Loading...
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 <linux/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_init_used(struct vhost_virtqueue *);
178int vhost_add_used(struct vhost_virtqueue *, unsigned int head, int len);
179int vhost_add_used_n(struct vhost_virtqueue *, struct vring_used_elem *heads,
180 unsigned count);
181void vhost_add_used_and_signal(struct vhost_dev *, struct vhost_virtqueue *,
182 unsigned int id, int len);
183void vhost_add_used_and_signal_n(struct vhost_dev *, struct vhost_virtqueue *,
184 struct vring_used_elem *heads, unsigned count);
185void vhost_signal(struct vhost_dev *, struct vhost_virtqueue *);
186void vhost_disable_notify(struct vhost_dev *, struct vhost_virtqueue *);
187bool vhost_enable_notify(struct vhost_dev *, struct vhost_virtqueue *);
188
189int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
190 unsigned int log_num, u64 len);
191void vhost_zerocopy_callback(void *arg);
192int vhost_zerocopy_signal_used(struct vhost_virtqueue *vq);
193
194#define vq_err(vq, fmt, ...) do { \
195 pr_debug(pr_fmt(fmt), ##__VA_ARGS__); \
196 if ((vq)->error_ctx) \
197 eventfd_signal((vq)->error_ctx, 1);\
198 } while (0)
199
200enum {
201 VHOST_FEATURES = (1ULL << VIRTIO_F_NOTIFY_ON_EMPTY) |
202 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) |
203 (1ULL << VIRTIO_RING_F_EVENT_IDX) |
204 (1ULL << VHOST_F_LOG_ALL) |
205 (1ULL << VHOST_NET_F_VIRTIO_NET_HDR) |
206 (1ULL << VIRTIO_NET_F_MRG_RXBUF),
207};
208
209static inline int vhost_has_feature(struct vhost_dev *dev, int bit)
210{
211 unsigned acked_features;
212
213 /* TODO: check that we are running from vhost_worker or dev mutex is
214 * held? */
215 acked_features = rcu_dereference_index_check(dev->acked_features, 1);
216 return acked_features & (1 << bit);
217}
218
219void vhost_enable_zcopy(int vq);
220
221#endif
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _VHOST_H
3#define _VHOST_H
4
5#include <linux/eventfd.h>
6#include <linux/vhost.h>
7#include <linux/mm.h>
8#include <linux/mutex.h>
9#include <linux/poll.h>
10#include <linux/file.h>
11#include <linux/uio.h>
12#include <linux/virtio_config.h>
13#include <linux/virtio_ring.h>
14#include <linux/atomic.h>
15#include <linux/vhost_iotlb.h>
16#include <linux/irqbypass.h>
17
18struct vhost_work;
19typedef void (*vhost_work_fn_t)(struct vhost_work *work);
20
21#define VHOST_WORK_QUEUED 1
22struct vhost_work {
23 struct llist_node node;
24 vhost_work_fn_t fn;
25 unsigned long flags;
26};
27
28/* Poll a file (eventfd or socket) */
29/* Note: there's nothing vhost specific about this structure. */
30struct vhost_poll {
31 poll_table table;
32 wait_queue_head_t *wqh;
33 wait_queue_entry_t wait;
34 struct vhost_work work;
35 __poll_t mask;
36 struct vhost_dev *dev;
37};
38
39void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn);
40void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work);
41bool vhost_has_work(struct vhost_dev *dev);
42
43void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn,
44 __poll_t mask, struct vhost_dev *dev);
45int vhost_poll_start(struct vhost_poll *poll, struct file *file);
46void vhost_poll_stop(struct vhost_poll *poll);
47void vhost_poll_queue(struct vhost_poll *poll);
48void vhost_dev_flush(struct vhost_dev *dev);
49
50struct vhost_log {
51 u64 addr;
52 u64 len;
53};
54
55enum vhost_uaddr_type {
56 VHOST_ADDR_DESC = 0,
57 VHOST_ADDR_AVAIL = 1,
58 VHOST_ADDR_USED = 2,
59 VHOST_NUM_ADDRS = 3,
60};
61
62struct vhost_vring_call {
63 struct eventfd_ctx *ctx;
64 struct irq_bypass_producer producer;
65};
66
67/* The virtqueue structure describes a queue attached to a device. */
68struct vhost_virtqueue {
69 struct vhost_dev *dev;
70
71 /* The actual ring of buffers. */
72 struct mutex mutex;
73 unsigned int num;
74 vring_desc_t __user *desc;
75 vring_avail_t __user *avail;
76 vring_used_t __user *used;
77 const struct vhost_iotlb_map *meta_iotlb[VHOST_NUM_ADDRS];
78 struct file *kick;
79 struct vhost_vring_call call_ctx;
80 struct eventfd_ctx *error_ctx;
81 struct eventfd_ctx *log_ctx;
82
83 struct vhost_poll poll;
84
85 /* The routine to call when the Guest pings us, or timeout. */
86 vhost_work_fn_t handle_kick;
87
88 /* Last available index we saw. */
89 u16 last_avail_idx;
90
91 /* Caches available index value from user. */
92 u16 avail_idx;
93
94 /* Last index we used. */
95 u16 last_used_idx;
96
97 /* Used flags */
98 u16 used_flags;
99
100 /* Last used index value we have signalled on */
101 u16 signalled_used;
102
103 /* Last used index value we have signalled on */
104 bool signalled_used_valid;
105
106 /* Log writes to used structure. */
107 bool log_used;
108 u64 log_addr;
109
110 struct iovec iov[UIO_MAXIOV];
111 struct iovec iotlb_iov[64];
112 struct iovec *indirect;
113 struct vring_used_elem *heads;
114 /* Protected by virtqueue mutex. */
115 struct vhost_iotlb *umem;
116 struct vhost_iotlb *iotlb;
117 void *private_data;
118 u64 acked_features;
119 u64 acked_backend_features;
120 /* Log write descriptors */
121 void __user *log_base;
122 struct vhost_log *log;
123 struct iovec log_iov[64];
124
125 /* Ring endianness. Defaults to legacy native endianness.
126 * Set to true when starting a modern virtio device. */
127 bool is_le;
128#ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
129 /* Ring endianness requested by userspace for cross-endian support. */
130 bool user_be;
131#endif
132 u32 busyloop_timeout;
133};
134
135struct vhost_msg_node {
136 union {
137 struct vhost_msg msg;
138 struct vhost_msg_v2 msg_v2;
139 };
140 struct vhost_virtqueue *vq;
141 struct list_head node;
142};
143
144struct vhost_dev {
145 struct mm_struct *mm;
146 struct mutex mutex;
147 struct vhost_virtqueue **vqs;
148 int nvqs;
149 struct eventfd_ctx *log_ctx;
150 struct llist_head work_list;
151 struct task_struct *worker;
152 struct vhost_iotlb *umem;
153 struct vhost_iotlb *iotlb;
154 spinlock_t iotlb_lock;
155 struct list_head read_list;
156 struct list_head pending_list;
157 wait_queue_head_t wait;
158 int iov_limit;
159 int weight;
160 int byte_weight;
161 u64 kcov_handle;
162 bool use_worker;
163 int (*msg_handler)(struct vhost_dev *dev, u32 asid,
164 struct vhost_iotlb_msg *msg);
165};
166
167bool vhost_exceeds_weight(struct vhost_virtqueue *vq, int pkts, int total_len);
168void vhost_dev_init(struct vhost_dev *, struct vhost_virtqueue **vqs,
169 int nvqs, int iov_limit, int weight, int byte_weight,
170 bool use_worker,
171 int (*msg_handler)(struct vhost_dev *dev, u32 asid,
172 struct vhost_iotlb_msg *msg));
173long vhost_dev_set_owner(struct vhost_dev *dev);
174bool vhost_dev_has_owner(struct vhost_dev *dev);
175long vhost_dev_check_owner(struct vhost_dev *);
176struct vhost_iotlb *vhost_dev_reset_owner_prepare(void);
177void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_iotlb *iotlb);
178void vhost_dev_cleanup(struct vhost_dev *);
179void vhost_dev_stop(struct vhost_dev *);
180long vhost_dev_ioctl(struct vhost_dev *, unsigned int ioctl, void __user *argp);
181long vhost_vring_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp);
182bool vhost_vq_access_ok(struct vhost_virtqueue *vq);
183bool vhost_log_access_ok(struct vhost_dev *);
184void vhost_clear_msg(struct vhost_dev *dev);
185
186int vhost_get_vq_desc(struct vhost_virtqueue *,
187 struct iovec iov[], unsigned int iov_count,
188 unsigned int *out_num, unsigned int *in_num,
189 struct vhost_log *log, unsigned int *log_num);
190void vhost_discard_vq_desc(struct vhost_virtqueue *, int n);
191
192bool vhost_vq_is_setup(struct vhost_virtqueue *vq);
193int vhost_vq_init_access(struct vhost_virtqueue *);
194int vhost_add_used(struct vhost_virtqueue *, unsigned int head, int len);
195int vhost_add_used_n(struct vhost_virtqueue *, struct vring_used_elem *heads,
196 unsigned count);
197void vhost_add_used_and_signal(struct vhost_dev *, struct vhost_virtqueue *,
198 unsigned int id, int len);
199void vhost_add_used_and_signal_n(struct vhost_dev *, struct vhost_virtqueue *,
200 struct vring_used_elem *heads, unsigned count);
201void vhost_signal(struct vhost_dev *, struct vhost_virtqueue *);
202void vhost_disable_notify(struct vhost_dev *, struct vhost_virtqueue *);
203bool vhost_vq_avail_empty(struct vhost_dev *, struct vhost_virtqueue *);
204bool vhost_enable_notify(struct vhost_dev *, struct vhost_virtqueue *);
205
206int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
207 unsigned int log_num, u64 len,
208 struct iovec *iov, int count);
209int vq_meta_prefetch(struct vhost_virtqueue *vq);
210
211struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type);
212void vhost_enqueue_msg(struct vhost_dev *dev,
213 struct list_head *head,
214 struct vhost_msg_node *node);
215struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev,
216 struct list_head *head);
217void vhost_set_backend_features(struct vhost_dev *dev, u64 features);
218
219__poll_t vhost_chr_poll(struct file *file, struct vhost_dev *dev,
220 poll_table *wait);
221ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to,
222 int noblock);
223ssize_t vhost_chr_write_iter(struct vhost_dev *dev,
224 struct iov_iter *from);
225int vhost_init_device_iotlb(struct vhost_dev *d, bool enabled);
226
227void vhost_iotlb_map_free(struct vhost_iotlb *iotlb,
228 struct vhost_iotlb_map *map);
229
230#define vq_err(vq, fmt, ...) do { \
231 pr_debug(pr_fmt(fmt), ##__VA_ARGS__); \
232 if ((vq)->error_ctx) \
233 eventfd_signal((vq)->error_ctx, 1);\
234 } while (0)
235
236enum {
237 VHOST_FEATURES = (1ULL << VIRTIO_F_NOTIFY_ON_EMPTY) |
238 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) |
239 (1ULL << VIRTIO_RING_F_EVENT_IDX) |
240 (1ULL << VHOST_F_LOG_ALL) |
241 (1ULL << VIRTIO_F_ANY_LAYOUT) |
242 (1ULL << VIRTIO_F_VERSION_1)
243};
244
245/**
246 * vhost_vq_set_backend - Set backend.
247 *
248 * @vq Virtqueue.
249 * @private_data The private data.
250 *
251 * Context: Need to call with vq->mutex acquired.
252 */
253static inline void vhost_vq_set_backend(struct vhost_virtqueue *vq,
254 void *private_data)
255{
256 vq->private_data = private_data;
257}
258
259/**
260 * vhost_vq_get_backend - Get backend.
261 *
262 * @vq Virtqueue.
263 *
264 * Context: Need to call with vq->mutex acquired.
265 * Return: Private data previously set with vhost_vq_set_backend.
266 */
267static inline void *vhost_vq_get_backend(struct vhost_virtqueue *vq)
268{
269 return vq->private_data;
270}
271
272static inline bool vhost_has_feature(struct vhost_virtqueue *vq, int bit)
273{
274 return vq->acked_features & (1ULL << bit);
275}
276
277static inline bool vhost_backend_has_feature(struct vhost_virtqueue *vq, int bit)
278{
279 return vq->acked_backend_features & (1ULL << bit);
280}
281
282#ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
283static inline bool vhost_is_little_endian(struct vhost_virtqueue *vq)
284{
285 return vq->is_le;
286}
287#else
288static inline bool vhost_is_little_endian(struct vhost_virtqueue *vq)
289{
290 return virtio_legacy_is_little_endian() || vq->is_le;
291}
292#endif
293
294/* Memory accessors */
295static inline u16 vhost16_to_cpu(struct vhost_virtqueue *vq, __virtio16 val)
296{
297 return __virtio16_to_cpu(vhost_is_little_endian(vq), val);
298}
299
300static inline __virtio16 cpu_to_vhost16(struct vhost_virtqueue *vq, u16 val)
301{
302 return __cpu_to_virtio16(vhost_is_little_endian(vq), val);
303}
304
305static inline u32 vhost32_to_cpu(struct vhost_virtqueue *vq, __virtio32 val)
306{
307 return __virtio32_to_cpu(vhost_is_little_endian(vq), val);
308}
309
310static inline __virtio32 cpu_to_vhost32(struct vhost_virtqueue *vq, u32 val)
311{
312 return __cpu_to_virtio32(vhost_is_little_endian(vq), val);
313}
314
315static inline u64 vhost64_to_cpu(struct vhost_virtqueue *vq, __virtio64 val)
316{
317 return __virtio64_to_cpu(vhost_is_little_endian(vq), val);
318}
319
320static inline __virtio64 cpu_to_vhost64(struct vhost_virtqueue *vq, u64 val)
321{
322 return __cpu_to_virtio64(vhost_is_little_endian(vq), val);
323}
324#endif