1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <crypto/hash.h>
3 #include <linux/export.h>
4 #include <linux/bvec.h>
5 #include <linux/fault-inject-usercopy.h>
7 #include <linux/pagemap.h>
8 #include <linux/highmem.h>
9 #include <linux/slab.h>
10 #include <linux/vmalloc.h>
11 #include <linux/splice.h>
12 #include <linux/compat.h>
13 #include <net/checksum.h>
14 #include <linux/scatterlist.h>
15 #include <linux/instrumented.h>
17 #define PIPE_PARANOIA /* for now */
19 /* covers ubuf and kbuf alike */
20 #define iterate_buf(i, n, base, len, off, __p, STEP) { \
21 size_t __maybe_unused off = 0; \
23 base = __p + i->iov_offset; \
25 i->iov_offset += len; \
29 /* covers iovec and kvec alike */
30 #define iterate_iovec(i, n, base, len, off, __p, STEP) { \
32 size_t skip = i->iov_offset; \
34 len = min(n, __p->iov_len - skip); \
36 base = __p->iov_base + skip; \
41 if (skip < __p->iov_len) \
47 i->iov_offset = skip; \
51 #define iterate_bvec(i, n, base, len, off, p, STEP) { \
53 unsigned skip = i->iov_offset; \
55 unsigned offset = p->bv_offset + skip; \
57 void *kaddr = kmap_local_page(p->bv_page + \
58 offset / PAGE_SIZE); \
59 base = kaddr + offset % PAGE_SIZE; \
60 len = min(min(n, (size_t)(p->bv_len - skip)), \
61 (size_t)(PAGE_SIZE - offset % PAGE_SIZE)); \
63 kunmap_local(kaddr); \
67 if (skip == p->bv_len) { \
75 i->iov_offset = skip; \
79 #define iterate_xarray(i, n, base, len, __off, STEP) { \
82 struct folio *folio; \
83 loff_t start = i->xarray_start + i->iov_offset; \
84 pgoff_t index = start / PAGE_SIZE; \
85 XA_STATE(xas, i->xarray, index); \
87 len = PAGE_SIZE - offset_in_page(start); \
89 xas_for_each(&xas, folio, ULONG_MAX) { \
92 if (xas_retry(&xas, folio)) \
94 if (WARN_ON(xa_is_value(folio))) \
96 if (WARN_ON(folio_test_hugetlb(folio))) \
98 offset = offset_in_folio(folio, start + __off); \
99 while (offset < folio_size(folio)) { \
100 base = kmap_local_folio(folio, offset); \
103 kunmap_local(base); \
107 if (left || n == 0) \
115 i->iov_offset += __off; \
119 #define __iterate_and_advance(i, n, base, len, off, I, K) { \
120 if (unlikely(i->count < n)) \
123 if (likely(iter_is_ubuf(i))) { \
126 iterate_buf(i, n, base, len, off, \
128 } else if (likely(iter_is_iovec(i))) { \
129 const struct iovec *iov = i->iov; \
132 iterate_iovec(i, n, base, len, off, \
134 i->nr_segs -= iov - i->iov; \
136 } else if (iov_iter_is_bvec(i)) { \
137 const struct bio_vec *bvec = i->bvec; \
140 iterate_bvec(i, n, base, len, off, \
142 i->nr_segs -= bvec - i->bvec; \
144 } else if (iov_iter_is_kvec(i)) { \
145 const struct kvec *kvec = i->kvec; \
148 iterate_iovec(i, n, base, len, off, \
150 i->nr_segs -= kvec - i->kvec; \
152 } else if (iov_iter_is_xarray(i)) { \
155 iterate_xarray(i, n, base, len, off, \
161 #define iterate_and_advance(i, n, base, len, off, I, K) \
162 __iterate_and_advance(i, n, base, len, off, I, ((void)(K),0))
164 static int copyout(void __user
*to
, const void *from
, size_t n
)
166 if (should_fail_usercopy())
168 if (access_ok(to
, n
)) {
169 instrument_copy_to_user(to
, from
, n
);
170 n
= raw_copy_to_user(to
, from
, n
);
175 static int copyin(void *to
, const void __user
*from
, size_t n
)
177 if (should_fail_usercopy())
179 if (access_ok(from
, n
)) {
180 instrument_copy_from_user(to
, from
, n
);
181 n
= raw_copy_from_user(to
, from
, n
);
186 static inline struct pipe_buffer
*pipe_buf(const struct pipe_inode_info
*pipe
,
189 return &pipe
->bufs
[slot
& (pipe
->ring_size
- 1)];
193 static bool sanity(const struct iov_iter
*i
)
195 struct pipe_inode_info
*pipe
= i
->pipe
;
196 unsigned int p_head
= pipe
->head
;
197 unsigned int p_tail
= pipe
->tail
;
198 unsigned int p_occupancy
= pipe_occupancy(p_head
, p_tail
);
199 unsigned int i_head
= i
->head
;
202 if (i
->last_offset
) {
203 struct pipe_buffer
*p
;
204 if (unlikely(p_occupancy
== 0))
205 goto Bad
; // pipe must be non-empty
206 if (unlikely(i_head
!= p_head
- 1))
207 goto Bad
; // must be at the last buffer...
209 p
= pipe_buf(pipe
, i_head
);
210 if (unlikely(p
->offset
+ p
->len
!= abs(i
->last_offset
)))
211 goto Bad
; // ... at the end of segment
213 if (i_head
!= p_head
)
214 goto Bad
; // must be right after the last buffer
218 printk(KERN_ERR
"idx = %d, offset = %d\n", i_head
, i
->last_offset
);
219 printk(KERN_ERR
"head = %d, tail = %d, buffers = %d\n",
220 p_head
, p_tail
, pipe
->ring_size
);
221 for (idx
= 0; idx
< pipe
->ring_size
; idx
++)
222 printk(KERN_ERR
"[%p %p %d %d]\n",
224 pipe
->bufs
[idx
].page
,
225 pipe
->bufs
[idx
].offset
,
226 pipe
->bufs
[idx
].len
);
231 #define sanity(i) true
234 static struct page
*push_anon(struct pipe_inode_info
*pipe
, unsigned size
)
236 struct page
*page
= alloc_page(GFP_USER
);
238 struct pipe_buffer
*buf
= pipe_buf(pipe
, pipe
->head
++);
239 *buf
= (struct pipe_buffer
) {
240 .ops
= &default_pipe_buf_ops
,
249 static void push_page(struct pipe_inode_info
*pipe
, struct page
*page
,
250 unsigned int offset
, unsigned int size
)
252 struct pipe_buffer
*buf
= pipe_buf(pipe
, pipe
->head
++);
253 *buf
= (struct pipe_buffer
) {
254 .ops
= &page_cache_pipe_buf_ops
,
262 static inline int last_offset(const struct pipe_buffer
*buf
)
264 if (buf
->ops
== &default_pipe_buf_ops
)
265 return buf
->len
; // buf->offset is 0 for those
267 return -(buf
->offset
+ buf
->len
);
270 static struct page
*append_pipe(struct iov_iter
*i
, size_t size
,
273 struct pipe_inode_info
*pipe
= i
->pipe
;
274 int offset
= i
->last_offset
;
275 struct pipe_buffer
*buf
;
278 if (offset
> 0 && offset
< PAGE_SIZE
) {
279 // some space in the last buffer; add to it
280 buf
= pipe_buf(pipe
, pipe
->head
- 1);
281 size
= min_t(size_t, size
, PAGE_SIZE
- offset
);
283 i
->last_offset
+= size
;
288 // OK, we need a new buffer
290 size
= min_t(size_t, size
, PAGE_SIZE
);
291 if (pipe_full(pipe
->head
, pipe
->tail
, pipe
->max_usage
))
293 page
= push_anon(pipe
, size
);
296 i
->head
= pipe
->head
- 1;
297 i
->last_offset
= size
;
302 static size_t copy_page_to_iter_pipe(struct page
*page
, size_t offset
, size_t bytes
,
305 struct pipe_inode_info
*pipe
= i
->pipe
;
306 unsigned int head
= pipe
->head
;
308 if (unlikely(bytes
> i
->count
))
311 if (unlikely(!bytes
))
317 if (offset
&& i
->last_offset
== -offset
) { // could we merge it?
318 struct pipe_buffer
*buf
= pipe_buf(pipe
, head
- 1);
319 if (buf
->page
== page
) {
321 i
->last_offset
-= bytes
;
326 if (pipe_full(pipe
->head
, pipe
->tail
, pipe
->max_usage
))
329 push_page(pipe
, page
, offset
, bytes
);
330 i
->last_offset
= -(offset
+ bytes
);
337 * fault_in_iov_iter_readable - fault in iov iterator for reading
339 * @size: maximum length
341 * Fault in one or more iovecs of the given iov_iter, to a maximum length of
342 * @size. For each iovec, fault in each page that constitutes the iovec.
344 * Returns the number of bytes not faulted in (like copy_to_user() and
347 * Always returns 0 for non-userspace iterators.
349 size_t fault_in_iov_iter_readable(const struct iov_iter
*i
, size_t size
)
351 if (iter_is_ubuf(i
)) {
352 size_t n
= min(size
, iov_iter_count(i
));
353 n
-= fault_in_readable(i
->ubuf
+ i
->iov_offset
, n
);
355 } else if (iter_is_iovec(i
)) {
356 size_t count
= min(size
, iov_iter_count(i
));
357 const struct iovec
*p
;
361 for (p
= i
->iov
, skip
= i
->iov_offset
; count
; p
++, skip
= 0) {
362 size_t len
= min(count
, p
->iov_len
- skip
);
367 ret
= fault_in_readable(p
->iov_base
+ skip
, len
);
376 EXPORT_SYMBOL(fault_in_iov_iter_readable
);
379 * fault_in_iov_iter_writeable - fault in iov iterator for writing
381 * @size: maximum length
383 * Faults in the iterator using get_user_pages(), i.e., without triggering
384 * hardware page faults. This is primarily useful when we already know that
385 * some or all of the pages in @i aren't in memory.
387 * Returns the number of bytes not faulted in, like copy_to_user() and
390 * Always returns 0 for non-user-space iterators.
392 size_t fault_in_iov_iter_writeable(const struct iov_iter
*i
, size_t size
)
394 if (iter_is_ubuf(i
)) {
395 size_t n
= min(size
, iov_iter_count(i
));
396 n
-= fault_in_safe_writeable(i
->ubuf
+ i
->iov_offset
, n
);
398 } else if (iter_is_iovec(i
)) {
399 size_t count
= min(size
, iov_iter_count(i
));
400 const struct iovec
*p
;
404 for (p
= i
->iov
, skip
= i
->iov_offset
; count
; p
++, skip
= 0) {
405 size_t len
= min(count
, p
->iov_len
- skip
);
410 ret
= fault_in_safe_writeable(p
->iov_base
+ skip
, len
);
419 EXPORT_SYMBOL(fault_in_iov_iter_writeable
);
421 void iov_iter_init(struct iov_iter
*i
, unsigned int direction
,
422 const struct iovec
*iov
, unsigned long nr_segs
,
425 WARN_ON(direction
& ~(READ
| WRITE
));
426 *i
= (struct iov_iter
) {
427 .iter_type
= ITER_IOVEC
,
430 .data_source
= direction
,
437 EXPORT_SYMBOL(iov_iter_init
);
439 // returns the offset in partial buffer (if any)
440 static inline unsigned int pipe_npages(const struct iov_iter
*i
, int *npages
)
442 struct pipe_inode_info
*pipe
= i
->pipe
;
443 int used
= pipe
->head
- pipe
->tail
;
444 int off
= i
->last_offset
;
446 *npages
= max((int)pipe
->max_usage
- used
, 0);
448 if (off
> 0 && off
< PAGE_SIZE
) { // anon and not full
455 static size_t copy_pipe_to_iter(const void *addr
, size_t bytes
,
458 unsigned int off
, chunk
;
460 if (unlikely(bytes
> i
->count
))
462 if (unlikely(!bytes
))
468 for (size_t n
= bytes
; n
; n
-= chunk
) {
469 struct page
*page
= append_pipe(i
, n
, &off
);
470 chunk
= min_t(size_t, n
, PAGE_SIZE
- off
);
473 memcpy_to_page(page
, off
, addr
, chunk
);
479 static __wsum
csum_and_memcpy(void *to
, const void *from
, size_t len
,
480 __wsum sum
, size_t off
)
482 __wsum next
= csum_partial_copy_nocheck(from
, to
, len
);
483 return csum_block_add(sum
, next
, off
);
486 static size_t csum_and_copy_to_pipe_iter(const void *addr
, size_t bytes
,
487 struct iov_iter
*i
, __wsum
*sump
)
491 unsigned int chunk
, r
;
493 if (unlikely(bytes
> i
->count
))
495 if (unlikely(!bytes
))
502 struct page
*page
= append_pipe(i
, bytes
, &r
);
507 chunk
= min_t(size_t, bytes
, PAGE_SIZE
- r
);
508 p
= kmap_local_page(page
);
509 sum
= csum_and_memcpy(p
+ r
, addr
+ off
, chunk
, sum
, off
);
518 size_t _copy_to_iter(const void *addr
, size_t bytes
, struct iov_iter
*i
)
520 if (unlikely(iov_iter_is_pipe(i
)))
521 return copy_pipe_to_iter(addr
, bytes
, i
);
522 if (user_backed_iter(i
))
524 iterate_and_advance(i
, bytes
, base
, len
, off
,
525 copyout(base
, addr
+ off
, len
),
526 memcpy(base
, addr
+ off
, len
)
531 EXPORT_SYMBOL(_copy_to_iter
);
533 #ifdef CONFIG_ARCH_HAS_COPY_MC
534 static int copyout_mc(void __user
*to
, const void *from
, size_t n
)
536 if (access_ok(to
, n
)) {
537 instrument_copy_to_user(to
, from
, n
);
538 n
= copy_mc_to_user((__force
void *) to
, from
, n
);
543 static size_t copy_mc_pipe_to_iter(const void *addr
, size_t bytes
,
547 unsigned int off
, chunk
;
549 if (unlikely(bytes
> i
->count
))
551 if (unlikely(!bytes
))
558 struct page
*page
= append_pipe(i
, bytes
, &off
);
564 chunk
= min_t(size_t, bytes
, PAGE_SIZE
- off
);
565 p
= kmap_local_page(page
);
566 rem
= copy_mc_to_kernel(p
+ off
, addr
+ xfer
, chunk
);
572 iov_iter_revert(i
, rem
);
580 * _copy_mc_to_iter - copy to iter with source memory error exception handling
581 * @addr: source kernel address
582 * @bytes: total transfer length
583 * @i: destination iterator
585 * The pmem driver deploys this for the dax operation
586 * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
587 * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
588 * successfully copied.
590 * The main differences between this and typical _copy_to_iter().
592 * * Typical tail/residue handling after a fault retries the copy
593 * byte-by-byte until the fault happens again. Re-triggering machine
594 * checks is potentially fatal so the implementation uses source
595 * alignment and poison alignment assumptions to avoid re-triggering
596 * hardware exceptions.
598 * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
599 * Compare to copy_to_iter() where only ITER_IOVEC attempts might return
602 * Return: number of bytes copied (may be %0)
604 size_t _copy_mc_to_iter(const void *addr
, size_t bytes
, struct iov_iter
*i
)
606 if (unlikely(iov_iter_is_pipe(i
)))
607 return copy_mc_pipe_to_iter(addr
, bytes
, i
);
608 if (user_backed_iter(i
))
610 __iterate_and_advance(i
, bytes
, base
, len
, off
,
611 copyout_mc(base
, addr
+ off
, len
),
612 copy_mc_to_kernel(base
, addr
+ off
, len
)
617 EXPORT_SYMBOL_GPL(_copy_mc_to_iter
);
618 #endif /* CONFIG_ARCH_HAS_COPY_MC */
620 size_t _copy_from_iter(void *addr
, size_t bytes
, struct iov_iter
*i
)
622 if (unlikely(iov_iter_is_pipe(i
))) {
626 if (user_backed_iter(i
))
628 iterate_and_advance(i
, bytes
, base
, len
, off
,
629 copyin(addr
+ off
, base
, len
),
630 memcpy(addr
+ off
, base
, len
)
635 EXPORT_SYMBOL(_copy_from_iter
);
637 size_t _copy_from_iter_nocache(void *addr
, size_t bytes
, struct iov_iter
*i
)
639 if (unlikely(iov_iter_is_pipe(i
))) {
643 iterate_and_advance(i
, bytes
, base
, len
, off
,
644 __copy_from_user_inatomic_nocache(addr
+ off
, base
, len
),
645 memcpy(addr
+ off
, base
, len
)
650 EXPORT_SYMBOL(_copy_from_iter_nocache
);
652 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
654 * _copy_from_iter_flushcache - write destination through cpu cache
655 * @addr: destination kernel address
656 * @bytes: total transfer length
657 * @i: source iterator
659 * The pmem driver arranges for filesystem-dax to use this facility via
660 * dax_copy_from_iter() for ensuring that writes to persistent memory
661 * are flushed through the CPU cache. It is differentiated from
662 * _copy_from_iter_nocache() in that guarantees all data is flushed for
663 * all iterator types. The _copy_from_iter_nocache() only attempts to
664 * bypass the cache for the ITER_IOVEC case, and on some archs may use
665 * instructions that strand dirty-data in the cache.
667 * Return: number of bytes copied (may be %0)
669 size_t _copy_from_iter_flushcache(void *addr
, size_t bytes
, struct iov_iter
*i
)
671 if (unlikely(iov_iter_is_pipe(i
))) {
675 iterate_and_advance(i
, bytes
, base
, len
, off
,
676 __copy_from_user_flushcache(addr
+ off
, base
, len
),
677 memcpy_flushcache(addr
+ off
, base
, len
)
682 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache
);
685 static inline bool page_copy_sane(struct page
*page
, size_t offset
, size_t n
)
688 size_t v
= n
+ offset
;
691 * The general case needs to access the page order in order
692 * to compute the page size.
693 * However, we mostly deal with order-0 pages and thus can
694 * avoid a possible cache line miss for requests that fit all
697 if (n
<= v
&& v
<= PAGE_SIZE
)
700 head
= compound_head(page
);
701 v
+= (page
- head
) << PAGE_SHIFT
;
703 if (likely(n
<= v
&& v
<= (page_size(head
))))
709 size_t copy_page_to_iter(struct page
*page
, size_t offset
, size_t bytes
,
713 if (unlikely(!page_copy_sane(page
, offset
, bytes
)))
715 if (unlikely(iov_iter_is_pipe(i
)))
716 return copy_page_to_iter_pipe(page
, offset
, bytes
, i
);
717 page
+= offset
/ PAGE_SIZE
; // first subpage
720 void *kaddr
= kmap_local_page(page
);
721 size_t n
= min(bytes
, (size_t)PAGE_SIZE
- offset
);
722 n
= _copy_to_iter(kaddr
+ offset
, n
, i
);
729 if (offset
== PAGE_SIZE
) {
736 EXPORT_SYMBOL(copy_page_to_iter
);
738 size_t copy_page_from_iter(struct page
*page
, size_t offset
, size_t bytes
,
742 if (!page_copy_sane(page
, offset
, bytes
))
744 page
+= offset
/ PAGE_SIZE
; // first subpage
747 void *kaddr
= kmap_local_page(page
);
748 size_t n
= min(bytes
, (size_t)PAGE_SIZE
- offset
);
749 n
= _copy_from_iter(kaddr
+ offset
, n
, i
);
756 if (offset
== PAGE_SIZE
) {
763 EXPORT_SYMBOL(copy_page_from_iter
);
765 static size_t pipe_zero(size_t bytes
, struct iov_iter
*i
)
767 unsigned int chunk
, off
;
769 if (unlikely(bytes
> i
->count
))
771 if (unlikely(!bytes
))
777 for (size_t n
= bytes
; n
; n
-= chunk
) {
778 struct page
*page
= append_pipe(i
, n
, &off
);
783 chunk
= min_t(size_t, n
, PAGE_SIZE
- off
);
784 p
= kmap_local_page(page
);
785 memset(p
+ off
, 0, chunk
);
791 size_t iov_iter_zero(size_t bytes
, struct iov_iter
*i
)
793 if (unlikely(iov_iter_is_pipe(i
)))
794 return pipe_zero(bytes
, i
);
795 iterate_and_advance(i
, bytes
, base
, len
, count
,
796 clear_user(base
, len
),
802 EXPORT_SYMBOL(iov_iter_zero
);
804 size_t copy_page_from_iter_atomic(struct page
*page
, unsigned offset
, size_t bytes
,
807 char *kaddr
= kmap_atomic(page
), *p
= kaddr
+ offset
;
808 if (unlikely(!page_copy_sane(page
, offset
, bytes
))) {
809 kunmap_atomic(kaddr
);
812 if (unlikely(iov_iter_is_pipe(i
) || iov_iter_is_discard(i
))) {
813 kunmap_atomic(kaddr
);
817 iterate_and_advance(i
, bytes
, base
, len
, off
,
818 copyin(p
+ off
, base
, len
),
819 memcpy(p
+ off
, base
, len
)
821 kunmap_atomic(kaddr
);
824 EXPORT_SYMBOL(copy_page_from_iter_atomic
);
826 static void pipe_advance(struct iov_iter
*i
, size_t size
)
828 struct pipe_inode_info
*pipe
= i
->pipe
;
829 int off
= i
->last_offset
;
832 pipe_discard_from(pipe
, i
->start_head
); // discard everything
837 struct pipe_buffer
*buf
= pipe_buf(pipe
, i
->head
);
838 if (off
) /* make it relative to the beginning of buffer */
839 size
+= abs(off
) - buf
->offset
;
840 if (size
<= buf
->len
) {
842 i
->last_offset
= last_offset(buf
);
849 pipe_discard_from(pipe
, i
->head
+ 1); // discard everything past this one
852 static void iov_iter_bvec_advance(struct iov_iter
*i
, size_t size
)
854 const struct bio_vec
*bvec
, *end
;
860 size
+= i
->iov_offset
;
862 for (bvec
= i
->bvec
, end
= bvec
+ i
->nr_segs
; bvec
< end
; bvec
++) {
863 if (likely(size
< bvec
->bv_len
))
865 size
-= bvec
->bv_len
;
867 i
->iov_offset
= size
;
868 i
->nr_segs
-= bvec
- i
->bvec
;
872 static void iov_iter_iovec_advance(struct iov_iter
*i
, size_t size
)
874 const struct iovec
*iov
, *end
;
880 size
+= i
->iov_offset
; // from beginning of current segment
881 for (iov
= i
->iov
, end
= iov
+ i
->nr_segs
; iov
< end
; iov
++) {
882 if (likely(size
< iov
->iov_len
))
884 size
-= iov
->iov_len
;
886 i
->iov_offset
= size
;
887 i
->nr_segs
-= iov
- i
->iov
;
891 void iov_iter_advance(struct iov_iter
*i
, size_t size
)
893 if (unlikely(i
->count
< size
))
895 if (likely(iter_is_ubuf(i
)) || unlikely(iov_iter_is_xarray(i
))) {
896 i
->iov_offset
+= size
;
898 } else if (likely(iter_is_iovec(i
) || iov_iter_is_kvec(i
))) {
899 /* iovec and kvec have identical layouts */
900 iov_iter_iovec_advance(i
, size
);
901 } else if (iov_iter_is_bvec(i
)) {
902 iov_iter_bvec_advance(i
, size
);
903 } else if (iov_iter_is_pipe(i
)) {
904 pipe_advance(i
, size
);
905 } else if (iov_iter_is_discard(i
)) {
909 EXPORT_SYMBOL(iov_iter_advance
);
911 void iov_iter_revert(struct iov_iter
*i
, size_t unroll
)
915 if (WARN_ON(unroll
> MAX_RW_COUNT
))
918 if (unlikely(iov_iter_is_pipe(i
))) {
919 struct pipe_inode_info
*pipe
= i
->pipe
;
920 unsigned int head
= pipe
->head
;
922 while (head
> i
->start_head
) {
923 struct pipe_buffer
*b
= pipe_buf(pipe
, --head
);
924 if (unroll
< b
->len
) {
926 i
->last_offset
= last_offset(b
);
931 pipe_buf_release(pipe
, b
);
938 if (unlikely(iov_iter_is_discard(i
)))
940 if (unroll
<= i
->iov_offset
) {
941 i
->iov_offset
-= unroll
;
944 unroll
-= i
->iov_offset
;
945 if (iov_iter_is_xarray(i
) || iter_is_ubuf(i
)) {
946 BUG(); /* We should never go beyond the start of the specified
947 * range since we might then be straying into pages that
950 } else if (iov_iter_is_bvec(i
)) {
951 const struct bio_vec
*bvec
= i
->bvec
;
953 size_t n
= (--bvec
)->bv_len
;
957 i
->iov_offset
= n
- unroll
;
962 } else { /* same logics for iovec and kvec */
963 const struct iovec
*iov
= i
->iov
;
965 size_t n
= (--iov
)->iov_len
;
969 i
->iov_offset
= n
- unroll
;
976 EXPORT_SYMBOL(iov_iter_revert
);
979 * Return the count of just the current iov_iter segment.
981 size_t iov_iter_single_seg_count(const struct iov_iter
*i
)
983 if (i
->nr_segs
> 1) {
984 if (likely(iter_is_iovec(i
) || iov_iter_is_kvec(i
)))
985 return min(i
->count
, i
->iov
->iov_len
- i
->iov_offset
);
986 if (iov_iter_is_bvec(i
))
987 return min(i
->count
, i
->bvec
->bv_len
- i
->iov_offset
);
991 EXPORT_SYMBOL(iov_iter_single_seg_count
);
993 void iov_iter_kvec(struct iov_iter
*i
, unsigned int direction
,
994 const struct kvec
*kvec
, unsigned long nr_segs
,
997 WARN_ON(direction
& ~(READ
| WRITE
));
998 *i
= (struct iov_iter
){
999 .iter_type
= ITER_KVEC
,
1000 .data_source
= direction
,
1007 EXPORT_SYMBOL(iov_iter_kvec
);
1009 void iov_iter_bvec(struct iov_iter
*i
, unsigned int direction
,
1010 const struct bio_vec
*bvec
, unsigned long nr_segs
,
1013 WARN_ON(direction
& ~(READ
| WRITE
));
1014 *i
= (struct iov_iter
){
1015 .iter_type
= ITER_BVEC
,
1016 .data_source
= direction
,
1023 EXPORT_SYMBOL(iov_iter_bvec
);
1025 void iov_iter_pipe(struct iov_iter
*i
, unsigned int direction
,
1026 struct pipe_inode_info
*pipe
,
1029 BUG_ON(direction
!= READ
);
1030 WARN_ON(pipe_full(pipe
->head
, pipe
->tail
, pipe
->ring_size
));
1031 *i
= (struct iov_iter
){
1032 .iter_type
= ITER_PIPE
,
1033 .data_source
= false,
1036 .start_head
= pipe
->head
,
1041 EXPORT_SYMBOL(iov_iter_pipe
);
1044 * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1045 * @i: The iterator to initialise.
1046 * @direction: The direction of the transfer.
1047 * @xarray: The xarray to access.
1048 * @start: The start file position.
1049 * @count: The size of the I/O buffer in bytes.
1051 * Set up an I/O iterator to either draw data out of the pages attached to an
1052 * inode or to inject data into those pages. The pages *must* be prevented
1053 * from evaporation, either by taking a ref on them or locking them by the
1056 void iov_iter_xarray(struct iov_iter
*i
, unsigned int direction
,
1057 struct xarray
*xarray
, loff_t start
, size_t count
)
1059 BUG_ON(direction
& ~1);
1060 *i
= (struct iov_iter
) {
1061 .iter_type
= ITER_XARRAY
,
1062 .data_source
= direction
,
1064 .xarray_start
= start
,
1069 EXPORT_SYMBOL(iov_iter_xarray
);
1072 * iov_iter_discard - Initialise an I/O iterator that discards data
1073 * @i: The iterator to initialise.
1074 * @direction: The direction of the transfer.
1075 * @count: The size of the I/O buffer in bytes.
1077 * Set up an I/O iterator that just discards everything that's written to it.
1078 * It's only available as a READ iterator.
1080 void iov_iter_discard(struct iov_iter
*i
, unsigned int direction
, size_t count
)
1082 BUG_ON(direction
!= READ
);
1083 *i
= (struct iov_iter
){
1084 .iter_type
= ITER_DISCARD
,
1085 .data_source
= false,
1090 EXPORT_SYMBOL(iov_iter_discard
);
1092 static bool iov_iter_aligned_iovec(const struct iov_iter
*i
, unsigned addr_mask
,
1095 size_t size
= i
->count
;
1096 size_t skip
= i
->iov_offset
;
1099 for (k
= 0; k
< i
->nr_segs
; k
++, skip
= 0) {
1100 size_t len
= i
->iov
[k
].iov_len
- skip
;
1106 if ((unsigned long)(i
->iov
[k
].iov_base
+ skip
) & addr_mask
)
1116 static bool iov_iter_aligned_bvec(const struct iov_iter
*i
, unsigned addr_mask
,
1119 size_t size
= i
->count
;
1120 unsigned skip
= i
->iov_offset
;
1123 for (k
= 0; k
< i
->nr_segs
; k
++, skip
= 0) {
1124 size_t len
= i
->bvec
[k
].bv_len
- skip
;
1130 if ((unsigned long)(i
->bvec
[k
].bv_offset
+ skip
) & addr_mask
)
1141 * iov_iter_is_aligned() - Check if the addresses and lengths of each segments
1142 * are aligned to the parameters.
1144 * @i: &struct iov_iter to restore
1145 * @addr_mask: bit mask to check against the iov element's addresses
1146 * @len_mask: bit mask to check against the iov element's lengths
1148 * Return: false if any addresses or lengths intersect with the provided masks
1150 bool iov_iter_is_aligned(const struct iov_iter
*i
, unsigned addr_mask
,
1153 if (likely(iter_is_ubuf(i
))) {
1154 if (i
->count
& len_mask
)
1156 if ((unsigned long)(i
->ubuf
+ i
->iov_offset
) & addr_mask
)
1161 if (likely(iter_is_iovec(i
) || iov_iter_is_kvec(i
)))
1162 return iov_iter_aligned_iovec(i
, addr_mask
, len_mask
);
1164 if (iov_iter_is_bvec(i
))
1165 return iov_iter_aligned_bvec(i
, addr_mask
, len_mask
);
1167 if (iov_iter_is_pipe(i
)) {
1168 size_t size
= i
->count
;
1170 if (size
& len_mask
)
1172 if (size
&& i
->last_offset
> 0) {
1173 if (i
->last_offset
& addr_mask
)
1180 if (iov_iter_is_xarray(i
)) {
1181 if (i
->count
& len_mask
)
1183 if ((i
->xarray_start
+ i
->iov_offset
) & addr_mask
)
1189 EXPORT_SYMBOL_GPL(iov_iter_is_aligned
);
1191 static unsigned long iov_iter_alignment_iovec(const struct iov_iter
*i
)
1193 unsigned long res
= 0;
1194 size_t size
= i
->count
;
1195 size_t skip
= i
->iov_offset
;
1198 for (k
= 0; k
< i
->nr_segs
; k
++, skip
= 0) {
1199 size_t len
= i
->iov
[k
].iov_len
- skip
;
1201 res
|= (unsigned long)i
->iov
[k
].iov_base
+ skip
;
1213 static unsigned long iov_iter_alignment_bvec(const struct iov_iter
*i
)
1216 size_t size
= i
->count
;
1217 unsigned skip
= i
->iov_offset
;
1220 for (k
= 0; k
< i
->nr_segs
; k
++, skip
= 0) {
1221 size_t len
= i
->bvec
[k
].bv_len
- skip
;
1222 res
|= (unsigned long)i
->bvec
[k
].bv_offset
+ skip
;
1233 unsigned long iov_iter_alignment(const struct iov_iter
*i
)
1235 if (likely(iter_is_ubuf(i
))) {
1236 size_t size
= i
->count
;
1238 return ((unsigned long)i
->ubuf
+ i
->iov_offset
) | size
;
1242 /* iovec and kvec have identical layouts */
1243 if (likely(iter_is_iovec(i
) || iov_iter_is_kvec(i
)))
1244 return iov_iter_alignment_iovec(i
);
1246 if (iov_iter_is_bvec(i
))
1247 return iov_iter_alignment_bvec(i
);
1249 if (iov_iter_is_pipe(i
)) {
1250 size_t size
= i
->count
;
1252 if (size
&& i
->last_offset
> 0)
1253 return size
| i
->last_offset
;
1257 if (iov_iter_is_xarray(i
))
1258 return (i
->xarray_start
+ i
->iov_offset
) | i
->count
;
1262 EXPORT_SYMBOL(iov_iter_alignment
);
1264 unsigned long iov_iter_gap_alignment(const struct iov_iter
*i
)
1266 unsigned long res
= 0;
1267 unsigned long v
= 0;
1268 size_t size
= i
->count
;
1271 if (iter_is_ubuf(i
))
1274 if (WARN_ON(!iter_is_iovec(i
)))
1277 for (k
= 0; k
< i
->nr_segs
; k
++) {
1278 if (i
->iov
[k
].iov_len
) {
1279 unsigned long base
= (unsigned long)i
->iov
[k
].iov_base
;
1280 if (v
) // if not the first one
1281 res
|= base
| v
; // this start | previous end
1282 v
= base
+ i
->iov
[k
].iov_len
;
1283 if (size
<= i
->iov
[k
].iov_len
)
1285 size
-= i
->iov
[k
].iov_len
;
1290 EXPORT_SYMBOL(iov_iter_gap_alignment
);
1292 static int want_pages_array(struct page
***res
, size_t size
,
1293 size_t start
, unsigned int maxpages
)
1295 unsigned int count
= DIV_ROUND_UP(size
+ start
, PAGE_SIZE
);
1297 if (count
> maxpages
)
1299 WARN_ON(!count
); // caller should've prevented that
1301 *res
= kvmalloc_array(count
, sizeof(struct page
*), GFP_KERNEL
);
1308 static ssize_t
pipe_get_pages(struct iov_iter
*i
,
1309 struct page
***pages
, size_t maxsize
, unsigned maxpages
,
1312 unsigned int npages
, count
, off
, chunk
;
1319 *start
= off
= pipe_npages(i
, &npages
);
1322 count
= want_pages_array(pages
, maxsize
, off
, min(npages
, maxpages
));
1326 for (npages
= 0, left
= maxsize
; npages
< count
; npages
++, left
-= chunk
) {
1327 struct page
*page
= append_pipe(i
, left
, &off
);
1330 chunk
= min_t(size_t, left
, PAGE_SIZE
- off
);
1331 get_page(*p
++ = page
);
1335 return maxsize
- left
;
1338 static ssize_t
iter_xarray_populate_pages(struct page
**pages
, struct xarray
*xa
,
1339 pgoff_t index
, unsigned int nr_pages
)
1341 XA_STATE(xas
, xa
, index
);
1343 unsigned int ret
= 0;
1346 for (page
= xas_load(&xas
); page
; page
= xas_next(&xas
)) {
1347 if (xas_retry(&xas
, page
))
1350 /* Has the page moved or been split? */
1351 if (unlikely(page
!= xas_reload(&xas
))) {
1356 pages
[ret
] = find_subpage(page
, xas
.xa_index
);
1357 get_page(pages
[ret
]);
1358 if (++ret
== nr_pages
)
1365 static ssize_t
iter_xarray_get_pages(struct iov_iter
*i
,
1366 struct page
***pages
, size_t maxsize
,
1367 unsigned maxpages
, size_t *_start_offset
)
1369 unsigned nr
, offset
, count
;
1373 pos
= i
->xarray_start
+ i
->iov_offset
;
1374 index
= pos
>> PAGE_SHIFT
;
1375 offset
= pos
& ~PAGE_MASK
;
1376 *_start_offset
= offset
;
1378 count
= want_pages_array(pages
, maxsize
, offset
, maxpages
);
1381 nr
= iter_xarray_populate_pages(*pages
, i
->xarray
, index
, count
);
1385 maxsize
= min_t(size_t, nr
* PAGE_SIZE
- offset
, maxsize
);
1386 i
->iov_offset
+= maxsize
;
1387 i
->count
-= maxsize
;
1391 /* must be done on non-empty ITER_UBUF or ITER_IOVEC one */
1392 static unsigned long first_iovec_segment(const struct iov_iter
*i
, size_t *size
)
1397 if (iter_is_ubuf(i
))
1398 return (unsigned long)i
->ubuf
+ i
->iov_offset
;
1400 for (k
= 0, skip
= i
->iov_offset
; k
< i
->nr_segs
; k
++, skip
= 0) {
1401 size_t len
= i
->iov
[k
].iov_len
- skip
;
1407 return (unsigned long)i
->iov
[k
].iov_base
+ skip
;
1409 BUG(); // if it had been empty, we wouldn't get called
1412 /* must be done on non-empty ITER_BVEC one */
1413 static struct page
*first_bvec_segment(const struct iov_iter
*i
,
1414 size_t *size
, size_t *start
)
1417 size_t skip
= i
->iov_offset
, len
;
1419 len
= i
->bvec
->bv_len
- skip
;
1422 skip
+= i
->bvec
->bv_offset
;
1423 page
= i
->bvec
->bv_page
+ skip
/ PAGE_SIZE
;
1424 *start
= skip
% PAGE_SIZE
;
1428 static ssize_t
__iov_iter_get_pages_alloc(struct iov_iter
*i
,
1429 struct page
***pages
, size_t maxsize
,
1430 unsigned int maxpages
, size_t *start
)
1434 if (maxsize
> i
->count
)
1438 if (maxsize
> MAX_RW_COUNT
)
1439 maxsize
= MAX_RW_COUNT
;
1441 if (likely(user_backed_iter(i
))) {
1442 unsigned int gup_flags
= 0;
1446 if (iov_iter_rw(i
) != WRITE
)
1447 gup_flags
|= FOLL_WRITE
;
1449 gup_flags
|= FOLL_NOFAULT
;
1451 addr
= first_iovec_segment(i
, &maxsize
);
1452 *start
= addr
% PAGE_SIZE
;
1454 n
= want_pages_array(pages
, maxsize
, *start
, maxpages
);
1457 res
= get_user_pages_fast(addr
, n
, gup_flags
, *pages
);
1458 if (unlikely(res
<= 0))
1460 maxsize
= min_t(size_t, maxsize
, res
* PAGE_SIZE
- *start
);
1461 iov_iter_advance(i
, maxsize
);
1464 if (iov_iter_is_bvec(i
)) {
1468 page
= first_bvec_segment(i
, &maxsize
, start
);
1469 n
= want_pages_array(pages
, maxsize
, *start
, maxpages
);
1473 for (int k
= 0; k
< n
; k
++)
1474 get_page(p
[k
] = page
+ k
);
1475 maxsize
= min_t(size_t, maxsize
, n
* PAGE_SIZE
- *start
);
1476 i
->count
-= maxsize
;
1477 i
->iov_offset
+= maxsize
;
1478 if (i
->iov_offset
== i
->bvec
->bv_len
) {
1485 if (iov_iter_is_pipe(i
))
1486 return pipe_get_pages(i
, pages
, maxsize
, maxpages
, start
);
1487 if (iov_iter_is_xarray(i
))
1488 return iter_xarray_get_pages(i
, pages
, maxsize
, maxpages
, start
);
1492 ssize_t
iov_iter_get_pages2(struct iov_iter
*i
,
1493 struct page
**pages
, size_t maxsize
, unsigned maxpages
,
1500 return __iov_iter_get_pages_alloc(i
, &pages
, maxsize
, maxpages
, start
);
1502 EXPORT_SYMBOL(iov_iter_get_pages2
);
1504 ssize_t
iov_iter_get_pages_alloc2(struct iov_iter
*i
,
1505 struct page
***pages
, size_t maxsize
,
1512 len
= __iov_iter_get_pages_alloc(i
, pages
, maxsize
, ~0U, start
);
1519 EXPORT_SYMBOL(iov_iter_get_pages_alloc2
);
1521 size_t csum_and_copy_from_iter(void *addr
, size_t bytes
, __wsum
*csum
,
1526 if (unlikely(iov_iter_is_pipe(i
) || iov_iter_is_discard(i
))) {
1530 iterate_and_advance(i
, bytes
, base
, len
, off
, ({
1531 next
= csum_and_copy_from_user(base
, addr
+ off
, len
);
1532 sum
= csum_block_add(sum
, next
, off
);
1535 sum
= csum_and_memcpy(addr
+ off
, base
, len
, sum
, off
);
1541 EXPORT_SYMBOL(csum_and_copy_from_iter
);
1543 size_t csum_and_copy_to_iter(const void *addr
, size_t bytes
, void *_csstate
,
1546 struct csum_state
*csstate
= _csstate
;
1549 if (unlikely(iov_iter_is_discard(i
))) {
1550 WARN_ON(1); /* for now */
1554 sum
= csum_shift(csstate
->csum
, csstate
->off
);
1555 if (unlikely(iov_iter_is_pipe(i
)))
1556 bytes
= csum_and_copy_to_pipe_iter(addr
, bytes
, i
, &sum
);
1557 else iterate_and_advance(i
, bytes
, base
, len
, off
, ({
1558 next
= csum_and_copy_to_user(addr
+ off
, base
, len
);
1559 sum
= csum_block_add(sum
, next
, off
);
1562 sum
= csum_and_memcpy(base
, addr
+ off
, len
, sum
, off
);
1565 csstate
->csum
= csum_shift(sum
, csstate
->off
);
1566 csstate
->off
+= bytes
;
1569 EXPORT_SYMBOL(csum_and_copy_to_iter
);
1571 size_t hash_and_copy_to_iter(const void *addr
, size_t bytes
, void *hashp
,
1574 #ifdef CONFIG_CRYPTO_HASH
1575 struct ahash_request
*hash
= hashp
;
1576 struct scatterlist sg
;
1579 copied
= copy_to_iter(addr
, bytes
, i
);
1580 sg_init_one(&sg
, addr
, copied
);
1581 ahash_request_set_crypt(hash
, &sg
, NULL
, copied
);
1582 crypto_ahash_update(hash
);
1588 EXPORT_SYMBOL(hash_and_copy_to_iter
);
1590 static int iov_npages(const struct iov_iter
*i
, int maxpages
)
1592 size_t skip
= i
->iov_offset
, size
= i
->count
;
1593 const struct iovec
*p
;
1596 for (p
= i
->iov
; size
; skip
= 0, p
++) {
1597 unsigned offs
= offset_in_page(p
->iov_base
+ skip
);
1598 size_t len
= min(p
->iov_len
- skip
, size
);
1602 npages
+= DIV_ROUND_UP(offs
+ len
, PAGE_SIZE
);
1603 if (unlikely(npages
> maxpages
))
1610 static int bvec_npages(const struct iov_iter
*i
, int maxpages
)
1612 size_t skip
= i
->iov_offset
, size
= i
->count
;
1613 const struct bio_vec
*p
;
1616 for (p
= i
->bvec
; size
; skip
= 0, p
++) {
1617 unsigned offs
= (p
->bv_offset
+ skip
) % PAGE_SIZE
;
1618 size_t len
= min(p
->bv_len
- skip
, size
);
1621 npages
+= DIV_ROUND_UP(offs
+ len
, PAGE_SIZE
);
1622 if (unlikely(npages
> maxpages
))
1628 int iov_iter_npages(const struct iov_iter
*i
, int maxpages
)
1630 if (unlikely(!i
->count
))
1632 if (likely(iter_is_ubuf(i
))) {
1633 unsigned offs
= offset_in_page(i
->ubuf
+ i
->iov_offset
);
1634 int npages
= DIV_ROUND_UP(offs
+ i
->count
, PAGE_SIZE
);
1635 return min(npages
, maxpages
);
1637 /* iovec and kvec have identical layouts */
1638 if (likely(iter_is_iovec(i
) || iov_iter_is_kvec(i
)))
1639 return iov_npages(i
, maxpages
);
1640 if (iov_iter_is_bvec(i
))
1641 return bvec_npages(i
, maxpages
);
1642 if (iov_iter_is_pipe(i
)) {
1648 pipe_npages(i
, &npages
);
1649 return min(npages
, maxpages
);
1651 if (iov_iter_is_xarray(i
)) {
1652 unsigned offset
= (i
->xarray_start
+ i
->iov_offset
) % PAGE_SIZE
;
1653 int npages
= DIV_ROUND_UP(offset
+ i
->count
, PAGE_SIZE
);
1654 return min(npages
, maxpages
);
1658 EXPORT_SYMBOL(iov_iter_npages
);
1660 const void *dup_iter(struct iov_iter
*new, struct iov_iter
*old
, gfp_t flags
)
1663 if (unlikely(iov_iter_is_pipe(new))) {
1667 if (iov_iter_is_bvec(new))
1668 return new->bvec
= kmemdup(new->bvec
,
1669 new->nr_segs
* sizeof(struct bio_vec
),
1671 else if (iov_iter_is_kvec(new) || iter_is_iovec(new))
1672 /* iovec and kvec have identical layout */
1673 return new->iov
= kmemdup(new->iov
,
1674 new->nr_segs
* sizeof(struct iovec
),
1678 EXPORT_SYMBOL(dup_iter
);
1680 static int copy_compat_iovec_from_user(struct iovec
*iov
,
1681 const struct iovec __user
*uvec
, unsigned long nr_segs
)
1683 const struct compat_iovec __user
*uiov
=
1684 (const struct compat_iovec __user
*)uvec
;
1685 int ret
= -EFAULT
, i
;
1687 if (!user_access_begin(uiov
, nr_segs
* sizeof(*uiov
)))
1690 for (i
= 0; i
< nr_segs
; i
++) {
1694 unsafe_get_user(len
, &uiov
[i
].iov_len
, uaccess_end
);
1695 unsafe_get_user(buf
, &uiov
[i
].iov_base
, uaccess_end
);
1697 /* check for compat_size_t not fitting in compat_ssize_t .. */
1702 iov
[i
].iov_base
= compat_ptr(buf
);
1703 iov
[i
].iov_len
= len
;
1712 static int copy_iovec_from_user(struct iovec
*iov
,
1713 const struct iovec __user
*uvec
, unsigned long nr_segs
)
1717 if (copy_from_user(iov
, uvec
, nr_segs
* sizeof(*uvec
)))
1719 for (seg
= 0; seg
< nr_segs
; seg
++) {
1720 if ((ssize_t
)iov
[seg
].iov_len
< 0)
1727 struct iovec
*iovec_from_user(const struct iovec __user
*uvec
,
1728 unsigned long nr_segs
, unsigned long fast_segs
,
1729 struct iovec
*fast_iov
, bool compat
)
1731 struct iovec
*iov
= fast_iov
;
1735 * SuS says "The readv() function *may* fail if the iovcnt argument was
1736 * less than or equal to 0, or greater than {IOV_MAX}. Linux has
1737 * traditionally returned zero for zero segments, so...
1741 if (nr_segs
> UIO_MAXIOV
)
1742 return ERR_PTR(-EINVAL
);
1743 if (nr_segs
> fast_segs
) {
1744 iov
= kmalloc_array(nr_segs
, sizeof(struct iovec
), GFP_KERNEL
);
1746 return ERR_PTR(-ENOMEM
);
1750 ret
= copy_compat_iovec_from_user(iov
, uvec
, nr_segs
);
1752 ret
= copy_iovec_from_user(iov
, uvec
, nr_segs
);
1754 if (iov
!= fast_iov
)
1756 return ERR_PTR(ret
);
1762 ssize_t
__import_iovec(int type
, const struct iovec __user
*uvec
,
1763 unsigned nr_segs
, unsigned fast_segs
, struct iovec
**iovp
,
1764 struct iov_iter
*i
, bool compat
)
1766 ssize_t total_len
= 0;
1770 iov
= iovec_from_user(uvec
, nr_segs
, fast_segs
, *iovp
, compat
);
1773 return PTR_ERR(iov
);
1777 * According to the Single Unix Specification we should return EINVAL if
1778 * an element length is < 0 when cast to ssize_t or if the total length
1779 * would overflow the ssize_t return value of the system call.
1781 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1784 for (seg
= 0; seg
< nr_segs
; seg
++) {
1785 ssize_t len
= (ssize_t
)iov
[seg
].iov_len
;
1787 if (!access_ok(iov
[seg
].iov_base
, len
)) {
1794 if (len
> MAX_RW_COUNT
- total_len
) {
1795 len
= MAX_RW_COUNT
- total_len
;
1796 iov
[seg
].iov_len
= len
;
1801 iov_iter_init(i
, type
, iov
, nr_segs
, total_len
);
1810 * import_iovec() - Copy an array of &struct iovec from userspace
1811 * into the kernel, check that it is valid, and initialize a new
1812 * &struct iov_iter iterator to access it.
1814 * @type: One of %READ or %WRITE.
1815 * @uvec: Pointer to the userspace array.
1816 * @nr_segs: Number of elements in userspace array.
1817 * @fast_segs: Number of elements in @iov.
1818 * @iovp: (input and output parameter) Pointer to pointer to (usually small
1819 * on-stack) kernel array.
1820 * @i: Pointer to iterator that will be initialized on success.
1822 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1823 * then this function places %NULL in *@iov on return. Otherwise, a new
1824 * array will be allocated and the result placed in *@iov. This means that
1825 * the caller may call kfree() on *@iov regardless of whether the small
1826 * on-stack array was used or not (and regardless of whether this function
1827 * returns an error or not).
1829 * Return: Negative error code on error, bytes imported on success
1831 ssize_t
import_iovec(int type
, const struct iovec __user
*uvec
,
1832 unsigned nr_segs
, unsigned fast_segs
,
1833 struct iovec
**iovp
, struct iov_iter
*i
)
1835 return __import_iovec(type
, uvec
, nr_segs
, fast_segs
, iovp
, i
,
1836 in_compat_syscall());
1838 EXPORT_SYMBOL(import_iovec
);
1840 int import_single_range(int rw
, void __user
*buf
, size_t len
,
1841 struct iovec
*iov
, struct iov_iter
*i
)
1843 if (len
> MAX_RW_COUNT
)
1845 if (unlikely(!access_ok(buf
, len
)))
1848 iov
->iov_base
= buf
;
1850 iov_iter_init(i
, rw
, iov
, 1, len
);
1853 EXPORT_SYMBOL(import_single_range
);
1856 * iov_iter_restore() - Restore a &struct iov_iter to the same state as when
1857 * iov_iter_save_state() was called.
1859 * @i: &struct iov_iter to restore
1860 * @state: state to restore from
1862 * Used after iov_iter_save_state() to bring restore @i, if operations may
1865 * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC
1867 void iov_iter_restore(struct iov_iter
*i
, struct iov_iter_state
*state
)
1869 if (WARN_ON_ONCE(!iov_iter_is_bvec(i
) && !iter_is_iovec(i
)) &&
1870 !iov_iter_is_kvec(i
) && !iter_is_ubuf(i
))
1872 i
->iov_offset
= state
->iov_offset
;
1873 i
->count
= state
->count
;
1874 if (iter_is_ubuf(i
))
1877 * For the *vec iters, nr_segs + iov is constant - if we increment
1878 * the vec, then we also decrement the nr_segs count. Hence we don't
1879 * need to track both of these, just one is enough and we can deduct
1880 * the other from that. ITER_KVEC and ITER_IOVEC are the same struct
1881 * size, so we can just increment the iov pointer as they are unionzed.
1882 * ITER_BVEC _may_ be the same size on some archs, but on others it is
1883 * not. Be safe and handle it separately.
1885 BUILD_BUG_ON(sizeof(struct iovec
) != sizeof(struct kvec
));
1886 if (iov_iter_is_bvec(i
))
1887 i
->bvec
-= state
->nr_segs
- i
->nr_segs
;
1889 i
->iov
-= state
->nr_segs
- i
->nr_segs
;
1890 i
->nr_segs
= state
->nr_segs
;