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);
78 /* Copy some bytes to/from the iovec. Returns num copied. */
79 static inline ssize_t vringh_iov_xfer(struct vringh *vrh,
80 struct vringh_kiov *iov,
81 void *ptr, size_t len,
82 int (*xfer)(const struct vringh *vrh,
83 void *addr, void *ptr,
88 while (len && iov->i < iov->used) {
91 partlen = min(iov->iov[iov->i].iov_len, len);
92 err = xfer(vrh, iov->iov[iov->i].iov_base, ptr, partlen);
98 iov->consumed += partlen;
99 iov->iov[iov->i].iov_len -= partlen;
100 iov->iov[iov->i].iov_base += partlen;
102 if (!iov->iov[iov->i].iov_len) {
103 /* Fix up old iov element then increment. */
104 iov->iov[iov->i].iov_len = iov->consumed;
105 iov->iov[iov->i].iov_base -= iov->consumed;
115 /* May reduce *len if range is shorter. */
116 static inline bool range_check(struct vringh *vrh, u64 addr, size_t *len,
117 struct vringh_range *range,
118 bool (*getrange)(struct vringh *,
119 u64, struct vringh_range *))
121 if (addr < range->start || addr > range->end_incl) {
122 if (!getrange(vrh, addr, range))
125 BUG_ON(addr < range->start || addr > range->end_incl);
127 /* To end of memory? */
128 if (unlikely(addr + *len == 0)) {
129 if (range->end_incl == -1ULL)
134 /* Otherwise, don't wrap. */
135 if (addr + *len < addr) {
136 vringh_bad("Wrapping descriptor %zu@0x%llx",
137 *len, (unsigned long long)addr);
141 if (unlikely(addr + *len - 1 > range->end_incl))
146 *len = range->end_incl + 1 - addr;
150 static inline bool no_range_check(struct vringh *vrh, u64 addr, size_t *len,
151 struct vringh_range *range,
152 bool (*getrange)(struct vringh *,
153 u64, struct vringh_range *))
158 /* No reason for this code to be inline. */
159 static int move_to_indirect(const struct vringh *vrh,
160 int *up_next, u16 *i, void *addr,
161 const struct vring_desc *desc,
162 struct vring_desc **descs, int *desc_max)
166 /* Indirect tables can't have indirect. */
167 if (*up_next != -1) {
168 vringh_bad("Multilevel indirect %u->%u", *up_next, *i);
172 len = vringh32_to_cpu(vrh, desc->len);
173 if (unlikely(len % sizeof(struct vring_desc))) {
174 vringh_bad("Strange indirect len %u", desc->len);
178 /* We will check this when we follow it! */
179 if (desc->flags & cpu_to_vringh16(vrh, VRING_DESC_F_NEXT))
180 *up_next = vringh16_to_cpu(vrh, desc->next);
184 *desc_max = len / sizeof(struct vring_desc);
186 /* Now, start at the first indirect. */
191 static int resize_iovec(struct vringh_kiov *iov, gfp_t gfp)
194 unsigned int flag, new_num = (iov->max_num & ~VRINGH_IOV_ALLOCATED) * 2;
199 flag = (iov->max_num & VRINGH_IOV_ALLOCATED);
201 new = krealloc_array(iov->iov, new_num,
202 sizeof(struct iovec), gfp);
204 new = kmalloc_array(new_num, sizeof(struct iovec), gfp);
206 memcpy(new, iov->iov,
207 iov->max_num * sizeof(struct iovec));
208 flag = VRINGH_IOV_ALLOCATED;
214 iov->max_num = (new_num | flag);
218 static u16 __cold return_from_indirect(const struct vringh *vrh, int *up_next,
219 struct vring_desc **descs, int *desc_max)
224 *descs = vrh->vring.desc;
225 *desc_max = vrh->vring.num;
229 static int slow_copy(struct vringh *vrh, void *dst, const void *src,
230 bool (*rcheck)(struct vringh *vrh, u64 addr, size_t *len,
231 struct vringh_range *range,
232 bool (*getrange)(struct vringh *vrh,
234 struct vringh_range *)),
235 bool (*getrange)(struct vringh *vrh,
237 struct vringh_range *r),
238 struct vringh_range *range,
239 int (*copy)(const struct vringh *vrh,
240 void *dst, const void *src, size_t len))
242 size_t part, len = sizeof(struct vring_desc);
249 addr = (u64)(unsigned long)src - range->offset;
251 if (!rcheck(vrh, addr, &part, range, getrange))
254 err = copy(vrh, dst, src, part);
266 __vringh_iov(struct vringh *vrh, u16 i,
267 struct vringh_kiov *riov,
268 struct vringh_kiov *wiov,
269 bool (*rcheck)(struct vringh *vrh, u64 addr, size_t *len,
270 struct vringh_range *range,
271 bool (*getrange)(struct vringh *, u64,
272 struct vringh_range *)),
273 bool (*getrange)(struct vringh *, u64, struct vringh_range *),
275 int (*copy)(const struct vringh *vrh,
276 void *dst, const void *src, size_t len))
278 int err, count = 0, up_next, desc_max;
279 struct vring_desc desc, *descs;
280 struct vringh_range range = { -1ULL, 0 }, slowrange;
283 /* We start traversing vring's descriptor table. */
284 descs = vrh->vring.desc;
285 desc_max = vrh->vring.num;
288 /* You must want something! */
289 if (WARN_ON(!riov && !wiov))
293 riov->i = riov->used = 0;
295 wiov->i = wiov->used = 0;
299 struct vringh_kiov *iov;
303 err = slow_copy(vrh, &desc, &descs[i], rcheck, getrange,
306 err = copy(vrh, &desc, &descs[i], sizeof(desc));
310 if (unlikely(desc.flags &
311 cpu_to_vringh16(vrh, VRING_DESC_F_INDIRECT))) {
312 u64 a = vringh64_to_cpu(vrh, desc.addr);
314 /* Make sure it's OK, and get offset. */
315 len = vringh32_to_cpu(vrh, desc.len);
316 if (!rcheck(vrh, a, &len, &range, getrange)) {
321 if (unlikely(len != vringh32_to_cpu(vrh, desc.len))) {
323 /* We need to save this range to use offset */
327 addr = (void *)(long)(a + range.offset);
328 err = move_to_indirect(vrh, &up_next, &i, addr, &desc,
335 if (count++ == vrh->vring.num) {
336 vringh_bad("Descriptor loop in %p", descs);
341 if (desc.flags & cpu_to_vringh16(vrh, VRING_DESC_F_WRITE))
345 if (unlikely(wiov && wiov->i)) {
346 vringh_bad("Readable desc %p after writable",
354 vringh_bad("Unexpected %s desc",
355 !wiov ? "writable" : "readable");
361 /* Make sure it's OK, and get offset. */
362 len = vringh32_to_cpu(vrh, desc.len);
363 if (!rcheck(vrh, vringh64_to_cpu(vrh, desc.addr), &len, &range,
368 addr = (void *)(unsigned long)(vringh64_to_cpu(vrh, desc.addr) +
371 if (unlikely(iov->used == (iov->max_num & ~VRINGH_IOV_ALLOCATED))) {
372 err = resize_iovec(iov, gfp);
377 iov->iov[iov->used].iov_base = addr;
378 iov->iov[iov->used].iov_len = len;
381 if (unlikely(len != vringh32_to_cpu(vrh, desc.len))) {
382 desc.len = cpu_to_vringh32(vrh,
383 vringh32_to_cpu(vrh, desc.len) - len);
384 desc.addr = cpu_to_vringh64(vrh,
385 vringh64_to_cpu(vrh, desc.addr) + len);
389 if (desc.flags & cpu_to_vringh16(vrh, VRING_DESC_F_NEXT)) {
390 i = vringh16_to_cpu(vrh, desc.next);
392 /* Just in case we need to finish traversing above. */
393 if (unlikely(up_next > 0)) {
394 i = return_from_indirect(vrh, &up_next,
402 vringh_bad("Chained index %u > %u", i, desc_max);
414 static inline int __vringh_complete(struct vringh *vrh,
415 const struct vring_used_elem *used,
416 unsigned int num_used,
417 int (*putu16)(const struct vringh *vrh,
418 __virtio16 *p, u16 val),
419 int (*putused)(const struct vringh *vrh,
420 struct vring_used_elem *dst,
421 const struct vring_used_elem
424 struct vring_used *used_ring;
428 used_ring = vrh->vring.used;
429 used_idx = vrh->last_used_idx + vrh->completed;
431 off = used_idx % vrh->vring.num;
433 /* Compiler knows num_used == 1 sometimes, hence extra check */
434 if (num_used > 1 && unlikely(off + num_used >= vrh->vring.num)) {
435 u16 part = vrh->vring.num - off;
436 err = putused(vrh, &used_ring->ring[off], used, part);
438 err = putused(vrh, &used_ring->ring[0], used + part,
441 err = putused(vrh, &used_ring->ring[off], used, num_used);
444 vringh_bad("Failed to write %u used entries %u at %p",
445 num_used, off, &used_ring->ring[off]);
449 /* Make sure buffer is written before we update index. */
450 virtio_wmb(vrh->weak_barriers);
452 err = putu16(vrh, &vrh->vring.used->idx, used_idx + num_used);
454 vringh_bad("Failed to update used index at %p",
455 &vrh->vring.used->idx);
459 vrh->completed += num_used;
464 static inline int __vringh_need_notify(struct vringh *vrh,
465 int (*getu16)(const struct vringh *vrh,
467 const __virtio16 *p))
473 /* Flush out used index update. This is paired with the
474 * barrier that the Guest executes when enabling
476 virtio_mb(vrh->weak_barriers);
478 /* Old-style, without event indices. */
479 if (!vrh->event_indices) {
481 err = getu16(vrh, &flags, &vrh->vring.avail->flags);
483 vringh_bad("Failed to get flags at %p",
484 &vrh->vring.avail->flags);
487 return (!(flags & VRING_AVAIL_F_NO_INTERRUPT));
490 /* Modern: we know when other side wants to know. */
491 err = getu16(vrh, &used_event, &vring_used_event(&vrh->vring));
493 vringh_bad("Failed to get used event idx at %p",
494 &vring_used_event(&vrh->vring));
498 /* Just in case we added so many that we wrap. */
499 if (unlikely(vrh->completed > 0xffff))
502 notify = vring_need_event(used_event,
503 vrh->last_used_idx + vrh->completed,
506 vrh->last_used_idx += vrh->completed;
511 static inline bool __vringh_notify_enable(struct vringh *vrh,
512 int (*getu16)(const struct vringh *vrh,
513 u16 *val, const __virtio16 *p),
514 int (*putu16)(const struct vringh *vrh,
515 __virtio16 *p, u16 val))
519 if (!vrh->event_indices) {
520 /* Old-school; update flags. */
521 if (putu16(vrh, &vrh->vring.used->flags, 0) != 0) {
522 vringh_bad("Clearing used flags %p",
523 &vrh->vring.used->flags);
527 if (putu16(vrh, &vring_avail_event(&vrh->vring),
528 vrh->last_avail_idx) != 0) {
529 vringh_bad("Updating avail event index %p",
530 &vring_avail_event(&vrh->vring));
535 /* They could have slipped one in as we were doing that: make
536 * sure it's written, then check again. */
537 virtio_mb(vrh->weak_barriers);
539 if (getu16(vrh, &avail, &vrh->vring.avail->idx) != 0) {
540 vringh_bad("Failed to check avail idx at %p",
541 &vrh->vring.avail->idx);
545 /* This is unlikely, so we just leave notifications enabled
546 * (if we're using event_indices, we'll only get one
547 * notification anyway). */
548 return avail == vrh->last_avail_idx;
551 static inline void __vringh_notify_disable(struct vringh *vrh,
552 int (*putu16)(const struct vringh *vrh,
553 __virtio16 *p, u16 val))
555 if (!vrh->event_indices) {
556 /* Old-school; update flags. */
557 if (putu16(vrh, &vrh->vring.used->flags,
558 VRING_USED_F_NO_NOTIFY)) {
559 vringh_bad("Setting used flags %p",
560 &vrh->vring.used->flags);
565 /* Userspace access helpers: in this case, addresses are really userspace. */
566 static inline int getu16_user(const struct vringh *vrh, u16 *val, const __virtio16 *p)
569 int rc = get_user(v, (__force __virtio16 __user *)p);
570 *val = vringh16_to_cpu(vrh, v);
574 static inline int putu16_user(const struct vringh *vrh, __virtio16 *p, u16 val)
576 __virtio16 v = cpu_to_vringh16(vrh, val);
577 return put_user(v, (__force __virtio16 __user *)p);
580 static inline int copydesc_user(const struct vringh *vrh,
581 void *dst, const void *src, size_t len)
583 return copy_from_user(dst, (__force void __user *)src, len) ?
587 static inline int putused_user(const struct vringh *vrh,
588 struct vring_used_elem *dst,
589 const struct vring_used_elem *src,
592 return copy_to_user((__force void __user *)dst, src,
593 sizeof(*dst) * num) ? -EFAULT : 0;
596 static inline int xfer_from_user(const struct vringh *vrh, void *src,
597 void *dst, size_t len)
599 return copy_from_user(dst, (__force void __user *)src, len) ?
603 static inline int xfer_to_user(const struct vringh *vrh,
604 void *dst, void *src, size_t len)
606 return copy_to_user((__force void __user *)dst, src, len) ?
611 * vringh_init_user - initialize a vringh for a userspace vring.
612 * @vrh: the vringh to initialize.
613 * @features: the feature bits for this ring.
614 * @num: the number of elements.
615 * @weak_barriers: true if we only need memory barriers, not I/O.
616 * @desc: the userpace descriptor pointer.
617 * @avail: the userpace avail pointer.
618 * @used: the userpace used pointer.
620 * Returns an error if num is invalid: you should check pointers
623 int vringh_init_user(struct vringh *vrh, u64 features,
624 unsigned int num, bool weak_barriers,
625 vring_desc_t __user *desc,
626 vring_avail_t __user *avail,
627 vring_used_t __user *used)
629 /* Sane power of 2 please! */
630 if (!num || num > 0xffff || (num & (num - 1))) {
631 vringh_bad("Bad ring size %u", num);
635 vrh->little_endian = (features & (1ULL << VIRTIO_F_VERSION_1));
636 vrh->event_indices = (features & (1 << VIRTIO_RING_F_EVENT_IDX));
637 vrh->weak_barriers = weak_barriers;
639 vrh->last_avail_idx = 0;
640 vrh->last_used_idx = 0;
641 vrh->vring.num = num;
642 /* vring expects kernel addresses, but only used via accessors. */
643 vrh->vring.desc = (__force struct vring_desc *)desc;
644 vrh->vring.avail = (__force struct vring_avail *)avail;
645 vrh->vring.used = (__force struct vring_used *)used;
648 EXPORT_SYMBOL(vringh_init_user);
651 * vringh_getdesc_user - get next available descriptor from userspace ring.
652 * @vrh: the userspace vring.
653 * @riov: where to put the readable descriptors (or NULL)
654 * @wiov: where to put the writable descriptors (or NULL)
655 * @getrange: function to call to check ranges.
656 * @head: head index we received, for passing to vringh_complete_user().
658 * Returns 0 if there was no descriptor, 1 if there was, or -errno.
660 * Note that on error return, you can tell the difference between an
661 * invalid ring and a single invalid descriptor: in the former case,
662 * *head will be vrh->vring.num. You may be able to ignore an invalid
663 * descriptor, but there's not much you can do with an invalid ring.
665 * Note that you may need to clean up riov and wiov, even on error!
667 int vringh_getdesc_user(struct vringh *vrh,
668 struct vringh_iov *riov,
669 struct vringh_iov *wiov,
670 bool (*getrange)(struct vringh *vrh,
671 u64 addr, struct vringh_range *r),
676 *head = vrh->vring.num;
677 err = __vringh_get_head(vrh, getu16_user, &vrh->last_avail_idx);
682 if (err == vrh->vring.num)
685 /* We need the layouts to be the identical for this to work */
686 BUILD_BUG_ON(sizeof(struct vringh_kiov) != sizeof(struct vringh_iov));
687 BUILD_BUG_ON(offsetof(struct vringh_kiov, iov) !=
688 offsetof(struct vringh_iov, iov));
689 BUILD_BUG_ON(offsetof(struct vringh_kiov, i) !=
690 offsetof(struct vringh_iov, i));
691 BUILD_BUG_ON(offsetof(struct vringh_kiov, used) !=
692 offsetof(struct vringh_iov, used));
693 BUILD_BUG_ON(offsetof(struct vringh_kiov, max_num) !=
694 offsetof(struct vringh_iov, max_num));
695 BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
696 BUILD_BUG_ON(offsetof(struct iovec, iov_base) !=
697 offsetof(struct kvec, iov_base));
698 BUILD_BUG_ON(offsetof(struct iovec, iov_len) !=
699 offsetof(struct kvec, iov_len));
700 BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_base)
701 != sizeof(((struct kvec *)NULL)->iov_base));
702 BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_len)
703 != sizeof(((struct kvec *)NULL)->iov_len));
706 err = __vringh_iov(vrh, *head, (struct vringh_kiov *)riov,
707 (struct vringh_kiov *)wiov,
708 range_check, getrange, GFP_KERNEL, copydesc_user);
714 EXPORT_SYMBOL(vringh_getdesc_user);
717 * vringh_iov_pull_user - copy bytes from vring_iov.
718 * @riov: the riov as passed to vringh_getdesc_user() (updated as we consume)
719 * @dst: the place to copy.
720 * @len: the maximum length to copy.
722 * Returns the bytes copied <= len or a negative errno.
724 ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
726 return vringh_iov_xfer(NULL, (struct vringh_kiov *)riov,
727 dst, len, xfer_from_user);
729 EXPORT_SYMBOL(vringh_iov_pull_user);
732 * vringh_iov_push_user - copy bytes into vring_iov.
733 * @wiov: the wiov as passed to vringh_getdesc_user() (updated as we consume)
734 * @src: the place to copy from.
735 * @len: the maximum length to copy.
737 * Returns the bytes copied <= len or a negative errno.
739 ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
740 const void *src, size_t len)
742 return vringh_iov_xfer(NULL, (struct vringh_kiov *)wiov,
743 (void *)src, len, xfer_to_user);
745 EXPORT_SYMBOL(vringh_iov_push_user);
748 * vringh_abandon_user - we've decided not to handle the descriptor(s).
750 * @num: the number of descriptors to put back (ie. num
751 * vringh_get_user() to undo).
753 * The next vringh_get_user() will return the old descriptor(s) again.
755 void vringh_abandon_user(struct vringh *vrh, unsigned int num)
757 /* We only update vring_avail_event(vr) when we want to be notified,
758 * so we haven't changed that yet. */
759 vrh->last_avail_idx -= num;
761 EXPORT_SYMBOL(vringh_abandon_user);
764 * vringh_complete_user - we've finished with descriptor, publish it.
766 * @head: the head as filled in by vringh_getdesc_user.
767 * @len: the length of data we have written.
769 * You should check vringh_need_notify_user() after one or more calls
772 int vringh_complete_user(struct vringh *vrh, u16 head, u32 len)
774 struct vring_used_elem used;
776 used.id = cpu_to_vringh32(vrh, head);
777 used.len = cpu_to_vringh32(vrh, len);
778 return __vringh_complete(vrh, &used, 1, putu16_user, putused_user);
780 EXPORT_SYMBOL(vringh_complete_user);
783 * vringh_complete_multi_user - we've finished with many descriptors.
785 * @used: the head, length pairs.
786 * @num_used: the number of used elements.
788 * You should check vringh_need_notify_user() after one or more calls
791 int vringh_complete_multi_user(struct vringh *vrh,
792 const struct vring_used_elem used[],
795 return __vringh_complete(vrh, used, num_used,
796 putu16_user, putused_user);
798 EXPORT_SYMBOL(vringh_complete_multi_user);
801 * vringh_notify_enable_user - we want to know if something changes.
804 * This always enables notifications, but returns false if there are
805 * now more buffers available in the vring.
807 bool vringh_notify_enable_user(struct vringh *vrh)
809 return __vringh_notify_enable(vrh, getu16_user, putu16_user);
811 EXPORT_SYMBOL(vringh_notify_enable_user);
814 * vringh_notify_disable_user - don't tell us if something changes.
817 * This is our normal running state: we disable and then only enable when
818 * we're going to sleep.
820 void vringh_notify_disable_user(struct vringh *vrh)
822 __vringh_notify_disable(vrh, putu16_user);
824 EXPORT_SYMBOL(vringh_notify_disable_user);
827 * vringh_need_notify_user - must we tell the other side about used buffers?
828 * @vrh: the vring we've called vringh_complete_user() on.
830 * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
832 int vringh_need_notify_user(struct vringh *vrh)
834 return __vringh_need_notify(vrh, getu16_user);
836 EXPORT_SYMBOL(vringh_need_notify_user);
838 /* Kernelspace access helpers. */
839 static inline int getu16_kern(const struct vringh *vrh,
840 u16 *val, const __virtio16 *p)
842 *val = vringh16_to_cpu(vrh, READ_ONCE(*p));
846 static inline int putu16_kern(const struct vringh *vrh, __virtio16 *p, u16 val)
848 WRITE_ONCE(*p, cpu_to_vringh16(vrh, val));
852 static inline int copydesc_kern(const struct vringh *vrh,
853 void *dst, const void *src, size_t len)
855 memcpy(dst, src, len);
859 static inline int putused_kern(const struct vringh *vrh,
860 struct vring_used_elem *dst,
861 const struct vring_used_elem *src,
864 memcpy(dst, src, num * sizeof(*dst));
868 static inline int xfer_kern(const struct vringh *vrh, void *src,
869 void *dst, size_t len)
871 memcpy(dst, src, len);
875 static inline int kern_xfer(const struct vringh *vrh, void *dst,
876 void *src, size_t len)
878 memcpy(dst, src, len);
883 * vringh_init_kern - initialize a vringh for a kernelspace vring.
884 * @vrh: the vringh to initialize.
885 * @features: the feature bits for this ring.
886 * @num: the number of elements.
887 * @weak_barriers: true if we only need memory barriers, not I/O.
888 * @desc: the userpace descriptor pointer.
889 * @avail: the userpace avail pointer.
890 * @used: the userpace used pointer.
892 * Returns an error if num is invalid.
894 int vringh_init_kern(struct vringh *vrh, u64 features,
895 unsigned int num, bool weak_barriers,
896 struct vring_desc *desc,
897 struct vring_avail *avail,
898 struct vring_used *used)
900 /* Sane power of 2 please! */
901 if (!num || num > 0xffff || (num & (num - 1))) {
902 vringh_bad("Bad ring size %u", num);
906 vrh->little_endian = (features & (1ULL << VIRTIO_F_VERSION_1));
907 vrh->event_indices = (features & (1 << VIRTIO_RING_F_EVENT_IDX));
908 vrh->weak_barriers = weak_barriers;
910 vrh->last_avail_idx = 0;
911 vrh->last_used_idx = 0;
912 vrh->vring.num = num;
913 vrh->vring.desc = desc;
914 vrh->vring.avail = avail;
915 vrh->vring.used = used;
918 EXPORT_SYMBOL(vringh_init_kern);
921 * vringh_getdesc_kern - get next available descriptor from kernelspace ring.
922 * @vrh: the kernelspace vring.
923 * @riov: where to put the readable descriptors (or NULL)
924 * @wiov: where to put the writable descriptors (or NULL)
925 * @head: head index we received, for passing to vringh_complete_kern().
926 * @gfp: flags for allocating larger riov/wiov.
928 * Returns 0 if there was no descriptor, 1 if there was, or -errno.
930 * Note that on error return, you can tell the difference between an
931 * invalid ring and a single invalid descriptor: in the former case,
932 * *head will be vrh->vring.num. You may be able to ignore an invalid
933 * descriptor, but there's not much you can do with an invalid ring.
935 * Note that you may need to clean up riov and wiov, even on error!
937 int vringh_getdesc_kern(struct vringh *vrh,
938 struct vringh_kiov *riov,
939 struct vringh_kiov *wiov,
945 err = __vringh_get_head(vrh, getu16_kern, &vrh->last_avail_idx);
950 if (err == vrh->vring.num)
954 err = __vringh_iov(vrh, *head, riov, wiov, no_range_check, NULL,
961 EXPORT_SYMBOL(vringh_getdesc_kern);
964 * vringh_iov_pull_kern - copy bytes from vring_iov.
965 * @riov: the riov as passed to vringh_getdesc_kern() (updated as we consume)
966 * @dst: the place to copy.
967 * @len: the maximum length to copy.
969 * Returns the bytes copied <= len or a negative errno.
971 ssize_t vringh_iov_pull_kern(struct vringh_kiov *riov, void *dst, size_t len)
973 return vringh_iov_xfer(NULL, riov, dst, len, xfer_kern);
975 EXPORT_SYMBOL(vringh_iov_pull_kern);
978 * vringh_iov_push_kern - copy bytes into vring_iov.
979 * @wiov: the wiov as passed to vringh_getdesc_kern() (updated as we consume)
980 * @src: the place to copy from.
981 * @len: the maximum length to copy.
983 * Returns the bytes copied <= len or a negative errno.
985 ssize_t vringh_iov_push_kern(struct vringh_kiov *wiov,
986 const void *src, size_t len)
988 return vringh_iov_xfer(NULL, wiov, (void *)src, len, kern_xfer);
990 EXPORT_SYMBOL(vringh_iov_push_kern);
993 * vringh_abandon_kern - we've decided not to handle the descriptor(s).
995 * @num: the number of descriptors to put back (ie. num
996 * vringh_get_kern() to undo).
998 * The next vringh_get_kern() will return the old descriptor(s) again.
1000 void vringh_abandon_kern(struct vringh *vrh, unsigned int num)
1002 /* We only update vring_avail_event(vr) when we want to be notified,
1003 * so we haven't changed that yet. */
1004 vrh->last_avail_idx -= num;
1006 EXPORT_SYMBOL(vringh_abandon_kern);
1009 * vringh_complete_kern - we've finished with descriptor, publish it.
1011 * @head: the head as filled in by vringh_getdesc_kern.
1012 * @len: the length of data we have written.
1014 * You should check vringh_need_notify_kern() after one or more calls
1017 int vringh_complete_kern(struct vringh *vrh, u16 head, u32 len)
1019 struct vring_used_elem used;
1021 used.id = cpu_to_vringh32(vrh, head);
1022 used.len = cpu_to_vringh32(vrh, len);
1024 return __vringh_complete(vrh, &used, 1, putu16_kern, putused_kern);
1026 EXPORT_SYMBOL(vringh_complete_kern);
1029 * vringh_notify_enable_kern - we want to know if something changes.
1032 * This always enables notifications, but returns false if there are
1033 * now more buffers available in the vring.
1035 bool vringh_notify_enable_kern(struct vringh *vrh)
1037 return __vringh_notify_enable(vrh, getu16_kern, putu16_kern);
1039 EXPORT_SYMBOL(vringh_notify_enable_kern);
1042 * vringh_notify_disable_kern - don't tell us if something changes.
1045 * This is our normal running state: we disable and then only enable when
1046 * we're going to sleep.
1048 void vringh_notify_disable_kern(struct vringh *vrh)
1050 __vringh_notify_disable(vrh, putu16_kern);
1052 EXPORT_SYMBOL(vringh_notify_disable_kern);
1055 * vringh_need_notify_kern - must we tell the other side about used buffers?
1056 * @vrh: the vring we've called vringh_complete_kern() on.
1058 * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
1060 int vringh_need_notify_kern(struct vringh *vrh)
1062 return __vringh_need_notify(vrh, getu16_kern);
1064 EXPORT_SYMBOL(vringh_need_notify_kern);
1066 #if IS_REACHABLE(CONFIG_VHOST_IOTLB)
1068 static int iotlb_translate(const struct vringh *vrh,
1069 u64 addr, u64 len, struct bio_vec iov[],
1070 int iov_size, u32 perm)
1072 struct vhost_iotlb_map *map;
1073 struct vhost_iotlb *iotlb = vrh->iotlb;
1080 if (unlikely(ret >= iov_size)) {
1085 map = vhost_iotlb_itree_first(iotlb, addr,
1087 if (!map || map->start > addr) {
1090 } else if (!(map->perm & perm)) {
1095 size = map->size - addr + map->start;
1096 pa = map->addr + addr - map->start;
1097 pfn = pa >> PAGE_SHIFT;
1098 iov[ret].bv_page = pfn_to_page(pfn);
1099 iov[ret].bv_len = min(len - s, size);
1100 iov[ret].bv_offset = pa & (PAGE_SIZE - 1);
1109 static inline int copy_from_iotlb(const struct vringh *vrh, void *dst,
1110 void *src, size_t len)
1112 struct iov_iter iter;
1113 struct bio_vec iov[16];
1116 ret = iotlb_translate(vrh, (u64)(uintptr_t)src,
1117 len, iov, 16, VHOST_MAP_RO);
1121 iov_iter_bvec(&iter, READ, iov, ret, len);
1123 ret = copy_from_iter(dst, len, &iter);
1128 static inline int copy_to_iotlb(const struct vringh *vrh, void *dst,
1129 void *src, size_t len)
1131 struct iov_iter iter;
1132 struct bio_vec iov[16];
1135 ret = iotlb_translate(vrh, (u64)(uintptr_t)dst,
1136 len, iov, 16, VHOST_MAP_WO);
1140 iov_iter_bvec(&iter, WRITE, iov, ret, len);
1142 return copy_to_iter(src, len, &iter);
1145 static inline int getu16_iotlb(const struct vringh *vrh,
1146 u16 *val, const __virtio16 *p)
1152 /* Atomic read is needed for getu16 */
1153 ret = iotlb_translate(vrh, (u64)(uintptr_t)p, sizeof(*p),
1154 &iov, 1, VHOST_MAP_RO);
1158 kaddr = kmap_atomic(iov.bv_page);
1159 from = kaddr + iov.bv_offset;
1160 *val = vringh16_to_cpu(vrh, READ_ONCE(*(__virtio16 *)from));
1161 kunmap_atomic(kaddr);
1166 static inline int putu16_iotlb(const struct vringh *vrh,
1167 __virtio16 *p, u16 val)
1173 /* Atomic write is needed for putu16 */
1174 ret = iotlb_translate(vrh, (u64)(uintptr_t)p, sizeof(*p),
1175 &iov, 1, VHOST_MAP_WO);
1179 kaddr = kmap_atomic(iov.bv_page);
1180 to = kaddr + iov.bv_offset;
1181 WRITE_ONCE(*(__virtio16 *)to, cpu_to_vringh16(vrh, val));
1182 kunmap_atomic(kaddr);
1187 static inline int copydesc_iotlb(const struct vringh *vrh,
1188 void *dst, const void *src, size_t len)
1192 ret = copy_from_iotlb(vrh, dst, (void *)src, len);
1199 static inline int xfer_from_iotlb(const struct vringh *vrh, void *src,
1200 void *dst, size_t len)
1204 ret = copy_from_iotlb(vrh, dst, src, len);
1211 static inline int xfer_to_iotlb(const struct vringh *vrh,
1212 void *dst, void *src, size_t len)
1216 ret = copy_to_iotlb(vrh, dst, src, len);
1223 static inline int putused_iotlb(const struct vringh *vrh,
1224 struct vring_used_elem *dst,
1225 const struct vring_used_elem *src,
1228 int size = num * sizeof(*dst);
1231 ret = copy_to_iotlb(vrh, dst, (void *)src, num * sizeof(*dst));
1239 * vringh_init_iotlb - initialize a vringh for a ring with IOTLB.
1240 * @vrh: the vringh to initialize.
1241 * @features: the feature bits for this ring.
1242 * @num: the number of elements.
1243 * @weak_barriers: true if we only need memory barriers, not I/O.
1244 * @desc: the userpace descriptor pointer.
1245 * @avail: the userpace avail pointer.
1246 * @used: the userpace used pointer.
1248 * Returns an error if num is invalid.
1250 int vringh_init_iotlb(struct vringh *vrh, u64 features,
1251 unsigned int num, bool weak_barriers,
1252 struct vring_desc *desc,
1253 struct vring_avail *avail,
1254 struct vring_used *used)
1256 return vringh_init_kern(vrh, features, num, weak_barriers,
1259 EXPORT_SYMBOL(vringh_init_iotlb);
1262 * vringh_set_iotlb - initialize a vringh for a ring with IOTLB.
1264 * @iotlb: iotlb associated with this vring
1266 void vringh_set_iotlb(struct vringh *vrh, struct vhost_iotlb *iotlb)
1270 EXPORT_SYMBOL(vringh_set_iotlb);
1273 * vringh_getdesc_iotlb - get next available descriptor from ring with
1275 * @vrh: the kernelspace vring.
1276 * @riov: where to put the readable descriptors (or NULL)
1277 * @wiov: where to put the writable descriptors (or NULL)
1278 * @head: head index we received, for passing to vringh_complete_iotlb().
1279 * @gfp: flags for allocating larger riov/wiov.
1281 * Returns 0 if there was no descriptor, 1 if there was, or -errno.
1283 * Note that on error return, you can tell the difference between an
1284 * invalid ring and a single invalid descriptor: in the former case,
1285 * *head will be vrh->vring.num. You may be able to ignore an invalid
1286 * descriptor, but there's not much you can do with an invalid ring.
1288 * Note that you may need to clean up riov and wiov, even on error!
1290 int vringh_getdesc_iotlb(struct vringh *vrh,
1291 struct vringh_kiov *riov,
1292 struct vringh_kiov *wiov,
1298 err = __vringh_get_head(vrh, getu16_iotlb, &vrh->last_avail_idx);
1303 if (err == vrh->vring.num)
1307 err = __vringh_iov(vrh, *head, riov, wiov, no_range_check, NULL,
1308 gfp, copydesc_iotlb);
1314 EXPORT_SYMBOL(vringh_getdesc_iotlb);
1317 * vringh_iov_pull_iotlb - copy bytes from vring_iov.
1319 * @riov: the riov as passed to vringh_getdesc_iotlb() (updated as we consume)
1320 * @dst: the place to copy.
1321 * @len: the maximum length to copy.
1323 * Returns the bytes copied <= len or a negative errno.
1325 ssize_t vringh_iov_pull_iotlb(struct vringh *vrh,
1326 struct vringh_kiov *riov,
1327 void *dst, size_t len)
1329 return vringh_iov_xfer(vrh, riov, dst, len, xfer_from_iotlb);
1331 EXPORT_SYMBOL(vringh_iov_pull_iotlb);
1334 * vringh_iov_push_iotlb - copy bytes into vring_iov.
1336 * @wiov: the wiov as passed to vringh_getdesc_iotlb() (updated as we consume)
1337 * @src: the place to copy from.
1338 * @len: the maximum length to copy.
1340 * Returns the bytes copied <= len or a negative errno.
1342 ssize_t vringh_iov_push_iotlb(struct vringh *vrh,
1343 struct vringh_kiov *wiov,
1344 const void *src, size_t len)
1346 return vringh_iov_xfer(vrh, wiov, (void *)src, len, xfer_to_iotlb);
1348 EXPORT_SYMBOL(vringh_iov_push_iotlb);
1351 * vringh_abandon_iotlb - we've decided not to handle the descriptor(s).
1353 * @num: the number of descriptors to put back (ie. num
1354 * vringh_get_iotlb() to undo).
1356 * The next vringh_get_iotlb() will return the old descriptor(s) again.
1358 void vringh_abandon_iotlb(struct vringh *vrh, unsigned int num)
1360 /* We only update vring_avail_event(vr) when we want to be notified,
1361 * so we haven't changed that yet.
1363 vrh->last_avail_idx -= num;
1365 EXPORT_SYMBOL(vringh_abandon_iotlb);
1368 * vringh_complete_iotlb - we've finished with descriptor, publish it.
1370 * @head: the head as filled in by vringh_getdesc_iotlb.
1371 * @len: the length of data we have written.
1373 * You should check vringh_need_notify_iotlb() after one or more calls
1376 int vringh_complete_iotlb(struct vringh *vrh, u16 head, u32 len)
1378 struct vring_used_elem used;
1380 used.id = cpu_to_vringh32(vrh, head);
1381 used.len = cpu_to_vringh32(vrh, len);
1383 return __vringh_complete(vrh, &used, 1, putu16_iotlb, putused_iotlb);
1385 EXPORT_SYMBOL(vringh_complete_iotlb);
1388 * vringh_notify_enable_iotlb - we want to know if something changes.
1391 * This always enables notifications, but returns false if there are
1392 * now more buffers available in the vring.
1394 bool vringh_notify_enable_iotlb(struct vringh *vrh)
1396 return __vringh_notify_enable(vrh, getu16_iotlb, putu16_iotlb);
1398 EXPORT_SYMBOL(vringh_notify_enable_iotlb);
1401 * vringh_notify_disable_iotlb - don't tell us if something changes.
1404 * This is our normal running state: we disable and then only enable when
1405 * we're going to sleep.
1407 void vringh_notify_disable_iotlb(struct vringh *vrh)
1409 __vringh_notify_disable(vrh, putu16_iotlb);
1411 EXPORT_SYMBOL(vringh_notify_disable_iotlb);
1414 * vringh_need_notify_iotlb - must we tell the other side about used buffers?
1415 * @vrh: the vring we've called vringh_complete_iotlb() on.
1417 * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
1419 int vringh_need_notify_iotlb(struct vringh *vrh)
1421 return __vringh_need_notify(vrh, getu16_iotlb);
1423 EXPORT_SYMBOL(vringh_need_notify_iotlb);
1427 MODULE_LICENSE("GPL");