vhost.h revision 8ea8cf89e19aeb596b818ee5f2bec8a8b0586b60 
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/skbuff.h>
11#include <linux/uio.h>
12#include <linux/virtio_config.h>
13#include <linux/virtio_ring.h>
14#include <asm/atomic.h>
15
16struct vhost_device;
17
18struct vhost_work;
19typedef void (*vhost_work_fn_t)(struct vhost_work *work);
20
21struct vhost_work {
22	struct list_head	  node;
23	vhost_work_fn_t		  fn;
24	wait_queue_head_t	  done;
25	int			  flushing;
26	unsigned		  queue_seq;
27	unsigned		  done_seq;
28};
29
30/* Poll a file (eventfd or socket) */
31/* Note: there's nothing vhost specific about this structure. */
32struct vhost_poll {
33	poll_table                table;
34	wait_queue_head_t        *wqh;
35	wait_queue_t              wait;
36	struct vhost_work	  work;
37	unsigned long		  mask;
38	struct vhost_dev	 *dev;
39};
40
41void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn,
42		     unsigned long mask, struct vhost_dev *dev);
43void vhost_poll_start(struct vhost_poll *poll, struct file *file);
44void vhost_poll_stop(struct vhost_poll *poll);
45void vhost_poll_flush(struct vhost_poll *poll);
46void vhost_poll_queue(struct vhost_poll *poll);
47
48struct vhost_log {
49	u64 addr;
50	u64 len;
51};
52
53/* The virtqueue structure describes a queue attached to a device. */
54struct vhost_virtqueue {
55	struct vhost_dev *dev;
56
57	/* The actual ring of buffers. */
58	struct mutex mutex;
59	unsigned int num;
60	struct vring_desc __user *desc;
61	struct vring_avail __user *avail;
62	struct vring_used __user *used;
63	struct file *kick;
64	struct file *call;
65	struct file *error;
66	struct eventfd_ctx *call_ctx;
67	struct eventfd_ctx *error_ctx;
68	struct eventfd_ctx *log_ctx;
69
70	struct vhost_poll poll;
71
72	/* The routine to call when the Guest pings us, or timeout. */
73	vhost_work_fn_t handle_kick;
74
75	/* Last available index we saw. */
76	u16 last_avail_idx;
77
78	/* Caches available index value from user. */
79	u16 avail_idx;
80
81	/* Last index we used. */
82	u16 last_used_idx;
83
84	/* Used flags */
85	u16 used_flags;
86
87	/* Last used index value we have signalled on */
88	u16 signalled_used;
89
90	/* Last used index value we have signalled on */
91	bool signalled_used_valid;
92
93	/* Log writes to used structure. */
94	bool log_used;
95	u64 log_addr;
96
97	struct iovec iov[UIO_MAXIOV];
98	/* hdr is used to store the virtio header.
99	 * Since each iovec has >= 1 byte length, we never need more than
100	 * header length entries to store the header. */
101	struct iovec hdr[sizeof(struct virtio_net_hdr_mrg_rxbuf)];
102	struct iovec *indirect;
103	size_t vhost_hlen;
104	size_t sock_hlen;
105	struct vring_used_elem *heads;
106	/* We use a kind of RCU to access private pointer.
107	 * All readers access it from worker, which makes it possible to
108	 * flush the vhost_work instead of synchronize_rcu. Therefore readers do
109	 * not need to call rcu_read_lock/rcu_read_unlock: the beginning of
110	 * vhost_work execution acts instead of rcu_read_lock() and the end of
111	 * vhost_work execution acts instead of rcu_read_unlock().
112	 * Writers use virtqueue mutex. */
113	void __rcu *private_data;
114	/* Log write descriptors */
115	void __user *log_base;
116	struct vhost_log *log;
117};
118
119struct vhost_dev {
120	/* Readers use RCU to access memory table pointer
121	 * log base pointer and features.
122	 * Writers use mutex below.*/
123	struct vhost_memory __rcu *memory;
124	struct mm_struct *mm;
125	struct mutex mutex;
126	unsigned acked_features;
127	struct vhost_virtqueue *vqs;
128	int nvqs;
129	struct file *log_file;
130	struct eventfd_ctx *log_ctx;
131	spinlock_t work_lock;
132	struct list_head work_list;
133	struct task_struct *worker;
134};
135
136long vhost_dev_init(struct vhost_dev *, struct vhost_virtqueue *vqs, int nvqs);
137long vhost_dev_check_owner(struct vhost_dev *);
138long vhost_dev_reset_owner(struct vhost_dev *);
139void vhost_dev_cleanup(struct vhost_dev *);
140long vhost_dev_ioctl(struct vhost_dev *, unsigned int ioctl, unsigned long arg);
141int vhost_vq_access_ok(struct vhost_virtqueue *vq);
142int vhost_log_access_ok(struct vhost_dev *);
143
144int vhost_get_vq_desc(struct vhost_dev *, struct vhost_virtqueue *,
145		      struct iovec iov[], unsigned int iov_count,
146		      unsigned int *out_num, unsigned int *in_num,
147		      struct vhost_log *log, unsigned int *log_num);
148void vhost_discard_vq_desc(struct vhost_virtqueue *, int n);
149
150int vhost_add_used(struct vhost_virtqueue *, unsigned int head, int len);
151int vhost_add_used_n(struct vhost_virtqueue *, struct vring_used_elem *heads,
152		     unsigned count);
153void vhost_add_used_and_signal(struct vhost_dev *, struct vhost_virtqueue *,
154			       unsigned int id, int len);
155void vhost_add_used_and_signal_n(struct vhost_dev *, struct vhost_virtqueue *,
156			       struct vring_used_elem *heads, unsigned count);
157void vhost_signal(struct vhost_dev *, struct vhost_virtqueue *);
158void vhost_disable_notify(struct vhost_dev *, struct vhost_virtqueue *);
159bool vhost_enable_notify(struct vhost_dev *, struct vhost_virtqueue *);
160
161int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
162		    unsigned int log_num, u64 len);
163
164#define vq_err(vq, fmt, ...) do {                                  \
165		pr_debug(pr_fmt(fmt), ##__VA_ARGS__);       \
166		if ((vq)->error_ctx)                               \
167				eventfd_signal((vq)->error_ctx, 1);\
168	} while (0)
169
170enum {
171	VHOST_FEATURES = (1ULL << VIRTIO_F_NOTIFY_ON_EMPTY) |
172			 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) |
173			 (1ULL << VIRTIO_RING_F_EVENT_IDX) |
174			 (1ULL << VHOST_F_LOG_ALL) |
175			 (1ULL << VHOST_NET_F_VIRTIO_NET_HDR) |
176			 (1ULL << VIRTIO_NET_F_MRG_RXBUF),
177};
178
179static inline int vhost_has_feature(struct vhost_dev *dev, int bit)
180{
181	unsigned acked_features;
182
183	/* TODO: check that we are running from vhost_worker or dev mutex is
184	 * held? */
185	acked_features = rcu_dereference_index_check(dev->acked_features, 1);
186	return acked_features & (1 << bit);
187}
188
189#endif
190