Loading...
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;
19struct vhost_task;
20typedef void (*vhost_work_fn_t)(struct vhost_work *work);
21
22#define VHOST_WORK_QUEUED 1
23struct vhost_work {
24 struct llist_node node;
25 vhost_work_fn_t fn;
26 unsigned long flags;
27};
28
29struct vhost_worker {
30 struct vhost_task *vtsk;
31 struct vhost_dev *dev;
32 /* Used to serialize device wide flushing with worker swapping. */
33 struct mutex mutex;
34 struct llist_head work_list;
35 u64 kcov_handle;
36 u32 id;
37 int attachment_cnt;
38 bool killed;
39};
40
41/* Poll a file (eventfd or socket) */
42/* Note: there's nothing vhost specific about this structure. */
43struct vhost_poll {
44 poll_table table;
45 wait_queue_head_t *wqh;
46 wait_queue_entry_t wait;
47 struct vhost_work work;
48 __poll_t mask;
49 struct vhost_dev *dev;
50 struct vhost_virtqueue *vq;
51};
52
53void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn,
54 __poll_t mask, struct vhost_dev *dev,
55 struct vhost_virtqueue *vq);
56int vhost_poll_start(struct vhost_poll *poll, struct file *file);
57void vhost_poll_stop(struct vhost_poll *poll);
58void vhost_poll_queue(struct vhost_poll *poll);
59
60void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn);
61void vhost_dev_flush(struct vhost_dev *dev);
62
63struct vhost_log {
64 u64 addr;
65 u64 len;
66};
67
68enum vhost_uaddr_type {
69 VHOST_ADDR_DESC = 0,
70 VHOST_ADDR_AVAIL = 1,
71 VHOST_ADDR_USED = 2,
72 VHOST_NUM_ADDRS = 3,
73};
74
75struct vhost_vring_call {
76 struct eventfd_ctx *ctx;
77 struct irq_bypass_producer producer;
78};
79
80/* The virtqueue structure describes a queue attached to a device. */
81struct vhost_virtqueue {
82 struct vhost_dev *dev;
83 struct vhost_worker __rcu *worker;
84
85 /* The actual ring of buffers. */
86 struct mutex mutex;
87 unsigned int num;
88 vring_desc_t __user *desc;
89 vring_avail_t __user *avail;
90 vring_used_t __user *used;
91 const struct vhost_iotlb_map *meta_iotlb[VHOST_NUM_ADDRS];
92 struct file *kick;
93 struct vhost_vring_call call_ctx;
94 struct eventfd_ctx *error_ctx;
95 struct eventfd_ctx *log_ctx;
96
97 struct vhost_poll poll;
98
99 /* The routine to call when the Guest pings us, or timeout. */
100 vhost_work_fn_t handle_kick;
101
102 /* Last available index we saw.
103 * Values are limited to 0x7fff, and the high bit is used as
104 * a wrap counter when using VIRTIO_F_RING_PACKED. */
105 u16 last_avail_idx;
106
107 /* Caches available index value from user. */
108 u16 avail_idx;
109
110 /* Last index we used.
111 * Values are limited to 0x7fff, and the high bit is used as
112 * a wrap counter when using VIRTIO_F_RING_PACKED. */
113 u16 last_used_idx;
114
115 /* Used flags */
116 u16 used_flags;
117
118 /* Last used index value we have signalled on */
119 u16 signalled_used;
120
121 /* Last used index value we have signalled on */
122 bool signalled_used_valid;
123
124 /* Log writes to used structure. */
125 bool log_used;
126 u64 log_addr;
127
128 struct iovec iov[UIO_MAXIOV];
129 struct iovec iotlb_iov[64];
130 struct iovec *indirect;
131 struct vring_used_elem *heads;
132 /* Protected by virtqueue mutex. */
133 struct vhost_iotlb *umem;
134 struct vhost_iotlb *iotlb;
135 void *private_data;
136 u64 acked_features;
137 u64 acked_backend_features;
138 /* Log write descriptors */
139 void __user *log_base;
140 struct vhost_log *log;
141 struct iovec log_iov[64];
142
143 /* Ring endianness. Defaults to legacy native endianness.
144 * Set to true when starting a modern virtio device. */
145 bool is_le;
146#ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
147 /* Ring endianness requested by userspace for cross-endian support. */
148 bool user_be;
149#endif
150 u32 busyloop_timeout;
151};
152
153struct vhost_msg_node {
154 union {
155 struct vhost_msg msg;
156 struct vhost_msg_v2 msg_v2;
157 };
158 struct vhost_virtqueue *vq;
159 struct list_head node;
160};
161
162struct vhost_dev {
163 struct mm_struct *mm;
164 struct mutex mutex;
165 struct vhost_virtqueue **vqs;
166 int nvqs;
167 struct eventfd_ctx *log_ctx;
168 struct vhost_iotlb *umem;
169 struct vhost_iotlb *iotlb;
170 spinlock_t iotlb_lock;
171 struct list_head read_list;
172 struct list_head pending_list;
173 wait_queue_head_t wait;
174 int iov_limit;
175 int weight;
176 int byte_weight;
177 struct xarray worker_xa;
178 bool use_worker;
179 int (*msg_handler)(struct vhost_dev *dev, u32 asid,
180 struct vhost_iotlb_msg *msg);
181};
182
183bool vhost_exceeds_weight(struct vhost_virtqueue *vq, int pkts, int total_len);
184void vhost_dev_init(struct vhost_dev *, struct vhost_virtqueue **vqs,
185 int nvqs, int iov_limit, int weight, int byte_weight,
186 bool use_worker,
187 int (*msg_handler)(struct vhost_dev *dev, u32 asid,
188 struct vhost_iotlb_msg *msg));
189long vhost_dev_set_owner(struct vhost_dev *dev);
190bool vhost_dev_has_owner(struct vhost_dev *dev);
191long vhost_dev_check_owner(struct vhost_dev *);
192struct vhost_iotlb *vhost_dev_reset_owner_prepare(void);
193void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_iotlb *iotlb);
194void vhost_dev_cleanup(struct vhost_dev *);
195void vhost_dev_stop(struct vhost_dev *);
196long vhost_dev_ioctl(struct vhost_dev *, unsigned int ioctl, void __user *argp);
197long vhost_vring_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp);
198long vhost_worker_ioctl(struct vhost_dev *dev, unsigned int ioctl,
199 void __user *argp);
200bool vhost_vq_access_ok(struct vhost_virtqueue *vq);
201bool vhost_log_access_ok(struct vhost_dev *);
202void vhost_clear_msg(struct vhost_dev *dev);
203
204int vhost_get_vq_desc(struct vhost_virtqueue *,
205 struct iovec iov[], unsigned int iov_size,
206 unsigned int *out_num, unsigned int *in_num,
207 struct vhost_log *log, unsigned int *log_num);
208void vhost_discard_vq_desc(struct vhost_virtqueue *, int n);
209
210bool vhost_vq_work_queue(struct vhost_virtqueue *vq, struct vhost_work *work);
211bool vhost_vq_has_work(struct vhost_virtqueue *vq);
212bool vhost_vq_is_setup(struct vhost_virtqueue *vq);
213int vhost_vq_init_access(struct vhost_virtqueue *);
214int vhost_add_used(struct vhost_virtqueue *, unsigned int head, int len);
215int vhost_add_used_n(struct vhost_virtqueue *, struct vring_used_elem *heads,
216 unsigned count);
217void vhost_add_used_and_signal(struct vhost_dev *, struct vhost_virtqueue *,
218 unsigned int id, int len);
219void vhost_add_used_and_signal_n(struct vhost_dev *, struct vhost_virtqueue *,
220 struct vring_used_elem *heads, unsigned count);
221void vhost_signal(struct vhost_dev *, struct vhost_virtqueue *);
222void vhost_disable_notify(struct vhost_dev *, struct vhost_virtqueue *);
223bool vhost_vq_avail_empty(struct vhost_dev *, struct vhost_virtqueue *);
224bool vhost_enable_notify(struct vhost_dev *, struct vhost_virtqueue *);
225
226int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
227 unsigned int log_num, u64 len,
228 struct iovec *iov, int count);
229int vq_meta_prefetch(struct vhost_virtqueue *vq);
230
231struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type);
232void vhost_enqueue_msg(struct vhost_dev *dev,
233 struct list_head *head,
234 struct vhost_msg_node *node);
235struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev,
236 struct list_head *head);
237void vhost_set_backend_features(struct vhost_dev *dev, u64 features);
238
239__poll_t vhost_chr_poll(struct file *file, struct vhost_dev *dev,
240 poll_table *wait);
241ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to,
242 int noblock);
243ssize_t vhost_chr_write_iter(struct vhost_dev *dev,
244 struct iov_iter *from);
245int vhost_init_device_iotlb(struct vhost_dev *d);
246
247void vhost_iotlb_map_free(struct vhost_iotlb *iotlb,
248 struct vhost_iotlb_map *map);
249
250#define vq_err(vq, fmt, ...) do { \
251 pr_debug(pr_fmt(fmt), ##__VA_ARGS__); \
252 if ((vq)->error_ctx) \
253 eventfd_signal((vq)->error_ctx);\
254 } while (0)
255
256enum {
257 VHOST_FEATURES = (1ULL << VIRTIO_F_NOTIFY_ON_EMPTY) |
258 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) |
259 (1ULL << VIRTIO_RING_F_EVENT_IDX) |
260 (1ULL << VHOST_F_LOG_ALL) |
261 (1ULL << VIRTIO_F_ANY_LAYOUT) |
262 (1ULL << VIRTIO_F_VERSION_1)
263};
264
265/**
266 * vhost_vq_set_backend - Set backend.
267 *
268 * @vq Virtqueue.
269 * @private_data The private data.
270 *
271 * Context: Need to call with vq->mutex acquired.
272 */
273static inline void vhost_vq_set_backend(struct vhost_virtqueue *vq,
274 void *private_data)
275{
276 vq->private_data = private_data;
277}
278
279/**
280 * vhost_vq_get_backend - Get backend.
281 *
282 * @vq Virtqueue.
283 *
284 * Context: Need to call with vq->mutex acquired.
285 * Return: Private data previously set with vhost_vq_set_backend.
286 */
287static inline void *vhost_vq_get_backend(struct vhost_virtqueue *vq)
288{
289 return vq->private_data;
290}
291
292static inline bool vhost_has_feature(struct vhost_virtqueue *vq, int bit)
293{
294 return vq->acked_features & (1ULL << bit);
295}
296
297static inline bool vhost_backend_has_feature(struct vhost_virtqueue *vq, int bit)
298{
299 return vq->acked_backend_features & (1ULL << bit);
300}
301
302#ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
303static inline bool vhost_is_little_endian(struct vhost_virtqueue *vq)
304{
305 return vq->is_le;
306}
307#else
308static inline bool vhost_is_little_endian(struct vhost_virtqueue *vq)
309{
310 return virtio_legacy_is_little_endian() || vq->is_le;
311}
312#endif
313
314/* Memory accessors */
315static inline u16 vhost16_to_cpu(struct vhost_virtqueue *vq, __virtio16 val)
316{
317 return __virtio16_to_cpu(vhost_is_little_endian(vq), val);
318}
319
320static inline __virtio16 cpu_to_vhost16(struct vhost_virtqueue *vq, u16 val)
321{
322 return __cpu_to_virtio16(vhost_is_little_endian(vq), val);
323}
324
325static inline u32 vhost32_to_cpu(struct vhost_virtqueue *vq, __virtio32 val)
326{
327 return __virtio32_to_cpu(vhost_is_little_endian(vq), val);
328}
329
330static inline __virtio32 cpu_to_vhost32(struct vhost_virtqueue *vq, u32 val)
331{
332 return __cpu_to_virtio32(vhost_is_little_endian(vq), val);
333}
334
335static inline u64 vhost64_to_cpu(struct vhost_virtqueue *vq, __virtio64 val)
336{
337 return __virtio64_to_cpu(vhost_is_little_endian(vq), val);
338}
339
340static inline __virtio64 cpu_to_vhost64(struct vhost_virtqueue *vq, u64 val)
341{
342 return __cpu_to_virtio64(vhost_is_little_endian(vq), val);
343}
344#endif
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 *, bool locked);
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(struct ubuf_info *);
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