2 * Helpers for the host side of a virtio ring.
4 * Since these may be in userspace, we use (inline) accessors.
6 #include <linux/compiler.h>
7 #include <linux/module.h>
8 #include <linux/vringh.h>
9 #include <linux/virtio_ring.h>
10 #include <linux/kernel.h>
11 #include <linux/ratelimit.h>
12 #include <linux/uaccess.h>
13 #include <linux/slab.h>
14 #include <linux/export.h>
15 #include <uapi/linux/virtio_config.h>
17 static __printf(1,2) __cold void vringh_bad(const char *fmt, ...)
19 static DEFINE_RATELIMIT_STATE(vringh_rs,
20 DEFAULT_RATELIMIT_INTERVAL,
21 DEFAULT_RATELIMIT_BURST);
22 if (__ratelimit(&vringh_rs)) {
25 printk(KERN_NOTICE "vringh:");
31 /* Returns vring->num if empty, -ve on error. */
32 static inline int __vringh_get_head(const struct vringh *vrh,
33 int (*getu16)(const struct vringh *vrh,
34 u16 *val, const __virtio16 *p),
37 u16 avail_idx, i, head;
40 err = getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
42 vringh_bad("Failed to access avail idx at %p",
43 &vrh->vring.avail->idx);
47 if (*last_avail_idx == avail_idx)
48 return vrh->vring.num;
50 /* Only get avail ring entries after they have been exposed by guest. */
51 virtio_rmb(vrh->weak_barriers);
53 i = *last_avail_idx & (vrh->vring.num - 1);
55 err = getu16(vrh, &head, &vrh->vring.avail->ring[i]);
57 vringh_bad("Failed to read head: idx %d address %p",
58 *last_avail_idx, &vrh->vring.avail->ring[i]);
62 if (head >= vrh->vring.num) {
63 vringh_bad("Guest says index %u > %u is available",
64 head, vrh->vring.num);
72 /* Copy some bytes to/from the iovec. Returns num copied. */
73 static inline ssize_t vringh_iov_xfer(struct vringh_kiov *iov,
74 void *ptr, size_t len,
75 int (*xfer)(void *addr, void *ptr,
80 while (len && iov->i < iov->used) {
83 partlen = min(iov->iov[iov->i].iov_len, len);
84 err = xfer(iov->iov[iov->i].iov_base, ptr, partlen);
90 iov->consumed += partlen;
91 iov->iov[iov->i].iov_len -= partlen;
92 iov->iov[iov->i].iov_base += partlen;
94 if (!iov->iov[iov->i].iov_len) {
95 /* Fix up old iov element then increment. */
96 iov->iov[iov->i].iov_len = iov->consumed;
97 iov->iov[iov->i].iov_base -= iov->consumed;
106 /* May reduce *len if range is shorter. */
107 static inline bool range_check(struct vringh *vrh, u64 addr, size_t *len,
108 struct vringh_range *range,
109 bool (*getrange)(struct vringh *,
110 u64, struct vringh_range *))
112 if (addr < range->start || addr > range->end_incl) {
113 if (!getrange(vrh, addr, range))
116 BUG_ON(addr < range->start || addr > range->end_incl);
118 /* To end of memory? */
119 if (unlikely(addr + *len == 0)) {
120 if (range->end_incl == -1ULL)
125 /* Otherwise, don't wrap. */
126 if (addr + *len < addr) {
127 vringh_bad("Wrapping descriptor %zu@0x%llx",
128 *len, (unsigned long long)addr);
132 if (unlikely(addr + *len - 1 > range->end_incl))
137 *len = range->end_incl + 1 - addr;
141 static inline bool no_range_check(struct vringh *vrh, u64 addr, size_t *len,
142 struct vringh_range *range,
143 bool (*getrange)(struct vringh *,
144 u64, struct vringh_range *))
149 /* No reason for this code to be inline. */
150 static int move_to_indirect(const struct vringh *vrh,
151 int *up_next, u16 *i, void *addr,
152 const struct vring_desc *desc,
153 struct vring_desc **descs, int *desc_max)
157 /* Indirect tables can't have indirect. */
158 if (*up_next != -1) {
159 vringh_bad("Multilevel indirect %u->%u", *up_next, *i);
163 len = vringh32_to_cpu(vrh, desc->len);
164 if (unlikely(len % sizeof(struct vring_desc))) {
165 vringh_bad("Strange indirect len %u", desc->len);
169 /* We will check this when we follow it! */
170 if (desc->flags & cpu_to_vringh16(vrh, VRING_DESC_F_NEXT))
171 *up_next = vringh16_to_cpu(vrh, desc->next);
175 *desc_max = len / sizeof(struct vring_desc);
177 /* Now, start at the first indirect. */
182 static int resize_iovec(struct vringh_kiov *iov, gfp_t gfp)
185 unsigned int flag, new_num = (iov->max_num & ~VRINGH_IOV_ALLOCATED) * 2;
190 flag = (iov->max_num & VRINGH_IOV_ALLOCATED);
192 new = krealloc(iov->iov, new_num * sizeof(struct iovec), gfp);
194 new = kmalloc_array(new_num, sizeof(struct iovec), gfp);
196 memcpy(new, iov->iov,
197 iov->max_num * sizeof(struct iovec));
198 flag = VRINGH_IOV_ALLOCATED;
204 iov->max_num = (new_num | flag);
208 static u16 __cold return_from_indirect(const struct vringh *vrh, int *up_next,
209 struct vring_desc **descs, int *desc_max)
214 *descs = vrh->vring.desc;
215 *desc_max = vrh->vring.num;
219 static int slow_copy(struct vringh *vrh, void *dst, const void *src,
220 bool (*rcheck)(struct vringh *vrh, u64 addr, size_t *len,
221 struct vringh_range *range,
222 bool (*getrange)(struct vringh *vrh,
224 struct vringh_range *)),
225 bool (*getrange)(struct vringh *vrh,
227 struct vringh_range *r),
228 struct vringh_range *range,
229 int (*copy)(void *dst, const void *src, size_t len))
231 size_t part, len = sizeof(struct vring_desc);
238 addr = (u64)(unsigned long)src - range->offset;
240 if (!rcheck(vrh, addr, &part, range, getrange))
243 err = copy(dst, src, part);
255 __vringh_iov(struct vringh *vrh, u16 i,
256 struct vringh_kiov *riov,
257 struct vringh_kiov *wiov,
258 bool (*rcheck)(struct vringh *vrh, u64 addr, size_t *len,
259 struct vringh_range *range,
260 bool (*getrange)(struct vringh *, u64,
261 struct vringh_range *)),
262 bool (*getrange)(struct vringh *, u64, struct vringh_range *),
264 int (*copy)(void *dst, const void *src, size_t len))
266 int err, count = 0, indirect_count = 0, up_next, desc_max;
267 struct vring_desc desc, *descs;
268 struct vringh_range range = { -1ULL, 0 }, slowrange;
271 /* We start traversing vring's descriptor table. */
272 descs = vrh->vring.desc;
273 desc_max = vrh->vring.num;
276 /* You must want something! */
277 if (WARN_ON(!riov && !wiov))
281 riov->i = riov->used = 0;
283 wiov->i = wiov->used = 0;
287 struct vringh_kiov *iov;
291 err = slow_copy(vrh, &desc, &descs[i], rcheck, getrange,
294 err = copy(&desc, &descs[i], sizeof(desc));
298 if (unlikely(desc.flags &
299 cpu_to_vringh16(vrh, VRING_DESC_F_INDIRECT))) {
300 u64 a = vringh64_to_cpu(vrh, desc.addr);
302 /* Make sure it's OK, and get offset. */
303 len = vringh32_to_cpu(vrh, desc.len);
304 if (!rcheck(vrh, a, &len, &range, getrange)) {
309 if (unlikely(len != vringh32_to_cpu(vrh, desc.len))) {
311 /* We need to save this range to use offset */
315 addr = (void *)(long)(a + range.offset);
316 err = move_to_indirect(vrh, &up_next, &i, addr, &desc,
328 if (count > vrh->vring.num || indirect_count > desc_max) {
329 vringh_bad("Descriptor loop in %p", descs);
334 if (desc.flags & cpu_to_vringh16(vrh, VRING_DESC_F_WRITE))
338 if (unlikely(wiov && wiov->used)) {
339 vringh_bad("Readable desc %p after writable",
347 vringh_bad("Unexpected %s desc",
348 !wiov ? "writable" : "readable");
354 /* Make sure it's OK, and get offset. */
355 len = vringh32_to_cpu(vrh, desc.len);
356 if (!rcheck(vrh, vringh64_to_cpu(vrh, desc.addr), &len, &range,
361 addr = (void *)(unsigned long)(vringh64_to_cpu(vrh, desc.addr) +
364 if (unlikely(iov->used == (iov->max_num & ~VRINGH_IOV_ALLOCATED))) {
365 err = resize_iovec(iov, gfp);
370 iov->iov[iov->used].iov_base = addr;
371 iov->iov[iov->used].iov_len = len;
374 if (unlikely(len != vringh32_to_cpu(vrh, desc.len))) {
375 desc.len = cpu_to_vringh32(vrh,
376 vringh32_to_cpu(vrh, desc.len) - len);
377 desc.addr = cpu_to_vringh64(vrh,
378 vringh64_to_cpu(vrh, desc.addr) + len);
382 if (desc.flags & cpu_to_vringh16(vrh, VRING_DESC_F_NEXT)) {
383 i = vringh16_to_cpu(vrh, desc.next);
385 /* Just in case we need to finish traversing above. */
386 if (unlikely(up_next > 0)) {
387 i = return_from_indirect(vrh, &up_next,
396 vringh_bad("Chained index %u > %u", i, desc_max);
408 static inline int __vringh_complete(struct vringh *vrh,
409 const struct vring_used_elem *used,
410 unsigned int num_used,
411 int (*putu16)(const struct vringh *vrh,
412 __virtio16 *p, u16 val),
413 int (*putused)(struct vring_used_elem *dst,
414 const struct vring_used_elem
417 struct vring_used *used_ring;
421 used_ring = vrh->vring.used;
422 used_idx = vrh->last_used_idx + vrh->completed;
424 off = used_idx % vrh->vring.num;
426 /* Compiler knows num_used == 1 sometimes, hence extra check */
427 if (num_used > 1 && unlikely(off + num_used >= vrh->vring.num)) {
428 u16 part = vrh->vring.num - off;
429 err = putused(&used_ring->ring[off], used, part);
431 err = putused(&used_ring->ring[0], used + part,
434 err = putused(&used_ring->ring[off], used, num_used);
437 vringh_bad("Failed to write %u used entries %u at %p",
438 num_used, off, &used_ring->ring[off]);
442 /* Make sure buffer is written before we update index. */
443 virtio_wmb(vrh->weak_barriers);
445 err = putu16(vrh, &vrh->vring.used->idx, used_idx + num_used);
447 vringh_bad("Failed to update used index at %p",
448 &vrh->vring.used->idx);
452 vrh->completed += num_used;
457 static inline int __vringh_need_notify(struct vringh *vrh,
458 int (*getu16)(const struct vringh *vrh,
460 const __virtio16 *p))
466 /* Flush out used index update. This is paired with the
467 * barrier that the Guest executes when enabling
469 virtio_mb(vrh->weak_barriers);
471 /* Old-style, without event indices. */
472 if (!vrh->event_indices) {
474 err = getu16(vrh, &flags, &vrh->vring.avail->flags);
476 vringh_bad("Failed to get flags at %p",
477 &vrh->vring.avail->flags);
480 return (!(flags & VRING_AVAIL_F_NO_INTERRUPT));
483 /* Modern: we know when other side wants to know. */
484 err = getu16(vrh, &used_event, &vring_used_event(&vrh->vring));
486 vringh_bad("Failed to get used event idx at %p",
487 &vring_used_event(&vrh->vring));
491 /* Just in case we added so many that we wrap. */
492 if (unlikely(vrh->completed > 0xffff))
495 notify = vring_need_event(used_event,
496 vrh->last_used_idx + vrh->completed,
499 vrh->last_used_idx += vrh->completed;
504 static inline bool __vringh_notify_enable(struct vringh *vrh,
505 int (*getu16)(const struct vringh *vrh,
506 u16 *val, const __virtio16 *p),
507 int (*putu16)(const struct vringh *vrh,
508 __virtio16 *p, u16 val))
512 if (!vrh->event_indices) {
513 /* Old-school; update flags. */
514 if (putu16(vrh, &vrh->vring.used->flags, 0) != 0) {
515 vringh_bad("Clearing used flags %p",
516 &vrh->vring.used->flags);
520 if (putu16(vrh, &vring_avail_event(&vrh->vring),
521 vrh->last_avail_idx) != 0) {
522 vringh_bad("Updating avail event index %p",
523 &vring_avail_event(&vrh->vring));
528 /* They could have slipped one in as we were doing that: make
529 * sure it's written, then check again. */
530 virtio_mb(vrh->weak_barriers);
532 if (getu16(vrh, &avail, &vrh->vring.avail->idx) != 0) {
533 vringh_bad("Failed to check avail idx at %p",
534 &vrh->vring.avail->idx);
538 /* This is unlikely, so we just leave notifications enabled
539 * (if we're using event_indices, we'll only get one
540 * notification anyway). */
541 return avail == vrh->last_avail_idx;
544 static inline void __vringh_notify_disable(struct vringh *vrh,
545 int (*putu16)(const struct vringh *vrh,
546 __virtio16 *p, u16 val))
548 if (!vrh->event_indices) {
549 /* Old-school; update flags. */
550 if (putu16(vrh, &vrh->vring.used->flags,
551 VRING_USED_F_NO_NOTIFY)) {
552 vringh_bad("Setting used flags %p",
553 &vrh->vring.used->flags);
558 /* Userspace access helpers: in this case, addresses are really userspace. */
559 static inline int getu16_user(const struct vringh *vrh, u16 *val, const __virtio16 *p)
562 int rc = get_user(v, (__force __virtio16 __user *)p);
563 *val = vringh16_to_cpu(vrh, v);
567 static inline int putu16_user(const struct vringh *vrh, __virtio16 *p, u16 val)
569 __virtio16 v = cpu_to_vringh16(vrh, val);
570 return put_user(v, (__force __virtio16 __user *)p);
573 static inline int copydesc_user(void *dst, const void *src, size_t len)
575 return copy_from_user(dst, (__force void __user *)src, len) ?
579 static inline int putused_user(struct vring_used_elem *dst,
580 const struct vring_used_elem *src,
583 return copy_to_user((__force void __user *)dst, src,
584 sizeof(*dst) * num) ? -EFAULT : 0;
587 static inline int xfer_from_user(void *src, void *dst, size_t len)
589 return copy_from_user(dst, (__force void __user *)src, len) ?
593 static inline int xfer_to_user(void *dst, void *src, size_t len)
595 return copy_to_user((__force void __user *)dst, src, len) ?
600 * vringh_init_user - initialize a vringh for a userspace vring.
601 * @vrh: the vringh to initialize.
602 * @features: the feature bits for this ring.
603 * @num: the number of elements.
604 * @weak_barriers: true if we only need memory barriers, not I/O.
605 * @desc: the userpace descriptor pointer.
606 * @avail: the userpace avail pointer.
607 * @used: the userpace used pointer.
609 * Returns an error if num is invalid: you should check pointers
612 int vringh_init_user(struct vringh *vrh, u64 features,
613 unsigned int num, bool weak_barriers,
614 struct vring_desc __user *desc,
615 struct vring_avail __user *avail,
616 struct vring_used __user *used)
618 /* Sane power of 2 please! */
619 if (!num || num > 0xffff || (num & (num - 1))) {
620 vringh_bad("Bad ring size %u", num);
624 vrh->little_endian = (features & (1ULL << VIRTIO_F_VERSION_1));
625 vrh->event_indices = (features & (1 << VIRTIO_RING_F_EVENT_IDX));
626 vrh->weak_barriers = weak_barriers;
628 vrh->last_avail_idx = 0;
629 vrh->last_used_idx = 0;
630 vrh->vring.num = num;
631 /* vring expects kernel addresses, but only used via accessors. */
632 vrh->vring.desc = (__force struct vring_desc *)desc;
633 vrh->vring.avail = (__force struct vring_avail *)avail;
634 vrh->vring.used = (__force struct vring_used *)used;
637 EXPORT_SYMBOL(vringh_init_user);
640 * vringh_getdesc_user - get next available descriptor from userspace ring.
641 * @vrh: the userspace vring.
642 * @riov: where to put the readable descriptors (or NULL)
643 * @wiov: where to put the writable descriptors (or NULL)
644 * @getrange: function to call to check ranges.
645 * @head: head index we received, for passing to vringh_complete_user().
647 * Returns 0 if there was no descriptor, 1 if there was, or -errno.
649 * Note that on error return, you can tell the difference between an
650 * invalid ring and a single invalid descriptor: in the former case,
651 * *head will be vrh->vring.num. You may be able to ignore an invalid
652 * descriptor, but there's not much you can do with an invalid ring.
654 * Note that you may need to clean up riov and wiov, even on error!
656 int vringh_getdesc_user(struct vringh *vrh,
657 struct vringh_iov *riov,
658 struct vringh_iov *wiov,
659 bool (*getrange)(struct vringh *vrh,
660 u64 addr, struct vringh_range *r),
665 *head = vrh->vring.num;
666 err = __vringh_get_head(vrh, getu16_user, &vrh->last_avail_idx);
671 if (err == vrh->vring.num)
674 /* We need the layouts to be the identical for this to work */
675 BUILD_BUG_ON(sizeof(struct vringh_kiov) != sizeof(struct vringh_iov));
676 BUILD_BUG_ON(offsetof(struct vringh_kiov, iov) !=
677 offsetof(struct vringh_iov, iov));
678 BUILD_BUG_ON(offsetof(struct vringh_kiov, i) !=
679 offsetof(struct vringh_iov, i));
680 BUILD_BUG_ON(offsetof(struct vringh_kiov, used) !=
681 offsetof(struct vringh_iov, used));
682 BUILD_BUG_ON(offsetof(struct vringh_kiov, max_num) !=
683 offsetof(struct vringh_iov, max_num));
684 BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
685 BUILD_BUG_ON(offsetof(struct iovec, iov_base) !=
686 offsetof(struct kvec, iov_base));
687 BUILD_BUG_ON(offsetof(struct iovec, iov_len) !=
688 offsetof(struct kvec, iov_len));
689 BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_base)
690 != sizeof(((struct kvec *)NULL)->iov_base));
691 BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_len)
692 != sizeof(((struct kvec *)NULL)->iov_len));
695 err = __vringh_iov(vrh, *head, (struct vringh_kiov *)riov,
696 (struct vringh_kiov *)wiov,
697 range_check, getrange, GFP_KERNEL, copydesc_user);
703 EXPORT_SYMBOL(vringh_getdesc_user);
706 * vringh_iov_pull_user - copy bytes from vring_iov.
707 * @riov: the riov as passed to vringh_getdesc_user() (updated as we consume)
708 * @dst: the place to copy.
709 * @len: the maximum length to copy.
711 * Returns the bytes copied <= len or a negative errno.
713 ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
715 return vringh_iov_xfer((struct vringh_kiov *)riov,
716 dst, len, xfer_from_user);
718 EXPORT_SYMBOL(vringh_iov_pull_user);
721 * vringh_iov_push_user - copy bytes into vring_iov.
722 * @wiov: the wiov as passed to vringh_getdesc_user() (updated as we consume)
723 * @dst: the place to copy.
724 * @len: the maximum length to copy.
726 * Returns the bytes copied <= len or a negative errno.
728 ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
729 const void *src, size_t len)
731 return vringh_iov_xfer((struct vringh_kiov *)wiov,
732 (void *)src, len, xfer_to_user);
734 EXPORT_SYMBOL(vringh_iov_push_user);
737 * vringh_abandon_user - we've decided not to handle the descriptor(s).
739 * @num: the number of descriptors to put back (ie. num
740 * vringh_get_user() to undo).
742 * The next vringh_get_user() will return the old descriptor(s) again.
744 void vringh_abandon_user(struct vringh *vrh, unsigned int num)
746 /* We only update vring_avail_event(vr) when we want to be notified,
747 * so we haven't changed that yet. */
748 vrh->last_avail_idx -= num;
750 EXPORT_SYMBOL(vringh_abandon_user);
753 * vringh_complete_user - we've finished with descriptor, publish it.
755 * @head: the head as filled in by vringh_getdesc_user.
756 * @len: the length of data we have written.
758 * You should check vringh_need_notify_user() after one or more calls
761 int vringh_complete_user(struct vringh *vrh, u16 head, u32 len)
763 struct vring_used_elem used;
765 used.id = cpu_to_vringh32(vrh, head);
766 used.len = cpu_to_vringh32(vrh, len);
767 return __vringh_complete(vrh, &used, 1, putu16_user, putused_user);
769 EXPORT_SYMBOL(vringh_complete_user);
772 * vringh_complete_multi_user - we've finished with many descriptors.
774 * @used: the head, length pairs.
775 * @num_used: the number of used elements.
777 * You should check vringh_need_notify_user() after one or more calls
780 int vringh_complete_multi_user(struct vringh *vrh,
781 const struct vring_used_elem used[],
784 return __vringh_complete(vrh, used, num_used,
785 putu16_user, putused_user);
787 EXPORT_SYMBOL(vringh_complete_multi_user);
790 * vringh_notify_enable_user - we want to know if something changes.
793 * This always enables notifications, but returns false if there are
794 * now more buffers available in the vring.
796 bool vringh_notify_enable_user(struct vringh *vrh)
798 return __vringh_notify_enable(vrh, getu16_user, putu16_user);
800 EXPORT_SYMBOL(vringh_notify_enable_user);
803 * vringh_notify_disable_user - don't tell us if something changes.
806 * This is our normal running state: we disable and then only enable when
807 * we're going to sleep.
809 void vringh_notify_disable_user(struct vringh *vrh)
811 __vringh_notify_disable(vrh, putu16_user);
813 EXPORT_SYMBOL(vringh_notify_disable_user);
816 * vringh_need_notify_user - must we tell the other side about used buffers?
817 * @vrh: the vring we've called vringh_complete_user() on.
819 * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
821 int vringh_need_notify_user(struct vringh *vrh)
823 return __vringh_need_notify(vrh, getu16_user);
825 EXPORT_SYMBOL(vringh_need_notify_user);
827 /* Kernelspace access helpers. */
828 static inline int getu16_kern(const struct vringh *vrh,
829 u16 *val, const __virtio16 *p)
831 *val = vringh16_to_cpu(vrh, READ_ONCE(*p));
835 static inline int putu16_kern(const struct vringh *vrh, __virtio16 *p, u16 val)
837 WRITE_ONCE(*p, cpu_to_vringh16(vrh, val));
841 static inline int copydesc_kern(void *dst, const void *src, size_t len)
843 memcpy(dst, src, len);
847 static inline int putused_kern(struct vring_used_elem *dst,
848 const struct vring_used_elem *src,
851 memcpy(dst, src, num * sizeof(*dst));
855 static inline int xfer_kern(void *src, void *dst, size_t len)
857 memcpy(dst, src, len);
862 * vringh_init_kern - initialize a vringh for a kernelspace vring.
863 * @vrh: the vringh to initialize.
864 * @features: the feature bits for this ring.
865 * @num: the number of elements.
866 * @weak_barriers: true if we only need memory barriers, not I/O.
867 * @desc: the userpace descriptor pointer.
868 * @avail: the userpace avail pointer.
869 * @used: the userpace used pointer.
871 * Returns an error if num is invalid.
873 int vringh_init_kern(struct vringh *vrh, u64 features,
874 unsigned int num, bool weak_barriers,
875 struct vring_desc *desc,
876 struct vring_avail *avail,
877 struct vring_used *used)
879 /* Sane power of 2 please! */
880 if (!num || num > 0xffff || (num & (num - 1))) {
881 vringh_bad("Bad ring size %u", num);
885 vrh->little_endian = (features & (1ULL << VIRTIO_F_VERSION_1));
886 vrh->event_indices = (features & (1 << VIRTIO_RING_F_EVENT_IDX));
887 vrh->weak_barriers = weak_barriers;
889 vrh->last_avail_idx = 0;
890 vrh->last_used_idx = 0;
891 vrh->vring.num = num;
892 vrh->vring.desc = desc;
893 vrh->vring.avail = avail;
894 vrh->vring.used = used;
897 EXPORT_SYMBOL(vringh_init_kern);
900 * vringh_getdesc_kern - get next available descriptor from kernelspace ring.
901 * @vrh: the kernelspace vring.
902 * @riov: where to put the readable descriptors (or NULL)
903 * @wiov: where to put the writable descriptors (or NULL)
904 * @head: head index we received, for passing to vringh_complete_kern().
905 * @gfp: flags for allocating larger riov/wiov.
907 * Returns 0 if there was no descriptor, 1 if there was, or -errno.
909 * Note that on error return, you can tell the difference between an
910 * invalid ring and a single invalid descriptor: in the former case,
911 * *head will be vrh->vring.num. You may be able to ignore an invalid
912 * descriptor, but there's not much you can do with an invalid ring.
914 * Note that you may need to clean up riov and wiov, even on error!
916 int vringh_getdesc_kern(struct vringh *vrh,
917 struct vringh_kiov *riov,
918 struct vringh_kiov *wiov,
924 err = __vringh_get_head(vrh, getu16_kern, &vrh->last_avail_idx);
929 if (err == vrh->vring.num)
933 err = __vringh_iov(vrh, *head, riov, wiov, no_range_check, NULL,
940 EXPORT_SYMBOL(vringh_getdesc_kern);
943 * vringh_iov_pull_kern - copy bytes from vring_iov.
944 * @riov: the riov as passed to vringh_getdesc_kern() (updated as we consume)
945 * @dst: the place to copy.
946 * @len: the maximum length to copy.
948 * Returns the bytes copied <= len or a negative errno.
950 ssize_t vringh_iov_pull_kern(struct vringh_kiov *riov, void *dst, size_t len)
952 return vringh_iov_xfer(riov, dst, len, xfer_kern);
954 EXPORT_SYMBOL(vringh_iov_pull_kern);
957 * vringh_iov_push_kern - copy bytes into vring_iov.
958 * @wiov: the wiov as passed to vringh_getdesc_kern() (updated as we consume)
959 * @dst: the place to copy.
960 * @len: the maximum length to copy.
962 * Returns the bytes copied <= len or a negative errno.
964 ssize_t vringh_iov_push_kern(struct vringh_kiov *wiov,
965 const void *src, size_t len)
967 return vringh_iov_xfer(wiov, (void *)src, len, xfer_kern);
969 EXPORT_SYMBOL(vringh_iov_push_kern);
972 * vringh_abandon_kern - we've decided not to handle the descriptor(s).
974 * @num: the number of descriptors to put back (ie. num
975 * vringh_get_kern() to undo).
977 * The next vringh_get_kern() will return the old descriptor(s) again.
979 void vringh_abandon_kern(struct vringh *vrh, unsigned int num)
981 /* We only update vring_avail_event(vr) when we want to be notified,
982 * so we haven't changed that yet. */
983 vrh->last_avail_idx -= num;
985 EXPORT_SYMBOL(vringh_abandon_kern);
988 * vringh_complete_kern - we've finished with descriptor, publish it.
990 * @head: the head as filled in by vringh_getdesc_kern.
991 * @len: the length of data we have written.
993 * You should check vringh_need_notify_kern() after one or more calls
996 int vringh_complete_kern(struct vringh *vrh, u16 head, u32 len)
998 struct vring_used_elem used;
1000 used.id = cpu_to_vringh32(vrh, head);
1001 used.len = cpu_to_vringh32(vrh, len);
1003 return __vringh_complete(vrh, &used, 1, putu16_kern, putused_kern);
1005 EXPORT_SYMBOL(vringh_complete_kern);
1008 * vringh_notify_enable_kern - we want to know if something changes.
1011 * This always enables notifications, but returns false if there are
1012 * now more buffers available in the vring.
1014 bool vringh_notify_enable_kern(struct vringh *vrh)
1016 return __vringh_notify_enable(vrh, getu16_kern, putu16_kern);
1018 EXPORT_SYMBOL(vringh_notify_enable_kern);
1021 * vringh_notify_disable_kern - don't tell us if something changes.
1024 * This is our normal running state: we disable and then only enable when
1025 * we're going to sleep.
1027 void vringh_notify_disable_kern(struct vringh *vrh)
1029 __vringh_notify_disable(vrh, putu16_kern);
1031 EXPORT_SYMBOL(vringh_notify_disable_kern);
1034 * vringh_need_notify_kern - must we tell the other side about used buffers?
1035 * @vrh: the vring we've called vringh_complete_kern() on.
1037 * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
1039 int vringh_need_notify_kern(struct vringh *vrh)
1041 return __vringh_need_notify(vrh, getu16_kern);
1043 EXPORT_SYMBOL(vringh_need_notify_kern);
1045 MODULE_LICENSE("GPL");