Loading...
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
18struct vhost_work;
19typedef void (*vhost_work_fn_t)(struct vhost_work *work);
20
21#define VHOST_WORK_QUEUED 1
22struct vhost_work {
23 struct llist_node node;
24 vhost_work_fn_t fn;
25 unsigned long flags;
26};
27
28/* Poll a file (eventfd or socket) */
29/* Note: there's nothing vhost specific about this structure. */
30struct vhost_poll {
31 poll_table table;
32 wait_queue_head_t *wqh;
33 wait_queue_entry_t wait;
34 struct vhost_work work;
35 __poll_t mask;
36 struct vhost_dev *dev;
37};
38
39void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn);
40void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work);
41bool vhost_has_work(struct vhost_dev *dev);
42
43void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn,
44 __poll_t mask, struct vhost_dev *dev);
45int vhost_poll_start(struct vhost_poll *poll, struct file *file);
46void vhost_poll_stop(struct vhost_poll *poll);
47void vhost_poll_flush(struct vhost_poll *poll);
48void vhost_poll_queue(struct vhost_poll *poll);
49void vhost_work_flush(struct vhost_dev *dev, struct vhost_work *work);
50long vhost_vring_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp);
51
52struct vhost_log {
53 u64 addr;
54 u64 len;
55};
56
57enum vhost_uaddr_type {
58 VHOST_ADDR_DESC = 0,
59 VHOST_ADDR_AVAIL = 1,
60 VHOST_ADDR_USED = 2,
61 VHOST_NUM_ADDRS = 3,
62};
63
64struct vhost_vring_call {
65 struct eventfd_ctx *ctx;
66 struct irq_bypass_producer producer;
67 spinlock_t ctx_lock;
68};
69
70/* The virtqueue structure describes a queue attached to a device. */
71struct vhost_virtqueue {
72 struct vhost_dev *dev;
73
74 /* The actual ring of buffers. */
75 struct mutex mutex;
76 unsigned int num;
77 vring_desc_t __user *desc;
78 vring_avail_t __user *avail;
79 vring_used_t __user *used;
80 const struct vhost_iotlb_map *meta_iotlb[VHOST_NUM_ADDRS];
81 struct file *kick;
82 struct vhost_vring_call call_ctx;
83 struct eventfd_ctx *error_ctx;
84 struct eventfd_ctx *log_ctx;
85
86 struct vhost_poll poll;
87
88 /* The routine to call when the Guest pings us, or timeout. */
89 vhost_work_fn_t handle_kick;
90
91 /* Last available index we saw. */
92 u16 last_avail_idx;
93
94 /* Caches available index value from user. */
95 u16 avail_idx;
96
97 /* Last index we used. */
98 u16 last_used_idx;
99
100 /* Used flags */
101 u16 used_flags;
102
103 /* Last used index value we have signalled on */
104 u16 signalled_used;
105
106 /* Last used index value we have signalled on */
107 bool signalled_used_valid;
108
109 /* Log writes to used structure. */
110 bool log_used;
111 u64 log_addr;
112
113 struct iovec iov[UIO_MAXIOV];
114 struct iovec iotlb_iov[64];
115 struct iovec *indirect;
116 struct vring_used_elem *heads;
117 /* Protected by virtqueue mutex. */
118 struct vhost_iotlb *umem;
119 struct vhost_iotlb *iotlb;
120 void *private_data;
121 u64 acked_features;
122 u64 acked_backend_features;
123 /* Log write descriptors */
124 void __user *log_base;
125 struct vhost_log *log;
126
127 /* Ring endianness. Defaults to legacy native endianness.
128 * Set to true when starting a modern virtio device. */
129 bool is_le;
130#ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
131 /* Ring endianness requested by userspace for cross-endian support. */
132 bool user_be;
133#endif
134 u32 busyloop_timeout;
135};
136
137struct vhost_msg_node {
138 union {
139 struct vhost_msg msg;
140 struct vhost_msg_v2 msg_v2;
141 };
142 struct vhost_virtqueue *vq;
143 struct list_head node;
144};
145
146struct vhost_dev {
147 struct mm_struct *mm;
148 struct mutex mutex;
149 struct vhost_virtqueue **vqs;
150 int nvqs;
151 struct eventfd_ctx *log_ctx;
152 struct llist_head work_list;
153 struct task_struct *worker;
154 struct vhost_iotlb *umem;
155 struct vhost_iotlb *iotlb;
156 spinlock_t iotlb_lock;
157 struct list_head read_list;
158 struct list_head pending_list;
159 wait_queue_head_t wait;
160 int iov_limit;
161 int weight;
162 int byte_weight;
163 u64 kcov_handle;
164 bool use_worker;
165 int (*msg_handler)(struct vhost_dev *dev,
166 struct vhost_iotlb_msg *msg);
167};
168
169bool vhost_exceeds_weight(struct vhost_virtqueue *vq, int pkts, int total_len);
170void vhost_dev_init(struct vhost_dev *, struct vhost_virtqueue **vqs,
171 int nvqs, int iov_limit, int weight, int byte_weight,
172 bool use_worker,
173 int (*msg_handler)(struct vhost_dev *dev,
174 struct vhost_iotlb_msg *msg));
175long vhost_dev_set_owner(struct vhost_dev *dev);
176bool vhost_dev_has_owner(struct vhost_dev *dev);
177long vhost_dev_check_owner(struct vhost_dev *);
178struct vhost_iotlb *vhost_dev_reset_owner_prepare(void);
179void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_iotlb *iotlb);
180void vhost_dev_cleanup(struct vhost_dev *);
181void vhost_dev_stop(struct vhost_dev *);
182long vhost_dev_ioctl(struct vhost_dev *, unsigned int ioctl, void __user *argp);
183long vhost_vring_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp);
184bool vhost_vq_access_ok(struct vhost_virtqueue *vq);
185bool vhost_log_access_ok(struct vhost_dev *);
186
187int vhost_get_vq_desc(struct vhost_virtqueue *,
188 struct iovec iov[], unsigned int iov_count,
189 unsigned int *out_num, unsigned int *in_num,
190 struct vhost_log *log, unsigned int *log_num);
191void vhost_discard_vq_desc(struct vhost_virtqueue *, int n);
192
193int vhost_vq_init_access(struct vhost_virtqueue *);
194int vhost_add_used(struct vhost_virtqueue *, unsigned int head, int len);
195int vhost_add_used_n(struct vhost_virtqueue *, struct vring_used_elem *heads,
196 unsigned count);
197void vhost_add_used_and_signal(struct vhost_dev *, struct vhost_virtqueue *,
198 unsigned int id, int len);
199void vhost_add_used_and_signal_n(struct vhost_dev *, struct vhost_virtqueue *,
200 struct vring_used_elem *heads, unsigned count);
201void vhost_signal(struct vhost_dev *, struct vhost_virtqueue *);
202void vhost_disable_notify(struct vhost_dev *, struct vhost_virtqueue *);
203bool vhost_vq_avail_empty(struct vhost_dev *, struct vhost_virtqueue *);
204bool vhost_enable_notify(struct vhost_dev *, struct vhost_virtqueue *);
205
206int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
207 unsigned int log_num, u64 len,
208 struct iovec *iov, int count);
209int vq_meta_prefetch(struct vhost_virtqueue *vq);
210
211struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type);
212void vhost_enqueue_msg(struct vhost_dev *dev,
213 struct list_head *head,
214 struct vhost_msg_node *node);
215struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev,
216 struct list_head *head);
217void vhost_set_backend_features(struct vhost_dev *dev, u64 features);
218
219__poll_t vhost_chr_poll(struct file *file, struct vhost_dev *dev,
220 poll_table *wait);
221ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to,
222 int noblock);
223ssize_t vhost_chr_write_iter(struct vhost_dev *dev,
224 struct iov_iter *from);
225int vhost_init_device_iotlb(struct vhost_dev *d, bool enabled);
226
227void vhost_iotlb_map_free(struct vhost_iotlb *iotlb,
228 struct vhost_iotlb_map *map);
229
230#define vq_err(vq, fmt, ...) do { \
231 pr_debug(pr_fmt(fmt), ##__VA_ARGS__); \
232 if ((vq)->error_ctx) \
233 eventfd_signal((vq)->error_ctx, 1);\
234 } while (0)
235
236enum {
237 VHOST_FEATURES = (1ULL << VIRTIO_F_NOTIFY_ON_EMPTY) |
238 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) |
239 (1ULL << VIRTIO_RING_F_EVENT_IDX) |
240 (1ULL << VHOST_F_LOG_ALL) |
241 (1ULL << VIRTIO_F_ANY_LAYOUT) |
242 (1ULL << VIRTIO_F_VERSION_1)
243};
244
245/**
246 * vhost_vq_set_backend - Set backend.
247 *
248 * @vq Virtqueue.
249 * @private_data The private data.
250 *
251 * Context: Need to call with vq->mutex acquired.
252 */
253static inline void vhost_vq_set_backend(struct vhost_virtqueue *vq,
254 void *private_data)
255{
256 vq->private_data = private_data;
257}
258
259/**
260 * vhost_vq_get_backend - Get backend.
261 *
262 * @vq Virtqueue.
263 *
264 * Context: Need to call with vq->mutex acquired.
265 * Return: Private data previously set with vhost_vq_set_backend.
266 */
267static inline void *vhost_vq_get_backend(struct vhost_virtqueue *vq)
268{
269 return vq->private_data;
270}
271
272static inline bool vhost_has_feature(struct vhost_virtqueue *vq, int bit)
273{
274 return vq->acked_features & (1ULL << bit);
275}
276
277static inline bool vhost_backend_has_feature(struct vhost_virtqueue *vq, int bit)
278{
279 return vq->acked_backend_features & (1ULL << bit);
280}
281
282#ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
283static inline bool vhost_is_little_endian(struct vhost_virtqueue *vq)
284{
285 return vq->is_le;
286}
287#else
288static inline bool vhost_is_little_endian(struct vhost_virtqueue *vq)
289{
290 return virtio_legacy_is_little_endian() || vq->is_le;
291}
292#endif
293
294/* Memory accessors */
295static inline u16 vhost16_to_cpu(struct vhost_virtqueue *vq, __virtio16 val)
296{
297 return __virtio16_to_cpu(vhost_is_little_endian(vq), val);
298}
299
300static inline __virtio16 cpu_to_vhost16(struct vhost_virtqueue *vq, u16 val)
301{
302 return __cpu_to_virtio16(vhost_is_little_endian(vq), val);
303}
304
305static inline u32 vhost32_to_cpu(struct vhost_virtqueue *vq, __virtio32 val)
306{
307 return __virtio32_to_cpu(vhost_is_little_endian(vq), val);
308}
309
310static inline __virtio32 cpu_to_vhost32(struct vhost_virtqueue *vq, u32 val)
311{
312 return __cpu_to_virtio32(vhost_is_little_endian(vq), val);
313}
314
315static inline u64 vhost64_to_cpu(struct vhost_virtqueue *vq, __virtio64 val)
316{
317 return __virtio64_to_cpu(vhost_is_little_endian(vq), val);
318}
319
320static inline __virtio64 cpu_to_vhost64(struct vhost_virtqueue *vq, u64 val)
321{
322 return __cpu_to_virtio64(vhost_is_little_endian(vq), val);
323}
324#endif
1#ifndef _VHOST_H
2#define _VHOST_H
3
4#include <linux/eventfd.h>
5#include <linux/vhost.h>
6#include <linux/mm.h>
7#include <linux/mutex.h>
8#include <linux/poll.h>
9#include <linux/file.h>
10#include <linux/uio.h>
11#include <linux/virtio_config.h>
12#include <linux/virtio_ring.h>
13#include <linux/atomic.h>
14
15struct vhost_work;
16typedef void (*vhost_work_fn_t)(struct vhost_work *work);
17
18#define VHOST_WORK_QUEUED 1
19struct vhost_work {
20 struct llist_node node;
21 vhost_work_fn_t fn;
22 wait_queue_head_t done;
23 int flushing;
24 unsigned queue_seq;
25 unsigned done_seq;
26 unsigned long flags;
27};
28
29/* Poll a file (eventfd or socket) */
30/* Note: there's nothing vhost specific about this structure. */
31struct vhost_poll {
32 poll_table table;
33 wait_queue_head_t *wqh;
34 wait_queue_t wait;
35 struct vhost_work work;
36 unsigned long mask;
37 struct vhost_dev *dev;
38};
39
40void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn);
41void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work);
42bool vhost_has_work(struct vhost_dev *dev);
43
44void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn,
45 unsigned long mask, struct vhost_dev *dev);
46int vhost_poll_start(struct vhost_poll *poll, struct file *file);
47void vhost_poll_stop(struct vhost_poll *poll);
48void vhost_poll_flush(struct vhost_poll *poll);
49void vhost_poll_queue(struct vhost_poll *poll);
50void vhost_work_flush(struct vhost_dev *dev, struct vhost_work *work);
51long vhost_vring_ioctl(struct vhost_dev *d, int ioctl, void __user *argp);
52
53struct vhost_log {
54 u64 addr;
55 u64 len;
56};
57
58#define START(node) ((node)->start)
59#define LAST(node) ((node)->last)
60
61struct vhost_umem_node {
62 struct rb_node rb;
63 struct list_head link;
64 __u64 start;
65 __u64 last;
66 __u64 size;
67 __u64 userspace_addr;
68 __u32 perm;
69 __u32 flags_padding;
70 __u64 __subtree_last;
71};
72
73struct vhost_umem {
74 struct rb_root umem_tree;
75 struct list_head umem_list;
76 int numem;
77};
78
79/* The virtqueue structure describes a queue attached to a device. */
80struct vhost_virtqueue {
81 struct vhost_dev *dev;
82
83 /* The actual ring of buffers. */
84 struct mutex mutex;
85 unsigned int num;
86 struct vring_desc __user *desc;
87 struct vring_avail __user *avail;
88 struct vring_used __user *used;
89 struct file *kick;
90 struct file *call;
91 struct file *error;
92 struct eventfd_ctx *call_ctx;
93 struct eventfd_ctx *error_ctx;
94 struct eventfd_ctx *log_ctx;
95
96 struct vhost_poll poll;
97
98 /* The routine to call when the Guest pings us, or timeout. */
99 vhost_work_fn_t handle_kick;
100
101 /* Last available index we saw. */
102 u16 last_avail_idx;
103
104 /* Caches available index value from user. */
105 u16 avail_idx;
106
107 /* Last index we used. */
108 u16 last_used_idx;
109
110 /* Last used evet we've seen */
111 u16 last_used_event;
112
113 /* Used flags */
114 u16 used_flags;
115
116 /* Last used index value we have signalled on */
117 u16 signalled_used;
118
119 /* Last used index value we have signalled on */
120 bool signalled_used_valid;
121
122 /* Log writes to used structure. */
123 bool log_used;
124 u64 log_addr;
125
126 struct iovec iov[UIO_MAXIOV];
127 struct iovec iotlb_iov[64];
128 struct iovec *indirect;
129 struct vring_used_elem *heads;
130 /* Protected by virtqueue mutex. */
131 struct vhost_umem *umem;
132 struct vhost_umem *iotlb;
133 void *private_data;
134 u64 acked_features;
135 /* Log write descriptors */
136 void __user *log_base;
137 struct vhost_log *log;
138
139 /* Ring endianness. Defaults to legacy native endianness.
140 * Set to true when starting a modern virtio device. */
141 bool is_le;
142#ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
143 /* Ring endianness requested by userspace for cross-endian support. */
144 bool user_be;
145#endif
146 u32 busyloop_timeout;
147};
148
149struct vhost_msg_node {
150 struct vhost_msg msg;
151 struct vhost_virtqueue *vq;
152 struct list_head node;
153};
154
155struct vhost_dev {
156 struct mm_struct *mm;
157 struct mutex mutex;
158 struct vhost_virtqueue **vqs;
159 int nvqs;
160 struct file *log_file;
161 struct eventfd_ctx *log_ctx;
162 struct llist_head work_list;
163 struct task_struct *worker;
164 struct vhost_umem *umem;
165 struct vhost_umem *iotlb;
166 spinlock_t iotlb_lock;
167 struct list_head read_list;
168 struct list_head pending_list;
169 wait_queue_head_t wait;
170};
171
172void vhost_dev_init(struct vhost_dev *, struct vhost_virtqueue **vqs, int nvqs);
173long vhost_dev_set_owner(struct vhost_dev *dev);
174bool vhost_dev_has_owner(struct vhost_dev *dev);
175long vhost_dev_check_owner(struct vhost_dev *);
176struct vhost_umem *vhost_dev_reset_owner_prepare(void);
177void vhost_dev_reset_owner(struct vhost_dev *, struct vhost_umem *);
178void vhost_dev_cleanup(struct vhost_dev *, bool locked);
179void vhost_dev_stop(struct vhost_dev *);
180long vhost_dev_ioctl(struct vhost_dev *, unsigned int ioctl, void __user *argp);
181long vhost_vring_ioctl(struct vhost_dev *d, int ioctl, void __user *argp);
182int vhost_vq_access_ok(struct vhost_virtqueue *vq);
183int vhost_log_access_ok(struct vhost_dev *);
184
185int vhost_get_vq_desc(struct vhost_virtqueue *,
186 struct iovec iov[], unsigned int iov_count,
187 unsigned int *out_num, unsigned int *in_num,
188 struct vhost_log *log, unsigned int *log_num);
189void vhost_discard_vq_desc(struct vhost_virtqueue *, int n);
190
191int vhost_vq_init_access(struct vhost_virtqueue *);
192int vhost_add_used(struct vhost_virtqueue *, unsigned int head, int len);
193int vhost_add_used_n(struct vhost_virtqueue *, struct vring_used_elem *heads,
194 unsigned count);
195void vhost_add_used_and_signal(struct vhost_dev *, struct vhost_virtqueue *,
196 unsigned int id, int len);
197void vhost_add_used_and_signal_n(struct vhost_dev *, struct vhost_virtqueue *,
198 struct vring_used_elem *heads, unsigned count);
199void vhost_signal(struct vhost_dev *, struct vhost_virtqueue *);
200void vhost_disable_notify(struct vhost_dev *, struct vhost_virtqueue *);
201bool vhost_vq_avail_empty(struct vhost_dev *, struct vhost_virtqueue *);
202bool vhost_enable_notify(struct vhost_dev *, struct vhost_virtqueue *);
203
204int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
205 unsigned int log_num, u64 len);
206int vq_iotlb_prefetch(struct vhost_virtqueue *vq);
207
208struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type);
209void vhost_enqueue_msg(struct vhost_dev *dev,
210 struct list_head *head,
211 struct vhost_msg_node *node);
212struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev,
213 struct list_head *head);
214unsigned int vhost_chr_poll(struct file *file, struct vhost_dev *dev,
215 poll_table *wait);
216ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to,
217 int noblock);
218ssize_t vhost_chr_write_iter(struct vhost_dev *dev,
219 struct iov_iter *from);
220int vhost_init_device_iotlb(struct vhost_dev *d, bool enabled);
221
222#define vq_err(vq, fmt, ...) do { \
223 pr_debug(pr_fmt(fmt), ##__VA_ARGS__); \
224 if ((vq)->error_ctx) \
225 eventfd_signal((vq)->error_ctx, 1);\
226 } while (0)
227
228enum {
229 VHOST_FEATURES = (1ULL << VIRTIO_F_NOTIFY_ON_EMPTY) |
230 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) |
231 (1ULL << VIRTIO_RING_F_EVENT_IDX) |
232 (1ULL << VHOST_F_LOG_ALL) |
233 (1ULL << VIRTIO_F_ANY_LAYOUT) |
234 (1ULL << VIRTIO_F_VERSION_1)
235};
236
237static inline bool vhost_has_feature(struct vhost_virtqueue *vq, int bit)
238{
239 return vq->acked_features & (1ULL << bit);
240}
241
242#ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
243static inline bool vhost_is_little_endian(struct vhost_virtqueue *vq)
244{
245 return vq->is_le;
246}
247#else
248static inline bool vhost_is_little_endian(struct vhost_virtqueue *vq)
249{
250 return virtio_legacy_is_little_endian() || vq->is_le;
251}
252#endif
253
254/* Memory accessors */
255static inline u16 vhost16_to_cpu(struct vhost_virtqueue *vq, __virtio16 val)
256{
257 return __virtio16_to_cpu(vhost_is_little_endian(vq), val);
258}
259
260static inline __virtio16 cpu_to_vhost16(struct vhost_virtqueue *vq, u16 val)
261{
262 return __cpu_to_virtio16(vhost_is_little_endian(vq), val);
263}
264
265static inline u32 vhost32_to_cpu(struct vhost_virtqueue *vq, __virtio32 val)
266{
267 return __virtio32_to_cpu(vhost_is_little_endian(vq), val);
268}
269
270static inline __virtio32 cpu_to_vhost32(struct vhost_virtqueue *vq, u32 val)
271{
272 return __cpu_to_virtio32(vhost_is_little_endian(vq), val);
273}
274
275static inline u64 vhost64_to_cpu(struct vhost_virtqueue *vq, __virtio64 val)
276{
277 return __virtio64_to_cpu(vhost_is_little_endian(vq), val);
278}
279
280static inline __virtio64 cpu_to_vhost64(struct vhost_virtqueue *vq, u64 val)
281{
282 return __cpu_to_virtio64(vhost_is_little_endian(vq), val);
283}
284#endif