drivers/vhost/vhost.h

   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/pagemap.h>
  16 #include <linux/mmu_notifier.h>
  17 #include <asm/cacheflush.h>
  18
  19 struct vhost_work;
  20 typedef void (*vhost_work_fn_t)(struct vhost_work *work);
  21
  22 #define VHOST_WORK_QUEUED 1
  23 struct vhost_work {
  24         struct llist_node         node;
  25         vhost_work_fn_t           fn;
  26         unsigned long             flags;
  27 };
  28
  29 /* Poll a file (eventfd or socket) */
  30 /* Note: there's nothing vhost specific about this structure. */
  31 struct vhost_poll {
  32         poll_table                table;
  33         wait_queue_head_t        *wqh;
  34         wait_queue_entry_t              wait;
  35         struct vhost_work         work;
  36         __poll_t                  mask;
  37         struct vhost_dev         *dev;
  38 };
  39
  40 void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn);
  41 void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work);
  42 bool vhost_has_work(struct vhost_dev *dev);
  43
  44 void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn,
  45                      __poll_t mask, struct vhost_dev *dev);
  46 int vhost_poll_start(struct vhost_poll *poll, struct file *file);
  47 void vhost_poll_stop(struct vhost_poll *poll);
  48 void vhost_poll_flush(struct vhost_poll *poll);
  49 void vhost_poll_queue(struct vhost_poll *poll);
  50 void vhost_work_flush(struct vhost_dev *dev, struct vhost_work *work);
  51 long vhost_vring_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp);
  52
  53 struct vhost_log {
  54         u64 addr;
  55         u64 len;
  56 };
  57
  58 #define START(node) ((node)->start)
  59 #define LAST(node) ((node)->last)
  60
  61 struct vhost_umem_node {
  62         struct rb_node rb;
  63         struct list_head link;
  64         __u64 start;
  65         __u64 last;
  66         __u64 size;
  67         __u64 userspace_addr;
  68         __u32 perm;
  69         __u32 flags_padding;
  70         __u64 __subtree_last;
  71 };
  72
  73 struct vhost_umem {
  74         struct rb_root_cached umem_tree;
  75         struct list_head umem_list;
  76         int numem;
  77 };
  78
  79 enum vhost_uaddr_type {
  80         VHOST_ADDR_DESC = 0,
  81         VHOST_ADDR_AVAIL = 1,
  82         VHOST_ADDR_USED = 2,
  83         VHOST_NUM_ADDRS = 3,
  84 };
  85
  86 struct vhost_map {
  87         int npages;
  88         void *addr;
  89         struct page **pages;
  90 };
  91
  92 struct vhost_uaddr {
  93         unsigned long uaddr;
  94         size_t size;
  95         bool write;
  96 };
  97
  98 #if defined(CONFIG_MMU_NOTIFIER) && ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 0
  99 #define VHOST_ARCH_CAN_ACCEL_UACCESS 0
 100 #else
 101 #define VHOST_ARCH_CAN_ACCEL_UACCESS 0
 102 #endif
 103
 104 /* The virtqueue structure describes a queue attached to a device. */
 105 struct vhost_virtqueue {
 106         struct vhost_dev *dev;
 107
 108         /* The actual ring of buffers. */
 109         struct mutex mutex;
 110         unsigned int num;
 111         struct vring_desc __user *desc;
 112         struct vring_avail __user *avail;
 113         struct vring_used __user *used;
 114
 115 #if VHOST_ARCH_CAN_ACCEL_UACCESS
 116         /* Read by memory accessors, modified by meta data
 117          * prefetching, MMU notifier and vring ioctl().
 118          * Synchonrized through mmu_lock (writers) and RCU (writers
 119          * and readers).
 120          */
 121         struct vhost_map __rcu *maps[VHOST_NUM_ADDRS];
 122         /* Read by MMU notifier, modified by vring ioctl(),
 123          * synchronized through MMU notifier
 124          * registering/unregistering.
 125          */
 126         struct vhost_uaddr uaddrs[VHOST_NUM_ADDRS];
 127 #endif
 128         const struct vhost_umem_node *meta_iotlb[VHOST_NUM_ADDRS];
 129
 130         struct file *kick;
 131         struct eventfd_ctx *call_ctx;
 132         struct eventfd_ctx *error_ctx;
 133         struct eventfd_ctx *log_ctx;
 134
 135         struct vhost_poll poll;
 136
 137         /* The routine to call when the Guest pings us, or timeout. */
 138         vhost_work_fn_t handle_kick;
 139
 140         /* Last available index we saw. */
 141         u16 last_avail_idx;
 142
 143         /* Caches available index value from user. */
 144         u16 avail_idx;
 145
 146         /* Last index we used. */
 147         u16 last_used_idx;
 148
 149         /* Used flags */
 150         u16 used_flags;
 151
 152         /* Last used index value we have signalled on */
 153         u16 signalled_used;
 154
 155         /* Last used index value we have signalled on */
 156         bool signalled_used_valid;
 157
 158         /* Log writes to used structure. */
 159         bool log_used;
 160         u64 log_addr;
 161
 162         struct iovec iov[UIO_MAXIOV];
 163         struct iovec iotlb_iov[64];
 164         struct iovec *indirect;
 165         struct vring_used_elem *heads;
 166         /* Protected by virtqueue mutex. */
 167         struct vhost_umem *umem;
 168         struct vhost_umem *iotlb;
 169         void *private_data;
 170         u64 acked_features;
 171         u64 acked_backend_features;
 172         /* Log write descriptors */
 173         void __user *log_base;
 174         struct vhost_log *log;
 175
 176         /* Ring endianness. Defaults to legacy native endianness.
 177          * Set to true when starting a modern virtio device. */
 178         bool is_le;
 179 #ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
 180         /* Ring endianness requested by userspace for cross-endian support. */
 181         bool user_be;
 182 #endif
 183         u32 busyloop_timeout;
 184         spinlock_t mmu_lock;
 185         int invalidate_count;
 186 };
 187
 188 struct vhost_msg_node {
 189   union {
 190           struct vhost_msg msg;
 191           struct vhost_msg_v2 msg_v2;
 192   };
 193   struct vhost_virtqueue *vq;
 194   struct list_head node;
 195 };
 196
 197 struct vhost_dev {
 198         struct mm_struct *mm;
 199 #ifdef CONFIG_MMU_NOTIFIER
 200         struct mmu_notifier mmu_notifier;
 201 #endif
 202         struct mutex mutex;
 203         struct vhost_virtqueue **vqs;
 204         int nvqs;
 205         struct eventfd_ctx *log_ctx;
 206         struct llist_head work_list;
 207         struct task_struct *worker;
 208         struct vhost_umem *umem;
 209         struct vhost_umem *iotlb;
 210         spinlock_t iotlb_lock;
 211         struct list_head read_list;
 212         struct list_head pending_list;
 213         wait_queue_head_t wait;
 214         int iov_limit;
 215         int weight;
 216         int byte_weight;
 217 };
 218
 219 bool vhost_exceeds_weight(struct vhost_virtqueue *vq, int pkts, int total_len);
 220 void vhost_dev_init(struct vhost_dev *, struct vhost_virtqueue **vqs,
 221                     int nvqs, int iov_limit, int weight, int byte_weight);
 222 long vhost_dev_set_owner(struct vhost_dev *dev);
 223 bool vhost_dev_has_owner(struct vhost_dev *dev);
 224 long vhost_dev_check_owner(struct vhost_dev *);
 225 struct vhost_umem *vhost_dev_reset_owner_prepare(void);
 226 void vhost_dev_reset_owner(struct vhost_dev *, struct vhost_umem *);
 227 void vhost_dev_cleanup(struct vhost_dev *);
 228 void vhost_dev_stop(struct vhost_dev *);
 229 long vhost_dev_ioctl(struct vhost_dev *, unsigned int ioctl, void __user *argp);
 230 long vhost_vring_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp);
 231 bool vhost_vq_access_ok(struct vhost_virtqueue *vq);
 232 bool vhost_log_access_ok(struct vhost_dev *);
 233
 234 int vhost_get_vq_desc(struct vhost_virtqueue *,
 235                       struct iovec iov[], unsigned int iov_count,
 236                       unsigned int *out_num, unsigned int *in_num,
 237                       struct vhost_log *log, unsigned int *log_num);
 238 void vhost_discard_vq_desc(struct vhost_virtqueue *, int n);
 239
 240 int vhost_vq_init_access(struct vhost_virtqueue *);
 241 int vhost_add_used(struct vhost_virtqueue *, unsigned int head, int len);
 242 int vhost_add_used_n(struct vhost_virtqueue *, struct vring_used_elem *heads,
 243                      unsigned count);
 244 void vhost_add_used_and_signal(struct vhost_dev *, struct vhost_virtqueue *,
 245                                unsigned int id, int len);
 246 void vhost_add_used_and_signal_n(struct vhost_dev *, struct vhost_virtqueue *,
 247                                struct vring_used_elem *heads, unsigned count);
 248 void vhost_signal(struct vhost_dev *, struct vhost_virtqueue *);
 249 void vhost_disable_notify(struct vhost_dev *, struct vhost_virtqueue *);
 250 bool vhost_vq_avail_empty(struct vhost_dev *, struct vhost_virtqueue *);
 251 bool vhost_enable_notify(struct vhost_dev *, struct vhost_virtqueue *);
 252
 253 int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
 254                     unsigned int log_num, u64 len,
 255                     struct iovec *iov, int count);
 256 int vq_meta_prefetch(struct vhost_virtqueue *vq);
 257
 258 struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type);
 259 void vhost_enqueue_msg(struct vhost_dev *dev,
 260                        struct list_head *head,
 261                        struct vhost_msg_node *node);
 262 struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev,
 263                                          struct list_head *head);
 264 __poll_t vhost_chr_poll(struct file *file, struct vhost_dev *dev,
 265                             poll_table *wait);
 266 ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to,
 267                             int noblock);
 268 ssize_t vhost_chr_write_iter(struct vhost_dev *dev,
 269                              struct iov_iter *from);
 270 int vhost_init_device_iotlb(struct vhost_dev *d, bool enabled);
 271
 272 #define vq_err(vq, fmt, ...) do {                                  \
 273                 pr_debug(pr_fmt(fmt), ##__VA_ARGS__);       \
 274                 if ((vq)->error_ctx)                               \
 275                                 eventfd_signal((vq)->error_ctx, 1);\
 276         } while (0)
 277
 278 enum {
 279         VHOST_FEATURES = (1ULL << VIRTIO_F_NOTIFY_ON_EMPTY) |
 280                          (1ULL << VIRTIO_RING_F_INDIRECT_DESC) |
 281                          (1ULL << VIRTIO_RING_F_EVENT_IDX) |
 282                          (1ULL << VHOST_F_LOG_ALL) |
 283                          (1ULL << VIRTIO_F_ANY_LAYOUT) |
 284                          (1ULL << VIRTIO_F_VERSION_1)
 285 };
 286
 287 static inline bool vhost_has_feature(struct vhost_virtqueue *vq, int bit)
 288 {
 289         return vq->acked_features & (1ULL << bit);
 290 }
 291
 292 static inline bool vhost_backend_has_feature(struct vhost_virtqueue *vq, int bit)
 293 {
 294         return vq->acked_backend_features & (1ULL << bit);
 295 }
 296
 297 #ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
 298 static inline bool vhost_is_little_endian(struct vhost_virtqueue *vq)
 299 {
 300         return vq->is_le;
 301 }
 302 #else
 303 static inline bool vhost_is_little_endian(struct vhost_virtqueue *vq)
 304 {
 305         return virtio_legacy_is_little_endian() || vq->is_le;
 306 }
 307 #endif
 308
 309 /* Memory accessors */
 310 static inline u16 vhost16_to_cpu(struct vhost_virtqueue *vq, __virtio16 val)
 311 {
 312         return __virtio16_to_cpu(vhost_is_little_endian(vq), val);
 313 }
 314
 315 static inline __virtio16 cpu_to_vhost16(struct vhost_virtqueue *vq, u16 val)
 316 {
 317         return __cpu_to_virtio16(vhost_is_little_endian(vq), val);
 318 }
 319
 320 static inline u32 vhost32_to_cpu(struct vhost_virtqueue *vq, __virtio32 val)
 321 {
 322         return __virtio32_to_cpu(vhost_is_little_endian(vq), val);
 323 }
 324
 325 static inline __virtio32 cpu_to_vhost32(struct vhost_virtqueue *vq, u32 val)
 326 {
 327         return __cpu_to_virtio32(vhost_is_little_endian(vq), val);
 328 }
 329
 330 static inline u64 vhost64_to_cpu(struct vhost_virtqueue *vq, __virtio64 val)
 331 {
 332         return __virtio64_to_cpu(vhost_is_little_endian(vq), val);
 333 }
 334
 335 static inline __virtio64 cpu_to_vhost64(struct vhost_virtqueue *vq, u64 val)
 336 {
 337         return __cpu_to_virtio64(vhost_is_little_endian(vq), val);
 338 }
 339 #endif