Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 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/workqueue.h> |
| 9 | #include <linux/poll.h> |
| 10 | #include <linux/file.h> |
| 11 | #include <linux/skbuff.h> |
| 12 | #include <linux/uio.h> |
| 13 | #include <linux/virtio_config.h> |
| 14 | #include <linux/virtio_ring.h> |
| 15 | |
| 16 | struct vhost_device; |
| 17 | |
| 18 | enum { |
| 19 | /* Enough place for all fragments, head, and virtio net header. */ |
| 20 | VHOST_NET_MAX_SG = MAX_SKB_FRAGS + 2, |
| 21 | }; |
| 22 | |
| 23 | /* Poll a file (eventfd or socket) */ |
| 24 | /* Note: there's nothing vhost specific about this structure. */ |
| 25 | struct vhost_poll { |
| 26 | poll_table table; |
| 27 | wait_queue_head_t *wqh; |
| 28 | wait_queue_t wait; |
| 29 | /* struct which will handle all actual work. */ |
| 30 | struct work_struct work; |
| 31 | unsigned long mask; |
| 32 | }; |
| 33 | |
| 34 | void vhost_poll_init(struct vhost_poll *poll, work_func_t func, |
| 35 | unsigned long mask); |
| 36 | void vhost_poll_start(struct vhost_poll *poll, struct file *file); |
| 37 | void vhost_poll_stop(struct vhost_poll *poll); |
| 38 | void vhost_poll_flush(struct vhost_poll *poll); |
| 39 | void vhost_poll_queue(struct vhost_poll *poll); |
| 40 | |
| 41 | struct vhost_log { |
| 42 | u64 addr; |
| 43 | u64 len; |
| 44 | }; |
| 45 | |
| 46 | /* The virtqueue structure describes a queue attached to a device. */ |
| 47 | struct vhost_virtqueue { |
| 48 | struct vhost_dev *dev; |
| 49 | |
| 50 | /* The actual ring of buffers. */ |
| 51 | struct mutex mutex; |
| 52 | unsigned int num; |
| 53 | struct vring_desc __user *desc; |
| 54 | struct vring_avail __user *avail; |
| 55 | struct vring_used __user *used; |
| 56 | struct file *kick; |
| 57 | struct file *call; |
| 58 | struct file *error; |
| 59 | struct eventfd_ctx *call_ctx; |
| 60 | struct eventfd_ctx *error_ctx; |
| 61 | struct eventfd_ctx *log_ctx; |
| 62 | |
| 63 | struct vhost_poll poll; |
| 64 | |
| 65 | /* The routine to call when the Guest pings us, or timeout. */ |
| 66 | work_func_t handle_kick; |
| 67 | |
| 68 | /* Last available index we saw. */ |
| 69 | u16 last_avail_idx; |
| 70 | |
| 71 | /* Caches available index value from user. */ |
| 72 | u16 avail_idx; |
| 73 | |
| 74 | /* Last index we used. */ |
| 75 | u16 last_used_idx; |
| 76 | |
| 77 | /* Used flags */ |
| 78 | u16 used_flags; |
| 79 | |
| 80 | /* Log writes to used structure. */ |
| 81 | bool log_used; |
| 82 | u64 log_addr; |
| 83 | |
| 84 | struct iovec indirect[VHOST_NET_MAX_SG]; |
| 85 | struct iovec iov[VHOST_NET_MAX_SG]; |
| 86 | struct iovec hdr[VHOST_NET_MAX_SG]; |
| 87 | size_t hdr_size; |
| 88 | /* We use a kind of RCU to access private pointer. |
| 89 | * All readers access it from workqueue, which makes it possible to |
| 90 | * flush the workqueue instead of synchronize_rcu. Therefore readers do |
| 91 | * not need to call rcu_read_lock/rcu_read_unlock: the beginning of |
| 92 | * work item execution acts instead of rcu_read_lock() and the end of |
| 93 | * work item execution acts instead of rcu_read_lock(). |
| 94 | * Writers use virtqueue mutex. */ |
| 95 | void *private_data; |
| 96 | /* Log write descriptors */ |
| 97 | void __user *log_base; |
| 98 | struct vhost_log log[VHOST_NET_MAX_SG]; |
| 99 | }; |
| 100 | |
| 101 | struct vhost_dev { |
| 102 | /* Readers use RCU to access memory table pointer |
| 103 | * log base pointer and features. |
| 104 | * Writers use mutex below.*/ |
| 105 | struct vhost_memory *memory; |
| 106 | struct mm_struct *mm; |
| 107 | struct mutex mutex; |
| 108 | unsigned acked_features; |
| 109 | struct vhost_virtqueue *vqs; |
| 110 | int nvqs; |
| 111 | struct file *log_file; |
| 112 | struct eventfd_ctx *log_ctx; |
| 113 | }; |
| 114 | |
| 115 | long vhost_dev_init(struct vhost_dev *, struct vhost_virtqueue *vqs, int nvqs); |
| 116 | long vhost_dev_check_owner(struct vhost_dev *); |
| 117 | long vhost_dev_reset_owner(struct vhost_dev *); |
| 118 | void vhost_dev_cleanup(struct vhost_dev *); |
| 119 | long vhost_dev_ioctl(struct vhost_dev *, unsigned int ioctl, unsigned long arg); |
| 120 | int vhost_vq_access_ok(struct vhost_virtqueue *vq); |
| 121 | int vhost_log_access_ok(struct vhost_dev *); |
| 122 | |
Michael S. Tsirkin | d5675bd | 2010-06-24 16:59:59 +0300 | [diff] [blame] | 123 | int vhost_get_vq_desc(struct vhost_dev *, struct vhost_virtqueue *, |
| 124 | struct iovec iov[], unsigned int iov_count, |
| 125 | unsigned int *out_num, unsigned int *in_num, |
| 126 | struct vhost_log *log, unsigned int *log_num); |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 127 | void vhost_discard_vq_desc(struct vhost_virtqueue *); |
| 128 | |
| 129 | int vhost_add_used(struct vhost_virtqueue *, unsigned int head, int len); |
| 130 | void vhost_signal(struct vhost_dev *, struct vhost_virtqueue *); |
| 131 | void vhost_add_used_and_signal(struct vhost_dev *, struct vhost_virtqueue *, |
| 132 | unsigned int head, int len); |
| 133 | void vhost_disable_notify(struct vhost_virtqueue *); |
| 134 | bool vhost_enable_notify(struct vhost_virtqueue *); |
| 135 | |
| 136 | int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log, |
| 137 | unsigned int log_num, u64 len); |
| 138 | |
| 139 | int vhost_init(void); |
| 140 | void vhost_cleanup(void); |
| 141 | |
| 142 | #define vq_err(vq, fmt, ...) do { \ |
| 143 | pr_debug(pr_fmt(fmt), ##__VA_ARGS__); \ |
| 144 | if ((vq)->error_ctx) \ |
| 145 | eventfd_signal((vq)->error_ctx, 1);\ |
| 146 | } while (0) |
| 147 | |
| 148 | enum { |
| 149 | VHOST_FEATURES = (1 << VIRTIO_F_NOTIFY_ON_EMPTY) | |
| 150 | (1 << VIRTIO_RING_F_INDIRECT_DESC) | |
| 151 | (1 << VHOST_F_LOG_ALL) | |
| 152 | (1 << VHOST_NET_F_VIRTIO_NET_HDR), |
| 153 | }; |
| 154 | |
| 155 | static inline int vhost_has_feature(struct vhost_dev *dev, int bit) |
| 156 | { |
| 157 | unsigned acked_features = rcu_dereference(dev->acked_features); |
| 158 | return acked_features & (1 << bit); |
| 159 | } |
| 160 | |
| 161 | #endif |