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 ubuf and kbuf alike */
20 #define iterate_buf(i, n, base, len, off, __p, STEP) { \
21 size_t __maybe_unused off = 0; \
23 base = __p + i->iov_offset; \
25 i->iov_offset += len; \
29 /* covers iovec and kvec alike */
30 #define iterate_iovec(i, n, base, len, off, __p, STEP) { \
32 size_t skip = i->iov_offset; \
34 len = min(n, __p->iov_len - skip); \
36 base = __p->iov_base + skip; \
41 if (skip < __p->iov_len) \
47 i->iov_offset = skip; \
51 #define iterate_bvec(i, n, base, len, off, p, STEP) { \
53 unsigned skip = i->iov_offset; \
55 unsigned offset = p->bv_offset + skip; \
57 void *kaddr = kmap_local_page(p->bv_page + \
58 offset / PAGE_SIZE); \
59 base = kaddr + offset % PAGE_SIZE; \
60 len = min(min(n, (size_t)(p->bv_len - skip)), \
61 (size_t)(PAGE_SIZE - offset % PAGE_SIZE)); \
63 kunmap_local(kaddr); \
67 if (skip == p->bv_len) { \
75 i->iov_offset = skip; \
79 #define iterate_xarray(i, n, base, len, __off, STEP) { \
82 struct folio *folio; \
83 loff_t start = i->xarray_start + i->iov_offset; \
84 pgoff_t index = start / PAGE_SIZE; \
85 XA_STATE(xas, i->xarray, index); \
87 len = PAGE_SIZE - offset_in_page(start); \
89 xas_for_each(&xas, folio, ULONG_MAX) { \
92 if (xas_retry(&xas, folio)) \
94 if (WARN_ON(xa_is_value(folio))) \
96 if (WARN_ON(folio_test_hugetlb(folio))) \
98 offset = offset_in_folio(folio, start + __off); \
99 while (offset < folio_size(folio)) { \
100 base = kmap_local_folio(folio, offset); \
103 kunmap_local(base); \
107 if (left || n == 0) \
115 i->iov_offset += __off; \
119 #define __iterate_and_advance(i, n, base, len, off, I, K) { \
120 if (unlikely(i->count < n)) \
123 if (likely(iter_is_ubuf(i))) { \
126 iterate_buf(i, n, base, len, off, \
128 } else if (likely(iter_is_iovec(i))) { \
129 const struct iovec *iov = i->iov; \
132 iterate_iovec(i, n, base, len, off, \
134 i->nr_segs -= iov - i->iov; \
136 } else if (iov_iter_is_bvec(i)) { \
137 const struct bio_vec *bvec = i->bvec; \
140 iterate_bvec(i, n, base, len, off, \
142 i->nr_segs -= bvec - i->bvec; \
144 } else if (iov_iter_is_kvec(i)) { \
145 const struct kvec *kvec = i->kvec; \
148 iterate_iovec(i, n, base, len, off, \
150 i->nr_segs -= kvec - i->kvec; \
152 } else if (iov_iter_is_xarray(i)) { \
155 iterate_xarray(i, n, base, len, off, \
161 #define iterate_and_advance(i, n, base, len, off, I, K) \
162 __iterate_and_advance(i, n, base, len, off, I, ((void)(K),0))
164 static int copyout(void __user *to, const void *from, size_t n)
166 if (should_fail_usercopy())
168 if (access_ok(to, n)) {
169 instrument_copy_to_user(to, from, n);
170 n = raw_copy_to_user(to, from, n);
175 static int copyin(void *to, const void __user *from, size_t n)
177 if (should_fail_usercopy())
179 if (access_ok(from, n)) {
180 instrument_copy_from_user(to, from, n);
181 n = raw_copy_from_user(to, from, n);
186 static inline struct pipe_buffer *pipe_buf(const struct pipe_inode_info *pipe,
189 return &pipe->bufs[slot & (pipe->ring_size - 1)];
193 static bool sanity(const struct iov_iter *i)
195 struct pipe_inode_info *pipe = i->pipe;
196 unsigned int p_head = pipe->head;
197 unsigned int p_tail = pipe->tail;
198 unsigned int p_occupancy = pipe_occupancy(p_head, p_tail);
199 unsigned int i_head = i->head;
202 if (i->last_offset) {
203 struct pipe_buffer *p;
204 if (unlikely(p_occupancy == 0))
205 goto Bad; // pipe must be non-empty
206 if (unlikely(i_head != p_head - 1))
207 goto Bad; // must be at the last buffer...
209 p = pipe_buf(pipe, i_head);
210 if (unlikely(p->offset + p->len != abs(i->last_offset)))
211 goto Bad; // ... at the end of segment
213 if (i_head != p_head)
214 goto Bad; // must be right after the last buffer
218 printk(KERN_ERR "idx = %d, offset = %d\n", i_head, i->last_offset);
219 printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n",
220 p_head, p_tail, pipe->ring_size);
221 for (idx = 0; idx < pipe->ring_size; idx++)
222 printk(KERN_ERR "[%p %p %d %d]\n",
224 pipe->bufs[idx].page,
225 pipe->bufs[idx].offset,
226 pipe->bufs[idx].len);
231 #define sanity(i) true
234 static struct page *push_anon(struct pipe_inode_info *pipe, unsigned size)
236 struct page *page = alloc_page(GFP_USER);
238 struct pipe_buffer *buf = pipe_buf(pipe, pipe->head++);
239 *buf = (struct pipe_buffer) {
240 .ops = &default_pipe_buf_ops,
249 static void push_page(struct pipe_inode_info *pipe, struct page *page,
250 unsigned int offset, unsigned int size)
252 struct pipe_buffer *buf = pipe_buf(pipe, pipe->head++);
253 *buf = (struct pipe_buffer) {
254 .ops = &page_cache_pipe_buf_ops,
262 static inline int last_offset(const struct pipe_buffer *buf)
264 if (buf->ops == &default_pipe_buf_ops)
265 return buf->len; // buf->offset is 0 for those
267 return -(buf->offset + buf->len);
270 static struct page *append_pipe(struct iov_iter *i, size_t size,
273 struct pipe_inode_info *pipe = i->pipe;
274 int offset = i->last_offset;
275 struct pipe_buffer *buf;
278 if (offset > 0 && offset < PAGE_SIZE) {
279 // some space in the last buffer; add to it
280 buf = pipe_buf(pipe, pipe->head - 1);
281 size = min_t(size_t, size, PAGE_SIZE - offset);
283 i->last_offset += size;
288 // OK, we need a new buffer
290 size = min_t(size_t, size, PAGE_SIZE);
291 if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
293 page = push_anon(pipe, size);
296 i->head = pipe->head - 1;
297 i->last_offset = size;
302 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
305 struct pipe_inode_info *pipe = i->pipe;
306 unsigned int head = pipe->head;
308 if (unlikely(bytes > i->count))
311 if (unlikely(!bytes))
317 if (offset && i->last_offset == -offset) { // could we merge it?
318 struct pipe_buffer *buf = pipe_buf(pipe, head - 1);
319 if (buf->page == page) {
321 i->last_offset -= bytes;
326 if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
329 push_page(pipe, page, offset, bytes);
330 i->last_offset = -(offset + bytes);
337 * fault_in_iov_iter_readable - fault in iov iterator for reading
339 * @size: maximum length
341 * Fault in one or more iovecs of the given iov_iter, to a maximum length of
342 * @size. For each iovec, fault in each page that constitutes the iovec.
344 * Returns the number of bytes not faulted in (like copy_to_user() and
347 * Always returns 0 for non-userspace iterators.
349 size_t fault_in_iov_iter_readable(const struct iov_iter *i, size_t size)
351 if (iter_is_ubuf(i)) {
352 size_t n = min(size, iov_iter_count(i));
353 n -= fault_in_readable(i->ubuf + i->iov_offset, n);
355 } else if (iter_is_iovec(i)) {
356 size_t count = min(size, iov_iter_count(i));
357 const struct iovec *p;
361 for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) {
362 size_t len = min(count, p->iov_len - skip);
367 ret = fault_in_readable(p->iov_base + skip, len);
376 EXPORT_SYMBOL(fault_in_iov_iter_readable);
379 * fault_in_iov_iter_writeable - fault in iov iterator for writing
381 * @size: maximum length
383 * Faults in the iterator using get_user_pages(), i.e., without triggering
384 * hardware page faults. This is primarily useful when we already know that
385 * some or all of the pages in @i aren't in memory.
387 * Returns the number of bytes not faulted in, like copy_to_user() and
390 * Always returns 0 for non-user-space iterators.
392 size_t fault_in_iov_iter_writeable(const struct iov_iter *i, size_t size)
394 if (iter_is_ubuf(i)) {
395 size_t n = min(size, iov_iter_count(i));
396 n -= fault_in_safe_writeable(i->ubuf + i->iov_offset, n);
398 } else if (iter_is_iovec(i)) {
399 size_t count = min(size, iov_iter_count(i));
400 const struct iovec *p;
404 for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) {
405 size_t len = min(count, p->iov_len - skip);
410 ret = fault_in_safe_writeable(p->iov_base + skip, len);
419 EXPORT_SYMBOL(fault_in_iov_iter_writeable);
421 void iov_iter_init(struct iov_iter *i, unsigned int direction,
422 const struct iovec *iov, unsigned long nr_segs,
425 WARN_ON(direction & ~(READ | WRITE));
426 *i = (struct iov_iter) {
427 .iter_type = ITER_IOVEC,
430 .data_source = direction,
437 EXPORT_SYMBOL(iov_iter_init);
439 // returns the offset in partial buffer (if any)
440 static inline unsigned int pipe_npages(const struct iov_iter *i, int *npages)
442 struct pipe_inode_info *pipe = i->pipe;
443 int used = pipe->head - pipe->tail;
444 int off = i->last_offset;
446 *npages = max((int)pipe->max_usage - used, 0);
448 if (off > 0 && off < PAGE_SIZE) { // anon and not full
455 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
458 unsigned int off, chunk;
460 if (unlikely(bytes > i->count))
462 if (unlikely(!bytes))
468 for (size_t n = bytes; n; n -= chunk) {
469 struct page *page = append_pipe(i, n, &off);
470 chunk = min_t(size_t, n, PAGE_SIZE - off);
473 memcpy_to_page(page, off, addr, chunk);
479 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
480 __wsum sum, size_t off)
482 __wsum next = csum_partial_copy_nocheck(from, to, len);
483 return csum_block_add(sum, next, off);
486 static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
487 struct iov_iter *i, __wsum *sump)
491 unsigned int chunk, r;
493 if (unlikely(bytes > i->count))
495 if (unlikely(!bytes))
502 struct page *page = append_pipe(i, bytes, &r);
507 chunk = min_t(size_t, bytes, PAGE_SIZE - r);
508 p = kmap_local_page(page);
509 sum = csum_and_memcpy(p + r, addr + off, chunk, sum, off);
518 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
520 if (unlikely(iov_iter_is_pipe(i)))
521 return copy_pipe_to_iter(addr, bytes, i);
522 if (user_backed_iter(i))
524 iterate_and_advance(i, bytes, base, len, off,
525 copyout(base, addr + off, len),
526 memcpy(base, addr + off, len)
531 EXPORT_SYMBOL(_copy_to_iter);
533 #ifdef CONFIG_ARCH_HAS_COPY_MC
534 static int copyout_mc(void __user *to, const void *from, size_t n)
536 if (access_ok(to, n)) {
537 instrument_copy_to_user(to, from, n);
538 n = copy_mc_to_user((__force void *) to, from, n);
543 static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes,
547 unsigned int off, chunk;
549 if (unlikely(bytes > i->count))
551 if (unlikely(!bytes))
558 struct page *page = append_pipe(i, bytes, &off);
564 chunk = min_t(size_t, bytes, PAGE_SIZE - off);
565 p = kmap_local_page(page);
566 rem = copy_mc_to_kernel(p + off, addr + xfer, chunk);
572 iov_iter_revert(i, rem);
580 * _copy_mc_to_iter - copy to iter with source memory error exception handling
581 * @addr: source kernel address
582 * @bytes: total transfer length
583 * @i: destination iterator
585 * The pmem driver deploys this for the dax operation
586 * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
587 * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
588 * successfully copied.
590 * The main differences between this and typical _copy_to_iter().
592 * * Typical tail/residue handling after a fault retries the copy
593 * byte-by-byte until the fault happens again. Re-triggering machine
594 * checks is potentially fatal so the implementation uses source
595 * alignment and poison alignment assumptions to avoid re-triggering
596 * hardware exceptions.
598 * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
599 * Compare to copy_to_iter() where only ITER_IOVEC attempts might return
602 * Return: number of bytes copied (may be %0)
604 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
606 if (unlikely(iov_iter_is_pipe(i)))
607 return copy_mc_pipe_to_iter(addr, bytes, i);
608 if (user_backed_iter(i))
610 __iterate_and_advance(i, bytes, base, len, off,
611 copyout_mc(base, addr + off, len),
612 copy_mc_to_kernel(base, addr + off, len)
617 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
618 #endif /* CONFIG_ARCH_HAS_COPY_MC */
620 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
622 if (unlikely(iov_iter_is_pipe(i))) {
626 if (user_backed_iter(i))
628 iterate_and_advance(i, bytes, base, len, off,
629 copyin(addr + off, base, len),
630 memcpy(addr + off, base, len)
635 EXPORT_SYMBOL(_copy_from_iter);
637 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
639 if (unlikely(iov_iter_is_pipe(i))) {
643 iterate_and_advance(i, bytes, base, len, off,
644 __copy_from_user_inatomic_nocache(addr + off, base, len),
645 memcpy(addr + off, base, len)
650 EXPORT_SYMBOL(_copy_from_iter_nocache);
652 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
654 * _copy_from_iter_flushcache - write destination through cpu cache
655 * @addr: destination kernel address
656 * @bytes: total transfer length
657 * @i: source iterator
659 * The pmem driver arranges for filesystem-dax to use this facility via
660 * dax_copy_from_iter() for ensuring that writes to persistent memory
661 * are flushed through the CPU cache. It is differentiated from
662 * _copy_from_iter_nocache() in that guarantees all data is flushed for
663 * all iterator types. The _copy_from_iter_nocache() only attempts to
664 * bypass the cache for the ITER_IOVEC case, and on some archs may use
665 * instructions that strand dirty-data in the cache.
667 * Return: number of bytes copied (may be %0)
669 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
671 if (unlikely(iov_iter_is_pipe(i))) {
675 iterate_and_advance(i, bytes, base, len, off,
676 __copy_from_user_flushcache(addr + off, base, len),
677 memcpy_flushcache(addr + off, base, len)
682 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
685 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
688 size_t v = n + offset;
691 * The general case needs to access the page order in order
692 * to compute the page size.
693 * However, we mostly deal with order-0 pages and thus can
694 * avoid a possible cache line miss for requests that fit all
697 if (n <= v && v <= PAGE_SIZE)
700 head = compound_head(page);
701 v += (page - head) << PAGE_SHIFT;
703 if (likely(n <= v && v <= (page_size(head))))
709 static size_t __copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
712 if (unlikely(iov_iter_is_pipe(i))) {
713 return copy_page_to_iter_pipe(page, offset, bytes, i);
715 void *kaddr = kmap_local_page(page);
716 size_t wanted = _copy_to_iter(kaddr + offset, bytes, i);
722 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
726 if (unlikely(!page_copy_sane(page, offset, bytes)))
728 page += offset / PAGE_SIZE; // first subpage
731 size_t n = __copy_page_to_iter(page, offset,
732 min(bytes, (size_t)PAGE_SIZE - offset), i);
738 if (offset == PAGE_SIZE) {
745 EXPORT_SYMBOL(copy_page_to_iter);
747 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
750 if (page_copy_sane(page, offset, bytes)) {
751 void *kaddr = kmap_local_page(page);
752 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
758 EXPORT_SYMBOL(copy_page_from_iter);
760 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
762 unsigned int chunk, off;
764 if (unlikely(bytes > i->count))
766 if (unlikely(!bytes))
772 for (size_t n = bytes; n; n -= chunk) {
773 struct page *page = append_pipe(i, n, &off);
778 chunk = min_t(size_t, n, PAGE_SIZE - off);
779 p = kmap_local_page(page);
780 memset(p + off, 0, chunk);
786 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
788 if (unlikely(iov_iter_is_pipe(i)))
789 return pipe_zero(bytes, i);
790 iterate_and_advance(i, bytes, base, len, count,
791 clear_user(base, len),
797 EXPORT_SYMBOL(iov_iter_zero);
799 size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes,
802 char *kaddr = kmap_atomic(page), *p = kaddr + offset;
803 if (unlikely(!page_copy_sane(page, offset, bytes))) {
804 kunmap_atomic(kaddr);
807 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
808 kunmap_atomic(kaddr);
812 iterate_and_advance(i, bytes, base, len, off,
813 copyin(p + off, base, len),
814 memcpy(p + off, base, len)
816 kunmap_atomic(kaddr);
819 EXPORT_SYMBOL(copy_page_from_iter_atomic);
821 static void pipe_advance(struct iov_iter *i, size_t size)
823 struct pipe_inode_info *pipe = i->pipe;
824 int off = i->last_offset;
827 pipe_discard_from(pipe, i->start_head); // discard everything
832 struct pipe_buffer *buf = pipe_buf(pipe, i->head);
833 if (off) /* make it relative to the beginning of buffer */
834 size += abs(off) - buf->offset;
835 if (size <= buf->len) {
837 i->last_offset = last_offset(buf);
844 pipe_discard_from(pipe, i->head + 1); // discard everything past this one
847 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
849 const struct bio_vec *bvec, *end;
855 size += i->iov_offset;
857 for (bvec = i->bvec, end = bvec + i->nr_segs; bvec < end; bvec++) {
858 if (likely(size < bvec->bv_len))
860 size -= bvec->bv_len;
862 i->iov_offset = size;
863 i->nr_segs -= bvec - i->bvec;
867 static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
869 const struct iovec *iov, *end;
875 size += i->iov_offset; // from beginning of current segment
876 for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) {
877 if (likely(size < iov->iov_len))
879 size -= iov->iov_len;
881 i->iov_offset = size;
882 i->nr_segs -= iov - i->iov;
886 void iov_iter_advance(struct iov_iter *i, size_t size)
888 if (unlikely(i->count < size))
890 if (likely(iter_is_ubuf(i)) || unlikely(iov_iter_is_xarray(i))) {
891 i->iov_offset += size;
893 } else if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
894 /* iovec and kvec have identical layouts */
895 iov_iter_iovec_advance(i, size);
896 } else if (iov_iter_is_bvec(i)) {
897 iov_iter_bvec_advance(i, size);
898 } else if (iov_iter_is_pipe(i)) {
899 pipe_advance(i, size);
900 } else if (iov_iter_is_discard(i)) {
904 EXPORT_SYMBOL(iov_iter_advance);
906 void iov_iter_revert(struct iov_iter *i, size_t unroll)
910 if (WARN_ON(unroll > MAX_RW_COUNT))
913 if (unlikely(iov_iter_is_pipe(i))) {
914 struct pipe_inode_info *pipe = i->pipe;
915 unsigned int head = pipe->head;
917 while (head > i->start_head) {
918 struct pipe_buffer *b = pipe_buf(pipe, --head);
919 if (unroll < b->len) {
921 i->last_offset = last_offset(b);
926 pipe_buf_release(pipe, b);
933 if (unlikely(iov_iter_is_discard(i)))
935 if (unroll <= i->iov_offset) {
936 i->iov_offset -= unroll;
939 unroll -= i->iov_offset;
940 if (iov_iter_is_xarray(i) || iter_is_ubuf(i)) {
941 BUG(); /* We should never go beyond the start of the specified
942 * range since we might then be straying into pages that
945 } else if (iov_iter_is_bvec(i)) {
946 const struct bio_vec *bvec = i->bvec;
948 size_t n = (--bvec)->bv_len;
952 i->iov_offset = n - unroll;
957 } else { /* same logics for iovec and kvec */
958 const struct iovec *iov = i->iov;
960 size_t n = (--iov)->iov_len;
964 i->iov_offset = n - unroll;
971 EXPORT_SYMBOL(iov_iter_revert);
974 * Return the count of just the current iov_iter segment.
976 size_t iov_iter_single_seg_count(const struct iov_iter *i)
978 if (i->nr_segs > 1) {
979 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
980 return min(i->count, i->iov->iov_len - i->iov_offset);
981 if (iov_iter_is_bvec(i))
982 return min(i->count, i->bvec->bv_len - i->iov_offset);
986 EXPORT_SYMBOL(iov_iter_single_seg_count);
988 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
989 const struct kvec *kvec, unsigned long nr_segs,
992 WARN_ON(direction & ~(READ | WRITE));
993 *i = (struct iov_iter){
994 .iter_type = ITER_KVEC,
995 .data_source = direction,
1002 EXPORT_SYMBOL(iov_iter_kvec);
1004 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1005 const struct bio_vec *bvec, unsigned long nr_segs,
1008 WARN_ON(direction & ~(READ | WRITE));
1009 *i = (struct iov_iter){
1010 .iter_type = ITER_BVEC,
1011 .data_source = direction,
1018 EXPORT_SYMBOL(iov_iter_bvec);
1020 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1021 struct pipe_inode_info *pipe,
1024 BUG_ON(direction != READ);
1025 WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1026 *i = (struct iov_iter){
1027 .iter_type = ITER_PIPE,
1028 .data_source = false,
1031 .start_head = pipe->head,
1036 EXPORT_SYMBOL(iov_iter_pipe);
1039 * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1040 * @i: The iterator to initialise.
1041 * @direction: The direction of the transfer.
1042 * @xarray: The xarray to access.
1043 * @start: The start file position.
1044 * @count: The size of the I/O buffer in bytes.
1046 * Set up an I/O iterator to either draw data out of the pages attached to an
1047 * inode or to inject data into those pages. The pages *must* be prevented
1048 * from evaporation, either by taking a ref on them or locking them by the
1051 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
1052 struct xarray *xarray, loff_t start, size_t count)
1054 BUG_ON(direction & ~1);
1055 *i = (struct iov_iter) {
1056 .iter_type = ITER_XARRAY,
1057 .data_source = direction,
1059 .xarray_start = start,
1064 EXPORT_SYMBOL(iov_iter_xarray);
1067 * iov_iter_discard - Initialise an I/O iterator that discards data
1068 * @i: The iterator to initialise.
1069 * @direction: The direction of the transfer.
1070 * @count: The size of the I/O buffer in bytes.
1072 * Set up an I/O iterator that just discards everything that's written to it.
1073 * It's only available as a READ iterator.
1075 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1077 BUG_ON(direction != READ);
1078 *i = (struct iov_iter){
1079 .iter_type = ITER_DISCARD,
1080 .data_source = false,
1085 EXPORT_SYMBOL(iov_iter_discard);
1087 static bool iov_iter_aligned_iovec(const struct iov_iter *i, unsigned addr_mask,
1090 size_t size = i->count;
1091 size_t skip = i->iov_offset;
1094 for (k = 0; k < i->nr_segs; k++, skip = 0) {
1095 size_t len = i->iov[k].iov_len - skip;
1101 if ((unsigned long)(i->iov[k].iov_base + skip) & addr_mask)
1111 static bool iov_iter_aligned_bvec(const struct iov_iter *i, unsigned addr_mask,
1114 size_t size = i->count;
1115 unsigned skip = i->iov_offset;
1118 for (k = 0; k < i->nr_segs; k++, skip = 0) {
1119 size_t len = i->bvec[k].bv_len - skip;
1125 if ((unsigned long)(i->bvec[k].bv_offset + skip) & addr_mask)
1136 * iov_iter_is_aligned() - Check if the addresses and lengths of each segments
1137 * are aligned to the parameters.
1139 * @i: &struct iov_iter to restore
1140 * @addr_mask: bit mask to check against the iov element's addresses
1141 * @len_mask: bit mask to check against the iov element's lengths
1143 * Return: false if any addresses or lengths intersect with the provided masks
1145 bool iov_iter_is_aligned(const struct iov_iter *i, unsigned addr_mask,
1148 if (likely(iter_is_ubuf(i))) {
1149 if (i->count & len_mask)
1151 if ((unsigned long)(i->ubuf + i->iov_offset) & addr_mask)
1156 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1157 return iov_iter_aligned_iovec(i, addr_mask, len_mask);
1159 if (iov_iter_is_bvec(i))
1160 return iov_iter_aligned_bvec(i, addr_mask, len_mask);
1162 if (iov_iter_is_pipe(i)) {
1163 size_t size = i->count;
1165 if (size & len_mask)
1167 if (size && i->last_offset > 0) {
1168 if (i->last_offset & addr_mask)
1175 if (iov_iter_is_xarray(i)) {
1176 if (i->count & len_mask)
1178 if ((i->xarray_start + i->iov_offset) & addr_mask)
1184 EXPORT_SYMBOL_GPL(iov_iter_is_aligned);
1186 static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
1188 unsigned long res = 0;
1189 size_t size = i->count;
1190 size_t skip = i->iov_offset;
1193 for (k = 0; k < i->nr_segs; k++, skip = 0) {
1194 size_t len = i->iov[k].iov_len - skip;
1196 res |= (unsigned long)i->iov[k].iov_base + skip;
1208 static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
1211 size_t size = i->count;
1212 unsigned skip = i->iov_offset;
1215 for (k = 0; k < i->nr_segs; k++, skip = 0) {
1216 size_t len = i->bvec[k].bv_len - skip;
1217 res |= (unsigned long)i->bvec[k].bv_offset + skip;
1228 unsigned long iov_iter_alignment(const struct iov_iter *i)
1230 if (likely(iter_is_ubuf(i))) {
1231 size_t size = i->count;
1233 return ((unsigned long)i->ubuf + i->iov_offset) | size;
1237 /* iovec and kvec have identical layouts */
1238 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1239 return iov_iter_alignment_iovec(i);
1241 if (iov_iter_is_bvec(i))
1242 return iov_iter_alignment_bvec(i);
1244 if (iov_iter_is_pipe(i)) {
1245 size_t size = i->count;
1247 if (size && i->last_offset > 0)
1248 return size | i->last_offset;
1252 if (iov_iter_is_xarray(i))
1253 return (i->xarray_start + i->iov_offset) | i->count;
1257 EXPORT_SYMBOL(iov_iter_alignment);
1259 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1261 unsigned long res = 0;
1262 unsigned long v = 0;
1263 size_t size = i->count;
1266 if (iter_is_ubuf(i))
1269 if (WARN_ON(!iter_is_iovec(i)))
1272 for (k = 0; k < i->nr_segs; k++) {
1273 if (i->iov[k].iov_len) {
1274 unsigned long base = (unsigned long)i->iov[k].iov_base;
1275 if (v) // if not the first one
1276 res |= base | v; // this start | previous end
1277 v = base + i->iov[k].iov_len;
1278 if (size <= i->iov[k].iov_len)
1280 size -= i->iov[k].iov_len;
1285 EXPORT_SYMBOL(iov_iter_gap_alignment);
1287 static int want_pages_array(struct page ***res, size_t size,
1288 size_t start, unsigned int maxpages)
1290 unsigned int count = DIV_ROUND_UP(size + start, PAGE_SIZE);
1292 if (count > maxpages)
1294 WARN_ON(!count); // caller should've prevented that
1296 *res = kvmalloc_array(count, sizeof(struct page *), GFP_KERNEL);
1303 static ssize_t pipe_get_pages(struct iov_iter *i,
1304 struct page ***pages, size_t maxsize, unsigned maxpages,
1307 unsigned int npages, count, off, chunk;
1314 *start = off = pipe_npages(i, &npages);
1317 count = want_pages_array(pages, maxsize, off, min(npages, maxpages));
1321 for (npages = 0, left = maxsize ; npages < count; npages++, left -= chunk) {
1322 struct page *page = append_pipe(i, left, &off);
1325 chunk = min_t(size_t, left, PAGE_SIZE - off);
1326 get_page(*p++ = page);
1330 return maxsize - left;
1333 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1334 pgoff_t index, unsigned int nr_pages)
1336 XA_STATE(xas, xa, index);
1338 unsigned int ret = 0;
1341 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1342 if (xas_retry(&xas, page))
1345 /* Has the page moved or been split? */
1346 if (unlikely(page != xas_reload(&xas))) {
1351 pages[ret] = find_subpage(page, xas.xa_index);
1352 get_page(pages[ret]);
1353 if (++ret == nr_pages)
1360 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1361 struct page ***pages, size_t maxsize,
1362 unsigned maxpages, size_t *_start_offset)
1364 unsigned nr, offset, count;
1368 pos = i->xarray_start + i->iov_offset;
1369 index = pos >> PAGE_SHIFT;
1370 offset = pos & ~PAGE_MASK;
1371 *_start_offset = offset;
1373 count = want_pages_array(pages, maxsize, offset, maxpages);
1376 nr = iter_xarray_populate_pages(*pages, i->xarray, index, count);
1380 maxsize = min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1381 iov_iter_advance(i, maxsize);
1385 /* must be done on non-empty ITER_UBUF or ITER_IOVEC one */
1386 static unsigned long first_iovec_segment(const struct iov_iter *i, size_t *size)
1391 if (iter_is_ubuf(i))
1392 return (unsigned long)i->ubuf + i->iov_offset;
1394 for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
1395 size_t len = i->iov[k].iov_len - skip;
1401 return (unsigned long)i->iov[k].iov_base + skip;
1403 BUG(); // if it had been empty, we wouldn't get called
1406 /* must be done on non-empty ITER_BVEC one */
1407 static struct page *first_bvec_segment(const struct iov_iter *i,
1408 size_t *size, size_t *start)
1411 size_t skip = i->iov_offset, len;
1413 len = i->bvec->bv_len - skip;
1416 skip += i->bvec->bv_offset;
1417 page = i->bvec->bv_page + skip / PAGE_SIZE;
1418 *start = skip % PAGE_SIZE;
1422 static ssize_t __iov_iter_get_pages_alloc(struct iov_iter *i,
1423 struct page ***pages, size_t maxsize,
1424 unsigned int maxpages, size_t *start)
1428 if (maxsize > i->count)
1432 if (maxsize > MAX_RW_COUNT)
1433 maxsize = MAX_RW_COUNT;
1435 if (likely(user_backed_iter(i))) {
1436 unsigned int gup_flags = 0;
1440 if (iov_iter_rw(i) != WRITE)
1441 gup_flags |= FOLL_WRITE;
1443 gup_flags |= FOLL_NOFAULT;
1445 addr = first_iovec_segment(i, &maxsize);
1446 *start = addr % PAGE_SIZE;
1448 n = want_pages_array(pages, maxsize, *start, maxpages);
1451 res = get_user_pages_fast(addr, n, gup_flags, *pages);
1452 if (unlikely(res <= 0))
1454 maxsize = min_t(size_t, maxsize, res * PAGE_SIZE - *start);
1455 iov_iter_advance(i, maxsize);
1458 if (iov_iter_is_bvec(i)) {
1462 page = first_bvec_segment(i, &maxsize, start);
1463 n = want_pages_array(pages, maxsize, *start, maxpages);
1467 for (int k = 0; k < n; k++)
1468 get_page(p[k] = page + k);
1469 maxsize = min_t(size_t, maxsize, n * PAGE_SIZE - *start);
1470 iov_iter_advance(i, maxsize);
1473 if (iov_iter_is_pipe(i))
1474 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1475 if (iov_iter_is_xarray(i))
1476 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1480 ssize_t iov_iter_get_pages2(struct iov_iter *i,
1481 struct page **pages, size_t maxsize, unsigned maxpages,
1488 return __iov_iter_get_pages_alloc(i, &pages, maxsize, maxpages, start);
1490 EXPORT_SYMBOL(iov_iter_get_pages2);
1492 ssize_t iov_iter_get_pages_alloc2(struct iov_iter *i,
1493 struct page ***pages, size_t maxsize,
1500 len = __iov_iter_get_pages_alloc(i, pages, maxsize, ~0U, start);
1507 EXPORT_SYMBOL(iov_iter_get_pages_alloc2);
1509 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1514 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1518 iterate_and_advance(i, bytes, base, len, off, ({
1519 next = csum_and_copy_from_user(base, addr + off, len);
1520 sum = csum_block_add(sum, next, off);
1523 sum = csum_and_memcpy(addr + off, base, len, sum, off);
1529 EXPORT_SYMBOL(csum_and_copy_from_iter);
1531 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1534 struct csum_state *csstate = _csstate;
1537 if (unlikely(iov_iter_is_discard(i))) {
1538 WARN_ON(1); /* for now */
1542 sum = csum_shift(csstate->csum, csstate->off);
1543 if (unlikely(iov_iter_is_pipe(i)))
1544 bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum);
1545 else iterate_and_advance(i, bytes, base, len, off, ({
1546 next = csum_and_copy_to_user(addr + off, base, len);
1547 sum = csum_block_add(sum, next, off);
1550 sum = csum_and_memcpy(base, addr + off, len, sum, off);
1553 csstate->csum = csum_shift(sum, csstate->off);
1554 csstate->off += bytes;
1557 EXPORT_SYMBOL(csum_and_copy_to_iter);
1559 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1562 #ifdef CONFIG_CRYPTO_HASH
1563 struct ahash_request *hash = hashp;
1564 struct scatterlist sg;
1567 copied = copy_to_iter(addr, bytes, i);
1568 sg_init_one(&sg, addr, copied);
1569 ahash_request_set_crypt(hash, &sg, NULL, copied);
1570 crypto_ahash_update(hash);
1576 EXPORT_SYMBOL(hash_and_copy_to_iter);
1578 static int iov_npages(const struct iov_iter *i, int maxpages)
1580 size_t skip = i->iov_offset, size = i->count;
1581 const struct iovec *p;
1584 for (p = i->iov; size; skip = 0, p++) {
1585 unsigned offs = offset_in_page(p->iov_base + skip);
1586 size_t len = min(p->iov_len - skip, size);
1590 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1591 if (unlikely(npages > maxpages))
1598 static int bvec_npages(const struct iov_iter *i, int maxpages)
1600 size_t skip = i->iov_offset, size = i->count;
1601 const struct bio_vec *p;
1604 for (p = i->bvec; size; skip = 0, p++) {
1605 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1606 size_t len = min(p->bv_len - skip, size);
1609 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1610 if (unlikely(npages > maxpages))
1616 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1618 if (unlikely(!i->count))
1620 if (likely(iter_is_ubuf(i))) {
1621 unsigned offs = offset_in_page(i->ubuf + i->iov_offset);
1622 int npages = DIV_ROUND_UP(offs + i->count, PAGE_SIZE);
1623 return min(npages, maxpages);
1625 /* iovec and kvec have identical layouts */
1626 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1627 return iov_npages(i, maxpages);
1628 if (iov_iter_is_bvec(i))
1629 return bvec_npages(i, maxpages);
1630 if (iov_iter_is_pipe(i)) {
1636 pipe_npages(i, &npages);
1637 return min(npages, maxpages);
1639 if (iov_iter_is_xarray(i)) {
1640 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1641 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1642 return min(npages, maxpages);
1646 EXPORT_SYMBOL(iov_iter_npages);
1648 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1651 if (unlikely(iov_iter_is_pipe(new))) {
1655 if (iov_iter_is_bvec(new))
1656 return new->bvec = kmemdup(new->bvec,
1657 new->nr_segs * sizeof(struct bio_vec),
1659 else if (iov_iter_is_kvec(new) || iter_is_iovec(new))
1660 /* iovec and kvec have identical layout */
1661 return new->iov = kmemdup(new->iov,
1662 new->nr_segs * sizeof(struct iovec),
1666 EXPORT_SYMBOL(dup_iter);
1668 static int copy_compat_iovec_from_user(struct iovec *iov,
1669 const struct iovec __user *uvec, unsigned long nr_segs)
1671 const struct compat_iovec __user *uiov =
1672 (const struct compat_iovec __user *)uvec;
1673 int ret = -EFAULT, i;
1675 if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1678 for (i = 0; i < nr_segs; i++) {
1682 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1683 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1685 /* check for compat_size_t not fitting in compat_ssize_t .. */
1690 iov[i].iov_base = compat_ptr(buf);
1691 iov[i].iov_len = len;
1700 static int copy_iovec_from_user(struct iovec *iov,
1701 const struct iovec __user *uvec, unsigned long nr_segs)
1705 if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1707 for (seg = 0; seg < nr_segs; seg++) {
1708 if ((ssize_t)iov[seg].iov_len < 0)
1715 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1716 unsigned long nr_segs, unsigned long fast_segs,
1717 struct iovec *fast_iov, bool compat)
1719 struct iovec *iov = fast_iov;
1723 * SuS says "The readv() function *may* fail if the iovcnt argument was
1724 * less than or equal to 0, or greater than {IOV_MAX}. Linux has
1725 * traditionally returned zero for zero segments, so...
1729 if (nr_segs > UIO_MAXIOV)
1730 return ERR_PTR(-EINVAL);
1731 if (nr_segs > fast_segs) {
1732 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1734 return ERR_PTR(-ENOMEM);
1738 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1740 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1742 if (iov != fast_iov)
1744 return ERR_PTR(ret);
1750 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1751 unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1752 struct iov_iter *i, bool compat)
1754 ssize_t total_len = 0;
1758 iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1761 return PTR_ERR(iov);
1765 * According to the Single Unix Specification we should return EINVAL if
1766 * an element length is < 0 when cast to ssize_t or if the total length
1767 * would overflow the ssize_t return value of the system call.
1769 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1772 for (seg = 0; seg < nr_segs; seg++) {
1773 ssize_t len = (ssize_t)iov[seg].iov_len;
1775 if (!access_ok(iov[seg].iov_base, len)) {
1782 if (len > MAX_RW_COUNT - total_len) {
1783 len = MAX_RW_COUNT - total_len;
1784 iov[seg].iov_len = len;
1789 iov_iter_init(i, type, iov, nr_segs, total_len);
1798 * import_iovec() - Copy an array of &struct iovec from userspace
1799 * into the kernel, check that it is valid, and initialize a new
1800 * &struct iov_iter iterator to access it.
1802 * @type: One of %READ or %WRITE.
1803 * @uvec: Pointer to the userspace array.
1804 * @nr_segs: Number of elements in userspace array.
1805 * @fast_segs: Number of elements in @iov.
1806 * @iovp: (input and output parameter) Pointer to pointer to (usually small
1807 * on-stack) kernel array.
1808 * @i: Pointer to iterator that will be initialized on success.
1810 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1811 * then this function places %NULL in *@iov on return. Otherwise, a new
1812 * array will be allocated and the result placed in *@iov. This means that
1813 * the caller may call kfree() on *@iov regardless of whether the small
1814 * on-stack array was used or not (and regardless of whether this function
1815 * returns an error or not).
1817 * Return: Negative error code on error, bytes imported on success
1819 ssize_t import_iovec(int type, const struct iovec __user *uvec,
1820 unsigned nr_segs, unsigned fast_segs,
1821 struct iovec **iovp, struct iov_iter *i)
1823 return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
1824 in_compat_syscall());
1826 EXPORT_SYMBOL(import_iovec);
1828 int import_single_range(int rw, void __user *buf, size_t len,
1829 struct iovec *iov, struct iov_iter *i)
1831 if (len > MAX_RW_COUNT)
1833 if (unlikely(!access_ok(buf, len)))
1836 iov->iov_base = buf;
1838 iov_iter_init(i, rw, iov, 1, len);
1841 EXPORT_SYMBOL(import_single_range);
1844 * iov_iter_restore() - Restore a &struct iov_iter to the same state as when
1845 * iov_iter_save_state() was called.
1847 * @i: &struct iov_iter to restore
1848 * @state: state to restore from
1850 * Used after iov_iter_save_state() to bring restore @i, if operations may
1853 * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC
1855 void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state)
1857 if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i)) &&
1858 !iov_iter_is_kvec(i) && !iter_is_ubuf(i))
1860 i->iov_offset = state->iov_offset;
1861 i->count = state->count;
1862 if (iter_is_ubuf(i))
1865 * For the *vec iters, nr_segs + iov is constant - if we increment
1866 * the vec, then we also decrement the nr_segs count. Hence we don't
1867 * need to track both of these, just one is enough and we can deduct
1868 * the other from that. ITER_KVEC and ITER_IOVEC are the same struct
1869 * size, so we can just increment the iov pointer as they are unionzed.
1870 * ITER_BVEC _may_ be the same size on some archs, but on others it is
1871 * not. Be safe and handle it separately.
1873 BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
1874 if (iov_iter_is_bvec(i))
1875 i->bvec -= state->nr_segs - i->nr_segs;
1877 i->iov -= state->nr_segs - i->nr_segs;
1878 i->nr_segs = state->nr_segs;