1 #include <linux/export.h>
3 #include <linux/pagemap.h>
4 #include <linux/slab.h>
5 #include <linux/vmalloc.h>
6 #include <linux/splice.h>
7 #include <net/checksum.h>
9 #define PIPE_PARANOIA /* for now */
11 #define iterate_iovec(i, n, __v, __p, skip, STEP) { \
15 __v.iov_len = min(n, __p->iov_len - skip); \
16 if (likely(__v.iov_len)) { \
17 __v.iov_base = __p->iov_base + skip; \
19 __v.iov_len -= left; \
20 skip += __v.iov_len; \
25 while (unlikely(!left && n)) { \
27 __v.iov_len = min(n, __p->iov_len); \
28 if (unlikely(!__v.iov_len)) \
30 __v.iov_base = __p->iov_base; \
32 __v.iov_len -= left; \
39 #define iterate_kvec(i, n, __v, __p, skip, STEP) { \
42 __v.iov_len = min(n, __p->iov_len - skip); \
43 if (likely(__v.iov_len)) { \
44 __v.iov_base = __p->iov_base + skip; \
46 skip += __v.iov_len; \
49 while (unlikely(n)) { \
51 __v.iov_len = min(n, __p->iov_len); \
52 if (unlikely(!__v.iov_len)) \
54 __v.iov_base = __p->iov_base; \
62 #define iterate_bvec(i, n, __v, __bi, skip, STEP) { \
63 struct bvec_iter __start; \
64 __start.bi_size = n; \
65 __start.bi_bvec_done = skip; \
67 for_each_bvec(__v, i->bvec, __bi, __start) { \
74 #define iterate_all_kinds(i, n, v, I, B, K) { \
75 size_t skip = i->iov_offset; \
76 if (unlikely(i->type & ITER_BVEC)) { \
78 struct bvec_iter __bi; \
79 iterate_bvec(i, n, v, __bi, skip, (B)) \
80 } else if (unlikely(i->type & ITER_KVEC)) { \
81 const struct kvec *kvec; \
83 iterate_kvec(i, n, v, kvec, skip, (K)) \
85 const struct iovec *iov; \
87 iterate_iovec(i, n, v, iov, skip, (I)) \
91 #define iterate_and_advance(i, n, v, I, B, K) { \
92 if (unlikely(i->count < n)) \
95 size_t skip = i->iov_offset; \
96 if (unlikely(i->type & ITER_BVEC)) { \
97 const struct bio_vec *bvec = i->bvec; \
99 struct bvec_iter __bi; \
100 iterate_bvec(i, n, v, __bi, skip, (B)) \
101 i->bvec = __bvec_iter_bvec(i->bvec, __bi); \
102 i->nr_segs -= i->bvec - bvec; \
103 skip = __bi.bi_bvec_done; \
104 } else if (unlikely(i->type & ITER_KVEC)) { \
105 const struct kvec *kvec; \
107 iterate_kvec(i, n, v, kvec, skip, (K)) \
108 if (skip == kvec->iov_len) { \
112 i->nr_segs -= kvec - i->kvec; \
115 const struct iovec *iov; \
117 iterate_iovec(i, n, v, iov, skip, (I)) \
118 if (skip == iov->iov_len) { \
122 i->nr_segs -= iov - i->iov; \
126 i->iov_offset = skip; \
130 static size_t copy_page_to_iter_iovec(struct page
*page
, size_t offset
, size_t bytes
,
133 size_t skip
, copy
, left
, wanted
;
134 const struct iovec
*iov
;
138 if (unlikely(bytes
> i
->count
))
141 if (unlikely(!bytes
))
146 skip
= i
->iov_offset
;
147 buf
= iov
->iov_base
+ skip
;
148 copy
= min(bytes
, iov
->iov_len
- skip
);
150 if (IS_ENABLED(CONFIG_HIGHMEM
) && !fault_in_pages_writeable(buf
, copy
)) {
151 kaddr
= kmap_atomic(page
);
152 from
= kaddr
+ offset
;
154 /* first chunk, usually the only one */
155 left
= __copy_to_user_inatomic(buf
, from
, copy
);
161 while (unlikely(!left
&& bytes
)) {
164 copy
= min(bytes
, iov
->iov_len
);
165 left
= __copy_to_user_inatomic(buf
, from
, copy
);
171 if (likely(!bytes
)) {
172 kunmap_atomic(kaddr
);
175 offset
= from
- kaddr
;
177 kunmap_atomic(kaddr
);
178 copy
= min(bytes
, iov
->iov_len
- skip
);
180 /* Too bad - revert to non-atomic kmap */
183 from
= kaddr
+ offset
;
184 left
= __copy_to_user(buf
, from
, copy
);
189 while (unlikely(!left
&& bytes
)) {
192 copy
= min(bytes
, iov
->iov_len
);
193 left
= __copy_to_user(buf
, from
, copy
);
202 if (skip
== iov
->iov_len
) {
206 i
->count
-= wanted
- bytes
;
207 i
->nr_segs
-= iov
- i
->iov
;
209 i
->iov_offset
= skip
;
210 return wanted
- bytes
;
213 static size_t copy_page_from_iter_iovec(struct page
*page
, size_t offset
, size_t bytes
,
216 size_t skip
, copy
, left
, wanted
;
217 const struct iovec
*iov
;
221 if (unlikely(bytes
> i
->count
))
224 if (unlikely(!bytes
))
229 skip
= i
->iov_offset
;
230 buf
= iov
->iov_base
+ skip
;
231 copy
= min(bytes
, iov
->iov_len
- skip
);
233 if (IS_ENABLED(CONFIG_HIGHMEM
) && !fault_in_pages_readable(buf
, copy
)) {
234 kaddr
= kmap_atomic(page
);
237 /* first chunk, usually the only one */
238 left
= __copy_from_user_inatomic(to
, buf
, copy
);
244 while (unlikely(!left
&& bytes
)) {
247 copy
= min(bytes
, iov
->iov_len
);
248 left
= __copy_from_user_inatomic(to
, buf
, copy
);
254 if (likely(!bytes
)) {
255 kunmap_atomic(kaddr
);
260 kunmap_atomic(kaddr
);
261 copy
= min(bytes
, iov
->iov_len
- skip
);
263 /* Too bad - revert to non-atomic kmap */
267 left
= __copy_from_user(to
, buf
, copy
);
272 while (unlikely(!left
&& bytes
)) {
275 copy
= min(bytes
, iov
->iov_len
);
276 left
= __copy_from_user(to
, buf
, copy
);
285 if (skip
== iov
->iov_len
) {
289 i
->count
-= wanted
- bytes
;
290 i
->nr_segs
-= iov
- i
->iov
;
292 i
->iov_offset
= skip
;
293 return wanted
- bytes
;
297 static bool sanity(const struct iov_iter
*i
)
299 struct pipe_inode_info
*pipe
= i
->pipe
;
301 int next
= pipe
->curbuf
+ pipe
->nrbufs
;
303 struct pipe_buffer
*p
;
304 if (unlikely(!pipe
->nrbufs
))
305 goto Bad
; // pipe must be non-empty
306 if (unlikely(idx
!= ((next
- 1) & (pipe
->buffers
- 1))))
307 goto Bad
; // must be at the last buffer...
309 p
= &pipe
->bufs
[idx
];
310 if (unlikely(p
->offset
+ p
->len
!= i
->iov_offset
))
311 goto Bad
; // ... at the end of segment
313 if (idx
!= (next
& (pipe
->buffers
- 1)))
314 goto Bad
; // must be right after the last buffer
318 printk(KERN_ERR
"idx = %d, offset = %zd\n", i
->idx
, i
->iov_offset
);
319 printk(KERN_ERR
"curbuf = %d, nrbufs = %d, buffers = %d\n",
320 pipe
->curbuf
, pipe
->nrbufs
, pipe
->buffers
);
321 for (idx
= 0; idx
< pipe
->buffers
; idx
++)
322 printk(KERN_ERR
"[%p %p %d %d]\n",
324 pipe
->bufs
[idx
].page
,
325 pipe
->bufs
[idx
].offset
,
326 pipe
->bufs
[idx
].len
);
331 #define sanity(i) true
334 static inline int next_idx(int idx
, struct pipe_inode_info
*pipe
)
336 return (idx
+ 1) & (pipe
->buffers
- 1);
339 static size_t copy_page_to_iter_pipe(struct page
*page
, size_t offset
, size_t bytes
,
342 struct pipe_inode_info
*pipe
= i
->pipe
;
343 struct pipe_buffer
*buf
;
347 if (unlikely(bytes
> i
->count
))
350 if (unlikely(!bytes
))
358 buf
= &pipe
->bufs
[idx
];
360 if (offset
== off
&& buf
->page
== page
) {
361 /* merge with the last one */
363 i
->iov_offset
+= bytes
;
366 idx
= next_idx(idx
, pipe
);
367 buf
= &pipe
->bufs
[idx
];
369 if (idx
== pipe
->curbuf
&& pipe
->nrbufs
)
372 buf
->ops
= &page_cache_pipe_buf_ops
;
373 get_page(buf
->page
= page
);
374 buf
->offset
= offset
;
376 i
->iov_offset
= offset
+ bytes
;
384 * Fault in one or more iovecs of the given iov_iter, to a maximum length of
385 * bytes. For each iovec, fault in each page that constitutes the iovec.
387 * Return 0 on success, or non-zero if the memory could not be accessed (i.e.
388 * because it is an invalid address).
390 int iov_iter_fault_in_readable(struct iov_iter
*i
, size_t bytes
)
392 size_t skip
= i
->iov_offset
;
393 const struct iovec
*iov
;
397 if (!(i
->type
& (ITER_BVEC
|ITER_KVEC
))) {
398 iterate_iovec(i
, bytes
, v
, iov
, skip
, ({
399 err
= fault_in_pages_readable(v
.iov_base
, v
.iov_len
);
406 EXPORT_SYMBOL(iov_iter_fault_in_readable
);
408 void iov_iter_init(struct iov_iter
*i
, int direction
,
409 const struct iovec
*iov
, unsigned long nr_segs
,
412 /* It will get better. Eventually... */
413 if (segment_eq(get_fs(), KERNEL_DS
)) {
414 direction
|= ITER_KVEC
;
416 i
->kvec
= (struct kvec
*)iov
;
421 i
->nr_segs
= nr_segs
;
425 EXPORT_SYMBOL(iov_iter_init
);
427 static void memcpy_from_page(char *to
, struct page
*page
, size_t offset
, size_t len
)
429 char *from
= kmap_atomic(page
);
430 memcpy(to
, from
+ offset
, len
);
434 static void memcpy_to_page(struct page
*page
, size_t offset
, const char *from
, size_t len
)
436 char *to
= kmap_atomic(page
);
437 memcpy(to
+ offset
, from
, len
);
441 static void memzero_page(struct page
*page
, size_t offset
, size_t len
)
443 char *addr
= kmap_atomic(page
);
444 memset(addr
+ offset
, 0, len
);
448 static inline bool allocated(struct pipe_buffer
*buf
)
450 return buf
->ops
== &default_pipe_buf_ops
;
453 static inline void data_start(const struct iov_iter
*i
, int *idxp
, size_t *offp
)
455 size_t off
= i
->iov_offset
;
457 if (off
&& (!allocated(&i
->pipe
->bufs
[idx
]) || off
== PAGE_SIZE
)) {
458 idx
= next_idx(idx
, i
->pipe
);
465 static size_t push_pipe(struct iov_iter
*i
, size_t size
,
466 int *idxp
, size_t *offp
)
468 struct pipe_inode_info
*pipe
= i
->pipe
;
473 if (unlikely(size
> i
->count
))
479 data_start(i
, &idx
, &off
);
483 left
-= PAGE_SIZE
- off
;
485 pipe
->bufs
[idx
].len
+= size
;
488 pipe
->bufs
[idx
].len
= PAGE_SIZE
;
489 idx
= next_idx(idx
, pipe
);
491 while (idx
!= pipe
->curbuf
|| !pipe
->nrbufs
) {
492 struct page
*page
= alloc_page(GFP_USER
);
496 pipe
->bufs
[idx
].ops
= &default_pipe_buf_ops
;
497 pipe
->bufs
[idx
].page
= page
;
498 pipe
->bufs
[idx
].offset
= 0;
499 if (left
<= PAGE_SIZE
) {
500 pipe
->bufs
[idx
].len
= left
;
503 pipe
->bufs
[idx
].len
= PAGE_SIZE
;
505 idx
= next_idx(idx
, pipe
);
510 static size_t copy_pipe_to_iter(const void *addr
, size_t bytes
,
513 struct pipe_inode_info
*pipe
= i
->pipe
;
520 bytes
= n
= push_pipe(i
, bytes
, &idx
, &off
);
523 for ( ; n
; idx
= next_idx(idx
, pipe
), off
= 0) {
524 size_t chunk
= min_t(size_t, n
, PAGE_SIZE
- off
);
525 memcpy_to_page(pipe
->bufs
[idx
].page
, off
, addr
, chunk
);
527 i
->iov_offset
= off
+ chunk
;
535 size_t copy_to_iter(const void *addr
, size_t bytes
, struct iov_iter
*i
)
537 const char *from
= addr
;
538 if (unlikely(i
->type
& ITER_PIPE
))
539 return copy_pipe_to_iter(addr
, bytes
, i
);
540 iterate_and_advance(i
, bytes
, v
,
541 __copy_to_user(v
.iov_base
, (from
+= v
.iov_len
) - v
.iov_len
,
543 memcpy_to_page(v
.bv_page
, v
.bv_offset
,
544 (from
+= v
.bv_len
) - v
.bv_len
, v
.bv_len
),
545 memcpy(v
.iov_base
, (from
+= v
.iov_len
) - v
.iov_len
, v
.iov_len
)
550 EXPORT_SYMBOL(copy_to_iter
);
552 size_t copy_from_iter(void *addr
, size_t bytes
, struct iov_iter
*i
)
555 if (unlikely(i
->type
& ITER_PIPE
)) {
559 iterate_and_advance(i
, bytes
, v
,
560 __copy_from_user((to
+= v
.iov_len
) - v
.iov_len
, v
.iov_base
,
562 memcpy_from_page((to
+= v
.bv_len
) - v
.bv_len
, v
.bv_page
,
563 v
.bv_offset
, v
.bv_len
),
564 memcpy((to
+= v
.iov_len
) - v
.iov_len
, v
.iov_base
, v
.iov_len
)
569 EXPORT_SYMBOL(copy_from_iter
);
571 size_t copy_from_iter_nocache(void *addr
, size_t bytes
, struct iov_iter
*i
)
574 if (unlikely(i
->type
& ITER_PIPE
)) {
578 iterate_and_advance(i
, bytes
, v
,
579 __copy_from_user_nocache((to
+= v
.iov_len
) - v
.iov_len
,
580 v
.iov_base
, v
.iov_len
),
581 memcpy_from_page((to
+= v
.bv_len
) - v
.bv_len
, v
.bv_page
,
582 v
.bv_offset
, v
.bv_len
),
583 memcpy((to
+= v
.iov_len
) - v
.iov_len
, v
.iov_base
, v
.iov_len
)
588 EXPORT_SYMBOL(copy_from_iter_nocache
);
590 size_t copy_page_to_iter(struct page
*page
, size_t offset
, size_t bytes
,
593 if (i
->type
& (ITER_BVEC
|ITER_KVEC
)) {
594 void *kaddr
= kmap_atomic(page
);
595 size_t wanted
= copy_to_iter(kaddr
+ offset
, bytes
, i
);
596 kunmap_atomic(kaddr
);
598 } else if (likely(!(i
->type
& ITER_PIPE
)))
599 return copy_page_to_iter_iovec(page
, offset
, bytes
, i
);
601 return copy_page_to_iter_pipe(page
, offset
, bytes
, i
);
603 EXPORT_SYMBOL(copy_page_to_iter
);
605 size_t copy_page_from_iter(struct page
*page
, size_t offset
, size_t bytes
,
608 if (unlikely(i
->type
& ITER_PIPE
)) {
612 if (i
->type
& (ITER_BVEC
|ITER_KVEC
)) {
613 void *kaddr
= kmap_atomic(page
);
614 size_t wanted
= copy_from_iter(kaddr
+ offset
, bytes
, i
);
615 kunmap_atomic(kaddr
);
618 return copy_page_from_iter_iovec(page
, offset
, bytes
, i
);
620 EXPORT_SYMBOL(copy_page_from_iter
);
622 static size_t pipe_zero(size_t bytes
, struct iov_iter
*i
)
624 struct pipe_inode_info
*pipe
= i
->pipe
;
631 bytes
= n
= push_pipe(i
, bytes
, &idx
, &off
);
635 for ( ; n
; idx
= next_idx(idx
, pipe
), off
= 0) {
636 size_t chunk
= min_t(size_t, n
, PAGE_SIZE
- off
);
637 memzero_page(pipe
->bufs
[idx
].page
, off
, chunk
);
639 i
->iov_offset
= off
+ chunk
;
646 size_t iov_iter_zero(size_t bytes
, struct iov_iter
*i
)
648 if (unlikely(i
->type
& ITER_PIPE
))
649 return pipe_zero(bytes
, i
);
650 iterate_and_advance(i
, bytes
, v
,
651 __clear_user(v
.iov_base
, v
.iov_len
),
652 memzero_page(v
.bv_page
, v
.bv_offset
, v
.bv_len
),
653 memset(v
.iov_base
, 0, v
.iov_len
)
658 EXPORT_SYMBOL(iov_iter_zero
);
660 size_t iov_iter_copy_from_user_atomic(struct page
*page
,
661 struct iov_iter
*i
, unsigned long offset
, size_t bytes
)
663 char *kaddr
= kmap_atomic(page
), *p
= kaddr
+ offset
;
664 if (unlikely(i
->type
& ITER_PIPE
)) {
665 kunmap_atomic(kaddr
);
669 iterate_all_kinds(i
, bytes
, v
,
670 __copy_from_user_inatomic((p
+= v
.iov_len
) - v
.iov_len
,
671 v
.iov_base
, v
.iov_len
),
672 memcpy_from_page((p
+= v
.bv_len
) - v
.bv_len
, v
.bv_page
,
673 v
.bv_offset
, v
.bv_len
),
674 memcpy((p
+= v
.iov_len
) - v
.iov_len
, v
.iov_base
, v
.iov_len
)
676 kunmap_atomic(kaddr
);
679 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic
);
681 static void pipe_advance(struct iov_iter
*i
, size_t size
)
683 struct pipe_inode_info
*pipe
= i
->pipe
;
684 struct pipe_buffer
*buf
;
686 size_t off
= i
->iov_offset
;
688 if (unlikely(i
->count
< size
))
692 if (off
) /* make it relative to the beginning of buffer */
693 size
+= off
- pipe
->bufs
[idx
].offset
;
695 buf
= &pipe
->bufs
[idx
];
696 if (size
<= buf
->len
)
699 idx
= next_idx(idx
, pipe
);
703 off
= i
->iov_offset
= buf
->offset
+ size
;
706 idx
= next_idx(idx
, pipe
);
708 int unused
= (pipe
->curbuf
+ pipe
->nrbufs
) & (pipe
->buffers
- 1);
709 /* [curbuf,unused) is in use. Free [idx,unused) */
710 while (idx
!= unused
) {
711 pipe_buf_release(pipe
, &pipe
->bufs
[idx
]);
712 idx
= next_idx(idx
, pipe
);
718 void iov_iter_advance(struct iov_iter
*i
, size_t size
)
720 if (unlikely(i
->type
& ITER_PIPE
)) {
721 pipe_advance(i
, size
);
724 iterate_and_advance(i
, size
, v
, 0, 0, 0)
726 EXPORT_SYMBOL(iov_iter_advance
);
729 * Return the count of just the current iov_iter segment.
731 size_t iov_iter_single_seg_count(const struct iov_iter
*i
)
733 if (unlikely(i
->type
& ITER_PIPE
))
734 return i
->count
; // it is a silly place, anyway
737 else if (i
->type
& ITER_BVEC
)
738 return min(i
->count
, i
->bvec
->bv_len
- i
->iov_offset
);
740 return min(i
->count
, i
->iov
->iov_len
- i
->iov_offset
);
742 EXPORT_SYMBOL(iov_iter_single_seg_count
);
744 void iov_iter_kvec(struct iov_iter
*i
, int direction
,
745 const struct kvec
*kvec
, unsigned long nr_segs
,
748 BUG_ON(!(direction
& ITER_KVEC
));
751 i
->nr_segs
= nr_segs
;
755 EXPORT_SYMBOL(iov_iter_kvec
);
757 void iov_iter_bvec(struct iov_iter
*i
, int direction
,
758 const struct bio_vec
*bvec
, unsigned long nr_segs
,
761 BUG_ON(!(direction
& ITER_BVEC
));
764 i
->nr_segs
= nr_segs
;
768 EXPORT_SYMBOL(iov_iter_bvec
);
770 void iov_iter_pipe(struct iov_iter
*i
, int direction
,
771 struct pipe_inode_info
*pipe
,
774 BUG_ON(direction
!= ITER_PIPE
);
777 i
->idx
= (pipe
->curbuf
+ pipe
->nrbufs
) & (pipe
->buffers
- 1);
781 EXPORT_SYMBOL(iov_iter_pipe
);
783 unsigned long iov_iter_alignment(const struct iov_iter
*i
)
785 unsigned long res
= 0;
786 size_t size
= i
->count
;
791 if (unlikely(i
->type
& ITER_PIPE
)) {
792 if (i
->iov_offset
&& allocated(&i
->pipe
->bufs
[i
->idx
]))
793 return size
| i
->iov_offset
;
796 iterate_all_kinds(i
, size
, v
,
797 (res
|= (unsigned long)v
.iov_base
| v
.iov_len
, 0),
798 res
|= v
.bv_offset
| v
.bv_len
,
799 res
|= (unsigned long)v
.iov_base
| v
.iov_len
803 EXPORT_SYMBOL(iov_iter_alignment
);
805 unsigned long iov_iter_gap_alignment(const struct iov_iter
*i
)
807 unsigned long res
= 0;
808 size_t size
= i
->count
;
812 if (unlikely(i
->type
& ITER_PIPE
)) {
817 iterate_all_kinds(i
, size
, v
,
818 (res
|= (!res
? 0 : (unsigned long)v
.iov_base
) |
819 (size
!= v
.iov_len
? size
: 0), 0),
820 (res
|= (!res
? 0 : (unsigned long)v
.bv_offset
) |
821 (size
!= v
.bv_len
? size
: 0)),
822 (res
|= (!res
? 0 : (unsigned long)v
.iov_base
) |
823 (size
!= v
.iov_len
? size
: 0))
827 EXPORT_SYMBOL(iov_iter_gap_alignment
);
829 static inline size_t __pipe_get_pages(struct iov_iter
*i
,
835 struct pipe_inode_info
*pipe
= i
->pipe
;
836 ssize_t n
= push_pipe(i
, maxsize
, &idx
, start
);
843 get_page(*pages
++ = pipe
->bufs
[idx
].page
);
844 idx
= next_idx(idx
, pipe
);
851 static ssize_t
pipe_get_pages(struct iov_iter
*i
,
852 struct page
**pages
, size_t maxsize
, unsigned maxpages
,
862 data_start(i
, &idx
, start
);
863 /* some of this one + all after this one */
864 npages
= ((i
->pipe
->curbuf
- idx
- 1) & (i
->pipe
->buffers
- 1)) + 1;
865 capacity
= min(npages
,maxpages
) * PAGE_SIZE
- *start
;
867 return __pipe_get_pages(i
, min(maxsize
, capacity
), pages
, idx
, start
);
870 ssize_t
iov_iter_get_pages(struct iov_iter
*i
,
871 struct page
**pages
, size_t maxsize
, unsigned maxpages
,
874 if (maxsize
> i
->count
)
880 if (unlikely(i
->type
& ITER_PIPE
))
881 return pipe_get_pages(i
, pages
, maxsize
, maxpages
, start
);
882 iterate_all_kinds(i
, maxsize
, v
, ({
883 unsigned long addr
= (unsigned long)v
.iov_base
;
884 size_t len
= v
.iov_len
+ (*start
= addr
& (PAGE_SIZE
- 1));
888 if (len
> maxpages
* PAGE_SIZE
)
889 len
= maxpages
* PAGE_SIZE
;
890 addr
&= ~(PAGE_SIZE
- 1);
891 n
= DIV_ROUND_UP(len
, PAGE_SIZE
);
892 res
= get_user_pages_fast(addr
, n
, (i
->type
& WRITE
) != WRITE
, pages
);
893 if (unlikely(res
< 0))
895 return (res
== n
? len
: res
* PAGE_SIZE
) - *start
;
897 /* can't be more than PAGE_SIZE */
898 *start
= v
.bv_offset
;
899 get_page(*pages
= v
.bv_page
);
907 EXPORT_SYMBOL(iov_iter_get_pages
);
909 static struct page
**get_pages_array(size_t n
)
911 struct page
**p
= kmalloc(n
* sizeof(struct page
*), GFP_KERNEL
);
913 p
= vmalloc(n
* sizeof(struct page
*));
917 static ssize_t
pipe_get_pages_alloc(struct iov_iter
*i
,
918 struct page
***pages
, size_t maxsize
,
929 data_start(i
, &idx
, start
);
930 /* some of this one + all after this one */
931 npages
= ((i
->pipe
->curbuf
- idx
- 1) & (i
->pipe
->buffers
- 1)) + 1;
932 n
= npages
* PAGE_SIZE
- *start
;
936 npages
= DIV_ROUND_UP(maxsize
+ *start
, PAGE_SIZE
);
937 p
= get_pages_array(npages
);
940 n
= __pipe_get_pages(i
, maxsize
, p
, idx
, start
);
948 ssize_t
iov_iter_get_pages_alloc(struct iov_iter
*i
,
949 struct page
***pages
, size_t maxsize
,
954 if (maxsize
> i
->count
)
960 if (unlikely(i
->type
& ITER_PIPE
))
961 return pipe_get_pages_alloc(i
, pages
, maxsize
, start
);
962 iterate_all_kinds(i
, maxsize
, v
, ({
963 unsigned long addr
= (unsigned long)v
.iov_base
;
964 size_t len
= v
.iov_len
+ (*start
= addr
& (PAGE_SIZE
- 1));
968 addr
&= ~(PAGE_SIZE
- 1);
969 n
= DIV_ROUND_UP(len
, PAGE_SIZE
);
970 p
= get_pages_array(n
);
973 res
= get_user_pages_fast(addr
, n
, (i
->type
& WRITE
) != WRITE
, p
);
974 if (unlikely(res
< 0)) {
979 return (res
== n
? len
: res
* PAGE_SIZE
) - *start
;
981 /* can't be more than PAGE_SIZE */
982 *start
= v
.bv_offset
;
983 *pages
= p
= get_pages_array(1);
986 get_page(*p
= v
.bv_page
);
994 EXPORT_SYMBOL(iov_iter_get_pages_alloc
);
996 size_t csum_and_copy_from_iter(void *addr
, size_t bytes
, __wsum
*csum
,
1003 if (unlikely(i
->type
& ITER_PIPE
)) {
1007 iterate_and_advance(i
, bytes
, v
, ({
1009 next
= csum_and_copy_from_user(v
.iov_base
,
1010 (to
+= v
.iov_len
) - v
.iov_len
,
1011 v
.iov_len
, 0, &err
);
1013 sum
= csum_block_add(sum
, next
, off
);
1016 err
? v
.iov_len
: 0;
1018 char *p
= kmap_atomic(v
.bv_page
);
1019 next
= csum_partial_copy_nocheck(p
+ v
.bv_offset
,
1020 (to
+= v
.bv_len
) - v
.bv_len
,
1023 sum
= csum_block_add(sum
, next
, off
);
1026 next
= csum_partial_copy_nocheck(v
.iov_base
,
1027 (to
+= v
.iov_len
) - v
.iov_len
,
1029 sum
= csum_block_add(sum
, next
, off
);
1036 EXPORT_SYMBOL(csum_and_copy_from_iter
);
1038 size_t csum_and_copy_to_iter(const void *addr
, size_t bytes
, __wsum
*csum
,
1041 const char *from
= addr
;
1045 if (unlikely(i
->type
& ITER_PIPE
)) {
1046 WARN_ON(1); /* for now */
1049 iterate_and_advance(i
, bytes
, v
, ({
1051 next
= csum_and_copy_to_user((from
+= v
.iov_len
) - v
.iov_len
,
1053 v
.iov_len
, 0, &err
);
1055 sum
= csum_block_add(sum
, next
, off
);
1058 err
? v
.iov_len
: 0;
1060 char *p
= kmap_atomic(v
.bv_page
);
1061 next
= csum_partial_copy_nocheck((from
+= v
.bv_len
) - v
.bv_len
,
1065 sum
= csum_block_add(sum
, next
, off
);
1068 next
= csum_partial_copy_nocheck((from
+= v
.iov_len
) - v
.iov_len
,
1071 sum
= csum_block_add(sum
, next
, off
);
1078 EXPORT_SYMBOL(csum_and_copy_to_iter
);
1080 int iov_iter_npages(const struct iov_iter
*i
, int maxpages
)
1082 size_t size
= i
->count
;
1088 if (unlikely(i
->type
& ITER_PIPE
)) {
1089 struct pipe_inode_info
*pipe
= i
->pipe
;
1096 data_start(i
, &idx
, &off
);
1097 /* some of this one + all after this one */
1098 npages
= ((pipe
->curbuf
- idx
- 1) & (pipe
->buffers
- 1)) + 1;
1099 if (npages
>= maxpages
)
1101 } else iterate_all_kinds(i
, size
, v
, ({
1102 unsigned long p
= (unsigned long)v
.iov_base
;
1103 npages
+= DIV_ROUND_UP(p
+ v
.iov_len
, PAGE_SIZE
)
1105 if (npages
>= maxpages
)
1109 if (npages
>= maxpages
)
1112 unsigned long p
= (unsigned long)v
.iov_base
;
1113 npages
+= DIV_ROUND_UP(p
+ v
.iov_len
, PAGE_SIZE
)
1115 if (npages
>= maxpages
)
1121 EXPORT_SYMBOL(iov_iter_npages
);
1123 const void *dup_iter(struct iov_iter
*new, struct iov_iter
*old
, gfp_t flags
)
1126 if (unlikely(new->type
& ITER_PIPE
)) {
1130 if (new->type
& ITER_BVEC
)
1131 return new->bvec
= kmemdup(new->bvec
,
1132 new->nr_segs
* sizeof(struct bio_vec
),
1135 /* iovec and kvec have identical layout */
1136 return new->iov
= kmemdup(new->iov
,
1137 new->nr_segs
* sizeof(struct iovec
),
1140 EXPORT_SYMBOL(dup_iter
);
1143 * import_iovec() - Copy an array of &struct iovec from userspace
1144 * into the kernel, check that it is valid, and initialize a new
1145 * &struct iov_iter iterator to access it.
1147 * @type: One of %READ or %WRITE.
1148 * @uvector: Pointer to the userspace array.
1149 * @nr_segs: Number of elements in userspace array.
1150 * @fast_segs: Number of elements in @iov.
1151 * @iov: (input and output parameter) Pointer to pointer to (usually small
1152 * on-stack) kernel array.
1153 * @i: Pointer to iterator that will be initialized on success.
1155 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1156 * then this function places %NULL in *@iov on return. Otherwise, a new
1157 * array will be allocated and the result placed in *@iov. This means that
1158 * the caller may call kfree() on *@iov regardless of whether the small
1159 * on-stack array was used or not (and regardless of whether this function
1160 * returns an error or not).
1162 * Return: 0 on success or negative error code on error.
1164 int import_iovec(int type
, const struct iovec __user
* uvector
,
1165 unsigned nr_segs
, unsigned fast_segs
,
1166 struct iovec
**iov
, struct iov_iter
*i
)
1170 n
= rw_copy_check_uvector(type
, uvector
, nr_segs
, fast_segs
,
1178 iov_iter_init(i
, type
, p
, nr_segs
, n
);
1179 *iov
= p
== *iov
? NULL
: p
;
1182 EXPORT_SYMBOL(import_iovec
);
1184 #ifdef CONFIG_COMPAT
1185 #include <linux/compat.h>
1187 int compat_import_iovec(int type
, const struct compat_iovec __user
* uvector
,
1188 unsigned nr_segs
, unsigned fast_segs
,
1189 struct iovec
**iov
, struct iov_iter
*i
)
1193 n
= compat_rw_copy_check_uvector(type
, uvector
, nr_segs
, fast_segs
,
1201 iov_iter_init(i
, type
, p
, nr_segs
, n
);
1202 *iov
= p
== *iov
? NULL
: p
;
1207 int import_single_range(int rw
, void __user
*buf
, size_t len
,
1208 struct iovec
*iov
, struct iov_iter
*i
)
1210 if (len
> MAX_RW_COUNT
)
1212 if (unlikely(!access_ok(!rw
, buf
, len
)))
1215 iov
->iov_base
= buf
;
1217 iov_iter_init(i
, rw
, iov
, 1, len
);
1220 EXPORT_SYMBOL(import_single_range
);