1 #include <linux/export.h>
3 #include <linux/pagemap.h>
4 #include <linux/slab.h>
5 #include <linux/vmalloc.h>
7 #define iterate_iovec(i, n, __v, __p, skip, STEP) { \
11 __v.iov_len = min(n, __p->iov_len - skip); \
12 if (likely(__v.iov_len)) { \
13 __v.iov_base = __p->iov_base + skip; \
15 __v.iov_len -= left; \
16 skip += __v.iov_len; \
21 while (unlikely(!left && n)) { \
23 __v.iov_len = min(n, __p->iov_len); \
24 if (unlikely(!__v.iov_len)) \
26 __v.iov_base = __p->iov_base; \
28 __v.iov_len -= left; \
35 #define iterate_bvec(i, n, __v, __p, skip, STEP) { \
38 __v.bv_len = min_t(size_t, n, __p->bv_len - skip); \
39 if (likely(__v.bv_len)) { \
40 __v.bv_page = __p->bv_page; \
41 __v.bv_offset = __p->bv_offset + skip; \
46 while (unlikely(n)) { \
48 __v.bv_len = min_t(size_t, n, __p->bv_len); \
49 if (unlikely(!__v.bv_len)) \
51 __v.bv_page = __p->bv_page; \
52 __v.bv_offset = __p->bv_offset; \
60 #define iterate_all_kinds(i, n, v, I, B) { \
61 size_t skip = i->iov_offset; \
62 if (unlikely(i->type & ITER_BVEC)) { \
63 const struct bio_vec *bvec; \
65 iterate_bvec(i, n, v, bvec, skip, (B)) \
67 const struct iovec *iov; \
69 iterate_iovec(i, n, v, iov, skip, (I)) \
73 #define iterate_and_advance(i, n, v, I, B) { \
74 size_t skip = i->iov_offset; \
75 if (unlikely(i->type & ITER_BVEC)) { \
76 const struct bio_vec *bvec; \
78 iterate_bvec(i, n, v, bvec, skip, (B)) \
79 if (skip == bvec->bv_len) { \
83 i->nr_segs -= bvec - i->bvec; \
86 const struct iovec *iov; \
88 iterate_iovec(i, n, v, iov, skip, (I)) \
89 if (skip == iov->iov_len) { \
93 i->nr_segs -= iov - i->iov; \
97 i->iov_offset = skip; \
100 static size_t copy_to_iter_iovec(void *from
, size_t bytes
, struct iov_iter
*i
)
102 size_t skip
, copy
, left
, wanted
;
103 const struct iovec
*iov
;
106 if (unlikely(bytes
> i
->count
))
109 if (unlikely(!bytes
))
114 skip
= i
->iov_offset
;
115 buf
= iov
->iov_base
+ skip
;
116 copy
= min(bytes
, iov
->iov_len
- skip
);
118 left
= __copy_to_user(buf
, from
, copy
);
123 while (unlikely(!left
&& bytes
)) {
126 copy
= min(bytes
, iov
->iov_len
);
127 left
= __copy_to_user(buf
, from
, copy
);
134 if (skip
== iov
->iov_len
) {
138 i
->count
-= wanted
- bytes
;
139 i
->nr_segs
-= iov
- i
->iov
;
141 i
->iov_offset
= skip
;
142 return wanted
- bytes
;
145 static size_t copy_from_iter_iovec(void *to
, size_t bytes
, struct iov_iter
*i
)
147 size_t skip
, copy
, left
, wanted
;
148 const struct iovec
*iov
;
151 if (unlikely(bytes
> i
->count
))
154 if (unlikely(!bytes
))
159 skip
= i
->iov_offset
;
160 buf
= iov
->iov_base
+ skip
;
161 copy
= min(bytes
, iov
->iov_len
- skip
);
163 left
= __copy_from_user(to
, buf
, copy
);
168 while (unlikely(!left
&& bytes
)) {
171 copy
= min(bytes
, iov
->iov_len
);
172 left
= __copy_from_user(to
, buf
, copy
);
179 if (skip
== iov
->iov_len
) {
183 i
->count
-= wanted
- bytes
;
184 i
->nr_segs
-= iov
- i
->iov
;
186 i
->iov_offset
= skip
;
187 return wanted
- bytes
;
190 static size_t copy_page_to_iter_iovec(struct page
*page
, size_t offset
, size_t bytes
,
193 size_t skip
, copy
, left
, wanted
;
194 const struct iovec
*iov
;
198 if (unlikely(bytes
> i
->count
))
201 if (unlikely(!bytes
))
206 skip
= i
->iov_offset
;
207 buf
= iov
->iov_base
+ skip
;
208 copy
= min(bytes
, iov
->iov_len
- skip
);
210 if (!fault_in_pages_writeable(buf
, copy
)) {
211 kaddr
= kmap_atomic(page
);
212 from
= kaddr
+ offset
;
214 /* first chunk, usually the only one */
215 left
= __copy_to_user_inatomic(buf
, from
, copy
);
221 while (unlikely(!left
&& bytes
)) {
224 copy
= min(bytes
, iov
->iov_len
);
225 left
= __copy_to_user_inatomic(buf
, from
, copy
);
231 if (likely(!bytes
)) {
232 kunmap_atomic(kaddr
);
235 offset
= from
- kaddr
;
237 kunmap_atomic(kaddr
);
238 copy
= min(bytes
, iov
->iov_len
- skip
);
240 /* Too bad - revert to non-atomic kmap */
242 from
= kaddr
+ offset
;
243 left
= __copy_to_user(buf
, from
, copy
);
248 while (unlikely(!left
&& bytes
)) {
251 copy
= min(bytes
, iov
->iov_len
);
252 left
= __copy_to_user(buf
, from
, copy
);
260 if (skip
== iov
->iov_len
) {
264 i
->count
-= wanted
- bytes
;
265 i
->nr_segs
-= iov
- i
->iov
;
267 i
->iov_offset
= skip
;
268 return wanted
- bytes
;
271 static size_t copy_page_from_iter_iovec(struct page
*page
, size_t offset
, size_t bytes
,
274 size_t skip
, copy
, left
, wanted
;
275 const struct iovec
*iov
;
279 if (unlikely(bytes
> i
->count
))
282 if (unlikely(!bytes
))
287 skip
= i
->iov_offset
;
288 buf
= iov
->iov_base
+ skip
;
289 copy
= min(bytes
, iov
->iov_len
- skip
);
291 if (!fault_in_pages_readable(buf
, copy
)) {
292 kaddr
= kmap_atomic(page
);
295 /* first chunk, usually the only one */
296 left
= __copy_from_user_inatomic(to
, buf
, copy
);
302 while (unlikely(!left
&& bytes
)) {
305 copy
= min(bytes
, iov
->iov_len
);
306 left
= __copy_from_user_inatomic(to
, buf
, copy
);
312 if (likely(!bytes
)) {
313 kunmap_atomic(kaddr
);
318 kunmap_atomic(kaddr
);
319 copy
= min(bytes
, iov
->iov_len
- skip
);
321 /* Too bad - revert to non-atomic kmap */
324 left
= __copy_from_user(to
, buf
, copy
);
329 while (unlikely(!left
&& bytes
)) {
332 copy
= min(bytes
, iov
->iov_len
);
333 left
= __copy_from_user(to
, buf
, copy
);
341 if (skip
== iov
->iov_len
) {
345 i
->count
-= wanted
- bytes
;
346 i
->nr_segs
-= iov
- i
->iov
;
348 i
->iov_offset
= skip
;
349 return wanted
- bytes
;
352 static size_t zero_iovec(size_t bytes
, struct iov_iter
*i
)
354 size_t skip
, copy
, left
, wanted
;
355 const struct iovec
*iov
;
358 if (unlikely(bytes
> i
->count
))
361 if (unlikely(!bytes
))
366 skip
= i
->iov_offset
;
367 buf
= iov
->iov_base
+ skip
;
368 copy
= min(bytes
, iov
->iov_len
- skip
);
370 left
= __clear_user(buf
, copy
);
375 while (unlikely(!left
&& bytes
)) {
378 copy
= min(bytes
, iov
->iov_len
);
379 left
= __clear_user(buf
, copy
);
385 if (skip
== iov
->iov_len
) {
389 i
->count
-= wanted
- bytes
;
390 i
->nr_segs
-= iov
- i
->iov
;
392 i
->iov_offset
= skip
;
393 return wanted
- bytes
;
397 * Fault in the first iovec of the given iov_iter, to a maximum length
398 * of bytes. Returns 0 on success, or non-zero if the memory could not be
399 * accessed (ie. because it is an invalid address).
401 * writev-intensive code may want this to prefault several iovecs -- that
402 * would be possible (callers must not rely on the fact that _only_ the
403 * first iovec will be faulted with the current implementation).
405 int iov_iter_fault_in_readable(struct iov_iter
*i
, size_t bytes
)
407 if (!(i
->type
& ITER_BVEC
)) {
408 char __user
*buf
= i
->iov
->iov_base
+ i
->iov_offset
;
409 bytes
= min(bytes
, i
->iov
->iov_len
- i
->iov_offset
);
410 return fault_in_pages_readable(buf
, bytes
);
414 EXPORT_SYMBOL(iov_iter_fault_in_readable
);
416 void iov_iter_init(struct iov_iter
*i
, int direction
,
417 const struct iovec
*iov
, unsigned long nr_segs
,
420 /* It will get better. Eventually... */
421 if (segment_eq(get_fs(), KERNEL_DS
))
422 direction
|= ITER_KVEC
;
425 i
->nr_segs
= nr_segs
;
429 EXPORT_SYMBOL(iov_iter_init
);
431 static ssize_t
get_pages_alloc_iovec(struct iov_iter
*i
,
432 struct page
***pages
, size_t maxsize
,
435 size_t offset
= i
->iov_offset
;
436 const struct iovec
*iov
= i
->iov
;
443 len
= iov
->iov_len
- offset
;
448 addr
= (unsigned long)iov
->iov_base
+ offset
;
449 len
+= *start
= addr
& (PAGE_SIZE
- 1);
450 addr
&= ~(PAGE_SIZE
- 1);
451 n
= (len
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
453 p
= kmalloc(n
* sizeof(struct page
*), GFP_KERNEL
);
455 p
= vmalloc(n
* sizeof(struct page
*));
459 res
= get_user_pages_fast(addr
, n
, (i
->type
& WRITE
) != WRITE
, p
);
460 if (unlikely(res
< 0)) {
465 return (res
== n
? len
: res
* PAGE_SIZE
) - *start
;
468 static void memcpy_from_page(char *to
, struct page
*page
, size_t offset
, size_t len
)
470 char *from
= kmap_atomic(page
);
471 memcpy(to
, from
+ offset
, len
);
475 static void memcpy_to_page(struct page
*page
, size_t offset
, char *from
, size_t len
)
477 char *to
= kmap_atomic(page
);
478 memcpy(to
+ offset
, from
, len
);
482 static void memzero_page(struct page
*page
, size_t offset
, size_t len
)
484 char *addr
= kmap_atomic(page
);
485 memset(addr
+ offset
, 0, len
);
489 static size_t copy_to_iter_bvec(void *from
, size_t bytes
, struct iov_iter
*i
)
491 size_t skip
, copy
, wanted
;
492 const struct bio_vec
*bvec
;
494 if (unlikely(bytes
> i
->count
))
497 if (unlikely(!bytes
))
502 skip
= i
->iov_offset
;
503 copy
= min_t(size_t, bytes
, bvec
->bv_len
- skip
);
505 memcpy_to_page(bvec
->bv_page
, skip
+ bvec
->bv_offset
, from
, copy
);
511 copy
= min(bytes
, (size_t)bvec
->bv_len
);
512 memcpy_to_page(bvec
->bv_page
, bvec
->bv_offset
, from
, copy
);
517 if (skip
== bvec
->bv_len
) {
521 i
->count
-= wanted
- bytes
;
522 i
->nr_segs
-= bvec
- i
->bvec
;
524 i
->iov_offset
= skip
;
525 return wanted
- bytes
;
528 static size_t copy_from_iter_bvec(void *to
, size_t bytes
, struct iov_iter
*i
)
530 size_t skip
, copy
, wanted
;
531 const struct bio_vec
*bvec
;
533 if (unlikely(bytes
> i
->count
))
536 if (unlikely(!bytes
))
541 skip
= i
->iov_offset
;
543 copy
= min(bytes
, bvec
->bv_len
- skip
);
545 memcpy_from_page(to
, bvec
->bv_page
, bvec
->bv_offset
+ skip
, copy
);
553 copy
= min(bytes
, (size_t)bvec
->bv_len
);
554 memcpy_from_page(to
, bvec
->bv_page
, bvec
->bv_offset
, copy
);
559 if (skip
== bvec
->bv_len
) {
564 i
->nr_segs
-= bvec
- i
->bvec
;
566 i
->iov_offset
= skip
;
570 static size_t copy_page_to_iter_bvec(struct page
*page
, size_t offset
,
571 size_t bytes
, struct iov_iter
*i
)
573 void *kaddr
= kmap_atomic(page
);
574 size_t wanted
= copy_to_iter_bvec(kaddr
+ offset
, bytes
, i
);
575 kunmap_atomic(kaddr
);
579 static size_t copy_page_from_iter_bvec(struct page
*page
, size_t offset
,
580 size_t bytes
, struct iov_iter
*i
)
582 void *kaddr
= kmap_atomic(page
);
583 size_t wanted
= copy_from_iter_bvec(kaddr
+ offset
, bytes
, i
);
584 kunmap_atomic(kaddr
);
588 static size_t zero_bvec(size_t bytes
, struct iov_iter
*i
)
590 size_t skip
, copy
, wanted
;
591 const struct bio_vec
*bvec
;
593 if (unlikely(bytes
> i
->count
))
596 if (unlikely(!bytes
))
601 skip
= i
->iov_offset
;
602 copy
= min_t(size_t, bytes
, bvec
->bv_len
- skip
);
604 memzero_page(bvec
->bv_page
, skip
+ bvec
->bv_offset
, copy
);
609 copy
= min(bytes
, (size_t)bvec
->bv_len
);
610 memzero_page(bvec
->bv_page
, bvec
->bv_offset
, copy
);
614 if (skip
== bvec
->bv_len
) {
618 i
->count
-= wanted
- bytes
;
619 i
->nr_segs
-= bvec
- i
->bvec
;
621 i
->iov_offset
= skip
;
622 return wanted
- bytes
;
625 static ssize_t
get_pages_alloc_bvec(struct iov_iter
*i
,
626 struct page
***pages
, size_t maxsize
,
629 const struct bio_vec
*bvec
= i
->bvec
;
630 size_t len
= bvec
->bv_len
- i
->iov_offset
;
635 *start
= bvec
->bv_offset
+ i
->iov_offset
;
637 *pages
= kmalloc(sizeof(struct page
*), GFP_KERNEL
);
641 get_page(**pages
= bvec
->bv_page
);
646 size_t copy_page_to_iter(struct page
*page
, size_t offset
, size_t bytes
,
649 if (i
->type
& ITER_BVEC
)
650 return copy_page_to_iter_bvec(page
, offset
, bytes
, i
);
652 return copy_page_to_iter_iovec(page
, offset
, bytes
, i
);
654 EXPORT_SYMBOL(copy_page_to_iter
);
656 size_t copy_page_from_iter(struct page
*page
, size_t offset
, size_t bytes
,
659 if (i
->type
& ITER_BVEC
)
660 return copy_page_from_iter_bvec(page
, offset
, bytes
, i
);
662 return copy_page_from_iter_iovec(page
, offset
, bytes
, i
);
664 EXPORT_SYMBOL(copy_page_from_iter
);
666 size_t copy_to_iter(void *addr
, size_t bytes
, struct iov_iter
*i
)
668 if (i
->type
& ITER_BVEC
)
669 return copy_to_iter_bvec(addr
, bytes
, i
);
671 return copy_to_iter_iovec(addr
, bytes
, i
);
673 EXPORT_SYMBOL(copy_to_iter
);
675 size_t copy_from_iter(void *addr
, size_t bytes
, struct iov_iter
*i
)
677 if (i
->type
& ITER_BVEC
)
678 return copy_from_iter_bvec(addr
, bytes
, i
);
680 return copy_from_iter_iovec(addr
, bytes
, i
);
682 EXPORT_SYMBOL(copy_from_iter
);
684 size_t iov_iter_zero(size_t bytes
, struct iov_iter
*i
)
686 if (i
->type
& ITER_BVEC
) {
687 return zero_bvec(bytes
, i
);
689 return zero_iovec(bytes
, i
);
692 EXPORT_SYMBOL(iov_iter_zero
);
694 size_t iov_iter_copy_from_user_atomic(struct page
*page
,
695 struct iov_iter
*i
, unsigned long offset
, size_t bytes
)
697 char *kaddr
= kmap_atomic(page
), *p
= kaddr
+ offset
;
698 iterate_all_kinds(i
, bytes
, v
,
699 __copy_from_user_inatomic((p
+= v
.iov_len
) - v
.iov_len
,
700 v
.iov_base
, v
.iov_len
),
701 memcpy_from_page((p
+= v
.bv_len
) - v
.bv_len
, v
.bv_page
,
702 v
.bv_offset
, v
.bv_len
)
704 kunmap_atomic(kaddr
);
707 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic
);
709 void iov_iter_advance(struct iov_iter
*i
, size_t size
)
711 iterate_and_advance(i
, size
, v
, 0, 0)
713 EXPORT_SYMBOL(iov_iter_advance
);
716 * Return the count of just the current iov_iter segment.
718 size_t iov_iter_single_seg_count(const struct iov_iter
*i
)
722 else if (i
->type
& ITER_BVEC
)
723 return min(i
->count
, i
->bvec
->bv_len
- i
->iov_offset
);
725 return min(i
->count
, i
->iov
->iov_len
- i
->iov_offset
);
727 EXPORT_SYMBOL(iov_iter_single_seg_count
);
729 unsigned long iov_iter_alignment(const struct iov_iter
*i
)
731 unsigned long res
= 0;
732 size_t size
= i
->count
;
737 iterate_all_kinds(i
, size
, v
,
738 (res
|= (unsigned long)v
.iov_base
| v
.iov_len
, 0),
739 res
|= v
.bv_offset
| v
.bv_len
743 EXPORT_SYMBOL(iov_iter_alignment
);
745 ssize_t
iov_iter_get_pages(struct iov_iter
*i
,
746 struct page
**pages
, size_t maxsize
, unsigned maxpages
,
749 if (maxsize
> i
->count
)
755 iterate_all_kinds(i
, maxsize
, v
, ({
756 unsigned long addr
= (unsigned long)v
.iov_base
;
757 size_t len
= v
.iov_len
+ (*start
= addr
& (PAGE_SIZE
- 1));
761 if (len
> maxpages
* PAGE_SIZE
)
762 len
= maxpages
* PAGE_SIZE
;
763 addr
&= ~(PAGE_SIZE
- 1);
764 n
= DIV_ROUND_UP(len
, PAGE_SIZE
);
765 res
= get_user_pages_fast(addr
, n
, (i
->type
& WRITE
) != WRITE
, pages
);
766 if (unlikely(res
< 0))
768 return (res
== n
? len
: res
* PAGE_SIZE
) - *start
;
770 /* can't be more than PAGE_SIZE */
771 *start
= v
.bv_offset
;
772 get_page(*pages
= v
.bv_page
);
778 EXPORT_SYMBOL(iov_iter_get_pages
);
780 ssize_t
iov_iter_get_pages_alloc(struct iov_iter
*i
,
781 struct page
***pages
, size_t maxsize
,
784 if (i
->type
& ITER_BVEC
)
785 return get_pages_alloc_bvec(i
, pages
, maxsize
, start
);
787 return get_pages_alloc_iovec(i
, pages
, maxsize
, start
);
789 EXPORT_SYMBOL(iov_iter_get_pages_alloc
);
791 int iov_iter_npages(const struct iov_iter
*i
, int maxpages
)
793 size_t size
= i
->count
;
799 iterate_all_kinds(i
, size
, v
, ({
800 unsigned long p
= (unsigned long)v
.iov_base
;
801 npages
+= DIV_ROUND_UP(p
+ v
.iov_len
, PAGE_SIZE
)
803 if (npages
>= maxpages
)
807 if (npages
>= maxpages
)
813 EXPORT_SYMBOL(iov_iter_npages
);