1 #include <linux/export.h>
2 #include <linux/bvec.h>
4 #include <linux/pagemap.h>
5 #include <linux/slab.h>
6 #include <linux/vmalloc.h>
7 #include <linux/splice.h>
8 #include <net/checksum.h>
10 #define PIPE_PARANOIA /* for now */
12 #define iterate_iovec(i, n, __v, __p, skip, STEP) { \
16 __v.iov_len = min(n, __p->iov_len - skip); \
17 if (likely(__v.iov_len)) { \
18 __v.iov_base = __p->iov_base + skip; \
20 __v.iov_len -= left; \
21 skip += __v.iov_len; \
26 while (unlikely(!left && n)) { \
28 __v.iov_len = min(n, __p->iov_len); \
29 if (unlikely(!__v.iov_len)) \
31 __v.iov_base = __p->iov_base; \
33 __v.iov_len -= left; \
40 #define iterate_kvec(i, n, __v, __p, skip, STEP) { \
43 __v.iov_len = min(n, __p->iov_len - skip); \
44 if (likely(__v.iov_len)) { \
45 __v.iov_base = __p->iov_base + skip; \
47 skip += __v.iov_len; \
50 while (unlikely(n)) { \
52 __v.iov_len = min(n, __p->iov_len); \
53 if (unlikely(!__v.iov_len)) \
55 __v.iov_base = __p->iov_base; \
63 #define iterate_bvec(i, n, __v, __bi, skip, STEP) { \
64 struct bvec_iter __start; \
65 __start.bi_size = n; \
66 __start.bi_bvec_done = skip; \
68 for_each_bvec(__v, i->bvec, __bi, __start) { \
75 #define iterate_all_kinds(i, n, v, I, B, K) { \
77 size_t skip = i->iov_offset; \
78 if (unlikely(i->type & ITER_BVEC)) { \
80 struct bvec_iter __bi; \
81 iterate_bvec(i, n, v, __bi, skip, (B)) \
82 } else if (unlikely(i->type & ITER_KVEC)) { \
83 const struct kvec *kvec; \
85 iterate_kvec(i, n, v, kvec, skip, (K)) \
87 const struct iovec *iov; \
89 iterate_iovec(i, n, v, iov, skip, (I)) \
94 #define iterate_and_advance(i, n, v, I, B, K) { \
95 if (unlikely(i->count < n)) \
98 size_t skip = i->iov_offset; \
99 if (unlikely(i->type & ITER_BVEC)) { \
100 const struct bio_vec *bvec = i->bvec; \
102 struct bvec_iter __bi; \
103 iterate_bvec(i, n, v, __bi, skip, (B)) \
104 i->bvec = __bvec_iter_bvec(i->bvec, __bi); \
105 i->nr_segs -= i->bvec - bvec; \
106 skip = __bi.bi_bvec_done; \
107 } else if (unlikely(i->type & ITER_KVEC)) { \
108 const struct kvec *kvec; \
110 iterate_kvec(i, n, v, kvec, skip, (K)) \
111 if (skip == kvec->iov_len) { \
115 i->nr_segs -= kvec - i->kvec; \
118 const struct iovec *iov; \
120 iterate_iovec(i, n, v, iov, skip, (I)) \
121 if (skip == iov->iov_len) { \
125 i->nr_segs -= iov - i->iov; \
129 i->iov_offset = skip; \
133 static int copyout(void __user
*to
, const void *from
, size_t n
)
135 if (access_ok(VERIFY_WRITE
, to
, n
)) {
136 kasan_check_read(from
, n
);
137 n
= raw_copy_to_user(to
, from
, n
);
142 static int copyin(void *to
, const void __user
*from
, size_t n
)
144 if (access_ok(VERIFY_READ
, from
, n
)) {
145 kasan_check_write(to
, n
);
146 n
= raw_copy_from_user(to
, from
, n
);
151 static size_t copy_page_to_iter_iovec(struct page
*page
, size_t offset
, size_t bytes
,
154 size_t skip
, copy
, left
, wanted
;
155 const struct iovec
*iov
;
159 if (unlikely(bytes
> i
->count
))
162 if (unlikely(!bytes
))
168 skip
= i
->iov_offset
;
169 buf
= iov
->iov_base
+ skip
;
170 copy
= min(bytes
, iov
->iov_len
- skip
);
172 if (IS_ENABLED(CONFIG_HIGHMEM
) && !fault_in_pages_writeable(buf
, copy
)) {
173 kaddr
= kmap_atomic(page
);
174 from
= kaddr
+ offset
;
176 /* first chunk, usually the only one */
177 left
= copyout(buf
, from
, copy
);
183 while (unlikely(!left
&& bytes
)) {
186 copy
= min(bytes
, iov
->iov_len
);
187 left
= copyout(buf
, from
, copy
);
193 if (likely(!bytes
)) {
194 kunmap_atomic(kaddr
);
197 offset
= from
- kaddr
;
199 kunmap_atomic(kaddr
);
200 copy
= min(bytes
, iov
->iov_len
- skip
);
202 /* Too bad - revert to non-atomic kmap */
205 from
= kaddr
+ offset
;
206 left
= copyout(buf
, from
, copy
);
211 while (unlikely(!left
&& bytes
)) {
214 copy
= min(bytes
, iov
->iov_len
);
215 left
= copyout(buf
, from
, copy
);
224 if (skip
== iov
->iov_len
) {
228 i
->count
-= wanted
- bytes
;
229 i
->nr_segs
-= iov
- i
->iov
;
231 i
->iov_offset
= skip
;
232 return wanted
- bytes
;
235 static size_t copy_page_from_iter_iovec(struct page
*page
, size_t offset
, size_t bytes
,
238 size_t skip
, copy
, left
, wanted
;
239 const struct iovec
*iov
;
243 if (unlikely(bytes
> i
->count
))
246 if (unlikely(!bytes
))
252 skip
= i
->iov_offset
;
253 buf
= iov
->iov_base
+ skip
;
254 copy
= min(bytes
, iov
->iov_len
- skip
);
256 if (IS_ENABLED(CONFIG_HIGHMEM
) && !fault_in_pages_readable(buf
, copy
)) {
257 kaddr
= kmap_atomic(page
);
260 /* first chunk, usually the only one */
261 left
= copyin(to
, buf
, copy
);
267 while (unlikely(!left
&& bytes
)) {
270 copy
= min(bytes
, iov
->iov_len
);
271 left
= copyin(to
, buf
, copy
);
277 if (likely(!bytes
)) {
278 kunmap_atomic(kaddr
);
283 kunmap_atomic(kaddr
);
284 copy
= min(bytes
, iov
->iov_len
- skip
);
286 /* Too bad - revert to non-atomic kmap */
290 left
= copyin(to
, buf
, copy
);
295 while (unlikely(!left
&& bytes
)) {
298 copy
= min(bytes
, iov
->iov_len
);
299 left
= copyin(to
, buf
, copy
);
308 if (skip
== iov
->iov_len
) {
312 i
->count
-= wanted
- bytes
;
313 i
->nr_segs
-= iov
- i
->iov
;
315 i
->iov_offset
= skip
;
316 return wanted
- bytes
;
320 static bool sanity(const struct iov_iter
*i
)
322 struct pipe_inode_info
*pipe
= i
->pipe
;
324 int next
= pipe
->curbuf
+ pipe
->nrbufs
;
326 struct pipe_buffer
*p
;
327 if (unlikely(!pipe
->nrbufs
))
328 goto Bad
; // pipe must be non-empty
329 if (unlikely(idx
!= ((next
- 1) & (pipe
->buffers
- 1))))
330 goto Bad
; // must be at the last buffer...
332 p
= &pipe
->bufs
[idx
];
333 if (unlikely(p
->offset
+ p
->len
!= i
->iov_offset
))
334 goto Bad
; // ... at the end of segment
336 if (idx
!= (next
& (pipe
->buffers
- 1)))
337 goto Bad
; // must be right after the last buffer
341 printk(KERN_ERR
"idx = %d, offset = %zd\n", i
->idx
, i
->iov_offset
);
342 printk(KERN_ERR
"curbuf = %d, nrbufs = %d, buffers = %d\n",
343 pipe
->curbuf
, pipe
->nrbufs
, pipe
->buffers
);
344 for (idx
= 0; idx
< pipe
->buffers
; idx
++)
345 printk(KERN_ERR
"[%p %p %d %d]\n",
347 pipe
->bufs
[idx
].page
,
348 pipe
->bufs
[idx
].offset
,
349 pipe
->bufs
[idx
].len
);
354 #define sanity(i) true
357 static inline int next_idx(int idx
, struct pipe_inode_info
*pipe
)
359 return (idx
+ 1) & (pipe
->buffers
- 1);
362 static size_t copy_page_to_iter_pipe(struct page
*page
, size_t offset
, size_t bytes
,
365 struct pipe_inode_info
*pipe
= i
->pipe
;
366 struct pipe_buffer
*buf
;
370 if (unlikely(bytes
> i
->count
))
373 if (unlikely(!bytes
))
381 buf
= &pipe
->bufs
[idx
];
383 if (offset
== off
&& buf
->page
== page
) {
384 /* merge with the last one */
386 i
->iov_offset
+= bytes
;
389 idx
= next_idx(idx
, pipe
);
390 buf
= &pipe
->bufs
[idx
];
392 if (idx
== pipe
->curbuf
&& pipe
->nrbufs
)
395 buf
->ops
= &page_cache_pipe_buf_ops
;
396 get_page(buf
->page
= page
);
397 buf
->offset
= offset
;
399 i
->iov_offset
= offset
+ bytes
;
407 * Fault in one or more iovecs of the given iov_iter, to a maximum length of
408 * bytes. For each iovec, fault in each page that constitutes the iovec.
410 * Return 0 on success, or non-zero if the memory could not be accessed (i.e.
411 * because it is an invalid address).
413 int iov_iter_fault_in_readable(struct iov_iter
*i
, size_t bytes
)
415 size_t skip
= i
->iov_offset
;
416 const struct iovec
*iov
;
420 if (!(i
->type
& (ITER_BVEC
|ITER_KVEC
))) {
421 iterate_iovec(i
, bytes
, v
, iov
, skip
, ({
422 err
= fault_in_pages_readable(v
.iov_base
, v
.iov_len
);
429 EXPORT_SYMBOL(iov_iter_fault_in_readable
);
431 void iov_iter_init(struct iov_iter
*i
, int direction
,
432 const struct iovec
*iov
, unsigned long nr_segs
,
435 /* It will get better. Eventually... */
436 if (uaccess_kernel()) {
437 direction
|= ITER_KVEC
;
439 i
->kvec
= (struct kvec
*)iov
;
444 i
->nr_segs
= nr_segs
;
448 EXPORT_SYMBOL(iov_iter_init
);
450 static void memcpy_from_page(char *to
, struct page
*page
, size_t offset
, size_t len
)
452 char *from
= kmap_atomic(page
);
453 memcpy(to
, from
+ offset
, len
);
457 static void memcpy_to_page(struct page
*page
, size_t offset
, const char *from
, size_t len
)
459 char *to
= kmap_atomic(page
);
460 memcpy(to
+ offset
, from
, len
);
464 static void memzero_page(struct page
*page
, size_t offset
, size_t len
)
466 char *addr
= kmap_atomic(page
);
467 memset(addr
+ offset
, 0, len
);
471 static inline bool allocated(struct pipe_buffer
*buf
)
473 return buf
->ops
== &default_pipe_buf_ops
;
476 static inline void data_start(const struct iov_iter
*i
, int *idxp
, size_t *offp
)
478 size_t off
= i
->iov_offset
;
480 if (off
&& (!allocated(&i
->pipe
->bufs
[idx
]) || off
== PAGE_SIZE
)) {
481 idx
= next_idx(idx
, i
->pipe
);
488 static size_t push_pipe(struct iov_iter
*i
, size_t size
,
489 int *idxp
, size_t *offp
)
491 struct pipe_inode_info
*pipe
= i
->pipe
;
496 if (unlikely(size
> i
->count
))
502 data_start(i
, &idx
, &off
);
506 left
-= PAGE_SIZE
- off
;
508 pipe
->bufs
[idx
].len
+= size
;
511 pipe
->bufs
[idx
].len
= PAGE_SIZE
;
512 idx
= next_idx(idx
, pipe
);
514 while (idx
!= pipe
->curbuf
|| !pipe
->nrbufs
) {
515 struct page
*page
= alloc_page(GFP_USER
);
519 pipe
->bufs
[idx
].ops
= &default_pipe_buf_ops
;
520 pipe
->bufs
[idx
].page
= page
;
521 pipe
->bufs
[idx
].offset
= 0;
522 if (left
<= PAGE_SIZE
) {
523 pipe
->bufs
[idx
].len
= left
;
526 pipe
->bufs
[idx
].len
= PAGE_SIZE
;
528 idx
= next_idx(idx
, pipe
);
533 static size_t copy_pipe_to_iter(const void *addr
, size_t bytes
,
536 struct pipe_inode_info
*pipe
= i
->pipe
;
543 bytes
= n
= push_pipe(i
, bytes
, &idx
, &off
);
546 for ( ; n
; idx
= next_idx(idx
, pipe
), off
= 0) {
547 size_t chunk
= min_t(size_t, n
, PAGE_SIZE
- off
);
548 memcpy_to_page(pipe
->bufs
[idx
].page
, off
, addr
, chunk
);
550 i
->iov_offset
= off
+ chunk
;
558 size_t _copy_to_iter(const void *addr
, size_t bytes
, struct iov_iter
*i
)
560 const char *from
= addr
;
561 if (unlikely(i
->type
& ITER_PIPE
))
562 return copy_pipe_to_iter(addr
, bytes
, i
);
563 if (iter_is_iovec(i
))
565 iterate_and_advance(i
, bytes
, v
,
566 copyout(v
.iov_base
, (from
+= v
.iov_len
) - v
.iov_len
, v
.iov_len
),
567 memcpy_to_page(v
.bv_page
, v
.bv_offset
,
568 (from
+= v
.bv_len
) - v
.bv_len
, v
.bv_len
),
569 memcpy(v
.iov_base
, (from
+= v
.iov_len
) - v
.iov_len
, v
.iov_len
)
574 EXPORT_SYMBOL(_copy_to_iter
);
576 size_t _copy_from_iter(void *addr
, size_t bytes
, struct iov_iter
*i
)
579 if (unlikely(i
->type
& ITER_PIPE
)) {
583 if (iter_is_iovec(i
))
585 iterate_and_advance(i
, bytes
, v
,
586 copyin((to
+= v
.iov_len
) - v
.iov_len
, v
.iov_base
, v
.iov_len
),
587 memcpy_from_page((to
+= v
.bv_len
) - v
.bv_len
, v
.bv_page
,
588 v
.bv_offset
, v
.bv_len
),
589 memcpy((to
+= v
.iov_len
) - v
.iov_len
, v
.iov_base
, v
.iov_len
)
594 EXPORT_SYMBOL(_copy_from_iter
);
596 bool _copy_from_iter_full(void *addr
, size_t bytes
, struct iov_iter
*i
)
599 if (unlikely(i
->type
& ITER_PIPE
)) {
603 if (unlikely(i
->count
< bytes
))
606 if (iter_is_iovec(i
))
608 iterate_all_kinds(i
, bytes
, v
, ({
609 if (copyin((to
+= v
.iov_len
) - v
.iov_len
,
610 v
.iov_base
, v
.iov_len
))
613 memcpy_from_page((to
+= v
.bv_len
) - v
.bv_len
, v
.bv_page
,
614 v
.bv_offset
, v
.bv_len
),
615 memcpy((to
+= v
.iov_len
) - v
.iov_len
, v
.iov_base
, v
.iov_len
)
618 iov_iter_advance(i
, bytes
);
621 EXPORT_SYMBOL(_copy_from_iter_full
);
623 size_t _copy_from_iter_nocache(void *addr
, size_t bytes
, struct iov_iter
*i
)
626 if (unlikely(i
->type
& ITER_PIPE
)) {
630 iterate_and_advance(i
, bytes
, v
,
631 __copy_from_user_inatomic_nocache((to
+= v
.iov_len
) - v
.iov_len
,
632 v
.iov_base
, v
.iov_len
),
633 memcpy_from_page((to
+= v
.bv_len
) - v
.bv_len
, v
.bv_page
,
634 v
.bv_offset
, v
.bv_len
),
635 memcpy((to
+= v
.iov_len
) - v
.iov_len
, v
.iov_base
, v
.iov_len
)
640 EXPORT_SYMBOL(_copy_from_iter_nocache
);
642 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
643 size_t _copy_from_iter_flushcache(void *addr
, size_t bytes
, struct iov_iter
*i
)
646 if (unlikely(i
->type
& ITER_PIPE
)) {
650 iterate_and_advance(i
, bytes
, v
,
651 __copy_from_user_flushcache((to
+= v
.iov_len
) - v
.iov_len
,
652 v
.iov_base
, v
.iov_len
),
653 memcpy_page_flushcache((to
+= v
.bv_len
) - v
.bv_len
, v
.bv_page
,
654 v
.bv_offset
, v
.bv_len
),
655 memcpy_flushcache((to
+= v
.iov_len
) - v
.iov_len
, v
.iov_base
,
661 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache
);
664 bool _copy_from_iter_full_nocache(void *addr
, size_t bytes
, struct iov_iter
*i
)
667 if (unlikely(i
->type
& ITER_PIPE
)) {
671 if (unlikely(i
->count
< bytes
))
673 iterate_all_kinds(i
, bytes
, v
, ({
674 if (__copy_from_user_inatomic_nocache((to
+= v
.iov_len
) - v
.iov_len
,
675 v
.iov_base
, v
.iov_len
))
678 memcpy_from_page((to
+= v
.bv_len
) - v
.bv_len
, v
.bv_page
,
679 v
.bv_offset
, v
.bv_len
),
680 memcpy((to
+= v
.iov_len
) - v
.iov_len
, v
.iov_base
, v
.iov_len
)
683 iov_iter_advance(i
, bytes
);
686 EXPORT_SYMBOL(_copy_from_iter_full_nocache
);
688 static inline bool page_copy_sane(struct page
*page
, size_t offset
, size_t n
)
691 size_t v
= n
+ offset
;
694 * The general case needs to access the page order in order
695 * to compute the page size.
696 * However, we mostly deal with order-0 pages and thus can
697 * avoid a possible cache line miss for requests that fit all
700 if (n
<= v
&& v
<= PAGE_SIZE
)
703 head
= compound_head(page
);
704 v
+= (page
- head
) << PAGE_SHIFT
;
706 if (likely(n
<= v
&& v
<= (PAGE_SIZE
<< compound_order(head
))))
712 size_t copy_page_to_iter(struct page
*page
, size_t offset
, size_t bytes
,
715 if (unlikely(!page_copy_sane(page
, offset
, bytes
)))
717 if (i
->type
& (ITER_BVEC
|ITER_KVEC
)) {
718 void *kaddr
= kmap_atomic(page
);
719 size_t wanted
= copy_to_iter(kaddr
+ offset
, bytes
, i
);
720 kunmap_atomic(kaddr
);
722 } else if (likely(!(i
->type
& ITER_PIPE
)))
723 return copy_page_to_iter_iovec(page
, offset
, bytes
, i
);
725 return copy_page_to_iter_pipe(page
, offset
, bytes
, i
);
727 EXPORT_SYMBOL(copy_page_to_iter
);
729 size_t copy_page_from_iter(struct page
*page
, size_t offset
, size_t bytes
,
732 if (unlikely(!page_copy_sane(page
, offset
, bytes
)))
734 if (unlikely(i
->type
& ITER_PIPE
)) {
738 if (i
->type
& (ITER_BVEC
|ITER_KVEC
)) {
739 void *kaddr
= kmap_atomic(page
);
740 size_t wanted
= _copy_from_iter(kaddr
+ offset
, bytes
, i
);
741 kunmap_atomic(kaddr
);
744 return copy_page_from_iter_iovec(page
, offset
, bytes
, i
);
746 EXPORT_SYMBOL(copy_page_from_iter
);
748 static size_t pipe_zero(size_t bytes
, struct iov_iter
*i
)
750 struct pipe_inode_info
*pipe
= i
->pipe
;
757 bytes
= n
= push_pipe(i
, bytes
, &idx
, &off
);
761 for ( ; n
; idx
= next_idx(idx
, pipe
), off
= 0) {
762 size_t chunk
= min_t(size_t, n
, PAGE_SIZE
- off
);
763 memzero_page(pipe
->bufs
[idx
].page
, off
, chunk
);
765 i
->iov_offset
= off
+ chunk
;
772 size_t iov_iter_zero(size_t bytes
, struct iov_iter
*i
)
774 if (unlikely(i
->type
& ITER_PIPE
))
775 return pipe_zero(bytes
, i
);
776 iterate_and_advance(i
, bytes
, v
,
777 clear_user(v
.iov_base
, v
.iov_len
),
778 memzero_page(v
.bv_page
, v
.bv_offset
, v
.bv_len
),
779 memset(v
.iov_base
, 0, v
.iov_len
)
784 EXPORT_SYMBOL(iov_iter_zero
);
786 size_t iov_iter_copy_from_user_atomic(struct page
*page
,
787 struct iov_iter
*i
, unsigned long offset
, size_t bytes
)
789 char *kaddr
= kmap_atomic(page
), *p
= kaddr
+ offset
;
790 if (unlikely(!page_copy_sane(page
, offset
, bytes
))) {
791 kunmap_atomic(kaddr
);
794 if (unlikely(i
->type
& ITER_PIPE
)) {
795 kunmap_atomic(kaddr
);
799 iterate_all_kinds(i
, bytes
, v
,
800 copyin((p
+= v
.iov_len
) - v
.iov_len
, v
.iov_base
, v
.iov_len
),
801 memcpy_from_page((p
+= v
.bv_len
) - v
.bv_len
, v
.bv_page
,
802 v
.bv_offset
, v
.bv_len
),
803 memcpy((p
+= v
.iov_len
) - v
.iov_len
, v
.iov_base
, v
.iov_len
)
805 kunmap_atomic(kaddr
);
808 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic
);
810 static inline void pipe_truncate(struct iov_iter
*i
)
812 struct pipe_inode_info
*pipe
= i
->pipe
;
814 size_t off
= i
->iov_offset
;
816 int nrbufs
= (idx
- pipe
->curbuf
) & (pipe
->buffers
- 1);
818 pipe
->bufs
[idx
].len
= off
- pipe
->bufs
[idx
].offset
;
819 idx
= next_idx(idx
, pipe
);
822 while (pipe
->nrbufs
> nrbufs
) {
823 pipe_buf_release(pipe
, &pipe
->bufs
[idx
]);
824 idx
= next_idx(idx
, pipe
);
830 static void pipe_advance(struct iov_iter
*i
, size_t size
)
832 struct pipe_inode_info
*pipe
= i
->pipe
;
833 if (unlikely(i
->count
< size
))
836 struct pipe_buffer
*buf
;
837 size_t off
= i
->iov_offset
, left
= size
;
839 if (off
) /* make it relative to the beginning of buffer */
840 left
+= off
- pipe
->bufs
[idx
].offset
;
842 buf
= &pipe
->bufs
[idx
];
843 if (left
<= buf
->len
)
846 idx
= next_idx(idx
, pipe
);
849 i
->iov_offset
= buf
->offset
+ left
;
852 /* ... and discard everything past that point */
856 void iov_iter_advance(struct iov_iter
*i
, size_t size
)
858 if (unlikely(i
->type
& ITER_PIPE
)) {
859 pipe_advance(i
, size
);
862 iterate_and_advance(i
, size
, v
, 0, 0, 0)
864 EXPORT_SYMBOL(iov_iter_advance
);
866 void iov_iter_revert(struct iov_iter
*i
, size_t unroll
)
870 if (WARN_ON(unroll
> MAX_RW_COUNT
))
873 if (unlikely(i
->type
& ITER_PIPE
)) {
874 struct pipe_inode_info
*pipe
= i
->pipe
;
876 size_t off
= i
->iov_offset
;
878 size_t n
= off
- pipe
->bufs
[idx
].offset
;
884 if (!unroll
&& idx
== i
->start_idx
) {
889 idx
= pipe
->buffers
- 1;
890 off
= pipe
->bufs
[idx
].offset
+ pipe
->bufs
[idx
].len
;
897 if (unroll
<= i
->iov_offset
) {
898 i
->iov_offset
-= unroll
;
901 unroll
-= i
->iov_offset
;
902 if (i
->type
& ITER_BVEC
) {
903 const struct bio_vec
*bvec
= i
->bvec
;
905 size_t n
= (--bvec
)->bv_len
;
909 i
->iov_offset
= n
- unroll
;
914 } else { /* same logics for iovec and kvec */
915 const struct iovec
*iov
= i
->iov
;
917 size_t n
= (--iov
)->iov_len
;
921 i
->iov_offset
= n
- unroll
;
928 EXPORT_SYMBOL(iov_iter_revert
);
931 * Return the count of just the current iov_iter segment.
933 size_t iov_iter_single_seg_count(const struct iov_iter
*i
)
935 if (unlikely(i
->type
& ITER_PIPE
))
936 return i
->count
; // it is a silly place, anyway
939 else if (i
->type
& ITER_BVEC
)
940 return min(i
->count
, i
->bvec
->bv_len
- i
->iov_offset
);
942 return min(i
->count
, i
->iov
->iov_len
- i
->iov_offset
);
944 EXPORT_SYMBOL(iov_iter_single_seg_count
);
946 void iov_iter_kvec(struct iov_iter
*i
, int direction
,
947 const struct kvec
*kvec
, unsigned long nr_segs
,
950 BUG_ON(!(direction
& ITER_KVEC
));
953 i
->nr_segs
= nr_segs
;
957 EXPORT_SYMBOL(iov_iter_kvec
);
959 void iov_iter_bvec(struct iov_iter
*i
, int direction
,
960 const struct bio_vec
*bvec
, unsigned long nr_segs
,
963 BUG_ON(!(direction
& ITER_BVEC
));
966 i
->nr_segs
= nr_segs
;
970 EXPORT_SYMBOL(iov_iter_bvec
);
972 void iov_iter_pipe(struct iov_iter
*i
, int direction
,
973 struct pipe_inode_info
*pipe
,
976 BUG_ON(direction
!= ITER_PIPE
);
977 WARN_ON(pipe
->nrbufs
== pipe
->buffers
);
980 i
->idx
= (pipe
->curbuf
+ pipe
->nrbufs
) & (pipe
->buffers
- 1);
983 i
->start_idx
= i
->idx
;
985 EXPORT_SYMBOL(iov_iter_pipe
);
987 unsigned long iov_iter_alignment(const struct iov_iter
*i
)
989 unsigned long res
= 0;
990 size_t size
= i
->count
;
992 if (unlikely(i
->type
& ITER_PIPE
)) {
993 if (size
&& i
->iov_offset
&& allocated(&i
->pipe
->bufs
[i
->idx
]))
994 return size
| i
->iov_offset
;
997 iterate_all_kinds(i
, size
, v
,
998 (res
|= (unsigned long)v
.iov_base
| v
.iov_len
, 0),
999 res
|= v
.bv_offset
| v
.bv_len
,
1000 res
|= (unsigned long)v
.iov_base
| v
.iov_len
1004 EXPORT_SYMBOL(iov_iter_alignment
);
1006 unsigned long iov_iter_gap_alignment(const struct iov_iter
*i
)
1008 unsigned long res
= 0;
1009 size_t size
= i
->count
;
1011 if (unlikely(i
->type
& ITER_PIPE
)) {
1016 iterate_all_kinds(i
, size
, v
,
1017 (res
|= (!res
? 0 : (unsigned long)v
.iov_base
) |
1018 (size
!= v
.iov_len
? size
: 0), 0),
1019 (res
|= (!res
? 0 : (unsigned long)v
.bv_offset
) |
1020 (size
!= v
.bv_len
? size
: 0)),
1021 (res
|= (!res
? 0 : (unsigned long)v
.iov_base
) |
1022 (size
!= v
.iov_len
? size
: 0))
1026 EXPORT_SYMBOL(iov_iter_gap_alignment
);
1028 static inline size_t __pipe_get_pages(struct iov_iter
*i
,
1030 struct page
**pages
,
1034 struct pipe_inode_info
*pipe
= i
->pipe
;
1035 ssize_t n
= push_pipe(i
, maxsize
, &idx
, start
);
1042 get_page(*pages
++ = pipe
->bufs
[idx
].page
);
1043 idx
= next_idx(idx
, pipe
);
1050 static ssize_t
pipe_get_pages(struct iov_iter
*i
,
1051 struct page
**pages
, size_t maxsize
, unsigned maxpages
,
1064 data_start(i
, &idx
, start
);
1065 /* some of this one + all after this one */
1066 npages
= ((i
->pipe
->curbuf
- idx
- 1) & (i
->pipe
->buffers
- 1)) + 1;
1067 capacity
= min(npages
,maxpages
) * PAGE_SIZE
- *start
;
1069 return __pipe_get_pages(i
, min(maxsize
, capacity
), pages
, idx
, start
);
1072 ssize_t
iov_iter_get_pages(struct iov_iter
*i
,
1073 struct page
**pages
, size_t maxsize
, unsigned maxpages
,
1076 if (maxsize
> i
->count
)
1079 if (unlikely(i
->type
& ITER_PIPE
))
1080 return pipe_get_pages(i
, pages
, maxsize
, maxpages
, start
);
1081 iterate_all_kinds(i
, maxsize
, v
, ({
1082 unsigned long addr
= (unsigned long)v
.iov_base
;
1083 size_t len
= v
.iov_len
+ (*start
= addr
& (PAGE_SIZE
- 1));
1087 if (len
> maxpages
* PAGE_SIZE
)
1088 len
= maxpages
* PAGE_SIZE
;
1089 addr
&= ~(PAGE_SIZE
- 1);
1090 n
= DIV_ROUND_UP(len
, PAGE_SIZE
);
1091 res
= get_user_pages_fast(addr
, n
, (i
->type
& WRITE
) != WRITE
, pages
);
1092 if (unlikely(res
< 0))
1094 return (res
== n
? len
: res
* PAGE_SIZE
) - *start
;
1096 /* can't be more than PAGE_SIZE */
1097 *start
= v
.bv_offset
;
1098 get_page(*pages
= v
.bv_page
);
1106 EXPORT_SYMBOL(iov_iter_get_pages
);
1108 static struct page
**get_pages_array(size_t n
)
1110 return kvmalloc_array(n
, sizeof(struct page
*), GFP_KERNEL
);
1113 static ssize_t
pipe_get_pages_alloc(struct iov_iter
*i
,
1114 struct page
***pages
, size_t maxsize
,
1128 data_start(i
, &idx
, start
);
1129 /* some of this one + all after this one */
1130 npages
= ((i
->pipe
->curbuf
- idx
- 1) & (i
->pipe
->buffers
- 1)) + 1;
1131 n
= npages
* PAGE_SIZE
- *start
;
1135 npages
= DIV_ROUND_UP(maxsize
+ *start
, PAGE_SIZE
);
1136 p
= get_pages_array(npages
);
1139 n
= __pipe_get_pages(i
, maxsize
, p
, idx
, start
);
1147 ssize_t
iov_iter_get_pages_alloc(struct iov_iter
*i
,
1148 struct page
***pages
, size_t maxsize
,
1153 if (maxsize
> i
->count
)
1156 if (unlikely(i
->type
& ITER_PIPE
))
1157 return pipe_get_pages_alloc(i
, pages
, maxsize
, start
);
1158 iterate_all_kinds(i
, maxsize
, v
, ({
1159 unsigned long addr
= (unsigned long)v
.iov_base
;
1160 size_t len
= v
.iov_len
+ (*start
= addr
& (PAGE_SIZE
- 1));
1164 addr
&= ~(PAGE_SIZE
- 1);
1165 n
= DIV_ROUND_UP(len
, PAGE_SIZE
);
1166 p
= get_pages_array(n
);
1169 res
= get_user_pages_fast(addr
, n
, (i
->type
& WRITE
) != WRITE
, p
);
1170 if (unlikely(res
< 0)) {
1175 return (res
== n
? len
: res
* PAGE_SIZE
) - *start
;
1177 /* can't be more than PAGE_SIZE */
1178 *start
= v
.bv_offset
;
1179 *pages
= p
= get_pages_array(1);
1182 get_page(*p
= v
.bv_page
);
1190 EXPORT_SYMBOL(iov_iter_get_pages_alloc
);
1192 size_t csum_and_copy_from_iter(void *addr
, size_t bytes
, __wsum
*csum
,
1199 if (unlikely(i
->type
& ITER_PIPE
)) {
1203 iterate_and_advance(i
, bytes
, v
, ({
1205 next
= csum_and_copy_from_user(v
.iov_base
,
1206 (to
+= v
.iov_len
) - v
.iov_len
,
1207 v
.iov_len
, 0, &err
);
1209 sum
= csum_block_add(sum
, next
, off
);
1212 err
? v
.iov_len
: 0;
1214 char *p
= kmap_atomic(v
.bv_page
);
1215 next
= csum_partial_copy_nocheck(p
+ v
.bv_offset
,
1216 (to
+= v
.bv_len
) - v
.bv_len
,
1219 sum
= csum_block_add(sum
, next
, off
);
1222 next
= csum_partial_copy_nocheck(v
.iov_base
,
1223 (to
+= v
.iov_len
) - v
.iov_len
,
1225 sum
= csum_block_add(sum
, next
, off
);
1232 EXPORT_SYMBOL(csum_and_copy_from_iter
);
1234 bool csum_and_copy_from_iter_full(void *addr
, size_t bytes
, __wsum
*csum
,
1241 if (unlikely(i
->type
& ITER_PIPE
)) {
1245 if (unlikely(i
->count
< bytes
))
1247 iterate_all_kinds(i
, bytes
, v
, ({
1249 next
= csum_and_copy_from_user(v
.iov_base
,
1250 (to
+= v
.iov_len
) - v
.iov_len
,
1251 v
.iov_len
, 0, &err
);
1254 sum
= csum_block_add(sum
, next
, off
);
1258 char *p
= kmap_atomic(v
.bv_page
);
1259 next
= csum_partial_copy_nocheck(p
+ v
.bv_offset
,
1260 (to
+= v
.bv_len
) - v
.bv_len
,
1263 sum
= csum_block_add(sum
, next
, off
);
1266 next
= csum_partial_copy_nocheck(v
.iov_base
,
1267 (to
+= v
.iov_len
) - v
.iov_len
,
1269 sum
= csum_block_add(sum
, next
, off
);
1274 iov_iter_advance(i
, bytes
);
1277 EXPORT_SYMBOL(csum_and_copy_from_iter_full
);
1279 size_t csum_and_copy_to_iter(const void *addr
, size_t bytes
, __wsum
*csum
,
1282 const char *from
= addr
;
1286 if (unlikely(i
->type
& ITER_PIPE
)) {
1287 WARN_ON(1); /* for now */
1290 iterate_and_advance(i
, bytes
, v
, ({
1292 next
= csum_and_copy_to_user((from
+= v
.iov_len
) - v
.iov_len
,
1294 v
.iov_len
, 0, &err
);
1296 sum
= csum_block_add(sum
, next
, off
);
1299 err
? v
.iov_len
: 0;
1301 char *p
= kmap_atomic(v
.bv_page
);
1302 next
= csum_partial_copy_nocheck((from
+= v
.bv_len
) - v
.bv_len
,
1306 sum
= csum_block_add(sum
, next
, off
);
1309 next
= csum_partial_copy_nocheck((from
+= v
.iov_len
) - v
.iov_len
,
1312 sum
= csum_block_add(sum
, next
, off
);
1319 EXPORT_SYMBOL(csum_and_copy_to_iter
);
1321 int iov_iter_npages(const struct iov_iter
*i
, int maxpages
)
1323 size_t size
= i
->count
;
1329 if (unlikely(i
->type
& ITER_PIPE
)) {
1330 struct pipe_inode_info
*pipe
= i
->pipe
;
1337 data_start(i
, &idx
, &off
);
1338 /* some of this one + all after this one */
1339 npages
= ((pipe
->curbuf
- idx
- 1) & (pipe
->buffers
- 1)) + 1;
1340 if (npages
>= maxpages
)
1342 } else iterate_all_kinds(i
, size
, v
, ({
1343 unsigned long p
= (unsigned long)v
.iov_base
;
1344 npages
+= DIV_ROUND_UP(p
+ v
.iov_len
, PAGE_SIZE
)
1346 if (npages
>= maxpages
)
1350 if (npages
>= maxpages
)
1353 unsigned long p
= (unsigned long)v
.iov_base
;
1354 npages
+= DIV_ROUND_UP(p
+ v
.iov_len
, PAGE_SIZE
)
1356 if (npages
>= maxpages
)
1362 EXPORT_SYMBOL(iov_iter_npages
);
1364 const void *dup_iter(struct iov_iter
*new, struct iov_iter
*old
, gfp_t flags
)
1367 if (unlikely(new->type
& ITER_PIPE
)) {
1371 if (new->type
& ITER_BVEC
)
1372 return new->bvec
= kmemdup(new->bvec
,
1373 new->nr_segs
* sizeof(struct bio_vec
),
1376 /* iovec and kvec have identical layout */
1377 return new->iov
= kmemdup(new->iov
,
1378 new->nr_segs
* sizeof(struct iovec
),
1381 EXPORT_SYMBOL(dup_iter
);
1384 * import_iovec() - Copy an array of &struct iovec from userspace
1385 * into the kernel, check that it is valid, and initialize a new
1386 * &struct iov_iter iterator to access it.
1388 * @type: One of %READ or %WRITE.
1389 * @uvector: Pointer to the userspace array.
1390 * @nr_segs: Number of elements in userspace array.
1391 * @fast_segs: Number of elements in @iov.
1392 * @iov: (input and output parameter) Pointer to pointer to (usually small
1393 * on-stack) kernel array.
1394 * @i: Pointer to iterator that will be initialized on success.
1396 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1397 * then this function places %NULL in *@iov on return. Otherwise, a new
1398 * array will be allocated and the result placed in *@iov. This means that
1399 * the caller may call kfree() on *@iov regardless of whether the small
1400 * on-stack array was used or not (and regardless of whether this function
1401 * returns an error or not).
1403 * Return: 0 on success or negative error code on error.
1405 int import_iovec(int type
, const struct iovec __user
* uvector
,
1406 unsigned nr_segs
, unsigned fast_segs
,
1407 struct iovec
**iov
, struct iov_iter
*i
)
1411 n
= rw_copy_check_uvector(type
, uvector
, nr_segs
, fast_segs
,
1419 iov_iter_init(i
, type
, p
, nr_segs
, n
);
1420 *iov
= p
== *iov
? NULL
: p
;
1423 EXPORT_SYMBOL(import_iovec
);
1425 #ifdef CONFIG_COMPAT
1426 #include <linux/compat.h>
1428 int compat_import_iovec(int type
, const struct compat_iovec __user
* uvector
,
1429 unsigned nr_segs
, unsigned fast_segs
,
1430 struct iovec
**iov
, struct iov_iter
*i
)
1434 n
= compat_rw_copy_check_uvector(type
, uvector
, nr_segs
, fast_segs
,
1442 iov_iter_init(i
, type
, p
, nr_segs
, n
);
1443 *iov
= p
== *iov
? NULL
: p
;
1448 int import_single_range(int rw
, void __user
*buf
, size_t len
,
1449 struct iovec
*iov
, struct iov_iter
*i
)
1451 if (len
> MAX_RW_COUNT
)
1453 if (unlikely(!access_ok(!rw
, buf
, len
)))
1456 iov
->iov_base
= buf
;
1458 iov_iter_init(i
, rw
, iov
, 1, len
);
1461 EXPORT_SYMBOL(import_single_range
);