthread_should_stop()) { | |
209 | spin_unlock_irq(&dev->work_lock); | |
210 | __set_current_state(TASK_RUNNING); | |
64e1c807 | 211 | break; |
c23f3445 TH |
212 | } |
213 | if (!list_empty(&dev->work_list)) { | |
214 | work = list_first_entry(&dev->work_list, | |
215 | struct vhost_work, node); | |
216 | list_del_init(&work->node); | |
217 | seq = work->queue_seq; | |
218 | } else | |
219 | work = NULL; | |
220 | spin_unlock_irq(&dev->work_lock); | |
221 | ||
222 | if (work) { | |
223 | __set_current_state(TASK_RUNNING); | |
224 | work->fn(work); | |
225 | } else | |
226 | schedule(); | |
227 | ||
228 | } | |
64e1c807 MT |
229 | unuse_mm(dev->mm); |
230 | return 0; | |
c23f3445 TH |
231 | } |
232 | ||
bab632d6 MT |
233 | static void vhost_vq_free_iovecs(struct vhost_virtqueue *vq) |
234 | { | |
235 | kfree(vq->indirect); | |
236 | vq->indirect = NULL; | |
237 | kfree(vq->log); | |
238 | vq->log = NULL; | |
239 | kfree(vq->heads); | |
240 | vq->heads = NULL; | |
241 | kfree(vq->ubuf_info); | |
242 | vq->ubuf_info = NULL; | |
243 | } | |
244 | ||
245 | void vhost_enable_zcopy(int vq) | |
246 | { | |
247 | vhost_zcopy_mask |= 0x1 << vq; | |
248 | } | |
249 | ||
e0e9b406 JW |
250 | /* Helper to allocate iovec buffers for all vqs. */ |
251 | static long vhost_dev_alloc_iovecs(struct vhost_dev *dev) | |
252 | { | |
253 | int i; | |
bab632d6 | 254 | bool zcopy; |
d47effe1 | 255 | |
e0e9b406 JW |
256 | for (i = 0; i < dev->nvqs; ++i) { |
257 | dev->vqs[i].indirect = kmalloc(sizeof *dev->vqs[i].indirect * | |
258 | UIO_MAXIOV, GFP_KERNEL); | |
259 | dev->vqs[i].log = kmalloc(sizeof *dev->vqs[i].log * UIO_MAXIOV, | |
260 | GFP_KERNEL); | |
261 | dev->vqs[i].heads = kmalloc(sizeof *dev->vqs[i].heads * | |
262 | UIO_MAXIOV, GFP_KERNEL); | |
bab632d6 MT |
263 | zcopy = vhost_zcopy_mask & (0x1 << i); |
264 | if (zcopy) | |
265 | dev->vqs[i].ubuf_info = | |
266 | kmalloc(sizeof *dev->vqs[i].ubuf_info * | |
267 | UIO_MAXIOV, GFP_KERNEL); | |
e0e9b406 | 268 | if (!dev->vqs[i].indirect || !dev->vqs[i].log || |
bab632d6 MT |
269 | !dev->vqs[i].heads || |
270 | (zcopy && !dev->vqs[i].ubuf_info)) | |
e0e9b406 JW |
271 | goto err_nomem; |
272 | } | |
273 | return 0; | |
d47effe1 | 274 | |
e0e9b406 | 275 | err_nomem: |
bab632d6 MT |
276 | for (; i >= 0; --i) |
277 | vhost_vq_free_iovecs(&dev->vqs[i]); | |
e0e9b406 JW |
278 | return -ENOMEM; |
279 | } | |
280 | ||
281 | static void vhost_dev_free_iovecs(struct vhost_dev *dev) | |
282 | { | |
283 | int i; | |
d47effe1 | 284 | |
bab632d6 MT |
285 | for (i = 0; i < dev->nvqs; ++i) |
286 | vhost_vq_free_iovecs(&dev->vqs[i]); | |
e0e9b406 JW |
287 | } |
288 | ||
3a4d5c94 MT |
289 | long vhost_dev_init(struct vhost_dev *dev, |
290 | struct vhost_virtqueue *vqs, int nvqs) | |
291 | { | |
292 | int i; | |
c23f3445 | 293 | |
3a4d5c94 MT |
294 | dev->vqs = vqs; |
295 | dev->nvqs = nvqs; | |
296 | mutex_init(&dev->mutex); | |
297 | dev->log_ctx = NULL; | |
298 | dev->log_file = NULL; | |
299 | dev->memory = NULL; | |
300 | dev->mm = NULL; | |
c23f3445 TH |
301 | spin_lock_init(&dev->work_lock); |
302 | INIT_LIST_HEAD(&dev->work_list); | |
303 | dev->worker = NULL; | |
3a4d5c94 MT |
304 | |
305 | for (i = 0; i < dev->nvqs; ++i) { | |
e0e9b406 JW |
306 | dev->vqs[i].log = NULL; |
307 | dev->vqs[i].indirect = NULL; | |
308 | dev->vqs[i].heads = NULL; | |
bab632d6 | 309 | dev->vqs[i].ubuf_info = NULL; |
3a4d5c94 MT |
310 | dev->vqs[i].dev = dev; |
311 | mutex_init(&dev->vqs[i].mutex); | |
312 | vhost_vq_reset(dev, dev->vqs + i); | |
313 | if (dev->vqs[i].handle_kick) | |
314 | vhost_poll_init(&dev->vqs[i].poll, | |
c23f3445 | 315 | dev->vqs[i].handle_kick, POLLIN, dev); |
3a4d5c94 | 316 | } |
c23f3445 | 317 | |
3a4d5c94 MT |
318 | return 0; |
319 | } | |
320 | ||
321 | /* Caller should have device mutex */ | |
322 | long vhost_dev_check_owner(struct vhost_dev *dev) | |
323 | { | |
324 | /* Are you the owner? If not, I don't think you mean to do that */ | |
325 | return dev->mm == current->mm ? 0 : -EPERM; | |
326 | } | |
327 | ||
87d6a412 | 328 | struct vhost_attach_cgroups_struct { |
d47effe1 KK |
329 | struct vhost_work work; |
330 | struct task_struct *owner; | |
331 | int ret; | |
87d6a412 MT |
332 | }; |
333 | ||
334 | static void vhost_attach_cgroups_work(struct vhost_work *work) | |
335 | { | |
d47effe1 KK |
336 | struct vhost_attach_cgroups_struct *s; |
337 | ||
338 | s = container_of(work, struct vhost_attach_cgroups_struct, work); | |
339 | s->ret = cgroup_attach_task_all(s->owner, current); | |
87d6a412 MT |
340 | } |
341 | ||
342 | static int vhost_attach_cgroups(struct vhost_dev *dev) | |
343 | { | |
d47effe1 KK |
344 | struct vhost_attach_cgroups_struct attach; |
345 | ||
346 | attach.owner = current; | |
347 | vhost_work_init(&attach.work, vhost_attach_cgroups_work); | |
348 | vhost_work_queue(dev, &attach.work); | |
349 | vhost_work_flush(dev, &attach.work); | |
350 | return attach.ret; | |
87d6a412 MT |
351 | } |
352 | ||
3a4d5c94 MT |
353 | /* Caller should have device mutex */ |
354 | static long vhost_dev_set_owner(struct vhost_dev *dev) | |
355 | { | |
c23f3445 TH |
356 | struct task_struct *worker; |
357 | int err; | |
d47effe1 | 358 | |
3a4d5c94 | 359 | /* Is there an owner already? */ |
c23f3445 TH |
360 | if (dev->mm) { |
361 | err = -EBUSY; | |
362 | goto err_mm; | |
363 | } | |
d47effe1 | 364 | |
3a4d5c94 MT |
365 | /* No owner, become one */ |
366 | dev->mm = get_task_mm(current); | |
c23f3445 TH |
367 | worker = kthread_create(vhost_worker, dev, "vhost-%d", current->pid); |
368 | if (IS_ERR(worker)) { | |
369 | err = PTR_ERR(worker); | |
370 | goto err_worker; | |
371 | } | |
372 | ||
373 | dev->worker = worker; | |
87d6a412 MT |
374 | wake_up_process(worker); /* avoid contributing to loadavg */ |
375 | ||
376 | err = vhost_attach_cgroups(dev); | |
9e3d1957 MT |
377 | if (err) |
378 | goto err_cgroup; | |
c23f3445 | 379 | |
e0e9b406 JW |
380 | err = vhost_dev_alloc_iovecs(dev); |
381 | if (err) | |
382 | goto err_cgroup; | |
383 | ||
3a4d5c94 | 384 | return 0; |
9e3d1957 MT |
385 | err_cgroup: |
386 | kthread_stop(worker); | |
615cc221 | 387 | dev->worker = NULL; |
c23f3445 TH |
388 | err_worker: |
389 | if (dev->mm) | |
390 | mmput(dev->mm); | |
391 | dev->mm = NULL; | |
392 | err_mm: | |
393 | return err; | |
3a4d5c94 MT |
394 | } |
395 | ||
396 | /* Caller should have device mutex */ | |
397 | long vhost_dev_reset_owner(struct vhost_dev *dev) | |
398 | { | |
399 | struct vhost_memory *memory; | |
400 | ||
401 | /* Restore memory to default empty mapping. */ | |
402 | memory = kmalloc(offsetof(struct vhost_memory, regions), GFP_KERNEL); | |
403 | if (!memory) | |
404 | return -ENOMEM; | |
405 | ||
406 | vhost_dev_cleanup(dev); | |
407 | ||
408 | memory->nregions = 0; | |
28457ee6 | 409 | RCU_INIT_POINTER(dev->memory, memory); |
3a4d5c94 MT |
410 | return 0; |
411 | } | |
412 | ||
bab632d6 MT |
413 | /* In case of DMA done not in order in lower device driver for some reason. |
414 | * upend_idx is used to track end of used idx, done_idx is used to track head | |
415 | * of used idx. Once lower device DMA done contiguously, we will signal KVM | |
416 | * guest used idx. | |
417 | */ | |
418 | int vhost_zerocopy_signal_used(struct vhost_virtqueue *vq) | |
419 | { | |
420 | int i; | |
421 | int j = 0; | |
422 | ||
423 | for (i = vq->done_idx; i != vq->upend_idx; i = (i + 1) % UIO_MAXIOV) { | |
424 | if ((vq->heads[i].len == VHOST_DMA_DONE_LEN)) { | |
425 | vq->heads[i].len = VHOST_DMA_CLEAR_LEN; | |
426 | vhost_add_used_and_signal(vq->dev, vq, | |
427 | vq->heads[i].id, 0); | |
428 | ++j; | |
429 | } else | |
430 | break; | |
431 | } | |
432 | if (j) | |
433 | vq->done_idx = i; | |
434 | return j; | |
435 | } | |
436 | ||
3a4d5c94 MT |
437 | /* Caller should have device mutex */ |
438 | void vhost_dev_cleanup(struct vhost_dev *dev) | |
439 | { | |
440 | int i; | |
d47effe1 | 441 | |
3a4d5c94 MT |
442 | for (i = 0; i < dev->nvqs; ++i) { |
443 | if (dev->vqs[i].kick && dev->vqs[i].handle_kick) { | |
444 | vhost_poll_stop(&dev->vqs[i].poll); | |
445 | vhost_poll_flush(&dev->vqs[i].poll); | |
446 | } | |
bab632d6 MT |
447 | /* Wait for all lower device DMAs done. */ |
448 | if (dev->vqs[i].ubufs) | |
449 | vhost_ubuf_put_and_wait(dev->vqs[i].ubufs); | |
450 | ||
451 | /* Signal guest as appropriate. */ | |
452 | vhost_zerocopy_signal_used(&dev->vqs[i]); | |
453 | ||
3a4d5c94 MT |
454 | if (dev->vqs[i].error_ctx) |
455 | eventfd_ctx_put(dev->vqs[i].error_ctx); | |
456 | if (dev->vqs[i].error) | |
457 | fput(dev->vqs[i].error); | |
458 | if (dev->vqs[i].kick) | |
459 | fput(dev->vqs[i].kick); | |
460 | if (dev->vqs[i].call_ctx) | |
461 | eventfd_ctx_put(dev->vqs[i].call_ctx); | |
462 | if (dev->vqs[i].call) | |
463 | fput(dev->vqs[i].call); | |
464 | vhost_vq_reset(dev, dev->vqs + i); | |
465 | } | |
e0e9b406 | 466 | vhost_dev_free_iovecs(dev); |
3a4d5c94 MT |
467 | if (dev->log_ctx) |
468 | eventfd_ctx_put(dev->log_ctx); | |
469 | dev->log_ctx = NULL; | |
470 | if (dev->log_file) | |
471 | fput(dev->log_file); | |
472 | dev->log_file = NULL; | |
473 | /* No one will access memory at this point */ | |
28457ee6 AB |
474 | kfree(rcu_dereference_protected(dev->memory, |
475 | lockdep_is_held(&dev->mutex))); | |
476 | RCU_INIT_POINTER(dev->memory, NULL); | |
c23f3445 | 477 | WARN_ON(!list_empty(&dev->work_list)); |
78b620ce ED |
478 | if (dev->worker) { |
479 | kthread_stop(dev->worker); | |
480 | dev->worker = NULL; | |
481 | } | |
533a19b4 MT |
482 | if (dev->mm) |
483 | mmput(dev->mm); | |
484 | dev->mm = NULL; | |
3a4d5c94 MT |
485 | } |
486 | ||
487 | static int log_access_ok(void __user *log_base, u64 addr, unsigned long sz) | |
488 | { | |
489 | u64 a = addr / VHOST_PAGE_SIZE / 8; | |
d47effe1 | 490 | |
3a4d5c94 MT |
491 | /* Make sure 64 bit math will not overflow. */ |
492 | if (a > ULONG_MAX - (unsigned long)log_base || | |
493 | a + (unsigned long)log_base > ULONG_MAX) | |
6d97e55f | 494 | return 0; |
3a4d5c94 MT |
495 | |
496 | return access_ok(VERIFY_WRITE, log_base + a, | |
497 | (sz + VHOST_PAGE_SIZE * 8 - 1) / VHOST_PAGE_SIZE / 8); | |
498 | } | |
499 | ||
500 | /* Caller should have vq mutex and device mutex. */ | |
501 | static int vq_memory_access_ok(void __user *log_base, struct vhost_memory *mem, | |
502 | int log_all) | |
503 | { | |
504 | int i; | |
179b284e | 505 | |
f8322fbe MT |
506 | if (!mem) |
507 | return 0; | |
179b284e | 508 | |
3a4d5c94 MT |
509 | for (i = 0; i < mem->nregions; ++i) { |
510 | struct vhost_memory_region *m = mem->regions + i; | |
511 | unsigned long a = m->userspace_addr; | |
512 | if (m->memory_size > ULONG_MAX) | |
513 | return 0; | |
514 | else if (!access_ok(VERIFY_WRITE, (void __user *)a, | |
515 | m->memory_size)) | |
516 | return 0; | |
517 | else if (log_all && !log_access_ok(log_base, | |
518 | m->guest_phys_addr, | |
519 | m->memory_size)) | |
520 | return 0; | |
521 | } | |
522 | return 1; | |
523 | } | |
524 | ||
525 | /* Can we switch to this memory table? */ | |
526 | /* Caller should have device mutex but not vq mutex */ | |
527 | static int memory_access_ok(struct vhost_dev *d, struct vhost_memory *mem, | |
528 | int log_all) | |
529 | { | |
530 | int i; | |
d47effe1 | 531 | |
3a4d5c94 MT |
532 | for (i = 0; i < d->nvqs; ++i) { |
533 | int ok; | |
534 | mutex_lock(&d->vqs[i].mutex); | |
535 | /* If ring is inactive, will check when it's enabled. */ | |
536 | if (d->vqs[i].private_data) | |
537 | ok = vq_memory_access_ok(d->vqs[i].log_base, mem, | |
538 | log_all); | |
539 | else | |
540 | ok = 1; | |
541 | mutex_unlock(&d->vqs[i].mutex); | |
542 | if (!ok) | |
543 | return 0; | |
544 | } | |
545 | return 1; | |
546 | } | |
547 | ||
8ea8cf89 | 548 | static int vq_access_ok(struct vhost_dev *d, unsigned int num, |
3a4d5c94 MT |
549 | struct vring_desc __user *desc, |
550 | struct vring_avail __user *avail, | |
551 | struct vring_used __user *used) | |
552 | { | |
8ea8cf89 | 553 | size_t s = vhost_has_feature(d, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; |
3a4d5c94 MT |
554 | return access_ok(VERIFY_READ, desc, num * sizeof *desc) && |
555 | access_ok(VERIFY_READ, avail, | |
8ea8cf89 | 556 | sizeof *avail + num * sizeof *avail->ring + s) && |
3a4d5c94 | 557 | access_ok(VERIFY_WRITE, used, |
8ea8cf89 | 558 | sizeof *used + num * sizeof *used->ring + s); |
3a4d5c94 MT |
559 | } |
560 | ||
561 | /* Can we log writes? */ | |
562 | /* Caller should have device mutex but not vq mutex */ | |
563 | int vhost_log_access_ok(struct vhost_dev *dev) | |
564 | { | |
28457ee6 AB |
565 | struct vhost_memory *mp; |
566 | ||
567 | mp = rcu_dereference_protected(dev->memory, | |
568 | lockdep_is_held(&dev->mutex)); | |
569 | return memory_access_ok(dev, mp, 1); | |
3a4d5c94 MT |
570 | } |
571 | ||
572 | /* Verify access for write logging. */ | |
573 | /* Caller should have vq mutex and device mutex */ | |
8ea8cf89 MT |
574 | static int vq_log_access_ok(struct vhost_dev *d, struct vhost_virtqueue *vq, |
575 | void __user *log_base) | |
3a4d5c94 | 576 | { |
28457ee6 | 577 | struct vhost_memory *mp; |
8ea8cf89 | 578 | size_t s = vhost_has_feature(d, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; |
28457ee6 AB |
579 | |
580 | mp = rcu_dereference_protected(vq->dev->memory, | |
581 | lockdep_is_held(&vq->mutex)); | |
582 | return vq_memory_access_ok(log_base, mp, | |
3a4d5c94 MT |
583 | vhost_has_feature(vq->dev, VHOST_F_LOG_ALL)) && |
584 | (!vq->log_used || log_access_ok(log_base, vq->log_addr, | |
585 | sizeof *vq->used + | |
8ea8cf89 | 586 | vq->num * sizeof *vq->used->ring + s)); |
3a4d5c94 MT |
587 | } |
588 | ||
589 | /* Can we start vq? */ | |
590 | /* Caller should have vq mutex and device mutex */ | |
591 | int vhost_vq_access_ok(struct vhost_virtqueue *vq) | |
592 | { | |
8ea8cf89 MT |
593 | return vq_access_ok(vq->dev, vq->num, vq->desc, vq->avail, vq->used) && |
594 | vq_log_access_ok(vq->dev, vq, vq->log_base); | |
3a4d5c94 MT |
595 | } |
596 | ||
597 | static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m) | |
598 | { | |
599 | struct vhost_memory mem, *newmem, *oldmem; | |
600 | unsigned long size = offsetof(struct vhost_memory, regions); | |
d47effe1 | 601 | |
7ad9c9d2 TY |
602 | if (copy_from_user(&mem, m, size)) |
603 | return -EFAULT; | |
3a4d5c94 MT |
604 | if (mem.padding) |
605 | return -EOPNOTSUPP; | |
606 | if (mem.nregions > VHOST_MEMORY_MAX_NREGIONS) | |
607 | return -E2BIG; | |
608 | newmem = kmalloc(size + mem.nregions * sizeof *m->regions, GFP_KERNEL); | |
609 | if (!newmem) | |
610 | return -ENOMEM; | |
611 | ||
612 | memcpy(newmem, &mem, size); | |
7ad9c9d2 TY |
613 | if (copy_from_user(newmem->regions, m->regions, |
614 | mem.nregions * sizeof *m->regions)) { | |
3a4d5c94 | 615 | kfree(newmem); |
7ad9c9d2 | 616 | return -EFAULT; |
3a4d5c94 MT |
617 | } |
618 | ||
d47effe1 KK |
619 | if (!memory_access_ok(d, newmem, |
620 | vhost_has_feature(d, VHOST_F_LOG_ALL))) { | |
a02c3789 | 621 | kfree(newmem); |
3a4d5c94 | 622 | return -EFAULT; |
a02c3789 | 623 | } |
28457ee6 AB |
624 | oldmem = rcu_dereference_protected(d->memory, |
625 | lockdep_is_held(&d->mutex)); | |
3a4d5c94 MT |
626 | rcu_assign_pointer(d->memory, newmem); |
627 | synchronize_rcu(); | |
628 | kfree(oldmem); | |
629 | return 0; | |
630 | } | |
631 | ||
f59281da | 632 | int vhost_init_used(struct vhost_virtqueue *vq) |
3a4d5c94 | 633 | { |
f59281da JW |
634 | int r; |
635 | if (!vq->private_data) | |
636 | return 0; | |
d47effe1 | 637 | |
f59281da | 638 | r = put_user(vq->used_flags, &vq->used->flags); |
3a4d5c94 MT |
639 | if (r) |
640 | return r; | |
8ea8cf89 | 641 | vq->signalled_used_valid = false; |
f59281da | 642 | return get_user(vq->last_used_idx, &vq->used->idx); |
3a4d5c94 MT |
643 | } |
644 | ||
645 | static long vhost_set_vring(struct vhost_dev *d, int ioctl, void __user *argp) | |
646 | { | |
647 | struct file *eventfp, *filep = NULL, | |
648 | *pollstart = NULL, *pollstop = NULL; | |
649 | struct eventfd_ctx *ctx = NULL; | |
650 | u32 __user *idxp = argp; | |
651 | struct vhost_virtqueue *vq; | |
652 | struct vhost_vring_state s; | |
653 | struct vhost_vring_file f; | |
654 | struct vhost_vring_addr a; | |
655 | u32 idx; | |
656 | long r; | |
657 | ||
658 | r = get_user(idx, idxp); | |
659 | if (r < 0) | |
660 | return r; | |
0f3d9a17 | 661 | if (idx >= d->nvqs) |
3a4d5c94 MT |
662 | return -ENOBUFS; |
663 | ||
664 | vq = d->vqs + idx; | |
665 | ||
666 | mutex_lock(&vq->mutex); | |
667 | ||
668 | switch (ioctl) { | |
669 | case VHOST_SET_VRING_NUM: | |
670 | /* Resizing ring with an active backend? | |
671 | * You don't want to do that. */ | |
672 | if (vq->private_data) { | |
673 | r = -EBUSY; | |
674 | break; | |
675 | } | |
7ad9c9d2 TY |
676 | if (copy_from_user(&s, argp, sizeof s)) { |
677 | r = -EFAULT; | |
3a4d5c94 | 678 | break; |
7ad9c9d2 | 679 | } |
3a4d5c94 MT |
680 | if (!s.num || s.num > 0xffff || (s.num & (s.num - 1))) { |
681 | r = -EINVAL; | |
682 | break; | |
683 | } | |
684 | vq->num = s.num; | |
685 | break; | |
686 | case VHOST_SET_VRING_BASE: | |
687 | /* Moving base with an active backend? | |
688 | * You don't want to do that. */ | |
689 | if (vq->private_data) { | |
690 | r = -EBUSY; | |
691 | break; | |
692 | } | |
7ad9c9d2 TY |
693 | if (copy_from_user(&s, argp, sizeof s)) { |
694 | r = -EFAULT; | |
3a4d5c94 | 695 | break; |
7ad9c9d2 | 696 | } |
3a4d5c94 MT |
697 | if (s.num > 0xffff) { |
698 | r = -EINVAL; | |
699 | break; | |
700 | } | |
701 | vq->last_avail_idx = s.num; | |
702 | /* Forget the cached index value. */ | |
703 | vq->avail_idx = vq->last_avail_idx; | |
704 | break; | |
705 | case VHOST_GET_VRING_BASE: | |
706 | s.index = idx; | |
707 | s.num = vq->last_avail_idx; | |
7ad9c9d2 TY |
708 | if (copy_to_user(argp, &s, sizeof s)) |
709 | r = -EFAULT; | |
3a4d5c94 MT |
710 | break; |
711 | case VHOST_SET_VRING_ADDR: | |
7ad9c9d2 TY |
712 | if (copy_from_user(&a, argp, sizeof a)) { |
713 | r = -EFAULT; | |
3a4d5c94 | 714 | break; |
7ad9c9d2 | 715 | } |
3a4d5c94 MT |
716 | if (a.flags & ~(0x1 << VHOST_VRING_F_LOG)) { |
717 | r = -EOPNOTSUPP; | |
718 | break; | |
719 | } | |
720 | /* For 32bit, verify that the top 32bits of the user | |
721 | data are set to zero. */ | |
722 | if ((u64)(unsigned long)a.desc_user_addr != a.desc_user_addr || | |
723 | (u64)(unsigned long)a.used_user_addr != a.used_user_addr || | |
724 | (u64)(unsigned long)a.avail_user_addr != a.avail_user_addr) { | |
725 | r = -EFAULT; | |
726 | break; | |
727 | } | |
728 | if ((a.avail_user_addr & (sizeof *vq->avail->ring - 1)) || | |
729 | (a.used_user_addr & (sizeof *vq->used->ring - 1)) || | |
730 | (a.log_guest_addr & (sizeof *vq->used->ring - 1))) { | |
731 | r = -EINVAL; | |
732 | break; | |
733 | } | |
734 | ||
735 | /* We only verify access here if backend is configured. | |
736 | * If it is not, we don't as size might not have been setup. | |
737 | * We will verify when backend is configured. */ | |
738 | if (vq->private_data) { | |
8ea8cf89 | 739 | if (!vq_access_ok(d, vq->num, |
3a4d5c94 MT |
740 | (void __user *)(unsigned long)a.desc_user_addr, |
741 | (void __user *)(unsigned long)a.avail_user_addr, | |
742 | (void __user *)(unsigned long)a.used_user_addr)) { | |
743 | r = -EINVAL; | |
744 | break; | |
745 | } | |
746 | ||
747 | /* Also validate log access for used ring if enabled. */ | |
748 | if ((a.flags & (0x1 << VHOST_VRING_F_LOG)) && | |
749 | !log_access_ok(vq->log_base, a.log_guest_addr, | |
750 | sizeof *vq->used + | |
751 | vq->num * sizeof *vq->used->ring)) { | |
752 | r = -EINVAL; | |
753 | break; | |
754 | } | |
755 | } | |
756 | ||
3a4d5c94 MT |
757 | vq->log_used = !!(a.flags & (0x1 << VHOST_VRING_F_LOG)); |
758 | vq->desc = (void __user *)(unsigned long)a.desc_user_addr; | |
759 | vq->avail = (void __user *)(unsigned long)a.avail_user_addr; | |
760 | vq->log_addr = a.log_guest_addr; | |
761 | vq->used = (void __user *)(unsigned long)a.used_user_addr; | |
762 | break; | |
763 | case VHOST_SET_VRING_KICK: | |
7ad9c9d2 TY |
764 | if (copy_from_user(&f, argp, sizeof f)) { |
765 | r = -EFAULT; | |
3a4d5c94 | 766 | break; |
7ad9c9d2 | 767 | } |
3a4d5c94 | 768 | eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd); |
535297a6 MT |
769 | if (IS_ERR(eventfp)) { |
770 | r = PTR_ERR(eventfp); | |
771 | break; | |
772 | } | |
3a4d5c94 MT |
773 | if (eventfp != vq->kick) { |
774 | pollstop = filep = vq->kick; | |
775 | pollstart = vq->kick = eventfp; | |
776 | } else | |
777 | filep = eventfp; | |
778 | break; | |
779 | case VHOST_SET_VRING_CALL: | |
7ad9c9d2 TY |
780 | if (copy_from_user(&f, argp, sizeof f)) { |
781 | r = -EFAULT; | |
3a4d5c94 | 782 | break; |
7ad9c9d2 | 783 | } |
3a4d5c94 | 784 | eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd); |
535297a6 MT |
785 | if (IS_ERR(eventfp)) { |
786 | r = PTR_ERR(eventfp); | |
787 | break; | |
788 | } | |
3a4d5c94 MT |
789 | if (eventfp != vq->call) { |
790 | filep = vq->call; | |
791 | ctx = vq->call_ctx; | |
792 | vq->call = eventfp; | |
793 | vq->call_ctx = eventfp ? | |
794 | eventfd_ctx_fileget(eventfp) : NULL; | |
795 | } else | |
796 | filep = eventfp; | |
797 | break; | |
798 | case VHOST_SET_VRING_ERR: | |
7ad9c9d2 TY |
799 | if (copy_from_user(&f, argp, sizeof f)) { |
800 | r = -EFAULT; | |
3a4d5c94 | 801 | break; |
7ad9c9d2 | 802 | } |
3a4d5c94 | 803 | eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd); |
535297a6 MT |
804 | if (IS_ERR(eventfp)) { |
805 | r = PTR_ERR(eventfp); | |
806 | break; | |
807 | } | |
3a4d5c94 MT |
808 | if (eventfp != vq->error) { |
809 | filep = vq->error; | |
810 | vq->error = eventfp; | |
811 | ctx = vq->error_ctx; | |
812 | vq->error_ctx = eventfp ? | |
813 | eventfd_ctx_fileget(eventfp) : NULL; | |
814 | } else | |
815 | filep = eventfp; | |
816 | break; | |
817 | default: | |
818 | r = -ENOIOCTLCMD; | |
819 | } | |
820 | ||
821 | if (pollstop && vq->handle_kick) | |
822 | vhost_poll_stop(&vq->poll); | |
823 | ||
824 | if (ctx) | |
825 | eventfd_ctx_put(ctx); | |
826 | if (filep) | |
827 | fput(filep); | |
828 | ||
829 | if (pollstart && vq->handle_kick) | |
830 | vhost_poll_start(&vq->poll, vq->kick); | |
831 | ||
832 | mutex_unlock(&vq->mutex); | |
833 | ||
834 | if (pollstop && vq->handle_kick) | |
835 | vhost_poll_flush(&vq->poll); | |
836 | return r; | |
837 | } | |
838 | ||
839 | /* Caller must have device mutex */ | |
840 | long vhost_dev_ioctl(struct vhost_dev *d, unsigned int ioctl, unsigned long arg) | |
841 | { | |
842 | void __user *argp = (void __user *)arg; | |
843 | struct file *eventfp, *filep = NULL; | |
844 | struct eventfd_ctx *ctx = NULL; | |
845 | u64 p; | |
846 | long r; | |
847 | int i, fd; | |
848 | ||
849 | /* If you are not the owner, you can become one */ | |
850 | if (ioctl == VHOST_SET_OWNER) { | |
851 | r = vhost_dev_set_owner(d); | |
852 | goto done; | |
853 | } | |
854 | ||
855 | /* You must be the owner to do anything else */ | |
856 | r = vhost_dev_check_owner(d); | |
857 | if (r) | |
858 | goto done; | |
859 | ||
860 | switch (ioctl) { | |
861 | case VHOST_SET_MEM_TABLE: | |
862 | r = vhost_set_memory(d, argp); | |
863 | break; | |
864 | case VHOST_SET_LOG_BASE: | |
7ad9c9d2 TY |
865 | if (copy_from_user(&p, argp, sizeof p)) { |
866 | r = -EFAULT; | |
3a4d5c94 | 867 | break; |
7ad9c9d2 | 868 | } |
3a4d5c94 MT |
869 | if ((u64)(unsigned long)p != p) { |
870 | r = -EFAULT; | |
871 | break; | |
872 | } | |
873 | for (i = 0; i < d->nvqs; ++i) { | |
874 | struct vhost_virtqueue *vq; | |
875 | void __user *base = (void __user *)(unsigned long)p; | |
876 | vq = d->vqs + i; | |
877 | mutex_lock(&vq->mutex); | |
878 | /* If ring is inactive, will check when it's enabled. */ | |
8ea8cf89 | 879 | if (vq->private_data && !vq_log_access_ok(d, vq, base)) |
3a4d5c94 MT |
880 | r = -EFAULT; |
881 | else | |
882 | vq->log_base = base; | |
883 | mutex_unlock(&vq->mutex); | |
884 | } | |
885 | break; | |
886 | case VHOST_SET_LOG_FD: | |
887 | r = get_user(fd, (int __user *)argp); | |
888 | if (r < 0) | |
889 | break; | |
890 | eventfp = fd == -1 ? NULL : eventfd_fget(fd); | |
891 | if (IS_ERR(eventfp)) { | |
892 | r = PTR_ERR(eventfp); | |
893 | break; | |
894 | } | |
895 | if (eventfp != d->log_file) { | |
896 | filep = d->log_file; | |
897 | ctx = d->log_ctx; | |
898 | d->log_ctx = eventfp ? | |
899 | eventfd_ctx_fileget(eventfp) : NULL; | |
900 | } else | |
901 | filep = eventfp; | |
902 | for (i = 0; i < d->nvqs; ++i) { | |
903 | mutex_lock(&d->vqs[i].mutex); | |
904 | d->vqs[i].log_ctx = d->log_ctx; | |
905 | mutex_unlock(&d->vqs[i].mutex); | |
906 | } | |
907 | if (ctx) | |
908 | eventfd_ctx_put(ctx); | |
909 | if (filep) | |
910 | fput(filep); | |
911 | break; | |
912 | default: | |
913 | r = vhost_set_vring(d, ioctl, argp); | |
914 | break; | |
915 | } | |
916 | done: | |
917 | return r; | |
918 | } | |
919 | ||
920 | static const struct vhost_memory_region *find_region(struct vhost_memory *mem, | |
921 | __u64 addr, __u32 len) | |
922 | { | |
923 | struct vhost_memory_region *reg; | |
924 | int i; | |
d47effe1 | 925 | |
3a4d5c94 MT |
926 | /* linear search is not brilliant, but we really have on the order of 6 |
927 | * regions in practice */ | |
928 | for (i = 0; i < mem->nregions; ++i) { | |
929 | reg = mem->regions + i; | |
930 | if (reg->guest_phys_addr <= addr && | |
931 | reg->guest_phys_addr + reg->memory_size - 1 >= addr) | |
932 | return reg; | |
933 | } | |
934 | return NULL; | |
935 | } | |
936 | ||
937 | /* TODO: This is really inefficient. We need something like get_user() | |
938 | * (instruction directly accesses the data, with an exception table entry | |
939 | * returning -EFAULT). See Documentation/x86/exception-tables.txt. | |
940 | */ | |
941 | static int set_bit_to_user(int nr, void __user *addr) | |
942 | { | |
943 | unsigned long log = (unsigned long)addr; | |
944 | struct page *page; | |
945 | void *base; | |
946 | int bit = nr + (log % PAGE_SIZE) * 8; | |
947 | int r; | |
d47effe1 | 948 | |
3a4d5c94 | 949 | r = get_user_pages_fast(log, 1, 1, &page); |
d6db3f5c | 950 | if (r < 0) |
3a4d5c94 | 951 | return r; |
d6db3f5c | 952 | BUG_ON(r != 1); |
3a4d5c94 MT |
953 | base = kmap_atomic(page, KM_USER0); |
954 | set_bit(bit, base); | |
955 | kunmap_atomic(base, KM_USER0); | |
956 | set_page_dirty_lock(page); | |
957 | put_page(page); | |
958 | return 0; | |
959 | } | |
960 | ||
961 | static int log_write(void __user *log_base, | |
962 | u64 write_address, u64 write_length) | |
963 | { | |
28831ee6 | 964 | u64 write_page = write_address / VHOST_PAGE_SIZE; |
3a4d5c94 | 965 | int r; |
d47effe1 | 966 | |
3a4d5c94 MT |
967 | if (!write_length) |
968 | return 0; | |
3bf9be40 | 969 | write_length += write_address % VHOST_PAGE_SIZE; |
3a4d5c94 MT |
970 | for (;;) { |
971 | u64 base = (u64)(unsigned long)log_base; | |
28831ee6 MT |
972 | u64 log = base + write_page / 8; |
973 | int bit = write_page % 8; | |
3a4d5c94 MT |
974 | if ((u64)(unsigned long)log != log) |
975 | return -EFAULT; | |
976 | r = set_bit_to_user(bit, (void __user *)(unsigned long)log); | |
977 | if (r < 0) | |
978 | return r; | |
979 | if (write_length <= VHOST_PAGE_SIZE) | |
980 | break; | |
981 | write_length -= VHOST_PAGE_SIZE; | |
28831ee6 | 982 | write_page += 1; |
3a4d5c94 MT |
983 | } |
984 | return r; | |
985 | } | |
986 | ||
987 | int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log, | |
988 | unsigned int log_num, u64 len) | |
989 | { | |
990 | int i, r; | |
991 | ||
992 | /* Make sure data written is seen before log. */ | |
5659338c | 993 | smp_wmb(); |
3a4d5c94 MT |
994 | for (i = 0; i < log_num; ++i) { |
995 | u64 l = min(log[i].len, len); | |
996 | r = log_write(vq->log_base, log[i].addr, l); | |
997 | if (r < 0) | |
998 | return r; | |
999 | len -= l; | |
5786aee8 MT |
1000 | if (!len) { |
1001 | if (vq->log_ctx) | |
1002 | eventfd_signal(vq->log_ctx, 1); | |
3a4d5c94 | 1003 | return 0; |
5786aee8 | 1004 | } |
3a4d5c94 | 1005 | } |
3a4d5c94 MT |
1006 | /* Length written exceeds what we have stored. This is a bug. */ |
1007 | BUG(); | |
1008 | return 0; | |
1009 | } | |
1010 | ||
a8d3782f CH |
1011 | static int translate_desc(struct vhost_dev *dev, u64 addr, u32 len, |
1012 | struct iovec iov[], int iov_size) | |
3a4d5c94 MT |
1013 | { |
1014 | const struct vhost_memory_region *reg; | |
1015 | struct vhost_memory *mem; | |
1016 | struct iovec *_iov; | |
1017 | u64 s = 0; | |
1018 | int ret = 0; | |
1019 | ||
1020 | rcu_read_lock(); | |
1021 | ||
1022 | mem = rcu_dereference(dev->memory); | |
1023 | while ((u64)len > s) { | |
1024 | u64 size; | |
7b3384fc | 1025 | if (unlikely(ret >= iov_size)) { |
3a4d5c94 MT |
1026 | ret = -ENOBUFS; |
1027 | break; | |
1028 | } | |
1029 | reg = find_region(mem, addr, len); | |
7b3384fc | 1030 | if (unlikely(!reg)) { |
3a4d5c94 MT |
1031 | ret = -EFAULT; |
1032 | break; | |
1033 | } | |
1034 | _iov = iov + ret; | |
1035 | size = reg->memory_size - addr + reg->guest_phys_addr; | |
1036 | _iov->iov_len = min((u64)len, size); | |
a8d3782f | 1037 | _iov->iov_base = (void __user *)(unsigned long) |
3a4d5c94 MT |
1038 | (reg->userspace_addr + addr - reg->guest_phys_addr); |
1039 | s += size; | |
1040 | addr += size; | |
1041 | ++ret; | |
1042 | } | |
1043 | ||
1044 | rcu_read_unlock(); | |
1045 | return ret; | |
1046 | } | |
1047 | ||
1048 | /* Each buffer in the virtqueues is actually a chain of descriptors. This | |
1049 | * function returns the next descriptor in the chain, | |
1050 | * or -1U if we're at the end. */ | |
1051 | static unsigned next_desc(struct vring_desc *desc) | |
1052 | { | |
1053 | unsigned int next; | |
1054 | ||
1055 | /* If this descriptor says it doesn't chain, we're done. */ | |
1056 | if (!(desc->flags & VRING_DESC_F_NEXT)) | |
1057 | return -1U; | |
1058 | ||
1059 | /* Check they're not leading us off end of descriptors. */ | |
1060 | next = desc->next; | |
1061 | /* Make sure compiler knows to grab that: we don't want it changing! */ | |
1062 | /* We will use the result as an index in an array, so most | |
1063 | * architectures only need a compiler barrier here. */ | |
1064 | read_barrier_depends(); | |
1065 | ||
1066 | return next; | |
1067 | } | |
1068 | ||
7b3384fc MT |
1069 | static int get_indirect(struct vhost_dev *dev, struct vhost_virtqueue *vq, |
1070 | struct iovec iov[], unsigned int iov_size, | |
1071 | unsigned int *out_num, unsigned int *in_num, | |
1072 | struct vhost_log *log, unsigned int *log_num, | |
1073 | struct vring_desc *indirect) | |
3a4d5c94 MT |
1074 | { |
1075 | struct vring_desc desc; | |
1076 | unsigned int i = 0, count, found = 0; | |
1077 | int ret; | |
1078 | ||
1079 | /* Sanity check */ | |
7b3384fc | 1080 | if (unlikely(indirect->len % sizeof desc)) { |
3a4d5c94 MT |
1081 | vq_err(vq, "Invalid length in indirect descriptor: " |
1082 | "len 0x%llx not multiple of 0x%zx\n", | |
1083 | (unsigned long long)indirect->len, | |
1084 | sizeof desc); | |
1085 | return -EINVAL; | |
1086 | } | |
1087 | ||
1088 | ret = translate_desc(dev, indirect->addr, indirect->len, vq->indirect, | |
e0e9b406 | 1089 | UIO_MAXIOV); |
7b3384fc | 1090 | if (unlikely(ret < 0)) { |
3a4d5c94 MT |
1091 | vq_err(vq, "Translation failure %d in indirect.\n", ret); |
1092 | return ret; | |
1093 | } | |
1094 | ||
1095 | /* We will use the result as an address to read from, so most | |
1096 | * architectures only need a compiler barrier here. */ | |
1097 | read_barrier_depends(); | |
1098 | ||
1099 | count = indirect->len / sizeof desc; | |
1100 | /* Buffers are chained via a 16 bit next field, so | |
1101 | * we can have at most 2^16 of these. */ | |
7b3384fc | 1102 | if (unlikely(count > USHRT_MAX + 1)) { |
3a4d5c94 MT |
1103 | vq_err(vq, "Indirect buffer length too big: %d\n", |
1104 | indirect->len); | |
1105 | return -E2BIG; | |
1106 | } | |
1107 | ||
1108 | do { | |
1109 | unsigned iov_count = *in_num + *out_num; | |
7b3384fc | 1110 | if (unlikely(++found > count)) { |
3a4d5c94 MT |
1111 | vq_err(vq, "Loop detected: last one at %u " |
1112 | "indirect size %u\n", | |
1113 | i, count); | |
1114 | return -EINVAL; | |
1115 | } | |
d47effe1 KK |
1116 | if (unlikely(memcpy_fromiovec((unsigned char *)&desc, |
1117 | vq->indirect, sizeof desc))) { | |
3a4d5c94 MT |
1118 | vq_err(vq, "Failed indirect descriptor: idx %d, %zx\n", |
1119 | i, (size_t)indirect->addr + i * sizeof desc); | |
1120 | return -EINVAL; | |
1121 | } | |
7b3384fc | 1122 | if (unlikely(desc.flags & VRING_DESC_F_INDIRECT)) { |
3a4d5c94 MT |
1123 | vq_err(vq, "Nested indirect descriptor: idx %d, %zx\n", |
1124 | i, (size_t)indirect->addr + i * sizeof desc); | |
1125 | return -EINVAL; | |
1126 | } | |
1127 | ||
1128 | ret = translate_desc(dev, desc.addr, desc.len, iov + iov_count, | |
1129 | iov_size - iov_count); | |
7b3384fc | 1130 | if (unlikely(ret < 0)) { |
3a4d5c94 MT |
1131 | vq_err(vq, "Translation failure %d indirect idx %d\n", |
1132 | ret, i); | |
1133 | return ret; | |
1134 | } | |
1135 | /* If this is an input descriptor, increment that count. */ | |
1136 | if (desc.flags & VRING_DESC_F_WRITE) { | |
1137 | *in_num += ret; | |
1138 | if (unlikely(log)) { | |
1139 | log[*log_num].addr = desc.addr; | |
1140 | log[*log_num].len = desc.len; | |
1141 | ++*log_num; | |
1142 | } | |
1143 | } else { | |
1144 | /* If it's an output descriptor, they're all supposed | |
1145 | * to come before any input descriptors. */ | |
7b3384fc | 1146 | if (unlikely(*in_num)) { |
3a4d5c94 MT |
1147 | vq_err(vq, "Indirect descriptor " |
1148 | "has out after in: idx %d\n", i); | |
1149 | return -EINVAL; | |
1150 | } | |
1151 | *out_num += ret; | |
1152 | } | |
1153 | } while ((i = next_desc(&desc)) != -1); | |
1154 | return 0; | |
1155 | } | |
1156 | ||
1157 | /* This looks in the virtqueue and for the first available buffer, and converts | |
1158 | * it to an iovec for convenient access. Since descriptors consist of some | |
1159 | * number of output then some number of input descriptors, it's actually two | |
1160 | * iovecs, but we pack them into one and note how many of each there were. | |
1161 | * | |
d5675bd2 MT |
1162 | * This function returns the descriptor number found, or vq->num (which is |
1163 | * never a valid descriptor number) if none was found. A negative code is | |
1164 | * returned on error. */ | |
1165 | int vhost_get_vq_desc(struct vhost_dev *dev, struct vhost_virtqueue *vq, | |
1166 | struct iovec iov[], unsigned int iov_size, | |
1167 | unsigned int *out_num, unsigned int *in_num, | |
1168 | struct vhost_log *log, unsigned int *log_num) | |
3a4d5c94 MT |
1169 | { |
1170 | struct vring_desc desc; | |
1171 | unsigned int i, head, found = 0; | |
1172 | u16 last_avail_idx; | |
1173 | int ret; | |
1174 | ||
1175 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
1176 | last_avail_idx = vq->last_avail_idx; | |
8b7347aa | 1177 | if (unlikely(__get_user(vq->avail_idx, &vq->avail->idx))) { |
3a4d5c94 MT |
1178 | vq_err(vq, "Failed to access avail idx at %p\n", |
1179 | &vq->avail->idx); | |
d5675bd2 | 1180 | return -EFAULT; |
3a4d5c94 MT |
1181 | } |
1182 | ||
7b3384fc | 1183 | if (unlikely((u16)(vq->avail_idx - last_avail_idx) > vq->num)) { |
3a4d5c94 MT |
1184 | vq_err(vq, "Guest moved used index from %u to %u", |
1185 | last_avail_idx, vq->avail_idx); | |
d5675bd2 | 1186 | return -EFAULT; |
3a4d5c94 MT |
1187 | } |
1188 | ||
1189 | /* If there's nothing new since last we looked, return invalid. */ | |
1190 | if (vq->avail_idx == last_avail_idx) | |
1191 | return vq->num; | |
1192 | ||
1193 | /* Only get avail ring entries after they have been exposed by guest. */ | |
5659338c | 1194 | smp_rmb(); |
3a4d5c94 MT |
1195 | |
1196 | /* Grab the next descriptor number they're advertising, and increment | |
1197 | * the index we've seen. */ | |
8b7347aa MT |
1198 | if (unlikely(__get_user(head, |
1199 | &vq->avail->ring[last_avail_idx % vq->num]))) { | |
3a4d5c94 MT |
1200 | vq_err(vq, "Failed to read head: idx %d address %p\n", |
1201 | last_avail_idx, | |
1202 | &vq->avail->ring[last_avail_idx % vq->num]); | |
d5675bd2 | 1203 | return -EFAULT; |
3a4d5c94 MT |
1204 | } |
1205 | ||
1206 | /* If their number is silly, that's an error. */ | |
7b3384fc | 1207 | if (unlikely(head >= vq->num)) { |
3a4d5c94 MT |
1208 | vq_err(vq, "Guest says index %u > %u is available", |
1209 | head, vq->num); | |
d5675bd2 | 1210 | return -EINVAL; |
3a4d5c94 MT |
1211 | } |
1212 | ||
1213 | /* When we start there are none of either input nor output. */ | |
1214 | *out_num = *in_num = 0; | |
1215 | if (unlikely(log)) | |
1216 | *log_num = 0; | |
1217 | ||
1218 | i = head; | |
1219 | do { | |
1220 | unsigned iov_count = *in_num + *out_num; | |
7b3384fc | 1221 | if (unlikely(i >= vq->num)) { |
3a4d5c94 MT |
1222 | vq_err(vq, "Desc index is %u > %u, head = %u", |
1223 | i, vq->num, head); | |
d5675bd2 | 1224 | return -EINVAL; |
3a4d5c94 | 1225 | } |
7b3384fc | 1226 | if (unlikely(++found > vq->num)) { |
3a4d5c94 MT |
1227 | vq_err(vq, "Loop detected: last one at %u " |
1228 | "vq size %u head %u\n", | |
1229 | i, vq->num, head); | |
d5675bd2 | 1230 | return -EINVAL; |
3a4d5c94 | 1231 | } |
fcc042a2 | 1232 | ret = __copy_from_user(&desc, vq->desc + i, sizeof desc); |
7b3384fc | 1233 | if (unlikely(ret)) { |
3a4d5c94 MT |
1234 | vq_err(vq, "Failed to get descriptor: idx %d addr %p\n", |
1235 | i, vq->desc + i); | |
d5675bd2 | 1236 | return -EFAULT; |
3a4d5c94 MT |
1237 | } |
1238 | if (desc.flags & VRING_DESC_F_INDIRECT) { | |
1239 | ret = get_indirect(dev, vq, iov, iov_size, | |
1240 | out_num, in_num, | |
1241 | log, log_num, &desc); | |
7b3384fc | 1242 | if (unlikely(ret < 0)) { |
3a4d5c94 MT |
1243 | vq_err(vq, "Failure detected " |
1244 | "in indirect descriptor at idx %d\n", i); | |
d5675bd2 | 1245 | return ret; |
3a4d5c94 MT |
1246 | } |
1247 | continue; | |
1248 | } | |
1249 | ||
1250 | ret = translate_desc(dev, desc.addr, desc.len, iov + iov_count, | |
1251 | iov_size - iov_count); | |
7b3384fc | 1252 | if (unlikely(ret < 0)) { |
3a4d5c94 MT |
1253 | vq_err(vq, "Translation failure %d descriptor idx %d\n", |
1254 | ret, i); | |
d5675bd2 | 1255 | return ret; |
3a4d5c94 MT |
1256 | } |
1257 | if (desc.flags & VRING_DESC_F_WRITE) { | |
1258 | /* If this is an input descriptor, | |
1259 | * increment that count. */ | |
1260 | *in_num += ret; | |
1261 | if (unlikely(log)) { | |
1262 | log[*log_num].addr = desc.addr; | |
1263 | log[*log_num].len = desc.len; | |
1264 | ++*log_num; | |
1265 | } | |
1266 | } else { | |
1267 | /* If it's an output descriptor, they're all supposed | |
1268 | * to come before any input descriptors. */ | |
7b3384fc | 1269 | if (unlikely(*in_num)) { |
3a4d5c94 MT |
1270 | vq_err(vq, "Descriptor has out after in: " |
1271 | "idx %d\n", i); | |
d5675bd2 | 1272 | return -EINVAL; |
3a4d5c94 MT |
1273 | } |
1274 | *out_num += ret; | |
1275 | } | |
1276 | } while ((i = next_desc(&desc)) != -1); | |
1277 | ||
1278 | /* On success, increment avail index. */ | |
1279 | vq->last_avail_idx++; | |
8ea8cf89 MT |
1280 | |
1281 | /* Assume notifications from guest are disabled at this point, | |
1282 | * if they aren't we would need to update avail_event index. */ | |
1283 | BUG_ON(!(vq->used_flags & VRING_USED_F_NO_NOTIFY)); | |
3a4d5c94 MT |
1284 | return head; |
1285 | } | |
1286 | ||
1287 | /* Reverse the effect of vhost_get_vq_desc. Useful for error handling. */ | |
8dd014ad | 1288 | void vhost_discard_vq_desc(struct vhost_virtqueue *vq, int n) |
3a4d5c94 | 1289 | { |
8dd014ad | 1290 | vq->last_avail_idx -= n; |
3a4d5c94 MT |
1291 | } |
1292 | ||
1293 | /* After we've used one of their buffers, we tell them about it. We'll then | |
1294 | * want to notify the guest, using eventfd. */ | |
1295 | int vhost_add_used(struct vhost_virtqueue *vq, unsigned int head, int len) | |
1296 | { | |
a8d3782f | 1297 | struct vring_used_elem __user *used; |
3a4d5c94 MT |
1298 | |
1299 | /* The virtqueue contains a ring of used buffers. Get a pointer to the | |
1300 | * next entry in that used ring. */ | |
1301 | used = &vq->used->ring[vq->last_used_idx % vq->num]; | |
8b7347aa | 1302 | if (__put_user(head, &used->id)) { |
3a4d5c94 MT |
1303 | vq_err(vq, "Failed to write used id"); |
1304 | return -EFAULT; | |
1305 | } | |
8b7347aa | 1306 | if (__put_user(len, &used->len)) { |
3a4d5c94 MT |
1307 | vq_err(vq, "Failed to write used len"); |
1308 | return -EFAULT; | |
1309 | } | |
1310 | /* Make sure buffer is written before we update index. */ | |
5659338c | 1311 | smp_wmb(); |
8b7347aa | 1312 | if (__put_user(vq->last_used_idx + 1, &vq->used->idx)) { |
3a4d5c94 MT |
1313 | vq_err(vq, "Failed to increment used idx"); |
1314 | return -EFAULT; | |
1315 | } | |
1316 | if (unlikely(vq->log_used)) { | |
1317 | /* Make sure data is seen before log. */ | |
5659338c | 1318 | smp_wmb(); |
86e9424d MT |
1319 | /* Log used ring entry write. */ |
1320 | log_write(vq->log_base, | |
a8d3782f CH |
1321 | vq->log_addr + |
1322 | ((void __user *)used - (void __user *)vq->used), | |
86e9424d MT |
1323 | sizeof *used); |
1324 | /* Log used index update. */ | |
1325 | log_write(vq->log_base, | |
1326 | vq->log_addr + offsetof(struct vring_used, idx), | |
1327 | sizeof vq->used->idx); | |
3a4d5c94 MT |
1328 | if (vq->log_ctx) |
1329 | eventfd_signal(vq->log_ctx, 1); | |
1330 | } | |
1331 | vq->last_used_idx++; | |
8ea8cf89 MT |
1332 | /* If the driver never bothers to signal in a very long while, |
1333 | * used index might wrap around. If that happens, invalidate | |
1334 | * signalled_used index we stored. TODO: make sure driver | |
1335 | * signals at least once in 2^16 and remove this. */ | |
1336 | if (unlikely(vq->last_used_idx == vq->signalled_used)) | |
1337 | vq->signalled_used_valid = false; | |
3a4d5c94 MT |
1338 | return 0; |
1339 | } | |
1340 | ||
8dd014ad DS |
1341 | static int __vhost_add_used_n(struct vhost_virtqueue *vq, |
1342 | struct vring_used_elem *heads, | |
1343 | unsigned count) | |
1344 | { | |
1345 | struct vring_used_elem __user *used; | |
8ea8cf89 | 1346 | u16 old, new; |
8dd014ad DS |
1347 | int start; |
1348 | ||
1349 | start = vq->last_used_idx % vq->num; | |
1350 | used = vq->used->ring + start; | |
dfe5ac5b | 1351 | if (__copy_to_user(used, heads, count * sizeof *used)) { |
8dd014ad DS |
1352 | vq_err(vq, "Failed to write used"); |
1353 | return -EFAULT; | |
1354 | } | |
1355 | if (unlikely(vq->log_used)) { | |
1356 | /* Make sure data is seen before log. */ | |
1357 | smp_wmb(); | |
1358 | /* Log used ring entry write. */ | |
1359 | log_write(vq->log_base, | |
1360 | vq->log_addr + | |
1361 | ((void __user *)used - (void __user *)vq->used), | |
1362 | count * sizeof *used); | |
1363 | } | |
8ea8cf89 MT |
1364 | old = vq->last_used_idx; |
1365 | new = (vq->last_used_idx += count); | |
1366 | /* If the driver never bothers to signal in a very long while, | |
1367 | * used index might wrap around. If that happens, invalidate | |
1368 | * signalled_used index we stored. TODO: make sure driver | |
1369 | * signals at least once in 2^16 and remove this. */ | |
1370 | if (unlikely((u16)(new - vq->signalled_used) < (u16)(new - old))) | |
1371 | vq->signalled_used_valid = false; | |
8dd014ad DS |
1372 | return 0; |
1373 | } | |
1374 | ||
1375 | /* After we've used one of their buffers, we tell them about it. We'll then | |
1376 | * want to notify the guest, using eventfd. */ | |
1377 | int vhost_add_used_n(struct vhost_virtqueue *vq, struct vring_used_elem *heads, | |
1378 | unsigned count) | |
1379 | { | |
1380 | int start, n, r; | |
1381 | ||
1382 | start = vq->last_used_idx % vq->num; | |
1383 | n = vq->num - start; | |
1384 | if (n < count) { | |
1385 | r = __vhost_add_used_n(vq, heads, n); | |
1386 | if (r < 0) | |
1387 | return r; | |
1388 | heads += n; | |
1389 | count -= n; | |
1390 | } | |
1391 | r = __vhost_add_used_n(vq, heads, count); | |
1392 | ||
1393 | /* Make sure buffer is written before we update index. */ | |
1394 | smp_wmb(); | |
1395 | if (put_user(vq->last_used_idx, &vq->used->idx)) { | |
1396 | vq_err(vq, "Failed to increment used idx"); | |
1397 | return -EFAULT; | |
1398 | } | |
1399 | if (unlikely(vq->log_used)) { | |
1400 | /* Log used index update. */ | |
1401 | log_write(vq->log_base, | |
1402 | vq->log_addr + offsetof(struct vring_used, idx), | |
1403 | sizeof vq->used->idx); | |
1404 | if (vq->log_ctx) | |
1405 | eventfd_signal(vq->log_ctx, 1); | |
1406 | } | |
1407 | return r; | |
1408 | } | |
1409 | ||
8ea8cf89 | 1410 | static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 | 1411 | { |
8ea8cf89 MT |
1412 | __u16 old, new, event; |
1413 | bool v; | |
0d499356 MT |
1414 | /* Flush out used index updates. This is paired |
1415 | * with the barrier that the Guest executes when enabling | |
1416 | * interrupts. */ | |
1417 | smp_mb(); | |
1418 | ||
8ea8cf89 MT |
1419 | if (vhost_has_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY) && |
1420 | unlikely(vq->avail_idx == vq->last_avail_idx)) | |
1421 | return true; | |
1422 | ||
1423 | if (!vhost_has_feature(dev, VIRTIO_RING_F_EVENT_IDX)) { | |
1424 | __u16 flags; | |
1425 | if (__get_user(flags, &vq->avail->flags)) { | |
1426 | vq_err(vq, "Failed to get flags"); | |
1427 | return true; | |
1428 | } | |
1429 | return !(flags & VRING_AVAIL_F_NO_INTERRUPT); | |
3a4d5c94 | 1430 | } |
8ea8cf89 MT |
1431 | old = vq->signalled_used; |
1432 | v = vq->signalled_used_valid; | |
1433 | new = vq->signalled_used = vq->last_used_idx; | |
1434 | vq->signalled_used_valid = true; | |
3a4d5c94 | 1435 | |
8ea8cf89 MT |
1436 | if (unlikely(!v)) |
1437 | return true; | |
3a4d5c94 | 1438 | |
8ea8cf89 MT |
1439 | if (get_user(event, vhost_used_event(vq))) { |
1440 | vq_err(vq, "Failed to get used event idx"); | |
1441 | return true; | |
1442 | } | |
1443 | return vring_need_event(event, new, old); | |
1444 | } | |
1445 | ||
1446 | /* This actually signals the guest, using eventfd. */ | |
1447 | void vhost_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq) | |
1448 | { | |
3a4d5c94 | 1449 | /* Signal the Guest tell them we used something up. */ |
8ea8cf89 | 1450 | if (vq->call_ctx && vhost_notify(dev, vq)) |
3a4d5c94 MT |
1451 | eventfd_signal(vq->call_ctx, 1); |
1452 | } | |
1453 | ||
1454 | /* And here's the combo meal deal. Supersize me! */ | |
1455 | void vhost_add_used_and_signal(struct vhost_dev *dev, | |
1456 | struct vhost_virtqueue *vq, | |
1457 | unsigned int head, int len) | |
1458 | { | |
1459 | vhost_add_used(vq, head, len); | |
1460 | vhost_signal(dev, vq); | |
1461 | } | |
1462 | ||
8dd014ad DS |
1463 | /* multi-buffer version of vhost_add_used_and_signal */ |
1464 | void vhost_add_used_and_signal_n(struct vhost_dev *dev, | |
1465 | struct vhost_virtqueue *vq, | |
1466 | struct vring_used_elem *heads, unsigned count) | |
1467 | { | |
1468 | vhost_add_used_n(vq, heads, count); | |
1469 | vhost_signal(dev, vq); | |
1470 | } | |
1471 | ||
3a4d5c94 | 1472 | /* OK, now we need to know about added descriptors. */ |
8ea8cf89 | 1473 | bool vhost_enable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 MT |
1474 | { |
1475 | u16 avail_idx; | |
1476 | int r; | |
d47effe1 | 1477 | |
3a4d5c94 MT |
1478 | if (!(vq->used_flags & VRING_USED_F_NO_NOTIFY)) |
1479 | return false; | |
1480 | vq->used_flags &= ~VRING_USED_F_NO_NOTIFY; | |
8ea8cf89 MT |
1481 | if (!vhost_has_feature(dev, VIRTIO_RING_F_EVENT_IDX)) { |
1482 | r = put_user(vq->used_flags, &vq->used->flags); | |
1483 | if (r) { | |
1484 | vq_err(vq, "Failed to enable notification at %p: %d\n", | |
1485 | &vq->used->flags, r); | |
1486 | return false; | |
1487 | } | |
1488 | } else { | |
1489 | r = put_user(vq->avail_idx, vhost_avail_event(vq)); | |
1490 | if (r) { | |
1491 | vq_err(vq, "Failed to update avail event index at %p: %d\n", | |
1492 | vhost_avail_event(vq), r); | |
1493 | return false; | |
1494 | } | |
1495 | } | |
1496 | if (unlikely(vq->log_used)) { | |
1497 | void __user *used; | |
1498 | /* Make sure data is seen before log. */ | |
1499 | smp_wmb(); | |
1500 | used = vhost_has_feature(dev, VIRTIO_RING_F_EVENT_IDX) ? | |
1501 | &vq->used->flags : vhost_avail_event(vq); | |
1502 | /* Log used flags or event index entry write. Both are 16 bit | |
1503 | * fields. */ | |
1504 | log_write(vq->log_base, vq->log_addr + | |
1505 | (used - (void __user *)vq->used), | |
1506 | sizeof(u16)); | |
1507 | if (vq->log_ctx) | |
1508 | eventfd_signal(vq->log_ctx, 1); | |
3a4d5c94 MT |
1509 | } |
1510 | /* They could have slipped one in as we were doing that: make | |
1511 | * sure it's written, then check again. */ | |
5659338c | 1512 | smp_mb(); |
8b7347aa | 1513 | r = __get_user(avail_idx, &vq->avail->idx); |
3a4d5c94 MT |
1514 | if (r) { |
1515 | vq_err(vq, "Failed to check avail idx at %p: %d\n", | |
1516 | &vq->avail->idx, r); | |
1517 | return false; | |
1518 | } | |
1519 | ||
8dd014ad | 1520 | return avail_idx != vq->avail_idx; |
3a4d5c94 MT |
1521 | } |
1522 | ||
1523 | /* We don't need to be notified again. */ | |
8ea8cf89 | 1524 | void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 MT |
1525 | { |
1526 | int r; | |
d47effe1 | 1527 | |
3a4d5c94 MT |
1528 | if (vq->used_flags & VRING_USED_F_NO_NOTIFY) |
1529 | return; | |
1530 | vq->used_flags |= VRING_USED_F_NO_NOTIFY; | |
8ea8cf89 MT |
1531 | if (!vhost_has_feature(dev, VIRTIO_RING_F_EVENT_IDX)) { |
1532 | r = put_user(vq->used_flags, &vq->used->flags); | |
1533 | if (r) | |
1534 | vq_err(vq, "Failed to enable notification at %p: %d\n", | |
1535 | &vq->used->flags, r); | |
1536 | } | |
3a4d5c94 | 1537 | } |
bab632d6 MT |
1538 | |
1539 | static void vhost_zerocopy_done_signal(struct kref *kref) | |
1540 | { | |
1541 | struct vhost_ubuf_ref *ubufs = container_of(kref, struct vhost_ubuf_ref, | |
1542 | kref); | |
1543 | wake_up(&ubufs->wait); | |
1544 | } | |
1545 | ||
1546 | struct vhost_ubuf_ref *vhost_ubuf_alloc(struct vhost_virtqueue *vq, | |
1547 | bool zcopy) | |
1548 | { | |
1549 | struct vhost_ubuf_ref *ubufs; | |
1550 | /* No zero copy backend? Nothing to count. */ | |
1551 | if (!zcopy) | |
1552 | return NULL; | |
1553 | ubufs = kmalloc(sizeof *ubufs, GFP_KERNEL); | |
1554 | if (!ubufs) | |
1555 | return ERR_PTR(-ENOMEM); | |
1556 | kref_init(&ubufs->kref); | |
1557 | kref_get(&ubufs->kref); | |
1558 | init_waitqueue_head(&ubufs->wait); | |
1559 | ubufs->vq = vq; | |
1560 | return ubufs; | |
1561 | } | |
1562 | ||
1563 | void vhost_ubuf_put(struct vhost_ubuf_ref *ubufs) | |
1564 | { | |
1565 | kref_put(&ubufs->kref, vhost_zerocopy_done_signal); | |
1566 | } | |
1567 | ||
1568 | void vhost_ubuf_put_and_wait(struct vhost_ubuf_ref *ubufs) | |
1569 | { | |
1570 | kref_put(&ubufs->kref, vhost_zerocopy_done_signal); | |
1571 | wait_event(ubufs->wait, !atomic_read(&ubufs->kref.refcount)); | |
1572 | kfree(ubufs); | |
1573 | } | |
1574 | ||
1575 | void vhost_zerocopy_callback(void *arg) | |
1576 | { | |
1577 | struct ubuf_info *ubuf = arg; | |
1578 | struct vhost_ubuf_ref *ubufs = ubuf->arg; | |
1579 | struct vhost_virtqueue *vq = ubufs->vq; | |
1580 | ||
1581 | /* set len = 1 to mark this desc buffers done DMA */ | |
1582 | vq->heads[ubuf->desc].len = VHOST_DMA_DONE_LEN; | |
1583 | kref_put(&ubufs->kref, vhost_zerocopy_done_signal); | |
1584 | } |