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 #define iterate_iovec(i, n, __v, __p, skip, STEP) { \
23 __v.iov_len = min(n, __p->iov_len - skip); \
24 if (likely(__v.iov_len)) { \
25 __v.iov_base = __p->iov_base + skip; \
27 __v.iov_len -= left; \
28 skip += __v.iov_len; \
33 while (unlikely(!left && n)) { \
35 __v.iov_len = min(n, __p->iov_len); \
36 if (unlikely(!__v.iov_len)) \
38 __v.iov_base = __p->iov_base; \
40 __v.iov_len -= left; \
47 #define iterate_kvec(i, n, __v, __p, skip, STEP) { \
50 __v.iov_len = min(n, __p->iov_len - skip); \
51 if (likely(__v.iov_len)) { \
52 __v.iov_base = __p->iov_base + skip; \
54 skip += __v.iov_len; \
57 while (unlikely(n)) { \
59 __v.iov_len = min(n, __p->iov_len); \
60 if (unlikely(!__v.iov_len)) \
62 __v.iov_base = __p->iov_base; \
70 #define iterate_bvec(i, n, __v, __bi, skip, STEP) { \
71 struct bvec_iter __start; \
72 __start.bi_size = n; \
73 __start.bi_bvec_done = skip; \
75 for_each_bvec(__v, i->bvec, __bi, __start) { \
80 #define iterate_xarray(i, n, __v, skip, STEP) { \
81 struct page *head = NULL; \
82 size_t wanted = n, seg, offset; \
83 loff_t start = i->xarray_start + skip; \
84 pgoff_t index = start >> PAGE_SHIFT; \
87 XA_STATE(xas, i->xarray, index); \
90 xas_for_each(&xas, head, ULONG_MAX) { \
91 if (xas_retry(&xas, head)) \
93 if (WARN_ON(xa_is_value(head))) \
95 if (WARN_ON(PageHuge(head))) \
97 for (j = (head->index < index) ? index - head->index : 0; \
98 j < thp_nr_pages(head); j++) { \
99 __v.bv_page = head + j; \
100 offset = (i->xarray_start + skip) & ~PAGE_MASK; \
101 seg = PAGE_SIZE - offset; \
102 __v.bv_offset = offset; \
103 __v.bv_len = min(n, seg); \
106 skip += __v.bv_len; \
117 #define iterate_all_kinds(i, n, v, I, B, K, X) { \
119 size_t skip = i->iov_offset; \
120 if (likely(iter_is_iovec(i))) { \
121 const struct iovec *iov; \
123 iterate_iovec(i, n, v, iov, skip, (I)) \
124 } else if (iov_iter_is_bvec(i)) { \
126 struct bvec_iter __bi; \
127 iterate_bvec(i, n, v, __bi, skip, (B)) \
128 } else if (iov_iter_is_kvec(i)) { \
129 const struct kvec *kvec; \
131 iterate_kvec(i, n, v, kvec, skip, (K)) \
132 } else if (iov_iter_is_xarray(i)) { \
134 iterate_xarray(i, n, v, skip, (X)); \
139 #define iterate_and_advance(i, n, v, I, B, K, X) { \
140 if (unlikely(i->count < n)) \
143 size_t skip = i->iov_offset; \
144 if (likely(iter_is_iovec(i))) { \
145 const struct iovec *iov; \
147 iterate_iovec(i, n, v, iov, skip, (I)) \
148 if (skip == iov->iov_len) { \
152 i->nr_segs -= iov - i->iov; \
154 } else if (iov_iter_is_bvec(i)) { \
155 const struct bio_vec *bvec = i->bvec; \
157 struct bvec_iter __bi; \
158 iterate_bvec(i, n, v, __bi, skip, (B)) \
159 i->bvec = __bvec_iter_bvec(i->bvec, __bi); \
160 i->nr_segs -= i->bvec - bvec; \
161 skip = __bi.bi_bvec_done; \
162 } else if (iov_iter_is_kvec(i)) { \
163 const struct kvec *kvec; \
165 iterate_kvec(i, n, v, kvec, skip, (K)) \
166 if (skip == kvec->iov_len) { \
170 i->nr_segs -= kvec - i->kvec; \
172 } else if (iov_iter_is_xarray(i)) { \
174 iterate_xarray(i, n, v, skip, (X)) \
177 i->iov_offset = skip; \
181 static int copyout(void __user *to, const void *from, size_t n)
183 if (should_fail_usercopy())
185 if (access_ok(to, n)) {
186 instrument_copy_to_user(to, from, n);
187 n = raw_copy_to_user(to, from, n);
192 static int copyin(void *to, const void __user *from, size_t n)
194 if (should_fail_usercopy())
196 if (access_ok(from, n)) {
197 instrument_copy_from_user(to, from, n);
198 n = raw_copy_from_user(to, from, n);
203 static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes,
206 size_t skip, copy, left, wanted;
207 const struct iovec *iov;
211 if (unlikely(bytes > i->count))
214 if (unlikely(!bytes))
220 skip = i->iov_offset;
221 buf = iov->iov_base + skip;
222 copy = min(bytes, iov->iov_len - skip);
224 if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_writeable(buf, copy)) {
225 kaddr = kmap_atomic(page);
226 from = kaddr + offset;
228 /* first chunk, usually the only one */
229 left = copyout(buf, from, copy);
235 while (unlikely(!left && bytes)) {
238 copy = min(bytes, iov->iov_len);
239 left = copyout(buf, from, copy);
245 if (likely(!bytes)) {
246 kunmap_atomic(kaddr);
249 offset = from - kaddr;
251 kunmap_atomic(kaddr);
252 copy = min(bytes, iov->iov_len - skip);
254 /* Too bad - revert to non-atomic kmap */
257 from = kaddr + offset;
258 left = copyout(buf, from, copy);
263 while (unlikely(!left && bytes)) {
266 copy = min(bytes, iov->iov_len);
267 left = copyout(buf, from, copy);
276 if (skip == iov->iov_len) {
280 i->count -= wanted - bytes;
281 i->nr_segs -= iov - i->iov;
283 i->iov_offset = skip;
284 return wanted - bytes;
287 static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes,
290 size_t skip, copy, left, wanted;
291 const struct iovec *iov;
295 if (unlikely(bytes > i->count))
298 if (unlikely(!bytes))
304 skip = i->iov_offset;
305 buf = iov->iov_base + skip;
306 copy = min(bytes, iov->iov_len - skip);
308 if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_readable(buf, copy)) {
309 kaddr = kmap_atomic(page);
312 /* first chunk, usually the only one */
313 left = copyin(to, buf, copy);
319 while (unlikely(!left && bytes)) {
322 copy = min(bytes, iov->iov_len);
323 left = copyin(to, buf, copy);
329 if (likely(!bytes)) {
330 kunmap_atomic(kaddr);
335 kunmap_atomic(kaddr);
336 copy = min(bytes, iov->iov_len - skip);
338 /* Too bad - revert to non-atomic kmap */
342 left = copyin(to, buf, copy);
347 while (unlikely(!left && bytes)) {
350 copy = min(bytes, iov->iov_len);
351 left = copyin(to, buf, copy);
360 if (skip == iov->iov_len) {
364 i->count -= wanted - bytes;
365 i->nr_segs -= iov - i->iov;
367 i->iov_offset = skip;
368 return wanted - bytes;
372 static bool sanity(const struct iov_iter *i)
374 struct pipe_inode_info *pipe = i->pipe;
375 unsigned int p_head = pipe->head;
376 unsigned int p_tail = pipe->tail;
377 unsigned int p_mask = pipe->ring_size - 1;
378 unsigned int p_occupancy = pipe_occupancy(p_head, p_tail);
379 unsigned int i_head = i->head;
383 struct pipe_buffer *p;
384 if (unlikely(p_occupancy == 0))
385 goto Bad; // pipe must be non-empty
386 if (unlikely(i_head != p_head - 1))
387 goto Bad; // must be at the last buffer...
389 p = &pipe->bufs[i_head & p_mask];
390 if (unlikely(p->offset + p->len != i->iov_offset))
391 goto Bad; // ... at the end of segment
393 if (i_head != p_head)
394 goto Bad; // must be right after the last buffer
398 printk(KERN_ERR "idx = %d, offset = %zd\n", i_head, i->iov_offset);
399 printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n",
400 p_head, p_tail, pipe->ring_size);
401 for (idx = 0; idx < pipe->ring_size; idx++)
402 printk(KERN_ERR "[%p %p %d %d]\n",
404 pipe->bufs[idx].page,
405 pipe->bufs[idx].offset,
406 pipe->bufs[idx].len);
411 #define sanity(i) true
414 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
417 struct pipe_inode_info *pipe = i->pipe;
418 struct pipe_buffer *buf;
419 unsigned int p_tail = pipe->tail;
420 unsigned int p_mask = pipe->ring_size - 1;
421 unsigned int i_head = i->head;
424 if (unlikely(bytes > i->count))
427 if (unlikely(!bytes))
434 buf = &pipe->bufs[i_head & p_mask];
436 if (offset == off && buf->page == page) {
437 /* merge with the last one */
439 i->iov_offset += bytes;
443 buf = &pipe->bufs[i_head & p_mask];
445 if (pipe_full(i_head, p_tail, pipe->max_usage))
448 buf->ops = &page_cache_pipe_buf_ops;
451 buf->offset = offset;
454 pipe->head = i_head + 1;
455 i->iov_offset = offset + bytes;
463 * Fault in one or more iovecs of the given iov_iter, to a maximum length of
464 * bytes. For each iovec, fault in each page that constitutes the iovec.
466 * Return 0 on success, or non-zero if the memory could not be accessed (i.e.
467 * because it is an invalid address).
469 int iov_iter_fault_in_readable(const struct iov_iter *i, size_t bytes)
471 if (iter_is_iovec(i)) {
472 const struct iovec *p;
475 if (bytes > i->count)
477 for (p = i->iov, skip = i->iov_offset; bytes; p++, skip = 0) {
478 size_t len = min(bytes, p->iov_len - skip);
483 err = fault_in_pages_readable(p->iov_base + skip, len);
491 EXPORT_SYMBOL(iov_iter_fault_in_readable);
493 void iov_iter_init(struct iov_iter *i, unsigned int direction,
494 const struct iovec *iov, unsigned long nr_segs,
497 WARN_ON(direction & ~(READ | WRITE));
498 WARN_ON_ONCE(uaccess_kernel());
499 *i = (struct iov_iter) {
500 .iter_type = ITER_IOVEC,
501 .data_source = direction,
508 EXPORT_SYMBOL(iov_iter_init);
510 static inline bool allocated(struct pipe_buffer *buf)
512 return buf->ops == &default_pipe_buf_ops;
515 static inline void data_start(const struct iov_iter *i,
516 unsigned int *iter_headp, size_t *offp)
518 unsigned int p_mask = i->pipe->ring_size - 1;
519 unsigned int iter_head = i->head;
520 size_t off = i->iov_offset;
522 if (off && (!allocated(&i->pipe->bufs[iter_head & p_mask]) ||
527 *iter_headp = iter_head;
531 static size_t push_pipe(struct iov_iter *i, size_t size,
532 int *iter_headp, size_t *offp)
534 struct pipe_inode_info *pipe = i->pipe;
535 unsigned int p_tail = pipe->tail;
536 unsigned int p_mask = pipe->ring_size - 1;
537 unsigned int iter_head;
541 if (unlikely(size > i->count))
547 data_start(i, &iter_head, &off);
548 *iter_headp = iter_head;
551 left -= PAGE_SIZE - off;
553 pipe->bufs[iter_head & p_mask].len += size;
556 pipe->bufs[iter_head & p_mask].len = PAGE_SIZE;
559 while (!pipe_full(iter_head, p_tail, pipe->max_usage)) {
560 struct pipe_buffer *buf = &pipe->bufs[iter_head & p_mask];
561 struct page *page = alloc_page(GFP_USER);
565 buf->ops = &default_pipe_buf_ops;
568 buf->len = min_t(ssize_t, left, PAGE_SIZE);
571 pipe->head = iter_head;
579 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
582 struct pipe_inode_info *pipe = i->pipe;
583 unsigned int p_mask = pipe->ring_size - 1;
590 bytes = n = push_pipe(i, bytes, &i_head, &off);
594 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
595 memcpy_to_page(pipe->bufs[i_head & p_mask].page, off, addr, chunk);
597 i->iov_offset = off + chunk;
607 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
608 __wsum sum, size_t off)
610 __wsum next = csum_partial_copy_nocheck(from, to, len);
611 return csum_block_add(sum, next, off);
614 static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
615 struct csum_state *csstate,
618 struct pipe_inode_info *pipe = i->pipe;
619 unsigned int p_mask = pipe->ring_size - 1;
620 __wsum sum = csstate->csum;
621 size_t off = csstate->off;
628 bytes = n = push_pipe(i, bytes, &i_head, &r);
632 size_t chunk = min_t(size_t, n, PAGE_SIZE - r);
633 char *p = kmap_atomic(pipe->bufs[i_head & p_mask].page);
634 sum = csum_and_memcpy(p + r, addr, chunk, sum, off);
637 i->iov_offset = r + chunk;
650 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
652 const char *from = addr;
653 if (unlikely(iov_iter_is_pipe(i)))
654 return copy_pipe_to_iter(addr, bytes, i);
655 if (iter_is_iovec(i))
657 iterate_and_advance(i, bytes, v,
658 copyout(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
659 memcpy_to_page(v.bv_page, v.bv_offset,
660 (from += v.bv_len) - v.bv_len, v.bv_len),
661 memcpy(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
662 memcpy_to_page(v.bv_page, v.bv_offset,
663 (from += v.bv_len) - v.bv_len, v.bv_len)
668 EXPORT_SYMBOL(_copy_to_iter);
670 #ifdef CONFIG_ARCH_HAS_COPY_MC
671 static int copyout_mc(void __user *to, const void *from, size_t n)
673 if (access_ok(to, n)) {
674 instrument_copy_to_user(to, from, n);
675 n = copy_mc_to_user((__force void *) to, from, n);
680 static unsigned long copy_mc_to_page(struct page *page, size_t offset,
681 const char *from, size_t len)
686 to = kmap_atomic(page);
687 ret = copy_mc_to_kernel(to + offset, from, len);
693 static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes,
696 struct pipe_inode_info *pipe = i->pipe;
697 unsigned int p_mask = pipe->ring_size - 1;
699 size_t n, off, xfer = 0;
704 bytes = n = push_pipe(i, bytes, &i_head, &off);
708 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
711 rem = copy_mc_to_page(pipe->bufs[i_head & p_mask].page,
714 i->iov_offset = off + chunk - rem;
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 * @iter: 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 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
752 const char *from = addr;
753 unsigned long rem, curr_addr, s_addr = (unsigned long) addr;
755 if (unlikely(iov_iter_is_pipe(i)))
756 return copy_mc_pipe_to_iter(addr, bytes, i);
757 if (iter_is_iovec(i))
759 iterate_and_advance(i, bytes, v,
760 copyout_mc(v.iov_base, (from += v.iov_len) - v.iov_len,
763 rem = copy_mc_to_page(v.bv_page, v.bv_offset,
764 (from += v.bv_len) - v.bv_len, v.bv_len);
766 curr_addr = (unsigned long) from;
767 bytes = curr_addr - s_addr - rem;
772 rem = copy_mc_to_kernel(v.iov_base, (from += v.iov_len)
773 - v.iov_len, v.iov_len);
775 curr_addr = (unsigned long) from;
776 bytes = curr_addr - s_addr - rem;
781 rem = copy_mc_to_page(v.bv_page, v.bv_offset,
782 (from += v.bv_len) - v.bv_len, v.bv_len);
784 curr_addr = (unsigned long) from;
785 bytes = curr_addr - s_addr - rem;
787 i->iov_offset += bytes;
796 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
797 #endif /* CONFIG_ARCH_HAS_COPY_MC */
799 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
802 if (unlikely(iov_iter_is_pipe(i))) {
806 if (iter_is_iovec(i))
808 iterate_and_advance(i, bytes, v,
809 copyin((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
810 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
811 v.bv_offset, v.bv_len),
812 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
813 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
814 v.bv_offset, v.bv_len)
819 EXPORT_SYMBOL(_copy_from_iter);
821 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
824 if (unlikely(iov_iter_is_pipe(i))) {
828 iterate_and_advance(i, bytes, v,
829 __copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
830 v.iov_base, v.iov_len),
831 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
832 v.bv_offset, v.bv_len),
833 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
834 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
835 v.bv_offset, v.bv_len)
840 EXPORT_SYMBOL(_copy_from_iter_nocache);
842 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
844 * _copy_from_iter_flushcache - write destination through cpu cache
845 * @addr: destination kernel address
846 * @bytes: total transfer length
847 * @iter: source iterator
849 * The pmem driver arranges for filesystem-dax to use this facility via
850 * dax_copy_from_iter() for ensuring that writes to persistent memory
851 * are flushed through the CPU cache. It is differentiated from
852 * _copy_from_iter_nocache() in that guarantees all data is flushed for
853 * all iterator types. The _copy_from_iter_nocache() only attempts to
854 * bypass the cache for the ITER_IOVEC case, and on some archs may use
855 * instructions that strand dirty-data in the cache.
857 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
860 if (unlikely(iov_iter_is_pipe(i))) {
864 iterate_and_advance(i, bytes, v,
865 __copy_from_user_flushcache((to += v.iov_len) - v.iov_len,
866 v.iov_base, v.iov_len),
867 memcpy_page_flushcache((to += v.bv_len) - v.bv_len, v.bv_page,
868 v.bv_offset, v.bv_len),
869 memcpy_flushcache((to += v.iov_len) - v.iov_len, v.iov_base,
871 memcpy_page_flushcache((to += v.bv_len) - v.bv_len, v.bv_page,
872 v.bv_offset, v.bv_len)
877 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
880 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
883 size_t v = n + offset;
886 * The general case needs to access the page order in order
887 * to compute the page size.
888 * However, we mostly deal with order-0 pages and thus can
889 * avoid a possible cache line miss for requests that fit all
892 if (n <= v && v <= PAGE_SIZE)
895 head = compound_head(page);
896 v += (page - head) << PAGE_SHIFT;
898 if (likely(n <= v && v <= (page_size(head))))
904 static size_t __copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
907 if (likely(iter_is_iovec(i)))
908 return copy_page_to_iter_iovec(page, offset, bytes, i);
909 if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
910 void *kaddr = kmap_atomic(page);
911 size_t wanted = copy_to_iter(kaddr + offset, bytes, i);
912 kunmap_atomic(kaddr);
915 if (iov_iter_is_pipe(i))
916 return copy_page_to_iter_pipe(page, offset, bytes, i);
917 if (unlikely(iov_iter_is_discard(i))) {
918 if (unlikely(i->count < bytes))
927 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
931 if (unlikely(!page_copy_sane(page, offset, bytes)))
933 page += offset / PAGE_SIZE; // first subpage
936 size_t n = __copy_page_to_iter(page, offset,
937 min(bytes, (size_t)PAGE_SIZE - offset), i);
943 if (offset == PAGE_SIZE) {
950 EXPORT_SYMBOL(copy_page_to_iter);
952 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
955 if (unlikely(!page_copy_sane(page, offset, bytes)))
957 if (likely(iter_is_iovec(i)))
958 return copy_page_from_iter_iovec(page, offset, bytes, i);
959 if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
960 void *kaddr = kmap_atomic(page);
961 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
962 kunmap_atomic(kaddr);
968 EXPORT_SYMBOL(copy_page_from_iter);
970 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
972 struct pipe_inode_info *pipe = i->pipe;
973 unsigned int p_mask = pipe->ring_size - 1;
980 bytes = n = push_pipe(i, bytes, &i_head, &off);
985 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
986 memzero_page(pipe->bufs[i_head & p_mask].page, off, chunk);
988 i->iov_offset = off + chunk;
997 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
999 if (unlikely(iov_iter_is_pipe(i)))
1000 return pipe_zero(bytes, i);
1001 iterate_and_advance(i, bytes, v,
1002 clear_user(v.iov_base, v.iov_len),
1003 memzero_page(v.bv_page, v.bv_offset, v.bv_len),
1004 memset(v.iov_base, 0, v.iov_len),
1005 memzero_page(v.bv_page, v.bv_offset, v.bv_len)
1010 EXPORT_SYMBOL(iov_iter_zero);
1012 size_t iov_iter_copy_from_user_atomic(struct page *page,
1013 struct iov_iter *i, unsigned long offset, size_t bytes)
1015 char *kaddr = kmap_atomic(page), *p = kaddr + offset;
1016 if (unlikely(!page_copy_sane(page, offset, bytes))) {
1017 kunmap_atomic(kaddr);
1020 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1021 kunmap_atomic(kaddr);
1025 iterate_all_kinds(i, bytes, v,
1026 copyin((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
1027 memcpy_from_page((p += v.bv_len) - v.bv_len, v.bv_page,
1028 v.bv_offset, v.bv_len),
1029 memcpy((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
1030 memcpy_from_page((p += v.bv_len) - v.bv_len, v.bv_page,
1031 v.bv_offset, v.bv_len)
1033 kunmap_atomic(kaddr);
1036 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
1038 static inline void pipe_truncate(struct iov_iter *i)
1040 struct pipe_inode_info *pipe = i->pipe;
1041 unsigned int p_tail = pipe->tail;
1042 unsigned int p_head = pipe->head;
1043 unsigned int p_mask = pipe->ring_size - 1;
1045 if (!pipe_empty(p_head, p_tail)) {
1046 struct pipe_buffer *buf;
1047 unsigned int i_head = i->head;
1048 size_t off = i->iov_offset;
1051 buf = &pipe->bufs[i_head & p_mask];
1052 buf->len = off - buf->offset;
1055 while (p_head != i_head) {
1057 pipe_buf_release(pipe, &pipe->bufs[p_head & p_mask]);
1060 pipe->head = p_head;
1064 static void pipe_advance(struct iov_iter *i, size_t size)
1066 struct pipe_inode_info *pipe = i->pipe;
1068 struct pipe_buffer *buf;
1069 unsigned int p_mask = pipe->ring_size - 1;
1070 unsigned int i_head = i->head;
1071 size_t off = i->iov_offset, left = size;
1073 if (off) /* make it relative to the beginning of buffer */
1074 left += off - pipe->bufs[i_head & p_mask].offset;
1076 buf = &pipe->bufs[i_head & p_mask];
1077 if (left <= buf->len)
1083 i->iov_offset = buf->offset + left;
1086 /* ... and discard everything past that point */
1090 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
1092 struct bvec_iter bi;
1094 bi.bi_size = i->count;
1095 bi.bi_bvec_done = i->iov_offset;
1097 bvec_iter_advance(i->bvec, &bi, size);
1099 i->bvec += bi.bi_idx;
1100 i->nr_segs -= bi.bi_idx;
1101 i->count = bi.bi_size;
1102 i->iov_offset = bi.bi_bvec_done;
1105 static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
1107 const struct iovec *iov, *end;
1113 size += i->iov_offset; // from beginning of current segment
1114 for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) {
1115 if (likely(size < iov->iov_len))
1117 size -= iov->iov_len;
1119 i->iov_offset = size;
1120 i->nr_segs -= iov - i->iov;
1124 void iov_iter_advance(struct iov_iter *i, size_t size)
1126 if (unlikely(i->count < size))
1128 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
1129 /* iovec and kvec have identical layouts */
1130 iov_iter_iovec_advance(i, size);
1131 } else if (iov_iter_is_bvec(i)) {
1132 iov_iter_bvec_advance(i, size);
1133 } else if (iov_iter_is_pipe(i)) {
1134 pipe_advance(i, size);
1135 } else if (unlikely(iov_iter_is_xarray(i))) {
1136 i->iov_offset += size;
1138 } else if (iov_iter_is_discard(i)) {
1142 EXPORT_SYMBOL(iov_iter_advance);
1144 void iov_iter_revert(struct iov_iter *i, size_t unroll)
1148 if (WARN_ON(unroll > MAX_RW_COUNT))
1151 if (unlikely(iov_iter_is_pipe(i))) {
1152 struct pipe_inode_info *pipe = i->pipe;
1153 unsigned int p_mask = pipe->ring_size - 1;
1154 unsigned int i_head = i->head;
1155 size_t off = i->iov_offset;
1157 struct pipe_buffer *b = &pipe->bufs[i_head & p_mask];
1158 size_t n = off - b->offset;
1164 if (!unroll && i_head == i->start_head) {
1169 b = &pipe->bufs[i_head & p_mask];
1170 off = b->offset + b->len;
1172 i->iov_offset = off;
1177 if (unlikely(iov_iter_is_discard(i)))
1179 if (unroll <= i->iov_offset) {
1180 i->iov_offset -= unroll;
1183 unroll -= i->iov_offset;
1184 if (iov_iter_is_xarray(i)) {
1185 BUG(); /* We should never go beyond the start of the specified
1186 * range since we might then be straying into pages that
1189 } else if (iov_iter_is_bvec(i)) {
1190 const struct bio_vec *bvec = i->bvec;
1192 size_t n = (--bvec)->bv_len;
1196 i->iov_offset = n - unroll;
1201 } else { /* same logics for iovec and kvec */
1202 const struct iovec *iov = i->iov;
1204 size_t n = (--iov)->iov_len;
1208 i->iov_offset = n - unroll;
1215 EXPORT_SYMBOL(iov_iter_revert);
1218 * Return the count of just the current iov_iter segment.
1220 size_t iov_iter_single_seg_count(const struct iov_iter *i)
1222 if (i->nr_segs > 1) {
1223 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1224 return min(i->count, i->iov->iov_len - i->iov_offset);
1225 if (iov_iter_is_bvec(i))
1226 return min(i->count, i->bvec->bv_len - i->iov_offset);
1230 EXPORT_SYMBOL(iov_iter_single_seg_count);
1232 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
1233 const struct kvec *kvec, unsigned long nr_segs,
1236 WARN_ON(direction & ~(READ | WRITE));
1237 *i = (struct iov_iter){
1238 .iter_type = ITER_KVEC,
1239 .data_source = direction,
1246 EXPORT_SYMBOL(iov_iter_kvec);
1248 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1249 const struct bio_vec *bvec, unsigned long nr_segs,
1252 WARN_ON(direction & ~(READ | WRITE));
1253 *i = (struct iov_iter){
1254 .iter_type = ITER_BVEC,
1255 .data_source = direction,
1262 EXPORT_SYMBOL(iov_iter_bvec);
1264 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1265 struct pipe_inode_info *pipe,
1268 BUG_ON(direction != READ);
1269 WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1270 *i = (struct iov_iter){
1271 .iter_type = ITER_PIPE,
1272 .data_source = false,
1275 .start_head = pipe->head,
1280 EXPORT_SYMBOL(iov_iter_pipe);
1283 * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1284 * @i: The iterator to initialise.
1285 * @direction: The direction of the transfer.
1286 * @xarray: The xarray to access.
1287 * @start: The start file position.
1288 * @count: The size of the I/O buffer in bytes.
1290 * Set up an I/O iterator to either draw data out of the pages attached to an
1291 * inode or to inject data into those pages. The pages *must* be prevented
1292 * from evaporation, either by taking a ref on them or locking them by the
1295 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
1296 struct xarray *xarray, loff_t start, size_t count)
1298 BUG_ON(direction & ~1);
1299 *i = (struct iov_iter) {
1300 .iter_type = ITER_XARRAY,
1301 .data_source = direction,
1303 .xarray_start = start,
1308 EXPORT_SYMBOL(iov_iter_xarray);
1311 * iov_iter_discard - Initialise an I/O iterator that discards data
1312 * @i: The iterator to initialise.
1313 * @direction: The direction of the transfer.
1314 * @count: The size of the I/O buffer in bytes.
1316 * Set up an I/O iterator that just discards everything that's written to it.
1317 * It's only available as a READ iterator.
1319 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1321 BUG_ON(direction != READ);
1322 *i = (struct iov_iter){
1323 .iter_type = ITER_DISCARD,
1324 .data_source = false,
1329 EXPORT_SYMBOL(iov_iter_discard);
1331 unsigned long iov_iter_alignment(const struct iov_iter *i)
1333 unsigned long res = 0;
1334 size_t size = i->count;
1336 if (unlikely(iov_iter_is_pipe(i))) {
1337 unsigned int p_mask = i->pipe->ring_size - 1;
1339 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->head & p_mask]))
1340 return size | i->iov_offset;
1343 if (unlikely(iov_iter_is_xarray(i)))
1344 return (i->xarray_start + i->iov_offset) | i->count;
1345 iterate_all_kinds(i, size, v,
1346 (res |= (unsigned long)v.iov_base | v.iov_len, 0),
1347 res |= v.bv_offset | v.bv_len,
1348 res |= (unsigned long)v.iov_base | v.iov_len,
1349 res |= v.bv_offset | v.bv_len
1353 EXPORT_SYMBOL(iov_iter_alignment);
1355 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1357 unsigned long res = 0;
1358 size_t size = i->count;
1360 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1365 iterate_all_kinds(i, size, v,
1366 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1367 (size != v.iov_len ? size : 0), 0),
1368 (res |= (!res ? 0 : (unsigned long)v.bv_offset) |
1369 (size != v.bv_len ? size : 0)),
1370 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1371 (size != v.iov_len ? size : 0)),
1372 (res |= (!res ? 0 : (unsigned long)v.bv_offset) |
1373 (size != v.bv_len ? size : 0))
1377 EXPORT_SYMBOL(iov_iter_gap_alignment);
1379 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1381 struct page **pages,
1385 struct pipe_inode_info *pipe = i->pipe;
1386 unsigned int p_mask = pipe->ring_size - 1;
1387 ssize_t n = push_pipe(i, maxsize, &iter_head, start);
1394 get_page(*pages++ = pipe->bufs[iter_head & p_mask].page);
1402 static ssize_t pipe_get_pages(struct iov_iter *i,
1403 struct page **pages, size_t maxsize, unsigned maxpages,
1406 unsigned int iter_head, npages;
1415 data_start(i, &iter_head, start);
1416 /* Amount of free space: some of this one + all after this one */
1417 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1418 capacity = min(npages, maxpages) * PAGE_SIZE - *start;
1420 return __pipe_get_pages(i, min(maxsize, capacity), pages, iter_head, start);
1423 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1424 pgoff_t index, unsigned int nr_pages)
1426 XA_STATE(xas, xa, index);
1428 unsigned int ret = 0;
1431 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1432 if (xas_retry(&xas, page))
1435 /* Has the page moved or been split? */
1436 if (unlikely(page != xas_reload(&xas))) {
1441 pages[ret] = find_subpage(page, xas.xa_index);
1442 get_page(pages[ret]);
1443 if (++ret == nr_pages)
1450 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1451 struct page **pages, size_t maxsize,
1452 unsigned maxpages, size_t *_start_offset)
1454 unsigned nr, offset;
1455 pgoff_t index, count;
1456 size_t size = maxsize, actual;
1459 if (!size || !maxpages)
1462 pos = i->xarray_start + i->iov_offset;
1463 index = pos >> PAGE_SHIFT;
1464 offset = pos & ~PAGE_MASK;
1465 *_start_offset = offset;
1468 if (size > PAGE_SIZE - offset) {
1469 size -= PAGE_SIZE - offset;
1470 count += size >> PAGE_SHIFT;
1476 if (count > maxpages)
1479 nr = iter_xarray_populate_pages(pages, i->xarray, index, count);
1483 actual = PAGE_SIZE * nr;
1485 if (nr == count && size > 0) {
1486 unsigned last_offset = (nr > 1) ? 0 : offset;
1487 actual -= PAGE_SIZE - (last_offset + size);
1492 ssize_t iov_iter_get_pages(struct iov_iter *i,
1493 struct page **pages, size_t maxsize, unsigned maxpages,
1496 if (maxsize > i->count)
1499 if (unlikely(iov_iter_is_pipe(i)))
1500 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1501 if (unlikely(iov_iter_is_xarray(i)))
1502 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1503 if (unlikely(iov_iter_is_discard(i)))
1506 iterate_all_kinds(i, maxsize, v, ({
1507 unsigned long addr = (unsigned long)v.iov_base;
1508 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1512 if (len > maxpages * PAGE_SIZE)
1513 len = maxpages * PAGE_SIZE;
1514 addr &= ~(PAGE_SIZE - 1);
1515 n = DIV_ROUND_UP(len, PAGE_SIZE);
1516 res = get_user_pages_fast(addr, n,
1517 iov_iter_rw(i) != WRITE ? FOLL_WRITE : 0,
1519 if (unlikely(res < 0))
1521 return (res == n ? len : res * PAGE_SIZE) - *start;
1523 /* can't be more than PAGE_SIZE */
1524 *start = v.bv_offset;
1525 get_page(*pages = v.bv_page);
1534 EXPORT_SYMBOL(iov_iter_get_pages);
1536 static struct page **get_pages_array(size_t n)
1538 return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1541 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1542 struct page ***pages, size_t maxsize,
1546 unsigned int iter_head, npages;
1555 data_start(i, &iter_head, start);
1556 /* Amount of free space: some of this one + all after this one */
1557 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1558 n = npages * PAGE_SIZE - *start;
1562 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1563 p = get_pages_array(npages);
1566 n = __pipe_get_pages(i, maxsize, p, iter_head, start);
1574 static ssize_t iter_xarray_get_pages_alloc(struct iov_iter *i,
1575 struct page ***pages, size_t maxsize,
1576 size_t *_start_offset)
1579 unsigned nr, offset;
1580 pgoff_t index, count;
1581 size_t size = maxsize, actual;
1587 pos = i->xarray_start + i->iov_offset;
1588 index = pos >> PAGE_SHIFT;
1589 offset = pos & ~PAGE_MASK;
1590 *_start_offset = offset;
1593 if (size > PAGE_SIZE - offset) {
1594 size -= PAGE_SIZE - offset;
1595 count += size >> PAGE_SHIFT;
1601 p = get_pages_array(count);
1606 nr = iter_xarray_populate_pages(p, i->xarray, index, count);
1610 actual = PAGE_SIZE * nr;
1612 if (nr == count && size > 0) {
1613 unsigned last_offset = (nr > 1) ? 0 : offset;
1614 actual -= PAGE_SIZE - (last_offset + size);
1619 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1620 struct page ***pages, size_t maxsize,
1625 if (maxsize > i->count)
1628 if (unlikely(iov_iter_is_pipe(i)))
1629 return pipe_get_pages_alloc(i, pages, maxsize, start);
1630 if (unlikely(iov_iter_is_xarray(i)))
1631 return iter_xarray_get_pages_alloc(i, pages, maxsize, start);
1632 if (unlikely(iov_iter_is_discard(i)))
1635 iterate_all_kinds(i, maxsize, v, ({
1636 unsigned long addr = (unsigned long)v.iov_base;
1637 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1641 addr &= ~(PAGE_SIZE - 1);
1642 n = DIV_ROUND_UP(len, PAGE_SIZE);
1643 p = get_pages_array(n);
1646 res = get_user_pages_fast(addr, n,
1647 iov_iter_rw(i) != WRITE ? FOLL_WRITE : 0, p);
1648 if (unlikely(res < 0)) {
1653 return (res == n ? len : res * PAGE_SIZE) - *start;
1655 /* can't be more than PAGE_SIZE */
1656 *start = v.bv_offset;
1657 *pages = p = get_pages_array(1);
1660 get_page(*p = v.bv_page);
1668 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1670 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1677 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1681 iterate_and_advance(i, bytes, v, ({
1682 next = csum_and_copy_from_user(v.iov_base,
1683 (to += v.iov_len) - v.iov_len,
1686 sum = csum_block_add(sum, next, off);
1689 next ? 0 : v.iov_len;
1691 char *p = kmap_atomic(v.bv_page);
1692 sum = csum_and_memcpy((to += v.bv_len) - v.bv_len,
1693 p + v.bv_offset, v.bv_len,
1698 sum = csum_and_memcpy((to += v.iov_len) - v.iov_len,
1699 v.iov_base, v.iov_len,
1703 char *p = kmap_atomic(v.bv_page);
1704 sum = csum_and_memcpy((to += v.bv_len) - v.bv_len,
1705 p + v.bv_offset, v.bv_len,
1714 EXPORT_SYMBOL(csum_and_copy_from_iter);
1716 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1719 struct csum_state *csstate = _csstate;
1720 const char *from = addr;
1724 if (unlikely(iov_iter_is_pipe(i)))
1725 return csum_and_copy_to_pipe_iter(addr, bytes, _csstate, i);
1727 sum = csstate->csum;
1729 if (unlikely(iov_iter_is_discard(i))) {
1730 WARN_ON(1); /* for now */
1733 iterate_and_advance(i, bytes, v, ({
1734 next = csum_and_copy_to_user((from += v.iov_len) - v.iov_len,
1738 sum = csum_block_add(sum, next, off);
1741 next ? 0 : v.iov_len;
1743 char *p = kmap_atomic(v.bv_page);
1744 sum = csum_and_memcpy(p + v.bv_offset,
1745 (from += v.bv_len) - v.bv_len,
1746 v.bv_len, sum, off);
1750 sum = csum_and_memcpy(v.iov_base,
1751 (from += v.iov_len) - v.iov_len,
1752 v.iov_len, sum, off);
1755 char *p = kmap_atomic(v.bv_page);
1756 sum = csum_and_memcpy(p + v.bv_offset,
1757 (from += v.bv_len) - v.bv_len,
1758 v.bv_len, sum, off);
1763 csstate->csum = sum;
1767 EXPORT_SYMBOL(csum_and_copy_to_iter);
1769 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1772 #ifdef CONFIG_CRYPTO_HASH
1773 struct ahash_request *hash = hashp;
1774 struct scatterlist sg;
1777 copied = copy_to_iter(addr, bytes, i);
1778 sg_init_one(&sg, addr, copied);
1779 ahash_request_set_crypt(hash, &sg, NULL, copied);
1780 crypto_ahash_update(hash);
1786 EXPORT_SYMBOL(hash_and_copy_to_iter);
1788 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1790 size_t size = i->count;
1795 if (unlikely(iov_iter_is_discard(i)))
1798 if (unlikely(iov_iter_is_pipe(i))) {
1799 struct pipe_inode_info *pipe = i->pipe;
1800 unsigned int iter_head;
1806 data_start(i, &iter_head, &off);
1807 /* some of this one + all after this one */
1808 npages = pipe_space_for_user(iter_head, pipe->tail, pipe);
1809 if (npages >= maxpages)
1811 } else if (unlikely(iov_iter_is_xarray(i))) {
1814 offset = (i->xarray_start + i->iov_offset) & ~PAGE_MASK;
1817 if (size > PAGE_SIZE - offset) {
1818 size -= PAGE_SIZE - offset;
1819 npages += size >> PAGE_SHIFT;
1824 if (npages >= maxpages)
1826 } else iterate_all_kinds(i, size, v, ({
1827 unsigned long p = (unsigned long)v.iov_base;
1828 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1830 if (npages >= maxpages)
1834 if (npages >= maxpages)
1837 unsigned long p = (unsigned long)v.iov_base;
1838 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1840 if (npages >= maxpages)
1847 EXPORT_SYMBOL(iov_iter_npages);
1849 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1852 if (unlikely(iov_iter_is_pipe(new))) {
1856 if (unlikely(iov_iter_is_discard(new) || iov_iter_is_xarray(new)))
1858 if (iov_iter_is_bvec(new))
1859 return new->bvec = kmemdup(new->bvec,
1860 new->nr_segs * sizeof(struct bio_vec),
1863 /* iovec and kvec have identical layout */
1864 return new->iov = kmemdup(new->iov,
1865 new->nr_segs * sizeof(struct iovec),
1868 EXPORT_SYMBOL(dup_iter);
1870 static int copy_compat_iovec_from_user(struct iovec *iov,
1871 const struct iovec __user *uvec, unsigned long nr_segs)
1873 const struct compat_iovec __user *uiov =
1874 (const struct compat_iovec __user *)uvec;
1875 int ret = -EFAULT, i;
1877 if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1880 for (i = 0; i < nr_segs; i++) {
1884 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1885 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1887 /* check for compat_size_t not fitting in compat_ssize_t .. */
1892 iov[i].iov_base = compat_ptr(buf);
1893 iov[i].iov_len = len;
1902 static int copy_iovec_from_user(struct iovec *iov,
1903 const struct iovec __user *uvec, unsigned long nr_segs)
1907 if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1909 for (seg = 0; seg < nr_segs; seg++) {
1910 if ((ssize_t)iov[seg].iov_len < 0)
1917 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1918 unsigned long nr_segs, unsigned long fast_segs,
1919 struct iovec *fast_iov, bool compat)
1921 struct iovec *iov = fast_iov;
1925 * SuS says "The readv() function *may* fail if the iovcnt argument was
1926 * less than or equal to 0, or greater than {IOV_MAX}. Linux has
1927 * traditionally returned zero for zero segments, so...
1931 if (nr_segs > UIO_MAXIOV)
1932 return ERR_PTR(-EINVAL);
1933 if (nr_segs > fast_segs) {
1934 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1936 return ERR_PTR(-ENOMEM);
1940 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1942 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1944 if (iov != fast_iov)
1946 return ERR_PTR(ret);
1952 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1953 unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1954 struct iov_iter *i, bool compat)
1956 ssize_t total_len = 0;
1960 iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1963 return PTR_ERR(iov);
1967 * According to the Single Unix Specification we should return EINVAL if
1968 * an element length is < 0 when cast to ssize_t or if the total length
1969 * would overflow the ssize_t return value of the system call.
1971 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1974 for (seg = 0; seg < nr_segs; seg++) {
1975 ssize_t len = (ssize_t)iov[seg].iov_len;
1977 if (!access_ok(iov[seg].iov_base, len)) {
1984 if (len > MAX_RW_COUNT - total_len) {
1985 len = MAX_RW_COUNT - total_len;
1986 iov[seg].iov_len = len;
1991 iov_iter_init(i, type, iov, nr_segs, total_len);
2000 * import_iovec() - Copy an array of &struct iovec from userspace
2001 * into the kernel, check that it is valid, and initialize a new
2002 * &struct iov_iter iterator to access it.
2004 * @type: One of %READ or %WRITE.
2005 * @uvec: Pointer to the userspace array.
2006 * @nr_segs: Number of elements in userspace array.
2007 * @fast_segs: Number of elements in @iov.
2008 * @iovp: (input and output parameter) Pointer to pointer to (usually small
2009 * on-stack) kernel array.
2010 * @i: Pointer to iterator that will be initialized on success.
2012 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
2013 * then this function places %NULL in *@iov on return. Otherwise, a new
2014 * array will be allocated and the result placed in *@iov. This means that
2015 * the caller may call kfree() on *@iov regardless of whether the small
2016 * on-stack array was used or not (and regardless of whether this function
2017 * returns an error or not).
2019 * Return: Negative error code on error, bytes imported on success
2021 ssize_t import_iovec(int type, const struct iovec __user *uvec,
2022 unsigned nr_segs, unsigned fast_segs,
2023 struct iovec **iovp, struct iov_iter *i)
2025 return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
2026 in_compat_syscall());
2028 EXPORT_SYMBOL(import_iovec);
2030 int import_single_range(int rw, void __user *buf, size_t len,
2031 struct iovec *iov, struct iov_iter *i)
2033 if (len > MAX_RW_COUNT)
2035 if (unlikely(!access_ok(buf, len)))
2038 iov->iov_base = buf;
2040 iov_iter_init(i, rw, iov, 1, len);
2043 EXPORT_SYMBOL(import_single_range);