1 // SPDX-License-Identifier: GPL-2.0-only
3 * Helpers for the host side of a virtio ring.
5 * Since these may be in userspace, we use (inline) accessors.
7 #include <linux/compiler.h>
8 #include <linux/module.h>
9 #include <linux/vringh.h>
10 #include <linux/virtio_ring.h>
11 #include <linux/kernel.h>
12 #include <linux/ratelimit.h>
13 #include <linux/uaccess.h>
14 #include <linux/slab.h>
15 #include <linux/export.h>
16 #if IS_REACHABLE(CONFIG_VHOST_IOTLB)
17 #include <linux/bvec.h>
18 #include <linux/highmem.h>
19 #include <linux/vhost_iotlb.h>
21 #include <uapi/linux/virtio_config.h>
23 static __printf(1,2) __cold void vringh_bad(const char *fmt, ...)
25 static DEFINE_RATELIMIT_STATE(vringh_rs,
26 DEFAULT_RATELIMIT_INTERVAL,
27 DEFAULT_RATELIMIT_BURST);
28 if (__ratelimit(&vringh_rs)) {
31 printk(KERN_NOTICE "vringh:");
37 /* Returns vring->num if empty, -ve on error. */
38 static inline int __vringh_get_head(const struct vringh *vrh,
39 int (*getu16)(const struct vringh *vrh,
40 u16 *val, const __virtio16 *p),
43 u16 avail_idx, i, head;
46 err = getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
48 vringh_bad("Failed to access avail idx at %p",
49 &vrh->vring.avail->idx);
53 if (*last_avail_idx == avail_idx)
54 return vrh->vring.num;
56 /* Only get avail ring entries after they have been exposed by guest. */
57 virtio_rmb(vrh->weak_barriers);
59 i = *last_avail_idx & (vrh->vring.num - 1);
61 err = getu16(vrh, &head, &vrh->vring.avail->ring[i]);
63 vringh_bad("Failed to read head: idx %d address %p",
64 *last_avail_idx, &vrh->vring.avail->ring[i]);
68 if (head >= vrh->vring.num) {
69 vringh_bad("Guest says index %u > %u is available",
70 head, vrh->vring.num);
79 * vringh_kiov_advance - skip bytes from vring_kiov
80 * @iov: an iov passed to vringh_getdesc_*() (updated as we consume)
81 * @len: the maximum length to advance
83 void vringh_kiov_advance(struct vringh_kiov *iov, size_t len)
85 while (len && iov->i < iov->used) {
86 size_t partlen = min(iov->iov[iov->i].iov_len, len);
88 iov->consumed += partlen;
89 iov->iov[iov->i].iov_len -= partlen;
90 iov->iov[iov->i].iov_base += partlen;
92 if (!iov->iov[iov->i].iov_len) {
93 /* Fix up old iov element then increment. */
94 iov->iov[iov->i].iov_len = iov->consumed;
95 iov->iov[iov->i].iov_base -= iov->consumed;
104 EXPORT_SYMBOL(vringh_kiov_advance);
106 /* Copy some bytes to/from the iovec. Returns num copied. */
107 static inline ssize_t vringh_iov_xfer(struct vringh *vrh,
108 struct vringh_kiov *iov,
109 void *ptr, size_t len,
110 int (*xfer)(const struct vringh *vrh,
111 void *addr, void *ptr,
116 while (len && iov->i < iov->used) {
119 partlen = min(iov->iov[iov->i].iov_len, len);
120 err = xfer(vrh, iov->iov[iov->i].iov_base, ptr, partlen);
127 vringh_kiov_advance(iov, partlen);
132 /* May reduce *len if range is shorter. */
133 static inline bool range_check(struct vringh *vrh, u64 addr, size_t *len,
134 struct vringh_range *range,
135 bool (*getrange)(struct vringh *,
136 u64, struct vringh_range *))
138 if (addr < range->start || addr > range->end_incl) {
139 if (!getrange(vrh, addr, range))
142 BUG_ON(addr < range->start || addr > range->end_incl);
144 /* To end of memory? */
145 if (unlikely(addr + *len == 0)) {
146 if (range->end_incl == -1ULL)
151 /* Otherwise, don't wrap. */
152 if (addr + *len < addr) {
153 vringh_bad("Wrapping descriptor %zu@0x%llx",
154 *len, (unsigned long long)addr);
158 if (unlikely(addr + *len - 1 > range->end_incl))
163 *len = range->end_incl + 1 - addr;
167 static inline bool no_range_check(struct vringh *vrh, u64 addr, size_t *len,
168 struct vringh_range *range,
169 bool (*getrange)(struct vringh *,
170 u64, struct vringh_range *))
175 /* No reason for this code to be inline. */
176 static int move_to_indirect(const struct vringh *vrh,
177 int *up_next, u16 *i, void *addr,
178 const struct vring_desc *desc,
179 struct vring_desc **descs, int *desc_max)
183 /* Indirect tables can't have indirect. */
184 if (*up_next != -1) {
185 vringh_bad("Multilevel indirect %u->%u", *up_next, *i);
189 len = vringh32_to_cpu(vrh, desc->len);
190 if (unlikely(len % sizeof(struct vring_desc))) {
191 vringh_bad("Strange indirect len %u", desc->len);
195 /* We will check this when we follow it! */
196 if (desc->flags & cpu_to_vringh16(vrh, VRING_DESC_F_NEXT))
197 *up_next = vringh16_to_cpu(vrh, desc->next);
201 *desc_max = len / sizeof(struct vring_desc);
203 /* Now, start at the first indirect. */
208 static int resize_iovec(struct vringh_kiov *iov, gfp_t gfp)
211 unsigned int flag, new_num = (iov->max_num & ~VRINGH_IOV_ALLOCATED) * 2;
216 flag = (iov->max_num & VRINGH_IOV_ALLOCATED);
218 new = krealloc_array(iov->iov, new_num,
219 sizeof(struct iovec), gfp);
221 new = kmalloc_array(new_num, sizeof(struct iovec), gfp);
223 memcpy(new, iov->iov,
224 iov->max_num * sizeof(struct iovec));
225 flag = VRINGH_IOV_ALLOCATED;
231 iov->max_num = (new_num | flag);
235 static u16 __cold return_from_indirect(const struct vringh *vrh, int *up_next,
236 struct vring_desc **descs, int *desc_max)
241 *descs = vrh->vring.desc;
242 *desc_max = vrh->vring.num;
246 static int slow_copy(struct vringh *vrh, void *dst, const void *src,
247 bool (*rcheck)(struct vringh *vrh, u64 addr, size_t *len,
248 struct vringh_range *range,
249 bool (*getrange)(struct vringh *vrh,
251 struct vringh_range *)),
252 bool (*getrange)(struct vringh *vrh,
254 struct vringh_range *r),
255 struct vringh_range *range,
256 int (*copy)(const struct vringh *vrh,
257 void *dst, const void *src, size_t len))
259 size_t part, len = sizeof(struct vring_desc);
266 addr = (u64)(unsigned long)src - range->offset;
268 if (!rcheck(vrh, addr, &part, range, getrange))
271 err = copy(vrh, dst, src, part);
283 __vringh_iov(struct vringh *vrh, u16 i,
284 struct vringh_kiov *riov,
285 struct vringh_kiov *wiov,
286 bool (*rcheck)(struct vringh *vrh, u64 addr, size_t *len,
287 struct vringh_range *range,
288 bool (*getrange)(struct vringh *, u64,
289 struct vringh_range *)),
290 bool (*getrange)(struct vringh *, u64, struct vringh_range *),
292 int (*copy)(const struct vringh *vrh,
293 void *dst, const void *src, size_t len))
295 int err, count = 0, indirect_count = 0, up_next, desc_max;
296 struct vring_desc desc, *descs;
297 struct vringh_range range = { -1ULL, 0 }, slowrange;
300 /* We start traversing vring's descriptor table. */
301 descs = vrh->vring.desc;
302 desc_max = vrh->vring.num;
305 /* You must want something! */
306 if (WARN_ON(!riov && !wiov))
310 riov->i = riov->used = riov->consumed = 0;
312 wiov->i = wiov->used = wiov->consumed = 0;
316 struct vringh_kiov *iov;
320 err = slow_copy(vrh, &desc, &descs[i], rcheck, getrange,
323 err = copy(vrh, &desc, &descs[i], sizeof(desc));
327 if (unlikely(desc.flags &
328 cpu_to_vringh16(vrh, VRING_DESC_F_INDIRECT))) {
329 u64 a = vringh64_to_cpu(vrh, desc.addr);
331 /* Make sure it's OK, and get offset. */
332 len = vringh32_to_cpu(vrh, desc.len);
333 if (!rcheck(vrh, a, &len, &range, getrange)) {
338 if (unlikely(len != vringh32_to_cpu(vrh, desc.len))) {
340 /* We need to save this range to use offset */
344 addr = (void *)(long)(a + range.offset);
345 err = move_to_indirect(vrh, &up_next, &i, addr, &desc,
357 if (count > vrh->vring.num || indirect_count > desc_max) {
358 vringh_bad("Descriptor loop in %p", descs);
363 if (desc.flags & cpu_to_vringh16(vrh, VRING_DESC_F_WRITE))
367 if (unlikely(wiov && wiov->used)) {
368 vringh_bad("Readable desc %p after writable",
376 vringh_bad("Unexpected %s desc",
377 !wiov ? "writable" : "readable");
383 /* Make sure it's OK, and get offset. */
384 len = vringh32_to_cpu(vrh, desc.len);
385 if (!rcheck(vrh, vringh64_to_cpu(vrh, desc.addr), &len, &range,
390 addr = (void *)(unsigned long)(vringh64_to_cpu(vrh, desc.addr) +
393 if (unlikely(iov->used == (iov->max_num & ~VRINGH_IOV_ALLOCATED))) {
394 err = resize_iovec(iov, gfp);
399 iov->iov[iov->used].iov_base = addr;
400 iov->iov[iov->used].iov_len = len;
403 if (unlikely(len != vringh32_to_cpu(vrh, desc.len))) {
404 desc.len = cpu_to_vringh32(vrh,
405 vringh32_to_cpu(vrh, desc.len) - len);
406 desc.addr = cpu_to_vringh64(vrh,
407 vringh64_to_cpu(vrh, desc.addr) + len);
411 if (desc.flags & cpu_to_vringh16(vrh, VRING_DESC_F_NEXT)) {
412 i = vringh16_to_cpu(vrh, desc.next);
414 /* Just in case we need to finish traversing above. */
415 if (unlikely(up_next > 0)) {
416 i = return_from_indirect(vrh, &up_next,
425 vringh_bad("Chained index %u > %u", i, desc_max);
437 static inline int __vringh_complete(struct vringh *vrh,
438 const struct vring_used_elem *used,
439 unsigned int num_used,
440 int (*putu16)(const struct vringh *vrh,
441 __virtio16 *p, u16 val),
442 int (*putused)(const struct vringh *vrh,
443 struct vring_used_elem *dst,
444 const struct vring_used_elem
447 struct vring_used *used_ring;
451 used_ring = vrh->vring.used;
452 used_idx = vrh->last_used_idx + vrh->completed;
454 off = used_idx % vrh->vring.num;
456 /* Compiler knows num_used == 1 sometimes, hence extra check */
457 if (num_used > 1 && unlikely(off + num_used >= vrh->vring.num)) {
458 u16 part = vrh->vring.num - off;
459 err = putused(vrh, &used_ring->ring[off], used, part);
461 err = putused(vrh, &used_ring->ring[0], used + part,
464 err = putused(vrh, &used_ring->ring[off], used, num_used);
467 vringh_bad("Failed to write %u used entries %u at %p",
468 num_used, off, &used_ring->ring[off]);
472 /* Make sure buffer is written before we update index. */
473 virtio_wmb(vrh->weak_barriers);
475 err = putu16(vrh, &vrh->vring.used->idx, used_idx + num_used);
477 vringh_bad("Failed to update used index at %p",
478 &vrh->vring.used->idx);
482 vrh->completed += num_used;
487 static inline int __vringh_need_notify(struct vringh *vrh,
488 int (*getu16)(const struct vringh *vrh,
490 const __virtio16 *p))
496 /* Flush out used index update. This is paired with the
497 * barrier that the Guest executes when enabling
499 virtio_mb(vrh->weak_barriers);
501 /* Old-style, without event indices. */
502 if (!vrh->event_indices) {
504 err = getu16(vrh, &flags, &vrh->vring.avail->flags);
506 vringh_bad("Failed to get flags at %p",
507 &vrh->vring.avail->flags);
510 return (!(flags & VRING_AVAIL_F_NO_INTERRUPT));
513 /* Modern: we know when other side wants to know. */
514 err = getu16(vrh, &used_event, &vring_used_event(&vrh->vring));
516 vringh_bad("Failed to get used event idx at %p",
517 &vring_used_event(&vrh->vring));
521 /* Just in case we added so many that we wrap. */
522 if (unlikely(vrh->completed > 0xffff))
525 notify = vring_need_event(used_event,
526 vrh->last_used_idx + vrh->completed,
529 vrh->last_used_idx += vrh->completed;
534 static inline bool __vringh_notify_enable(struct vringh *vrh,
535 int (*getu16)(const struct vringh *vrh,
536 u16 *val, const __virtio16 *p),
537 int (*putu16)(const struct vringh *vrh,
538 __virtio16 *p, u16 val))
542 if (!vrh->event_indices) {
543 /* Old-school; update flags. */
544 if (putu16(vrh, &vrh->vring.used->flags, 0) != 0) {
545 vringh_bad("Clearing used flags %p",
546 &vrh->vring.used->flags);
550 if (putu16(vrh, &vring_avail_event(&vrh->vring),
551 vrh->last_avail_idx) != 0) {
552 vringh_bad("Updating avail event index %p",
553 &vring_avail_event(&vrh->vring));
558 /* They could have slipped one in as we were doing that: make
559 * sure it's written, then check again. */
560 virtio_mb(vrh->weak_barriers);
562 if (getu16(vrh, &avail, &vrh->vring.avail->idx) != 0) {
563 vringh_bad("Failed to check avail idx at %p",
564 &vrh->vring.avail->idx);
568 /* This is unlikely, so we just leave notifications enabled
569 * (if we're using event_indices, we'll only get one
570 * notification anyway). */
571 return avail == vrh->last_avail_idx;
574 static inline void __vringh_notify_disable(struct vringh *vrh,
575 int (*putu16)(const struct vringh *vrh,
576 __virtio16 *p, u16 val))
578 if (!vrh->event_indices) {
579 /* Old-school; update flags. */
580 if (putu16(vrh, &vrh->vring.used->flags,
581 VRING_USED_F_NO_NOTIFY)) {
582 vringh_bad("Setting used flags %p",
583 &vrh->vring.used->flags);
588 /* Userspace access helpers: in this case, addresses are really userspace. */
589 static inline int getu16_user(const struct vringh *vrh, u16 *val, const __virtio16 *p)
592 int rc = get_user(v, (__force __virtio16 __user *)p);
593 *val = vringh16_to_cpu(vrh, v);
597 static inline int putu16_user(const struct vringh *vrh, __virtio16 *p, u16 val)
599 __virtio16 v = cpu_to_vringh16(vrh, val);
600 return put_user(v, (__force __virtio16 __user *)p);
603 static inline int copydesc_user(const struct vringh *vrh,
604 void *dst, const void *src, size_t len)
606 return copy_from_user(dst, (__force void __user *)src, len) ?
610 static inline int putused_user(const struct vringh *vrh,
611 struct vring_used_elem *dst,
612 const struct vring_used_elem *src,
615 return copy_to_user((__force void __user *)dst, src,
616 sizeof(*dst) * num) ? -EFAULT : 0;
619 static inline int xfer_from_user(const struct vringh *vrh, void *src,
620 void *dst, size_t len)
622 return copy_from_user(dst, (__force void __user *)src, len) ?
626 static inline int xfer_to_user(const struct vringh *vrh,
627 void *dst, void *src, size_t len)
629 return copy_to_user((__force void __user *)dst, src, len) ?
634 * vringh_init_user - initialize a vringh for a userspace vring.
635 * @vrh: the vringh to initialize.
636 * @features: the feature bits for this ring.
637 * @num: the number of elements.
638 * @weak_barriers: true if we only need memory barriers, not I/O.
639 * @desc: the userpace descriptor pointer.
640 * @avail: the userpace avail pointer.
641 * @used: the userpace used pointer.
643 * Returns an error if num is invalid: you should check pointers
646 int vringh_init_user(struct vringh *vrh, u64 features,
647 unsigned int num, bool weak_barriers,
648 vring_desc_t __user *desc,
649 vring_avail_t __user *avail,
650 vring_used_t __user *used)
652 /* Sane power of 2 please! */
653 if (!num || num > 0xffff || (num & (num - 1))) {
654 vringh_bad("Bad ring size %u", num);
658 vrh->little_endian = (features & (1ULL << VIRTIO_F_VERSION_1));
659 vrh->event_indices = (features & (1 << VIRTIO_RING_F_EVENT_IDX));
660 vrh->weak_barriers = weak_barriers;
662 vrh->last_avail_idx = 0;
663 vrh->last_used_idx = 0;
664 vrh->vring.num = num;
665 /* vring expects kernel addresses, but only used via accessors. */
666 vrh->vring.desc = (__force struct vring_desc *)desc;
667 vrh->vring.avail = (__force struct vring_avail *)avail;
668 vrh->vring.used = (__force struct vring_used *)used;
671 EXPORT_SYMBOL(vringh_init_user);
674 * vringh_getdesc_user - get next available descriptor from userspace ring.
675 * @vrh: the userspace vring.
676 * @riov: where to put the readable descriptors (or NULL)
677 * @wiov: where to put the writable descriptors (or NULL)
678 * @getrange: function to call to check ranges.
679 * @head: head index we received, for passing to vringh_complete_user().
681 * Returns 0 if there was no descriptor, 1 if there was, or -errno.
683 * Note that on error return, you can tell the difference between an
684 * invalid ring and a single invalid descriptor: in the former case,
685 * *head will be vrh->vring.num. You may be able to ignore an invalid
686 * descriptor, but there's not much you can do with an invalid ring.
688 * Note that you can reuse riov and wiov with subsequent calls. Content is
689 * overwritten and memory reallocated if more space is needed.
690 * When you don't have to use riov and wiov anymore, you should clean up them
691 * calling vringh_iov_cleanup() to release the memory, even on error!
693 int vringh_getdesc_user(struct vringh *vrh,
694 struct vringh_iov *riov,
695 struct vringh_iov *wiov,
696 bool (*getrange)(struct vringh *vrh,
697 u64 addr, struct vringh_range *r),
702 *head = vrh->vring.num;
703 err = __vringh_get_head(vrh, getu16_user, &vrh->last_avail_idx);
708 if (err == vrh->vring.num)
711 /* We need the layouts to be the identical for this to work */
712 BUILD_BUG_ON(sizeof(struct vringh_kiov) != sizeof(struct vringh_iov));
713 BUILD_BUG_ON(offsetof(struct vringh_kiov, iov) !=
714 offsetof(struct vringh_iov, iov));
715 BUILD_BUG_ON(offsetof(struct vringh_kiov, i) !=
716 offsetof(struct vringh_iov, i));
717 BUILD_BUG_ON(offsetof(struct vringh_kiov, used) !=
718 offsetof(struct vringh_iov, used));
719 BUILD_BUG_ON(offsetof(struct vringh_kiov, max_num) !=
720 offsetof(struct vringh_iov, max_num));
721 BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
722 BUILD_BUG_ON(offsetof(struct iovec, iov_base) !=
723 offsetof(struct kvec, iov_base));
724 BUILD_BUG_ON(offsetof(struct iovec, iov_len) !=
725 offsetof(struct kvec, iov_len));
726 BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_base)
727 != sizeof(((struct kvec *)NULL)->iov_base));
728 BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_len)
729 != sizeof(((struct kvec *)NULL)->iov_len));
732 err = __vringh_iov(vrh, *head, (struct vringh_kiov *)riov,
733 (struct vringh_kiov *)wiov,
734 range_check, getrange, GFP_KERNEL, copydesc_user);
740 EXPORT_SYMBOL(vringh_getdesc_user);
743 * vringh_iov_pull_user - copy bytes from vring_iov.
744 * @riov: the riov as passed to vringh_getdesc_user() (updated as we consume)
745 * @dst: the place to copy.
746 * @len: the maximum length to copy.
748 * Returns the bytes copied <= len or a negative errno.
750 ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
752 return vringh_iov_xfer(NULL, (struct vringh_kiov *)riov,
753 dst, len, xfer_from_user);
755 EXPORT_SYMBOL(vringh_iov_pull_user);
758 * vringh_iov_push_user - copy bytes into vring_iov.
759 * @wiov: the wiov as passed to vringh_getdesc_user() (updated as we consume)
760 * @src: the place to copy from.
761 * @len: the maximum length to copy.
763 * Returns the bytes copied <= len or a negative errno.
765 ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
766 const void *src, size_t len)
768 return vringh_iov_xfer(NULL, (struct vringh_kiov *)wiov,
769 (void *)src, len, xfer_to_user);
771 EXPORT_SYMBOL(vringh_iov_push_user);
774 * vringh_abandon_user - we've decided not to handle the descriptor(s).
776 * @num: the number of descriptors to put back (ie. num
777 * vringh_get_user() to undo).
779 * The next vringh_get_user() will return the old descriptor(s) again.
781 void vringh_abandon_user(struct vringh *vrh, unsigned int num)
783 /* We only update vring_avail_event(vr) when we want to be notified,
784 * so we haven't changed that yet. */
785 vrh->last_avail_idx -= num;
787 EXPORT_SYMBOL(vringh_abandon_user);
790 * vringh_complete_user - we've finished with descriptor, publish it.
792 * @head: the head as filled in by vringh_getdesc_user.
793 * @len: the length of data we have written.
795 * You should check vringh_need_notify_user() after one or more calls
798 int vringh_complete_user(struct vringh *vrh, u16 head, u32 len)
800 struct vring_used_elem used;
802 used.id = cpu_to_vringh32(vrh, head);
803 used.len = cpu_to_vringh32(vrh, len);
804 return __vringh_complete(vrh, &used, 1, putu16_user, putused_user);
806 EXPORT_SYMBOL(vringh_complete_user);
809 * vringh_complete_multi_user - we've finished with many descriptors.
811 * @used: the head, length pairs.
812 * @num_used: the number of used elements.
814 * You should check vringh_need_notify_user() after one or more calls
817 int vringh_complete_multi_user(struct vringh *vrh,
818 const struct vring_used_elem used[],
821 return __vringh_complete(vrh, used, num_used,
822 putu16_user, putused_user);
824 EXPORT_SYMBOL(vringh_complete_multi_user);
827 * vringh_notify_enable_user - we want to know if something changes.
830 * This always enables notifications, but returns false if there are
831 * now more buffers available in the vring.
833 bool vringh_notify_enable_user(struct vringh *vrh)
835 return __vringh_notify_enable(vrh, getu16_user, putu16_user);
837 EXPORT_SYMBOL(vringh_notify_enable_user);
840 * vringh_notify_disable_user - don't tell us if something changes.
843 * This is our normal running state: we disable and then only enable when
844 * we're going to sleep.
846 void vringh_notify_disable_user(struct vringh *vrh)
848 __vringh_notify_disable(vrh, putu16_user);
850 EXPORT_SYMBOL(vringh_notify_disable_user);
853 * vringh_need_notify_user - must we tell the other side about used buffers?
854 * @vrh: the vring we've called vringh_complete_user() on.
856 * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
858 int vringh_need_notify_user(struct vringh *vrh)
860 return __vringh_need_notify(vrh, getu16_user);
862 EXPORT_SYMBOL(vringh_need_notify_user);
864 /* Kernelspace access helpers. */
865 static inline int getu16_kern(const struct vringh *vrh,
866 u16 *val, const __virtio16 *p)
868 *val = vringh16_to_cpu(vrh, READ_ONCE(*p));
872 static inline int putu16_kern(const struct vringh *vrh, __virtio16 *p, u16 val)
874 WRITE_ONCE(*p, cpu_to_vringh16(vrh, val));
878 static inline int copydesc_kern(const struct vringh *vrh,
879 void *dst, const void *src, size_t len)
881 memcpy(dst, src, len);
885 static inline int putused_kern(const struct vringh *vrh,
886 struct vring_used_elem *dst,
887 const struct vring_used_elem *src,
890 memcpy(dst, src, num * sizeof(*dst));
894 static inline int xfer_kern(const struct vringh *vrh, void *src,
895 void *dst, size_t len)
897 memcpy(dst, src, len);
901 static inline int kern_xfer(const struct vringh *vrh, void *dst,
902 void *src, size_t len)
904 memcpy(dst, src, len);
909 * vringh_init_kern - initialize a vringh for a kernelspace vring.
910 * @vrh: the vringh to initialize.
911 * @features: the feature bits for this ring.
912 * @num: the number of elements.
913 * @weak_barriers: true if we only need memory barriers, not I/O.
914 * @desc: the userpace descriptor pointer.
915 * @avail: the userpace avail pointer.
916 * @used: the userpace used pointer.
918 * Returns an error if num is invalid.
920 int vringh_init_kern(struct vringh *vrh, u64 features,
921 unsigned int num, bool weak_barriers,
922 struct vring_desc *desc,
923 struct vring_avail *avail,
924 struct vring_used *used)
926 /* Sane power of 2 please! */
927 if (!num || num > 0xffff || (num & (num - 1))) {
928 vringh_bad("Bad ring size %u", num);
932 vrh->little_endian = (features & (1ULL << VIRTIO_F_VERSION_1));
933 vrh->event_indices = (features & (1 << VIRTIO_RING_F_EVENT_IDX));
934 vrh->weak_barriers = weak_barriers;
936 vrh->last_avail_idx = 0;
937 vrh->last_used_idx = 0;
938 vrh->vring.num = num;
939 vrh->vring.desc = desc;
940 vrh->vring.avail = avail;
941 vrh->vring.used = used;
944 EXPORT_SYMBOL(vringh_init_kern);
947 * vringh_getdesc_kern - get next available descriptor from kernelspace ring.
948 * @vrh: the kernelspace vring.
949 * @riov: where to put the readable descriptors (or NULL)
950 * @wiov: where to put the writable descriptors (or NULL)
951 * @head: head index we received, for passing to vringh_complete_kern().
952 * @gfp: flags for allocating larger riov/wiov.
954 * Returns 0 if there was no descriptor, 1 if there was, or -errno.
956 * Note that on error return, you can tell the difference between an
957 * invalid ring and a single invalid descriptor: in the former case,
958 * *head will be vrh->vring.num. You may be able to ignore an invalid
959 * descriptor, but there's not much you can do with an invalid ring.
961 * Note that you can reuse riov and wiov with subsequent calls. Content is
962 * overwritten and memory reallocated if more space is needed.
963 * When you don't have to use riov and wiov anymore, you should clean up them
964 * calling vringh_kiov_cleanup() to release the memory, even on error!
966 int vringh_getdesc_kern(struct vringh *vrh,
967 struct vringh_kiov *riov,
968 struct vringh_kiov *wiov,
974 err = __vringh_get_head(vrh, getu16_kern, &vrh->last_avail_idx);
979 if (err == vrh->vring.num)
983 err = __vringh_iov(vrh, *head, riov, wiov, no_range_check, NULL,
990 EXPORT_SYMBOL(vringh_getdesc_kern);
993 * vringh_iov_pull_kern - copy bytes from vring_iov.
994 * @riov: the riov as passed to vringh_getdesc_kern() (updated as we consume)
995 * @dst: the place to copy.
996 * @len: the maximum length to copy.
998 * Returns the bytes copied <= len or a negative errno.
1000 ssize_t vringh_iov_pull_kern(struct vringh_kiov *riov, void *dst, size_t len)
1002 return vringh_iov_xfer(NULL, riov, dst, len, xfer_kern);
1004 EXPORT_SYMBOL(vringh_iov_pull_kern);
1007 * vringh_iov_push_kern - copy bytes into vring_iov.
1008 * @wiov: the wiov as passed to vringh_getdesc_kern() (updated as we consume)
1009 * @src: the place to copy from.
1010 * @len: the maximum length to copy.
1012 * Returns the bytes copied <= len or a negative errno.
1014 ssize_t vringh_iov_push_kern(struct vringh_kiov *wiov,
1015 const void *src, size_t len)
1017 return vringh_iov_xfer(NULL, wiov, (void *)src, len, kern_xfer);
1019 EXPORT_SYMBOL(vringh_iov_push_kern);
1022 * vringh_abandon_kern - we've decided not to handle the descriptor(s).
1024 * @num: the number of descriptors to put back (ie. num
1025 * vringh_get_kern() to undo).
1027 * The next vringh_get_kern() will return the old descriptor(s) again.
1029 void vringh_abandon_kern(struct vringh *vrh, unsigned int num)
1031 /* We only update vring_avail_event(vr) when we want to be notified,
1032 * so we haven't changed that yet. */
1033 vrh->last_avail_idx -= num;
1035 EXPORT_SYMBOL(vringh_abandon_kern);
1038 * vringh_complete_kern - we've finished with descriptor, publish it.
1040 * @head: the head as filled in by vringh_getdesc_kern.
1041 * @len: the length of data we have written.
1043 * You should check vringh_need_notify_kern() after one or more calls
1046 int vringh_complete_kern(struct vringh *vrh, u16 head, u32 len)
1048 struct vring_used_elem used;
1050 used.id = cpu_to_vringh32(vrh, head);
1051 used.len = cpu_to_vringh32(vrh, len);
1053 return __vringh_complete(vrh, &used, 1, putu16_kern, putused_kern);
1055 EXPORT_SYMBOL(vringh_complete_kern);
1058 * vringh_notify_enable_kern - we want to know if something changes.
1061 * This always enables notifications, but returns false if there are
1062 * now more buffers available in the vring.
1064 bool vringh_notify_enable_kern(struct vringh *vrh)
1066 return __vringh_notify_enable(vrh, getu16_kern, putu16_kern);
1068 EXPORT_SYMBOL(vringh_notify_enable_kern);
1071 * vringh_notify_disable_kern - don't tell us if something changes.
1074 * This is our normal running state: we disable and then only enable when
1075 * we're going to sleep.
1077 void vringh_notify_disable_kern(struct vringh *vrh)
1079 __vringh_notify_disable(vrh, putu16_kern);
1081 EXPORT_SYMBOL(vringh_notify_disable_kern);
1084 * vringh_need_notify_kern - must we tell the other side about used buffers?
1085 * @vrh: the vring we've called vringh_complete_kern() on.
1087 * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
1089 int vringh_need_notify_kern(struct vringh *vrh)
1091 return __vringh_need_notify(vrh, getu16_kern);
1093 EXPORT_SYMBOL(vringh_need_notify_kern);
1095 #if IS_REACHABLE(CONFIG_VHOST_IOTLB)
1097 static int iotlb_translate(const struct vringh *vrh,
1098 u64 addr, u64 len, u64 *translated,
1099 struct bio_vec iov[],
1100 int iov_size, u32 perm)
1102 struct vhost_iotlb_map *map;
1103 struct vhost_iotlb *iotlb = vrh->iotlb;
1107 spin_lock(vrh->iotlb_lock);
1112 if (unlikely(ret >= iov_size)) {
1117 map = vhost_iotlb_itree_first(iotlb, addr,
1119 if (!map || map->start > addr) {
1122 } else if (!(map->perm & perm)) {
1127 size = map->size - addr + map->start;
1128 pa = map->addr + addr - map->start;
1129 pfn = pa >> PAGE_SHIFT;
1130 iov[ret].bv_page = pfn_to_page(pfn);
1131 iov[ret].bv_len = min(len - s, size);
1132 iov[ret].bv_offset = pa & (PAGE_SIZE - 1);
1138 spin_unlock(vrh->iotlb_lock);
1141 *translated = min(len, s);
1146 static inline int copy_from_iotlb(const struct vringh *vrh, void *dst,
1147 void *src, size_t len)
1149 u64 total_translated = 0;
1151 while (total_translated < len) {
1152 struct bio_vec iov[16];
1153 struct iov_iter iter;
1157 ret = iotlb_translate(vrh, (u64)(uintptr_t)src,
1158 len - total_translated, &translated,
1159 iov, ARRAY_SIZE(iov), VHOST_MAP_RO);
1160 if (ret == -ENOBUFS)
1161 ret = ARRAY_SIZE(iov);
1165 iov_iter_bvec(&iter, READ, iov, ret, translated);
1167 ret = copy_from_iter(dst, translated, &iter);
1173 total_translated += translated;
1176 return total_translated;
1179 static inline int copy_to_iotlb(const struct vringh *vrh, void *dst,
1180 void *src, size_t len)
1182 u64 total_translated = 0;
1184 while (total_translated < len) {
1185 struct bio_vec iov[16];
1186 struct iov_iter iter;
1190 ret = iotlb_translate(vrh, (u64)(uintptr_t)dst,
1191 len - total_translated, &translated,
1192 iov, ARRAY_SIZE(iov), VHOST_MAP_WO);
1193 if (ret == -ENOBUFS)
1194 ret = ARRAY_SIZE(iov);
1198 iov_iter_bvec(&iter, WRITE, iov, ret, translated);
1200 ret = copy_to_iter(src, translated, &iter);
1206 total_translated += translated;
1209 return total_translated;
1212 static inline int getu16_iotlb(const struct vringh *vrh,
1213 u16 *val, const __virtio16 *p)
1219 /* Atomic read is needed for getu16 */
1220 ret = iotlb_translate(vrh, (u64)(uintptr_t)p, sizeof(*p), NULL,
1221 &iov, 1, VHOST_MAP_RO);
1225 kaddr = kmap_atomic(iov.bv_page);
1226 from = kaddr + iov.bv_offset;
1227 *val = vringh16_to_cpu(vrh, READ_ONCE(*(__virtio16 *)from));
1228 kunmap_atomic(kaddr);
1233 static inline int putu16_iotlb(const struct vringh *vrh,
1234 __virtio16 *p, u16 val)
1240 /* Atomic write is needed for putu16 */
1241 ret = iotlb_translate(vrh, (u64)(uintptr_t)p, sizeof(*p), NULL,
1242 &iov, 1, VHOST_MAP_WO);
1246 kaddr = kmap_atomic(iov.bv_page);
1247 to = kaddr + iov.bv_offset;
1248 WRITE_ONCE(*(__virtio16 *)to, cpu_to_vringh16(vrh, val));
1249 kunmap_atomic(kaddr);
1254 static inline int copydesc_iotlb(const struct vringh *vrh,
1255 void *dst, const void *src, size_t len)
1259 ret = copy_from_iotlb(vrh, dst, (void *)src, len);
1266 static inline int xfer_from_iotlb(const struct vringh *vrh, void *src,
1267 void *dst, size_t len)
1271 ret = copy_from_iotlb(vrh, dst, src, len);
1278 static inline int xfer_to_iotlb(const struct vringh *vrh,
1279 void *dst, void *src, size_t len)
1283 ret = copy_to_iotlb(vrh, dst, src, len);
1290 static inline int putused_iotlb(const struct vringh *vrh,
1291 struct vring_used_elem *dst,
1292 const struct vring_used_elem *src,
1295 int size = num * sizeof(*dst);
1298 ret = copy_to_iotlb(vrh, dst, (void *)src, num * sizeof(*dst));
1306 * vringh_init_iotlb - initialize a vringh for a ring with IOTLB.
1307 * @vrh: the vringh to initialize.
1308 * @features: the feature bits for this ring.
1309 * @num: the number of elements.
1310 * @weak_barriers: true if we only need memory barriers, not I/O.
1311 * @desc: the userpace descriptor pointer.
1312 * @avail: the userpace avail pointer.
1313 * @used: the userpace used pointer.
1315 * Returns an error if num is invalid.
1317 int vringh_init_iotlb(struct vringh *vrh, u64 features,
1318 unsigned int num, bool weak_barriers,
1319 struct vring_desc *desc,
1320 struct vring_avail *avail,
1321 struct vring_used *used)
1323 return vringh_init_kern(vrh, features, num, weak_barriers,
1326 EXPORT_SYMBOL(vringh_init_iotlb);
1329 * vringh_set_iotlb - initialize a vringh for a ring with IOTLB.
1331 * @iotlb: iotlb associated with this vring
1332 * @iotlb_lock: spinlock to synchronize the iotlb accesses
1334 void vringh_set_iotlb(struct vringh *vrh, struct vhost_iotlb *iotlb,
1335 spinlock_t *iotlb_lock)
1338 vrh->iotlb_lock = iotlb_lock;
1340 EXPORT_SYMBOL(vringh_set_iotlb);
1343 * vringh_getdesc_iotlb - get next available descriptor from ring with
1345 * @vrh: the kernelspace vring.
1346 * @riov: where to put the readable descriptors (or NULL)
1347 * @wiov: where to put the writable descriptors (or NULL)
1348 * @head: head index we received, for passing to vringh_complete_iotlb().
1349 * @gfp: flags for allocating larger riov/wiov.
1351 * Returns 0 if there was no descriptor, 1 if there was, or -errno.
1353 * Note that on error return, you can tell the difference between an
1354 * invalid ring and a single invalid descriptor: in the former case,
1355 * *head will be vrh->vring.num. You may be able to ignore an invalid
1356 * descriptor, but there's not much you can do with an invalid ring.
1358 * Note that you can reuse riov and wiov with subsequent calls. Content is
1359 * overwritten and memory reallocated if more space is needed.
1360 * When you don't have to use riov and wiov anymore, you should clean up them
1361 * calling vringh_kiov_cleanup() to release the memory, even on error!
1363 int vringh_getdesc_iotlb(struct vringh *vrh,
1364 struct vringh_kiov *riov,
1365 struct vringh_kiov *wiov,
1371 err = __vringh_get_head(vrh, getu16_iotlb, &vrh->last_avail_idx);
1376 if (err == vrh->vring.num)
1380 err = __vringh_iov(vrh, *head, riov, wiov, no_range_check, NULL,
1381 gfp, copydesc_iotlb);
1387 EXPORT_SYMBOL(vringh_getdesc_iotlb);
1390 * vringh_iov_pull_iotlb - copy bytes from vring_iov.
1392 * @riov: the riov as passed to vringh_getdesc_iotlb() (updated as we consume)
1393 * @dst: the place to copy.
1394 * @len: the maximum length to copy.
1396 * Returns the bytes copied <= len or a negative errno.
1398 ssize_t vringh_iov_pull_iotlb(struct vringh *vrh,
1399 struct vringh_kiov *riov,
1400 void *dst, size_t len)
1402 return vringh_iov_xfer(vrh, riov, dst, len, xfer_from_iotlb);
1404 EXPORT_SYMBOL(vringh_iov_pull_iotlb);
1407 * vringh_iov_push_iotlb - copy bytes into vring_iov.
1409 * @wiov: the wiov as passed to vringh_getdesc_iotlb() (updated as we consume)
1410 * @src: the place to copy from.
1411 * @len: the maximum length to copy.
1413 * Returns the bytes copied <= len or a negative errno.
1415 ssize_t vringh_iov_push_iotlb(struct vringh *vrh,
1416 struct vringh_kiov *wiov,
1417 const void *src, size_t len)
1419 return vringh_iov_xfer(vrh, wiov, (void *)src, len, xfer_to_iotlb);
1421 EXPORT_SYMBOL(vringh_iov_push_iotlb);
1424 * vringh_abandon_iotlb - we've decided not to handle the descriptor(s).
1426 * @num: the number of descriptors to put back (ie. num
1427 * vringh_get_iotlb() to undo).
1429 * The next vringh_get_iotlb() will return the old descriptor(s) again.
1431 void vringh_abandon_iotlb(struct vringh *vrh, unsigned int num)
1433 /* We only update vring_avail_event(vr) when we want to be notified,
1434 * so we haven't changed that yet.
1436 vrh->last_avail_idx -= num;
1438 EXPORT_SYMBOL(vringh_abandon_iotlb);
1441 * vringh_complete_iotlb - we've finished with descriptor, publish it.
1443 * @head: the head as filled in by vringh_getdesc_iotlb.
1444 * @len: the length of data we have written.
1446 * You should check vringh_need_notify_iotlb() after one or more calls
1449 int vringh_complete_iotlb(struct vringh *vrh, u16 head, u32 len)
1451 struct vring_used_elem used;
1453 used.id = cpu_to_vringh32(vrh, head);
1454 used.len = cpu_to_vringh32(vrh, len);
1456 return __vringh_complete(vrh, &used, 1, putu16_iotlb, putused_iotlb);
1458 EXPORT_SYMBOL(vringh_complete_iotlb);
1461 * vringh_notify_enable_iotlb - we want to know if something changes.
1464 * This always enables notifications, but returns false if there are
1465 * now more buffers available in the vring.
1467 bool vringh_notify_enable_iotlb(struct vringh *vrh)
1469 return __vringh_notify_enable(vrh, getu16_iotlb, putu16_iotlb);
1471 EXPORT_SYMBOL(vringh_notify_enable_iotlb);
1474 * vringh_notify_disable_iotlb - don't tell us if something changes.
1477 * This is our normal running state: we disable and then only enable when
1478 * we're going to sleep.
1480 void vringh_notify_disable_iotlb(struct vringh *vrh)
1482 __vringh_notify_disable(vrh, putu16_iotlb);
1484 EXPORT_SYMBOL(vringh_notify_disable_iotlb);
1487 * vringh_need_notify_iotlb - must we tell the other side about used buffers?
1488 * @vrh: the vring we've called vringh_complete_iotlb() on.
1490 * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
1492 int vringh_need_notify_iotlb(struct vringh *vrh)
1494 return __vringh_need_notify(vrh, getu16_iotlb);
1496 EXPORT_SYMBOL(vringh_need_notify_iotlb);
1500 MODULE_LICENSE("GPL");