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 folio *folio; \
73 loff_t start = i->xarray_start + i->iov_offset; \
74 pgoff_t index = start / PAGE_SIZE; \
75 XA_STATE(xas, i->xarray, index); \
77 len = PAGE_SIZE - offset_in_page(start); \
79 xas_for_each(&xas, folio, ULONG_MAX) { \
82 if (xas_retry(&xas, folio)) \
84 if (WARN_ON(xa_is_value(folio))) \
86 if (WARN_ON(folio_test_hugetlb(folio))) \
88 offset = offset_in_folio(folio, start + __off); \
89 while (offset < folio_size(folio)) { \
90 base = kmap_local_folio(folio, offset); \
105 i->iov_offset += __off; \
109 #define __iterate_and_advance(i, n, base, len, off, I, K) { \
110 if (unlikely(i->count < n)) \
113 if (likely(iter_is_iovec(i))) { \
114 const struct iovec *iov = i->iov; \
117 iterate_iovec(i, n, base, len, off, \
119 i->nr_segs -= iov - i->iov; \
121 } else if (iov_iter_is_bvec(i)) { \
122 const struct bio_vec *bvec = i->bvec; \
125 iterate_bvec(i, n, base, len, off, \
127 i->nr_segs -= bvec - i->bvec; \
129 } else if (iov_iter_is_kvec(i)) { \
130 const struct kvec *kvec = i->kvec; \
133 iterate_iovec(i, n, base, len, off, \
135 i->nr_segs -= kvec - i->kvec; \
137 } else if (iov_iter_is_xarray(i)) { \
140 iterate_xarray(i, n, base, len, off, \
146 #define iterate_and_advance(i, n, base, len, off, I, K) \
147 __iterate_and_advance(i, n, base, len, off, I, ((void)(K),0))
149 static int copyout(void __user *to, const void *from, size_t n)
151 if (should_fail_usercopy())
153 if (access_ok(to, n)) {
154 instrument_copy_to_user(to, from, n);
155 n = raw_copy_to_user(to, from, n);
160 static int copyin(void *to, const void __user *from, size_t n)
162 if (should_fail_usercopy())
164 if (access_ok(from, n)) {
165 instrument_copy_from_user(to, from, n);
166 n = raw_copy_from_user(to, from, n);
171 static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes,
174 size_t skip, copy, left, wanted;
175 const struct iovec *iov;
179 if (unlikely(bytes > i->count))
182 if (unlikely(!bytes))
188 skip = i->iov_offset;
189 buf = iov->iov_base + skip;
190 copy = min(bytes, iov->iov_len - skip);
192 if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_writeable(buf, copy)) {
193 kaddr = kmap_atomic(page);
194 from = kaddr + offset;
196 /* first chunk, usually the only one */
197 left = copyout(buf, from, copy);
203 while (unlikely(!left && bytes)) {
206 copy = min(bytes, iov->iov_len);
207 left = copyout(buf, from, copy);
213 if (likely(!bytes)) {
214 kunmap_atomic(kaddr);
217 offset = from - kaddr;
219 kunmap_atomic(kaddr);
220 copy = min(bytes, iov->iov_len - skip);
222 /* Too bad - revert to non-atomic kmap */
225 from = kaddr + offset;
226 left = copyout(buf, from, copy);
231 while (unlikely(!left && bytes)) {
234 copy = min(bytes, iov->iov_len);
235 left = copyout(buf, from, copy);
244 if (skip == iov->iov_len) {
248 i->count -= wanted - bytes;
249 i->nr_segs -= iov - i->iov;
251 i->iov_offset = skip;
252 return wanted - bytes;
255 static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes,
258 size_t skip, copy, left, wanted;
259 const struct iovec *iov;
263 if (unlikely(bytes > i->count))
266 if (unlikely(!bytes))
272 skip = i->iov_offset;
273 buf = iov->iov_base + skip;
274 copy = min(bytes, iov->iov_len - skip);
276 if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_readable(buf, copy)) {
277 kaddr = kmap_atomic(page);
280 /* first chunk, usually the only one */
281 left = copyin(to, buf, copy);
287 while (unlikely(!left && bytes)) {
290 copy = min(bytes, iov->iov_len);
291 left = copyin(to, buf, copy);
297 if (likely(!bytes)) {
298 kunmap_atomic(kaddr);
303 kunmap_atomic(kaddr);
304 copy = min(bytes, iov->iov_len - skip);
306 /* Too bad - revert to non-atomic kmap */
310 left = copyin(to, buf, copy);
315 while (unlikely(!left && bytes)) {
318 copy = min(bytes, iov->iov_len);
319 left = copyin(to, buf, copy);
328 if (skip == iov->iov_len) {
332 i->count -= wanted - bytes;
333 i->nr_segs -= iov - i->iov;
335 i->iov_offset = skip;
336 return wanted - bytes;
340 static bool sanity(const struct iov_iter *i)
342 struct pipe_inode_info *pipe = i->pipe;
343 unsigned int p_head = pipe->head;
344 unsigned int p_tail = pipe->tail;
345 unsigned int p_mask = pipe->ring_size - 1;
346 unsigned int p_occupancy = pipe_occupancy(p_head, p_tail);
347 unsigned int i_head = i->head;
351 struct pipe_buffer *p;
352 if (unlikely(p_occupancy == 0))
353 goto Bad; // pipe must be non-empty
354 if (unlikely(i_head != p_head - 1))
355 goto Bad; // must be at the last buffer...
357 p = &pipe->bufs[i_head & p_mask];
358 if (unlikely(p->offset + p->len != i->iov_offset))
359 goto Bad; // ... at the end of segment
361 if (i_head != p_head)
362 goto Bad; // must be right after the last buffer
366 printk(KERN_ERR "idx = %d, offset = %zd\n", i_head, i->iov_offset);
367 printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n",
368 p_head, p_tail, pipe->ring_size);
369 for (idx = 0; idx < pipe->ring_size; idx++)
370 printk(KERN_ERR "[%p %p %d %d]\n",
372 pipe->bufs[idx].page,
373 pipe->bufs[idx].offset,
374 pipe->bufs[idx].len);
379 #define sanity(i) true
382 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
385 struct pipe_inode_info *pipe = i->pipe;
386 struct pipe_buffer *buf;
387 unsigned int p_tail = pipe->tail;
388 unsigned int p_mask = pipe->ring_size - 1;
389 unsigned int i_head = i->head;
392 if (unlikely(bytes > i->count))
395 if (unlikely(!bytes))
402 buf = &pipe->bufs[i_head & p_mask];
404 if (offset == off && buf->page == page) {
405 /* merge with the last one */
407 i->iov_offset += bytes;
411 buf = &pipe->bufs[i_head & p_mask];
413 if (pipe_full(i_head, p_tail, pipe->max_usage))
416 buf->ops = &page_cache_pipe_buf_ops;
420 buf->offset = offset;
423 pipe->head = i_head + 1;
424 i->iov_offset = offset + bytes;
432 * fault_in_iov_iter_readable - fault in iov iterator for reading
434 * @size: maximum length
436 * Fault in one or more iovecs of the given iov_iter, to a maximum length of
437 * @size. For each iovec, fault in each page that constitutes the iovec.
439 * Returns the number of bytes not faulted in (like copy_to_user() and
442 * Always returns 0 for non-userspace iterators.
444 size_t fault_in_iov_iter_readable(const struct iov_iter *i, size_t size)
446 if (iter_is_iovec(i)) {
447 size_t count = min(size, iov_iter_count(i));
448 const struct iovec *p;
452 for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) {
453 size_t len = min(count, p->iov_len - skip);
458 ret = fault_in_readable(p->iov_base + skip, len);
467 EXPORT_SYMBOL(fault_in_iov_iter_readable);
470 * fault_in_iov_iter_writeable - fault in iov iterator for writing
472 * @size: maximum length
474 * Faults in the iterator using get_user_pages(), i.e., without triggering
475 * hardware page faults. This is primarily useful when we already know that
476 * some or all of the pages in @i aren't in memory.
478 * Returns the number of bytes not faulted in, like copy_to_user() and
481 * Always returns 0 for non-user-space iterators.
483 size_t fault_in_iov_iter_writeable(const struct iov_iter *i, size_t size)
485 if (iter_is_iovec(i)) {
486 size_t count = min(size, iov_iter_count(i));
487 const struct iovec *p;
491 for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) {
492 size_t len = min(count, p->iov_len - skip);
497 ret = fault_in_safe_writeable(p->iov_base + skip, len);
506 EXPORT_SYMBOL(fault_in_iov_iter_writeable);
508 void iov_iter_init(struct iov_iter *i, unsigned int direction,
509 const struct iovec *iov, unsigned long nr_segs,
512 WARN_ON(direction & ~(READ | WRITE));
513 *i = (struct iov_iter) {
514 .iter_type = ITER_IOVEC,
516 .data_source = direction,
523 EXPORT_SYMBOL(iov_iter_init);
525 static inline bool allocated(struct pipe_buffer *buf)
527 return buf->ops == &default_pipe_buf_ops;
530 static inline void data_start(const struct iov_iter *i,
531 unsigned int *iter_headp, size_t *offp)
533 unsigned int p_mask = i->pipe->ring_size - 1;
534 unsigned int iter_head = i->head;
535 size_t off = i->iov_offset;
537 if (off && (!allocated(&i->pipe->bufs[iter_head & p_mask]) ||
542 *iter_headp = iter_head;
546 static size_t push_pipe(struct iov_iter *i, size_t size,
547 int *iter_headp, size_t *offp)
549 struct pipe_inode_info *pipe = i->pipe;
550 unsigned int p_tail = pipe->tail;
551 unsigned int p_mask = pipe->ring_size - 1;
552 unsigned int iter_head;
556 if (unlikely(size > i->count))
562 data_start(i, &iter_head, &off);
563 *iter_headp = iter_head;
566 left -= PAGE_SIZE - off;
568 pipe->bufs[iter_head & p_mask].len += size;
571 pipe->bufs[iter_head & p_mask].len = PAGE_SIZE;
574 while (!pipe_full(iter_head, p_tail, pipe->max_usage)) {
575 struct pipe_buffer *buf = &pipe->bufs[iter_head & p_mask];
576 struct page *page = alloc_page(GFP_USER);
580 buf->ops = &default_pipe_buf_ops;
584 buf->len = min_t(ssize_t, left, PAGE_SIZE);
587 pipe->head = iter_head;
595 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
598 struct pipe_inode_info *pipe = i->pipe;
599 unsigned int p_mask = pipe->ring_size - 1;
606 bytes = n = push_pipe(i, bytes, &i_head, &off);
610 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
611 memcpy_to_page(pipe->bufs[i_head & p_mask].page, off, addr, chunk);
613 i->iov_offset = off + chunk;
623 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
624 __wsum sum, size_t off)
626 __wsum next = csum_partial_copy_nocheck(from, to, len);
627 return csum_block_add(sum, next, off);
630 static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
631 struct iov_iter *i, __wsum *sump)
633 struct pipe_inode_info *pipe = i->pipe;
634 unsigned int p_mask = pipe->ring_size - 1;
643 bytes = push_pipe(i, bytes, &i_head, &r);
645 size_t chunk = min_t(size_t, bytes, PAGE_SIZE - r);
646 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
647 sum = csum_and_memcpy(p + r, addr + off, chunk, sum, off);
650 i->iov_offset = r + chunk;
661 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
663 if (unlikely(iov_iter_is_pipe(i)))
664 return copy_pipe_to_iter(addr, bytes, i);
665 if (iter_is_iovec(i))
667 iterate_and_advance(i, bytes, base, len, off,
668 copyout(base, addr + off, len),
669 memcpy(base, addr + off, len)
674 EXPORT_SYMBOL(_copy_to_iter);
676 #ifdef CONFIG_ARCH_HAS_COPY_MC
677 static int copyout_mc(void __user *to, const void *from, size_t n)
679 if (access_ok(to, n)) {
680 instrument_copy_to_user(to, from, n);
681 n = copy_mc_to_user((__force void *) to, from, n);
686 static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes,
689 struct pipe_inode_info *pipe = i->pipe;
690 unsigned int p_mask = pipe->ring_size - 1;
692 size_t n, off, xfer = 0;
697 n = push_pipe(i, bytes, &i_head, &off);
699 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
700 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
702 rem = copy_mc_to_kernel(p + off, addr + xfer, chunk);
706 i->iov_offset = off + chunk;
719 * _copy_mc_to_iter - copy to iter with source memory error exception handling
720 * @addr: source kernel address
721 * @bytes: total transfer length
722 * @i: destination iterator
724 * The pmem driver deploys this for the dax operation
725 * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
726 * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
727 * successfully copied.
729 * The main differences between this and typical _copy_to_iter().
731 * * Typical tail/residue handling after a fault retries the copy
732 * byte-by-byte until the fault happens again. Re-triggering machine
733 * checks is potentially fatal so the implementation uses source
734 * alignment and poison alignment assumptions to avoid re-triggering
735 * hardware exceptions.
737 * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
738 * Compare to copy_to_iter() where only ITER_IOVEC attempts might return
741 * Return: number of bytes copied (may be %0)
743 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
745 if (unlikely(iov_iter_is_pipe(i)))
746 return copy_mc_pipe_to_iter(addr, bytes, i);
747 if (iter_is_iovec(i))
749 __iterate_and_advance(i, bytes, base, len, off,
750 copyout_mc(base, addr + off, len),
751 copy_mc_to_kernel(base, addr + off, len)
756 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
757 #endif /* CONFIG_ARCH_HAS_COPY_MC */
759 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
761 if (unlikely(iov_iter_is_pipe(i))) {
765 if (iter_is_iovec(i))
767 iterate_and_advance(i, bytes, base, len, off,
768 copyin(addr + off, base, len),
769 memcpy(addr + off, base, len)
774 EXPORT_SYMBOL(_copy_from_iter);
776 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
778 if (unlikely(iov_iter_is_pipe(i))) {
782 iterate_and_advance(i, bytes, base, len, off,
783 __copy_from_user_inatomic_nocache(addr + off, base, len),
784 memcpy(addr + off, base, len)
789 EXPORT_SYMBOL(_copy_from_iter_nocache);
791 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
793 * _copy_from_iter_flushcache - write destination through cpu cache
794 * @addr: destination kernel address
795 * @bytes: total transfer length
796 * @i: source iterator
798 * The pmem driver arranges for filesystem-dax to use this facility via
799 * dax_copy_from_iter() for ensuring that writes to persistent memory
800 * are flushed through the CPU cache. It is differentiated from
801 * _copy_from_iter_nocache() in that guarantees all data is flushed for
802 * all iterator types. The _copy_from_iter_nocache() only attempts to
803 * bypass the cache for the ITER_IOVEC case, and on some archs may use
804 * instructions that strand dirty-data in the cache.
806 * Return: number of bytes copied (may be %0)
808 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
810 if (unlikely(iov_iter_is_pipe(i))) {
814 iterate_and_advance(i, bytes, base, len, off,
815 __copy_from_user_flushcache(addr + off, base, len),
816 memcpy_flushcache(addr + off, base, len)
821 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
824 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
827 size_t v = n + offset;
830 * The general case needs to access the page order in order
831 * to compute the page size.
832 * However, we mostly deal with order-0 pages and thus can
833 * avoid a possible cache line miss for requests that fit all
836 if (n <= v && v <= PAGE_SIZE)
839 head = compound_head(page);
840 v += (page - head) << PAGE_SHIFT;
842 if (likely(n <= v && v <= (page_size(head))))
848 static size_t __copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
851 if (likely(iter_is_iovec(i)))
852 return copy_page_to_iter_iovec(page, offset, bytes, i);
853 if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
854 void *kaddr = kmap_local_page(page);
855 size_t wanted = _copy_to_iter(kaddr + offset, bytes, i);
859 if (iov_iter_is_pipe(i))
860 return copy_page_to_iter_pipe(page, offset, bytes, i);
861 if (unlikely(iov_iter_is_discard(i))) {
862 if (unlikely(i->count < bytes))
871 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
875 if (unlikely(!page_copy_sane(page, offset, bytes)))
877 page += offset / PAGE_SIZE; // first subpage
880 size_t n = __copy_page_to_iter(page, offset,
881 min(bytes, (size_t)PAGE_SIZE - offset), i);
887 if (offset == PAGE_SIZE) {
894 EXPORT_SYMBOL(copy_page_to_iter);
896 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
899 if (unlikely(!page_copy_sane(page, offset, bytes)))
901 if (likely(iter_is_iovec(i)))
902 return copy_page_from_iter_iovec(page, offset, bytes, i);
903 if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
904 void *kaddr = kmap_local_page(page);
905 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
912 EXPORT_SYMBOL(copy_page_from_iter);
914 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
916 struct pipe_inode_info *pipe = i->pipe;
917 unsigned int p_mask = pipe->ring_size - 1;
924 bytes = n = push_pipe(i, bytes, &i_head, &off);
929 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
930 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
931 memset(p + off, 0, chunk);
934 i->iov_offset = off + chunk;
943 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
945 if (unlikely(iov_iter_is_pipe(i)))
946 return pipe_zero(bytes, i);
947 iterate_and_advance(i, bytes, base, len, count,
948 clear_user(base, len),
954 EXPORT_SYMBOL(iov_iter_zero);
956 size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes,
959 char *kaddr = kmap_atomic(page), *p = kaddr + offset;
960 if (unlikely(!page_copy_sane(page, offset, bytes))) {
961 kunmap_atomic(kaddr);
964 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
965 kunmap_atomic(kaddr);
969 iterate_and_advance(i, bytes, base, len, off,
970 copyin(p + off, base, len),
971 memcpy(p + off, base, len)
973 kunmap_atomic(kaddr);
976 EXPORT_SYMBOL(copy_page_from_iter_atomic);
978 static inline void pipe_truncate(struct iov_iter *i)
980 struct pipe_inode_info *pipe = i->pipe;
981 unsigned int p_tail = pipe->tail;
982 unsigned int p_head = pipe->head;
983 unsigned int p_mask = pipe->ring_size - 1;
985 if (!pipe_empty(p_head, p_tail)) {
986 struct pipe_buffer *buf;
987 unsigned int i_head = i->head;
988 size_t off = i->iov_offset;
991 buf = &pipe->bufs[i_head & p_mask];
992 buf->len = off - buf->offset;
995 while (p_head != i_head) {
997 pipe_buf_release(pipe, &pipe->bufs[p_head & p_mask]);
1000 pipe->head = p_head;
1004 static void pipe_advance(struct iov_iter *i, size_t size)
1006 struct pipe_inode_info *pipe = i->pipe;
1008 struct pipe_buffer *buf;
1009 unsigned int p_mask = pipe->ring_size - 1;
1010 unsigned int i_head = i->head;
1011 size_t off = i->iov_offset, left = size;
1013 if (off) /* make it relative to the beginning of buffer */
1014 left += off - pipe->bufs[i_head & p_mask].offset;
1016 buf = &pipe->bufs[i_head & p_mask];
1017 if (left <= buf->len)
1023 i->iov_offset = buf->offset + left;
1026 /* ... and discard everything past that point */
1030 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
1032 struct bvec_iter bi;
1034 bi.bi_size = i->count;
1035 bi.bi_bvec_done = i->iov_offset;
1037 bvec_iter_advance(i->bvec, &bi, size);
1039 i->bvec += bi.bi_idx;
1040 i->nr_segs -= bi.bi_idx;
1041 i->count = bi.bi_size;
1042 i->iov_offset = bi.bi_bvec_done;
1045 static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
1047 const struct iovec *iov, *end;
1053 size += i->iov_offset; // from beginning of current segment
1054 for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) {
1055 if (likely(size < iov->iov_len))
1057 size -= iov->iov_len;
1059 i->iov_offset = size;
1060 i->nr_segs -= iov - i->iov;
1064 void iov_iter_advance(struct iov_iter *i, size_t size)
1066 if (unlikely(i->count < size))
1068 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
1069 /* iovec and kvec have identical layouts */
1070 iov_iter_iovec_advance(i, size);
1071 } else if (iov_iter_is_bvec(i)) {
1072 iov_iter_bvec_advance(i, size);
1073 } else if (iov_iter_is_pipe(i)) {
1074 pipe_advance(i, size);
1075 } else if (unlikely(iov_iter_is_xarray(i))) {
1076 i->iov_offset += size;
1078 } else if (iov_iter_is_discard(i)) {
1082 EXPORT_SYMBOL(iov_iter_advance);
1084 void iov_iter_revert(struct iov_iter *i, size_t unroll)
1088 if (WARN_ON(unroll > MAX_RW_COUNT))
1091 if (unlikely(iov_iter_is_pipe(i))) {
1092 struct pipe_inode_info *pipe = i->pipe;
1093 unsigned int p_mask = pipe->ring_size - 1;
1094 unsigned int i_head = i->head;
1095 size_t off = i->iov_offset;
1097 struct pipe_buffer *b = &pipe->bufs[i_head & p_mask];
1098 size_t n = off - b->offset;
1104 if (!unroll && i_head == i->start_head) {
1109 b = &pipe->bufs[i_head & p_mask];
1110 off = b->offset + b->len;
1112 i->iov_offset = off;
1117 if (unlikely(iov_iter_is_discard(i)))
1119 if (unroll <= i->iov_offset) {
1120 i->iov_offset -= unroll;
1123 unroll -= i->iov_offset;
1124 if (iov_iter_is_xarray(i)) {
1125 BUG(); /* We should never go beyond the start of the specified
1126 * range since we might then be straying into pages that
1129 } else if (iov_iter_is_bvec(i)) {
1130 const struct bio_vec *bvec = i->bvec;
1132 size_t n = (--bvec)->bv_len;
1136 i->iov_offset = n - unroll;
1141 } else { /* same logics for iovec and kvec */
1142 const struct iovec *iov = i->iov;
1144 size_t n = (--iov)->iov_len;
1148 i->iov_offset = n - unroll;
1155 EXPORT_SYMBOL(iov_iter_revert);
1158 * Return the count of just the current iov_iter segment.
1160 size_t iov_iter_single_seg_count(const struct iov_iter *i)
1162 if (i->nr_segs > 1) {
1163 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1164 return min(i->count, i->iov->iov_len - i->iov_offset);
1165 if (iov_iter_is_bvec(i))
1166 return min(i->count, i->bvec->bv_len - i->iov_offset);
1170 EXPORT_SYMBOL(iov_iter_single_seg_count);
1172 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
1173 const struct kvec *kvec, unsigned long nr_segs,
1176 WARN_ON(direction & ~(READ | WRITE));
1177 *i = (struct iov_iter){
1178 .iter_type = ITER_KVEC,
1179 .data_source = direction,
1186 EXPORT_SYMBOL(iov_iter_kvec);
1188 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1189 const struct bio_vec *bvec, unsigned long nr_segs,
1192 WARN_ON(direction & ~(READ | WRITE));
1193 *i = (struct iov_iter){
1194 .iter_type = ITER_BVEC,
1195 .data_source = direction,
1202 EXPORT_SYMBOL(iov_iter_bvec);
1204 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1205 struct pipe_inode_info *pipe,
1208 BUG_ON(direction != READ);
1209 WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1210 *i = (struct iov_iter){
1211 .iter_type = ITER_PIPE,
1212 .data_source = false,
1215 .start_head = pipe->head,
1220 EXPORT_SYMBOL(iov_iter_pipe);
1223 * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1224 * @i: The iterator to initialise.
1225 * @direction: The direction of the transfer.
1226 * @xarray: The xarray to access.
1227 * @start: The start file position.
1228 * @count: The size of the I/O buffer in bytes.
1230 * Set up an I/O iterator to either draw data out of the pages attached to an
1231 * inode or to inject data into those pages. The pages *must* be prevented
1232 * from evaporation, either by taking a ref on them or locking them by the
1235 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
1236 struct xarray *xarray, loff_t start, size_t count)
1238 BUG_ON(direction & ~1);
1239 *i = (struct iov_iter) {
1240 .iter_type = ITER_XARRAY,
1241 .data_source = direction,
1243 .xarray_start = start,
1248 EXPORT_SYMBOL(iov_iter_xarray);
1251 * iov_iter_discard - Initialise an I/O iterator that discards data
1252 * @i: The iterator to initialise.
1253 * @direction: The direction of the transfer.
1254 * @count: The size of the I/O buffer in bytes.
1256 * Set up an I/O iterator that just discards everything that's written to it.
1257 * It's only available as a READ iterator.
1259 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1261 BUG_ON(direction != READ);
1262 *i = (struct iov_iter){
1263 .iter_type = ITER_DISCARD,
1264 .data_source = false,
1269 EXPORT_SYMBOL(iov_iter_discard);
1271 static bool iov_iter_aligned_iovec(const struct iov_iter *i, unsigned addr_mask,
1274 size_t size = i->count;
1275 size_t skip = i->iov_offset;
1278 for (k = 0; k < i->nr_segs; k++, skip = 0) {
1279 size_t len = i->iov[k].iov_len - skip;
1285 if ((unsigned long)(i->iov[k].iov_base + skip) & addr_mask)
1295 static bool iov_iter_aligned_bvec(const struct iov_iter *i, unsigned addr_mask,
1298 size_t size = i->count;
1299 unsigned skip = i->iov_offset;
1302 for (k = 0; k < i->nr_segs; k++, skip = 0) {
1303 size_t len = i->bvec[k].bv_len - skip;
1309 if ((unsigned long)(i->bvec[k].bv_offset + skip) & addr_mask)
1320 * iov_iter_is_aligned() - Check if the addresses and lengths of each segments
1321 * are aligned to the parameters.
1323 * @i: &struct iov_iter to restore
1324 * @addr_mask: bit mask to check against the iov element's addresses
1325 * @len_mask: bit mask to check against the iov element's lengths
1327 * Return: false if any addresses or lengths intersect with the provided masks
1329 bool iov_iter_is_aligned(const struct iov_iter *i, unsigned addr_mask,
1332 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1333 return iov_iter_aligned_iovec(i, addr_mask, len_mask);
1335 if (iov_iter_is_bvec(i))
1336 return iov_iter_aligned_bvec(i, addr_mask, len_mask);
1338 if (iov_iter_is_pipe(i)) {
1339 unsigned int p_mask = i->pipe->ring_size - 1;
1340 size_t size = i->count;
1342 if (size & len_mask)
1344 if (size && allocated(&i->pipe->bufs[i->head & p_mask])) {
1345 if (i->iov_offset & addr_mask)
1352 if (iov_iter_is_xarray(i)) {
1353 if (i->count & len_mask)
1355 if ((i->xarray_start + i->iov_offset) & addr_mask)
1361 EXPORT_SYMBOL_GPL(iov_iter_is_aligned);
1363 static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
1365 unsigned long res = 0;
1366 size_t size = i->count;
1367 size_t skip = i->iov_offset;
1370 for (k = 0; k < i->nr_segs; k++, skip = 0) {
1371 size_t len = i->iov[k].iov_len - skip;
1373 res |= (unsigned long)i->iov[k].iov_base + skip;
1385 static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
1388 size_t size = i->count;
1389 unsigned skip = i->iov_offset;
1392 for (k = 0; k < i->nr_segs; k++, skip = 0) {
1393 size_t len = i->bvec[k].bv_len - skip;
1394 res |= (unsigned long)i->bvec[k].bv_offset + skip;
1405 unsigned long iov_iter_alignment(const struct iov_iter *i)
1407 /* iovec and kvec have identical layouts */
1408 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1409 return iov_iter_alignment_iovec(i);
1411 if (iov_iter_is_bvec(i))
1412 return iov_iter_alignment_bvec(i);
1414 if (iov_iter_is_pipe(i)) {
1415 unsigned int p_mask = i->pipe->ring_size - 1;
1416 size_t size = i->count;
1418 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->head & p_mask]))
1419 return size | i->iov_offset;
1423 if (iov_iter_is_xarray(i))
1424 return (i->xarray_start + i->iov_offset) | i->count;
1428 EXPORT_SYMBOL(iov_iter_alignment);
1430 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1432 unsigned long res = 0;
1433 unsigned long v = 0;
1434 size_t size = i->count;
1437 if (WARN_ON(!iter_is_iovec(i)))
1440 for (k = 0; k < i->nr_segs; k++) {
1441 if (i->iov[k].iov_len) {
1442 unsigned long base = (unsigned long)i->iov[k].iov_base;
1443 if (v) // if not the first one
1444 res |= base | v; // this start | previous end
1445 v = base + i->iov[k].iov_len;
1446 if (size <= i->iov[k].iov_len)
1448 size -= i->iov[k].iov_len;
1453 EXPORT_SYMBOL(iov_iter_gap_alignment);
1455 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1457 struct page **pages,
1461 struct pipe_inode_info *pipe = i->pipe;
1462 unsigned int p_mask = pipe->ring_size - 1;
1463 ssize_t n = push_pipe(i, maxsize, &iter_head, start);
1470 get_page(*pages++ = pipe->bufs[iter_head & p_mask].page);
1478 static ssize_t pipe_get_pages(struct iov_iter *i,
1479 struct page **pages, size_t maxsize, unsigned maxpages,
1482 unsigned int iter_head, npages;
1488 data_start(i, &iter_head, start);
1489 /* Amount of free space: some of this one + all after this one */
1490 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1491 capacity = min(npages, maxpages) * PAGE_SIZE - *start;
1493 return __pipe_get_pages(i, min(maxsize, capacity), pages, iter_head, start);
1496 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1497 pgoff_t index, unsigned int nr_pages)
1499 XA_STATE(xas, xa, index);
1501 unsigned int ret = 0;
1504 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1505 if (xas_retry(&xas, page))
1508 /* Has the page moved or been split? */
1509 if (unlikely(page != xas_reload(&xas))) {
1514 pages[ret] = find_subpage(page, xas.xa_index);
1515 get_page(pages[ret]);
1516 if (++ret == nr_pages)
1523 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1524 struct page **pages, size_t maxsize,
1525 unsigned maxpages, size_t *_start_offset)
1527 unsigned nr, offset;
1528 pgoff_t index, count;
1529 size_t size = maxsize;
1532 if (!size || !maxpages)
1535 pos = i->xarray_start + i->iov_offset;
1536 index = pos >> PAGE_SHIFT;
1537 offset = pos & ~PAGE_MASK;
1538 *_start_offset = offset;
1541 if (size > PAGE_SIZE - offset) {
1542 size -= PAGE_SIZE - offset;
1543 count += size >> PAGE_SHIFT;
1549 if (count > maxpages)
1552 nr = iter_xarray_populate_pages(pages, i->xarray, index, count);
1556 return min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1559 /* must be done on non-empty ITER_IOVEC one */
1560 static unsigned long first_iovec_segment(const struct iov_iter *i,
1561 size_t *size, size_t *start,
1562 size_t maxsize, unsigned maxpages)
1567 for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
1568 unsigned long addr = (unsigned long)i->iov[k].iov_base + skip;
1569 size_t len = i->iov[k].iov_len - skip;
1575 len += (*start = addr % PAGE_SIZE);
1576 if (len > maxpages * PAGE_SIZE)
1577 len = maxpages * PAGE_SIZE;
1579 return addr & PAGE_MASK;
1581 BUG(); // if it had been empty, we wouldn't get called
1584 /* must be done on non-empty ITER_BVEC one */
1585 static struct page *first_bvec_segment(const struct iov_iter *i,
1586 size_t *size, size_t *start,
1587 size_t maxsize, unsigned maxpages)
1590 size_t skip = i->iov_offset, len;
1592 len = i->bvec->bv_len - skip;
1595 skip += i->bvec->bv_offset;
1596 page = i->bvec->bv_page + skip / PAGE_SIZE;
1597 len += (*start = skip % PAGE_SIZE);
1598 if (len > maxpages * PAGE_SIZE)
1599 len = maxpages * PAGE_SIZE;
1604 ssize_t iov_iter_get_pages(struct iov_iter *i,
1605 struct page **pages, size_t maxsize, unsigned maxpages,
1611 if (maxsize > i->count)
1616 if (likely(iter_is_iovec(i))) {
1617 unsigned int gup_flags = 0;
1620 if (iov_iter_rw(i) != WRITE)
1621 gup_flags |= FOLL_WRITE;
1623 gup_flags |= FOLL_NOFAULT;
1625 addr = first_iovec_segment(i, &len, start, maxsize, maxpages);
1626 n = DIV_ROUND_UP(len, PAGE_SIZE);
1627 res = get_user_pages_fast(addr, n, gup_flags, pages);
1628 if (unlikely(res <= 0))
1630 return (res == n ? len : res * PAGE_SIZE) - *start;
1632 if (iov_iter_is_bvec(i)) {
1635 page = first_bvec_segment(i, &len, start, maxsize, maxpages);
1636 n = DIV_ROUND_UP(len, PAGE_SIZE);
1638 get_page(*pages++ = page++);
1639 return len - *start;
1641 if (iov_iter_is_pipe(i))
1642 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1643 if (iov_iter_is_xarray(i))
1644 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1647 EXPORT_SYMBOL(iov_iter_get_pages);
1649 static struct page **get_pages_array(size_t n)
1651 return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1654 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1655 struct page ***pages, size_t maxsize,
1659 unsigned int iter_head, npages;
1665 data_start(i, &iter_head, start);
1666 /* Amount of free space: some of this one + all after this one */
1667 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1668 n = npages * PAGE_SIZE - *start;
1672 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1673 p = get_pages_array(npages);
1676 n = __pipe_get_pages(i, maxsize, p, iter_head, start);
1684 static ssize_t iter_xarray_get_pages_alloc(struct iov_iter *i,
1685 struct page ***pages, size_t maxsize,
1686 size_t *_start_offset)
1689 unsigned nr, offset;
1690 pgoff_t index, count;
1691 size_t size = maxsize;
1697 pos = i->xarray_start + i->iov_offset;
1698 index = pos >> PAGE_SHIFT;
1699 offset = pos & ~PAGE_MASK;
1700 *_start_offset = offset;
1703 if (size > PAGE_SIZE - offset) {
1704 size -= PAGE_SIZE - offset;
1705 count += size >> PAGE_SHIFT;
1711 p = get_pages_array(count);
1716 nr = iter_xarray_populate_pages(p, i->xarray, index, count);
1720 return min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1723 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1724 struct page ***pages, size_t maxsize,
1731 if (maxsize > i->count)
1736 if (likely(iter_is_iovec(i))) {
1737 unsigned int gup_flags = 0;
1740 if (iov_iter_rw(i) != WRITE)
1741 gup_flags |= FOLL_WRITE;
1743 gup_flags |= FOLL_NOFAULT;
1745 addr = first_iovec_segment(i, &len, start, maxsize, ~0U);
1746 n = DIV_ROUND_UP(len, PAGE_SIZE);
1747 p = get_pages_array(n);
1750 res = get_user_pages_fast(addr, n, gup_flags, p);
1751 if (unlikely(res <= 0)) {
1757 return (res == n ? len : res * PAGE_SIZE) - *start;
1759 if (iov_iter_is_bvec(i)) {
1762 page = first_bvec_segment(i, &len, start, maxsize, ~0U);
1763 n = DIV_ROUND_UP(len, PAGE_SIZE);
1764 *pages = p = get_pages_array(n);
1768 get_page(*p++ = page++);
1769 return len - *start;
1771 if (iov_iter_is_pipe(i))
1772 return pipe_get_pages_alloc(i, pages, maxsize, start);
1773 if (iov_iter_is_xarray(i))
1774 return iter_xarray_get_pages_alloc(i, pages, maxsize, start);
1777 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1779 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1784 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1788 iterate_and_advance(i, bytes, base, len, off, ({
1789 next = csum_and_copy_from_user(base, addr + off, len);
1790 sum = csum_block_add(sum, next, off);
1793 sum = csum_and_memcpy(addr + off, base, len, sum, off);
1799 EXPORT_SYMBOL(csum_and_copy_from_iter);
1801 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1804 struct csum_state *csstate = _csstate;
1807 if (unlikely(iov_iter_is_discard(i))) {
1808 WARN_ON(1); /* for now */
1812 sum = csum_shift(csstate->csum, csstate->off);
1813 if (unlikely(iov_iter_is_pipe(i)))
1814 bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum);
1815 else iterate_and_advance(i, bytes, base, len, off, ({
1816 next = csum_and_copy_to_user(addr + off, base, len);
1817 sum = csum_block_add(sum, next, off);
1820 sum = csum_and_memcpy(base, addr + off, len, sum, off);
1823 csstate->csum = csum_shift(sum, csstate->off);
1824 csstate->off += bytes;
1827 EXPORT_SYMBOL(csum_and_copy_to_iter);
1829 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1832 #ifdef CONFIG_CRYPTO_HASH
1833 struct ahash_request *hash = hashp;
1834 struct scatterlist sg;
1837 copied = copy_to_iter(addr, bytes, i);
1838 sg_init_one(&sg, addr, copied);
1839 ahash_request_set_crypt(hash, &sg, NULL, copied);
1840 crypto_ahash_update(hash);
1846 EXPORT_SYMBOL(hash_and_copy_to_iter);
1848 static int iov_npages(const struct iov_iter *i, int maxpages)
1850 size_t skip = i->iov_offset, size = i->count;
1851 const struct iovec *p;
1854 for (p = i->iov; size; skip = 0, p++) {
1855 unsigned offs = offset_in_page(p->iov_base + skip);
1856 size_t len = min(p->iov_len - skip, size);
1860 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1861 if (unlikely(npages > maxpages))
1868 static int bvec_npages(const struct iov_iter *i, int maxpages)
1870 size_t skip = i->iov_offset, size = i->count;
1871 const struct bio_vec *p;
1874 for (p = i->bvec; size; skip = 0, p++) {
1875 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1876 size_t len = min(p->bv_len - skip, size);
1879 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1880 if (unlikely(npages > maxpages))
1886 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1888 if (unlikely(!i->count))
1890 /* iovec and kvec have identical layouts */
1891 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1892 return iov_npages(i, maxpages);
1893 if (iov_iter_is_bvec(i))
1894 return bvec_npages(i, maxpages);
1895 if (iov_iter_is_pipe(i)) {
1896 unsigned int iter_head;
1903 data_start(i, &iter_head, &off);
1904 /* some of this one + all after this one */
1905 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1906 return min(npages, maxpages);
1908 if (iov_iter_is_xarray(i)) {
1909 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1910 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1911 return min(npages, maxpages);
1915 EXPORT_SYMBOL(iov_iter_npages);
1917 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1920 if (unlikely(iov_iter_is_pipe(new))) {
1924 if (unlikely(iov_iter_is_discard(new) || iov_iter_is_xarray(new)))
1926 if (iov_iter_is_bvec(new))
1927 return new->bvec = kmemdup(new->bvec,
1928 new->nr_segs * sizeof(struct bio_vec),
1931 /* iovec and kvec have identical layout */
1932 return new->iov = kmemdup(new->iov,
1933 new->nr_segs * sizeof(struct iovec),
1936 EXPORT_SYMBOL(dup_iter);
1938 static int copy_compat_iovec_from_user(struct iovec *iov,
1939 const struct iovec __user *uvec, unsigned long nr_segs)
1941 const struct compat_iovec __user *uiov =
1942 (const struct compat_iovec __user *)uvec;
1943 int ret = -EFAULT, i;
1945 if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1948 for (i = 0; i < nr_segs; i++) {
1952 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1953 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1955 /* check for compat_size_t not fitting in compat_ssize_t .. */
1960 iov[i].iov_base = compat_ptr(buf);
1961 iov[i].iov_len = len;
1970 static int copy_iovec_from_user(struct iovec *iov,
1971 const struct iovec __user *uvec, unsigned long nr_segs)
1975 if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1977 for (seg = 0; seg < nr_segs; seg++) {
1978 if ((ssize_t)iov[seg].iov_len < 0)
1985 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1986 unsigned long nr_segs, unsigned long fast_segs,
1987 struct iovec *fast_iov, bool compat)
1989 struct iovec *iov = fast_iov;
1993 * SuS says "The readv() function *may* fail if the iovcnt argument was
1994 * less than or equal to 0, or greater than {IOV_MAX}. Linux has
1995 * traditionally returned zero for zero segments, so...
1999 if (nr_segs > UIO_MAXIOV)
2000 return ERR_PTR(-EINVAL);
2001 if (nr_segs > fast_segs) {
2002 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
2004 return ERR_PTR(-ENOMEM);
2008 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
2010 ret = copy_iovec_from_user(iov, uvec, nr_segs);
2012 if (iov != fast_iov)
2014 return ERR_PTR(ret);
2020 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
2021 unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
2022 struct iov_iter *i, bool compat)
2024 ssize_t total_len = 0;
2028 iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
2031 return PTR_ERR(iov);
2035 * According to the Single Unix Specification we should return EINVAL if
2036 * an element length is < 0 when cast to ssize_t or if the total length
2037 * would overflow the ssize_t return value of the system call.
2039 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
2042 for (seg = 0; seg < nr_segs; seg++) {
2043 ssize_t len = (ssize_t)iov[seg].iov_len;
2045 if (!access_ok(iov[seg].iov_base, len)) {
2052 if (len > MAX_RW_COUNT - total_len) {
2053 len = MAX_RW_COUNT - total_len;
2054 iov[seg].iov_len = len;
2059 iov_iter_init(i, type, iov, nr_segs, total_len);
2068 * import_iovec() - Copy an array of &struct iovec from userspace
2069 * into the kernel, check that it is valid, and initialize a new
2070 * &struct iov_iter iterator to access it.
2072 * @type: One of %READ or %WRITE.
2073 * @uvec: Pointer to the userspace array.
2074 * @nr_segs: Number of elements in userspace array.
2075 * @fast_segs: Number of elements in @iov.
2076 * @iovp: (input and output parameter) Pointer to pointer to (usually small
2077 * on-stack) kernel array.
2078 * @i: Pointer to iterator that will be initialized on success.
2080 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
2081 * then this function places %NULL in *@iov on return. Otherwise, a new
2082 * array will be allocated and the result placed in *@iov. This means that
2083 * the caller may call kfree() on *@iov regardless of whether the small
2084 * on-stack array was used or not (and regardless of whether this function
2085 * returns an error or not).
2087 * Return: Negative error code on error, bytes imported on success
2089 ssize_t import_iovec(int type, const struct iovec __user *uvec,
2090 unsigned nr_segs, unsigned fast_segs,
2091 struct iovec **iovp, struct iov_iter *i)
2093 return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
2094 in_compat_syscall());
2096 EXPORT_SYMBOL(import_iovec);
2098 int import_single_range(int rw, void __user *buf, size_t len,
2099 struct iovec *iov, struct iov_iter *i)
2101 if (len > MAX_RW_COUNT)
2103 if (unlikely(!access_ok(buf, len)))
2106 iov->iov_base = buf;
2108 iov_iter_init(i, rw, iov, 1, len);
2111 EXPORT_SYMBOL(import_single_range);
2114 * iov_iter_restore() - Restore a &struct iov_iter to the same state as when
2115 * iov_iter_save_state() was called.
2117 * @i: &struct iov_iter to restore
2118 * @state: state to restore from
2120 * Used after iov_iter_save_state() to bring restore @i, if operations may
2123 * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC
2125 void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state)
2127 if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i)) &&
2128 !iov_iter_is_kvec(i))
2130 i->iov_offset = state->iov_offset;
2131 i->count = state->count;
2133 * For the *vec iters, nr_segs + iov is constant - if we increment
2134 * the vec, then we also decrement the nr_segs count. Hence we don't
2135 * need to track both of these, just one is enough and we can deduct
2136 * the other from that. ITER_KVEC and ITER_IOVEC are the same struct
2137 * size, so we can just increment the iov pointer as they are unionzed.
2138 * ITER_BVEC _may_ be the same size on some archs, but on others it is
2139 * not. Be safe and handle it separately.
2141 BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
2142 if (iov_iter_is_bvec(i))
2143 i->bvec -= state->nr_segs - i->nr_segs;
2145 i->iov -= state->nr_segs - i->nr_segs;
2146 i->nr_segs = state->nr_segs;