1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _VHOST_H
3 #define _VHOST_H
4
5 #include <linux/eventfd.h>
6 #include <linux/vhost.h>
7 #include <linux/mm.h>
8 #include <linux/mutex.h>
9 #include <linux/poll.h>
10 #include <linux/file.h>
11 #include <linux/uio.h>
12 #include <linux/virtio_config.h>
13 #include <linux/virtio_ring.h>
14 #include <linux/atomic.h>
15 #include <linux/vhost_iotlb.h>
16 #include <linux/irqbypass.h>
17 #include <linux/unroll.h>
18
19 struct vhost_work;
20 struct vhost_task;
21 typedef void (*vhost_work_fn_t)(struct vhost_work *work);
22
23 #define VHOST_WORK_QUEUED 1
24 struct vhost_work {
25 struct llist_node node;
26 vhost_work_fn_t fn;
27 unsigned long flags;
28 };
29
30 struct vhost_worker;
31 struct vhost_dev;
32 struct vhost_msg_node;
33
34 struct vhost_worker_ops {
35 int (*create)(struct vhost_worker *worker, struct vhost_dev *dev,
36 const char *name);
37 void (*stop)(struct vhost_worker *worker);
38 void (*wakeup)(struct vhost_worker *worker);
39 };
40
41 struct vhost_worker {
42 struct task_struct *kthread_task;
43 struct vhost_task *vtsk;
44 struct vhost_dev *dev;
45 /* Used to serialize device wide flushing with worker swapping. */
46 struct mutex mutex;
47 struct llist_head work_list;
48 struct kcov_common_handle_id kcov_handle;
49 u32 id;
50 int attachment_cnt;
51 bool killed;
52 const struct vhost_worker_ops *ops;
53 };
54
55 /* Poll a file (eventfd or socket) */
56 /* Note: there's nothing vhost specific about this structure. */
57 struct vhost_poll {
58 poll_table table;
59 wait_queue_head_t *wqh;
60 wait_queue_entry_t wait;
61 struct vhost_work work;
62 __poll_t mask;
63 struct vhost_dev *dev;
64 struct vhost_virtqueue *vq;
65 };
66
67 void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn,
68 __poll_t mask, struct vhost_dev *dev,
69 struct vhost_virtqueue *vq);
70 int vhost_poll_start(struct vhost_poll *poll, struct file *file);
71 void vhost_poll_stop(struct vhost_poll *poll);
72 void vhost_poll_queue(struct vhost_poll *poll);
73
74 void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn);
75 void vhost_dev_flush(struct vhost_dev *dev);
76
77 struct vhost_log {
78 u64 addr;
79 u64 len;
80 };
81
82 enum vhost_uaddr_type {
83 VHOST_ADDR_DESC = 0,
84 VHOST_ADDR_AVAIL = 1,
85 VHOST_ADDR_USED = 2,
86 VHOST_NUM_ADDRS = 3,
87 };
88
89 struct vhost_vring_call {
90 struct eventfd_ctx *ctx;
91 struct irq_bypass_producer producer;
92 };
93
94 /* The virtqueue structure describes a queue attached to a device. */
95 struct vhost_virtqueue {
96 struct vhost_dev *dev;
97 struct vhost_worker __rcu *worker;
98
99 /* The actual ring of buffers. */
100 struct mutex mutex;
101 unsigned int num;
102 vring_desc_t __user *desc;
103 vring_avail_t __user *avail;
104 vring_used_t __user *used;
105 const struct vhost_iotlb_map *meta_iotlb[VHOST_NUM_ADDRS];
106 struct file *kick;
107 struct vhost_vring_call call_ctx;
108 struct eventfd_ctx *error_ctx;
109 struct eventfd_ctx *log_ctx;
110
111 struct vhost_poll poll;
112
113 /* The routine to call when the Guest pings us, or timeout. */
114 vhost_work_fn_t handle_kick;
115
116 /* Last available index we saw.
117 * Values are limited to 0x7fff, and the high bit is used as
118 * a wrap counter when using VIRTIO_F_RING_PACKED. */
119 u16 last_avail_idx;
120 /* Next avail ring head when VIRTIO_F_IN_ORDER is negoitated */
121 u16 next_avail_head;
122
123 /* Caches available index value from user. */
124 u16 avail_idx;
125
126 /* Last index we used.
127 * Values are limited to 0x7fff, and the high bit is used as
128 * a wrap counter when using VIRTIO_F_RING_PACKED. */
129 u16 last_used_idx;
130
131 /* Used flags */
132 u16 used_flags;
133
134 /* Last used index value we have signalled on */
135 u16 signalled_used;
136
137 /* Last used index value we have signalled on */
138 bool signalled_used_valid;
139
140 /* Log writes to used structure. */
141 bool log_used;
142 u64 log_addr;
143
144 struct iovec iov[UIO_MAXIOV];
145 struct iovec iotlb_iov[64];
146 struct iovec *indirect;
147 struct vring_used_elem *heads;
148 u16 *nheads;
149 /* Protected by virtqueue mutex. */
150 struct vhost_iotlb *umem;
151 struct vhost_iotlb *iotlb;
152 /* Protected by dev->iotlb_lock. */
153 struct vhost_msg_node *iotlb_miss;
154 void *private_data;
155 VIRTIO_DECLARE_FEATURES(acked_features);
156 u64 acked_backend_features;
157 /* Log write descriptors */
158 void __user *log_base;
159 struct vhost_log *log;
160 struct iovec log_iov[64];
161
162 /* Ring endianness. Defaults to legacy native endianness.
163 * Set to true when starting a modern virtio device. */
164 bool is_le;
165 #ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
166 /* Ring endianness requested by userspace for cross-endian support. */
167 bool user_be;
168 #endif
169 u32 busyloop_timeout;
170 };
171
172 struct vhost_msg_node {
173 union {
174 struct vhost_msg msg;
175 struct vhost_msg_v2 msg_v2;
176 };
177 struct vhost_virtqueue *vq;
178 struct list_head node;
179 };
180
181 struct vhost_dev {
182 struct mm_struct *mm;
183 struct mutex mutex;
184 struct vhost_virtqueue **vqs;
185 int nvqs;
186 struct eventfd_ctx *log_ctx;
187 struct vhost_iotlb *umem;
188 struct vhost_iotlb *iotlb;
189 spinlock_t iotlb_lock;
190 struct list_head read_list;
191 struct list_head pending_list;
192 wait_queue_head_t wait;
193 int iov_limit;
194 int weight;
195 int byte_weight;
196 struct xarray worker_xa;
197 bool use_worker;
198 /*
199 * If fork_owner is true we use vhost_tasks to create
200 * the worker so all settings/limits like cgroups, NPROC,
201 * scheduler, etc are inherited from the owner. If false,
202 * we use kthreads and only attach to the same cgroups
203 * as the owner for compat with older kernels.
204 * here we use true as default value.
205 * The default value is set by fork_from_owner_default
206 */
207 bool fork_owner;
208 int (*msg_handler)(struct vhost_dev *dev, u32 asid,
209 struct vhost_iotlb_msg *msg);
210 };
211
212 bool vhost_exceeds_weight(struct vhost_virtqueue *vq, int pkts, int total_len);
213 void vhost_dev_init(struct vhost_dev *, struct vhost_virtqueue **vqs,
214 int nvqs, int iov_limit, int weight, int byte_weight,
215 bool use_worker,
216 int (*msg_handler)(struct vhost_dev *dev, u32 asid,
217 struct vhost_iotlb_msg *msg));
218 long vhost_dev_set_owner(struct vhost_dev *dev);
219 bool vhost_dev_has_owner(struct vhost_dev *dev);
220 long vhost_dev_check_owner(struct vhost_dev *);
221 struct vhost_iotlb *vhost_dev_reset_owner_prepare(void);
222 void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_iotlb *iotlb);
223 void vhost_dev_cleanup(struct vhost_dev *);
224 void vhost_dev_stop(struct vhost_dev *);
225 long vhost_dev_ioctl(struct vhost_dev *, unsigned int ioctl, void __user *argp);
226 long vhost_vring_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp);
227 long vhost_worker_ioctl(struct vhost_dev *dev, unsigned int ioctl,
228 void __user *argp);
229 bool vhost_vq_access_ok(struct vhost_virtqueue *vq);
230 bool vhost_log_access_ok(struct vhost_dev *);
231 void vhost_clear_msg(struct vhost_dev *dev);
232
233 int vhost_get_vq_desc(struct vhost_virtqueue *,
234 struct iovec iov[], unsigned int iov_size,
235 unsigned int *out_num, unsigned int *in_num,
236 struct vhost_log *log, unsigned int *log_num);
237
238 int vhost_get_vq_desc_n(struct vhost_virtqueue *vq,
239 struct iovec iov[], unsigned int iov_size,
240 unsigned int *out_num, unsigned int *in_num,
241 struct vhost_log *log, unsigned int *log_num,
242 unsigned int *ndesc);
243
244 void vhost_discard_vq_desc(struct vhost_virtqueue *, int nbuf,
245 unsigned int ndesc);
246
247 bool vhost_vq_work_queue(struct vhost_virtqueue *vq, struct vhost_work *work);
248 bool vhost_vq_has_work(struct vhost_virtqueue *vq);
249 bool vhost_vq_is_setup(struct vhost_virtqueue *vq);
250 int vhost_vq_init_access(struct vhost_virtqueue *);
251 int vhost_add_used(struct vhost_virtqueue *, unsigned int head, int len);
252 int vhost_add_used_n(struct vhost_virtqueue *, struct vring_used_elem *heads,
253 u16 *nheads, unsigned count);
254 void vhost_add_used_and_signal(struct vhost_dev *, struct vhost_virtqueue *,
255 unsigned int id, int len);
256 void vhost_add_used_and_signal_n(struct vhost_dev *, struct vhost_virtqueue *,
257 struct vring_used_elem *heads, u16 *nheads,
258 unsigned count);
259 void vhost_signal(struct vhost_dev *, struct vhost_virtqueue *);
260 void vhost_disable_notify(struct vhost_dev *, struct vhost_virtqueue *);
261 bool vhost_vq_avail_empty(struct vhost_dev *, struct vhost_virtqueue *);
262 bool vhost_enable_notify(struct vhost_dev *, struct vhost_virtqueue *);
263
264 int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
265 unsigned int log_num, u64 len,
266 struct iovec *iov, int count);
267 int vq_meta_prefetch(struct vhost_virtqueue *vq);
268
269 struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type);
270 void vhost_enqueue_msg(struct vhost_dev *dev,
271 struct list_head *head,
272 struct vhost_msg_node *node);
273 struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev,
274 struct list_head *head);
275 void vhost_set_backend_features(struct vhost_dev *dev, u64 features);
276
277 __poll_t vhost_chr_poll(struct file *file, struct vhost_dev *dev,
278 poll_table *wait);
279 ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to,
280 int noblock);
281 ssize_t vhost_chr_write_iter(struct vhost_dev *dev,
282 struct iov_iter *from);
283 void vhost_clear_device_iotlb(struct vhost_dev *d);
284 int vhost_init_device_iotlb(struct vhost_dev *d);
285
286 void vhost_iotlb_map_free(struct vhost_iotlb *iotlb,
287 struct vhost_iotlb_map *map);
288
289 #define vq_err(vq, fmt, ...) do { \
290 pr_debug(pr_fmt(fmt), ##__VA_ARGS__); \
291 if ((vq)->error_ctx) \
292 eventfd_signal((vq)->error_ctx);\
293 } while (0)
294
295 #define VHOST_FEATURES \
296 VIRTIO_F_NOTIFY_ON_EMPTY, \
297 VIRTIO_RING_F_INDIRECT_DESC, \
298 VIRTIO_RING_F_EVENT_IDX, \
299 VHOST_F_LOG_ALL, \
300 VIRTIO_F_ANY_LAYOUT, \
301 VIRTIO_F_VERSION_1
302
vhost_features_u64(const int * features,int size,int idx)303 static inline u64 vhost_features_u64(const int *features, int size, int idx)
304 {
305 u64 res = 0;
306
307 unrolled_count(VIRTIO_FEATURES_BITS)
308 for (int i = 0; i < size; ++i) {
309 int bit = features[i];
310
311 if (virtio_features_chk_bit(bit) && VIRTIO_U64(bit) == idx)
312 res |= VIRTIO_BIT(bit);
313 }
314 return res;
315 }
316
317 #define VHOST_FEATURES_U64(features, idx) \
318 vhost_features_u64(features, ARRAY_SIZE(features), idx)
319
320 #define DEFINE_VHOST_FEATURES_ARRAY_ENTRY(idx, features) \
321 [idx] = VHOST_FEATURES_U64(features, idx),
322
323 #define DEFINE_VHOST_FEATURES_ARRAY(array, features) \
324 u64 array[VIRTIO_FEATURES_U64S] = { \
325 UNROLL(VIRTIO_FEATURES_U64S, \
326 DEFINE_VHOST_FEATURES_ARRAY_ENTRY, features) \
327 }
328
329 /**
330 * vhost_vq_set_backend - Set backend.
331 *
332 * @vq Virtqueue.
333 * @private_data The private data.
334 *
335 * Context: Need to call with vq->mutex acquired.
336 */
vhost_vq_set_backend(struct vhost_virtqueue * vq,void * private_data)337 static inline void vhost_vq_set_backend(struct vhost_virtqueue *vq,
338 void *private_data)
339 {
340 vq->private_data = private_data;
341 }
342
343 /**
344 * vhost_vq_get_backend - Get backend.
345 *
346 * @vq Virtqueue.
347 *
348 * Context: Need to call with vq->mutex acquired.
349 * Return: Private data previously set with vhost_vq_set_backend.
350 */
vhost_vq_get_backend(struct vhost_virtqueue * vq)351 static inline void *vhost_vq_get_backend(struct vhost_virtqueue *vq)
352 {
353 return vq->private_data;
354 }
355
vhost_has_feature(struct vhost_virtqueue * vq,int bit)356 static inline bool vhost_has_feature(struct vhost_virtqueue *vq, int bit)
357 {
358 return virtio_features_test_bit(vq->acked_features_array, bit);
359 }
360
vhost_backend_has_feature(struct vhost_virtqueue * vq,int bit)361 static inline bool vhost_backend_has_feature(struct vhost_virtqueue *vq, int bit)
362 {
363 return vq->acked_backend_features & (1ULL << bit);
364 }
365
366 #ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
vhost_is_little_endian(struct vhost_virtqueue * vq)367 static inline bool vhost_is_little_endian(struct vhost_virtqueue *vq)
368 {
369 return vq->is_le;
370 }
371 #else
vhost_is_little_endian(struct vhost_virtqueue * vq)372 static inline bool vhost_is_little_endian(struct vhost_virtqueue *vq)
373 {
374 return virtio_legacy_is_little_endian() || vq->is_le;
375 }
376 #endif
377
378 /* Memory accessors */
vhost16_to_cpu(struct vhost_virtqueue * vq,__virtio16 val)379 static inline u16 vhost16_to_cpu(struct vhost_virtqueue *vq, __virtio16 val)
380 {
381 return __virtio16_to_cpu(vhost_is_little_endian(vq), val);
382 }
383
cpu_to_vhost16(struct vhost_virtqueue * vq,u16 val)384 static inline __virtio16 cpu_to_vhost16(struct vhost_virtqueue *vq, u16 val)
385 {
386 return __cpu_to_virtio16(vhost_is_little_endian(vq), val);
387 }
388
vhost32_to_cpu(struct vhost_virtqueue * vq,__virtio32 val)389 static inline u32 vhost32_to_cpu(struct vhost_virtqueue *vq, __virtio32 val)
390 {
391 return __virtio32_to_cpu(vhost_is_little_endian(vq), val);
392 }
393
cpu_to_vhost32(struct vhost_virtqueue * vq,u32 val)394 static inline __virtio32 cpu_to_vhost32(struct vhost_virtqueue *vq, u32 val)
395 {
396 return __cpu_to_virtio32(vhost_is_little_endian(vq), val);
397 }
398
vhost64_to_cpu(struct vhost_virtqueue * vq,__virtio64 val)399 static inline u64 vhost64_to_cpu(struct vhost_virtqueue *vq, __virtio64 val)
400 {
401 return __virtio64_to_cpu(vhost_is_little_endian(vq), val);
402 }
403
cpu_to_vhost64(struct vhost_virtqueue * vq,u64 val)404 static inline __virtio64 cpu_to_vhost64(struct vhost_virtqueue *vq, u64 val)
405 {
406 return __cpu_to_virtio64(vhost_is_little_endian(vq), val);
407 }
408 #endif
409