1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <crypto/hash.h>
3 #include <linux/export.h>
4 #include <linux/bvec.h>
5 #include <linux/fault-inject-usercopy.h>
7 #include <linux/pagemap.h>
8 #include <linux/highmem.h>
9 #include <linux/slab.h>
10 #include <linux/vmalloc.h>
11 #include <linux/splice.h>
12 #include <linux/compat.h>
13 #include <net/checksum.h>
14 #include <linux/scatterlist.h>
15 #include <linux/instrumented.h>
17 #define PIPE_PARANOIA /* for now */
19 /* covers iovec and kvec alike */
20 #define iterate_iovec(i, n, base, len, off, __p, STEP) { \
22 size_t skip = i->iov_offset; \
24 len = min(n, __p->iov_len - skip); \
26 base = __p->iov_base + skip; \
31 if (skip < __p->iov_len) \
37 i->iov_offset = skip; \
41 #define iterate_bvec(i, n, base, len, off, p, STEP) { \
43 unsigned skip = i->iov_offset; \
45 unsigned offset = p->bv_offset + skip; \
47 void *kaddr = kmap_local_page(p->bv_page + \
48 offset / PAGE_SIZE); \
49 base = kaddr + offset % PAGE_SIZE; \
50 len = min(min(n, (size_t)(p->bv_len - skip)), \
51 (size_t)(PAGE_SIZE - offset % PAGE_SIZE)); \
53 kunmap_local(kaddr); \
57 if (skip == p->bv_len) { \
65 i->iov_offset = skip; \
69 #define iterate_xarray(i, n, base, len, __off, STEP) { \
72 struct page *head = NULL; \
73 loff_t start = i->xarray_start + i->iov_offset; \
74 unsigned offset = start % PAGE_SIZE; \
75 pgoff_t index = start / PAGE_SIZE; \
78 XA_STATE(xas, i->xarray, index); \
81 xas_for_each(&xas, head, ULONG_MAX) { \
83 if (xas_retry(&xas, head)) \
85 if (WARN_ON(xa_is_value(head))) \
87 if (WARN_ON(PageHuge(head))) \
89 for (j = (head->index < index) ? index - head->index : 0; \
90 j < thp_nr_pages(head); j++) { \
91 void *kaddr = kmap_local_page(head + j); \
92 base = kaddr + offset; \
93 len = PAGE_SIZE - offset; \
96 kunmap_local(kaddr); \
100 if (left || n == 0) \
107 i->iov_offset += __off; \
111 #define __iterate_and_advance(i, n, base, len, off, I, K) { \
112 if (unlikely(i->count < n)) \
115 if (likely(iter_is_iovec(i))) { \
116 const struct iovec *iov = i->iov; \
119 iterate_iovec(i, n, base, len, off, \
121 i->nr_segs -= iov - i->iov; \
123 } else if (iov_iter_is_bvec(i)) { \
124 const struct bio_vec *bvec = i->bvec; \
127 iterate_bvec(i, n, base, len, off, \
129 i->nr_segs -= bvec - i->bvec; \
131 } else if (iov_iter_is_kvec(i)) { \
132 const struct kvec *kvec = i->kvec; \
135 iterate_iovec(i, n, base, len, off, \
137 i->nr_segs -= kvec - i->kvec; \
139 } else if (iov_iter_is_xarray(i)) { \
142 iterate_xarray(i, n, base, len, off, \
148 #define iterate_and_advance(i, n, base, len, off, I, K) \
149 __iterate_and_advance(i, n, base, len, off, I, ((void)(K),0))
151 static int copyout(void __user *to, const void *from, size_t n)
153 if (should_fail_usercopy())
155 if (access_ok(to, n)) {
156 instrument_copy_to_user(to, from, n);
157 n = raw_copy_to_user(to, from, n);
162 static int copyin(void *to, const void __user *from, size_t n)
164 if (should_fail_usercopy())
166 if (access_ok(from, n)) {
167 instrument_copy_from_user(to, from, n);
168 n = raw_copy_from_user(to, from, n);
173 static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes,
176 size_t skip, copy, left, wanted;
177 const struct iovec *iov;
181 if (unlikely(bytes > i->count))
184 if (unlikely(!bytes))
190 skip = i->iov_offset;
191 buf = iov->iov_base + skip;
192 copy = min(bytes, iov->iov_len - skip);
194 if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_writeable(buf, copy)) {
195 kaddr = kmap_atomic(page);
196 from = kaddr + offset;
198 /* first chunk, usually the only one */
199 left = copyout(buf, from, copy);
205 while (unlikely(!left && bytes)) {
208 copy = min(bytes, iov->iov_len);
209 left = copyout(buf, from, copy);
215 if (likely(!bytes)) {
216 kunmap_atomic(kaddr);
219 offset = from - kaddr;
221 kunmap_atomic(kaddr);
222 copy = min(bytes, iov->iov_len - skip);
224 /* Too bad - revert to non-atomic kmap */
227 from = kaddr + offset;
228 left = copyout(buf, from, copy);
233 while (unlikely(!left && bytes)) {
236 copy = min(bytes, iov->iov_len);
237 left = copyout(buf, from, copy);
246 if (skip == iov->iov_len) {
250 i->count -= wanted - bytes;
251 i->nr_segs -= iov - i->iov;
253 i->iov_offset = skip;
254 return wanted - bytes;
257 static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes,
260 size_t skip, copy, left, wanted;
261 const struct iovec *iov;
265 if (unlikely(bytes > i->count))
268 if (unlikely(!bytes))
274 skip = i->iov_offset;
275 buf = iov->iov_base + skip;
276 copy = min(bytes, iov->iov_len - skip);
278 if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_readable(buf, copy)) {
279 kaddr = kmap_atomic(page);
282 /* first chunk, usually the only one */
283 left = copyin(to, buf, copy);
289 while (unlikely(!left && bytes)) {
292 copy = min(bytes, iov->iov_len);
293 left = copyin(to, buf, copy);
299 if (likely(!bytes)) {
300 kunmap_atomic(kaddr);
305 kunmap_atomic(kaddr);
306 copy = min(bytes, iov->iov_len - skip);
308 /* Too bad - revert to non-atomic kmap */
312 left = copyin(to, buf, copy);
317 while (unlikely(!left && bytes)) {
320 copy = min(bytes, iov->iov_len);
321 left = copyin(to, buf, copy);
330 if (skip == iov->iov_len) {
334 i->count -= wanted - bytes;
335 i->nr_segs -= iov - i->iov;
337 i->iov_offset = skip;
338 return wanted - bytes;
342 static bool sanity(const struct iov_iter *i)
344 struct pipe_inode_info *pipe = i->pipe;
345 unsigned int p_head = pipe->head;
346 unsigned int p_tail = pipe->tail;
347 unsigned int p_mask = pipe->ring_size - 1;
348 unsigned int p_occupancy = pipe_occupancy(p_head, p_tail);
349 unsigned int i_head = i->head;
353 struct pipe_buffer *p;
354 if (unlikely(p_occupancy == 0))
355 goto Bad; // pipe must be non-empty
356 if (unlikely(i_head != p_head - 1))
357 goto Bad; // must be at the last buffer...
359 p = &pipe->bufs[i_head & p_mask];
360 if (unlikely(p->offset + p->len != i->iov_offset))
361 goto Bad; // ... at the end of segment
363 if (i_head != p_head)
364 goto Bad; // must be right after the last buffer
368 printk(KERN_ERR "idx = %d, offset = %zd\n", i_head, i->iov_offset);
369 printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n",
370 p_head, p_tail, pipe->ring_size);
371 for (idx = 0; idx < pipe->ring_size; idx++)
372 printk(KERN_ERR "[%p %p %d %d]\n",
374 pipe->bufs[idx].page,
375 pipe->bufs[idx].offset,
376 pipe->bufs[idx].len);
381 #define sanity(i) true
384 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
387 struct pipe_inode_info *pipe = i->pipe;
388 struct pipe_buffer *buf;
389 unsigned int p_tail = pipe->tail;
390 unsigned int p_mask = pipe->ring_size - 1;
391 unsigned int i_head = i->head;
394 if (unlikely(bytes > i->count))
397 if (unlikely(!bytes))
404 buf = &pipe->bufs[i_head & p_mask];
406 if (offset == off && buf->page == page) {
407 /* merge with the last one */
409 i->iov_offset += bytes;
413 buf = &pipe->bufs[i_head & p_mask];
415 if (pipe_full(i_head, p_tail, pipe->max_usage))
418 buf->ops = &page_cache_pipe_buf_ops;
421 buf->offset = offset;
424 pipe->head = i_head + 1;
425 i->iov_offset = offset + bytes;
433 * Fault in one or more iovecs of the given iov_iter, to a maximum length of
434 * bytes. For each iovec, fault in each page that constitutes the iovec.
436 * Return 0 on success, or non-zero if the memory could not be accessed (i.e.
437 * because it is an invalid address).
439 int iov_iter_fault_in_readable(const struct iov_iter *i, size_t bytes)
441 if (iter_is_iovec(i)) {
442 const struct iovec *p;
445 if (bytes > i->count)
447 for (p = i->iov, skip = i->iov_offset; bytes; p++, skip = 0) {
448 size_t len = min(bytes, p->iov_len - skip);
453 err = fault_in_pages_readable(p->iov_base + skip, len);
461 EXPORT_SYMBOL(iov_iter_fault_in_readable);
463 void iov_iter_init(struct iov_iter *i, unsigned int direction,
464 const struct iovec *iov, unsigned long nr_segs,
467 WARN_ON(direction & ~(READ | WRITE));
468 WARN_ON_ONCE(uaccess_kernel());
469 *i = (struct iov_iter) {
470 .iter_type = ITER_IOVEC,
471 .data_source = direction,
478 EXPORT_SYMBOL(iov_iter_init);
480 static inline bool allocated(struct pipe_buffer *buf)
482 return buf->ops == &default_pipe_buf_ops;
485 static inline void data_start(const struct iov_iter *i,
486 unsigned int *iter_headp, size_t *offp)
488 unsigned int p_mask = i->pipe->ring_size - 1;
489 unsigned int iter_head = i->head;
490 size_t off = i->iov_offset;
492 if (off && (!allocated(&i->pipe->bufs[iter_head & p_mask]) ||
497 *iter_headp = iter_head;
501 static size_t push_pipe(struct iov_iter *i, size_t size,
502 int *iter_headp, size_t *offp)
504 struct pipe_inode_info *pipe = i->pipe;
505 unsigned int p_tail = pipe->tail;
506 unsigned int p_mask = pipe->ring_size - 1;
507 unsigned int iter_head;
511 if (unlikely(size > i->count))
517 data_start(i, &iter_head, &off);
518 *iter_headp = iter_head;
521 left -= PAGE_SIZE - off;
523 pipe->bufs[iter_head & p_mask].len += size;
526 pipe->bufs[iter_head & p_mask].len = PAGE_SIZE;
529 while (!pipe_full(iter_head, p_tail, pipe->max_usage)) {
530 struct pipe_buffer *buf = &pipe->bufs[iter_head & p_mask];
531 struct page *page = alloc_page(GFP_USER);
535 buf->ops = &default_pipe_buf_ops;
538 buf->len = min_t(ssize_t, left, PAGE_SIZE);
541 pipe->head = iter_head;
549 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
552 struct pipe_inode_info *pipe = i->pipe;
553 unsigned int p_mask = pipe->ring_size - 1;
560 bytes = n = push_pipe(i, bytes, &i_head, &off);
564 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
565 memcpy_to_page(pipe->bufs[i_head & p_mask].page, off, addr, chunk);
567 i->iov_offset = off + chunk;
577 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
578 __wsum sum, size_t off)
580 __wsum next = csum_partial_copy_nocheck(from, to, len);
581 return csum_block_add(sum, next, off);
584 static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
585 struct iov_iter *i, __wsum *sump)
587 struct pipe_inode_info *pipe = i->pipe;
588 unsigned int p_mask = pipe->ring_size - 1;
597 bytes = push_pipe(i, bytes, &i_head, &r);
599 size_t chunk = min_t(size_t, bytes, PAGE_SIZE - r);
600 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
601 sum = csum_and_memcpy(p + r, addr + off, chunk, sum, off);
604 i->iov_offset = r + chunk;
615 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
617 if (unlikely(iov_iter_is_pipe(i)))
618 return copy_pipe_to_iter(addr, bytes, i);
619 if (iter_is_iovec(i))
621 iterate_and_advance(i, bytes, base, len, off,
622 copyout(base, addr + off, len),
623 memcpy(base, addr + off, len)
628 EXPORT_SYMBOL(_copy_to_iter);
630 #ifdef CONFIG_ARCH_HAS_COPY_MC
631 static int copyout_mc(void __user *to, const void *from, size_t n)
633 if (access_ok(to, n)) {
634 instrument_copy_to_user(to, from, n);
635 n = copy_mc_to_user((__force void *) to, from, n);
640 static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes,
643 struct pipe_inode_info *pipe = i->pipe;
644 unsigned int p_mask = pipe->ring_size - 1;
646 size_t n, off, xfer = 0;
651 n = push_pipe(i, bytes, &i_head, &off);
653 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
654 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
656 rem = copy_mc_to_kernel(p + off, addr + xfer, chunk);
660 i->iov_offset = off + chunk;
673 * _copy_mc_to_iter - copy to iter with source memory error exception handling
674 * @addr: source kernel address
675 * @bytes: total transfer length
676 * @iter: destination iterator
678 * The pmem driver deploys this for the dax operation
679 * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
680 * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
681 * successfully copied.
683 * The main differences between this and typical _copy_to_iter().
685 * * Typical tail/residue handling after a fault retries the copy
686 * byte-by-byte until the fault happens again. Re-triggering machine
687 * checks is potentially fatal so the implementation uses source
688 * alignment and poison alignment assumptions to avoid re-triggering
689 * hardware exceptions.
691 * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
692 * Compare to copy_to_iter() where only ITER_IOVEC attempts might return
695 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
697 if (unlikely(iov_iter_is_pipe(i)))
698 return copy_mc_pipe_to_iter(addr, bytes, i);
699 if (iter_is_iovec(i))
701 __iterate_and_advance(i, bytes, base, len, off,
702 copyout_mc(base, addr + off, len),
703 copy_mc_to_kernel(base, addr + off, len)
708 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
709 #endif /* CONFIG_ARCH_HAS_COPY_MC */
711 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
713 if (unlikely(iov_iter_is_pipe(i))) {
717 if (iter_is_iovec(i))
719 iterate_and_advance(i, bytes, base, len, off,
720 copyin(addr + off, base, len),
721 memcpy(addr + off, base, len)
726 EXPORT_SYMBOL(_copy_from_iter);
728 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
730 if (unlikely(iov_iter_is_pipe(i))) {
734 iterate_and_advance(i, bytes, base, len, off,
735 __copy_from_user_inatomic_nocache(addr + off, base, len),
736 memcpy(addr + off, base, len)
741 EXPORT_SYMBOL(_copy_from_iter_nocache);
743 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
745 * _copy_from_iter_flushcache - write destination through cpu cache
746 * @addr: destination kernel address
747 * @bytes: total transfer length
748 * @iter: source iterator
750 * The pmem driver arranges for filesystem-dax to use this facility via
751 * dax_copy_from_iter() for ensuring that writes to persistent memory
752 * are flushed through the CPU cache. It is differentiated from
753 * _copy_from_iter_nocache() in that guarantees all data is flushed for
754 * all iterator types. The _copy_from_iter_nocache() only attempts to
755 * bypass the cache for the ITER_IOVEC case, and on some archs may use
756 * instructions that strand dirty-data in the cache.
758 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
760 if (unlikely(iov_iter_is_pipe(i))) {
764 iterate_and_advance(i, bytes, base, len, off,
765 __copy_from_user_flushcache(addr + off, base, len),
766 memcpy_flushcache(addr + off, base, len)
771 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
774 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
777 size_t v = n + offset;
780 * The general case needs to access the page order in order
781 * to compute the page size.
782 * However, we mostly deal with order-0 pages and thus can
783 * avoid a possible cache line miss for requests that fit all
786 if (n <= v && v <= PAGE_SIZE)
789 head = compound_head(page);
790 v += (page - head) << PAGE_SHIFT;
792 if (likely(n <= v && v <= (page_size(head))))
798 static size_t __copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
801 if (likely(iter_is_iovec(i)))
802 return copy_page_to_iter_iovec(page, offset, bytes, i);
803 if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
804 void *kaddr = kmap_local_page(page);
805 size_t wanted = _copy_to_iter(kaddr + offset, bytes, i);
809 if (iov_iter_is_pipe(i))
810 return copy_page_to_iter_pipe(page, offset, bytes, i);
811 if (unlikely(iov_iter_is_discard(i))) {
812 if (unlikely(i->count < bytes))
821 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
825 if (unlikely(!page_copy_sane(page, offset, bytes)))
827 page += offset / PAGE_SIZE; // first subpage
830 size_t n = __copy_page_to_iter(page, offset,
831 min(bytes, (size_t)PAGE_SIZE - offset), i);
837 if (offset == PAGE_SIZE) {
844 EXPORT_SYMBOL(copy_page_to_iter);
846 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
849 if (unlikely(!page_copy_sane(page, offset, bytes)))
851 if (likely(iter_is_iovec(i)))
852 return copy_page_from_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_from_iter(kaddr + offset, bytes, i);
862 EXPORT_SYMBOL(copy_page_from_iter);
864 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
866 struct pipe_inode_info *pipe = i->pipe;
867 unsigned int p_mask = pipe->ring_size - 1;
874 bytes = n = push_pipe(i, bytes, &i_head, &off);
879 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
880 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
881 memset(p + off, 0, chunk);
884 i->iov_offset = off + chunk;
893 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
895 if (unlikely(iov_iter_is_pipe(i)))
896 return pipe_zero(bytes, i);
897 iterate_and_advance(i, bytes, base, len, count,
898 clear_user(base, len),
904 EXPORT_SYMBOL(iov_iter_zero);
906 size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes,
909 char *kaddr = kmap_atomic(page), *p = kaddr + offset;
910 if (unlikely(!page_copy_sane(page, offset, bytes))) {
911 kunmap_atomic(kaddr);
914 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
915 kunmap_atomic(kaddr);
919 iterate_and_advance(i, bytes, base, len, off,
920 copyin(p + off, base, len),
921 memcpy(p + off, base, len)
923 kunmap_atomic(kaddr);
926 EXPORT_SYMBOL(copy_page_from_iter_atomic);
928 static inline void pipe_truncate(struct iov_iter *i)
930 struct pipe_inode_info *pipe = i->pipe;
931 unsigned int p_tail = pipe->tail;
932 unsigned int p_head = pipe->head;
933 unsigned int p_mask = pipe->ring_size - 1;
935 if (!pipe_empty(p_head, p_tail)) {
936 struct pipe_buffer *buf;
937 unsigned int i_head = i->head;
938 size_t off = i->iov_offset;
941 buf = &pipe->bufs[i_head & p_mask];
942 buf->len = off - buf->offset;
945 while (p_head != i_head) {
947 pipe_buf_release(pipe, &pipe->bufs[p_head & p_mask]);
954 static void pipe_advance(struct iov_iter *i, size_t size)
956 struct pipe_inode_info *pipe = i->pipe;
958 struct pipe_buffer *buf;
959 unsigned int p_mask = pipe->ring_size - 1;
960 unsigned int i_head = i->head;
961 size_t off = i->iov_offset, left = size;
963 if (off) /* make it relative to the beginning of buffer */
964 left += off - pipe->bufs[i_head & p_mask].offset;
966 buf = &pipe->bufs[i_head & p_mask];
967 if (left <= buf->len)
973 i->iov_offset = buf->offset + left;
976 /* ... and discard everything past that point */
980 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
984 bi.bi_size = i->count;
985 bi.bi_bvec_done = i->iov_offset;
987 bvec_iter_advance(i->bvec, &bi, size);
989 i->bvec += bi.bi_idx;
990 i->nr_segs -= bi.bi_idx;
991 i->count = bi.bi_size;
992 i->iov_offset = bi.bi_bvec_done;
995 static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
997 const struct iovec *iov, *end;
1003 size += i->iov_offset; // from beginning of current segment
1004 for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) {
1005 if (likely(size < iov->iov_len))
1007 size -= iov->iov_len;
1009 i->iov_offset = size;
1010 i->nr_segs -= iov - i->iov;
1014 void iov_iter_advance(struct iov_iter *i, size_t size)
1016 if (unlikely(i->count < size))
1018 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
1019 /* iovec and kvec have identical layouts */
1020 iov_iter_iovec_advance(i, size);
1021 } else if (iov_iter_is_bvec(i)) {
1022 iov_iter_bvec_advance(i, size);
1023 } else if (iov_iter_is_pipe(i)) {
1024 pipe_advance(i, size);
1025 } else if (unlikely(iov_iter_is_xarray(i))) {
1026 i->iov_offset += size;
1028 } else if (iov_iter_is_discard(i)) {
1032 EXPORT_SYMBOL(iov_iter_advance);
1034 void iov_iter_revert(struct iov_iter *i, size_t unroll)
1038 if (WARN_ON(unroll > MAX_RW_COUNT))
1041 if (unlikely(iov_iter_is_pipe(i))) {
1042 struct pipe_inode_info *pipe = i->pipe;
1043 unsigned int p_mask = pipe->ring_size - 1;
1044 unsigned int i_head = i->head;
1045 size_t off = i->iov_offset;
1047 struct pipe_buffer *b = &pipe->bufs[i_head & p_mask];
1048 size_t n = off - b->offset;
1054 if (!unroll && i_head == i->start_head) {
1059 b = &pipe->bufs[i_head & p_mask];
1060 off = b->offset + b->len;
1062 i->iov_offset = off;
1067 if (unlikely(iov_iter_is_discard(i)))
1069 if (unroll <= i->iov_offset) {
1070 i->iov_offset -= unroll;
1073 unroll -= i->iov_offset;
1074 if (iov_iter_is_xarray(i)) {
1075 BUG(); /* We should never go beyond the start of the specified
1076 * range since we might then be straying into pages that
1079 } else if (iov_iter_is_bvec(i)) {
1080 const struct bio_vec *bvec = i->bvec;
1082 size_t n = (--bvec)->bv_len;
1086 i->iov_offset = n - unroll;
1091 } else { /* same logics for iovec and kvec */
1092 const struct iovec *iov = i->iov;
1094 size_t n = (--iov)->iov_len;
1098 i->iov_offset = n - unroll;
1105 EXPORT_SYMBOL(iov_iter_revert);
1108 * Return the count of just the current iov_iter segment.
1110 size_t iov_iter_single_seg_count(const struct iov_iter *i)
1112 if (i->nr_segs > 1) {
1113 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1114 return min(i->count, i->iov->iov_len - i->iov_offset);
1115 if (iov_iter_is_bvec(i))
1116 return min(i->count, i->bvec->bv_len - i->iov_offset);
1120 EXPORT_SYMBOL(iov_iter_single_seg_count);
1122 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
1123 const struct kvec *kvec, unsigned long nr_segs,
1126 WARN_ON(direction & ~(READ | WRITE));
1127 *i = (struct iov_iter){
1128 .iter_type = ITER_KVEC,
1129 .data_source = direction,
1136 EXPORT_SYMBOL(iov_iter_kvec);
1138 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1139 const struct bio_vec *bvec, unsigned long nr_segs,
1142 WARN_ON(direction & ~(READ | WRITE));
1143 *i = (struct iov_iter){
1144 .iter_type = ITER_BVEC,
1145 .data_source = direction,
1152 EXPORT_SYMBOL(iov_iter_bvec);
1154 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1155 struct pipe_inode_info *pipe,
1158 BUG_ON(direction != READ);
1159 WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1160 *i = (struct iov_iter){
1161 .iter_type = ITER_PIPE,
1162 .data_source = false,
1165 .start_head = pipe->head,
1170 EXPORT_SYMBOL(iov_iter_pipe);
1173 * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1174 * @i: The iterator to initialise.
1175 * @direction: The direction of the transfer.
1176 * @xarray: The xarray to access.
1177 * @start: The start file position.
1178 * @count: The size of the I/O buffer in bytes.
1180 * Set up an I/O iterator to either draw data out of the pages attached to an
1181 * inode or to inject data into those pages. The pages *must* be prevented
1182 * from evaporation, either by taking a ref on them or locking them by the
1185 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
1186 struct xarray *xarray, loff_t start, size_t count)
1188 BUG_ON(direction & ~1);
1189 *i = (struct iov_iter) {
1190 .iter_type = ITER_XARRAY,
1191 .data_source = direction,
1193 .xarray_start = start,
1198 EXPORT_SYMBOL(iov_iter_xarray);
1201 * iov_iter_discard - Initialise an I/O iterator that discards data
1202 * @i: The iterator to initialise.
1203 * @direction: The direction of the transfer.
1204 * @count: The size of the I/O buffer in bytes.
1206 * Set up an I/O iterator that just discards everything that's written to it.
1207 * It's only available as a READ iterator.
1209 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1211 BUG_ON(direction != READ);
1212 *i = (struct iov_iter){
1213 .iter_type = ITER_DISCARD,
1214 .data_source = false,
1219 EXPORT_SYMBOL(iov_iter_discard);
1221 static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
1223 unsigned long res = 0;
1224 size_t size = i->count;
1225 size_t skip = i->iov_offset;
1228 for (k = 0; k < i->nr_segs; k++, skip = 0) {
1229 size_t len = i->iov[k].iov_len - skip;
1231 res |= (unsigned long)i->iov[k].iov_base + skip;
1243 static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
1246 size_t size = i->count;
1247 unsigned skip = i->iov_offset;
1250 for (k = 0; k < i->nr_segs; k++, skip = 0) {
1251 size_t len = i->bvec[k].bv_len - skip;
1252 res |= (unsigned long)i->bvec[k].bv_offset + skip;
1263 unsigned long iov_iter_alignment(const struct iov_iter *i)
1265 /* iovec and kvec have identical layouts */
1266 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1267 return iov_iter_alignment_iovec(i);
1269 if (iov_iter_is_bvec(i))
1270 return iov_iter_alignment_bvec(i);
1272 if (iov_iter_is_pipe(i)) {
1273 unsigned int p_mask = i->pipe->ring_size - 1;
1274 size_t size = i->count;
1276 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->head & p_mask]))
1277 return size | i->iov_offset;
1281 if (iov_iter_is_xarray(i))
1282 return (i->xarray_start + i->iov_offset) | i->count;
1286 EXPORT_SYMBOL(iov_iter_alignment);
1288 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1290 unsigned long res = 0;
1291 unsigned long v = 0;
1292 size_t size = i->count;
1295 if (WARN_ON(!iter_is_iovec(i)))
1298 for (k = 0; k < i->nr_segs; k++) {
1299 if (i->iov[k].iov_len) {
1300 unsigned long base = (unsigned long)i->iov[k].iov_base;
1301 if (v) // if not the first one
1302 res |= base | v; // this start | previous end
1303 v = base + i->iov[k].iov_len;
1304 if (size <= i->iov[k].iov_len)
1306 size -= i->iov[k].iov_len;
1311 EXPORT_SYMBOL(iov_iter_gap_alignment);
1313 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1315 struct page **pages,
1319 struct pipe_inode_info *pipe = i->pipe;
1320 unsigned int p_mask = pipe->ring_size - 1;
1321 ssize_t n = push_pipe(i, maxsize, &iter_head, start);
1328 get_page(*pages++ = pipe->bufs[iter_head & p_mask].page);
1336 static ssize_t pipe_get_pages(struct iov_iter *i,
1337 struct page **pages, size_t maxsize, unsigned maxpages,
1340 unsigned int iter_head, npages;
1346 data_start(i, &iter_head, start);
1347 /* Amount of free space: some of this one + all after this one */
1348 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1349 capacity = min(npages, maxpages) * PAGE_SIZE - *start;
1351 return __pipe_get_pages(i, min(maxsize, capacity), pages, iter_head, start);
1354 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1355 pgoff_t index, unsigned int nr_pages)
1357 XA_STATE(xas, xa, index);
1359 unsigned int ret = 0;
1362 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1363 if (xas_retry(&xas, page))
1366 /* Has the page moved or been split? */
1367 if (unlikely(page != xas_reload(&xas))) {
1372 pages[ret] = find_subpage(page, xas.xa_index);
1373 get_page(pages[ret]);
1374 if (++ret == nr_pages)
1381 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1382 struct page **pages, size_t maxsize,
1383 unsigned maxpages, size_t *_start_offset)
1385 unsigned nr, offset;
1386 pgoff_t index, count;
1387 size_t size = maxsize, actual;
1390 if (!size || !maxpages)
1393 pos = i->xarray_start + i->iov_offset;
1394 index = pos >> PAGE_SHIFT;
1395 offset = pos & ~PAGE_MASK;
1396 *_start_offset = offset;
1399 if (size > PAGE_SIZE - offset) {
1400 size -= PAGE_SIZE - offset;
1401 count += size >> PAGE_SHIFT;
1407 if (count > maxpages)
1410 nr = iter_xarray_populate_pages(pages, i->xarray, index, count);
1414 actual = PAGE_SIZE * nr;
1416 if (nr == count && size > 0) {
1417 unsigned last_offset = (nr > 1) ? 0 : offset;
1418 actual -= PAGE_SIZE - (last_offset + size);
1423 /* must be done on non-empty ITER_IOVEC one */
1424 static unsigned long first_iovec_segment(const struct iov_iter *i,
1425 size_t *size, size_t *start,
1426 size_t maxsize, unsigned maxpages)
1431 for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
1432 unsigned long addr = (unsigned long)i->iov[k].iov_base + skip;
1433 size_t len = i->iov[k].iov_len - skip;
1439 len += (*start = addr % PAGE_SIZE);
1440 if (len > maxpages * PAGE_SIZE)
1441 len = maxpages * PAGE_SIZE;
1443 return addr & PAGE_MASK;
1445 BUG(); // if it had been empty, we wouldn't get called
1448 /* must be done on non-empty ITER_BVEC one */
1449 static struct page *first_bvec_segment(const struct iov_iter *i,
1450 size_t *size, size_t *start,
1451 size_t maxsize, unsigned maxpages)
1454 size_t skip = i->iov_offset, len;
1456 len = i->bvec->bv_len - skip;
1459 skip += i->bvec->bv_offset;
1460 page = i->bvec->bv_page + skip / PAGE_SIZE;
1461 len += (*start = skip % PAGE_SIZE);
1462 if (len > maxpages * PAGE_SIZE)
1463 len = maxpages * PAGE_SIZE;
1468 ssize_t iov_iter_get_pages(struct iov_iter *i,
1469 struct page **pages, size_t maxsize, unsigned maxpages,
1475 if (maxsize > i->count)
1480 if (likely(iter_is_iovec(i))) {
1483 addr = first_iovec_segment(i, &len, start, maxsize, maxpages);
1484 n = DIV_ROUND_UP(len, PAGE_SIZE);
1485 res = get_user_pages_fast(addr, n,
1486 iov_iter_rw(i) != WRITE ? FOLL_WRITE : 0,
1488 if (unlikely(res < 0))
1490 return (res == n ? len : res * PAGE_SIZE) - *start;
1492 if (iov_iter_is_bvec(i)) {
1495 page = first_bvec_segment(i, &len, start, maxsize, maxpages);
1496 n = DIV_ROUND_UP(len, PAGE_SIZE);
1498 get_page(*pages++ = page++);
1499 return len - *start;
1501 if (iov_iter_is_pipe(i))
1502 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1503 if (iov_iter_is_xarray(i))
1504 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1507 EXPORT_SYMBOL(iov_iter_get_pages);
1509 static struct page **get_pages_array(size_t n)
1511 return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1514 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1515 struct page ***pages, size_t maxsize,
1519 unsigned int iter_head, npages;
1525 data_start(i, &iter_head, start);
1526 /* Amount of free space: some of this one + all after this one */
1527 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1528 n = npages * PAGE_SIZE - *start;
1532 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1533 p = get_pages_array(npages);
1536 n = __pipe_get_pages(i, maxsize, p, iter_head, start);
1544 static ssize_t iter_xarray_get_pages_alloc(struct iov_iter *i,
1545 struct page ***pages, size_t maxsize,
1546 size_t *_start_offset)
1549 unsigned nr, offset;
1550 pgoff_t index, count;
1551 size_t size = maxsize, actual;
1557 pos = i->xarray_start + i->iov_offset;
1558 index = pos >> PAGE_SHIFT;
1559 offset = pos & ~PAGE_MASK;
1560 *_start_offset = offset;
1563 if (size > PAGE_SIZE - offset) {
1564 size -= PAGE_SIZE - offset;
1565 count += size >> PAGE_SHIFT;
1571 p = get_pages_array(count);
1576 nr = iter_xarray_populate_pages(p, i->xarray, index, count);
1580 actual = PAGE_SIZE * nr;
1582 if (nr == count && size > 0) {
1583 unsigned last_offset = (nr > 1) ? 0 : offset;
1584 actual -= PAGE_SIZE - (last_offset + size);
1589 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1590 struct page ***pages, size_t maxsize,
1597 if (maxsize > i->count)
1602 if (likely(iter_is_iovec(i))) {
1605 addr = first_iovec_segment(i, &len, start, maxsize, ~0U);
1606 n = DIV_ROUND_UP(len, PAGE_SIZE);
1607 p = get_pages_array(n);
1610 res = get_user_pages_fast(addr, n,
1611 iov_iter_rw(i) != WRITE ? FOLL_WRITE : 0, p);
1612 if (unlikely(res < 0)) {
1617 return (res == n ? len : res * PAGE_SIZE) - *start;
1619 if (iov_iter_is_bvec(i)) {
1622 page = first_bvec_segment(i, &len, start, maxsize, ~0U);
1623 n = DIV_ROUND_UP(len, PAGE_SIZE);
1624 *pages = p = get_pages_array(n);
1628 get_page(*p++ = page++);
1629 return len - *start;
1631 if (iov_iter_is_pipe(i))
1632 return pipe_get_pages_alloc(i, pages, maxsize, start);
1633 if (iov_iter_is_xarray(i))
1634 return iter_xarray_get_pages_alloc(i, pages, maxsize, start);
1637 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1639 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1644 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1648 iterate_and_advance(i, bytes, base, len, off, ({
1649 next = csum_and_copy_from_user(base, addr + off, len);
1650 sum = csum_block_add(sum, next, off);
1653 sum = csum_and_memcpy(addr + off, base, len, sum, off);
1659 EXPORT_SYMBOL(csum_and_copy_from_iter);
1661 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1664 struct csum_state *csstate = _csstate;
1667 if (unlikely(iov_iter_is_discard(i))) {
1668 WARN_ON(1); /* for now */
1672 sum = csum_shift(csstate->csum, csstate->off);
1673 if (unlikely(iov_iter_is_pipe(i)))
1674 bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum);
1675 else iterate_and_advance(i, bytes, base, len, off, ({
1676 next = csum_and_copy_to_user(addr + off, base, len);
1677 sum = csum_block_add(sum, next, off);
1680 sum = csum_and_memcpy(base, addr + off, len, sum, off);
1683 csstate->csum = csum_shift(sum, csstate->off);
1684 csstate->off += bytes;
1687 EXPORT_SYMBOL(csum_and_copy_to_iter);
1689 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1692 #ifdef CONFIG_CRYPTO_HASH
1693 struct ahash_request *hash = hashp;
1694 struct scatterlist sg;
1697 copied = copy_to_iter(addr, bytes, i);
1698 sg_init_one(&sg, addr, copied);
1699 ahash_request_set_crypt(hash, &sg, NULL, copied);
1700 crypto_ahash_update(hash);
1706 EXPORT_SYMBOL(hash_and_copy_to_iter);
1708 static int iov_npages(const struct iov_iter *i, int maxpages)
1710 size_t skip = i->iov_offset, size = i->count;
1711 const struct iovec *p;
1714 for (p = i->iov; size; skip = 0, p++) {
1715 unsigned offs = offset_in_page(p->iov_base + skip);
1716 size_t len = min(p->iov_len - skip, size);
1720 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1721 if (unlikely(npages > maxpages))
1728 static int bvec_npages(const struct iov_iter *i, int maxpages)
1730 size_t skip = i->iov_offset, size = i->count;
1731 const struct bio_vec *p;
1734 for (p = i->bvec; size; skip = 0, p++) {
1735 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1736 size_t len = min(p->bv_len - skip, size);
1739 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1740 if (unlikely(npages > maxpages))
1746 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1748 if (unlikely(!i->count))
1750 /* iovec and kvec have identical layouts */
1751 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1752 return iov_npages(i, maxpages);
1753 if (iov_iter_is_bvec(i))
1754 return bvec_npages(i, maxpages);
1755 if (iov_iter_is_pipe(i)) {
1756 unsigned int iter_head;
1763 data_start(i, &iter_head, &off);
1764 /* some of this one + all after this one */
1765 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1766 return min(npages, maxpages);
1768 if (iov_iter_is_xarray(i)) {
1769 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1770 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1771 return min(npages, maxpages);
1775 EXPORT_SYMBOL(iov_iter_npages);
1777 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1780 if (unlikely(iov_iter_is_pipe(new))) {
1784 if (unlikely(iov_iter_is_discard(new) || iov_iter_is_xarray(new)))
1786 if (iov_iter_is_bvec(new))
1787 return new->bvec = kmemdup(new->bvec,
1788 new->nr_segs * sizeof(struct bio_vec),
1791 /* iovec and kvec have identical layout */
1792 return new->iov = kmemdup(new->iov,
1793 new->nr_segs * sizeof(struct iovec),
1796 EXPORT_SYMBOL(dup_iter);
1798 static int copy_compat_iovec_from_user(struct iovec *iov,
1799 const struct iovec __user *uvec, unsigned long nr_segs)
1801 const struct compat_iovec __user *uiov =
1802 (const struct compat_iovec __user *)uvec;
1803 int ret = -EFAULT, i;
1805 if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1808 for (i = 0; i < nr_segs; i++) {
1812 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1813 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1815 /* check for compat_size_t not fitting in compat_ssize_t .. */
1820 iov[i].iov_base = compat_ptr(buf);
1821 iov[i].iov_len = len;
1830 static int copy_iovec_from_user(struct iovec *iov,
1831 const struct iovec __user *uvec, unsigned long nr_segs)
1835 if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1837 for (seg = 0; seg < nr_segs; seg++) {
1838 if ((ssize_t)iov[seg].iov_len < 0)
1845 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1846 unsigned long nr_segs, unsigned long fast_segs,
1847 struct iovec *fast_iov, bool compat)
1849 struct iovec *iov = fast_iov;
1853 * SuS says "The readv() function *may* fail if the iovcnt argument was
1854 * less than or equal to 0, or greater than {IOV_MAX}. Linux has
1855 * traditionally returned zero for zero segments, so...
1859 if (nr_segs > UIO_MAXIOV)
1860 return ERR_PTR(-EINVAL);
1861 if (nr_segs > fast_segs) {
1862 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1864 return ERR_PTR(-ENOMEM);
1868 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1870 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1872 if (iov != fast_iov)
1874 return ERR_PTR(ret);
1880 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1881 unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1882 struct iov_iter *i, bool compat)
1884 ssize_t total_len = 0;
1888 iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1891 return PTR_ERR(iov);
1895 * According to the Single Unix Specification we should return EINVAL if
1896 * an element length is < 0 when cast to ssize_t or if the total length
1897 * would overflow the ssize_t return value of the system call.
1899 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1902 for (seg = 0; seg < nr_segs; seg++) {
1903 ssize_t len = (ssize_t)iov[seg].iov_len;
1905 if (!access_ok(iov[seg].iov_base, len)) {
1912 if (len > MAX_RW_COUNT - total_len) {
1913 len = MAX_RW_COUNT - total_len;
1914 iov[seg].iov_len = len;
1919 iov_iter_init(i, type, iov, nr_segs, total_len);
1928 * import_iovec() - Copy an array of &struct iovec from userspace
1929 * into the kernel, check that it is valid, and initialize a new
1930 * &struct iov_iter iterator to access it.
1932 * @type: One of %READ or %WRITE.
1933 * @uvec: Pointer to the userspace array.
1934 * @nr_segs: Number of elements in userspace array.
1935 * @fast_segs: Number of elements in @iov.
1936 * @iovp: (input and output parameter) Pointer to pointer to (usually small
1937 * on-stack) kernel array.
1938 * @i: Pointer to iterator that will be initialized on success.
1940 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1941 * then this function places %NULL in *@iov on return. Otherwise, a new
1942 * array will be allocated and the result placed in *@iov. This means that
1943 * the caller may call kfree() on *@iov regardless of whether the small
1944 * on-stack array was used or not (and regardless of whether this function
1945 * returns an error or not).
1947 * Return: Negative error code on error, bytes imported on success
1949 ssize_t import_iovec(int type, const struct iovec __user *uvec,
1950 unsigned nr_segs, unsigned fast_segs,
1951 struct iovec **iovp, struct iov_iter *i)
1953 return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
1954 in_compat_syscall());
1956 EXPORT_SYMBOL(import_iovec);
1958 int import_single_range(int rw, void __user *buf, size_t len,
1959 struct iovec *iov, struct iov_iter *i)
1961 if (len > MAX_RW_COUNT)
1963 if (unlikely(!access_ok(buf, len)))
1966 iov->iov_base = buf;
1968 iov_iter_init(i, rw, iov, 1, len);
1971 EXPORT_SYMBOL(import_single_range);