Loading...
Note: File does not exist in v3.1.
1/*
2 * Helpers for the host side of a virtio ring.
3 *
4 * Since these may be in userspace, we use (inline) accessors.
5 */
6#include <linux/module.h>
7#include <linux/vringh.h>
8#include <linux/virtio_ring.h>
9#include <linux/kernel.h>
10#include <linux/ratelimit.h>
11#include <linux/uaccess.h>
12#include <linux/slab.h>
13#include <linux/export.h>
14#include <uapi/linux/virtio_config.h>
15
16static __printf(1,2) __cold void vringh_bad(const char *fmt, ...)
17{
18 static DEFINE_RATELIMIT_STATE(vringh_rs,
19 DEFAULT_RATELIMIT_INTERVAL,
20 DEFAULT_RATELIMIT_BURST);
21 if (__ratelimit(&vringh_rs)) {
22 va_list ap;
23 va_start(ap, fmt);
24 printk(KERN_NOTICE "vringh:");
25 vprintk(fmt, ap);
26 va_end(ap);
27 }
28}
29
30/* Returns vring->num if empty, -ve on error. */
31static inline int __vringh_get_head(const struct vringh *vrh,
32 int (*getu16)(const struct vringh *vrh,
33 u16 *val, const __virtio16 *p),
34 u16 *last_avail_idx)
35{
36 u16 avail_idx, i, head;
37 int err;
38
39 err = getu16(vrh, &avail_idx, &vrh->vring.avail->idx);
40 if (err) {
41 vringh_bad("Failed to access avail idx at %p",
42 &vrh->vring.avail->idx);
43 return err;
44 }
45
46 if (*last_avail_idx == avail_idx)
47 return vrh->vring.num;
48
49 /* Only get avail ring entries after they have been exposed by guest. */
50 virtio_rmb(vrh->weak_barriers);
51
52 i = *last_avail_idx & (vrh->vring.num - 1);
53
54 err = getu16(vrh, &head, &vrh->vring.avail->ring[i]);
55 if (err) {
56 vringh_bad("Failed to read head: idx %d address %p",
57 *last_avail_idx, &vrh->vring.avail->ring[i]);
58 return err;
59 }
60
61 if (head >= vrh->vring.num) {
62 vringh_bad("Guest says index %u > %u is available",
63 head, vrh->vring.num);
64 return -EINVAL;
65 }
66
67 (*last_avail_idx)++;
68 return head;
69}
70
71/* Copy some bytes to/from the iovec. Returns num copied. */
72static inline ssize_t vringh_iov_xfer(struct vringh_kiov *iov,
73 void *ptr, size_t len,
74 int (*xfer)(void *addr, void *ptr,
75 size_t len))
76{
77 int err, done = 0;
78
79 while (len && iov->i < iov->used) {
80 size_t partlen;
81
82 partlen = min(iov->iov[iov->i].iov_len, len);
83 err = xfer(iov->iov[iov->i].iov_base, ptr, partlen);
84 if (err)
85 return err;
86 done += partlen;
87 len -= partlen;
88 ptr += partlen;
89 iov->consumed += partlen;
90 iov->iov[iov->i].iov_len -= partlen;
91 iov->iov[iov->i].iov_base += partlen;
92
93 if (!iov->iov[iov->i].iov_len) {
94 /* Fix up old iov element then increment. */
95 iov->iov[iov->i].iov_len = iov->consumed;
96 iov->iov[iov->i].iov_base -= iov->consumed;
97
98 iov->consumed = 0;
99 iov->i++;
100 }
101 }
102 return done;
103}
104
105/* May reduce *len if range is shorter. */
106static inline bool range_check(struct vringh *vrh, u64 addr, size_t *len,
107 struct vringh_range *range,
108 bool (*getrange)(struct vringh *,
109 u64, struct vringh_range *))
110{
111 if (addr < range->start || addr > range->end_incl) {
112 if (!getrange(vrh, addr, range))
113 return false;
114 }
115 BUG_ON(addr < range->start || addr > range->end_incl);
116
117 /* To end of memory? */
118 if (unlikely(addr + *len == 0)) {
119 if (range->end_incl == -1ULL)
120 return true;
121 goto truncate;
122 }
123
124 /* Otherwise, don't wrap. */
125 if (addr + *len < addr) {
126 vringh_bad("Wrapping descriptor %zu@0x%llx",
127 *len, (unsigned long long)addr);
128 return false;
129 }
130
131 if (unlikely(addr + *len - 1 > range->end_incl))
132 goto truncate;
133 return true;
134
135truncate:
136 *len = range->end_incl + 1 - addr;
137 return true;
138}
139
140static inline bool no_range_check(struct vringh *vrh, u64 addr, size_t *len,
141 struct vringh_range *range,
142 bool (*getrange)(struct vringh *,
143 u64, struct vringh_range *))
144{
145 return true;
146}
147
148/* No reason for this code to be inline. */
149static int move_to_indirect(const struct vringh *vrh,
150 int *up_next, u16 *i, void *addr,
151 const struct vring_desc *desc,
152 struct vring_desc **descs, int *desc_max)
153{
154 u32 len;
155
156 /* Indirect tables can't have indirect. */
157 if (*up_next != -1) {
158 vringh_bad("Multilevel indirect %u->%u", *up_next, *i);
159 return -EINVAL;
160 }
161
162 len = vringh32_to_cpu(vrh, desc->len);
163 if (unlikely(len % sizeof(struct vring_desc))) {
164 vringh_bad("Strange indirect len %u", desc->len);
165 return -EINVAL;
166 }
167
168 /* We will check this when we follow it! */
169 if (desc->flags & cpu_to_vringh16(vrh, VRING_DESC_F_NEXT))
170 *up_next = vringh16_to_cpu(vrh, desc->next);
171 else
172 *up_next = -2;
173 *descs = addr;
174 *desc_max = len / sizeof(struct vring_desc);
175
176 /* Now, start at the first indirect. */
177 *i = 0;
178 return 0;
179}
180
181static int resize_iovec(struct vringh_kiov *iov, gfp_t gfp)
182{
183 struct kvec *new;
184 unsigned int flag, new_num = (iov->max_num & ~VRINGH_IOV_ALLOCATED) * 2;
185
186 if (new_num < 8)
187 new_num = 8;
188
189 flag = (iov->max_num & VRINGH_IOV_ALLOCATED);
190 if (flag)
191 new = krealloc(iov->iov, new_num * sizeof(struct iovec), gfp);
192 else {
193 new = kmalloc(new_num * sizeof(struct iovec), gfp);
194 if (new) {
195 memcpy(new, iov->iov,
196 iov->max_num * sizeof(struct iovec));
197 flag = VRINGH_IOV_ALLOCATED;
198 }
199 }
200 if (!new)
201 return -ENOMEM;
202 iov->iov = new;
203 iov->max_num = (new_num | flag);
204 return 0;
205}
206
207static u16 __cold return_from_indirect(const struct vringh *vrh, int *up_next,
208 struct vring_desc **descs, int *desc_max)
209{
210 u16 i = *up_next;
211
212 *up_next = -1;
213 *descs = vrh->vring.desc;
214 *desc_max = vrh->vring.num;
215 return i;
216}
217
218static int slow_copy(struct vringh *vrh, void *dst, const void *src,
219 bool (*rcheck)(struct vringh *vrh, u64 addr, size_t *len,
220 struct vringh_range *range,
221 bool (*getrange)(struct vringh *vrh,
222 u64,
223 struct vringh_range *)),
224 bool (*getrange)(struct vringh *vrh,
225 u64 addr,
226 struct vringh_range *r),
227 struct vringh_range *range,
228 int (*copy)(void *dst, const void *src, size_t len))
229{
230 size_t part, len = sizeof(struct vring_desc);
231
232 do {
233 u64 addr;
234 int err;
235
236 part = len;
237 addr = (u64)(unsigned long)src - range->offset;
238
239 if (!rcheck(vrh, addr, &part, range, getrange))
240 return -EINVAL;
241
242 err = copy(dst, src, part);
243 if (err)
244 return err;
245
246 dst += part;
247 src += part;
248 len -= part;
249 } while (len);
250 return 0;
251}
252
253static inline int
254__vringh_iov(struct vringh *vrh, u16 i,
255 struct vringh_kiov *riov,
256 struct vringh_kiov *wiov,
257 bool (*rcheck)(struct vringh *vrh, u64 addr, size_t *len,
258 struct vringh_range *range,
259 bool (*getrange)(struct vringh *, u64,
260 struct vringh_range *)),
261 bool (*getrange)(struct vringh *, u64, struct vringh_range *),
262 gfp_t gfp,
263 int (*copy)(void *dst, const void *src, size_t len))
264{
265 int err, count = 0, up_next, desc_max;
266 struct vring_desc desc, *descs;
267 struct vringh_range range = { -1ULL, 0 }, slowrange;
268 bool slow = false;
269
270 /* We start traversing vring's descriptor table. */
271 descs = vrh->vring.desc;
272 desc_max = vrh->vring.num;
273 up_next = -1;
274
275 if (riov)
276 riov->i = riov->used = 0;
277 else if (wiov)
278 wiov->i = wiov->used = 0;
279 else
280 /* You must want something! */
281 BUG();
282
283 for (;;) {
284 void *addr;
285 struct vringh_kiov *iov;
286 size_t len;
287
288 if (unlikely(slow))
289 err = slow_copy(vrh, &desc, &descs[i], rcheck, getrange,
290 &slowrange, copy);
291 else
292 err = copy(&desc, &descs[i], sizeof(desc));
293 if (unlikely(err))
294 goto fail;
295
296 if (unlikely(desc.flags &
297 cpu_to_vringh16(vrh, VRING_DESC_F_INDIRECT))) {
298 u64 a = vringh64_to_cpu(vrh, desc.addr);
299
300 /* Make sure it's OK, and get offset. */
301 len = vringh32_to_cpu(vrh, desc.len);
302 if (!rcheck(vrh, a, &len, &range, getrange)) {
303 err = -EINVAL;
304 goto fail;
305 }
306
307 if (unlikely(len != vringh32_to_cpu(vrh, desc.len))) {
308 slow = true;
309 /* We need to save this range to use offset */
310 slowrange = range;
311 }
312
313 addr = (void *)(long)(a + range.offset);
314 err = move_to_indirect(vrh, &up_next, &i, addr, &desc,
315 &descs, &desc_max);
316 if (err)
317 goto fail;
318 continue;
319 }
320
321 if (count++ == vrh->vring.num) {
322 vringh_bad("Descriptor loop in %p", descs);
323 err = -ELOOP;
324 goto fail;
325 }
326
327 if (desc.flags & cpu_to_vringh16(vrh, VRING_DESC_F_WRITE))
328 iov = wiov;
329 else {
330 iov = riov;
331 if (unlikely(wiov && wiov->i)) {
332 vringh_bad("Readable desc %p after writable",
333 &descs[i]);
334 err = -EINVAL;
335 goto fail;
336 }
337 }
338
339 if (!iov) {
340 vringh_bad("Unexpected %s desc",
341 !wiov ? "writable" : "readable");
342 err = -EPROTO;
343 goto fail;
344 }
345
346 again:
347 /* Make sure it's OK, and get offset. */
348 len = vringh32_to_cpu(vrh, desc.len);
349 if (!rcheck(vrh, vringh64_to_cpu(vrh, desc.addr), &len, &range,
350 getrange)) {
351 err = -EINVAL;
352 goto fail;
353 }
354 addr = (void *)(unsigned long)(vringh64_to_cpu(vrh, desc.addr) +
355 range.offset);
356
357 if (unlikely(iov->used == (iov->max_num & ~VRINGH_IOV_ALLOCATED))) {
358 err = resize_iovec(iov, gfp);
359 if (err)
360 goto fail;
361 }
362
363 iov->iov[iov->used].iov_base = addr;
364 iov->iov[iov->used].iov_len = len;
365 iov->used++;
366
367 if (unlikely(len != vringh32_to_cpu(vrh, desc.len))) {
368 desc.len = cpu_to_vringh32(vrh,
369 vringh32_to_cpu(vrh, desc.len) - len);
370 desc.addr = cpu_to_vringh64(vrh,
371 vringh64_to_cpu(vrh, desc.addr) + len);
372 goto again;
373 }
374
375 if (desc.flags & cpu_to_vringh16(vrh, VRING_DESC_F_NEXT)) {
376 i = vringh16_to_cpu(vrh, desc.next);
377 } else {
378 /* Just in case we need to finish traversing above. */
379 if (unlikely(up_next > 0)) {
380 i = return_from_indirect(vrh, &up_next,
381 &descs, &desc_max);
382 slow = false;
383 } else
384 break;
385 }
386
387 if (i >= desc_max) {
388 vringh_bad("Chained index %u > %u", i, desc_max);
389 err = -EINVAL;
390 goto fail;
391 }
392 }
393
394 return 0;
395
396fail:
397 return err;
398}
399
400static inline int __vringh_complete(struct vringh *vrh,
401 const struct vring_used_elem *used,
402 unsigned int num_used,
403 int (*putu16)(const struct vringh *vrh,
404 __virtio16 *p, u16 val),
405 int (*putused)(struct vring_used_elem *dst,
406 const struct vring_used_elem
407 *src, unsigned num))
408{
409 struct vring_used *used_ring;
410 int err;
411 u16 used_idx, off;
412
413 used_ring = vrh->vring.used;
414 used_idx = vrh->last_used_idx + vrh->completed;
415
416 off = used_idx % vrh->vring.num;
417
418 /* Compiler knows num_used == 1 sometimes, hence extra check */
419 if (num_used > 1 && unlikely(off + num_used >= vrh->vring.num)) {
420 u16 part = vrh->vring.num - off;
421 err = putused(&used_ring->ring[off], used, part);
422 if (!err)
423 err = putused(&used_ring->ring[0], used + part,
424 num_used - part);
425 } else
426 err = putused(&used_ring->ring[off], used, num_used);
427
428 if (err) {
429 vringh_bad("Failed to write %u used entries %u at %p",
430 num_used, off, &used_ring->ring[off]);
431 return err;
432 }
433
434 /* Make sure buffer is written before we update index. */
435 virtio_wmb(vrh->weak_barriers);
436
437 err = putu16(vrh, &vrh->vring.used->idx, used_idx + num_used);
438 if (err) {
439 vringh_bad("Failed to update used index at %p",
440 &vrh->vring.used->idx);
441 return err;
442 }
443
444 vrh->completed += num_used;
445 return 0;
446}
447
448
449static inline int __vringh_need_notify(struct vringh *vrh,
450 int (*getu16)(const struct vringh *vrh,
451 u16 *val,
452 const __virtio16 *p))
453{
454 bool notify;
455 u16 used_event;
456 int err;
457
458 /* Flush out used index update. This is paired with the
459 * barrier that the Guest executes when enabling
460 * interrupts. */
461 virtio_mb(vrh->weak_barriers);
462
463 /* Old-style, without event indices. */
464 if (!vrh->event_indices) {
465 u16 flags;
466 err = getu16(vrh, &flags, &vrh->vring.avail->flags);
467 if (err) {
468 vringh_bad("Failed to get flags at %p",
469 &vrh->vring.avail->flags);
470 return err;
471 }
472 return (!(flags & VRING_AVAIL_F_NO_INTERRUPT));
473 }
474
475 /* Modern: we know when other side wants to know. */
476 err = getu16(vrh, &used_event, &vring_used_event(&vrh->vring));
477 if (err) {
478 vringh_bad("Failed to get used event idx at %p",
479 &vring_used_event(&vrh->vring));
480 return err;
481 }
482
483 /* Just in case we added so many that we wrap. */
484 if (unlikely(vrh->completed > 0xffff))
485 notify = true;
486 else
487 notify = vring_need_event(used_event,
488 vrh->last_used_idx + vrh->completed,
489 vrh->last_used_idx);
490
491 vrh->last_used_idx += vrh->completed;
492 vrh->completed = 0;
493 return notify;
494}
495
496static inline bool __vringh_notify_enable(struct vringh *vrh,
497 int (*getu16)(const struct vringh *vrh,
498 u16 *val, const __virtio16 *p),
499 int (*putu16)(const struct vringh *vrh,
500 __virtio16 *p, u16 val))
501{
502 u16 avail;
503
504 if (!vrh->event_indices) {
505 /* Old-school; update flags. */
506 if (putu16(vrh, &vrh->vring.used->flags, 0) != 0) {
507 vringh_bad("Clearing used flags %p",
508 &vrh->vring.used->flags);
509 return true;
510 }
511 } else {
512 if (putu16(vrh, &vring_avail_event(&vrh->vring),
513 vrh->last_avail_idx) != 0) {
514 vringh_bad("Updating avail event index %p",
515 &vring_avail_event(&vrh->vring));
516 return true;
517 }
518 }
519
520 /* They could have slipped one in as we were doing that: make
521 * sure it's written, then check again. */
522 virtio_mb(vrh->weak_barriers);
523
524 if (getu16(vrh, &avail, &vrh->vring.avail->idx) != 0) {
525 vringh_bad("Failed to check avail idx at %p",
526 &vrh->vring.avail->idx);
527 return true;
528 }
529
530 /* This is unlikely, so we just leave notifications enabled
531 * (if we're using event_indices, we'll only get one
532 * notification anyway). */
533 return avail == vrh->last_avail_idx;
534}
535
536static inline void __vringh_notify_disable(struct vringh *vrh,
537 int (*putu16)(const struct vringh *vrh,
538 __virtio16 *p, u16 val))
539{
540 if (!vrh->event_indices) {
541 /* Old-school; update flags. */
542 if (putu16(vrh, &vrh->vring.used->flags,
543 VRING_USED_F_NO_NOTIFY)) {
544 vringh_bad("Setting used flags %p",
545 &vrh->vring.used->flags);
546 }
547 }
548}
549
550/* Userspace access helpers: in this case, addresses are really userspace. */
551static inline int getu16_user(const struct vringh *vrh, u16 *val, const __virtio16 *p)
552{
553 __virtio16 v = 0;
554 int rc = get_user(v, (__force __virtio16 __user *)p);
555 *val = vringh16_to_cpu(vrh, v);
556 return rc;
557}
558
559static inline int putu16_user(const struct vringh *vrh, __virtio16 *p, u16 val)
560{
561 __virtio16 v = cpu_to_vringh16(vrh, val);
562 return put_user(v, (__force __virtio16 __user *)p);
563}
564
565static inline int copydesc_user(void *dst, const void *src, size_t len)
566{
567 return copy_from_user(dst, (__force void __user *)src, len) ?
568 -EFAULT : 0;
569}
570
571static inline int putused_user(struct vring_used_elem *dst,
572 const struct vring_used_elem *src,
573 unsigned int num)
574{
575 return copy_to_user((__force void __user *)dst, src,
576 sizeof(*dst) * num) ? -EFAULT : 0;
577}
578
579static inline int xfer_from_user(void *src, void *dst, size_t len)
580{
581 return copy_from_user(dst, (__force void __user *)src, len) ?
582 -EFAULT : 0;
583}
584
585static inline int xfer_to_user(void *dst, void *src, size_t len)
586{
587 return copy_to_user((__force void __user *)dst, src, len) ?
588 -EFAULT : 0;
589}
590
591/**
592 * vringh_init_user - initialize a vringh for a userspace vring.
593 * @vrh: the vringh to initialize.
594 * @features: the feature bits for this ring.
595 * @num: the number of elements.
596 * @weak_barriers: true if we only need memory barriers, not I/O.
597 * @desc: the userpace descriptor pointer.
598 * @avail: the userpace avail pointer.
599 * @used: the userpace used pointer.
600 *
601 * Returns an error if num is invalid: you should check pointers
602 * yourself!
603 */
604int vringh_init_user(struct vringh *vrh, u64 features,
605 unsigned int num, bool weak_barriers,
606 struct vring_desc __user *desc,
607 struct vring_avail __user *avail,
608 struct vring_used __user *used)
609{
610 /* Sane power of 2 please! */
611 if (!num || num > 0xffff || (num & (num - 1))) {
612 vringh_bad("Bad ring size %u", num);
613 return -EINVAL;
614 }
615
616 vrh->little_endian = (features & (1ULL << VIRTIO_F_VERSION_1));
617 vrh->event_indices = (features & (1 << VIRTIO_RING_F_EVENT_IDX));
618 vrh->weak_barriers = weak_barriers;
619 vrh->completed = 0;
620 vrh->last_avail_idx = 0;
621 vrh->last_used_idx = 0;
622 vrh->vring.num = num;
623 /* vring expects kernel addresses, but only used via accessors. */
624 vrh->vring.desc = (__force struct vring_desc *)desc;
625 vrh->vring.avail = (__force struct vring_avail *)avail;
626 vrh->vring.used = (__force struct vring_used *)used;
627 return 0;
628}
629EXPORT_SYMBOL(vringh_init_user);
630
631/**
632 * vringh_getdesc_user - get next available descriptor from userspace ring.
633 * @vrh: the userspace vring.
634 * @riov: where to put the readable descriptors (or NULL)
635 * @wiov: where to put the writable descriptors (or NULL)
636 * @getrange: function to call to check ranges.
637 * @head: head index we received, for passing to vringh_complete_user().
638 *
639 * Returns 0 if there was no descriptor, 1 if there was, or -errno.
640 *
641 * Note that on error return, you can tell the difference between an
642 * invalid ring and a single invalid descriptor: in the former case,
643 * *head will be vrh->vring.num. You may be able to ignore an invalid
644 * descriptor, but there's not much you can do with an invalid ring.
645 *
646 * Note that you may need to clean up riov and wiov, even on error!
647 */
648int vringh_getdesc_user(struct vringh *vrh,
649 struct vringh_iov *riov,
650 struct vringh_iov *wiov,
651 bool (*getrange)(struct vringh *vrh,
652 u64 addr, struct vringh_range *r),
653 u16 *head)
654{
655 int err;
656
657 *head = vrh->vring.num;
658 err = __vringh_get_head(vrh, getu16_user, &vrh->last_avail_idx);
659 if (err < 0)
660 return err;
661
662 /* Empty... */
663 if (err == vrh->vring.num)
664 return 0;
665
666 /* We need the layouts to be the identical for this to work */
667 BUILD_BUG_ON(sizeof(struct vringh_kiov) != sizeof(struct vringh_iov));
668 BUILD_BUG_ON(offsetof(struct vringh_kiov, iov) !=
669 offsetof(struct vringh_iov, iov));
670 BUILD_BUG_ON(offsetof(struct vringh_kiov, i) !=
671 offsetof(struct vringh_iov, i));
672 BUILD_BUG_ON(offsetof(struct vringh_kiov, used) !=
673 offsetof(struct vringh_iov, used));
674 BUILD_BUG_ON(offsetof(struct vringh_kiov, max_num) !=
675 offsetof(struct vringh_iov, max_num));
676 BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
677 BUILD_BUG_ON(offsetof(struct iovec, iov_base) !=
678 offsetof(struct kvec, iov_base));
679 BUILD_BUG_ON(offsetof(struct iovec, iov_len) !=
680 offsetof(struct kvec, iov_len));
681 BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_base)
682 != sizeof(((struct kvec *)NULL)->iov_base));
683 BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_len)
684 != sizeof(((struct kvec *)NULL)->iov_len));
685
686 *head = err;
687 err = __vringh_iov(vrh, *head, (struct vringh_kiov *)riov,
688 (struct vringh_kiov *)wiov,
689 range_check, getrange, GFP_KERNEL, copydesc_user);
690 if (err)
691 return err;
692
693 return 1;
694}
695EXPORT_SYMBOL(vringh_getdesc_user);
696
697/**
698 * vringh_iov_pull_user - copy bytes from vring_iov.
699 * @riov: the riov as passed to vringh_getdesc_user() (updated as we consume)
700 * @dst: the place to copy.
701 * @len: the maximum length to copy.
702 *
703 * Returns the bytes copied <= len or a negative errno.
704 */
705ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
706{
707 return vringh_iov_xfer((struct vringh_kiov *)riov,
708 dst, len, xfer_from_user);
709}
710EXPORT_SYMBOL(vringh_iov_pull_user);
711
712/**
713 * vringh_iov_push_user - copy bytes into vring_iov.
714 * @wiov: the wiov as passed to vringh_getdesc_user() (updated as we consume)
715 * @dst: the place to copy.
716 * @len: the maximum length to copy.
717 *
718 * Returns the bytes copied <= len or a negative errno.
719 */
720ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
721 const void *src, size_t len)
722{
723 return vringh_iov_xfer((struct vringh_kiov *)wiov,
724 (void *)src, len, xfer_to_user);
725}
726EXPORT_SYMBOL(vringh_iov_push_user);
727
728/**
729 * vringh_abandon_user - we've decided not to handle the descriptor(s).
730 * @vrh: the vring.
731 * @num: the number of descriptors to put back (ie. num
732 * vringh_get_user() to undo).
733 *
734 * The next vringh_get_user() will return the old descriptor(s) again.
735 */
736void vringh_abandon_user(struct vringh *vrh, unsigned int num)
737{
738 /* We only update vring_avail_event(vr) when we want to be notified,
739 * so we haven't changed that yet. */
740 vrh->last_avail_idx -= num;
741}
742EXPORT_SYMBOL(vringh_abandon_user);
743
744/**
745 * vringh_complete_user - we've finished with descriptor, publish it.
746 * @vrh: the vring.
747 * @head: the head as filled in by vringh_getdesc_user.
748 * @len: the length of data we have written.
749 *
750 * You should check vringh_need_notify_user() after one or more calls
751 * to this function.
752 */
753int vringh_complete_user(struct vringh *vrh, u16 head, u32 len)
754{
755 struct vring_used_elem used;
756
757 used.id = cpu_to_vringh32(vrh, head);
758 used.len = cpu_to_vringh32(vrh, len);
759 return __vringh_complete(vrh, &used, 1, putu16_user, putused_user);
760}
761EXPORT_SYMBOL(vringh_complete_user);
762
763/**
764 * vringh_complete_multi_user - we've finished with many descriptors.
765 * @vrh: the vring.
766 * @used: the head, length pairs.
767 * @num_used: the number of used elements.
768 *
769 * You should check vringh_need_notify_user() after one or more calls
770 * to this function.
771 */
772int vringh_complete_multi_user(struct vringh *vrh,
773 const struct vring_used_elem used[],
774 unsigned num_used)
775{
776 return __vringh_complete(vrh, used, num_used,
777 putu16_user, putused_user);
778}
779EXPORT_SYMBOL(vringh_complete_multi_user);
780
781/**
782 * vringh_notify_enable_user - we want to know if something changes.
783 * @vrh: the vring.
784 *
785 * This always enables notifications, but returns false if there are
786 * now more buffers available in the vring.
787 */
788bool vringh_notify_enable_user(struct vringh *vrh)
789{
790 return __vringh_notify_enable(vrh, getu16_user, putu16_user);
791}
792EXPORT_SYMBOL(vringh_notify_enable_user);
793
794/**
795 * vringh_notify_disable_user - don't tell us if something changes.
796 * @vrh: the vring.
797 *
798 * This is our normal running state: we disable and then only enable when
799 * we're going to sleep.
800 */
801void vringh_notify_disable_user(struct vringh *vrh)
802{
803 __vringh_notify_disable(vrh, putu16_user);
804}
805EXPORT_SYMBOL(vringh_notify_disable_user);
806
807/**
808 * vringh_need_notify_user - must we tell the other side about used buffers?
809 * @vrh: the vring we've called vringh_complete_user() on.
810 *
811 * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
812 */
813int vringh_need_notify_user(struct vringh *vrh)
814{
815 return __vringh_need_notify(vrh, getu16_user);
816}
817EXPORT_SYMBOL(vringh_need_notify_user);
818
819/* Kernelspace access helpers. */
820static inline int getu16_kern(const struct vringh *vrh,
821 u16 *val, const __virtio16 *p)
822{
823 *val = vringh16_to_cpu(vrh, ACCESS_ONCE(*p));
824 return 0;
825}
826
827static inline int putu16_kern(const struct vringh *vrh, __virtio16 *p, u16 val)
828{
829 ACCESS_ONCE(*p) = cpu_to_vringh16(vrh, val);
830 return 0;
831}
832
833static inline int copydesc_kern(void *dst, const void *src, size_t len)
834{
835 memcpy(dst, src, len);
836 return 0;
837}
838
839static inline int putused_kern(struct vring_used_elem *dst,
840 const struct vring_used_elem *src,
841 unsigned int num)
842{
843 memcpy(dst, src, num * sizeof(*dst));
844 return 0;
845}
846
847static inline int xfer_kern(void *src, void *dst, size_t len)
848{
849 memcpy(dst, src, len);
850 return 0;
851}
852
853/**
854 * vringh_init_kern - initialize a vringh for a kernelspace vring.
855 * @vrh: the vringh to initialize.
856 * @features: the feature bits for this ring.
857 * @num: the number of elements.
858 * @weak_barriers: true if we only need memory barriers, not I/O.
859 * @desc: the userpace descriptor pointer.
860 * @avail: the userpace avail pointer.
861 * @used: the userpace used pointer.
862 *
863 * Returns an error if num is invalid.
864 */
865int vringh_init_kern(struct vringh *vrh, u64 features,
866 unsigned int num, bool weak_barriers,
867 struct vring_desc *desc,
868 struct vring_avail *avail,
869 struct vring_used *used)
870{
871 /* Sane power of 2 please! */
872 if (!num || num > 0xffff || (num & (num - 1))) {
873 vringh_bad("Bad ring size %u", num);
874 return -EINVAL;
875 }
876
877 vrh->little_endian = (features & (1ULL << VIRTIO_F_VERSION_1));
878 vrh->event_indices = (features & (1 << VIRTIO_RING_F_EVENT_IDX));
879 vrh->weak_barriers = weak_barriers;
880 vrh->completed = 0;
881 vrh->last_avail_idx = 0;
882 vrh->last_used_idx = 0;
883 vrh->vring.num = num;
884 vrh->vring.desc = desc;
885 vrh->vring.avail = avail;
886 vrh->vring.used = used;
887 return 0;
888}
889EXPORT_SYMBOL(vringh_init_kern);
890
891/**
892 * vringh_getdesc_kern - get next available descriptor from kernelspace ring.
893 * @vrh: the kernelspace vring.
894 * @riov: where to put the readable descriptors (or NULL)
895 * @wiov: where to put the writable descriptors (or NULL)
896 * @head: head index we received, for passing to vringh_complete_kern().
897 * @gfp: flags for allocating larger riov/wiov.
898 *
899 * Returns 0 if there was no descriptor, 1 if there was, or -errno.
900 *
901 * Note that on error return, you can tell the difference between an
902 * invalid ring and a single invalid descriptor: in the former case,
903 * *head will be vrh->vring.num. You may be able to ignore an invalid
904 * descriptor, but there's not much you can do with an invalid ring.
905 *
906 * Note that you may need to clean up riov and wiov, even on error!
907 */
908int vringh_getdesc_kern(struct vringh *vrh,
909 struct vringh_kiov *riov,
910 struct vringh_kiov *wiov,
911 u16 *head,
912 gfp_t gfp)
913{
914 int err;
915
916 err = __vringh_get_head(vrh, getu16_kern, &vrh->last_avail_idx);
917 if (err < 0)
918 return err;
919
920 /* Empty... */
921 if (err == vrh->vring.num)
922 return 0;
923
924 *head = err;
925 err = __vringh_iov(vrh, *head, riov, wiov, no_range_check, NULL,
926 gfp, copydesc_kern);
927 if (err)
928 return err;
929
930 return 1;
931}
932EXPORT_SYMBOL(vringh_getdesc_kern);
933
934/**
935 * vringh_iov_pull_kern - copy bytes from vring_iov.
936 * @riov: the riov as passed to vringh_getdesc_kern() (updated as we consume)
937 * @dst: the place to copy.
938 * @len: the maximum length to copy.
939 *
940 * Returns the bytes copied <= len or a negative errno.
941 */
942ssize_t vringh_iov_pull_kern(struct vringh_kiov *riov, void *dst, size_t len)
943{
944 return vringh_iov_xfer(riov, dst, len, xfer_kern);
945}
946EXPORT_SYMBOL(vringh_iov_pull_kern);
947
948/**
949 * vringh_iov_push_kern - copy bytes into vring_iov.
950 * @wiov: the wiov as passed to vringh_getdesc_kern() (updated as we consume)
951 * @dst: the place to copy.
952 * @len: the maximum length to copy.
953 *
954 * Returns the bytes copied <= len or a negative errno.
955 */
956ssize_t vringh_iov_push_kern(struct vringh_kiov *wiov,
957 const void *src, size_t len)
958{
959 return vringh_iov_xfer(wiov, (void *)src, len, xfer_kern);
960}
961EXPORT_SYMBOL(vringh_iov_push_kern);
962
963/**
964 * vringh_abandon_kern - we've decided not to handle the descriptor(s).
965 * @vrh: the vring.
966 * @num: the number of descriptors to put back (ie. num
967 * vringh_get_kern() to undo).
968 *
969 * The next vringh_get_kern() will return the old descriptor(s) again.
970 */
971void vringh_abandon_kern(struct vringh *vrh, unsigned int num)
972{
973 /* We only update vring_avail_event(vr) when we want to be notified,
974 * so we haven't changed that yet. */
975 vrh->last_avail_idx -= num;
976}
977EXPORT_SYMBOL(vringh_abandon_kern);
978
979/**
980 * vringh_complete_kern - we've finished with descriptor, publish it.
981 * @vrh: the vring.
982 * @head: the head as filled in by vringh_getdesc_kern.
983 * @len: the length of data we have written.
984 *
985 * You should check vringh_need_notify_kern() after one or more calls
986 * to this function.
987 */
988int vringh_complete_kern(struct vringh *vrh, u16 head, u32 len)
989{
990 struct vring_used_elem used;
991
992 used.id = cpu_to_vringh32(vrh, head);
993 used.len = cpu_to_vringh32(vrh, len);
994
995 return __vringh_complete(vrh, &used, 1, putu16_kern, putused_kern);
996}
997EXPORT_SYMBOL(vringh_complete_kern);
998
999/**
1000 * vringh_notify_enable_kern - we want to know if something changes.
1001 * @vrh: the vring.
1002 *
1003 * This always enables notifications, but returns false if there are
1004 * now more buffers available in the vring.
1005 */
1006bool vringh_notify_enable_kern(struct vringh *vrh)
1007{
1008 return __vringh_notify_enable(vrh, getu16_kern, putu16_kern);
1009}
1010EXPORT_SYMBOL(vringh_notify_enable_kern);
1011
1012/**
1013 * vringh_notify_disable_kern - don't tell us if something changes.
1014 * @vrh: the vring.
1015 *
1016 * This is our normal running state: we disable and then only enable when
1017 * we're going to sleep.
1018 */
1019void vringh_notify_disable_kern(struct vringh *vrh)
1020{
1021 __vringh_notify_disable(vrh, putu16_kern);
1022}
1023EXPORT_SYMBOL(vringh_notify_disable_kern);
1024
1025/**
1026 * vringh_need_notify_kern - must we tell the other side about used buffers?
1027 * @vrh: the vring we've called vringh_complete_kern() on.
1028 *
1029 * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
1030 */
1031int vringh_need_notify_kern(struct vringh *vrh)
1032{
1033 return __vringh_need_notify(vrh, getu16_kern);
1034}
1035EXPORT_SYMBOL(vringh_need_notify_kern);
1036
1037MODULE_LICENSE("GPL");