vhost.h revision bab632d69ee48a106e779b60cc01adfe80a72807 
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/* This is for zerocopy, used buffer len is set to 1 when lower device DMA
17 * done */
18#define VHOST_DMA_DONE_LEN	1
19#define VHOST_DMA_CLEAR_LEN	0
20
21struct vhost_device;
22
23struct vhost_work;
24typedef void (*vhost_work_fn_t)(struct vhost_work *work);
25
26struct vhost_work {
27	struct list_head	  node;
28	vhost_work_fn_t		  fn;
29	wait_queue_head_t	  done;
30	int			  flushing;
31	unsigned		  queue_seq;
32	unsigned		  done_seq;
33};
34
35/* Poll a file (eventfd or socket) */
36/* Note: there's nothing vhost specific about this structure. */
37struct vhost_poll {
38	poll_table                table;
39	wait_queue_head_t        *wqh;
40	wait_queue_t              wait;
41	struct vhost_work	  work;
42	unsigned long		  mask;
43	struct vhost_dev	 *dev;
44};
45
46void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn,
47		     unsigned long mask, struct vhost_dev *dev);
48void vhost_poll_start(struct vhost_poll *poll, struct file *file);
49void vhost_poll_stop(struct vhost_poll *poll);
50void vhost_poll_flush(struct vhost_poll *poll);
51void vhost_poll_queue(struct vhost_poll *poll);
52
53struct vhost_log {
54	u64 addr;
55	u64 len;
56};
57
58struct vhost_virtqueue;
59
60struct vhost_ubuf_ref {
61	struct kref kref;
62	wait_queue_head_t wait;
63	struct vhost_virtqueue *vq;
64};
65
66struct vhost_ubuf_ref *vhost_ubuf_alloc(struct vhost_virtqueue *, bool zcopy);
67void vhost_ubuf_put(struct vhost_ubuf_ref *);
68void vhost_ubuf_put_and_wait(struct vhost_ubuf_ref *);
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	struct vring_desc __user *desc;
78	struct vring_avail __user *avail;
79	struct vring_used __user *used;
80	struct file *kick;
81	struct file *call;
82	struct file *error;
83	struct eventfd_ctx *call_ctx;
84	struct eventfd_ctx *error_ctx;
85	struct eventfd_ctx *log_ctx;
86
87	struct vhost_poll poll;
88
89	/* The routine to call when the Guest pings us, or timeout. */
90	vhost_work_fn_t handle_kick;
91
92	/* Last available index we saw. */
93	u16 last_avail_idx;
94
95	/* Caches available index value from user. */
96	u16 avail_idx;
97
98	/* Last index we used. */
99	u16 last_used_idx;
100
101	/* Used flags */
102	u16 used_flags;
103
104	/* Last used index value we have signalled on */
105	u16 signalled_used;
106
107	/* Last used index value we have signalled on */
108	bool signalled_used_valid;
109
110	/* Log writes to used structure. */
111	bool log_used;
112	u64 log_addr;
113
114	struct iovec iov[UIO_MAXIOV];
115	/* hdr is used to store the virtio header.
116	 * Since each iovec has >= 1 byte length, we never need more than
117	 * header length entries to store the header. */
118	struct iovec hdr[sizeof(struct virtio_net_hdr_mrg_rxbuf)];
119	struct iovec *indirect;
120	size_t vhost_hlen;
121	size_t sock_hlen;
122	struct vring_used_elem *heads;
123	/* We use a kind of RCU to access private pointer.
124	 * All readers access it from worker, which makes it possible to
125	 * flush the vhost_work instead of synchronize_rcu. Therefore readers do
126	 * not need to call rcu_read_lock/rcu_read_unlock: the beginning of
127	 * vhost_work execution acts instead of rcu_read_lock() and the end of
128	 * vhost_work execution acts instead of rcu_read_unlock().
129	 * Writers use virtqueue mutex. */
130	void __rcu *private_data;
131	/* Log write descriptors */
132	void __user *log_base;
133	struct vhost_log *log;
134	/* vhost zerocopy support fields below: */
135	/* last used idx for outstanding DMA zerocopy buffers */
136	int upend_idx;
137	/* first used idx for DMA done zerocopy buffers */
138	int done_idx;
139	/* an array of userspace buffers info */
140	struct ubuf_info *ubuf_info;
141	/* Reference counting for outstanding ubufs.
142	 * Protected by vq mutex. Writers must also take device mutex. */
143	struct vhost_ubuf_ref *ubufs;
144};
145
146struct vhost_dev {
147	/* Readers use RCU to access memory table pointer
148	 * log base pointer and features.
149	 * Writers use mutex below.*/
150	struct vhost_memory __rcu *memory;
151	struct mm_struct *mm;
152	struct mutex mutex;
153	unsigned acked_features;
154	struct vhost_virtqueue *vqs;
155	int nvqs;
156	struct file *log_file;
157	struct eventfd_ctx *log_ctx;
158	spinlock_t work_lock;
159	struct list_head work_list;
160	struct task_struct *worker;
161};
162
163long vhost_dev_init(struct vhost_dev *, struct vhost_virtqueue *vqs, int nvqs);
164long vhost_dev_check_owner(struct vhost_dev *);
165long vhost_dev_reset_owner(struct vhost_dev *);
166void vhost_dev_cleanup(struct vhost_dev *);
167long vhost_dev_ioctl(struct vhost_dev *, unsigned int ioctl, unsigned long arg);
168int vhost_vq_access_ok(struct vhost_virtqueue *vq);
169int vhost_log_access_ok(struct vhost_dev *);
170
171int vhost_get_vq_desc(struct vhost_dev *, struct vhost_virtqueue *,
172		      struct iovec iov[], unsigned int iov_count,
173		      unsigned int *out_num, unsigned int *in_num,
174		      struct vhost_log *log, unsigned int *log_num);
175void vhost_discard_vq_desc(struct vhost_virtqueue *, int n);
176
177int vhost_add_used(struct vhost_virtqueue *, unsigned int head, int len);
178int vhost_add_used_n(struct vhost_virtqueue *, struct vring_used_elem *heads,
179		     unsigned count);
180void vhost_add_used_and_signal(struct vhost_dev *, struct vhost_virtqueue *,
181			       unsigned int id, int len);
182void vhost_add_used_and_signal_n(struct vhost_dev *, struct vhost_virtqueue *,
183			       struct vring_used_elem *heads, unsigned count);
184void vhost_signal(struct vhost_dev *, struct vhost_virtqueue *);
185void vhost_disable_notify(struct vhost_dev *, struct vhost_virtqueue *);
186bool vhost_enable_notify(struct vhost_dev *, struct vhost_virtqueue *);
187
188int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
189		    unsigned int log_num, u64 len);
190void vhost_zerocopy_callback(void *arg);
191int vhost_zerocopy_signal_used(struct vhost_virtqueue *vq);
192
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);\
197	} while (0)
198
199enum {
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			 (1ULL << VHOST_NET_F_VIRTIO_NET_HDR) |
205			 (1ULL << VIRTIO_NET_F_MRG_RXBUF),
206};
207
208static inline int vhost_has_feature(struct vhost_dev *dev, int bit)
209{
210	unsigned acked_features;
211
212	/* TODO: check that we are running from vhost_worker or dev mutex is
213	 * held? */
214	acked_features = rcu_dereference_index_check(dev->acked_features, 1);
215	return acked_features & (1 << bit);
216}
217
218void vhost_enable_zcopy(int vq);
219
220#endif
221