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_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_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_iov_iter_readable - fault in iov iterator for reading
436 * @size: maximum length
438 * Fault in one or more iovecs of the given iov_iter, to a maximum length of
439 * @size. For each iovec, fault in each page that constitutes the iovec.
441 * Returns the number of bytes not faulted in (like copy_to_user() and
444 * Always returns 0 for non-userspace iterators.
446 size_t fault_in_iov_iter_readable(const struct iov_iter
*i
, size_t size
)
448 if (iter_is_iovec(i
)) {
449 size_t count
= min(size
, iov_iter_count(i
));
450 const struct iovec
*p
;
454 for (p
= i
->iov
, skip
= i
->iov_offset
; count
; p
++, skip
= 0) {
455 size_t len
= min(count
, p
->iov_len
- skip
);
460 ret
= fault_in_readable(p
->iov_base
+ skip
, len
);
469 EXPORT_SYMBOL(fault_in_iov_iter_readable
);
472 * fault_in_iov_iter_writeable - fault in iov iterator for writing
474 * @size: maximum length
476 * Faults in the iterator using get_user_pages(), i.e., without triggering
477 * hardware page faults. This is primarily useful when we already know that
478 * some or all of the pages in @i aren't in memory.
480 * Returns the number of bytes not faulted in, like copy_to_user() and
483 * Always returns 0 for non-user-space iterators.
485 size_t fault_in_iov_iter_writeable(const struct iov_iter
*i
, size_t size
)
487 if (iter_is_iovec(i
)) {
488 size_t count
= min(size
, iov_iter_count(i
));
489 const struct iovec
*p
;
493 for (p
= i
->iov
, skip
= i
->iov_offset
; count
; p
++, skip
= 0) {
494 size_t len
= min(count
, p
->iov_len
- skip
);
499 ret
= fault_in_safe_writeable(p
->iov_base
+ skip
, len
);
508 EXPORT_SYMBOL(fault_in_iov_iter_writeable
);
510 void iov_iter_init(struct iov_iter
*i
, unsigned int direction
,
511 const struct iovec
*iov
, unsigned long nr_segs
,
514 WARN_ON(direction
& ~(READ
| WRITE
));
515 *i
= (struct iov_iter
) {
516 .iter_type
= ITER_IOVEC
,
518 .data_source
= direction
,
525 EXPORT_SYMBOL(iov_iter_init
);
527 static inline bool allocated(struct pipe_buffer
*buf
)
529 return buf
->ops
== &default_pipe_buf_ops
;
532 static inline void data_start(const struct iov_iter
*i
,
533 unsigned int *iter_headp
, size_t *offp
)
535 unsigned int p_mask
= i
->pipe
->ring_size
- 1;
536 unsigned int iter_head
= i
->head
;
537 size_t off
= i
->iov_offset
;
539 if (off
&& (!allocated(&i
->pipe
->bufs
[iter_head
& p_mask
]) ||
544 *iter_headp
= iter_head
;
548 static size_t push_pipe(struct iov_iter
*i
, size_t size
,
549 int *iter_headp
, size_t *offp
)
551 struct pipe_inode_info
*pipe
= i
->pipe
;
552 unsigned int p_tail
= pipe
->tail
;
553 unsigned int p_mask
= pipe
->ring_size
- 1;
554 unsigned int iter_head
;
558 if (unlikely(size
> i
->count
))
564 data_start(i
, &iter_head
, &off
);
565 *iter_headp
= iter_head
;
568 left
-= PAGE_SIZE
- off
;
570 pipe
->bufs
[iter_head
& p_mask
].len
+= size
;
573 pipe
->bufs
[iter_head
& p_mask
].len
= PAGE_SIZE
;
576 while (!pipe_full(iter_head
, p_tail
, pipe
->max_usage
)) {
577 struct pipe_buffer
*buf
= &pipe
->bufs
[iter_head
& p_mask
];
578 struct page
*page
= alloc_page(GFP_USER
);
582 buf
->ops
= &default_pipe_buf_ops
;
586 buf
->len
= min_t(ssize_t
, left
, PAGE_SIZE
);
589 pipe
->head
= iter_head
;
597 static size_t copy_pipe_to_iter(const void *addr
, size_t bytes
,
600 struct pipe_inode_info
*pipe
= i
->pipe
;
601 unsigned int p_mask
= pipe
->ring_size
- 1;
608 bytes
= n
= push_pipe(i
, bytes
, &i_head
, &off
);
612 size_t chunk
= min_t(size_t, n
, PAGE_SIZE
- off
);
613 memcpy_to_page(pipe
->bufs
[i_head
& p_mask
].page
, off
, addr
, chunk
);
615 i
->iov_offset
= off
+ chunk
;
625 static __wsum
csum_and_memcpy(void *to
, const void *from
, size_t len
,
626 __wsum sum
, size_t off
)
628 __wsum next
= csum_partial_copy_nocheck(from
, to
, len
);
629 return csum_block_add(sum
, next
, off
);
632 static size_t csum_and_copy_to_pipe_iter(const void *addr
, size_t bytes
,
633 struct iov_iter
*i
, __wsum
*sump
)
635 struct pipe_inode_info
*pipe
= i
->pipe
;
636 unsigned int p_mask
= pipe
->ring_size
- 1;
645 bytes
= push_pipe(i
, bytes
, &i_head
, &r
);
647 size_t chunk
= min_t(size_t, bytes
, PAGE_SIZE
- r
);
648 char *p
= kmap_local_page(pipe
->bufs
[i_head
& p_mask
].page
);
649 sum
= csum_and_memcpy(p
+ r
, addr
+ off
, chunk
, sum
, off
);
652 i
->iov_offset
= r
+ chunk
;
663 size_t _copy_to_iter(const void *addr
, size_t bytes
, struct iov_iter
*i
)
665 if (unlikely(iov_iter_is_pipe(i
)))
666 return copy_pipe_to_iter(addr
, bytes
, i
);
667 if (iter_is_iovec(i
))
669 iterate_and_advance(i
, bytes
, base
, len
, off
,
670 copyout(base
, addr
+ off
, len
),
671 memcpy(base
, addr
+ off
, len
)
676 EXPORT_SYMBOL(_copy_to_iter
);
678 #ifdef CONFIG_ARCH_HAS_COPY_MC
679 static int copyout_mc(void __user
*to
, const void *from
, size_t n
)
681 if (access_ok(to
, n
)) {
682 instrument_copy_to_user(to
, from
, n
);
683 n
= copy_mc_to_user((__force
void *) to
, from
, n
);
688 static size_t copy_mc_pipe_to_iter(const void *addr
, size_t bytes
,
691 struct pipe_inode_info
*pipe
= i
->pipe
;
692 unsigned int p_mask
= pipe
->ring_size
- 1;
694 unsigned int valid
= pipe
->head
;
695 size_t n
, off
, xfer
= 0;
700 n
= push_pipe(i
, bytes
, &i_head
, &off
);
702 size_t chunk
= min_t(size_t, n
, PAGE_SIZE
- off
);
703 char *p
= kmap_local_page(pipe
->bufs
[i_head
& p_mask
].page
);
705 rem
= copy_mc_to_kernel(p
+ off
, addr
+ xfer
, chunk
);
710 i
->iov_offset
= off
+ chunk
;
715 pipe
->bufs
[i_head
& p_mask
].len
-= rem
;
716 pipe_discard_from(pipe
, valid
);
728 * _copy_mc_to_iter - copy to iter with source memory error exception handling
729 * @addr: source kernel address
730 * @bytes: total transfer length
731 * @i: destination iterator
733 * The pmem driver deploys this for the dax operation
734 * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
735 * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
736 * successfully copied.
738 * The main differences between this and typical _copy_to_iter().
740 * * Typical tail/residue handling after a fault retries the copy
741 * byte-by-byte until the fault happens again. Re-triggering machine
742 * checks is potentially fatal so the implementation uses source
743 * alignment and poison alignment assumptions to avoid re-triggering
744 * hardware exceptions.
746 * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
747 * Compare to copy_to_iter() where only ITER_IOVEC attempts might return
750 * Return: number of bytes copied (may be %0)
752 size_t _copy_mc_to_iter(const void *addr
, size_t bytes
, struct iov_iter
*i
)
754 if (unlikely(iov_iter_is_pipe(i
)))
755 return copy_mc_pipe_to_iter(addr
, bytes
, i
);
756 if (iter_is_iovec(i
))
758 __iterate_and_advance(i
, bytes
, base
, len
, off
,
759 copyout_mc(base
, addr
+ off
, len
),
760 copy_mc_to_kernel(base
, addr
+ off
, len
)
765 EXPORT_SYMBOL_GPL(_copy_mc_to_iter
);
766 #endif /* CONFIG_ARCH_HAS_COPY_MC */
768 size_t _copy_from_iter(void *addr
, size_t bytes
, struct iov_iter
*i
)
770 if (unlikely(iov_iter_is_pipe(i
))) {
774 if (iter_is_iovec(i
))
776 iterate_and_advance(i
, bytes
, base
, len
, off
,
777 copyin(addr
+ off
, base
, len
),
778 memcpy(addr
+ off
, base
, len
)
783 EXPORT_SYMBOL(_copy_from_iter
);
785 size_t _copy_from_iter_nocache(void *addr
, size_t bytes
, struct iov_iter
*i
)
787 if (unlikely(iov_iter_is_pipe(i
))) {
791 iterate_and_advance(i
, bytes
, base
, len
, off
,
792 __copy_from_user_inatomic_nocache(addr
+ off
, base
, len
),
793 memcpy(addr
+ off
, base
, len
)
798 EXPORT_SYMBOL(_copy_from_iter_nocache
);
800 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
802 * _copy_from_iter_flushcache - write destination through cpu cache
803 * @addr: destination kernel address
804 * @bytes: total transfer length
805 * @i: source iterator
807 * The pmem driver arranges for filesystem-dax to use this facility via
808 * dax_copy_from_iter() for ensuring that writes to persistent memory
809 * are flushed through the CPU cache. It is differentiated from
810 * _copy_from_iter_nocache() in that guarantees all data is flushed for
811 * all iterator types. The _copy_from_iter_nocache() only attempts to
812 * bypass the cache for the ITER_IOVEC case, and on some archs may use
813 * instructions that strand dirty-data in the cache.
815 * Return: number of bytes copied (may be %0)
817 size_t _copy_from_iter_flushcache(void *addr
, size_t bytes
, struct iov_iter
*i
)
819 if (unlikely(iov_iter_is_pipe(i
))) {
823 iterate_and_advance(i
, bytes
, base
, len
, off
,
824 __copy_from_user_flushcache(addr
+ off
, base
, len
),
825 memcpy_flushcache(addr
+ off
, base
, len
)
830 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache
);
833 static inline bool page_copy_sane(struct page
*page
, size_t offset
, size_t n
)
836 size_t v
= n
+ offset
;
839 * The general case needs to access the page order in order
840 * to compute the page size.
841 * However, we mostly deal with order-0 pages and thus can
842 * avoid a possible cache line miss for requests that fit all
845 if (n
<= v
&& v
<= PAGE_SIZE
)
848 head
= compound_head(page
);
849 v
+= (page
- head
) << PAGE_SHIFT
;
851 if (likely(n
<= v
&& v
<= (page_size(head
))))
857 static size_t __copy_page_to_iter(struct page
*page
, size_t offset
, size_t bytes
,
860 if (likely(iter_is_iovec(i
)))
861 return copy_page_to_iter_iovec(page
, offset
, bytes
, i
);
862 if (iov_iter_is_bvec(i
) || iov_iter_is_kvec(i
) || iov_iter_is_xarray(i
)) {
863 void *kaddr
= kmap_local_page(page
);
864 size_t wanted
= _copy_to_iter(kaddr
+ offset
, bytes
, i
);
868 if (iov_iter_is_pipe(i
))
869 return copy_page_to_iter_pipe(page
, offset
, bytes
, i
);
870 if (unlikely(iov_iter_is_discard(i
))) {
871 if (unlikely(i
->count
< bytes
))
880 size_t copy_page_to_iter(struct page
*page
, size_t offset
, size_t bytes
,
884 if (unlikely(!page_copy_sane(page
, offset
, bytes
)))
886 page
+= offset
/ PAGE_SIZE
; // first subpage
889 size_t n
= __copy_page_to_iter(page
, offset
,
890 min(bytes
, (size_t)PAGE_SIZE
- offset
), i
);
896 if (offset
== PAGE_SIZE
) {
903 EXPORT_SYMBOL(copy_page_to_iter
);
905 size_t copy_page_from_iter(struct page
*page
, size_t offset
, size_t bytes
,
908 if (unlikely(!page_copy_sane(page
, offset
, bytes
)))
910 if (likely(iter_is_iovec(i
)))
911 return copy_page_from_iter_iovec(page
, offset
, bytes
, i
);
912 if (iov_iter_is_bvec(i
) || iov_iter_is_kvec(i
) || iov_iter_is_xarray(i
)) {
913 void *kaddr
= kmap_local_page(page
);
914 size_t wanted
= _copy_from_iter(kaddr
+ offset
, bytes
, i
);
921 EXPORT_SYMBOL(copy_page_from_iter
);
923 static size_t pipe_zero(size_t bytes
, struct iov_iter
*i
)
925 struct pipe_inode_info
*pipe
= i
->pipe
;
926 unsigned int p_mask
= pipe
->ring_size
- 1;
933 bytes
= n
= push_pipe(i
, bytes
, &i_head
, &off
);
938 size_t chunk
= min_t(size_t, n
, PAGE_SIZE
- off
);
939 char *p
= kmap_local_page(pipe
->bufs
[i_head
& p_mask
].page
);
940 memset(p
+ off
, 0, chunk
);
943 i
->iov_offset
= off
+ chunk
;
952 size_t iov_iter_zero(size_t bytes
, struct iov_iter
*i
)
954 if (unlikely(iov_iter_is_pipe(i
)))
955 return pipe_zero(bytes
, i
);
956 iterate_and_advance(i
, bytes
, base
, len
, count
,
957 clear_user(base
, len
),
963 EXPORT_SYMBOL(iov_iter_zero
);
965 size_t copy_page_from_iter_atomic(struct page
*page
, unsigned offset
, size_t bytes
,
968 char *kaddr
= kmap_atomic(page
), *p
= kaddr
+ offset
;
969 if (unlikely(!page_copy_sane(page
, offset
, bytes
))) {
970 kunmap_atomic(kaddr
);
973 if (unlikely(iov_iter_is_pipe(i
) || iov_iter_is_discard(i
))) {
974 kunmap_atomic(kaddr
);
978 iterate_and_advance(i
, bytes
, base
, len
, off
,
979 copyin(p
+ off
, base
, len
),
980 memcpy(p
+ off
, base
, len
)
982 kunmap_atomic(kaddr
);
985 EXPORT_SYMBOL(copy_page_from_iter_atomic
);
987 static inline void pipe_truncate(struct iov_iter
*i
)
989 struct pipe_inode_info
*pipe
= i
->pipe
;
990 unsigned int p_tail
= pipe
->tail
;
991 unsigned int p_head
= pipe
->head
;
992 unsigned int p_mask
= pipe
->ring_size
- 1;
994 if (!pipe_empty(p_head
, p_tail
)) {
995 struct pipe_buffer
*buf
;
996 unsigned int i_head
= i
->head
;
997 size_t off
= i
->iov_offset
;
1000 buf
= &pipe
->bufs
[i_head
& p_mask
];
1001 buf
->len
= off
- buf
->offset
;
1004 while (p_head
!= i_head
) {
1006 pipe_buf_release(pipe
, &pipe
->bufs
[p_head
& p_mask
]);
1009 pipe
->head
= p_head
;
1013 static void pipe_advance(struct iov_iter
*i
, size_t size
)
1015 struct pipe_inode_info
*pipe
= i
->pipe
;
1017 struct pipe_buffer
*buf
;
1018 unsigned int p_mask
= pipe
->ring_size
- 1;
1019 unsigned int i_head
= i
->head
;
1020 size_t off
= i
->iov_offset
, left
= size
;
1022 if (off
) /* make it relative to the beginning of buffer */
1023 left
+= off
- pipe
->bufs
[i_head
& p_mask
].offset
;
1025 buf
= &pipe
->bufs
[i_head
& p_mask
];
1026 if (left
<= buf
->len
)
1032 i
->iov_offset
= buf
->offset
+ left
;
1035 /* ... and discard everything past that point */
1039 static void iov_iter_bvec_advance(struct iov_iter
*i
, size_t size
)
1041 struct bvec_iter bi
;
1043 bi
.bi_size
= i
->count
;
1044 bi
.bi_bvec_done
= i
->iov_offset
;
1046 bvec_iter_advance(i
->bvec
, &bi
, size
);
1048 i
->bvec
+= bi
.bi_idx
;
1049 i
->nr_segs
-= bi
.bi_idx
;
1050 i
->count
= bi
.bi_size
;
1051 i
->iov_offset
= bi
.bi_bvec_done
;
1054 static void iov_iter_iovec_advance(struct iov_iter
*i
, size_t size
)
1056 const struct iovec
*iov
, *end
;
1062 size
+= i
->iov_offset
; // from beginning of current segment
1063 for (iov
= i
->iov
, end
= iov
+ i
->nr_segs
; iov
< end
; iov
++) {
1064 if (likely(size
< iov
->iov_len
))
1066 size
-= iov
->iov_len
;
1068 i
->iov_offset
= size
;
1069 i
->nr_segs
-= iov
- i
->iov
;
1073 void iov_iter_advance(struct iov_iter
*i
, size_t size
)
1075 if (unlikely(i
->count
< size
))
1077 if (likely(iter_is_iovec(i
) || iov_iter_is_kvec(i
))) {
1078 /* iovec and kvec have identical layouts */
1079 iov_iter_iovec_advance(i
, size
);
1080 } else if (iov_iter_is_bvec(i
)) {
1081 iov_iter_bvec_advance(i
, size
);
1082 } else if (iov_iter_is_pipe(i
)) {
1083 pipe_advance(i
, size
);
1084 } else if (unlikely(iov_iter_is_xarray(i
))) {
1085 i
->iov_offset
+= size
;
1087 } else if (iov_iter_is_discard(i
)) {
1091 EXPORT_SYMBOL(iov_iter_advance
);
1093 void iov_iter_revert(struct iov_iter
*i
, size_t unroll
)
1097 if (WARN_ON(unroll
> MAX_RW_COUNT
))
1100 if (unlikely(iov_iter_is_pipe(i
))) {
1101 struct pipe_inode_info
*pipe
= i
->pipe
;
1102 unsigned int p_mask
= pipe
->ring_size
- 1;
1103 unsigned int i_head
= i
->head
;
1104 size_t off
= i
->iov_offset
;
1106 struct pipe_buffer
*b
= &pipe
->bufs
[i_head
& p_mask
];
1107 size_t n
= off
- b
->offset
;
1113 if (!unroll
&& i_head
== i
->start_head
) {
1118 b
= &pipe
->bufs
[i_head
& p_mask
];
1119 off
= b
->offset
+ b
->len
;
1121 i
->iov_offset
= off
;
1126 if (unlikely(iov_iter_is_discard(i
)))
1128 if (unroll
<= i
->iov_offset
) {
1129 i
->iov_offset
-= unroll
;
1132 unroll
-= i
->iov_offset
;
1133 if (iov_iter_is_xarray(i
)) {
1134 BUG(); /* We should never go beyond the start of the specified
1135 * range since we might then be straying into pages that
1138 } else if (iov_iter_is_bvec(i
)) {
1139 const struct bio_vec
*bvec
= i
->bvec
;
1141 size_t n
= (--bvec
)->bv_len
;
1145 i
->iov_offset
= n
- unroll
;
1150 } else { /* same logics for iovec and kvec */
1151 const struct iovec
*iov
= i
->iov
;
1153 size_t n
= (--iov
)->iov_len
;
1157 i
->iov_offset
= n
- unroll
;
1164 EXPORT_SYMBOL(iov_iter_revert
);
1167 * Return the count of just the current iov_iter segment.
1169 size_t iov_iter_single_seg_count(const struct iov_iter
*i
)
1171 if (i
->nr_segs
> 1) {
1172 if (likely(iter_is_iovec(i
) || iov_iter_is_kvec(i
)))
1173 return min(i
->count
, i
->iov
->iov_len
- i
->iov_offset
);
1174 if (iov_iter_is_bvec(i
))
1175 return min(i
->count
, i
->bvec
->bv_len
- i
->iov_offset
);
1179 EXPORT_SYMBOL(iov_iter_single_seg_count
);
1181 void iov_iter_kvec(struct iov_iter
*i
, unsigned int direction
,
1182 const struct kvec
*kvec
, unsigned long nr_segs
,
1185 WARN_ON(direction
& ~(READ
| WRITE
));
1186 *i
= (struct iov_iter
){
1187 .iter_type
= ITER_KVEC
,
1188 .data_source
= direction
,
1195 EXPORT_SYMBOL(iov_iter_kvec
);
1197 void iov_iter_bvec(struct iov_iter
*i
, unsigned int direction
,
1198 const struct bio_vec
*bvec
, unsigned long nr_segs
,
1201 WARN_ON(direction
& ~(READ
| WRITE
));
1202 *i
= (struct iov_iter
){
1203 .iter_type
= ITER_BVEC
,
1204 .data_source
= direction
,
1211 EXPORT_SYMBOL(iov_iter_bvec
);
1213 void iov_iter_pipe(struct iov_iter
*i
, unsigned int direction
,
1214 struct pipe_inode_info
*pipe
,
1217 BUG_ON(direction
!= READ
);
1218 WARN_ON(pipe_full(pipe
->head
, pipe
->tail
, pipe
->ring_size
));
1219 *i
= (struct iov_iter
){
1220 .iter_type
= ITER_PIPE
,
1221 .data_source
= false,
1224 .start_head
= pipe
->head
,
1229 EXPORT_SYMBOL(iov_iter_pipe
);
1232 * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1233 * @i: The iterator to initialise.
1234 * @direction: The direction of the transfer.
1235 * @xarray: The xarray to access.
1236 * @start: The start file position.
1237 * @count: The size of the I/O buffer in bytes.
1239 * Set up an I/O iterator to either draw data out of the pages attached to an
1240 * inode or to inject data into those pages. The pages *must* be prevented
1241 * from evaporation, either by taking a ref on them or locking them by the
1244 void iov_iter_xarray(struct iov_iter
*i
, unsigned int direction
,
1245 struct xarray
*xarray
, loff_t start
, size_t count
)
1247 BUG_ON(direction
& ~1);
1248 *i
= (struct iov_iter
) {
1249 .iter_type
= ITER_XARRAY
,
1250 .data_source
= direction
,
1252 .xarray_start
= start
,
1257 EXPORT_SYMBOL(iov_iter_xarray
);
1260 * iov_iter_discard - Initialise an I/O iterator that discards data
1261 * @i: The iterator to initialise.
1262 * @direction: The direction of the transfer.
1263 * @count: The size of the I/O buffer in bytes.
1265 * Set up an I/O iterator that just discards everything that's written to it.
1266 * It's only available as a READ iterator.
1268 void iov_iter_discard(struct iov_iter
*i
, unsigned int direction
, size_t count
)
1270 BUG_ON(direction
!= READ
);
1271 *i
= (struct iov_iter
){
1272 .iter_type
= ITER_DISCARD
,
1273 .data_source
= false,
1278 EXPORT_SYMBOL(iov_iter_discard
);
1280 static unsigned long iov_iter_alignment_iovec(const struct iov_iter
*i
)
1282 unsigned long res
= 0;
1283 size_t size
= i
->count
;
1284 size_t skip
= i
->iov_offset
;
1287 for (k
= 0; k
< i
->nr_segs
; k
++, skip
= 0) {
1288 size_t len
= i
->iov
[k
].iov_len
- skip
;
1290 res
|= (unsigned long)i
->iov
[k
].iov_base
+ skip
;
1302 static unsigned long iov_iter_alignment_bvec(const struct iov_iter
*i
)
1305 size_t size
= i
->count
;
1306 unsigned skip
= i
->iov_offset
;
1309 for (k
= 0; k
< i
->nr_segs
; k
++, skip
= 0) {
1310 size_t len
= i
->bvec
[k
].bv_len
- skip
;
1311 res
|= (unsigned long)i
->bvec
[k
].bv_offset
+ skip
;
1322 unsigned long iov_iter_alignment(const struct iov_iter
*i
)
1324 /* iovec and kvec have identical layouts */
1325 if (likely(iter_is_iovec(i
) || iov_iter_is_kvec(i
)))
1326 return iov_iter_alignment_iovec(i
);
1328 if (iov_iter_is_bvec(i
))
1329 return iov_iter_alignment_bvec(i
);
1331 if (iov_iter_is_pipe(i
)) {
1332 unsigned int p_mask
= i
->pipe
->ring_size
- 1;
1333 size_t size
= i
->count
;
1335 if (size
&& i
->iov_offset
&& allocated(&i
->pipe
->bufs
[i
->head
& p_mask
]))
1336 return size
| i
->iov_offset
;
1340 if (iov_iter_is_xarray(i
))
1341 return (i
->xarray_start
+ i
->iov_offset
) | i
->count
;
1345 EXPORT_SYMBOL(iov_iter_alignment
);
1347 unsigned long iov_iter_gap_alignment(const struct iov_iter
*i
)
1349 unsigned long res
= 0;
1350 unsigned long v
= 0;
1351 size_t size
= i
->count
;
1354 if (WARN_ON(!iter_is_iovec(i
)))
1357 for (k
= 0; k
< i
->nr_segs
; k
++) {
1358 if (i
->iov
[k
].iov_len
) {
1359 unsigned long base
= (unsigned long)i
->iov
[k
].iov_base
;
1360 if (v
) // if not the first one
1361 res
|= base
| v
; // this start | previous end
1362 v
= base
+ i
->iov
[k
].iov_len
;
1363 if (size
<= i
->iov
[k
].iov_len
)
1365 size
-= i
->iov
[k
].iov_len
;
1370 EXPORT_SYMBOL(iov_iter_gap_alignment
);
1372 static inline ssize_t
__pipe_get_pages(struct iov_iter
*i
,
1374 struct page
**pages
,
1378 struct pipe_inode_info
*pipe
= i
->pipe
;
1379 unsigned int p_mask
= pipe
->ring_size
- 1;
1380 ssize_t n
= push_pipe(i
, maxsize
, &iter_head
, start
);
1387 get_page(*pages
++ = pipe
->bufs
[iter_head
& p_mask
].page
);
1395 static ssize_t
pipe_get_pages(struct iov_iter
*i
,
1396 struct page
**pages
, size_t maxsize
, unsigned maxpages
,
1399 unsigned int iter_head
, npages
;
1405 data_start(i
, &iter_head
, start
);
1406 /* Amount of free space: some of this one + all after this one */
1407 npages
= pipe_space_for_user(iter_head
, i
->pipe
->tail
, i
->pipe
);
1408 capacity
= min(npages
, maxpages
) * PAGE_SIZE
- *start
;
1410 return __pipe_get_pages(i
, min(maxsize
, capacity
), pages
, iter_head
, start
);
1413 static ssize_t
iter_xarray_populate_pages(struct page
**pages
, struct xarray
*xa
,
1414 pgoff_t index
, unsigned int nr_pages
)
1416 XA_STATE(xas
, xa
, index
);
1418 unsigned int ret
= 0;
1421 for (page
= xas_load(&xas
); page
; page
= xas_next(&xas
)) {
1422 if (xas_retry(&xas
, page
))
1425 /* Has the page moved or been split? */
1426 if (unlikely(page
!= xas_reload(&xas
))) {
1431 pages
[ret
] = find_subpage(page
, xas
.xa_index
);
1432 get_page(pages
[ret
]);
1433 if (++ret
== nr_pages
)
1440 static ssize_t
iter_xarray_get_pages(struct iov_iter
*i
,
1441 struct page
**pages
, size_t maxsize
,
1442 unsigned maxpages
, size_t *_start_offset
)
1444 unsigned nr
, offset
;
1445 pgoff_t index
, count
;
1446 size_t size
= maxsize
;
1449 if (!size
|| !maxpages
)
1452 pos
= i
->xarray_start
+ i
->iov_offset
;
1453 index
= pos
>> PAGE_SHIFT
;
1454 offset
= pos
& ~PAGE_MASK
;
1455 *_start_offset
= offset
;
1458 if (size
> PAGE_SIZE
- offset
) {
1459 size
-= PAGE_SIZE
- offset
;
1460 count
+= size
>> PAGE_SHIFT
;
1466 if (count
> maxpages
)
1469 nr
= iter_xarray_populate_pages(pages
, i
->xarray
, index
, count
);
1473 return min_t(size_t, nr
* PAGE_SIZE
- offset
, maxsize
);
1476 /* must be done on non-empty ITER_IOVEC one */
1477 static unsigned long first_iovec_segment(const struct iov_iter
*i
,
1478 size_t *size
, size_t *start
,
1479 size_t maxsize
, unsigned maxpages
)
1484 for (k
= 0, skip
= i
->iov_offset
; k
< i
->nr_segs
; k
++, skip
= 0) {
1485 unsigned long addr
= (unsigned long)i
->iov
[k
].iov_base
+ skip
;
1486 size_t len
= i
->iov
[k
].iov_len
- skip
;
1492 len
+= (*start
= addr
% PAGE_SIZE
);
1493 if (len
> maxpages
* PAGE_SIZE
)
1494 len
= maxpages
* PAGE_SIZE
;
1496 return addr
& PAGE_MASK
;
1498 BUG(); // if it had been empty, we wouldn't get called
1501 /* must be done on non-empty ITER_BVEC one */
1502 static struct page
*first_bvec_segment(const struct iov_iter
*i
,
1503 size_t *size
, size_t *start
,
1504 size_t maxsize
, unsigned maxpages
)
1507 size_t skip
= i
->iov_offset
, len
;
1509 len
= i
->bvec
->bv_len
- skip
;
1512 skip
+= i
->bvec
->bv_offset
;
1513 page
= i
->bvec
->bv_page
+ skip
/ PAGE_SIZE
;
1514 len
+= (*start
= skip
% PAGE_SIZE
);
1515 if (len
> maxpages
* PAGE_SIZE
)
1516 len
= maxpages
* PAGE_SIZE
;
1521 ssize_t
iov_iter_get_pages(struct iov_iter
*i
,
1522 struct page
**pages
, size_t maxsize
, unsigned maxpages
,
1528 if (maxsize
> i
->count
)
1533 if (likely(iter_is_iovec(i
))) {
1534 unsigned int gup_flags
= 0;
1537 if (iov_iter_rw(i
) != WRITE
)
1538 gup_flags
|= FOLL_WRITE
;
1540 gup_flags
|= FOLL_NOFAULT
;
1542 addr
= first_iovec_segment(i
, &len
, start
, maxsize
, maxpages
);
1543 n
= DIV_ROUND_UP(len
, PAGE_SIZE
);
1544 res
= get_user_pages_fast(addr
, n
, gup_flags
, pages
);
1545 if (unlikely(res
<= 0))
1547 return (res
== n
? len
: res
* PAGE_SIZE
) - *start
;
1549 if (iov_iter_is_bvec(i
)) {
1552 page
= first_bvec_segment(i
, &len
, start
, maxsize
, maxpages
);
1553 n
= DIV_ROUND_UP(len
, PAGE_SIZE
);
1555 get_page(*pages
++ = page
++);
1556 return len
- *start
;
1558 if (iov_iter_is_pipe(i
))
1559 return pipe_get_pages(i
, pages
, maxsize
, maxpages
, start
);
1560 if (iov_iter_is_xarray(i
))
1561 return iter_xarray_get_pages(i
, pages
, maxsize
, maxpages
, start
);
1564 EXPORT_SYMBOL(iov_iter_get_pages
);
1566 static struct page
**get_pages_array(size_t n
)
1568 return kvmalloc_array(n
, sizeof(struct page
*), GFP_KERNEL
);
1571 static ssize_t
pipe_get_pages_alloc(struct iov_iter
*i
,
1572 struct page
***pages
, size_t maxsize
,
1576 unsigned int iter_head
, npages
;
1582 data_start(i
, &iter_head
, start
);
1583 /* Amount of free space: some of this one + all after this one */
1584 npages
= pipe_space_for_user(iter_head
, i
->pipe
->tail
, i
->pipe
);
1585 n
= npages
* PAGE_SIZE
- *start
;
1589 npages
= DIV_ROUND_UP(maxsize
+ *start
, PAGE_SIZE
);
1590 p
= get_pages_array(npages
);
1593 n
= __pipe_get_pages(i
, maxsize
, p
, iter_head
, start
);
1601 static ssize_t
iter_xarray_get_pages_alloc(struct iov_iter
*i
,
1602 struct page
***pages
, size_t maxsize
,
1603 size_t *_start_offset
)
1606 unsigned nr
, offset
;
1607 pgoff_t index
, count
;
1608 size_t size
= maxsize
;
1614 pos
= i
->xarray_start
+ i
->iov_offset
;
1615 index
= pos
>> PAGE_SHIFT
;
1616 offset
= pos
& ~PAGE_MASK
;
1617 *_start_offset
= offset
;
1620 if (size
> PAGE_SIZE
- offset
) {
1621 size
-= PAGE_SIZE
- offset
;
1622 count
+= size
>> PAGE_SHIFT
;
1628 p
= get_pages_array(count
);
1633 nr
= iter_xarray_populate_pages(p
, i
->xarray
, index
, count
);
1637 return min_t(size_t, nr
* PAGE_SIZE
- offset
, maxsize
);
1640 ssize_t
iov_iter_get_pages_alloc(struct iov_iter
*i
,
1641 struct page
***pages
, size_t maxsize
,
1648 if (maxsize
> i
->count
)
1653 if (likely(iter_is_iovec(i
))) {
1654 unsigned int gup_flags
= 0;
1657 if (iov_iter_rw(i
) != WRITE
)
1658 gup_flags
|= FOLL_WRITE
;
1660 gup_flags
|= FOLL_NOFAULT
;
1662 addr
= first_iovec_segment(i
, &len
, start
, maxsize
, ~0U);
1663 n
= DIV_ROUND_UP(len
, PAGE_SIZE
);
1664 p
= get_pages_array(n
);
1667 res
= get_user_pages_fast(addr
, n
, gup_flags
, p
);
1668 if (unlikely(res
<= 0)) {
1674 return (res
== n
? len
: res
* PAGE_SIZE
) - *start
;
1676 if (iov_iter_is_bvec(i
)) {
1679 page
= first_bvec_segment(i
, &len
, start
, maxsize
, ~0U);
1680 n
= DIV_ROUND_UP(len
, PAGE_SIZE
);
1681 *pages
= p
= get_pages_array(n
);
1685 get_page(*p
++ = page
++);
1686 return len
- *start
;
1688 if (iov_iter_is_pipe(i
))
1689 return pipe_get_pages_alloc(i
, pages
, maxsize
, start
);
1690 if (iov_iter_is_xarray(i
))
1691 return iter_xarray_get_pages_alloc(i
, pages
, maxsize
, start
);
1694 EXPORT_SYMBOL(iov_iter_get_pages_alloc
);
1696 size_t csum_and_copy_from_iter(void *addr
, size_t bytes
, __wsum
*csum
,
1701 if (unlikely(iov_iter_is_pipe(i
) || iov_iter_is_discard(i
))) {
1705 iterate_and_advance(i
, bytes
, base
, len
, off
, ({
1706 next
= csum_and_copy_from_user(base
, addr
+ off
, len
);
1707 sum
= csum_block_add(sum
, next
, off
);
1710 sum
= csum_and_memcpy(addr
+ off
, base
, len
, sum
, off
);
1716 EXPORT_SYMBOL(csum_and_copy_from_iter
);
1718 size_t csum_and_copy_to_iter(const void *addr
, size_t bytes
, void *_csstate
,
1721 struct csum_state
*csstate
= _csstate
;
1724 if (unlikely(iov_iter_is_discard(i
))) {
1725 WARN_ON(1); /* for now */
1729 sum
= csum_shift(csstate
->csum
, csstate
->off
);
1730 if (unlikely(iov_iter_is_pipe(i
)))
1731 bytes
= csum_and_copy_to_pipe_iter(addr
, bytes
, i
, &sum
);
1732 else iterate_and_advance(i
, bytes
, base
, len
, off
, ({
1733 next
= csum_and_copy_to_user(addr
+ off
, base
, len
);
1734 sum
= csum_block_add(sum
, next
, off
);
1737 sum
= csum_and_memcpy(base
, addr
+ off
, len
, sum
, off
);
1740 csstate
->csum
= csum_shift(sum
, csstate
->off
);
1741 csstate
->off
+= bytes
;
1744 EXPORT_SYMBOL(csum_and_copy_to_iter
);
1746 size_t hash_and_copy_to_iter(const void *addr
, size_t bytes
, void *hashp
,
1749 #ifdef CONFIG_CRYPTO_HASH
1750 struct ahash_request
*hash
= hashp
;
1751 struct scatterlist sg
;
1754 copied
= copy_to_iter(addr
, bytes
, i
);
1755 sg_init_one(&sg
, addr
, copied
);
1756 ahash_request_set_crypt(hash
, &sg
, NULL
, copied
);
1757 crypto_ahash_update(hash
);
1763 EXPORT_SYMBOL(hash_and_copy_to_iter
);
1765 static int iov_npages(const struct iov_iter
*i
, int maxpages
)
1767 size_t skip
= i
->iov_offset
, size
= i
->count
;
1768 const struct iovec
*p
;
1771 for (p
= i
->iov
; size
; skip
= 0, p
++) {
1772 unsigned offs
= offset_in_page(p
->iov_base
+ skip
);
1773 size_t len
= min(p
->iov_len
- skip
, size
);
1777 npages
+= DIV_ROUND_UP(offs
+ len
, PAGE_SIZE
);
1778 if (unlikely(npages
> maxpages
))
1785 static int bvec_npages(const struct iov_iter
*i
, int maxpages
)
1787 size_t skip
= i
->iov_offset
, size
= i
->count
;
1788 const struct bio_vec
*p
;
1791 for (p
= i
->bvec
; size
; skip
= 0, p
++) {
1792 unsigned offs
= (p
->bv_offset
+ skip
) % PAGE_SIZE
;
1793 size_t len
= min(p
->bv_len
- skip
, size
);
1796 npages
+= DIV_ROUND_UP(offs
+ len
, PAGE_SIZE
);
1797 if (unlikely(npages
> maxpages
))
1803 int iov_iter_npages(const struct iov_iter
*i
, int maxpages
)
1805 if (unlikely(!i
->count
))
1807 /* iovec and kvec have identical layouts */
1808 if (likely(iter_is_iovec(i
) || iov_iter_is_kvec(i
)))
1809 return iov_npages(i
, maxpages
);
1810 if (iov_iter_is_bvec(i
))
1811 return bvec_npages(i
, maxpages
);
1812 if (iov_iter_is_pipe(i
)) {
1813 unsigned int iter_head
;
1820 data_start(i
, &iter_head
, &off
);
1821 /* some of this one + all after this one */
1822 npages
= pipe_space_for_user(iter_head
, i
->pipe
->tail
, i
->pipe
);
1823 return min(npages
, maxpages
);
1825 if (iov_iter_is_xarray(i
)) {
1826 unsigned offset
= (i
->xarray_start
+ i
->iov_offset
) % PAGE_SIZE
;
1827 int npages
= DIV_ROUND_UP(offset
+ i
->count
, PAGE_SIZE
);
1828 return min(npages
, maxpages
);
1832 EXPORT_SYMBOL(iov_iter_npages
);
1834 const void *dup_iter(struct iov_iter
*new, struct iov_iter
*old
, gfp_t flags
)
1837 if (unlikely(iov_iter_is_pipe(new))) {
1841 if (unlikely(iov_iter_is_discard(new) || iov_iter_is_xarray(new)))
1843 if (iov_iter_is_bvec(new))
1844 return new->bvec
= kmemdup(new->bvec
,
1845 new->nr_segs
* sizeof(struct bio_vec
),
1848 /* iovec and kvec have identical layout */
1849 return new->iov
= kmemdup(new->iov
,
1850 new->nr_segs
* sizeof(struct iovec
),
1853 EXPORT_SYMBOL(dup_iter
);
1855 static int copy_compat_iovec_from_user(struct iovec
*iov
,
1856 const struct iovec __user
*uvec
, unsigned long nr_segs
)
1858 const struct compat_iovec __user
*uiov
=
1859 (const struct compat_iovec __user
*)uvec
;
1860 int ret
= -EFAULT
, i
;
1862 if (!user_access_begin(uiov
, nr_segs
* sizeof(*uiov
)))
1865 for (i
= 0; i
< nr_segs
; i
++) {
1869 unsafe_get_user(len
, &uiov
[i
].iov_len
, uaccess_end
);
1870 unsafe_get_user(buf
, &uiov
[i
].iov_base
, uaccess_end
);
1872 /* check for compat_size_t not fitting in compat_ssize_t .. */
1877 iov
[i
].iov_base
= compat_ptr(buf
);
1878 iov
[i
].iov_len
= len
;
1887 static int copy_iovec_from_user(struct iovec
*iov
,
1888 const struct iovec __user
*uvec
, unsigned long nr_segs
)
1892 if (copy_from_user(iov
, uvec
, nr_segs
* sizeof(*uvec
)))
1894 for (seg
= 0; seg
< nr_segs
; seg
++) {
1895 if ((ssize_t
)iov
[seg
].iov_len
< 0)
1902 struct iovec
*iovec_from_user(const struct iovec __user
*uvec
,
1903 unsigned long nr_segs
, unsigned long fast_segs
,
1904 struct iovec
*fast_iov
, bool compat
)
1906 struct iovec
*iov
= fast_iov
;
1910 * SuS says "The readv() function *may* fail if the iovcnt argument was
1911 * less than or equal to 0, or greater than {IOV_MAX}. Linux has
1912 * traditionally returned zero for zero segments, so...
1916 if (nr_segs
> UIO_MAXIOV
)
1917 return ERR_PTR(-EINVAL
);
1918 if (nr_segs
> fast_segs
) {
1919 iov
= kmalloc_array(nr_segs
, sizeof(struct iovec
), GFP_KERNEL
);
1921 return ERR_PTR(-ENOMEM
);
1925 ret
= copy_compat_iovec_from_user(iov
, uvec
, nr_segs
);
1927 ret
= copy_iovec_from_user(iov
, uvec
, nr_segs
);
1929 if (iov
!= fast_iov
)
1931 return ERR_PTR(ret
);
1937 ssize_t
__import_iovec(int type
, const struct iovec __user
*uvec
,
1938 unsigned nr_segs
, unsigned fast_segs
, struct iovec
**iovp
,
1939 struct iov_iter
*i
, bool compat
)
1941 ssize_t total_len
= 0;
1945 iov
= iovec_from_user(uvec
, nr_segs
, fast_segs
, *iovp
, compat
);
1948 return PTR_ERR(iov
);
1952 * According to the Single Unix Specification we should return EINVAL if
1953 * an element length is < 0 when cast to ssize_t or if the total length
1954 * would overflow the ssize_t return value of the system call.
1956 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1959 for (seg
= 0; seg
< nr_segs
; seg
++) {
1960 ssize_t len
= (ssize_t
)iov
[seg
].iov_len
;
1962 if (!access_ok(iov
[seg
].iov_base
, len
)) {
1969 if (len
> MAX_RW_COUNT
- total_len
) {
1970 len
= MAX_RW_COUNT
- total_len
;
1971 iov
[seg
].iov_len
= len
;
1976 iov_iter_init(i
, type
, iov
, nr_segs
, total_len
);
1985 * import_iovec() - Copy an array of &struct iovec from userspace
1986 * into the kernel, check that it is valid, and initialize a new
1987 * &struct iov_iter iterator to access it.
1989 * @type: One of %READ or %WRITE.
1990 * @uvec: Pointer to the userspace array.
1991 * @nr_segs: Number of elements in userspace array.
1992 * @fast_segs: Number of elements in @iov.
1993 * @iovp: (input and output parameter) Pointer to pointer to (usually small
1994 * on-stack) kernel array.
1995 * @i: Pointer to iterator that will be initialized on success.
1997 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1998 * then this function places %NULL in *@iov on return. Otherwise, a new
1999 * array will be allocated and the result placed in *@iov. This means that
2000 * the caller may call kfree() on *@iov regardless of whether the small
2001 * on-stack array was used or not (and regardless of whether this function
2002 * returns an error or not).
2004 * Return: Negative error code on error, bytes imported on success
2006 ssize_t
import_iovec(int type
, const struct iovec __user
*uvec
,
2007 unsigned nr_segs
, unsigned fast_segs
,
2008 struct iovec
**iovp
, struct iov_iter
*i
)
2010 return __import_iovec(type
, uvec
, nr_segs
, fast_segs
, iovp
, i
,
2011 in_compat_syscall());
2013 EXPORT_SYMBOL(import_iovec
);
2015 int import_single_range(int rw
, void __user
*buf
, size_t len
,
2016 struct iovec
*iov
, struct iov_iter
*i
)
2018 if (len
> MAX_RW_COUNT
)
2020 if (unlikely(!access_ok(buf
, len
)))
2023 iov
->iov_base
= buf
;
2025 iov_iter_init(i
, rw
, iov
, 1, len
);
2028 EXPORT_SYMBOL(import_single_range
);
2031 * iov_iter_restore() - Restore a &struct iov_iter to the same state as when
2032 * iov_iter_save_state() was called.
2034 * @i: &struct iov_iter to restore
2035 * @state: state to restore from
2037 * Used after iov_iter_save_state() to bring restore @i, if operations may
2040 * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC
2042 void iov_iter_restore(struct iov_iter
*i
, struct iov_iter_state
*state
)
2044 if (WARN_ON_ONCE(!iov_iter_is_bvec(i
) && !iter_is_iovec(i
)) &&
2045 !iov_iter_is_kvec(i
))
2047 i
->iov_offset
= state
->iov_offset
;
2048 i
->count
= state
->count
;
2050 * For the *vec iters, nr_segs + iov is constant - if we increment
2051 * the vec, then we also decrement the nr_segs count. Hence we don't
2052 * need to track both of these, just one is enough and we can deduct
2053 * the other from that. ITER_KVEC and ITER_IOVEC are the same struct
2054 * size, so we can just increment the iov pointer as they are unionzed.
2055 * ITER_BVEC _may_ be the same size on some archs, but on others it is
2056 * not. Be safe and handle it separately.
2058 BUILD_BUG_ON(sizeof(struct iovec
) != sizeof(struct kvec
));
2059 if (iov_iter_is_bvec(i
))
2060 i
->bvec
-= state
->nr_segs
- i
->nr_segs
;
2062 i
->iov
-= state
->nr_segs
- i
->nr_segs
;
2063 i
->nr_segs
= state
->nr_segs
;