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 
16 struct vhost_device;
17 
18 struct vhost_work;
19 typedef void (*vhost_work_fn_t)(struct vhost_work *work);
20 
21 struct 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. */
32 struct 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 
41 void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn,
42 		     unsigned long mask, struct vhost_dev *dev);
43 void vhost_poll_start(struct vhost_poll *poll, struct file *file);
44 void vhost_poll_stop(struct vhost_poll *poll);
45 void vhost_poll_flush(struct vhost_poll *poll);
46 void vhost_poll_queue(struct vhost_poll *poll);
47 
48 struct vhost_log {
49 	u64 addr;
50 	u64 len;
51 };
52 
53 /* The virtqueue structure describes a queue attached to a device. */
54 struct 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 	/* Log writes to used structure. */
88 	bool log_used;
89 	u64 log_addr;
90 
91 	struct iovec iov[UIO_MAXIOV];
92 	/* hdr is used to store the virtio header.
93 	 * Since each iovec has >= 1 byte length, we never need more than
94 	 * header length entries to store the header. */
95 	struct iovec hdr[sizeof(struct virtio_net_hdr_mrg_rxbuf)];
96 	struct iovec *indirect;
97 	size_t vhost_hlen;
98 	size_t sock_hlen;
99 	struct vring_used_elem *heads;
100 	/* We use a kind of RCU to access private pointer.
101 	 * All readers access it from worker, which makes it possible to
102 	 * flush the vhost_work instead of synchronize_rcu. Therefore readers do
103 	 * not need to call rcu_read_lock/rcu_read_unlock: the beginning of
104 	 * vhost_work execution acts instead of rcu_read_lock() and the end of
105 	 * vhost_work execution acts instead of rcu_read_unlock().
106 	 * Writers use virtqueue mutex. */
107 	void __rcu *private_data;
108 	/* Log write descriptors */
109 	void __user *log_base;
110 	struct vhost_log *log;
111 };
112 
113 struct vhost_dev {
114 	/* Readers use RCU to access memory table pointer
115 	 * log base pointer and features.
116 	 * Writers use mutex below.*/
117 	struct vhost_memory __rcu *memory;
118 	struct mm_struct *mm;
119 	struct mutex mutex;
120 	unsigned acked_features;
121 	struct vhost_virtqueue *vqs;
122 	int nvqs;
123 	struct file *log_file;
124 	struct eventfd_ctx *log_ctx;
125 	spinlock_t work_lock;
126 	struct list_head work_list;
127 	struct task_struct *worker;
128 };
129 
130 long vhost_dev_init(struct vhost_dev *, struct vhost_virtqueue *vqs, int nvqs);
131 long vhost_dev_check_owner(struct vhost_dev *);
132 long vhost_dev_reset_owner(struct vhost_dev *);
133 void vhost_dev_cleanup(struct vhost_dev *);
134 long vhost_dev_ioctl(struct vhost_dev *, unsigned int ioctl, unsigned long arg);
135 int vhost_vq_access_ok(struct vhost_virtqueue *vq);
136 int vhost_log_access_ok(struct vhost_dev *);
137 
138 int vhost_get_vq_desc(struct vhost_dev *, struct vhost_virtqueue *,
139 		      struct iovec iov[], unsigned int iov_count,
140 		      unsigned int *out_num, unsigned int *in_num,
141 		      struct vhost_log *log, unsigned int *log_num);
142 void vhost_discard_vq_desc(struct vhost_virtqueue *, int n);
143 
144 int vhost_add_used(struct vhost_virtqueue *, unsigned int head, int len);
145 int vhost_add_used_n(struct vhost_virtqueue *, struct vring_used_elem *heads,
146 		     unsigned count);
147 void vhost_add_used_and_signal(struct vhost_dev *, struct vhost_virtqueue *,
148 			       unsigned int id, int len);
149 void vhost_add_used_and_signal_n(struct vhost_dev *, struct vhost_virtqueue *,
150 			       struct vring_used_elem *heads, unsigned count);
151 void vhost_signal(struct vhost_dev *, struct vhost_virtqueue *);
152 void vhost_disable_notify(struct vhost_virtqueue *);
153 bool vhost_enable_notify(struct vhost_virtqueue *);
154 
155 int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
156 		    unsigned int log_num, u64 len);
157 
158 #define vq_err(vq, fmt, ...) do {                                  \
159 		pr_debug(pr_fmt(fmt), ##__VA_ARGS__);       \
160 		if ((vq)->error_ctx)                               \
161 				eventfd_signal((vq)->error_ctx, 1);\
162 	} while (0)
163 
164 enum {
165 	VHOST_FEATURES = (1 << VIRTIO_F_NOTIFY_ON_EMPTY) |
166 			 (1 << VIRTIO_RING_F_INDIRECT_DESC) |
167 			 (1 << VHOST_F_LOG_ALL) |
168 			 (1 << VHOST_NET_F_VIRTIO_NET_HDR) |
169 			 (1 << VIRTIO_NET_F_MRG_RXBUF),
170 };
171 
vhost_has_feature(struct vhost_dev * dev,int bit)172 static inline int vhost_has_feature(struct vhost_dev *dev, int bit)
173 {
174 	unsigned acked_features;
175 
176 	/* TODO: check that we are running from vhost_worker or dev mutex is
177 	 * held? */
178 	acked_features = rcu_dereference_index_check(dev->acked_features, 1);
179 	return acked_features & (1 << bit);
180 }
181 
182 #endif
183