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