Commit | Line | Data |
---|---|---|
3a4d5c94 MT |
1 | /* Copyright (C) 2009 Red Hat, Inc. |
2 | * Copyright (C) 2006 Rusty Russell IBM Corporation | |
3 | * | |
4 | * Author: Michael S. Tsirkin <mst@redhat.com> | |
5 | * | |
6 | * Inspiration, some code, and most witty comments come from | |
61516587 | 7 | * Documentation/virtual/lguest/lguest.c, by Rusty Russell |
3a4d5c94 MT |
8 | * |
9 | * This work is licensed under the terms of the GNU GPL, version 2. | |
10 | * | |
11 | * Generic code for virtio server in host kernel. | |
12 | */ | |
13 | ||
14 | #include <linux/eventfd.h> | |
15 | #include <linux/vhost.h> | |
35596b27 | 16 | #include <linux/uio.h> |
3a4d5c94 | 17 | #include <linux/mm.h> |
64e1c807 | 18 | #include <linux/mmu_context.h> |
3a4d5c94 MT |
19 | #include <linux/miscdevice.h> |
20 | #include <linux/mutex.h> | |
3a4d5c94 MT |
21 | #include <linux/poll.h> |
22 | #include <linux/file.h> | |
23 | #include <linux/highmem.h> | |
5a0e3ad6 | 24 | #include <linux/slab.h> |
4de7255f | 25 | #include <linux/vmalloc.h> |
c23f3445 | 26 | #include <linux/kthread.h> |
9e3d1957 | 27 | #include <linux/cgroup.h> |
6ac1afbf | 28 | #include <linux/module.h> |
bcfeacab | 29 | #include <linux/sort.h> |
a9709d68 | 30 | #include <linux/interval_tree_generic.h> |
3a4d5c94 | 31 | |
3a4d5c94 MT |
32 | #include "vhost.h" |
33 | ||
c9ce42f7 IM |
34 | static ushort max_mem_regions = 64; |
35 | module_param(max_mem_regions, ushort, 0444); | |
36 | MODULE_PARM_DESC(max_mem_regions, | |
37 | "Maximum number of memory regions in memory map. (default: 64)"); | |
6b1e6cc7 JW |
38 | static int max_iotlb_entries = 2048; |
39 | module_param(max_iotlb_entries, int, 0444); | |
40 | MODULE_PARM_DESC(max_iotlb_entries, | |
41 | "Maximum number of iotlb entries. (default: 2048)"); | |
c9ce42f7 | 42 | |
3a4d5c94 | 43 | enum { |
3a4d5c94 MT |
44 | VHOST_MEMORY_F_LOG = 0x1, |
45 | }; | |
46 | ||
3b1bbe89 MT |
47 | #define vhost_used_event(vq) ((__virtio16 __user *)&vq->avail->ring[vq->num]) |
48 | #define vhost_avail_event(vq) ((__virtio16 __user *)&vq->used->ring[vq->num]) | |
8ea8cf89 | 49 | |
a9709d68 JW |
50 | INTERVAL_TREE_DEFINE(struct vhost_umem_node, |
51 | rb, __u64, __subtree_last, | |
2f952c01 | 52 | START, LAST, static inline, vhost_umem_interval_tree); |
a9709d68 | 53 | |
2751c988 | 54 | #ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY |
c5072037 | 55 | static void vhost_disable_cross_endian(struct vhost_virtqueue *vq) |
2751c988 GK |
56 | { |
57 | vq->user_be = !virtio_legacy_is_little_endian(); | |
58 | } | |
59 | ||
c5072037 GK |
60 | static void vhost_enable_cross_endian_big(struct vhost_virtqueue *vq) |
61 | { | |
62 | vq->user_be = true; | |
63 | } | |
64 | ||
65 | static void vhost_enable_cross_endian_little(struct vhost_virtqueue *vq) | |
66 | { | |
67 | vq->user_be = false; | |
68 | } | |
69 | ||
2751c988 GK |
70 | static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp) |
71 | { | |
72 | struct vhost_vring_state s; | |
73 | ||
74 | if (vq->private_data) | |
75 | return -EBUSY; | |
76 | ||
77 | if (copy_from_user(&s, argp, sizeof(s))) | |
78 | return -EFAULT; | |
79 | ||
80 | if (s.num != VHOST_VRING_LITTLE_ENDIAN && | |
81 | s.num != VHOST_VRING_BIG_ENDIAN) | |
82 | return -EINVAL; | |
83 | ||
c5072037 GK |
84 | if (s.num == VHOST_VRING_BIG_ENDIAN) |
85 | vhost_enable_cross_endian_big(vq); | |
86 | else | |
87 | vhost_enable_cross_endian_little(vq); | |
2751c988 GK |
88 | |
89 | return 0; | |
90 | } | |
91 | ||
92 | static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx, | |
93 | int __user *argp) | |
94 | { | |
95 | struct vhost_vring_state s = { | |
96 | .index = idx, | |
97 | .num = vq->user_be | |
98 | }; | |
99 | ||
100 | if (copy_to_user(argp, &s, sizeof(s))) | |
101 | return -EFAULT; | |
102 | ||
103 | return 0; | |
104 | } | |
105 | ||
106 | static void vhost_init_is_le(struct vhost_virtqueue *vq) | |
107 | { | |
108 | /* Note for legacy virtio: user_be is initialized at reset time | |
109 | * according to the host endianness. If userspace does not set an | |
110 | * explicit endianness, the default behavior is native endian, as | |
111 | * expected by legacy virtio. | |
112 | */ | |
113 | vq->is_le = vhost_has_feature(vq, VIRTIO_F_VERSION_1) || !vq->user_be; | |
114 | } | |
115 | #else | |
c5072037 | 116 | static void vhost_disable_cross_endian(struct vhost_virtqueue *vq) |
2751c988 GK |
117 | { |
118 | } | |
119 | ||
120 | static long vhost_set_vring_endian(struct vhost_virtqueue *vq, int __user *argp) | |
121 | { | |
122 | return -ENOIOCTLCMD; | |
123 | } | |
124 | ||
125 | static long vhost_get_vring_endian(struct vhost_virtqueue *vq, u32 idx, | |
126 | int __user *argp) | |
127 | { | |
128 | return -ENOIOCTLCMD; | |
129 | } | |
130 | ||
131 | static void vhost_init_is_le(struct vhost_virtqueue *vq) | |
132 | { | |
cda8bba0 HP |
133 | vq->is_le = vhost_has_feature(vq, VIRTIO_F_VERSION_1) |
134 | || virtio_legacy_is_little_endian(); | |
2751c988 GK |
135 | } |
136 | #endif /* CONFIG_VHOST_CROSS_ENDIAN_LEGACY */ | |
137 | ||
c5072037 GK |
138 | static void vhost_reset_is_le(struct vhost_virtqueue *vq) |
139 | { | |
cda8bba0 | 140 | vhost_init_is_le(vq); |
c5072037 GK |
141 | } |
142 | ||
7235acdb JW |
143 | struct vhost_flush_struct { |
144 | struct vhost_work work; | |
145 | struct completion wait_event; | |
146 | }; | |
147 | ||
148 | static void vhost_flush_work(struct vhost_work *work) | |
149 | { | |
150 | struct vhost_flush_struct *s; | |
151 | ||
152 | s = container_of(work, struct vhost_flush_struct, work); | |
153 | complete(&s->wait_event); | |
154 | } | |
155 | ||
3a4d5c94 MT |
156 | static void vhost_poll_func(struct file *file, wait_queue_head_t *wqh, |
157 | poll_table *pt) | |
158 | { | |
159 | struct vhost_poll *poll; | |
3a4d5c94 | 160 | |
d47effe1 | 161 | poll = container_of(pt, struct vhost_poll, table); |
3a4d5c94 MT |
162 | poll->wqh = wqh; |
163 | add_wait_queue(wqh, &poll->wait); | |
164 | } | |
165 | ||
166 | static int vhost_poll_wakeup(wait_queue_t *wait, unsigned mode, int sync, | |
167 | void *key) | |
168 | { | |
c23f3445 TH |
169 | struct vhost_poll *poll = container_of(wait, struct vhost_poll, wait); |
170 | ||
3a4d5c94 MT |
171 | if (!((unsigned long)key & poll->mask)) |
172 | return 0; | |
173 | ||
c23f3445 | 174 | vhost_poll_queue(poll); |
3a4d5c94 MT |
175 | return 0; |
176 | } | |
177 | ||
163049ae | 178 | void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn) |
87d6a412 | 179 | { |
04b96e55 | 180 | clear_bit(VHOST_WORK_QUEUED, &work->flags); |
87d6a412 MT |
181 | work->fn = fn; |
182 | init_waitqueue_head(&work->done); | |
87d6a412 | 183 | } |
6ac1afbf | 184 | EXPORT_SYMBOL_GPL(vhost_work_init); |
87d6a412 | 185 | |
3a4d5c94 | 186 | /* Init poll structure */ |
c23f3445 TH |
187 | void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn, |
188 | unsigned long mask, struct vhost_dev *dev) | |
3a4d5c94 | 189 | { |
3a4d5c94 MT |
190 | init_waitqueue_func_entry(&poll->wait, vhost_poll_wakeup); |
191 | init_poll_funcptr(&poll->table, vhost_poll_func); | |
192 | poll->mask = mask; | |
c23f3445 | 193 | poll->dev = dev; |
2b8b328b | 194 | poll->wqh = NULL; |
c23f3445 | 195 | |
87d6a412 | 196 | vhost_work_init(&poll->work, fn); |
3a4d5c94 | 197 | } |
6ac1afbf | 198 | EXPORT_SYMBOL_GPL(vhost_poll_init); |
3a4d5c94 MT |
199 | |
200 | /* Start polling a file. We add ourselves to file's wait queue. The caller must | |
201 | * keep a reference to a file until after vhost_poll_stop is called. */ | |
2b8b328b | 202 | int vhost_poll_start(struct vhost_poll *poll, struct file *file) |
3a4d5c94 MT |
203 | { |
204 | unsigned long mask; | |
2b8b328b | 205 | int ret = 0; |
d47effe1 | 206 | |
70181d51 JW |
207 | if (poll->wqh) |
208 | return 0; | |
209 | ||
3a4d5c94 MT |
210 | mask = file->f_op->poll(file, &poll->table); |
211 | if (mask) | |
212 | vhost_poll_wakeup(&poll->wait, 0, 0, (void *)mask); | |
2b8b328b JW |
213 | if (mask & POLLERR) { |
214 | if (poll->wqh) | |
215 | remove_wait_queue(poll->wqh, &poll->wait); | |
216 | ret = -EINVAL; | |
217 | } | |
218 | ||
219 | return ret; | |
3a4d5c94 | 220 | } |
6ac1afbf | 221 | EXPORT_SYMBOL_GPL(vhost_poll_start); |
3a4d5c94 MT |
222 | |
223 | /* Stop polling a file. After this function returns, it becomes safe to drop the | |
224 | * file reference. You must also flush afterwards. */ | |
225 | void vhost_poll_stop(struct vhost_poll *poll) | |
226 | { | |
2b8b328b JW |
227 | if (poll->wqh) { |
228 | remove_wait_queue(poll->wqh, &poll->wait); | |
229 | poll->wqh = NULL; | |
230 | } | |
3a4d5c94 | 231 | } |
6ac1afbf | 232 | EXPORT_SYMBOL_GPL(vhost_poll_stop); |
3a4d5c94 | 233 | |
7235acdb | 234 | void vhost_work_flush(struct vhost_dev *dev, struct vhost_work *work) |
0174b0c3 | 235 | { |
7235acdb | 236 | struct vhost_flush_struct flush; |
d47effe1 | 237 | |
7235acdb JW |
238 | if (dev->worker) { |
239 | init_completion(&flush.wait_event); | |
240 | vhost_work_init(&flush.work, vhost_flush_work); | |
0174b0c3 | 241 | |
7235acdb JW |
242 | vhost_work_queue(dev, &flush.work); |
243 | wait_for_completion(&flush.wait_event); | |
244 | } | |
3a4d5c94 | 245 | } |
6ac1afbf | 246 | EXPORT_SYMBOL_GPL(vhost_work_flush); |
3a4d5c94 | 247 | |
87d6a412 MT |
248 | /* Flush any work that has been scheduled. When calling this, don't hold any |
249 | * locks that are also used by the callback. */ | |
250 | void vhost_poll_flush(struct vhost_poll *poll) | |
251 | { | |
252 | vhost_work_flush(poll->dev, &poll->work); | |
253 | } | |
6ac1afbf | 254 | EXPORT_SYMBOL_GPL(vhost_poll_flush); |
87d6a412 | 255 | |
163049ae | 256 | void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work) |
3a4d5c94 | 257 | { |
04b96e55 JW |
258 | if (!dev->worker) |
259 | return; | |
c23f3445 | 260 | |
04b96e55 JW |
261 | if (!test_and_set_bit(VHOST_WORK_QUEUED, &work->flags)) { |
262 | /* We can only add the work to the list after we're | |
263 | * sure it was not in the list. | |
635abf01 | 264 | * test_and_set_bit() implies a memory barrier. |
04b96e55 | 265 | */ |
04b96e55 | 266 | llist_add(&work->node, &dev->work_list); |
c23f3445 TH |
267 | wake_up_process(dev->worker); |
268 | } | |
3a4d5c94 | 269 | } |
6ac1afbf | 270 | EXPORT_SYMBOL_GPL(vhost_work_queue); |
3a4d5c94 | 271 | |
526d3e7f JW |
272 | /* A lockless hint for busy polling code to exit the loop */ |
273 | bool vhost_has_work(struct vhost_dev *dev) | |
274 | { | |
04b96e55 | 275 | return !llist_empty(&dev->work_list); |
526d3e7f JW |
276 | } |
277 | EXPORT_SYMBOL_GPL(vhost_has_work); | |
278 | ||
87d6a412 MT |
279 | void vhost_poll_queue(struct vhost_poll *poll) |
280 | { | |
281 | vhost_work_queue(poll->dev, &poll->work); | |
282 | } | |
6ac1afbf | 283 | EXPORT_SYMBOL_GPL(vhost_poll_queue); |
87d6a412 | 284 | |
3a4d5c94 MT |
285 | static void vhost_vq_reset(struct vhost_dev *dev, |
286 | struct vhost_virtqueue *vq) | |
287 | { | |
288 | vq->num = 1; | |
289 | vq->desc = NULL; | |
290 | vq->avail = NULL; | |
291 | vq->used = NULL; | |
292 | vq->last_avail_idx = 0; | |
809ecb9b | 293 | vq->last_used_event = 0; |
3a4d5c94 MT |
294 | vq->avail_idx = 0; |
295 | vq->last_used_idx = 0; | |
8ea8cf89 MT |
296 | vq->signalled_used = 0; |
297 | vq->signalled_used_valid = false; | |
3a4d5c94 | 298 | vq->used_flags = 0; |
3a4d5c94 MT |
299 | vq->log_used = false; |
300 | vq->log_addr = -1ull; | |
3a4d5c94 | 301 | vq->private_data = NULL; |
ea16c514 | 302 | vq->acked_features = 0; |
3a4d5c94 MT |
303 | vq->log_base = NULL; |
304 | vq->error_ctx = NULL; | |
305 | vq->error = NULL; | |
306 | vq->kick = NULL; | |
307 | vq->call_ctx = NULL; | |
308 | vq->call = NULL; | |
73a99f08 | 309 | vq->log_ctx = NULL; |
c5072037 GK |
310 | vhost_reset_is_le(vq); |
311 | vhost_disable_cross_endian(vq); | |
03088137 | 312 | vq->busyloop_timeout = 0; |
a9709d68 | 313 | vq->umem = NULL; |
6b1e6cc7 | 314 | vq->iotlb = NULL; |
3a4d5c94 MT |
315 | } |
316 | ||
c23f3445 TH |
317 | static int vhost_worker(void *data) |
318 | { | |
319 | struct vhost_dev *dev = data; | |
04b96e55 JW |
320 | struct vhost_work *work, *work_next; |
321 | struct llist_node *node; | |
d7ffde35 | 322 | mm_segment_t oldfs = get_fs(); |
c23f3445 | 323 | |
d7ffde35 | 324 | set_fs(USER_DS); |
64e1c807 MT |
325 | use_mm(dev->mm); |
326 | ||
c23f3445 TH |
327 | for (;;) { |
328 | /* mb paired w/ kthread_stop */ | |
329 | set_current_state(TASK_INTERRUPTIBLE); | |
330 | ||
c23f3445 | 331 | if (kthread_should_stop()) { |
c23f3445 | 332 | __set_current_state(TASK_RUNNING); |
64e1c807 | 333 | break; |
c23f3445 | 334 | } |
04b96e55 JW |
335 | |
336 | node = llist_del_all(&dev->work_list); | |
337 | if (!node) | |
338 | schedule(); | |
339 | ||
340 | node = llist_reverse_order(node); | |
341 | /* make sure flag is seen after deletion */ | |
342 | smp_wmb(); | |
343 | llist_for_each_entry_safe(work, work_next, node, node) { | |
344 | clear_bit(VHOST_WORK_QUEUED, &work->flags); | |
c23f3445 TH |
345 | __set_current_state(TASK_RUNNING); |
346 | work->fn(work); | |
d550dda1 NHE |
347 | if (need_resched()) |
348 | schedule(); | |
04b96e55 | 349 | } |
c23f3445 | 350 | } |
64e1c807 | 351 | unuse_mm(dev->mm); |
d7ffde35 | 352 | set_fs(oldfs); |
64e1c807 | 353 | return 0; |
c23f3445 TH |
354 | } |
355 | ||
bab632d6 MT |
356 | static void vhost_vq_free_iovecs(struct vhost_virtqueue *vq) |
357 | { | |
358 | kfree(vq->indirect); | |
359 | vq->indirect = NULL; | |
360 | kfree(vq->log); | |
361 | vq->log = NULL; | |
362 | kfree(vq->heads); | |
363 | vq->heads = NULL; | |
bab632d6 MT |
364 | } |
365 | ||
e0e9b406 JW |
366 | /* Helper to allocate iovec buffers for all vqs. */ |
367 | static long vhost_dev_alloc_iovecs(struct vhost_dev *dev) | |
368 | { | |
6d5e6aa8 | 369 | struct vhost_virtqueue *vq; |
e0e9b406 | 370 | int i; |
d47effe1 | 371 | |
e0e9b406 | 372 | for (i = 0; i < dev->nvqs; ++i) { |
6d5e6aa8 AH |
373 | vq = dev->vqs[i]; |
374 | vq->indirect = kmalloc(sizeof *vq->indirect * UIO_MAXIOV, | |
375 | GFP_KERNEL); | |
376 | vq->log = kmalloc(sizeof *vq->log * UIO_MAXIOV, GFP_KERNEL); | |
377 | vq->heads = kmalloc(sizeof *vq->heads * UIO_MAXIOV, GFP_KERNEL); | |
378 | if (!vq->indirect || !vq->log || !vq->heads) | |
e0e9b406 JW |
379 | goto err_nomem; |
380 | } | |
381 | return 0; | |
d47effe1 | 382 | |
e0e9b406 | 383 | err_nomem: |
bab632d6 | 384 | for (; i >= 0; --i) |
3ab2e420 | 385 | vhost_vq_free_iovecs(dev->vqs[i]); |
e0e9b406 JW |
386 | return -ENOMEM; |
387 | } | |
388 | ||
389 | static void vhost_dev_free_iovecs(struct vhost_dev *dev) | |
390 | { | |
391 | int i; | |
d47effe1 | 392 | |
bab632d6 | 393 | for (i = 0; i < dev->nvqs; ++i) |
3ab2e420 | 394 | vhost_vq_free_iovecs(dev->vqs[i]); |
e0e9b406 JW |
395 | } |
396 | ||
59566b6e | 397 | void vhost_dev_init(struct vhost_dev *dev, |
3ab2e420 | 398 | struct vhost_virtqueue **vqs, int nvqs) |
3a4d5c94 | 399 | { |
6d5e6aa8 | 400 | struct vhost_virtqueue *vq; |
3a4d5c94 | 401 | int i; |
c23f3445 | 402 | |
3a4d5c94 MT |
403 | dev->vqs = vqs; |
404 | dev->nvqs = nvqs; | |
405 | mutex_init(&dev->mutex); | |
406 | dev->log_ctx = NULL; | |
407 | dev->log_file = NULL; | |
a9709d68 | 408 | dev->umem = NULL; |
6b1e6cc7 | 409 | dev->iotlb = NULL; |
3a4d5c94 | 410 | dev->mm = NULL; |
c23f3445 | 411 | dev->worker = NULL; |
04b96e55 | 412 | init_llist_head(&dev->work_list); |
6b1e6cc7 JW |
413 | init_waitqueue_head(&dev->wait); |
414 | INIT_LIST_HEAD(&dev->read_list); | |
415 | INIT_LIST_HEAD(&dev->pending_list); | |
416 | spin_lock_init(&dev->iotlb_lock); | |
04b96e55 | 417 | |
3a4d5c94 MT |
418 | |
419 | for (i = 0; i < dev->nvqs; ++i) { | |
6d5e6aa8 AH |
420 | vq = dev->vqs[i]; |
421 | vq->log = NULL; | |
422 | vq->indirect = NULL; | |
423 | vq->heads = NULL; | |
424 | vq->dev = dev; | |
425 | mutex_init(&vq->mutex); | |
426 | vhost_vq_reset(dev, vq); | |
427 | if (vq->handle_kick) | |
428 | vhost_poll_init(&vq->poll, vq->handle_kick, | |
429 | POLLIN, dev); | |
3a4d5c94 | 430 | } |
3a4d5c94 | 431 | } |
6ac1afbf | 432 | EXPORT_SYMBOL_GPL(vhost_dev_init); |
3a4d5c94 MT |
433 | |
434 | /* Caller should have device mutex */ | |
435 | long vhost_dev_check_owner(struct vhost_dev *dev) | |
436 | { | |
437 | /* Are you the owner? If not, I don't think you mean to do that */ | |
438 | return dev->mm == current->mm ? 0 : -EPERM; | |
439 | } | |
6ac1afbf | 440 | EXPORT_SYMBOL_GPL(vhost_dev_check_owner); |
3a4d5c94 | 441 | |
87d6a412 | 442 | struct vhost_attach_cgroups_struct { |
d47effe1 KK |
443 | struct vhost_work work; |
444 | struct task_struct *owner; | |
445 | int ret; | |
87d6a412 MT |
446 | }; |
447 | ||
448 | static void vhost_attach_cgroups_work(struct vhost_work *work) | |
449 | { | |
d47effe1 KK |
450 | struct vhost_attach_cgroups_struct *s; |
451 | ||
452 | s = container_of(work, struct vhost_attach_cgroups_struct, work); | |
453 | s->ret = cgroup_attach_task_all(s->owner, current); | |
87d6a412 MT |
454 | } |
455 | ||
456 | static int vhost_attach_cgroups(struct vhost_dev *dev) | |
457 | { | |
d47effe1 KK |
458 | struct vhost_attach_cgroups_struct attach; |
459 | ||
460 | attach.owner = current; | |
461 | vhost_work_init(&attach.work, vhost_attach_cgroups_work); | |
462 | vhost_work_queue(dev, &attach.work); | |
463 | vhost_work_flush(dev, &attach.work); | |
464 | return attach.ret; | |
87d6a412 MT |
465 | } |
466 | ||
05c05351 MT |
467 | /* Caller should have device mutex */ |
468 | bool vhost_dev_has_owner(struct vhost_dev *dev) | |
469 | { | |
470 | return dev->mm; | |
471 | } | |
6ac1afbf | 472 | EXPORT_SYMBOL_GPL(vhost_dev_has_owner); |
05c05351 | 473 | |
3a4d5c94 | 474 | /* Caller should have device mutex */ |
54db63c2 | 475 | long vhost_dev_set_owner(struct vhost_dev *dev) |
3a4d5c94 | 476 | { |
c23f3445 TH |
477 | struct task_struct *worker; |
478 | int err; | |
d47effe1 | 479 | |
3a4d5c94 | 480 | /* Is there an owner already? */ |
05c05351 | 481 | if (vhost_dev_has_owner(dev)) { |
c23f3445 TH |
482 | err = -EBUSY; |
483 | goto err_mm; | |
484 | } | |
d47effe1 | 485 | |
3a4d5c94 MT |
486 | /* No owner, become one */ |
487 | dev->mm = get_task_mm(current); | |
c23f3445 TH |
488 | worker = kthread_create(vhost_worker, dev, "vhost-%d", current->pid); |
489 | if (IS_ERR(worker)) { | |
490 | err = PTR_ERR(worker); | |
491 | goto err_worker; | |
492 | } | |
493 | ||
494 | dev->worker = worker; | |
87d6a412 MT |
495 | wake_up_process(worker); /* avoid contributing to loadavg */ |
496 | ||
497 | err = vhost_attach_cgroups(dev); | |
9e3d1957 MT |
498 | if (err) |
499 | goto err_cgroup; | |
c23f3445 | 500 | |
e0e9b406 JW |
501 | err = vhost_dev_alloc_iovecs(dev); |
502 | if (err) | |
503 | goto err_cgroup; | |
504 | ||
3a4d5c94 | 505 | return 0; |
9e3d1957 MT |
506 | err_cgroup: |
507 | kthread_stop(worker); | |
615cc221 | 508 | dev->worker = NULL; |
c23f3445 TH |
509 | err_worker: |
510 | if (dev->mm) | |
511 | mmput(dev->mm); | |
512 | dev->mm = NULL; | |
513 | err_mm: | |
514 | return err; | |
3a4d5c94 | 515 | } |
6ac1afbf | 516 | EXPORT_SYMBOL_GPL(vhost_dev_set_owner); |
3a4d5c94 | 517 | |
a9709d68 | 518 | static void *vhost_kvzalloc(unsigned long size) |
3a4d5c94 | 519 | { |
a9709d68 JW |
520 | void *n = kzalloc(size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
521 | ||
522 | if (!n) | |
523 | n = vzalloc(size); | |
524 | return n; | |
525 | } | |
526 | ||
527 | struct vhost_umem *vhost_dev_reset_owner_prepare(void) | |
528 | { | |
529 | return vhost_kvzalloc(sizeof(struct vhost_umem)); | |
150b9e51 | 530 | } |
6ac1afbf | 531 | EXPORT_SYMBOL_GPL(vhost_dev_reset_owner_prepare); |
3a4d5c94 | 532 | |
150b9e51 | 533 | /* Caller should have device mutex */ |
a9709d68 | 534 | void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_umem *umem) |
150b9e51 | 535 | { |
47283bef MT |
536 | int i; |
537 | ||
ea5d4046 | 538 | vhost_dev_cleanup(dev, true); |
3a4d5c94 | 539 | |
150b9e51 | 540 | /* Restore memory to default empty mapping. */ |
a9709d68 JW |
541 | INIT_LIST_HEAD(&umem->umem_list); |
542 | dev->umem = umem; | |
47283bef MT |
543 | /* We don't need VQ locks below since vhost_dev_cleanup makes sure |
544 | * VQs aren't running. | |
545 | */ | |
546 | for (i = 0; i < dev->nvqs; ++i) | |
a9709d68 | 547 | dev->vqs[i]->umem = umem; |
3a4d5c94 | 548 | } |
6ac1afbf | 549 | EXPORT_SYMBOL_GPL(vhost_dev_reset_owner); |
3a4d5c94 | 550 | |
b211616d | 551 | void vhost_dev_stop(struct vhost_dev *dev) |
bab632d6 MT |
552 | { |
553 | int i; | |
b211616d MT |
554 | |
555 | for (i = 0; i < dev->nvqs; ++i) { | |
3ab2e420 AH |
556 | if (dev->vqs[i]->kick && dev->vqs[i]->handle_kick) { |
557 | vhost_poll_stop(&dev->vqs[i]->poll); | |
558 | vhost_poll_flush(&dev->vqs[i]->poll); | |
b211616d | 559 | } |
bab632d6 | 560 | } |
bab632d6 | 561 | } |
6ac1afbf | 562 | EXPORT_SYMBOL_GPL(vhost_dev_stop); |
bab632d6 | 563 | |
6b1e6cc7 JW |
564 | static void vhost_umem_free(struct vhost_umem *umem, |
565 | struct vhost_umem_node *node) | |
566 | { | |
567 | vhost_umem_interval_tree_remove(node, &umem->umem_tree); | |
568 | list_del(&node->link); | |
569 | kfree(node); | |
570 | umem->numem--; | |
571 | } | |
572 | ||
a9709d68 JW |
573 | static void vhost_umem_clean(struct vhost_umem *umem) |
574 | { | |
575 | struct vhost_umem_node *node, *tmp; | |
576 | ||
577 | if (!umem) | |
578 | return; | |
579 | ||
6b1e6cc7 JW |
580 | list_for_each_entry_safe(node, tmp, &umem->umem_list, link) |
581 | vhost_umem_free(umem, node); | |
582 | ||
a9709d68 JW |
583 | kvfree(umem); |
584 | } | |
585 | ||
6b1e6cc7 JW |
586 | static void vhost_clear_msg(struct vhost_dev *dev) |
587 | { | |
588 | struct vhost_msg_node *node, *n; | |
589 | ||
590 | spin_lock(&dev->iotlb_lock); | |
591 | ||
592 | list_for_each_entry_safe(node, n, &dev->read_list, node) { | |
593 | list_del(&node->node); | |
594 | kfree(node); | |
595 | } | |
596 | ||
597 | list_for_each_entry_safe(node, n, &dev->pending_list, node) { | |
598 | list_del(&node->node); | |
599 | kfree(node); | |
600 | } | |
601 | ||
602 | spin_unlock(&dev->iotlb_lock); | |
603 | } | |
604 | ||
ea5d4046 MT |
605 | /* Caller should have device mutex if and only if locked is set */ |
606 | void vhost_dev_cleanup(struct vhost_dev *dev, bool locked) | |
3a4d5c94 MT |
607 | { |
608 | int i; | |
d47effe1 | 609 | |
3a4d5c94 | 610 | for (i = 0; i < dev->nvqs; ++i) { |
3ab2e420 AH |
611 | if (dev->vqs[i]->error_ctx) |
612 | eventfd_ctx_put(dev->vqs[i]->error_ctx); | |
613 | if (dev->vqs[i]->error) | |
614 | fput(dev->vqs[i]->error); | |
615 | if (dev->vqs[i]->kick) | |
616 | fput(dev->vqs[i]->kick); | |
617 | if (dev->vqs[i]->call_ctx) | |
618 | eventfd_ctx_put(dev->vqs[i]->call_ctx); | |
619 | if (dev->vqs[i]->call) | |
620 | fput(dev->vqs[i]->call); | |
621 | vhost_vq_reset(dev, dev->vqs[i]); | |
3a4d5c94 | 622 | } |
e0e9b406 | 623 | vhost_dev_free_iovecs(dev); |
3a4d5c94 MT |
624 | if (dev->log_ctx) |
625 | eventfd_ctx_put(dev->log_ctx); | |
626 | dev->log_ctx = NULL; | |
627 | if (dev->log_file) | |
628 | fput(dev->log_file); | |
629 | dev->log_file = NULL; | |
630 | /* No one will access memory at this point */ | |
a9709d68 JW |
631 | vhost_umem_clean(dev->umem); |
632 | dev->umem = NULL; | |
6b1e6cc7 JW |
633 | vhost_umem_clean(dev->iotlb); |
634 | dev->iotlb = NULL; | |
635 | vhost_clear_msg(dev); | |
636 | wake_up_interruptible_poll(&dev->wait, POLLIN | POLLRDNORM); | |
04b96e55 | 637 | WARN_ON(!llist_empty(&dev->work_list)); |
78b620ce ED |
638 | if (dev->worker) { |
639 | kthread_stop(dev->worker); | |
640 | dev->worker = NULL; | |
641 | } | |
533a19b4 MT |
642 | if (dev->mm) |
643 | mmput(dev->mm); | |
644 | dev->mm = NULL; | |
3a4d5c94 | 645 | } |
6ac1afbf | 646 | EXPORT_SYMBOL_GPL(vhost_dev_cleanup); |
3a4d5c94 MT |
647 | |
648 | static int log_access_ok(void __user *log_base, u64 addr, unsigned long sz) | |
649 | { | |
650 | u64 a = addr / VHOST_PAGE_SIZE / 8; | |
d47effe1 | 651 | |
3a4d5c94 MT |
652 | /* Make sure 64 bit math will not overflow. */ |
653 | if (a > ULONG_MAX - (unsigned long)log_base || | |
654 | a + (unsigned long)log_base > ULONG_MAX) | |
6d97e55f | 655 | return 0; |
3a4d5c94 MT |
656 | |
657 | return access_ok(VERIFY_WRITE, log_base + a, | |
658 | (sz + VHOST_PAGE_SIZE * 8 - 1) / VHOST_PAGE_SIZE / 8); | |
659 | } | |
660 | ||
ec33d031 MT |
661 | static bool vhost_overflow(u64 uaddr, u64 size) |
662 | { | |
663 | /* Make sure 64 bit math will not overflow. */ | |
664 | return uaddr > ULONG_MAX || size > ULONG_MAX || uaddr > ULONG_MAX - size; | |
665 | } | |
666 | ||
3a4d5c94 | 667 | /* Caller should have vq mutex and device mutex. */ |
a9709d68 | 668 | static int vq_memory_access_ok(void __user *log_base, struct vhost_umem *umem, |
3a4d5c94 MT |
669 | int log_all) |
670 | { | |
a9709d68 | 671 | struct vhost_umem_node *node; |
179b284e | 672 | |
a9709d68 | 673 | if (!umem) |
f8322fbe | 674 | return 0; |
179b284e | 675 | |
a9709d68 JW |
676 | list_for_each_entry(node, &umem->umem_list, link) { |
677 | unsigned long a = node->userspace_addr; | |
678 | ||
ec33d031 | 679 | if (vhost_overflow(node->userspace_addr, node->size)) |
3a4d5c94 | 680 | return 0; |
ec33d031 MT |
681 | |
682 | ||
683 | if (!access_ok(VERIFY_WRITE, (void __user *)a, | |
a9709d68 | 684 | node->size)) |
3a4d5c94 MT |
685 | return 0; |
686 | else if (log_all && !log_access_ok(log_base, | |
a9709d68 JW |
687 | node->start, |
688 | node->size)) | |
3a4d5c94 MT |
689 | return 0; |
690 | } | |
691 | return 1; | |
692 | } | |
693 | ||
694 | /* Can we switch to this memory table? */ | |
695 | /* Caller should have device mutex but not vq mutex */ | |
a9709d68 | 696 | static int memory_access_ok(struct vhost_dev *d, struct vhost_umem *umem, |
3a4d5c94 MT |
697 | int log_all) |
698 | { | |
699 | int i; | |
d47effe1 | 700 | |
3a4d5c94 MT |
701 | for (i = 0; i < d->nvqs; ++i) { |
702 | int ok; | |
ea16c514 MT |
703 | bool log; |
704 | ||
3ab2e420 | 705 | mutex_lock(&d->vqs[i]->mutex); |
ea16c514 | 706 | log = log_all || vhost_has_feature(d->vqs[i], VHOST_F_LOG_ALL); |
3a4d5c94 | 707 | /* If ring is inactive, will check when it's enabled. */ |
3ab2e420 | 708 | if (d->vqs[i]->private_data) |
a9709d68 JW |
709 | ok = vq_memory_access_ok(d->vqs[i]->log_base, |
710 | umem, log); | |
3a4d5c94 MT |
711 | else |
712 | ok = 1; | |
3ab2e420 | 713 | mutex_unlock(&d->vqs[i]->mutex); |
3a4d5c94 MT |
714 | if (!ok) |
715 | return 0; | |
716 | } | |
717 | return 1; | |
718 | } | |
719 | ||
6b1e6cc7 JW |
720 | static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len, |
721 | struct iovec iov[], int iov_size, int access); | |
bfe2bc51 | 722 | |
72952cc0 | 723 | static int vhost_copy_to_user(struct vhost_virtqueue *vq, void __user *to, |
bfe2bc51 JW |
724 | const void *from, unsigned size) |
725 | { | |
6b1e6cc7 | 726 | int ret; |
bfe2bc51 | 727 | |
6b1e6cc7 JW |
728 | if (!vq->iotlb) |
729 | return __copy_to_user(to, from, size); | |
730 | else { | |
731 | /* This function should be called after iotlb | |
732 | * prefetch, which means we're sure that all vq | |
733 | * could be access through iotlb. So -EAGAIN should | |
734 | * not happen in this case. | |
735 | */ | |
736 | /* TODO: more fast path */ | |
737 | struct iov_iter t; | |
738 | ret = translate_desc(vq, (u64)(uintptr_t)to, size, vq->iotlb_iov, | |
739 | ARRAY_SIZE(vq->iotlb_iov), | |
740 | VHOST_ACCESS_WO); | |
741 | if (ret < 0) | |
742 | goto out; | |
743 | iov_iter_init(&t, WRITE, vq->iotlb_iov, ret, size); | |
744 | ret = copy_to_iter(from, size, &t); | |
745 | if (ret == size) | |
746 | ret = 0; | |
747 | } | |
748 | out: | |
749 | return ret; | |
750 | } | |
bfe2bc51 JW |
751 | |
752 | static int vhost_copy_from_user(struct vhost_virtqueue *vq, void *to, | |
72952cc0 | 753 | void __user *from, unsigned size) |
bfe2bc51 | 754 | { |
6b1e6cc7 JW |
755 | int ret; |
756 | ||
757 | if (!vq->iotlb) | |
758 | return __copy_from_user(to, from, size); | |
759 | else { | |
760 | /* This function should be called after iotlb | |
761 | * prefetch, which means we're sure that vq | |
762 | * could be access through iotlb. So -EAGAIN should | |
763 | * not happen in this case. | |
764 | */ | |
765 | /* TODO: more fast path */ | |
766 | struct iov_iter f; | |
767 | ret = translate_desc(vq, (u64)(uintptr_t)from, size, vq->iotlb_iov, | |
768 | ARRAY_SIZE(vq->iotlb_iov), | |
769 | VHOST_ACCESS_RO); | |
770 | if (ret < 0) { | |
771 | vq_err(vq, "IOTLB translation failure: uaddr " | |
772 | "%p size 0x%llx\n", from, | |
773 | (unsigned long long) size); | |
774 | goto out; | |
775 | } | |
776 | iov_iter_init(&f, READ, vq->iotlb_iov, ret, size); | |
777 | ret = copy_from_iter(to, size, &f); | |
778 | if (ret == size) | |
779 | ret = 0; | |
780 | } | |
781 | ||
782 | out: | |
783 | return ret; | |
784 | } | |
785 | ||
786 | static void __user *__vhost_get_user(struct vhost_virtqueue *vq, | |
72952cc0 | 787 | void __user *addr, unsigned size) |
6b1e6cc7 JW |
788 | { |
789 | int ret; | |
790 | ||
791 | /* This function should be called after iotlb | |
792 | * prefetch, which means we're sure that vq | |
793 | * could be access through iotlb. So -EAGAIN should | |
794 | * not happen in this case. | |
795 | */ | |
796 | /* TODO: more fast path */ | |
797 | ret = translate_desc(vq, (u64)(uintptr_t)addr, size, vq->iotlb_iov, | |
798 | ARRAY_SIZE(vq->iotlb_iov), | |
799 | VHOST_ACCESS_RO); | |
800 | if (ret < 0) { | |
801 | vq_err(vq, "IOTLB translation failure: uaddr " | |
802 | "%p size 0x%llx\n", addr, | |
803 | (unsigned long long) size); | |
804 | return NULL; | |
805 | } | |
806 | ||
807 | if (ret != 1 || vq->iotlb_iov[0].iov_len != size) { | |
808 | vq_err(vq, "Non atomic userspace memory access: uaddr " | |
809 | "%p size 0x%llx\n", addr, | |
810 | (unsigned long long) size); | |
811 | return NULL; | |
812 | } | |
813 | ||
814 | return vq->iotlb_iov[0].iov_base; | |
815 | } | |
816 | ||
817 | #define vhost_put_user(vq, x, ptr) \ | |
818 | ({ \ | |
819 | int ret = -EFAULT; \ | |
820 | if (!vq->iotlb) { \ | |
821 | ret = __put_user(x, ptr); \ | |
822 | } else { \ | |
823 | __typeof__(ptr) to = \ | |
824 | (__typeof__(ptr)) __vhost_get_user(vq, ptr, sizeof(*ptr)); \ | |
825 | if (to != NULL) \ | |
826 | ret = __put_user(x, to); \ | |
827 | else \ | |
828 | ret = -EFAULT; \ | |
829 | } \ | |
830 | ret; \ | |
831 | }) | |
832 | ||
833 | #define vhost_get_user(vq, x, ptr) \ | |
834 | ({ \ | |
835 | int ret; \ | |
836 | if (!vq->iotlb) { \ | |
837 | ret = __get_user(x, ptr); \ | |
838 | } else { \ | |
839 | __typeof__(ptr) from = \ | |
840 | (__typeof__(ptr)) __vhost_get_user(vq, ptr, sizeof(*ptr)); \ | |
841 | if (from != NULL) \ | |
842 | ret = __get_user(x, from); \ | |
843 | else \ | |
844 | ret = -EFAULT; \ | |
845 | } \ | |
846 | ret; \ | |
847 | }) | |
848 | ||
849 | static void vhost_dev_lock_vqs(struct vhost_dev *d) | |
850 | { | |
851 | int i = 0; | |
852 | for (i = 0; i < d->nvqs; ++i) | |
853 | mutex_lock(&d->vqs[i]->mutex); | |
854 | } | |
855 | ||
856 | static void vhost_dev_unlock_vqs(struct vhost_dev *d) | |
857 | { | |
858 | int i = 0; | |
859 | for (i = 0; i < d->nvqs; ++i) | |
860 | mutex_unlock(&d->vqs[i]->mutex); | |
861 | } | |
862 | ||
863 | static int vhost_new_umem_range(struct vhost_umem *umem, | |
864 | u64 start, u64 size, u64 end, | |
865 | u64 userspace_addr, int perm) | |
866 | { | |
867 | struct vhost_umem_node *tmp, *node = kmalloc(sizeof(*node), GFP_ATOMIC); | |
868 | ||
869 | if (!node) | |
870 | return -ENOMEM; | |
871 | ||
872 | if (umem->numem == max_iotlb_entries) { | |
873 | tmp = list_first_entry(&umem->umem_list, typeof(*tmp), link); | |
874 | vhost_umem_free(umem, tmp); | |
875 | } | |
876 | ||
877 | node->start = start; | |
878 | node->size = size; | |
879 | node->last = end; | |
880 | node->userspace_addr = userspace_addr; | |
881 | node->perm = perm; | |
882 | INIT_LIST_HEAD(&node->link); | |
883 | list_add_tail(&node->link, &umem->umem_list); | |
884 | vhost_umem_interval_tree_insert(node, &umem->umem_tree); | |
885 | umem->numem++; | |
886 | ||
887 | return 0; | |
888 | } | |
889 | ||
890 | static void vhost_del_umem_range(struct vhost_umem *umem, | |
891 | u64 start, u64 end) | |
892 | { | |
893 | struct vhost_umem_node *node; | |
894 | ||
895 | while ((node = vhost_umem_interval_tree_iter_first(&umem->umem_tree, | |
896 | start, end))) | |
897 | vhost_umem_free(umem, node); | |
898 | } | |
899 | ||
900 | static void vhost_iotlb_notify_vq(struct vhost_dev *d, | |
901 | struct vhost_iotlb_msg *msg) | |
902 | { | |
903 | struct vhost_msg_node *node, *n; | |
904 | ||
905 | spin_lock(&d->iotlb_lock); | |
906 | ||
907 | list_for_each_entry_safe(node, n, &d->pending_list, node) { | |
908 | struct vhost_iotlb_msg *vq_msg = &node->msg.iotlb; | |
909 | if (msg->iova <= vq_msg->iova && | |
910 | msg->iova + msg->size - 1 > vq_msg->iova && | |
911 | vq_msg->type == VHOST_IOTLB_MISS) { | |
912 | vhost_poll_queue(&node->vq->poll); | |
913 | list_del(&node->node); | |
914 | kfree(node); | |
915 | } | |
916 | } | |
917 | ||
918 | spin_unlock(&d->iotlb_lock); | |
919 | } | |
920 | ||
921 | static int umem_access_ok(u64 uaddr, u64 size, int access) | |
922 | { | |
923 | unsigned long a = uaddr; | |
924 | ||
ec33d031 MT |
925 | /* Make sure 64 bit math will not overflow. */ |
926 | if (vhost_overflow(uaddr, size)) | |
927 | return -EFAULT; | |
928 | ||
6b1e6cc7 JW |
929 | if ((access & VHOST_ACCESS_RO) && |
930 | !access_ok(VERIFY_READ, (void __user *)a, size)) | |
931 | return -EFAULT; | |
932 | if ((access & VHOST_ACCESS_WO) && | |
933 | !access_ok(VERIFY_WRITE, (void __user *)a, size)) | |
934 | return -EFAULT; | |
935 | return 0; | |
936 | } | |
937 | ||
72952cc0 MT |
938 | static int vhost_process_iotlb_msg(struct vhost_dev *dev, |
939 | struct vhost_iotlb_msg *msg) | |
6b1e6cc7 JW |
940 | { |
941 | int ret = 0; | |
942 | ||
943 | vhost_dev_lock_vqs(dev); | |
944 | switch (msg->type) { | |
945 | case VHOST_IOTLB_UPDATE: | |
946 | if (!dev->iotlb) { | |
947 | ret = -EFAULT; | |
948 | break; | |
949 | } | |
950 | if (umem_access_ok(msg->uaddr, msg->size, msg->perm)) { | |
951 | ret = -EFAULT; | |
952 | break; | |
953 | } | |
954 | if (vhost_new_umem_range(dev->iotlb, msg->iova, msg->size, | |
955 | msg->iova + msg->size - 1, | |
956 | msg->uaddr, msg->perm)) { | |
957 | ret = -ENOMEM; | |
958 | break; | |
959 | } | |
960 | vhost_iotlb_notify_vq(dev, msg); | |
961 | break; | |
962 | case VHOST_IOTLB_INVALIDATE: | |
963 | vhost_del_umem_range(dev->iotlb, msg->iova, | |
964 | msg->iova + msg->size - 1); | |
965 | break; | |
966 | default: | |
967 | ret = -EINVAL; | |
968 | break; | |
969 | } | |
970 | ||
971 | vhost_dev_unlock_vqs(dev); | |
972 | return ret; | |
973 | } | |
974 | ssize_t vhost_chr_write_iter(struct vhost_dev *dev, | |
975 | struct iov_iter *from) | |
976 | { | |
977 | struct vhost_msg_node node; | |
978 | unsigned size = sizeof(struct vhost_msg); | |
979 | size_t ret; | |
980 | int err; | |
981 | ||
982 | if (iov_iter_count(from) < size) | |
983 | return 0; | |
984 | ret = copy_from_iter(&node.msg, size, from); | |
985 | if (ret != size) | |
986 | goto done; | |
987 | ||
988 | switch (node.msg.type) { | |
989 | case VHOST_IOTLB_MSG: | |
990 | err = vhost_process_iotlb_msg(dev, &node.msg.iotlb); | |
991 | if (err) | |
992 | ret = err; | |
993 | break; | |
994 | default: | |
995 | ret = -EINVAL; | |
996 | break; | |
997 | } | |
998 | ||
999 | done: | |
1000 | return ret; | |
1001 | } | |
1002 | EXPORT_SYMBOL(vhost_chr_write_iter); | |
1003 | ||
1004 | unsigned int vhost_chr_poll(struct file *file, struct vhost_dev *dev, | |
1005 | poll_table *wait) | |
1006 | { | |
1007 | unsigned int mask = 0; | |
1008 | ||
1009 | poll_wait(file, &dev->wait, wait); | |
1010 | ||
1011 | if (!list_empty(&dev->read_list)) | |
1012 | mask |= POLLIN | POLLRDNORM; | |
1013 | ||
1014 | return mask; | |
1015 | } | |
1016 | EXPORT_SYMBOL(vhost_chr_poll); | |
1017 | ||
1018 | ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to, | |
1019 | int noblock) | |
1020 | { | |
1021 | DEFINE_WAIT(wait); | |
1022 | struct vhost_msg_node *node; | |
1023 | ssize_t ret = 0; | |
1024 | unsigned size = sizeof(struct vhost_msg); | |
1025 | ||
1026 | if (iov_iter_count(to) < size) | |
1027 | return 0; | |
1028 | ||
1029 | while (1) { | |
1030 | if (!noblock) | |
1031 | prepare_to_wait(&dev->wait, &wait, | |
1032 | TASK_INTERRUPTIBLE); | |
1033 | ||
1034 | node = vhost_dequeue_msg(dev, &dev->read_list); | |
1035 | if (node) | |
1036 | break; | |
1037 | if (noblock) { | |
1038 | ret = -EAGAIN; | |
1039 | break; | |
1040 | } | |
1041 | if (signal_pending(current)) { | |
1042 | ret = -ERESTARTSYS; | |
1043 | break; | |
1044 | } | |
1045 | if (!dev->iotlb) { | |
1046 | ret = -EBADFD; | |
1047 | break; | |
1048 | } | |
1049 | ||
1050 | schedule(); | |
1051 | } | |
1052 | ||
1053 | if (!noblock) | |
1054 | finish_wait(&dev->wait, &wait); | |
1055 | ||
1056 | if (node) { | |
1057 | ret = copy_to_iter(&node->msg, size, to); | |
1058 | ||
1059 | if (ret != size || node->msg.type != VHOST_IOTLB_MISS) { | |
1060 | kfree(node); | |
1061 | return ret; | |
1062 | } | |
1063 | ||
1064 | vhost_enqueue_msg(dev, &dev->pending_list, node); | |
1065 | } | |
1066 | ||
1067 | return ret; | |
1068 | } | |
1069 | EXPORT_SYMBOL_GPL(vhost_chr_read_iter); | |
1070 | ||
1071 | static int vhost_iotlb_miss(struct vhost_virtqueue *vq, u64 iova, int access) | |
1072 | { | |
1073 | struct vhost_dev *dev = vq->dev; | |
1074 | struct vhost_msg_node *node; | |
1075 | struct vhost_iotlb_msg *msg; | |
1076 | ||
1077 | node = vhost_new_msg(vq, VHOST_IOTLB_MISS); | |
1078 | if (!node) | |
1079 | return -ENOMEM; | |
1080 | ||
1081 | msg = &node->msg.iotlb; | |
1082 | msg->type = VHOST_IOTLB_MISS; | |
1083 | msg->iova = iova; | |
1084 | msg->perm = access; | |
1085 | ||
1086 | vhost_enqueue_msg(dev, &dev->read_list, node); | |
1087 | ||
1088 | return 0; | |
bfe2bc51 JW |
1089 | } |
1090 | ||
ea16c514 | 1091 | static int vq_access_ok(struct vhost_virtqueue *vq, unsigned int num, |
3a4d5c94 MT |
1092 | struct vring_desc __user *desc, |
1093 | struct vring_avail __user *avail, | |
1094 | struct vring_used __user *used) | |
6b1e6cc7 | 1095 | |
3a4d5c94 | 1096 | { |
ea16c514 | 1097 | size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; |
6b1e6cc7 | 1098 | |
3a4d5c94 MT |
1099 | return access_ok(VERIFY_READ, desc, num * sizeof *desc) && |
1100 | access_ok(VERIFY_READ, avail, | |
8ea8cf89 | 1101 | sizeof *avail + num * sizeof *avail->ring + s) && |
3a4d5c94 | 1102 | access_ok(VERIFY_WRITE, used, |
8ea8cf89 | 1103 | sizeof *used + num * sizeof *used->ring + s); |
3a4d5c94 MT |
1104 | } |
1105 | ||
6b1e6cc7 JW |
1106 | static int iotlb_access_ok(struct vhost_virtqueue *vq, |
1107 | int access, u64 addr, u64 len) | |
1108 | { | |
1109 | const struct vhost_umem_node *node; | |
1110 | struct vhost_umem *umem = vq->iotlb; | |
1111 | u64 s = 0, size; | |
1112 | ||
1113 | while (len > s) { | |
1114 | node = vhost_umem_interval_tree_iter_first(&umem->umem_tree, | |
1115 | addr, | |
1116 | addr + len - 1); | |
1117 | if (node == NULL || node->start > addr) { | |
1118 | vhost_iotlb_miss(vq, addr, access); | |
1119 | return false; | |
1120 | } else if (!(node->perm & access)) { | |
1121 | /* Report the possible access violation by | |
1122 | * request another translation from userspace. | |
1123 | */ | |
1124 | return false; | |
1125 | } | |
1126 | ||
1127 | size = node->size - addr + node->start; | |
1128 | s += size; | |
1129 | addr += size; | |
1130 | } | |
1131 | ||
1132 | return true; | |
1133 | } | |
1134 | ||
1135 | int vq_iotlb_prefetch(struct vhost_virtqueue *vq) | |
1136 | { | |
1137 | size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; | |
1138 | unsigned int num = vq->num; | |
1139 | ||
1140 | if (!vq->iotlb) | |
1141 | return 1; | |
1142 | ||
1143 | return iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->desc, | |
1144 | num * sizeof *vq->desc) && | |
1145 | iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->avail, | |
1146 | sizeof *vq->avail + | |
1147 | num * sizeof *vq->avail->ring + s) && | |
1148 | iotlb_access_ok(vq, VHOST_ACCESS_WO, (u64)(uintptr_t)vq->used, | |
1149 | sizeof *vq->used + | |
1150 | num * sizeof *vq->used->ring + s); | |
1151 | } | |
1152 | EXPORT_SYMBOL_GPL(vq_iotlb_prefetch); | |
1153 | ||
3a4d5c94 MT |
1154 | /* Can we log writes? */ |
1155 | /* Caller should have device mutex but not vq mutex */ | |
1156 | int vhost_log_access_ok(struct vhost_dev *dev) | |
1157 | { | |
a9709d68 | 1158 | return memory_access_ok(dev, dev->umem, 1); |
3a4d5c94 | 1159 | } |
6ac1afbf | 1160 | EXPORT_SYMBOL_GPL(vhost_log_access_ok); |
3a4d5c94 MT |
1161 | |
1162 | /* Verify access for write logging. */ | |
1163 | /* Caller should have vq mutex and device mutex */ | |
ea16c514 | 1164 | static int vq_log_access_ok(struct vhost_virtqueue *vq, |
8ea8cf89 | 1165 | void __user *log_base) |
3a4d5c94 | 1166 | { |
ea16c514 | 1167 | size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; |
28457ee6 | 1168 | |
a9709d68 | 1169 | return vq_memory_access_ok(log_base, vq->umem, |
ea16c514 | 1170 | vhost_has_feature(vq, VHOST_F_LOG_ALL)) && |
3a4d5c94 MT |
1171 | (!vq->log_used || log_access_ok(log_base, vq->log_addr, |
1172 | sizeof *vq->used + | |
8ea8cf89 | 1173 | vq->num * sizeof *vq->used->ring + s)); |
3a4d5c94 MT |
1174 | } |
1175 | ||
1176 | /* Can we start vq? */ | |
1177 | /* Caller should have vq mutex and device mutex */ | |
1178 | int vhost_vq_access_ok(struct vhost_virtqueue *vq) | |
1179 | { | |
6b1e6cc7 JW |
1180 | if (vq->iotlb) { |
1181 | /* When device IOTLB was used, the access validation | |
1182 | * will be validated during prefetching. | |
1183 | */ | |
1184 | return 1; | |
1185 | } | |
ea16c514 MT |
1186 | return vq_access_ok(vq, vq->num, vq->desc, vq->avail, vq->used) && |
1187 | vq_log_access_ok(vq, vq->log_base); | |
3a4d5c94 | 1188 | } |
6ac1afbf | 1189 | EXPORT_SYMBOL_GPL(vhost_vq_access_ok); |
3a4d5c94 | 1190 | |
6b1e6cc7 JW |
1191 | static struct vhost_umem *vhost_umem_alloc(void) |
1192 | { | |
1193 | struct vhost_umem *umem = vhost_kvzalloc(sizeof(*umem)); | |
1194 | ||
1195 | if (!umem) | |
1196 | return NULL; | |
1197 | ||
1198 | umem->umem_tree = RB_ROOT; | |
1199 | umem->numem = 0; | |
1200 | INIT_LIST_HEAD(&umem->umem_list); | |
1201 | ||
1202 | return umem; | |
1203 | } | |
1204 | ||
3a4d5c94 MT |
1205 | static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m) |
1206 | { | |
a9709d68 JW |
1207 | struct vhost_memory mem, *newmem; |
1208 | struct vhost_memory_region *region; | |
a9709d68 | 1209 | struct vhost_umem *newumem, *oldumem; |
3a4d5c94 | 1210 | unsigned long size = offsetof(struct vhost_memory, regions); |
98f9ca0a | 1211 | int i; |
d47effe1 | 1212 | |
7ad9c9d2 TY |
1213 | if (copy_from_user(&mem, m, size)) |
1214 | return -EFAULT; | |
3a4d5c94 MT |
1215 | if (mem.padding) |
1216 | return -EOPNOTSUPP; | |
c9ce42f7 | 1217 | if (mem.nregions > max_mem_regions) |
3a4d5c94 | 1218 | return -E2BIG; |
4de7255f | 1219 | newmem = vhost_kvzalloc(size + mem.nregions * sizeof(*m->regions)); |
3a4d5c94 MT |
1220 | if (!newmem) |
1221 | return -ENOMEM; | |
1222 | ||
1223 | memcpy(newmem, &mem, size); | |
7ad9c9d2 TY |
1224 | if (copy_from_user(newmem->regions, m->regions, |
1225 | mem.nregions * sizeof *m->regions)) { | |
bcfeacab | 1226 | kvfree(newmem); |
7ad9c9d2 | 1227 | return -EFAULT; |
3a4d5c94 MT |
1228 | } |
1229 | ||
6b1e6cc7 | 1230 | newumem = vhost_umem_alloc(); |
a9709d68 | 1231 | if (!newumem) { |
4de7255f | 1232 | kvfree(newmem); |
a9709d68 JW |
1233 | return -ENOMEM; |
1234 | } | |
1235 | ||
a9709d68 JW |
1236 | for (region = newmem->regions; |
1237 | region < newmem->regions + mem.nregions; | |
1238 | region++) { | |
6b1e6cc7 JW |
1239 | if (vhost_new_umem_range(newumem, |
1240 | region->guest_phys_addr, | |
1241 | region->memory_size, | |
1242 | region->guest_phys_addr + | |
1243 | region->memory_size - 1, | |
1244 | region->userspace_addr, | |
1245 | VHOST_ACCESS_RW)) | |
a9709d68 | 1246 | goto err; |
a02c3789 | 1247 | } |
a9709d68 JW |
1248 | |
1249 | if (!memory_access_ok(d, newumem, 0)) | |
1250 | goto err; | |
1251 | ||
1252 | oldumem = d->umem; | |
1253 | d->umem = newumem; | |
98f9ca0a | 1254 | |
47283bef | 1255 | /* All memory accesses are done under some VQ mutex. */ |
98f9ca0a MT |
1256 | for (i = 0; i < d->nvqs; ++i) { |
1257 | mutex_lock(&d->vqs[i]->mutex); | |
a9709d68 | 1258 | d->vqs[i]->umem = newumem; |
98f9ca0a MT |
1259 | mutex_unlock(&d->vqs[i]->mutex); |
1260 | } | |
a9709d68 JW |
1261 | |
1262 | kvfree(newmem); | |
1263 | vhost_umem_clean(oldumem); | |
3a4d5c94 | 1264 | return 0; |
a9709d68 JW |
1265 | |
1266 | err: | |
1267 | vhost_umem_clean(newumem); | |
1268 | kvfree(newmem); | |
1269 | return -EFAULT; | |
3a4d5c94 MT |
1270 | } |
1271 | ||
935cdee7 | 1272 | long vhost_vring_ioctl(struct vhost_dev *d, int ioctl, void __user *argp) |
3a4d5c94 | 1273 | { |
cecb46f1 AV |
1274 | struct file *eventfp, *filep = NULL; |
1275 | bool pollstart = false, pollstop = false; | |
3a4d5c94 MT |
1276 | struct eventfd_ctx *ctx = NULL; |
1277 | u32 __user *idxp = argp; | |
1278 | struct vhost_virtqueue *vq; | |
1279 | struct vhost_vring_state s; | |
1280 | struct vhost_vring_file f; | |
1281 | struct vhost_vring_addr a; | |
1282 | u32 idx; | |
1283 | long r; | |
1284 | ||
1285 | r = get_user(idx, idxp); | |
1286 | if (r < 0) | |
1287 | return r; | |
0f3d9a17 | 1288 | if (idx >= d->nvqs) |
3a4d5c94 MT |
1289 | return -ENOBUFS; |
1290 | ||
3ab2e420 | 1291 | vq = d->vqs[idx]; |
3a4d5c94 MT |
1292 | |
1293 | mutex_lock(&vq->mutex); | |
1294 | ||
1295 | switch (ioctl) { | |
1296 | case VHOST_SET_VRING_NUM: | |
1297 | /* Resizing ring with an active backend? | |
1298 | * You don't want to do that. */ | |
1299 | if (vq->private_data) { | |
1300 | r = -EBUSY; | |
1301 | break; | |
1302 | } | |
7ad9c9d2 TY |
1303 | if (copy_from_user(&s, argp, sizeof s)) { |
1304 | r = -EFAULT; | |
3a4d5c94 | 1305 | break; |
7ad9c9d2 | 1306 | } |
3a4d5c94 MT |
1307 | if (!s.num || s.num > 0xffff || (s.num & (s.num - 1))) { |
1308 | r = -EINVAL; | |
1309 | break; | |
1310 | } | |
1311 | vq->num = s.num; | |
1312 | break; | |
1313 | case VHOST_SET_VRING_BASE: | |
1314 | /* Moving base with an active backend? | |
1315 | * You don't want to do that. */ | |
1316 | if (vq->private_data) { | |
1317 | r = -EBUSY; | |
1318 | break; | |
1319 | } | |
7ad9c9d2 TY |
1320 | if (copy_from_user(&s, argp, sizeof s)) { |
1321 | r = -EFAULT; | |
3a4d5c94 | 1322 | break; |
7ad9c9d2 | 1323 | } |
3a4d5c94 MT |
1324 | if (s.num > 0xffff) { |
1325 | r = -EINVAL; | |
1326 | break; | |
1327 | } | |
809ecb9b | 1328 | vq->last_avail_idx = vq->last_used_event = s.num; |
3a4d5c94 MT |
1329 | /* Forget the cached index value. */ |
1330 | vq->avail_idx = vq->last_avail_idx; | |
1331 | break; | |
1332 | case VHOST_GET_VRING_BASE: | |
1333 | s.index = idx; | |
1334 | s.num = vq->last_avail_idx; | |
7ad9c9d2 TY |
1335 | if (copy_to_user(argp, &s, sizeof s)) |
1336 | r = -EFAULT; | |
3a4d5c94 MT |
1337 | break; |
1338 | case VHOST_SET_VRING_ADDR: | |
7ad9c9d2 TY |
1339 | if (copy_from_user(&a, argp, sizeof a)) { |
1340 | r = -EFAULT; | |
3a4d5c94 | 1341 | break; |
7ad9c9d2 | 1342 | } |
3a4d5c94 MT |
1343 | if (a.flags & ~(0x1 << VHOST_VRING_F_LOG)) { |
1344 | r = -EOPNOTSUPP; | |
1345 | break; | |
1346 | } | |
1347 | /* For 32bit, verify that the top 32bits of the user | |
1348 | data are set to zero. */ | |
1349 | if ((u64)(unsigned long)a.desc_user_addr != a.desc_user_addr || | |
1350 | (u64)(unsigned long)a.used_user_addr != a.used_user_addr || | |
1351 | (u64)(unsigned long)a.avail_user_addr != a.avail_user_addr) { | |
1352 | r = -EFAULT; | |
1353 | break; | |
1354 | } | |
5d9a07b0 MT |
1355 | |
1356 | /* Make sure it's safe to cast pointers to vring types. */ | |
1357 | BUILD_BUG_ON(__alignof__ *vq->avail > VRING_AVAIL_ALIGN_SIZE); | |
1358 | BUILD_BUG_ON(__alignof__ *vq->used > VRING_USED_ALIGN_SIZE); | |
1359 | if ((a.avail_user_addr & (VRING_AVAIL_ALIGN_SIZE - 1)) || | |
1360 | (a.used_user_addr & (VRING_USED_ALIGN_SIZE - 1)) || | |
d5424838 | 1361 | (a.log_guest_addr & (VRING_USED_ALIGN_SIZE - 1))) { |
3a4d5c94 MT |
1362 | r = -EINVAL; |
1363 | break; | |
1364 | } | |
1365 | ||
1366 | /* We only verify access here if backend is configured. | |
1367 | * If it is not, we don't as size might not have been setup. | |
1368 | * We will verify when backend is configured. */ | |
1369 | if (vq->private_data) { | |
ea16c514 | 1370 | if (!vq_access_ok(vq, vq->num, |
3a4d5c94 MT |
1371 | (void __user *)(unsigned long)a.desc_user_addr, |
1372 | (void __user *)(unsigned long)a.avail_user_addr, | |
1373 | (void __user *)(unsigned long)a.used_user_addr)) { | |
1374 | r = -EINVAL; | |
1375 | break; | |
1376 | } | |
1377 | ||
1378 | /* Also validate log access for used ring if enabled. */ | |
1379 | if ((a.flags & (0x1 << VHOST_VRING_F_LOG)) && | |
1380 | !log_access_ok(vq->log_base, a.log_guest_addr, | |
1381 | sizeof *vq->used + | |
1382 | vq->num * sizeof *vq->used->ring)) { | |
1383 | r = -EINVAL; | |
1384 | break; | |
1385 | } | |
1386 | } | |
1387 | ||
3a4d5c94 MT |
1388 | vq->log_used = !!(a.flags & (0x1 << VHOST_VRING_F_LOG)); |
1389 | vq->desc = (void __user *)(unsigned long)a.desc_user_addr; | |
1390 | vq->avail = (void __user *)(unsigned long)a.avail_user_addr; | |
1391 | vq->log_addr = a.log_guest_addr; | |
1392 | vq->used = (void __user *)(unsigned long)a.used_user_addr; | |
1393 | break; | |
1394 | case VHOST_SET_VRING_KICK: | |
7ad9c9d2 TY |
1395 | if (copy_from_user(&f, argp, sizeof f)) { |
1396 | r = -EFAULT; | |
3a4d5c94 | 1397 | break; |
7ad9c9d2 | 1398 | } |
3a4d5c94 | 1399 | eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd); |
535297a6 MT |
1400 | if (IS_ERR(eventfp)) { |
1401 | r = PTR_ERR(eventfp); | |
1402 | break; | |
1403 | } | |
3a4d5c94 | 1404 | if (eventfp != vq->kick) { |
cecb46f1 AV |
1405 | pollstop = (filep = vq->kick) != NULL; |
1406 | pollstart = (vq->kick = eventfp) != NULL; | |
3a4d5c94 MT |
1407 | } else |
1408 | filep = eventfp; | |
1409 | break; | |
1410 | case VHOST_SET_VRING_CALL: | |
7ad9c9d2 TY |
1411 | if (copy_from_user(&f, argp, sizeof f)) { |
1412 | r = -EFAULT; | |
3a4d5c94 | 1413 | break; |
7ad9c9d2 | 1414 | } |
3a4d5c94 | 1415 | eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd); |
535297a6 MT |
1416 | if (IS_ERR(eventfp)) { |
1417 | r = PTR_ERR(eventfp); | |
1418 | break; | |
1419 | } | |
3a4d5c94 MT |
1420 | if (eventfp != vq->call) { |
1421 | filep = vq->call; | |
1422 | ctx = vq->call_ctx; | |
1423 | vq->call = eventfp; | |
1424 | vq->call_ctx = eventfp ? | |
1425 | eventfd_ctx_fileget(eventfp) : NULL; | |
1426 | } else | |
1427 | filep = eventfp; | |
1428 | break; | |
1429 | case VHOST_SET_VRING_ERR: | |
7ad9c9d2 TY |
1430 | if (copy_from_user(&f, argp, sizeof f)) { |
1431 | r = -EFAULT; | |
3a4d5c94 | 1432 | break; |
7ad9c9d2 | 1433 | } |
3a4d5c94 | 1434 | eventfp = f.fd == -1 ? NULL : eventfd_fget(f.fd); |
535297a6 MT |
1435 | if (IS_ERR(eventfp)) { |
1436 | r = PTR_ERR(eventfp); | |
1437 | break; | |
1438 | } | |
3a4d5c94 MT |
1439 | if (eventfp != vq->error) { |
1440 | filep = vq->error; | |
1441 | vq->error = eventfp; | |
1442 | ctx = vq->error_ctx; | |
1443 | vq->error_ctx = eventfp ? | |
1444 | eventfd_ctx_fileget(eventfp) : NULL; | |
1445 | } else | |
1446 | filep = eventfp; | |
1447 | break; | |
2751c988 GK |
1448 | case VHOST_SET_VRING_ENDIAN: |
1449 | r = vhost_set_vring_endian(vq, argp); | |
1450 | break; | |
1451 | case VHOST_GET_VRING_ENDIAN: | |
1452 | r = vhost_get_vring_endian(vq, idx, argp); | |
1453 | break; | |
03088137 JW |
1454 | case VHOST_SET_VRING_BUSYLOOP_TIMEOUT: |
1455 | if (copy_from_user(&s, argp, sizeof(s))) { | |
1456 | r = -EFAULT; | |
1457 | break; | |
1458 | } | |
1459 | vq->busyloop_timeout = s.num; | |
1460 | break; | |
1461 | case VHOST_GET_VRING_BUSYLOOP_TIMEOUT: | |
1462 | s.index = idx; | |
1463 | s.num = vq->busyloop_timeout; | |
1464 | if (copy_to_user(argp, &s, sizeof(s))) | |
1465 | r = -EFAULT; | |
1466 | break; | |
3a4d5c94 MT |
1467 | default: |
1468 | r = -ENOIOCTLCMD; | |
1469 | } | |
1470 | ||
1471 | if (pollstop && vq->handle_kick) | |
1472 | vhost_poll_stop(&vq->poll); | |
1473 | ||
1474 | if (ctx) | |
1475 | eventfd_ctx_put(ctx); | |
1476 | if (filep) | |
1477 | fput(filep); | |
1478 | ||
1479 | if (pollstart && vq->handle_kick) | |
2b8b328b | 1480 | r = vhost_poll_start(&vq->poll, vq->kick); |
3a4d5c94 MT |
1481 | |
1482 | mutex_unlock(&vq->mutex); | |
1483 | ||
1484 | if (pollstop && vq->handle_kick) | |
1485 | vhost_poll_flush(&vq->poll); | |
1486 | return r; | |
1487 | } | |
6ac1afbf | 1488 | EXPORT_SYMBOL_GPL(vhost_vring_ioctl); |
3a4d5c94 | 1489 | |
6b1e6cc7 JW |
1490 | int vhost_init_device_iotlb(struct vhost_dev *d, bool enabled) |
1491 | { | |
1492 | struct vhost_umem *niotlb, *oiotlb; | |
1493 | int i; | |
1494 | ||
1495 | niotlb = vhost_umem_alloc(); | |
1496 | if (!niotlb) | |
1497 | return -ENOMEM; | |
1498 | ||
1499 | oiotlb = d->iotlb; | |
1500 | d->iotlb = niotlb; | |
1501 | ||
1502 | for (i = 0; i < d->nvqs; ++i) { | |
1503 | mutex_lock(&d->vqs[i]->mutex); | |
1504 | d->vqs[i]->iotlb = niotlb; | |
1505 | mutex_unlock(&d->vqs[i]->mutex); | |
1506 | } | |
1507 | ||
1508 | vhost_umem_clean(oiotlb); | |
1509 | ||
1510 | return 0; | |
1511 | } | |
1512 | EXPORT_SYMBOL_GPL(vhost_init_device_iotlb); | |
1513 | ||
3a4d5c94 | 1514 | /* Caller must have device mutex */ |
935cdee7 | 1515 | long vhost_dev_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp) |
3a4d5c94 | 1516 | { |
3a4d5c94 MT |
1517 | struct file *eventfp, *filep = NULL; |
1518 | struct eventfd_ctx *ctx = NULL; | |
1519 | u64 p; | |
1520 | long r; | |
1521 | int i, fd; | |
1522 | ||
1523 | /* If you are not the owner, you can become one */ | |
1524 | if (ioctl == VHOST_SET_OWNER) { | |
1525 | r = vhost_dev_set_owner(d); | |
1526 | goto done; | |
1527 | } | |
1528 | ||
1529 | /* You must be the owner to do anything else */ | |
1530 | r = vhost_dev_check_owner(d); | |
1531 | if (r) | |
1532 | goto done; | |
1533 | ||
1534 | switch (ioctl) { | |
1535 | case VHOST_SET_MEM_TABLE: | |
1536 | r = vhost_set_memory(d, argp); | |
1537 | break; | |
1538 | case VHOST_SET_LOG_BASE: | |
7ad9c9d2 TY |
1539 | if (copy_from_user(&p, argp, sizeof p)) { |
1540 | r = -EFAULT; | |
3a4d5c94 | 1541 | break; |
7ad9c9d2 | 1542 | } |
3a4d5c94 MT |
1543 | if ((u64)(unsigned long)p != p) { |
1544 | r = -EFAULT; | |
1545 | break; | |
1546 | } | |
1547 | for (i = 0; i < d->nvqs; ++i) { | |
1548 | struct vhost_virtqueue *vq; | |
1549 | void __user *base = (void __user *)(unsigned long)p; | |
3ab2e420 | 1550 | vq = d->vqs[i]; |
3a4d5c94 MT |
1551 | mutex_lock(&vq->mutex); |
1552 | /* If ring is inactive, will check when it's enabled. */ | |
ea16c514 | 1553 | if (vq->private_data && !vq_log_access_ok(vq, base)) |
3a4d5c94 MT |
1554 | r = -EFAULT; |
1555 | else | |
1556 | vq->log_base = base; | |
1557 | mutex_unlock(&vq->mutex); | |
1558 | } | |
1559 | break; | |
1560 | case VHOST_SET_LOG_FD: | |
1561 | r = get_user(fd, (int __user *)argp); | |
1562 | if (r < 0) | |
1563 | break; | |
1564 | eventfp = fd == -1 ? NULL : eventfd_fget(fd); | |
1565 | if (IS_ERR(eventfp)) { | |
1566 | r = PTR_ERR(eventfp); | |
1567 | break; | |
1568 | } | |
1569 | if (eventfp != d->log_file) { | |
1570 | filep = d->log_file; | |
7932c0bd | 1571 | d->log_file = eventfp; |
3a4d5c94 MT |
1572 | ctx = d->log_ctx; |
1573 | d->log_ctx = eventfp ? | |
1574 | eventfd_ctx_fileget(eventfp) : NULL; | |
1575 | } else | |
1576 | filep = eventfp; | |
1577 | for (i = 0; i < d->nvqs; ++i) { | |
3ab2e420 AH |
1578 | mutex_lock(&d->vqs[i]->mutex); |
1579 | d->vqs[i]->log_ctx = d->log_ctx; | |
1580 | mutex_unlock(&d->vqs[i]->mutex); | |
3a4d5c94 MT |
1581 | } |
1582 | if (ctx) | |
1583 | eventfd_ctx_put(ctx); | |
1584 | if (filep) | |
1585 | fput(filep); | |
1586 | break; | |
1587 | default: | |
935cdee7 | 1588 | r = -ENOIOCTLCMD; |
3a4d5c94 MT |
1589 | break; |
1590 | } | |
1591 | done: | |
1592 | return r; | |
1593 | } | |
6ac1afbf | 1594 | EXPORT_SYMBOL_GPL(vhost_dev_ioctl); |
3a4d5c94 | 1595 | |
3a4d5c94 MT |
1596 | /* TODO: This is really inefficient. We need something like get_user() |
1597 | * (instruction directly accesses the data, with an exception table entry | |
1598 | * returning -EFAULT). See Documentation/x86/exception-tables.txt. | |
1599 | */ | |
1600 | static int set_bit_to_user(int nr, void __user *addr) | |
1601 | { | |
1602 | unsigned long log = (unsigned long)addr; | |
1603 | struct page *page; | |
1604 | void *base; | |
1605 | int bit = nr + (log % PAGE_SIZE) * 8; | |
1606 | int r; | |
d47effe1 | 1607 | |
3a4d5c94 | 1608 | r = get_user_pages_fast(log, 1, 1, &page); |
d6db3f5c | 1609 | if (r < 0) |
3a4d5c94 | 1610 | return r; |
d6db3f5c | 1611 | BUG_ON(r != 1); |
c6daa7ff | 1612 | base = kmap_atomic(page); |
3a4d5c94 | 1613 | set_bit(bit, base); |
c6daa7ff | 1614 | kunmap_atomic(base); |
3a4d5c94 MT |
1615 | set_page_dirty_lock(page); |
1616 | put_page(page); | |
1617 | return 0; | |
1618 | } | |
1619 | ||
1620 | static int log_write(void __user *log_base, | |
1621 | u64 write_address, u64 write_length) | |
1622 | { | |
28831ee6 | 1623 | u64 write_page = write_address / VHOST_PAGE_SIZE; |
3a4d5c94 | 1624 | int r; |
d47effe1 | 1625 | |
3a4d5c94 MT |
1626 | if (!write_length) |
1627 | return 0; | |
3bf9be40 | 1628 | write_length += write_address % VHOST_PAGE_SIZE; |
3a4d5c94 MT |
1629 | for (;;) { |
1630 | u64 base = (u64)(unsigned long)log_base; | |
28831ee6 MT |
1631 | u64 log = base + write_page / 8; |
1632 | int bit = write_page % 8; | |
3a4d5c94 MT |
1633 | if ((u64)(unsigned long)log != log) |
1634 | return -EFAULT; | |
1635 | r = set_bit_to_user(bit, (void __user *)(unsigned long)log); | |
1636 | if (r < 0) | |
1637 | return r; | |
1638 | if (write_length <= VHOST_PAGE_SIZE) | |
1639 | break; | |
1640 | write_length -= VHOST_PAGE_SIZE; | |
28831ee6 | 1641 | write_page += 1; |
3a4d5c94 MT |
1642 | } |
1643 | return r; | |
1644 | } | |
1645 | ||
1646 | int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log, | |
1647 | unsigned int log_num, u64 len) | |
1648 | { | |
1649 | int i, r; | |
1650 | ||
1651 | /* Make sure data written is seen before log. */ | |
5659338c | 1652 | smp_wmb(); |
3a4d5c94 MT |
1653 | for (i = 0; i < log_num; ++i) { |
1654 | u64 l = min(log[i].len, len); | |
1655 | r = log_write(vq->log_base, log[i].addr, l); | |
1656 | if (r < 0) | |
1657 | return r; | |
1658 | len -= l; | |
5786aee8 MT |
1659 | if (!len) { |
1660 | if (vq->log_ctx) | |
1661 | eventfd_signal(vq->log_ctx, 1); | |
3a4d5c94 | 1662 | return 0; |
5786aee8 | 1663 | } |
3a4d5c94 | 1664 | } |
3a4d5c94 MT |
1665 | /* Length written exceeds what we have stored. This is a bug. */ |
1666 | BUG(); | |
1667 | return 0; | |
1668 | } | |
6ac1afbf | 1669 | EXPORT_SYMBOL_GPL(vhost_log_write); |
3a4d5c94 | 1670 | |
2723feaa JW |
1671 | static int vhost_update_used_flags(struct vhost_virtqueue *vq) |
1672 | { | |
1673 | void __user *used; | |
bfe2bc51 JW |
1674 | if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->used_flags), |
1675 | &vq->used->flags) < 0) | |
2723feaa JW |
1676 | return -EFAULT; |
1677 | if (unlikely(vq->log_used)) { | |
1678 | /* Make sure the flag is seen before log. */ | |
1679 | smp_wmb(); | |
1680 | /* Log used flag write. */ | |
1681 | used = &vq->used->flags; | |
1682 | log_write(vq->log_base, vq->log_addr + | |
1683 | (used - (void __user *)vq->used), | |
1684 | sizeof vq->used->flags); | |
1685 | if (vq->log_ctx) | |
1686 | eventfd_signal(vq->log_ctx, 1); | |
1687 | } | |
1688 | return 0; | |
1689 | } | |
1690 | ||
1691 | static int vhost_update_avail_event(struct vhost_virtqueue *vq, u16 avail_event) | |
1692 | { | |
bfe2bc51 JW |
1693 | if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->avail_idx), |
1694 | vhost_avail_event(vq))) | |
2723feaa JW |
1695 | return -EFAULT; |
1696 | if (unlikely(vq->log_used)) { | |
1697 | void __user *used; | |
1698 | /* Make sure the event is seen before log. */ | |
1699 | smp_wmb(); | |
1700 | /* Log avail event write */ | |
1701 | used = vhost_avail_event(vq); | |
1702 | log_write(vq->log_base, vq->log_addr + | |
1703 | (used - (void __user *)vq->used), | |
1704 | sizeof *vhost_avail_event(vq)); | |
1705 | if (vq->log_ctx) | |
1706 | eventfd_signal(vq->log_ctx, 1); | |
1707 | } | |
1708 | return 0; | |
1709 | } | |
1710 | ||
80f7d030 | 1711 | int vhost_vq_init_access(struct vhost_virtqueue *vq) |
2723feaa | 1712 | { |
3b1bbe89 | 1713 | __virtio16 last_used_idx; |
2723feaa | 1714 | int r; |
e1f33be9 GK |
1715 | bool is_le = vq->is_le; |
1716 | ||
cda8bba0 | 1717 | if (!vq->private_data) |
2723feaa | 1718 | return 0; |
2751c988 GK |
1719 | |
1720 | vhost_init_is_le(vq); | |
2723feaa JW |
1721 | |
1722 | r = vhost_update_used_flags(vq); | |
1723 | if (r) | |
e1f33be9 | 1724 | goto err; |
2723feaa | 1725 | vq->signalled_used_valid = false; |
6b1e6cc7 JW |
1726 | if (!vq->iotlb && |
1727 | !access_ok(VERIFY_READ, &vq->used->idx, sizeof vq->used->idx)) { | |
e1f33be9 GK |
1728 | r = -EFAULT; |
1729 | goto err; | |
1730 | } | |
bfe2bc51 | 1731 | r = vhost_get_user(vq, last_used_idx, &vq->used->idx); |
6b1e6cc7 JW |
1732 | if (r) { |
1733 | vq_err(vq, "Can't access used idx at %p\n", | |
1734 | &vq->used->idx); | |
e1f33be9 | 1735 | goto err; |
6b1e6cc7 | 1736 | } |
3b1bbe89 | 1737 | vq->last_used_idx = vhost16_to_cpu(vq, last_used_idx); |
64f7f051 | 1738 | return 0; |
6b1e6cc7 | 1739 | |
e1f33be9 GK |
1740 | err: |
1741 | vq->is_le = is_le; | |
1742 | return r; | |
2723feaa | 1743 | } |
80f7d030 | 1744 | EXPORT_SYMBOL_GPL(vhost_vq_init_access); |
2723feaa | 1745 | |
47283bef | 1746 | static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len, |
6b1e6cc7 | 1747 | struct iovec iov[], int iov_size, int access) |
3a4d5c94 | 1748 | { |
a9709d68 | 1749 | const struct vhost_umem_node *node; |
6b1e6cc7 JW |
1750 | struct vhost_dev *dev = vq->dev; |
1751 | struct vhost_umem *umem = dev->iotlb ? dev->iotlb : dev->umem; | |
3a4d5c94 MT |
1752 | struct iovec *_iov; |
1753 | u64 s = 0; | |
1754 | int ret = 0; | |
1755 | ||
3a4d5c94 MT |
1756 | while ((u64)len > s) { |
1757 | u64 size; | |
7b3384fc | 1758 | if (unlikely(ret >= iov_size)) { |
3a4d5c94 MT |
1759 | ret = -ENOBUFS; |
1760 | break; | |
1761 | } | |
6b1e6cc7 | 1762 | |
a9709d68 JW |
1763 | node = vhost_umem_interval_tree_iter_first(&umem->umem_tree, |
1764 | addr, addr + len - 1); | |
1765 | if (node == NULL || node->start > addr) { | |
6b1e6cc7 JW |
1766 | if (umem != dev->iotlb) { |
1767 | ret = -EFAULT; | |
1768 | break; | |
1769 | } | |
1770 | ret = -EAGAIN; | |
1771 | break; | |
1772 | } else if (!(node->perm & access)) { | |
1773 | ret = -EPERM; | |
3a4d5c94 MT |
1774 | break; |
1775 | } | |
6b1e6cc7 | 1776 | |
3a4d5c94 | 1777 | _iov = iov + ret; |
a9709d68 | 1778 | size = node->size - addr + node->start; |
bd97120f | 1779 | _iov->iov_len = min((u64)len - s, size); |
a8d3782f | 1780 | _iov->iov_base = (void __user *)(unsigned long) |
a9709d68 | 1781 | (node->userspace_addr + addr - node->start); |
3a4d5c94 MT |
1782 | s += size; |
1783 | addr += size; | |
1784 | ++ret; | |
1785 | } | |
1786 | ||
6b1e6cc7 JW |
1787 | if (ret == -EAGAIN) |
1788 | vhost_iotlb_miss(vq, addr, access); | |
3a4d5c94 MT |
1789 | return ret; |
1790 | } | |
1791 | ||
1792 | /* Each buffer in the virtqueues is actually a chain of descriptors. This | |
1793 | * function returns the next descriptor in the chain, | |
1794 | * or -1U if we're at the end. */ | |
3b1bbe89 | 1795 | static unsigned next_desc(struct vhost_virtqueue *vq, struct vring_desc *desc) |
3a4d5c94 MT |
1796 | { |
1797 | unsigned int next; | |
1798 | ||
1799 | /* If this descriptor says it doesn't chain, we're done. */ | |
3b1bbe89 | 1800 | if (!(desc->flags & cpu_to_vhost16(vq, VRING_DESC_F_NEXT))) |
3a4d5c94 MT |
1801 | return -1U; |
1802 | ||
1803 | /* Check they're not leading us off end of descriptors. */ | |
3b1bbe89 | 1804 | next = vhost16_to_cpu(vq, desc->next); |
3a4d5c94 MT |
1805 | /* Make sure compiler knows to grab that: we don't want it changing! */ |
1806 | /* We will use the result as an index in an array, so most | |
1807 | * architectures only need a compiler barrier here. */ | |
1808 | read_barrier_depends(); | |
1809 | ||
1810 | return next; | |
1811 | } | |
1812 | ||
47283bef | 1813 | static int get_indirect(struct vhost_virtqueue *vq, |
7b3384fc MT |
1814 | struct iovec iov[], unsigned int iov_size, |
1815 | unsigned int *out_num, unsigned int *in_num, | |
1816 | struct vhost_log *log, unsigned int *log_num, | |
1817 | struct vring_desc *indirect) | |
3a4d5c94 MT |
1818 | { |
1819 | struct vring_desc desc; | |
1820 | unsigned int i = 0, count, found = 0; | |
3b1bbe89 | 1821 | u32 len = vhost32_to_cpu(vq, indirect->len); |
aad9a1ce | 1822 | struct iov_iter from; |
6b1e6cc7 | 1823 | int ret, access; |
3a4d5c94 MT |
1824 | |
1825 | /* Sanity check */ | |
3b1bbe89 | 1826 | if (unlikely(len % sizeof desc)) { |
3a4d5c94 MT |
1827 | vq_err(vq, "Invalid length in indirect descriptor: " |
1828 | "len 0x%llx not multiple of 0x%zx\n", | |
3b1bbe89 | 1829 | (unsigned long long)len, |
3a4d5c94 MT |
1830 | sizeof desc); |
1831 | return -EINVAL; | |
1832 | } | |
1833 | ||
3b1bbe89 | 1834 | ret = translate_desc(vq, vhost64_to_cpu(vq, indirect->addr), len, vq->indirect, |
6b1e6cc7 | 1835 | UIO_MAXIOV, VHOST_ACCESS_RO); |
7b3384fc | 1836 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
1837 | if (ret != -EAGAIN) |
1838 | vq_err(vq, "Translation failure %d in indirect.\n", ret); | |
3a4d5c94 MT |
1839 | return ret; |
1840 | } | |
aad9a1ce | 1841 | iov_iter_init(&from, READ, vq->indirect, ret, len); |
3a4d5c94 MT |
1842 | |
1843 | /* We will use the result as an address to read from, so most | |
1844 | * architectures only need a compiler barrier here. */ | |
1845 | read_barrier_depends(); | |
1846 | ||
3b1bbe89 | 1847 | count = len / sizeof desc; |
3a4d5c94 MT |
1848 | /* Buffers are chained via a 16 bit next field, so |
1849 | * we can have at most 2^16 of these. */ | |
7b3384fc | 1850 | if (unlikely(count > USHRT_MAX + 1)) { |
3a4d5c94 MT |
1851 | vq_err(vq, "Indirect buffer length too big: %d\n", |
1852 | indirect->len); | |
1853 | return -E2BIG; | |
1854 | } | |
1855 | ||
1856 | do { | |
1857 | unsigned iov_count = *in_num + *out_num; | |
7b3384fc | 1858 | if (unlikely(++found > count)) { |
3a4d5c94 MT |
1859 | vq_err(vq, "Loop detected: last one at %u " |
1860 | "indirect size %u\n", | |
1861 | i, count); | |
1862 | return -EINVAL; | |
1863 | } | |
cbbd26b8 | 1864 | if (unlikely(!copy_from_iter_full(&desc, sizeof(desc), &from))) { |
3a4d5c94 | 1865 | vq_err(vq, "Failed indirect descriptor: idx %d, %zx\n", |
3b1bbe89 | 1866 | i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc); |
3a4d5c94 MT |
1867 | return -EINVAL; |
1868 | } | |
3b1bbe89 | 1869 | if (unlikely(desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT))) { |
3a4d5c94 | 1870 | vq_err(vq, "Nested indirect descriptor: idx %d, %zx\n", |
3b1bbe89 | 1871 | i, (size_t)vhost64_to_cpu(vq, indirect->addr) + i * sizeof desc); |
3a4d5c94 MT |
1872 | return -EINVAL; |
1873 | } | |
1874 | ||
6b1e6cc7 JW |
1875 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE)) |
1876 | access = VHOST_ACCESS_WO; | |
1877 | else | |
1878 | access = VHOST_ACCESS_RO; | |
1879 | ||
3b1bbe89 MT |
1880 | ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr), |
1881 | vhost32_to_cpu(vq, desc.len), iov + iov_count, | |
6b1e6cc7 | 1882 | iov_size - iov_count, access); |
7b3384fc | 1883 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
1884 | if (ret != -EAGAIN) |
1885 | vq_err(vq, "Translation failure %d indirect idx %d\n", | |
1886 | ret, i); | |
3a4d5c94 MT |
1887 | return ret; |
1888 | } | |
1889 | /* If this is an input descriptor, increment that count. */ | |
6b1e6cc7 | 1890 | if (access == VHOST_ACCESS_WO) { |
3a4d5c94 MT |
1891 | *in_num += ret; |
1892 | if (unlikely(log)) { | |
3b1bbe89 MT |
1893 | log[*log_num].addr = vhost64_to_cpu(vq, desc.addr); |
1894 | log[*log_num].len = vhost32_to_cpu(vq, desc.len); | |
3a4d5c94 MT |
1895 | ++*log_num; |
1896 | } | |
1897 | } else { | |
1898 | /* If it's an output descriptor, they're all supposed | |
1899 | * to come before any input descriptors. */ | |
7b3384fc | 1900 | if (unlikely(*in_num)) { |
3a4d5c94 MT |
1901 | vq_err(vq, "Indirect descriptor " |
1902 | "has out after in: idx %d\n", i); | |
1903 | return -EINVAL; | |
1904 | } | |
1905 | *out_num += ret; | |
1906 | } | |
3b1bbe89 | 1907 | } while ((i = next_desc(vq, &desc)) != -1); |
3a4d5c94 MT |
1908 | return 0; |
1909 | } | |
1910 | ||
1911 | /* This looks in the virtqueue and for the first available buffer, and converts | |
1912 | * it to an iovec for convenient access. Since descriptors consist of some | |
1913 | * number of output then some number of input descriptors, it's actually two | |
1914 | * iovecs, but we pack them into one and note how many of each there were. | |
1915 | * | |
d5675bd2 MT |
1916 | * This function returns the descriptor number found, or vq->num (which is |
1917 | * never a valid descriptor number) if none was found. A negative code is | |
1918 | * returned on error. */ | |
47283bef | 1919 | int vhost_get_vq_desc(struct vhost_virtqueue *vq, |
d5675bd2 MT |
1920 | struct iovec iov[], unsigned int iov_size, |
1921 | unsigned int *out_num, unsigned int *in_num, | |
1922 | struct vhost_log *log, unsigned int *log_num) | |
3a4d5c94 MT |
1923 | { |
1924 | struct vring_desc desc; | |
1925 | unsigned int i, head, found = 0; | |
1926 | u16 last_avail_idx; | |
3b1bbe89 MT |
1927 | __virtio16 avail_idx; |
1928 | __virtio16 ring_head; | |
6b1e6cc7 | 1929 | int ret, access; |
3a4d5c94 MT |
1930 | |
1931 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
1932 | last_avail_idx = vq->last_avail_idx; | |
3a4d5c94 | 1933 | |
e3b56cdd JW |
1934 | if (vq->avail_idx == vq->last_avail_idx) { |
1935 | if (unlikely(vhost_get_user(vq, avail_idx, &vq->avail->idx))) { | |
1936 | vq_err(vq, "Failed to access avail idx at %p\n", | |
1937 | &vq->avail->idx); | |
1938 | return -EFAULT; | |
1939 | } | |
1940 | vq->avail_idx = vhost16_to_cpu(vq, avail_idx); | |
3a4d5c94 | 1941 | |
e3b56cdd JW |
1942 | if (unlikely((u16)(vq->avail_idx - last_avail_idx) > vq->num)) { |
1943 | vq_err(vq, "Guest moved used index from %u to %u", | |
1944 | last_avail_idx, vq->avail_idx); | |
1945 | return -EFAULT; | |
1946 | } | |
1947 | ||
1948 | /* If there's nothing new since last we looked, return | |
1949 | * invalid. | |
1950 | */ | |
1951 | if (vq->avail_idx == last_avail_idx) | |
1952 | return vq->num; | |
3a4d5c94 | 1953 | |
e3b56cdd JW |
1954 | /* Only get avail ring entries after they have been |
1955 | * exposed by guest. | |
1956 | */ | |
1957 | smp_rmb(); | |
1958 | } | |
3a4d5c94 MT |
1959 | |
1960 | /* Grab the next descriptor number they're advertising, and increment | |
1961 | * the index we've seen. */ | |
bfe2bc51 JW |
1962 | if (unlikely(vhost_get_user(vq, ring_head, |
1963 | &vq->avail->ring[last_avail_idx & (vq->num - 1)]))) { | |
3a4d5c94 MT |
1964 | vq_err(vq, "Failed to read head: idx %d address %p\n", |
1965 | last_avail_idx, | |
1966 | &vq->avail->ring[last_avail_idx % vq->num]); | |
d5675bd2 | 1967 | return -EFAULT; |
3a4d5c94 MT |
1968 | } |
1969 | ||
3b1bbe89 MT |
1970 | head = vhost16_to_cpu(vq, ring_head); |
1971 | ||
3a4d5c94 | 1972 | /* If their number is silly, that's an error. */ |
7b3384fc | 1973 | if (unlikely(head >= vq->num)) { |
3a4d5c94 MT |
1974 | vq_err(vq, "Guest says index %u > %u is available", |
1975 | head, vq->num); | |
d5675bd2 | 1976 | return -EINVAL; |
3a4d5c94 MT |
1977 | } |
1978 | ||
1979 | /* When we start there are none of either input nor output. */ | |
1980 | *out_num = *in_num = 0; | |
1981 | if (unlikely(log)) | |
1982 | *log_num = 0; | |
1983 | ||
1984 | i = head; | |
1985 | do { | |
1986 | unsigned iov_count = *in_num + *out_num; | |
7b3384fc | 1987 | if (unlikely(i >= vq->num)) { |
3a4d5c94 MT |
1988 | vq_err(vq, "Desc index is %u > %u, head = %u", |
1989 | i, vq->num, head); | |
d5675bd2 | 1990 | return -EINVAL; |
3a4d5c94 | 1991 | } |
7b3384fc | 1992 | if (unlikely(++found > vq->num)) { |
3a4d5c94 MT |
1993 | vq_err(vq, "Loop detected: last one at %u " |
1994 | "vq size %u head %u\n", | |
1995 | i, vq->num, head); | |
d5675bd2 | 1996 | return -EINVAL; |
3a4d5c94 | 1997 | } |
bfe2bc51 JW |
1998 | ret = vhost_copy_from_user(vq, &desc, vq->desc + i, |
1999 | sizeof desc); | |
7b3384fc | 2000 | if (unlikely(ret)) { |
3a4d5c94 MT |
2001 | vq_err(vq, "Failed to get descriptor: idx %d addr %p\n", |
2002 | i, vq->desc + i); | |
d5675bd2 | 2003 | return -EFAULT; |
3a4d5c94 | 2004 | } |
3b1bbe89 | 2005 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT)) { |
47283bef | 2006 | ret = get_indirect(vq, iov, iov_size, |
3a4d5c94 MT |
2007 | out_num, in_num, |
2008 | log, log_num, &desc); | |
7b3384fc | 2009 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
2010 | if (ret != -EAGAIN) |
2011 | vq_err(vq, "Failure detected " | |
2012 | "in indirect descriptor at idx %d\n", i); | |
d5675bd2 | 2013 | return ret; |
3a4d5c94 MT |
2014 | } |
2015 | continue; | |
2016 | } | |
2017 | ||
6b1e6cc7 JW |
2018 | if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE)) |
2019 | access = VHOST_ACCESS_WO; | |
2020 | else | |
2021 | access = VHOST_ACCESS_RO; | |
3b1bbe89 MT |
2022 | ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr), |
2023 | vhost32_to_cpu(vq, desc.len), iov + iov_count, | |
6b1e6cc7 | 2024 | iov_size - iov_count, access); |
7b3384fc | 2025 | if (unlikely(ret < 0)) { |
6b1e6cc7 JW |
2026 | if (ret != -EAGAIN) |
2027 | vq_err(vq, "Translation failure %d descriptor idx %d\n", | |
2028 | ret, i); | |
d5675bd2 | 2029 | return ret; |
3a4d5c94 | 2030 | } |
6b1e6cc7 | 2031 | if (access == VHOST_ACCESS_WO) { |
3a4d5c94 MT |
2032 | /* If this is an input descriptor, |
2033 | * increment that count. */ | |
2034 | *in_num += ret; | |
2035 | if (unlikely(log)) { | |
3b1bbe89 MT |
2036 | log[*log_num].addr = vhost64_to_cpu(vq, desc.addr); |
2037 | log[*log_num].len = vhost32_to_cpu(vq, desc.len); | |
3a4d5c94 MT |
2038 | ++*log_num; |
2039 | } | |
2040 | } else { | |
2041 | /* If it's an output descriptor, they're all supposed | |
2042 | * to come before any input descriptors. */ | |
7b3384fc | 2043 | if (unlikely(*in_num)) { |
3a4d5c94 MT |
2044 | vq_err(vq, "Descriptor has out after in: " |
2045 | "idx %d\n", i); | |
d5675bd2 | 2046 | return -EINVAL; |
3a4d5c94 MT |
2047 | } |
2048 | *out_num += ret; | |
2049 | } | |
3b1bbe89 | 2050 | } while ((i = next_desc(vq, &desc)) != -1); |
3a4d5c94 MT |
2051 | |
2052 | /* On success, increment avail index. */ | |
2053 | vq->last_avail_idx++; | |
8ea8cf89 MT |
2054 | |
2055 | /* Assume notifications from guest are disabled at this point, | |
2056 | * if they aren't we would need to update avail_event index. */ | |
2057 | BUG_ON(!(vq->used_flags & VRING_USED_F_NO_NOTIFY)); | |
3a4d5c94 MT |
2058 | return head; |
2059 | } | |
6ac1afbf | 2060 | EXPORT_SYMBOL_GPL(vhost_get_vq_desc); |
3a4d5c94 MT |
2061 | |
2062 | /* Reverse the effect of vhost_get_vq_desc. Useful for error handling. */ | |
8dd014ad | 2063 | void vhost_discard_vq_desc(struct vhost_virtqueue *vq, int n) |
3a4d5c94 | 2064 | { |
8dd014ad | 2065 | vq->last_avail_idx -= n; |
3a4d5c94 | 2066 | } |
6ac1afbf | 2067 | EXPORT_SYMBOL_GPL(vhost_discard_vq_desc); |
3a4d5c94 MT |
2068 | |
2069 | /* After we've used one of their buffers, we tell them about it. We'll then | |
2070 | * want to notify the guest, using eventfd. */ | |
2071 | int vhost_add_used(struct vhost_virtqueue *vq, unsigned int head, int len) | |
2072 | { | |
3b1bbe89 MT |
2073 | struct vring_used_elem heads = { |
2074 | cpu_to_vhost32(vq, head), | |
2075 | cpu_to_vhost32(vq, len) | |
2076 | }; | |
3a4d5c94 | 2077 | |
c49e4e57 | 2078 | return vhost_add_used_n(vq, &heads, 1); |
3a4d5c94 | 2079 | } |
6ac1afbf | 2080 | EXPORT_SYMBOL_GPL(vhost_add_used); |
3a4d5c94 | 2081 | |
8dd014ad DS |
2082 | static int __vhost_add_used_n(struct vhost_virtqueue *vq, |
2083 | struct vring_used_elem *heads, | |
2084 | unsigned count) | |
2085 | { | |
2086 | struct vring_used_elem __user *used; | |
8ea8cf89 | 2087 | u16 old, new; |
8dd014ad DS |
2088 | int start; |
2089 | ||
5fba13b5 | 2090 | start = vq->last_used_idx & (vq->num - 1); |
8dd014ad | 2091 | used = vq->used->ring + start; |
c49e4e57 | 2092 | if (count == 1) { |
bfe2bc51 | 2093 | if (vhost_put_user(vq, heads[0].id, &used->id)) { |
c49e4e57 JW |
2094 | vq_err(vq, "Failed to write used id"); |
2095 | return -EFAULT; | |
2096 | } | |
bfe2bc51 | 2097 | if (vhost_put_user(vq, heads[0].len, &used->len)) { |
c49e4e57 JW |
2098 | vq_err(vq, "Failed to write used len"); |
2099 | return -EFAULT; | |
2100 | } | |
bfe2bc51 | 2101 | } else if (vhost_copy_to_user(vq, used, heads, count * sizeof *used)) { |
8dd014ad DS |
2102 | vq_err(vq, "Failed to write used"); |
2103 | return -EFAULT; | |
2104 | } | |
2105 | if (unlikely(vq->log_used)) { | |
2106 | /* Make sure data is seen before log. */ | |
2107 | smp_wmb(); | |
2108 | /* Log used ring entry write. */ | |
2109 | log_write(vq->log_base, | |
2110 | vq->log_addr + | |
2111 | ((void __user *)used - (void __user *)vq->used), | |
2112 | count * sizeof *used); | |
2113 | } | |
8ea8cf89 MT |
2114 | old = vq->last_used_idx; |
2115 | new = (vq->last_used_idx += count); | |
2116 | /* If the driver never bothers to signal in a very long while, | |
2117 | * used index might wrap around. If that happens, invalidate | |
2118 | * signalled_used index we stored. TODO: make sure driver | |
2119 | * signals at least once in 2^16 and remove this. */ | |
2120 | if (unlikely((u16)(new - vq->signalled_used) < (u16)(new - old))) | |
2121 | vq->signalled_used_valid = false; | |
8dd014ad DS |
2122 | return 0; |
2123 | } | |
2124 | ||
2125 | /* After we've used one of their buffers, we tell them about it. We'll then | |
2126 | * want to notify the guest, using eventfd. */ | |
2127 | int vhost_add_used_n(struct vhost_virtqueue *vq, struct vring_used_elem *heads, | |
2128 | unsigned count) | |
2129 | { | |
2130 | int start, n, r; | |
2131 | ||
5fba13b5 | 2132 | start = vq->last_used_idx & (vq->num - 1); |
8dd014ad DS |
2133 | n = vq->num - start; |
2134 | if (n < count) { | |
2135 | r = __vhost_add_used_n(vq, heads, n); | |
2136 | if (r < 0) | |
2137 | return r; | |
2138 | heads += n; | |
2139 | count -= n; | |
2140 | } | |
2141 | r = __vhost_add_used_n(vq, heads, count); | |
2142 | ||
2143 | /* Make sure buffer is written before we update index. */ | |
2144 | smp_wmb(); | |
bfe2bc51 JW |
2145 | if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->last_used_idx), |
2146 | &vq->used->idx)) { | |
8dd014ad DS |
2147 | vq_err(vq, "Failed to increment used idx"); |
2148 | return -EFAULT; | |
2149 | } | |
2150 | if (unlikely(vq->log_used)) { | |
2151 | /* Log used index update. */ | |
2152 | log_write(vq->log_base, | |
2153 | vq->log_addr + offsetof(struct vring_used, idx), | |
2154 | sizeof vq->used->idx); | |
2155 | if (vq->log_ctx) | |
2156 | eventfd_signal(vq->log_ctx, 1); | |
2157 | } | |
2158 | return r; | |
2159 | } | |
6ac1afbf | 2160 | EXPORT_SYMBOL_GPL(vhost_add_used_n); |
8dd014ad | 2161 | |
8ea8cf89 | 2162 | static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 | 2163 | { |
3b1bbe89 MT |
2164 | __u16 old, new; |
2165 | __virtio16 event; | |
8ea8cf89 | 2166 | bool v; |
0d499356 | 2167 | |
ea16c514 | 2168 | if (vhost_has_feature(vq, VIRTIO_F_NOTIFY_ON_EMPTY) && |
8ea8cf89 MT |
2169 | unlikely(vq->avail_idx == vq->last_avail_idx)) |
2170 | return true; | |
2171 | ||
ea16c514 | 2172 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
3b1bbe89 | 2173 | __virtio16 flags; |
809ecb9b JW |
2174 | /* Flush out used index updates. This is paired |
2175 | * with the barrier that the Guest executes when enabling | |
2176 | * interrupts. */ | |
2177 | smp_mb(); | |
bfe2bc51 | 2178 | if (vhost_get_user(vq, flags, &vq->avail->flags)) { |
8ea8cf89 MT |
2179 | vq_err(vq, "Failed to get flags"); |
2180 | return true; | |
2181 | } | |
3b1bbe89 | 2182 | return !(flags & cpu_to_vhost16(vq, VRING_AVAIL_F_NO_INTERRUPT)); |
3a4d5c94 | 2183 | } |
8ea8cf89 MT |
2184 | old = vq->signalled_used; |
2185 | v = vq->signalled_used_valid; | |
2186 | new = vq->signalled_used = vq->last_used_idx; | |
2187 | vq->signalled_used_valid = true; | |
3a4d5c94 | 2188 | |
8ea8cf89 MT |
2189 | if (unlikely(!v)) |
2190 | return true; | |
3a4d5c94 | 2191 | |
809ecb9b JW |
2192 | /* We're sure if the following conditions are met, there's no |
2193 | * need to notify guest: | |
2194 | * 1) cached used event is ahead of new | |
2195 | * 2) old to new updating does not cross cached used event. */ | |
2196 | if (vring_need_event(vq->last_used_event, new + vq->num, new) && | |
2197 | !vring_need_event(vq->last_used_event, new, old)) | |
2198 | return false; | |
2199 | ||
2200 | /* Flush out used index updates. This is paired | |
2201 | * with the barrier that the Guest executes when enabling | |
2202 | * interrupts. */ | |
2203 | smp_mb(); | |
2204 | ||
bfe2bc51 | 2205 | if (vhost_get_user(vq, event, vhost_used_event(vq))) { |
8ea8cf89 MT |
2206 | vq_err(vq, "Failed to get used event idx"); |
2207 | return true; | |
2208 | } | |
809ecb9b JW |
2209 | vq->last_used_event = vhost16_to_cpu(vq, event); |
2210 | ||
2211 | return vring_need_event(vq->last_used_event, new, old); | |
8ea8cf89 MT |
2212 | } |
2213 | ||
2214 | /* This actually signals the guest, using eventfd. */ | |
2215 | void vhost_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq) | |
2216 | { | |
3a4d5c94 | 2217 | /* Signal the Guest tell them we used something up. */ |
8ea8cf89 | 2218 | if (vq->call_ctx && vhost_notify(dev, vq)) |
3a4d5c94 MT |
2219 | eventfd_signal(vq->call_ctx, 1); |
2220 | } | |
6ac1afbf | 2221 | EXPORT_SYMBOL_GPL(vhost_signal); |
3a4d5c94 MT |
2222 | |
2223 | /* And here's the combo meal deal. Supersize me! */ | |
2224 | void vhost_add_used_and_signal(struct vhost_dev *dev, | |
2225 | struct vhost_virtqueue *vq, | |
2226 | unsigned int head, int len) | |
2227 | { | |
2228 | vhost_add_used(vq, head, len); | |
2229 | vhost_signal(dev, vq); | |
2230 | } | |
6ac1afbf | 2231 | EXPORT_SYMBOL_GPL(vhost_add_used_and_signal); |
3a4d5c94 | 2232 | |
8dd014ad DS |
2233 | /* multi-buffer version of vhost_add_used_and_signal */ |
2234 | void vhost_add_used_and_signal_n(struct vhost_dev *dev, | |
2235 | struct vhost_virtqueue *vq, | |
2236 | struct vring_used_elem *heads, unsigned count) | |
2237 | { | |
2238 | vhost_add_used_n(vq, heads, count); | |
2239 | vhost_signal(dev, vq); | |
2240 | } | |
6ac1afbf | 2241 | EXPORT_SYMBOL_GPL(vhost_add_used_and_signal_n); |
8dd014ad | 2242 | |
d4a60603 JW |
2243 | /* return true if we're sure that avaiable ring is empty */ |
2244 | bool vhost_vq_avail_empty(struct vhost_dev *dev, struct vhost_virtqueue *vq) | |
2245 | { | |
2246 | __virtio16 avail_idx; | |
2247 | int r; | |
2248 | ||
bfe2bc51 | 2249 | r = vhost_get_user(vq, avail_idx, &vq->avail->idx); |
d4a60603 JW |
2250 | if (r) |
2251 | return false; | |
2252 | ||
2253 | return vhost16_to_cpu(vq, avail_idx) == vq->avail_idx; | |
2254 | } | |
2255 | EXPORT_SYMBOL_GPL(vhost_vq_avail_empty); | |
2256 | ||
3a4d5c94 | 2257 | /* OK, now we need to know about added descriptors. */ |
8ea8cf89 | 2258 | bool vhost_enable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 | 2259 | { |
3b1bbe89 | 2260 | __virtio16 avail_idx; |
3a4d5c94 | 2261 | int r; |
d47effe1 | 2262 | |
3a4d5c94 MT |
2263 | if (!(vq->used_flags & VRING_USED_F_NO_NOTIFY)) |
2264 | return false; | |
2265 | vq->used_flags &= ~VRING_USED_F_NO_NOTIFY; | |
ea16c514 | 2266 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
2723feaa | 2267 | r = vhost_update_used_flags(vq); |
8ea8cf89 MT |
2268 | if (r) { |
2269 | vq_err(vq, "Failed to enable notification at %p: %d\n", | |
2270 | &vq->used->flags, r); | |
2271 | return false; | |
2272 | } | |
2273 | } else { | |
2723feaa | 2274 | r = vhost_update_avail_event(vq, vq->avail_idx); |
8ea8cf89 MT |
2275 | if (r) { |
2276 | vq_err(vq, "Failed to update avail event index at %p: %d\n", | |
2277 | vhost_avail_event(vq), r); | |
2278 | return false; | |
2279 | } | |
2280 | } | |
3a4d5c94 MT |
2281 | /* They could have slipped one in as we were doing that: make |
2282 | * sure it's written, then check again. */ | |
5659338c | 2283 | smp_mb(); |
bfe2bc51 | 2284 | r = vhost_get_user(vq, avail_idx, &vq->avail->idx); |
3a4d5c94 MT |
2285 | if (r) { |
2286 | vq_err(vq, "Failed to check avail idx at %p: %d\n", | |
2287 | &vq->avail->idx, r); | |
2288 | return false; | |
2289 | } | |
2290 | ||
3b1bbe89 | 2291 | return vhost16_to_cpu(vq, avail_idx) != vq->avail_idx; |
3a4d5c94 | 2292 | } |
6ac1afbf | 2293 | EXPORT_SYMBOL_GPL(vhost_enable_notify); |
3a4d5c94 MT |
2294 | |
2295 | /* We don't need to be notified again. */ | |
8ea8cf89 | 2296 | void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) |
3a4d5c94 MT |
2297 | { |
2298 | int r; | |
d47effe1 | 2299 | |
3a4d5c94 MT |
2300 | if (vq->used_flags & VRING_USED_F_NO_NOTIFY) |
2301 | return; | |
2302 | vq->used_flags |= VRING_USED_F_NO_NOTIFY; | |
ea16c514 | 2303 | if (!vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX)) { |
2723feaa | 2304 | r = vhost_update_used_flags(vq); |
8ea8cf89 MT |
2305 | if (r) |
2306 | vq_err(vq, "Failed to enable notification at %p: %d\n", | |
2307 | &vq->used->flags, r); | |
2308 | } | |
3a4d5c94 | 2309 | } |
6ac1afbf AH |
2310 | EXPORT_SYMBOL_GPL(vhost_disable_notify); |
2311 | ||
6b1e6cc7 JW |
2312 | /* Create a new message. */ |
2313 | struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type) | |
2314 | { | |
2315 | struct vhost_msg_node *node = kmalloc(sizeof *node, GFP_KERNEL); | |
2316 | if (!node) | |
2317 | return NULL; | |
2318 | node->vq = vq; | |
2319 | node->msg.type = type; | |
2320 | return node; | |
2321 | } | |
2322 | EXPORT_SYMBOL_GPL(vhost_new_msg); | |
2323 | ||
2324 | void vhost_enqueue_msg(struct vhost_dev *dev, struct list_head *head, | |
2325 | struct vhost_msg_node *node) | |
2326 | { | |
2327 | spin_lock(&dev->iotlb_lock); | |
2328 | list_add_tail(&node->node, head); | |
2329 | spin_unlock(&dev->iotlb_lock); | |
2330 | ||
2331 | wake_up_interruptible_poll(&dev->wait, POLLIN | POLLRDNORM); | |
2332 | } | |
2333 | EXPORT_SYMBOL_GPL(vhost_enqueue_msg); | |
2334 | ||
2335 | struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev, | |
2336 | struct list_head *head) | |
2337 | { | |
2338 | struct vhost_msg_node *node = NULL; | |
2339 | ||
2340 | spin_lock(&dev->iotlb_lock); | |
2341 | if (!list_empty(head)) { | |
2342 | node = list_first_entry(head, struct vhost_msg_node, | |
2343 | node); | |
2344 | list_del(&node->node); | |
2345 | } | |
2346 | spin_unlock(&dev->iotlb_lock); | |
2347 | ||
2348 | return node; | |
2349 | } | |
2350 | EXPORT_SYMBOL_GPL(vhost_dequeue_msg); | |
2351 | ||
2352 | ||
6ac1afbf AH |
2353 | static int __init vhost_init(void) |
2354 | { | |
2355 | return 0; | |
2356 | } | |
2357 | ||
2358 | static void __exit vhost_exit(void) | |
2359 | { | |
2360 | } | |
2361 | ||
2362 | module_init(vhost_init); | |
2363 | module_exit(vhost_exit); | |
2364 | ||
2365 | MODULE_VERSION("0.0.1"); | |
2366 | MODULE_LICENSE("GPL v2"); | |
2367 | MODULE_AUTHOR("Michael S. Tsirkin"); | |
2368 | MODULE_DESCRIPTION("Host kernel accelerator for virtio"); |