4 #include <linux/eventfd.h>
5 #include <linux/vhost.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>
18 typedef void (*vhost_work_fn_t
)(struct vhost_work
*work
);
21 struct list_head node
;
23 wait_queue_head_t done
;
29 /* Poll a file (eventfd or socket) */
30 /* Note: there's nothing vhost specific about this structure. */
33 wait_queue_head_t
*wqh
;
35 struct vhost_work work
;
37 struct vhost_dev
*dev
;
40 void vhost_work_init(struct vhost_work
*work
, vhost_work_fn_t fn
);
41 void vhost_work_queue(struct vhost_dev
*dev
, struct vhost_work
*work
);
43 void vhost_poll_init(struct vhost_poll
*poll
, vhost_work_fn_t fn
,
44 unsigned long mask
, struct vhost_dev
*dev
);
45 int vhost_poll_start(struct vhost_poll
*poll
, struct file
*file
);
46 void vhost_poll_stop(struct vhost_poll
*poll
);
47 void vhost_poll_flush(struct vhost_poll
*poll
);
48 void vhost_poll_queue(struct vhost_poll
*poll
);
55 struct vhost_virtqueue
;
57 struct vhost_ubuf_ref
{
59 wait_queue_head_t wait
;
60 struct vhost_virtqueue
*vq
;
63 struct vhost_ubuf_ref
*vhost_ubuf_alloc(struct vhost_virtqueue
*, bool zcopy
);
64 void vhost_ubuf_put(struct vhost_ubuf_ref
*);
65 void vhost_ubuf_put_and_wait(struct vhost_ubuf_ref
*);
69 /* The virtqueue structure describes a queue attached to a device. */
70 struct vhost_virtqueue
{
71 struct vhost_dev
*dev
;
73 /* The actual ring of buffers. */
76 struct vring_desc __user
*desc
;
77 struct vring_avail __user
*avail
;
78 struct vring_used __user
*used
;
82 struct eventfd_ctx
*call_ctx
;
83 struct eventfd_ctx
*error_ctx
;
84 struct eventfd_ctx
*log_ctx
;
86 struct vhost_poll poll
;
88 /* The routine to call when the Guest pings us, or timeout. */
89 vhost_work_fn_t handle_kick
;
91 /* Last available index we saw. */
94 /* Caches available index value from user. */
97 /* Last index we used. */
103 /* Last used index value we have signalled on */
106 /* Last used index value we have signalled on */
107 bool signalled_used_valid
;
109 /* Log writes to used structure. */
113 struct iovec iov
[UIO_MAXIOV
];
114 /* hdr is used to store the virtio header.
115 * Since each iovec has >= 1 byte length, we never need more than
116 * header length entries to store the header. */
117 struct iovec hdr
[sizeof(struct virtio_net_hdr_mrg_rxbuf
)];
118 struct iovec
*indirect
;
121 struct vring_used_elem
*heads
;
122 /* We use a kind of RCU to access private pointer.
123 * All readers access it from worker, which makes it possible to
124 * flush the vhost_work instead of synchronize_rcu. Therefore readers do
125 * not need to call rcu_read_lock/rcu_read_unlock: the beginning of
126 * vhost_work execution acts instead of rcu_read_lock() and the end of
127 * vhost_work execution acts instead of rcu_read_unlock().
128 * Writers use virtqueue mutex. */
129 void __rcu
*private_data
;
130 /* Log write descriptors */
131 void __user
*log_base
;
132 struct vhost_log
*log
;
133 /* vhost zerocopy support fields below: */
134 /* last used idx for outstanding DMA zerocopy buffers */
136 /* first used idx for DMA done zerocopy buffers */
138 /* an array of userspace buffers info */
139 struct ubuf_info
*ubuf_info
;
140 /* Reference counting for outstanding ubufs.
141 * Protected by vq mutex. Writers must also take device mutex. */
142 struct vhost_ubuf_ref
*ubufs
;
146 /* Readers use RCU to access memory table pointer
147 * log base pointer and features.
148 * Writers use mutex below.*/
149 struct vhost_memory __rcu
*memory
;
150 struct mm_struct
*mm
;
152 unsigned acked_features
;
153 struct vhost_virtqueue
*vqs
;
155 struct file
*log_file
;
156 struct eventfd_ctx
*log_ctx
;
157 spinlock_t work_lock
;
158 struct list_head work_list
;
159 struct task_struct
*worker
;
162 long vhost_dev_init(struct vhost_dev
*, struct vhost_virtqueue
*vqs
, int nvqs
);
163 long vhost_dev_check_owner(struct vhost_dev
*);
164 long vhost_dev_reset_owner(struct vhost_dev
*);
165 void vhost_dev_cleanup(struct vhost_dev
*, bool locked
);
166 void vhost_dev_stop(struct vhost_dev
*);
167 long vhost_dev_ioctl(struct vhost_dev
*, unsigned int ioctl
, void __user
*argp
);
168 long vhost_vring_ioctl(struct vhost_dev
*d
, int ioctl
, void __user
*argp
);
169 int vhost_vq_access_ok(struct vhost_virtqueue
*vq
);
170 int vhost_log_access_ok(struct vhost_dev
*);
172 int vhost_get_vq_desc(struct vhost_dev
*, struct vhost_virtqueue
*,
173 struct iovec iov
[], unsigned int iov_count
,
174 unsigned int *out_num
, unsigned int *in_num
,
175 struct vhost_log
*log
, unsigned int *log_num
);
176 void vhost_discard_vq_desc(struct vhost_virtqueue
*, int n
);
178 int vhost_init_used(struct vhost_virtqueue
*);
179 int vhost_add_used(struct vhost_virtqueue
*, unsigned int head
, int len
);
180 int vhost_add_used_n(struct vhost_virtqueue
*, struct vring_used_elem
*heads
,
182 void vhost_add_used_and_signal(struct vhost_dev
*, struct vhost_virtqueue
*,
183 unsigned int id
, int len
);
184 void vhost_add_used_and_signal_n(struct vhost_dev
*, struct vhost_virtqueue
*,
185 struct vring_used_elem
*heads
, unsigned count
);
186 void vhost_signal(struct vhost_dev
*, struct vhost_virtqueue
*);
187 void vhost_disable_notify(struct vhost_dev
*, struct vhost_virtqueue
*);
188 bool vhost_enable_notify(struct vhost_dev
*, struct vhost_virtqueue
*);
190 int vhost_log_write(struct vhost_virtqueue
*vq
, struct vhost_log
*log
,
191 unsigned int log_num
, u64 len
);
193 #define vq_err(vq, fmt, ...) do { \
194 pr_debug(pr_fmt(fmt), ##__VA_ARGS__); \
195 if ((vq)->error_ctx) \
196 eventfd_signal((vq)->error_ctx, 1);\
200 VHOST_FEATURES
= (1ULL << VIRTIO_F_NOTIFY_ON_EMPTY
) |
201 (1ULL << VIRTIO_RING_F_INDIRECT_DESC
) |
202 (1ULL << VIRTIO_RING_F_EVENT_IDX
) |
203 (1ULL << VHOST_F_LOG_ALL
),
204 VHOST_NET_FEATURES
= VHOST_FEATURES
|
205 (1ULL << VHOST_NET_F_VIRTIO_NET_HDR
) |
206 (1ULL << VIRTIO_NET_F_MRG_RXBUF
),
209 static inline int vhost_has_feature(struct vhost_dev
*dev
, int bit
)
211 unsigned acked_features
;
213 /* TODO: check that we are running from vhost_worker or dev mutex is
215 acked_features
= rcu_dereference_index_check(dev
->acked_features
, 1);
216 return acked_features
& (1 << bit
);
219 void vhost_enable_zcopy(int vq
);