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 iovec and kvec alike */
20 #define iterate_iovec(i, n, base, len, off, __p, STEP) { \
22 size_t skip = i->iov_offset; \
24 len = min(n, __p->iov_len - skip); \
26 base = __p->iov_base + skip; \
31 if (skip < __p->iov_len) \
37 i->iov_offset = skip; \
41 #define iterate_bvec(i, n, base, len, off, p, STEP) { \
43 unsigned skip = i->iov_offset; \
45 unsigned offset = p->bv_offset + skip; \
47 void *kaddr = kmap_local_page(p->bv_page + \
48 offset / PAGE_SIZE); \
49 base = kaddr + offset % PAGE_SIZE; \
50 len = min(min(n, (size_t)(p->bv_len - skip)), \
51 (size_t)(PAGE_SIZE - offset % PAGE_SIZE)); \
53 kunmap_local(kaddr); \
57 if (skip == p->bv_len) { \
65 i->iov_offset = skip; \
69 #define iterate_xarray(i, n, base, len, __off, STEP) { \
72 struct page *head = NULL; \
73 loff_t start = i->xarray_start + i->iov_offset; \
74 unsigned offset = start % PAGE_SIZE; \
75 pgoff_t index = start / PAGE_SIZE; \
78 XA_STATE(xas, i->xarray, index); \
81 xas_for_each(&xas, head, ULONG_MAX) { \
83 if (xas_retry(&xas, head)) \
85 if (WARN_ON(xa_is_value(head))) \
87 if (WARN_ON(PageHuge(head))) \
89 for (j = (head->index < index) ? index - head->index : 0; \
90 j < thp_nr_pages(head); j++) { \
91 void *kaddr = kmap_local_page(head + j); \
92 base = kaddr + offset; \
93 len = PAGE_SIZE - offset; \
96 kunmap_local(kaddr); \
100 if (left || n == 0) \
107 i->iov_offset += __off; \
111 #define __iterate_and_advance(i, n, base, len, off, I, K) { \
112 if (unlikely(i->count < n)) \
115 if (likely(iter_is_iovec(i))) { \
116 const struct iovec *iov = i->iov; \
119 iterate_iovec(i, n, base, len, off, \
121 i->nr_segs -= iov - i->iov; \
123 } else if (iov_iter_is_bvec(i)) { \
124 const struct bio_vec *bvec = i->bvec; \
127 iterate_bvec(i, n, base, len, off, \
129 i->nr_segs -= bvec - i->bvec; \
131 } else if (iov_iter_is_kvec(i)) { \
132 const struct kvec *kvec = i->kvec; \
135 iterate_iovec(i, n, base, len, off, \
137 i->nr_segs -= kvec - i->kvec; \
139 } else if (iov_iter_is_xarray(i)) { \
142 iterate_xarray(i, n, base, len, off, \
148 #define iterate_and_advance(i, n, base, len, off, I, K) \
149 __iterate_and_advance(i, n, base, len, off, I, ((void)(K),0))
151 static int copyout(void __user
*to
, const void *from
, size_t n
)
153 if (should_fail_usercopy())
155 if (access_ok(to
, n
)) {
156 instrument_copy_to_user(to
, from
, n
);
157 n
= raw_copy_to_user(to
, from
, n
);
162 static int copyin(void *to
, const void __user
*from
, size_t n
)
164 if (should_fail_usercopy())
166 if (access_ok(from
, n
)) {
167 instrument_copy_from_user(to
, from
, n
);
168 n
= raw_copy_from_user(to
, from
, n
);
173 static size_t copy_page_to_iter_iovec(struct page
*page
, size_t offset
, size_t bytes
,
176 size_t skip
, copy
, left
, wanted
;
177 const struct iovec
*iov
;
181 if (unlikely(bytes
> i
->count
))
184 if (unlikely(!bytes
))
190 skip
= i
->iov_offset
;
191 buf
= iov
->iov_base
+ skip
;
192 copy
= min(bytes
, iov
->iov_len
- skip
);
194 if (IS_ENABLED(CONFIG_HIGHMEM
) && !fault_in_pages_writeable(buf
, copy
)) {
195 kaddr
= kmap_atomic(page
);
196 from
= kaddr
+ offset
;
198 /* first chunk, usually the only one */
199 left
= copyout(buf
, from
, copy
);
205 while (unlikely(!left
&& bytes
)) {
208 copy
= min(bytes
, iov
->iov_len
);
209 left
= copyout(buf
, from
, copy
);
215 if (likely(!bytes
)) {
216 kunmap_atomic(kaddr
);
219 offset
= from
- kaddr
;
221 kunmap_atomic(kaddr
);
222 copy
= min(bytes
, iov
->iov_len
- skip
);
224 /* Too bad - revert to non-atomic kmap */
227 from
= kaddr
+ offset
;
228 left
= copyout(buf
, from
, copy
);
233 while (unlikely(!left
&& bytes
)) {
236 copy
= min(bytes
, iov
->iov_len
);
237 left
= copyout(buf
, from
, copy
);
246 if (skip
== iov
->iov_len
) {
250 i
->count
-= wanted
- bytes
;
251 i
->nr_segs
-= iov
- i
->iov
;
253 i
->iov_offset
= skip
;
254 return wanted
- bytes
;
257 static size_t copy_page_from_iter_iovec(struct page
*page
, size_t offset
, size_t bytes
,
260 size_t skip
, copy
, left
, wanted
;
261 const struct iovec
*iov
;
265 if (unlikely(bytes
> i
->count
))
268 if (unlikely(!bytes
))
274 skip
= i
->iov_offset
;
275 buf
= iov
->iov_base
+ skip
;
276 copy
= min(bytes
, iov
->iov_len
- skip
);
278 if (IS_ENABLED(CONFIG_HIGHMEM
) && !fault_in_pages_readable(buf
, copy
)) {
279 kaddr
= kmap_atomic(page
);
282 /* first chunk, usually the only one */
283 left
= copyin(to
, buf
, copy
);
289 while (unlikely(!left
&& bytes
)) {
292 copy
= min(bytes
, iov
->iov_len
);
293 left
= copyin(to
, buf
, copy
);
299 if (likely(!bytes
)) {
300 kunmap_atomic(kaddr
);
305 kunmap_atomic(kaddr
);
306 copy
= min(bytes
, iov
->iov_len
- skip
);
308 /* Too bad - revert to non-atomic kmap */
312 left
= copyin(to
, buf
, copy
);
317 while (unlikely(!left
&& bytes
)) {
320 copy
= min(bytes
, iov
->iov_len
);
321 left
= copyin(to
, buf
, copy
);
330 if (skip
== iov
->iov_len
) {
334 i
->count
-= wanted
- bytes
;
335 i
->nr_segs
-= iov
- i
->iov
;
337 i
->iov_offset
= skip
;
338 return wanted
- bytes
;
342 static bool sanity(const struct iov_iter
*i
)
344 struct pipe_inode_info
*pipe
= i
->pipe
;
345 unsigned int p_head
= pipe
->head
;
346 unsigned int p_tail
= pipe
->tail
;
347 unsigned int p_mask
= pipe
->ring_size
- 1;
348 unsigned int p_occupancy
= pipe_occupancy(p_head
, p_tail
);
349 unsigned int i_head
= i
->head
;
353 struct pipe_buffer
*p
;
354 if (unlikely(p_occupancy
== 0))
355 goto Bad
; // pipe must be non-empty
356 if (unlikely(i_head
!= p_head
- 1))
357 goto Bad
; // must be at the last buffer...
359 p
= &pipe
->bufs
[i_head
& p_mask
];
360 if (unlikely(p
->offset
+ p
->len
!= i
->iov_offset
))
361 goto Bad
; // ... at the end of segment
363 if (i_head
!= p_head
)
364 goto Bad
; // must be right after the last buffer
368 printk(KERN_ERR
"idx = %d, offset = %zd\n", i_head
, i
->iov_offset
);
369 printk(KERN_ERR
"head = %d, tail = %d, buffers = %d\n",
370 p_head
, p_tail
, pipe
->ring_size
);
371 for (idx
= 0; idx
< pipe
->ring_size
; idx
++)
372 printk(KERN_ERR
"[%p %p %d %d]\n",
374 pipe
->bufs
[idx
].page
,
375 pipe
->bufs
[idx
].offset
,
376 pipe
->bufs
[idx
].len
);
381 #define sanity(i) true
384 static size_t copy_page_to_iter_pipe(struct page
*page
, size_t offset
, size_t bytes
,
387 struct pipe_inode_info
*pipe
= i
->pipe
;
388 struct pipe_buffer
*buf
;
389 unsigned int p_tail
= pipe
->tail
;
390 unsigned int p_mask
= pipe
->ring_size
- 1;
391 unsigned int i_head
= i
->head
;
394 if (unlikely(bytes
> i
->count
))
397 if (unlikely(!bytes
))
404 buf
= &pipe
->bufs
[i_head
& p_mask
];
406 if (offset
== off
&& buf
->page
== page
) {
407 /* merge with the last one */
409 i
->iov_offset
+= bytes
;
413 buf
= &pipe
->bufs
[i_head
& p_mask
];
415 if (pipe_full(i_head
, p_tail
, pipe
->max_usage
))
418 buf
->ops
= &page_cache_pipe_buf_ops
;
422 buf
->offset
= offset
;
425 pipe
->head
= i_head
+ 1;
426 i
->iov_offset
= offset
+ bytes
;
434 * Fault in one or more iovecs of the given iov_iter, to a maximum length of
435 * bytes. For each iovec, fault in each page that constitutes the iovec.
437 * Return 0 on success, or non-zero if the memory could not be accessed (i.e.
438 * because it is an invalid address).
440 int iov_iter_fault_in_readable(const struct iov_iter
*i
, size_t bytes
)
442 if (iter_is_iovec(i
)) {
443 const struct iovec
*p
;
446 if (bytes
> i
->count
)
448 for (p
= i
->iov
, skip
= i
->iov_offset
; bytes
; p
++, skip
= 0) {
449 size_t len
= min(bytes
, p
->iov_len
- skip
);
454 err
= fault_in_pages_readable(p
->iov_base
+ skip
, len
);
462 EXPORT_SYMBOL(iov_iter_fault_in_readable
);
464 void iov_iter_init(struct iov_iter
*i
, unsigned int direction
,
465 const struct iovec
*iov
, unsigned long nr_segs
,
468 WARN_ON(direction
& ~(READ
| WRITE
));
469 *i
= (struct iov_iter
) {
470 .iter_type
= ITER_IOVEC
,
471 .data_source
= direction
,
478 EXPORT_SYMBOL(iov_iter_init
);
480 static inline bool allocated(struct pipe_buffer
*buf
)
482 return buf
->ops
== &default_pipe_buf_ops
;
485 static inline void data_start(const struct iov_iter
*i
,
486 unsigned int *iter_headp
, size_t *offp
)
488 unsigned int p_mask
= i
->pipe
->ring_size
- 1;
489 unsigned int iter_head
= i
->head
;
490 size_t off
= i
->iov_offset
;
492 if (off
&& (!allocated(&i
->pipe
->bufs
[iter_head
& p_mask
]) ||
497 *iter_headp
= iter_head
;
501 static size_t push_pipe(struct iov_iter
*i
, size_t size
,
502 int *iter_headp
, size_t *offp
)
504 struct pipe_inode_info
*pipe
= i
->pipe
;
505 unsigned int p_tail
= pipe
->tail
;
506 unsigned int p_mask
= pipe
->ring_size
- 1;
507 unsigned int iter_head
;
511 if (unlikely(size
> i
->count
))
517 data_start(i
, &iter_head
, &off
);
518 *iter_headp
= iter_head
;
521 left
-= PAGE_SIZE
- off
;
523 pipe
->bufs
[iter_head
& p_mask
].len
+= size
;
526 pipe
->bufs
[iter_head
& p_mask
].len
= PAGE_SIZE
;
529 while (!pipe_full(iter_head
, p_tail
, pipe
->max_usage
)) {
530 struct pipe_buffer
*buf
= &pipe
->bufs
[iter_head
& p_mask
];
531 struct page
*page
= alloc_page(GFP_USER
);
535 buf
->ops
= &default_pipe_buf_ops
;
539 buf
->len
= min_t(ssize_t
, left
, PAGE_SIZE
);
542 pipe
->head
= iter_head
;
550 static size_t copy_pipe_to_iter(const void *addr
, size_t bytes
,
553 struct pipe_inode_info
*pipe
= i
->pipe
;
554 unsigned int p_mask
= pipe
->ring_size
- 1;
561 bytes
= n
= push_pipe(i
, bytes
, &i_head
, &off
);
565 size_t chunk
= min_t(size_t, n
, PAGE_SIZE
- off
);
566 memcpy_to_page(pipe
->bufs
[i_head
& p_mask
].page
, off
, addr
, chunk
);
568 i
->iov_offset
= off
+ chunk
;
578 static __wsum
csum_and_memcpy(void *to
, const void *from
, size_t len
,
579 __wsum sum
, size_t off
)
581 __wsum next
= csum_partial_copy_nocheck(from
, to
, len
);
582 return csum_block_add(sum
, next
, off
);
585 static size_t csum_and_copy_to_pipe_iter(const void *addr
, size_t bytes
,
586 struct iov_iter
*i
, __wsum
*sump
)
588 struct pipe_inode_info
*pipe
= i
->pipe
;
589 unsigned int p_mask
= pipe
->ring_size
- 1;
598 bytes
= push_pipe(i
, bytes
, &i_head
, &r
);
600 size_t chunk
= min_t(size_t, bytes
, PAGE_SIZE
- r
);
601 char *p
= kmap_local_page(pipe
->bufs
[i_head
& p_mask
].page
);
602 sum
= csum_and_memcpy(p
+ r
, addr
+ off
, chunk
, sum
, off
);
605 i
->iov_offset
= r
+ chunk
;
616 size_t _copy_to_iter(const void *addr
, size_t bytes
, struct iov_iter
*i
)
618 if (unlikely(iov_iter_is_pipe(i
)))
619 return copy_pipe_to_iter(addr
, bytes
, i
);
620 if (iter_is_iovec(i
))
622 iterate_and_advance(i
, bytes
, base
, len
, off
,
623 copyout(base
, addr
+ off
, len
),
624 memcpy(base
, addr
+ off
, len
)
629 EXPORT_SYMBOL(_copy_to_iter
);
631 #ifdef CONFIG_ARCH_HAS_COPY_MC
632 static int copyout_mc(void __user
*to
, const void *from
, size_t n
)
634 if (access_ok(to
, n
)) {
635 instrument_copy_to_user(to
, from
, n
);
636 n
= copy_mc_to_user((__force
void *) to
, from
, n
);
641 static size_t copy_mc_pipe_to_iter(const void *addr
, size_t bytes
,
644 struct pipe_inode_info
*pipe
= i
->pipe
;
645 unsigned int p_mask
= pipe
->ring_size
- 1;
647 size_t n
, off
, xfer
= 0;
652 n
= push_pipe(i
, bytes
, &i_head
, &off
);
654 size_t chunk
= min_t(size_t, n
, PAGE_SIZE
- off
);
655 char *p
= kmap_local_page(pipe
->bufs
[i_head
& p_mask
].page
);
657 rem
= copy_mc_to_kernel(p
+ off
, addr
+ xfer
, chunk
);
661 i
->iov_offset
= off
+ chunk
;
674 * _copy_mc_to_iter - copy to iter with source memory error exception handling
675 * @addr: source kernel address
676 * @bytes: total transfer length
677 * @i: destination iterator
679 * The pmem driver deploys this for the dax operation
680 * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
681 * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
682 * successfully copied.
684 * The main differences between this and typical _copy_to_iter().
686 * * Typical tail/residue handling after a fault retries the copy
687 * byte-by-byte until the fault happens again. Re-triggering machine
688 * checks is potentially fatal so the implementation uses source
689 * alignment and poison alignment assumptions to avoid re-triggering
690 * hardware exceptions.
692 * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
693 * Compare to copy_to_iter() where only ITER_IOVEC attempts might return
696 * Return: number of bytes copied (may be %0)
698 size_t _copy_mc_to_iter(const void *addr
, size_t bytes
, struct iov_iter
*i
)
700 if (unlikely(iov_iter_is_pipe(i
)))
701 return copy_mc_pipe_to_iter(addr
, bytes
, i
);
702 if (iter_is_iovec(i
))
704 __iterate_and_advance(i
, bytes
, base
, len
, off
,
705 copyout_mc(base
, addr
+ off
, len
),
706 copy_mc_to_kernel(base
, addr
+ off
, len
)
711 EXPORT_SYMBOL_GPL(_copy_mc_to_iter
);
712 #endif /* CONFIG_ARCH_HAS_COPY_MC */
714 size_t _copy_from_iter(void *addr
, size_t bytes
, struct iov_iter
*i
)
716 if (unlikely(iov_iter_is_pipe(i
))) {
720 if (iter_is_iovec(i
))
722 iterate_and_advance(i
, bytes
, base
, len
, off
,
723 copyin(addr
+ off
, base
, len
),
724 memcpy(addr
+ off
, base
, len
)
729 EXPORT_SYMBOL(_copy_from_iter
);
731 size_t _copy_from_iter_nocache(void *addr
, size_t bytes
, struct iov_iter
*i
)
733 if (unlikely(iov_iter_is_pipe(i
))) {
737 iterate_and_advance(i
, bytes
, base
, len
, off
,
738 __copy_from_user_inatomic_nocache(addr
+ off
, base
, len
),
739 memcpy(addr
+ off
, base
, len
)
744 EXPORT_SYMBOL(_copy_from_iter_nocache
);
746 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
748 * _copy_from_iter_flushcache - write destination through cpu cache
749 * @addr: destination kernel address
750 * @bytes: total transfer length
751 * @i: source iterator
753 * The pmem driver arranges for filesystem-dax to use this facility via
754 * dax_copy_from_iter() for ensuring that writes to persistent memory
755 * are flushed through the CPU cache. It is differentiated from
756 * _copy_from_iter_nocache() in that guarantees all data is flushed for
757 * all iterator types. The _copy_from_iter_nocache() only attempts to
758 * bypass the cache for the ITER_IOVEC case, and on some archs may use
759 * instructions that strand dirty-data in the cache.
761 * Return: number of bytes copied (may be %0)
763 size_t _copy_from_iter_flushcache(void *addr
, size_t bytes
, struct iov_iter
*i
)
765 if (unlikely(iov_iter_is_pipe(i
))) {
769 iterate_and_advance(i
, bytes
, base
, len
, off
,
770 __copy_from_user_flushcache(addr
+ off
, base
, len
),
771 memcpy_flushcache(addr
+ off
, base
, len
)
776 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache
);
779 static inline bool page_copy_sane(struct page
*page
, size_t offset
, size_t n
)
782 size_t v
= n
+ offset
;
785 * The general case needs to access the page order in order
786 * to compute the page size.
787 * However, we mostly deal with order-0 pages and thus can
788 * avoid a possible cache line miss for requests that fit all
791 if (n
<= v
&& v
<= PAGE_SIZE
)
794 head
= compound_head(page
);
795 v
+= (page
- head
) << PAGE_SHIFT
;
797 if (likely(n
<= v
&& v
<= (page_size(head
))))
803 static size_t __copy_page_to_iter(struct page
*page
, size_t offset
, size_t bytes
,
806 if (likely(iter_is_iovec(i
)))
807 return copy_page_to_iter_iovec(page
, offset
, bytes
, i
);
808 if (iov_iter_is_bvec(i
) || iov_iter_is_kvec(i
) || iov_iter_is_xarray(i
)) {
809 void *kaddr
= kmap_local_page(page
);
810 size_t wanted
= _copy_to_iter(kaddr
+ offset
, bytes
, i
);
814 if (iov_iter_is_pipe(i
))
815 return copy_page_to_iter_pipe(page
, offset
, bytes
, i
);
816 if (unlikely(iov_iter_is_discard(i
))) {
817 if (unlikely(i
->count
< bytes
))
826 size_t copy_page_to_iter(struct page
*page
, size_t offset
, size_t bytes
,
830 if (unlikely(!page_copy_sane(page
, offset
, bytes
)))
832 page
+= offset
/ PAGE_SIZE
; // first subpage
835 size_t n
= __copy_page_to_iter(page
, offset
,
836 min(bytes
, (size_t)PAGE_SIZE
- offset
), i
);
842 if (offset
== PAGE_SIZE
) {
849 EXPORT_SYMBOL(copy_page_to_iter
);
851 size_t copy_page_from_iter(struct page
*page
, size_t offset
, size_t bytes
,
854 if (unlikely(!page_copy_sane(page
, offset
, bytes
)))
856 if (likely(iter_is_iovec(i
)))
857 return copy_page_from_iter_iovec(page
, offset
, bytes
, i
);
858 if (iov_iter_is_bvec(i
) || iov_iter_is_kvec(i
) || iov_iter_is_xarray(i
)) {
859 void *kaddr
= kmap_local_page(page
);
860 size_t wanted
= _copy_from_iter(kaddr
+ offset
, bytes
, i
);
867 EXPORT_SYMBOL(copy_page_from_iter
);
869 static size_t pipe_zero(size_t bytes
, struct iov_iter
*i
)
871 struct pipe_inode_info
*pipe
= i
->pipe
;
872 unsigned int p_mask
= pipe
->ring_size
- 1;
879 bytes
= n
= push_pipe(i
, bytes
, &i_head
, &off
);
884 size_t chunk
= min_t(size_t, n
, PAGE_SIZE
- off
);
885 char *p
= kmap_local_page(pipe
->bufs
[i_head
& p_mask
].page
);
886 memset(p
+ off
, 0, chunk
);
889 i
->iov_offset
= off
+ chunk
;
898 size_t iov_iter_zero(size_t bytes
, struct iov_iter
*i
)
900 if (unlikely(iov_iter_is_pipe(i
)))
901 return pipe_zero(bytes
, i
);
902 iterate_and_advance(i
, bytes
, base
, len
, count
,
903 clear_user(base
, len
),
909 EXPORT_SYMBOL(iov_iter_zero
);
911 size_t copy_page_from_iter_atomic(struct page
*page
, unsigned offset
, size_t bytes
,
914 char *kaddr
= kmap_atomic(page
), *p
= kaddr
+ offset
;
915 if (unlikely(!page_copy_sane(page
, offset
, bytes
))) {
916 kunmap_atomic(kaddr
);
919 if (unlikely(iov_iter_is_pipe(i
) || iov_iter_is_discard(i
))) {
920 kunmap_atomic(kaddr
);
924 iterate_and_advance(i
, bytes
, base
, len
, off
,
925 copyin(p
+ off
, base
, len
),
926 memcpy(p
+ off
, base
, len
)
928 kunmap_atomic(kaddr
);
931 EXPORT_SYMBOL(copy_page_from_iter_atomic
);
933 static inline void pipe_truncate(struct iov_iter
*i
)
935 struct pipe_inode_info
*pipe
= i
->pipe
;
936 unsigned int p_tail
= pipe
->tail
;
937 unsigned int p_head
= pipe
->head
;
938 unsigned int p_mask
= pipe
->ring_size
- 1;
940 if (!pipe_empty(p_head
, p_tail
)) {
941 struct pipe_buffer
*buf
;
942 unsigned int i_head
= i
->head
;
943 size_t off
= i
->iov_offset
;
946 buf
= &pipe
->bufs
[i_head
& p_mask
];
947 buf
->len
= off
- buf
->offset
;
950 while (p_head
!= i_head
) {
952 pipe_buf_release(pipe
, &pipe
->bufs
[p_head
& p_mask
]);
959 static void pipe_advance(struct iov_iter
*i
, size_t size
)
961 struct pipe_inode_info
*pipe
= i
->pipe
;
963 struct pipe_buffer
*buf
;
964 unsigned int p_mask
= pipe
->ring_size
- 1;
965 unsigned int i_head
= i
->head
;
966 size_t off
= i
->iov_offset
, left
= size
;
968 if (off
) /* make it relative to the beginning of buffer */
969 left
+= off
- pipe
->bufs
[i_head
& p_mask
].offset
;
971 buf
= &pipe
->bufs
[i_head
& p_mask
];
972 if (left
<= buf
->len
)
978 i
->iov_offset
= buf
->offset
+ left
;
981 /* ... and discard everything past that point */
985 static void iov_iter_bvec_advance(struct iov_iter
*i
, size_t size
)
989 bi
.bi_size
= i
->count
;
990 bi
.bi_bvec_done
= i
->iov_offset
;
992 bvec_iter_advance(i
->bvec
, &bi
, size
);
994 i
->bvec
+= bi
.bi_idx
;
995 i
->nr_segs
-= bi
.bi_idx
;
996 i
->count
= bi
.bi_size
;
997 i
->iov_offset
= bi
.bi_bvec_done
;
1000 static void iov_iter_iovec_advance(struct iov_iter
*i
, size_t size
)
1002 const struct iovec
*iov
, *end
;
1008 size
+= i
->iov_offset
; // from beginning of current segment
1009 for (iov
= i
->iov
, end
= iov
+ i
->nr_segs
; iov
< end
; iov
++) {
1010 if (likely(size
< iov
->iov_len
))
1012 size
-= iov
->iov_len
;
1014 i
->iov_offset
= size
;
1015 i
->nr_segs
-= iov
- i
->iov
;
1019 void iov_iter_advance(struct iov_iter
*i
, size_t size
)
1021 if (unlikely(i
->count
< size
))
1023 if (likely(iter_is_iovec(i
) || iov_iter_is_kvec(i
))) {
1024 /* iovec and kvec have identical layouts */
1025 iov_iter_iovec_advance(i
, size
);
1026 } else if (iov_iter_is_bvec(i
)) {
1027 iov_iter_bvec_advance(i
, size
);
1028 } else if (iov_iter_is_pipe(i
)) {
1029 pipe_advance(i
, size
);
1030 } else if (unlikely(iov_iter_is_xarray(i
))) {
1031 i
->iov_offset
+= size
;
1033 } else if (iov_iter_is_discard(i
)) {
1037 EXPORT_SYMBOL(iov_iter_advance
);
1039 void iov_iter_revert(struct iov_iter
*i
, size_t unroll
)
1043 if (WARN_ON(unroll
> MAX_RW_COUNT
))
1046 if (unlikely(iov_iter_is_pipe(i
))) {
1047 struct pipe_inode_info
*pipe
= i
->pipe
;
1048 unsigned int p_mask
= pipe
->ring_size
- 1;
1049 unsigned int i_head
= i
->head
;
1050 size_t off
= i
->iov_offset
;
1052 struct pipe_buffer
*b
= &pipe
->bufs
[i_head
& p_mask
];
1053 size_t n
= off
- b
->offset
;
1059 if (!unroll
&& i_head
== i
->start_head
) {
1064 b
= &pipe
->bufs
[i_head
& p_mask
];
1065 off
= b
->offset
+ b
->len
;
1067 i
->iov_offset
= off
;
1072 if (unlikely(iov_iter_is_discard(i
)))
1074 if (unroll
<= i
->iov_offset
) {
1075 i
->iov_offset
-= unroll
;
1078 unroll
-= i
->iov_offset
;
1079 if (iov_iter_is_xarray(i
)) {
1080 BUG(); /* We should never go beyond the start of the specified
1081 * range since we might then be straying into pages that
1084 } else if (iov_iter_is_bvec(i
)) {
1085 const struct bio_vec
*bvec
= i
->bvec
;
1087 size_t n
= (--bvec
)->bv_len
;
1091 i
->iov_offset
= n
- unroll
;
1096 } else { /* same logics for iovec and kvec */
1097 const struct iovec
*iov
= i
->iov
;
1099 size_t n
= (--iov
)->iov_len
;
1103 i
->iov_offset
= n
- unroll
;
1110 EXPORT_SYMBOL(iov_iter_revert
);
1113 * Return the count of just the current iov_iter segment.
1115 size_t iov_iter_single_seg_count(const struct iov_iter
*i
)
1117 if (i
->nr_segs
> 1) {
1118 if (likely(iter_is_iovec(i
) || iov_iter_is_kvec(i
)))
1119 return min(i
->count
, i
->iov
->iov_len
- i
->iov_offset
);
1120 if (iov_iter_is_bvec(i
))
1121 return min(i
->count
, i
->bvec
->bv_len
- i
->iov_offset
);
1125 EXPORT_SYMBOL(iov_iter_single_seg_count
);
1127 void iov_iter_kvec(struct iov_iter
*i
, unsigned int direction
,
1128 const struct kvec
*kvec
, unsigned long nr_segs
,
1131 WARN_ON(direction
& ~(READ
| WRITE
));
1132 *i
= (struct iov_iter
){
1133 .iter_type
= ITER_KVEC
,
1134 .data_source
= direction
,
1141 EXPORT_SYMBOL(iov_iter_kvec
);
1143 void iov_iter_bvec(struct iov_iter
*i
, unsigned int direction
,
1144 const struct bio_vec
*bvec
, unsigned long nr_segs
,
1147 WARN_ON(direction
& ~(READ
| WRITE
));
1148 *i
= (struct iov_iter
){
1149 .iter_type
= ITER_BVEC
,
1150 .data_source
= direction
,
1157 EXPORT_SYMBOL(iov_iter_bvec
);
1159 void iov_iter_pipe(struct iov_iter
*i
, unsigned int direction
,
1160 struct pipe_inode_info
*pipe
,
1163 BUG_ON(direction
!= READ
);
1164 WARN_ON(pipe_full(pipe
->head
, pipe
->tail
, pipe
->ring_size
));
1165 *i
= (struct iov_iter
){
1166 .iter_type
= ITER_PIPE
,
1167 .data_source
= false,
1170 .start_head
= pipe
->head
,
1175 EXPORT_SYMBOL(iov_iter_pipe
);
1178 * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1179 * @i: The iterator to initialise.
1180 * @direction: The direction of the transfer.
1181 * @xarray: The xarray to access.
1182 * @start: The start file position.
1183 * @count: The size of the I/O buffer in bytes.
1185 * Set up an I/O iterator to either draw data out of the pages attached to an
1186 * inode or to inject data into those pages. The pages *must* be prevented
1187 * from evaporation, either by taking a ref on them or locking them by the
1190 void iov_iter_xarray(struct iov_iter
*i
, unsigned int direction
,
1191 struct xarray
*xarray
, loff_t start
, size_t count
)
1193 BUG_ON(direction
& ~1);
1194 *i
= (struct iov_iter
) {
1195 .iter_type
= ITER_XARRAY
,
1196 .data_source
= direction
,
1198 .xarray_start
= start
,
1203 EXPORT_SYMBOL(iov_iter_xarray
);
1206 * iov_iter_discard - Initialise an I/O iterator that discards data
1207 * @i: The iterator to initialise.
1208 * @direction: The direction of the transfer.
1209 * @count: The size of the I/O buffer in bytes.
1211 * Set up an I/O iterator that just discards everything that's written to it.
1212 * It's only available as a READ iterator.
1214 void iov_iter_discard(struct iov_iter
*i
, unsigned int direction
, size_t count
)
1216 BUG_ON(direction
!= READ
);
1217 *i
= (struct iov_iter
){
1218 .iter_type
= ITER_DISCARD
,
1219 .data_source
= false,
1224 EXPORT_SYMBOL(iov_iter_discard
);
1226 static unsigned long iov_iter_alignment_iovec(const struct iov_iter
*i
)
1228 unsigned long res
= 0;
1229 size_t size
= i
->count
;
1230 size_t skip
= i
->iov_offset
;
1233 for (k
= 0; k
< i
->nr_segs
; k
++, skip
= 0) {
1234 size_t len
= i
->iov
[k
].iov_len
- skip
;
1236 res
|= (unsigned long)i
->iov
[k
].iov_base
+ skip
;
1248 static unsigned long iov_iter_alignment_bvec(const struct iov_iter
*i
)
1251 size_t size
= i
->count
;
1252 unsigned skip
= i
->iov_offset
;
1255 for (k
= 0; k
< i
->nr_segs
; k
++, skip
= 0) {
1256 size_t len
= i
->bvec
[k
].bv_len
- skip
;
1257 res
|= (unsigned long)i
->bvec
[k
].bv_offset
+ skip
;
1268 unsigned long iov_iter_alignment(const struct iov_iter
*i
)
1270 /* iovec and kvec have identical layouts */
1271 if (likely(iter_is_iovec(i
) || iov_iter_is_kvec(i
)))
1272 return iov_iter_alignment_iovec(i
);
1274 if (iov_iter_is_bvec(i
))
1275 return iov_iter_alignment_bvec(i
);
1277 if (iov_iter_is_pipe(i
)) {
1278 unsigned int p_mask
= i
->pipe
->ring_size
- 1;
1279 size_t size
= i
->count
;
1281 if (size
&& i
->iov_offset
&& allocated(&i
->pipe
->bufs
[i
->head
& p_mask
]))
1282 return size
| i
->iov_offset
;
1286 if (iov_iter_is_xarray(i
))
1287 return (i
->xarray_start
+ i
->iov_offset
) | i
->count
;
1291 EXPORT_SYMBOL(iov_iter_alignment
);
1293 unsigned long iov_iter_gap_alignment(const struct iov_iter
*i
)
1295 unsigned long res
= 0;
1296 unsigned long v
= 0;
1297 size_t size
= i
->count
;
1300 if (WARN_ON(!iter_is_iovec(i
)))
1303 for (k
= 0; k
< i
->nr_segs
; k
++) {
1304 if (i
->iov
[k
].iov_len
) {
1305 unsigned long base
= (unsigned long)i
->iov
[k
].iov_base
;
1306 if (v
) // if not the first one
1307 res
|= base
| v
; // this start | previous end
1308 v
= base
+ i
->iov
[k
].iov_len
;
1309 if (size
<= i
->iov
[k
].iov_len
)
1311 size
-= i
->iov
[k
].iov_len
;
1316 EXPORT_SYMBOL(iov_iter_gap_alignment
);
1318 static inline ssize_t
__pipe_get_pages(struct iov_iter
*i
,
1320 struct page
**pages
,
1324 struct pipe_inode_info
*pipe
= i
->pipe
;
1325 unsigned int p_mask
= pipe
->ring_size
- 1;
1326 ssize_t n
= push_pipe(i
, maxsize
, &iter_head
, start
);
1333 get_page(*pages
++ = pipe
->bufs
[iter_head
& p_mask
].page
);
1341 static ssize_t
pipe_get_pages(struct iov_iter
*i
,
1342 struct page
**pages
, size_t maxsize
, unsigned maxpages
,
1345 unsigned int iter_head
, npages
;
1351 data_start(i
, &iter_head
, start
);
1352 /* Amount of free space: some of this one + all after this one */
1353 npages
= pipe_space_for_user(iter_head
, i
->pipe
->tail
, i
->pipe
);
1354 capacity
= min(npages
, maxpages
) * PAGE_SIZE
- *start
;
1356 return __pipe_get_pages(i
, min(maxsize
, capacity
), pages
, iter_head
, start
);
1359 static ssize_t
iter_xarray_populate_pages(struct page
**pages
, struct xarray
*xa
,
1360 pgoff_t index
, unsigned int nr_pages
)
1362 XA_STATE(xas
, xa
, index
);
1364 unsigned int ret
= 0;
1367 for (page
= xas_load(&xas
); page
; page
= xas_next(&xas
)) {
1368 if (xas_retry(&xas
, page
))
1371 /* Has the page moved or been split? */
1372 if (unlikely(page
!= xas_reload(&xas
))) {
1377 pages
[ret
] = find_subpage(page
, xas
.xa_index
);
1378 get_page(pages
[ret
]);
1379 if (++ret
== nr_pages
)
1386 static ssize_t
iter_xarray_get_pages(struct iov_iter
*i
,
1387 struct page
**pages
, size_t maxsize
,
1388 unsigned maxpages
, size_t *_start_offset
)
1390 unsigned nr
, offset
;
1391 pgoff_t index
, count
;
1392 size_t size
= maxsize
, actual
;
1395 if (!size
|| !maxpages
)
1398 pos
= i
->xarray_start
+ i
->iov_offset
;
1399 index
= pos
>> PAGE_SHIFT
;
1400 offset
= pos
& ~PAGE_MASK
;
1401 *_start_offset
= offset
;
1404 if (size
> PAGE_SIZE
- offset
) {
1405 size
-= PAGE_SIZE
- offset
;
1406 count
+= size
>> PAGE_SHIFT
;
1412 if (count
> maxpages
)
1415 nr
= iter_xarray_populate_pages(pages
, i
->xarray
, index
, count
);
1419 actual
= PAGE_SIZE
* nr
;
1421 if (nr
== count
&& size
> 0) {
1422 unsigned last_offset
= (nr
> 1) ? 0 : offset
;
1423 actual
-= PAGE_SIZE
- (last_offset
+ size
);
1428 /* must be done on non-empty ITER_IOVEC one */
1429 static unsigned long first_iovec_segment(const struct iov_iter
*i
,
1430 size_t *size
, size_t *start
,
1431 size_t maxsize
, unsigned maxpages
)
1436 for (k
= 0, skip
= i
->iov_offset
; k
< i
->nr_segs
; k
++, skip
= 0) {
1437 unsigned long addr
= (unsigned long)i
->iov
[k
].iov_base
+ skip
;
1438 size_t len
= i
->iov
[k
].iov_len
- skip
;
1444 len
+= (*start
= addr
% PAGE_SIZE
);
1445 if (len
> maxpages
* PAGE_SIZE
)
1446 len
= maxpages
* PAGE_SIZE
;
1448 return addr
& PAGE_MASK
;
1450 BUG(); // if it had been empty, we wouldn't get called
1453 /* must be done on non-empty ITER_BVEC one */
1454 static struct page
*first_bvec_segment(const struct iov_iter
*i
,
1455 size_t *size
, size_t *start
,
1456 size_t maxsize
, unsigned maxpages
)
1459 size_t skip
= i
->iov_offset
, len
;
1461 len
= i
->bvec
->bv_len
- skip
;
1464 skip
+= i
->bvec
->bv_offset
;
1465 page
= i
->bvec
->bv_page
+ skip
/ PAGE_SIZE
;
1466 len
+= (*start
= skip
% PAGE_SIZE
);
1467 if (len
> maxpages
* PAGE_SIZE
)
1468 len
= maxpages
* PAGE_SIZE
;
1473 ssize_t
iov_iter_get_pages(struct iov_iter
*i
,
1474 struct page
**pages
, size_t maxsize
, unsigned maxpages
,
1480 if (maxsize
> i
->count
)
1485 if (likely(iter_is_iovec(i
))) {
1488 addr
= first_iovec_segment(i
, &len
, start
, maxsize
, maxpages
);
1489 n
= DIV_ROUND_UP(len
, PAGE_SIZE
);
1490 res
= get_user_pages_fast(addr
, n
,
1491 iov_iter_rw(i
) != WRITE
? FOLL_WRITE
: 0,
1493 if (unlikely(res
<= 0))
1495 return (res
== n
? len
: res
* PAGE_SIZE
) - *start
;
1497 if (iov_iter_is_bvec(i
)) {
1500 page
= first_bvec_segment(i
, &len
, start
, maxsize
, maxpages
);
1501 n
= DIV_ROUND_UP(len
, PAGE_SIZE
);
1503 get_page(*pages
++ = page
++);
1504 return len
- *start
;
1506 if (iov_iter_is_pipe(i
))
1507 return pipe_get_pages(i
, pages
, maxsize
, maxpages
, start
);
1508 if (iov_iter_is_xarray(i
))
1509 return iter_xarray_get_pages(i
, pages
, maxsize
, maxpages
, start
);
1512 EXPORT_SYMBOL(iov_iter_get_pages
);
1514 static struct page
**get_pages_array(size_t n
)
1516 return kvmalloc_array(n
, sizeof(struct page
*), GFP_KERNEL
);
1519 static ssize_t
pipe_get_pages_alloc(struct iov_iter
*i
,
1520 struct page
***pages
, size_t maxsize
,
1524 unsigned int iter_head
, npages
;
1530 data_start(i
, &iter_head
, start
);
1531 /* Amount of free space: some of this one + all after this one */
1532 npages
= pipe_space_for_user(iter_head
, i
->pipe
->tail
, i
->pipe
);
1533 n
= npages
* PAGE_SIZE
- *start
;
1537 npages
= DIV_ROUND_UP(maxsize
+ *start
, PAGE_SIZE
);
1538 p
= get_pages_array(npages
);
1541 n
= __pipe_get_pages(i
, maxsize
, p
, iter_head
, start
);
1549 static ssize_t
iter_xarray_get_pages_alloc(struct iov_iter
*i
,
1550 struct page
***pages
, size_t maxsize
,
1551 size_t *_start_offset
)
1554 unsigned nr
, offset
;
1555 pgoff_t index
, count
;
1556 size_t size
= maxsize
, actual
;
1562 pos
= i
->xarray_start
+ i
->iov_offset
;
1563 index
= pos
>> PAGE_SHIFT
;
1564 offset
= pos
& ~PAGE_MASK
;
1565 *_start_offset
= offset
;
1568 if (size
> PAGE_SIZE
- offset
) {
1569 size
-= PAGE_SIZE
- offset
;
1570 count
+= size
>> PAGE_SHIFT
;
1576 p
= get_pages_array(count
);
1581 nr
= iter_xarray_populate_pages(p
, i
->xarray
, index
, count
);
1585 actual
= PAGE_SIZE
* nr
;
1587 if (nr
== count
&& size
> 0) {
1588 unsigned last_offset
= (nr
> 1) ? 0 : offset
;
1589 actual
-= PAGE_SIZE
- (last_offset
+ size
);
1594 ssize_t
iov_iter_get_pages_alloc(struct iov_iter
*i
,
1595 struct page
***pages
, size_t maxsize
,
1602 if (maxsize
> i
->count
)
1607 if (likely(iter_is_iovec(i
))) {
1610 addr
= first_iovec_segment(i
, &len
, start
, maxsize
, ~0U);
1611 n
= DIV_ROUND_UP(len
, PAGE_SIZE
);
1612 p
= get_pages_array(n
);
1615 res
= get_user_pages_fast(addr
, n
,
1616 iov_iter_rw(i
) != WRITE
? FOLL_WRITE
: 0, p
);
1617 if (unlikely(res
<= 0)) {
1623 return (res
== n
? len
: res
* PAGE_SIZE
) - *start
;
1625 if (iov_iter_is_bvec(i
)) {
1628 page
= first_bvec_segment(i
, &len
, start
, maxsize
, ~0U);
1629 n
= DIV_ROUND_UP(len
, PAGE_SIZE
);
1630 *pages
= p
= get_pages_array(n
);
1634 get_page(*p
++ = page
++);
1635 return len
- *start
;
1637 if (iov_iter_is_pipe(i
))
1638 return pipe_get_pages_alloc(i
, pages
, maxsize
, start
);
1639 if (iov_iter_is_xarray(i
))
1640 return iter_xarray_get_pages_alloc(i
, pages
, maxsize
, start
);
1643 EXPORT_SYMBOL(iov_iter_get_pages_alloc
);
1645 size_t csum_and_copy_from_iter(void *addr
, size_t bytes
, __wsum
*csum
,
1650 if (unlikely(iov_iter_is_pipe(i
) || iov_iter_is_discard(i
))) {
1654 iterate_and_advance(i
, bytes
, base
, len
, off
, ({
1655 next
= csum_and_copy_from_user(base
, addr
+ off
, len
);
1656 sum
= csum_block_add(sum
, next
, off
);
1659 sum
= csum_and_memcpy(addr
+ off
, base
, len
, sum
, off
);
1665 EXPORT_SYMBOL(csum_and_copy_from_iter
);
1667 size_t csum_and_copy_to_iter(const void *addr
, size_t bytes
, void *_csstate
,
1670 struct csum_state
*csstate
= _csstate
;
1673 if (unlikely(iov_iter_is_discard(i
))) {
1674 WARN_ON(1); /* for now */
1678 sum
= csum_shift(csstate
->csum
, csstate
->off
);
1679 if (unlikely(iov_iter_is_pipe(i
)))
1680 bytes
= csum_and_copy_to_pipe_iter(addr
, bytes
, i
, &sum
);
1681 else iterate_and_advance(i
, bytes
, base
, len
, off
, ({
1682 next
= csum_and_copy_to_user(addr
+ off
, base
, len
);
1683 sum
= csum_block_add(sum
, next
, off
);
1686 sum
= csum_and_memcpy(base
, addr
+ off
, len
, sum
, off
);
1689 csstate
->csum
= csum_shift(sum
, csstate
->off
);
1690 csstate
->off
+= bytes
;
1693 EXPORT_SYMBOL(csum_and_copy_to_iter
);
1695 size_t hash_and_copy_to_iter(const void *addr
, size_t bytes
, void *hashp
,
1698 #ifdef CONFIG_CRYPTO_HASH
1699 struct ahash_request
*hash
= hashp
;
1700 struct scatterlist sg
;
1703 copied
= copy_to_iter(addr
, bytes
, i
);
1704 sg_init_one(&sg
, addr
, copied
);
1705 ahash_request_set_crypt(hash
, &sg
, NULL
, copied
);
1706 crypto_ahash_update(hash
);
1712 EXPORT_SYMBOL(hash_and_copy_to_iter
);
1714 static int iov_npages(const struct iov_iter
*i
, int maxpages
)
1716 size_t skip
= i
->iov_offset
, size
= i
->count
;
1717 const struct iovec
*p
;
1720 for (p
= i
->iov
; size
; skip
= 0, p
++) {
1721 unsigned offs
= offset_in_page(p
->iov_base
+ skip
);
1722 size_t len
= min(p
->iov_len
- skip
, size
);
1726 npages
+= DIV_ROUND_UP(offs
+ len
, PAGE_SIZE
);
1727 if (unlikely(npages
> maxpages
))
1734 static int bvec_npages(const struct iov_iter
*i
, int maxpages
)
1736 size_t skip
= i
->iov_offset
, size
= i
->count
;
1737 const struct bio_vec
*p
;
1740 for (p
= i
->bvec
; size
; skip
= 0, p
++) {
1741 unsigned offs
= (p
->bv_offset
+ skip
) % PAGE_SIZE
;
1742 size_t len
= min(p
->bv_len
- skip
, size
);
1745 npages
+= DIV_ROUND_UP(offs
+ len
, PAGE_SIZE
);
1746 if (unlikely(npages
> maxpages
))
1752 int iov_iter_npages(const struct iov_iter
*i
, int maxpages
)
1754 if (unlikely(!i
->count
))
1756 /* iovec and kvec have identical layouts */
1757 if (likely(iter_is_iovec(i
) || iov_iter_is_kvec(i
)))
1758 return iov_npages(i
, maxpages
);
1759 if (iov_iter_is_bvec(i
))
1760 return bvec_npages(i
, maxpages
);
1761 if (iov_iter_is_pipe(i
)) {
1762 unsigned int iter_head
;
1769 data_start(i
, &iter_head
, &off
);
1770 /* some of this one + all after this one */
1771 npages
= pipe_space_for_user(iter_head
, i
->pipe
->tail
, i
->pipe
);
1772 return min(npages
, maxpages
);
1774 if (iov_iter_is_xarray(i
)) {
1775 unsigned offset
= (i
->xarray_start
+ i
->iov_offset
) % PAGE_SIZE
;
1776 int npages
= DIV_ROUND_UP(offset
+ i
->count
, PAGE_SIZE
);
1777 return min(npages
, maxpages
);
1781 EXPORT_SYMBOL(iov_iter_npages
);
1783 const void *dup_iter(struct iov_iter
*new, struct iov_iter
*old
, gfp_t flags
)
1786 if (unlikely(iov_iter_is_pipe(new))) {
1790 if (unlikely(iov_iter_is_discard(new) || iov_iter_is_xarray(new)))
1792 if (iov_iter_is_bvec(new))
1793 return new->bvec
= kmemdup(new->bvec
,
1794 new->nr_segs
* sizeof(struct bio_vec
),
1797 /* iovec and kvec have identical layout */
1798 return new->iov
= kmemdup(new->iov
,
1799 new->nr_segs
* sizeof(struct iovec
),
1802 EXPORT_SYMBOL(dup_iter
);
1804 static int copy_compat_iovec_from_user(struct iovec
*iov
,
1805 const struct iovec __user
*uvec
, unsigned long nr_segs
)
1807 const struct compat_iovec __user
*uiov
=
1808 (const struct compat_iovec __user
*)uvec
;
1809 int ret
= -EFAULT
, i
;
1811 if (!user_access_begin(uiov
, nr_segs
* sizeof(*uiov
)))
1814 for (i
= 0; i
< nr_segs
; i
++) {
1818 unsafe_get_user(len
, &uiov
[i
].iov_len
, uaccess_end
);
1819 unsafe_get_user(buf
, &uiov
[i
].iov_base
, uaccess_end
);
1821 /* check for compat_size_t not fitting in compat_ssize_t .. */
1826 iov
[i
].iov_base
= compat_ptr(buf
);
1827 iov
[i
].iov_len
= len
;
1836 static int copy_iovec_from_user(struct iovec
*iov
,
1837 const struct iovec __user
*uvec
, unsigned long nr_segs
)
1841 if (copy_from_user(iov
, uvec
, nr_segs
* sizeof(*uvec
)))
1843 for (seg
= 0; seg
< nr_segs
; seg
++) {
1844 if ((ssize_t
)iov
[seg
].iov_len
< 0)
1851 struct iovec
*iovec_from_user(const struct iovec __user
*uvec
,
1852 unsigned long nr_segs
, unsigned long fast_segs
,
1853 struct iovec
*fast_iov
, bool compat
)
1855 struct iovec
*iov
= fast_iov
;
1859 * SuS says "The readv() function *may* fail if the iovcnt argument was
1860 * less than or equal to 0, or greater than {IOV_MAX}. Linux has
1861 * traditionally returned zero for zero segments, so...
1865 if (nr_segs
> UIO_MAXIOV
)
1866 return ERR_PTR(-EINVAL
);
1867 if (nr_segs
> fast_segs
) {
1868 iov
= kmalloc_array(nr_segs
, sizeof(struct iovec
), GFP_KERNEL
);
1870 return ERR_PTR(-ENOMEM
);
1874 ret
= copy_compat_iovec_from_user(iov
, uvec
, nr_segs
);
1876 ret
= copy_iovec_from_user(iov
, uvec
, nr_segs
);
1878 if (iov
!= fast_iov
)
1880 return ERR_PTR(ret
);
1886 ssize_t
__import_iovec(int type
, const struct iovec __user
*uvec
,
1887 unsigned nr_segs
, unsigned fast_segs
, struct iovec
**iovp
,
1888 struct iov_iter
*i
, bool compat
)
1890 ssize_t total_len
= 0;
1894 iov
= iovec_from_user(uvec
, nr_segs
, fast_segs
, *iovp
, compat
);
1897 return PTR_ERR(iov
);
1901 * According to the Single Unix Specification we should return EINVAL if
1902 * an element length is < 0 when cast to ssize_t or if the total length
1903 * would overflow the ssize_t return value of the system call.
1905 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1908 for (seg
= 0; seg
< nr_segs
; seg
++) {
1909 ssize_t len
= (ssize_t
)iov
[seg
].iov_len
;
1911 if (!access_ok(iov
[seg
].iov_base
, len
)) {
1918 if (len
> MAX_RW_COUNT
- total_len
) {
1919 len
= MAX_RW_COUNT
- total_len
;
1920 iov
[seg
].iov_len
= len
;
1925 iov_iter_init(i
, type
, iov
, nr_segs
, total_len
);
1934 * import_iovec() - Copy an array of &struct iovec from userspace
1935 * into the kernel, check that it is valid, and initialize a new
1936 * &struct iov_iter iterator to access it.
1938 * @type: One of %READ or %WRITE.
1939 * @uvec: Pointer to the userspace array.
1940 * @nr_segs: Number of elements in userspace array.
1941 * @fast_segs: Number of elements in @iov.
1942 * @iovp: (input and output parameter) Pointer to pointer to (usually small
1943 * on-stack) kernel array.
1944 * @i: Pointer to iterator that will be initialized on success.
1946 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1947 * then this function places %NULL in *@iov on return. Otherwise, a new
1948 * array will be allocated and the result placed in *@iov. This means that
1949 * the caller may call kfree() on *@iov regardless of whether the small
1950 * on-stack array was used or not (and regardless of whether this function
1951 * returns an error or not).
1953 * Return: Negative error code on error, bytes imported on success
1955 ssize_t
import_iovec(int type
, const struct iovec __user
*uvec
,
1956 unsigned nr_segs
, unsigned fast_segs
,
1957 struct iovec
**iovp
, struct iov_iter
*i
)
1959 return __import_iovec(type
, uvec
, nr_segs
, fast_segs
, iovp
, i
,
1960 in_compat_syscall());
1962 EXPORT_SYMBOL(import_iovec
);
1964 int import_single_range(int rw
, void __user
*buf
, size_t len
,
1965 struct iovec
*iov
, struct iov_iter
*i
)
1967 if (len
> MAX_RW_COUNT
)
1969 if (unlikely(!access_ok(buf
, len
)))
1972 iov
->iov_base
= buf
;
1974 iov_iter_init(i
, rw
, iov
, 1, len
);
1977 EXPORT_SYMBOL(import_single_range
);
1980 * iov_iter_restore() - Restore a &struct iov_iter to the same state as when
1981 * iov_iter_save_state() was called.
1983 * @i: &struct iov_iter to restore
1984 * @state: state to restore from
1986 * Used after iov_iter_save_state() to bring restore @i, if operations may
1989 * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC
1991 void iov_iter_restore(struct iov_iter
*i
, struct iov_iter_state
*state
)
1993 if (WARN_ON_ONCE(!iov_iter_is_bvec(i
) && !iter_is_iovec(i
)) &&
1994 !iov_iter_is_kvec(i
))
1996 i
->iov_offset
= state
->iov_offset
;
1997 i
->count
= state
->count
;
1999 * For the *vec iters, nr_segs + iov is constant - if we increment
2000 * the vec, then we also decrement the nr_segs count. Hence we don't
2001 * need to track both of these, just one is enough and we can deduct
2002 * the other from that. ITER_KVEC and ITER_IOVEC are the same struct
2003 * size, so we can just increment the iov pointer as they are unionzed.
2004 * ITER_BVEC _may_ be the same size on some archs, but on others it is
2005 * not. Be safe and handle it separately.
2007 BUILD_BUG_ON(sizeof(struct iovec
) != sizeof(struct kvec
));
2008 if (iov_iter_is_bvec(i
))
2009 i
->bvec
-= state
->nr_segs
- i
->nr_segs
;
2011 i
->iov
-= state
->nr_segs
- i
->nr_segs
;
2012 i
->nr_segs
= state
->nr_segs
;