1 #include <linux/export.h>
2 #include <linux/bvec.h>
4 #include <linux/pagemap.h>
5 #include <linux/slab.h>
6 #include <linux/vmalloc.h>
7 #include <linux/splice.h>
8 #include <net/checksum.h>
10 #define PIPE_PARANOIA /* for now */
12 #define iterate_iovec(i, n, __v, __p, skip, STEP) { \
16 __v.iov_len = min(n, __p->iov_len - skip); \
17 if (likely(__v.iov_len)) { \
18 __v.iov_base = __p->iov_base + skip; \
20 __v.iov_len -= left; \
21 skip += __v.iov_len; \
26 while (unlikely(!left && n)) { \
28 __v.iov_len = min(n, __p->iov_len); \
29 if (unlikely(!__v.iov_len)) \
31 __v.iov_base = __p->iov_base; \
33 __v.iov_len -= left; \
40 #define iterate_kvec(i, n, __v, __p, skip, STEP) { \
43 __v.iov_len = min(n, __p->iov_len - skip); \
44 if (likely(__v.iov_len)) { \
45 __v.iov_base = __p->iov_base + skip; \
47 skip += __v.iov_len; \
50 while (unlikely(n)) { \
52 __v.iov_len = min(n, __p->iov_len); \
53 if (unlikely(!__v.iov_len)) \
55 __v.iov_base = __p->iov_base; \
63 #define iterate_bvec(i, n, __v, __bi, skip, STEP) { \
64 struct bvec_iter __start; \
65 __start.bi_size = n; \
66 __start.bi_bvec_done = skip; \
68 for_each_bvec(__v, i->bvec, __bi, __start) { \
75 #define iterate_all_kinds(i, n, v, I, B, K) { \
77 size_t skip = i->iov_offset; \
78 if (unlikely(i->type & ITER_BVEC)) { \
80 struct bvec_iter __bi; \
81 iterate_bvec(i, n, v, __bi, skip, (B)) \
82 } else if (unlikely(i->type & ITER_KVEC)) { \
83 const struct kvec *kvec; \
85 iterate_kvec(i, n, v, kvec, skip, (K)) \
87 const struct iovec *iov; \
89 iterate_iovec(i, n, v, iov, skip, (I)) \
94 #define iterate_and_advance(i, n, v, I, B, K) { \
95 if (unlikely(i->count < n)) \
98 size_t skip = i->iov_offset; \
99 if (unlikely(i->type & ITER_BVEC)) { \
100 const struct bio_vec *bvec = i->bvec; \
102 struct bvec_iter __bi; \
103 iterate_bvec(i, n, v, __bi, skip, (B)) \
104 i->bvec = __bvec_iter_bvec(i->bvec, __bi); \
105 i->nr_segs -= i->bvec - bvec; \
106 skip = __bi.bi_bvec_done; \
107 } else if (unlikely(i->type & ITER_KVEC)) { \
108 const struct kvec *kvec; \
110 iterate_kvec(i, n, v, kvec, skip, (K)) \
111 if (skip == kvec->iov_len) { \
115 i->nr_segs -= kvec - i->kvec; \
118 const struct iovec *iov; \
120 iterate_iovec(i, n, v, iov, skip, (I)) \
121 if (skip == iov->iov_len) { \
125 i->nr_segs -= iov - i->iov; \
129 i->iov_offset = skip; \
133 static int copyout(void __user *to, const void *from, size_t n)
135 if (access_ok(VERIFY_WRITE, to, n)) {
136 kasan_check_read(from, n);
137 n = raw_copy_to_user(to, from, n);
142 static int copyin(void *to, const void __user *from, size_t n)
144 if (access_ok(VERIFY_READ, from, n)) {
145 kasan_check_write(to, n);
146 n = raw_copy_from_user(to, from, n);
151 static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes,
154 size_t skip, copy, left, wanted;
155 const struct iovec *iov;
159 if (unlikely(bytes > i->count))
162 if (unlikely(!bytes))
168 skip = i->iov_offset;
169 buf = iov->iov_base + skip;
170 copy = min(bytes, iov->iov_len - skip);
172 if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_writeable(buf, copy)) {
173 kaddr = kmap_atomic(page);
174 from = kaddr + offset;
176 /* first chunk, usually the only one */
177 left = copyout(buf, from, copy);
183 while (unlikely(!left && bytes)) {
186 copy = min(bytes, iov->iov_len);
187 left = copyout(buf, from, copy);
193 if (likely(!bytes)) {
194 kunmap_atomic(kaddr);
197 offset = from - kaddr;
199 kunmap_atomic(kaddr);
200 copy = min(bytes, iov->iov_len - skip);
202 /* Too bad - revert to non-atomic kmap */
205 from = kaddr + offset;
206 left = copyout(buf, from, copy);
211 while (unlikely(!left && bytes)) {
214 copy = min(bytes, iov->iov_len);
215 left = copyout(buf, from, copy);
224 if (skip == iov->iov_len) {
228 i->count -= wanted - bytes;
229 i->nr_segs -= iov - i->iov;
231 i->iov_offset = skip;
232 return wanted - bytes;
235 static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes,
238 size_t skip, copy, left, wanted;
239 const struct iovec *iov;
243 if (unlikely(bytes > i->count))
246 if (unlikely(!bytes))
252 skip = i->iov_offset;
253 buf = iov->iov_base + skip;
254 copy = min(bytes, iov->iov_len - skip);
256 if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_readable(buf, copy)) {
257 kaddr = kmap_atomic(page);
260 /* first chunk, usually the only one */
261 left = copyin(to, buf, copy);
267 while (unlikely(!left && bytes)) {
270 copy = min(bytes, iov->iov_len);
271 left = copyin(to, buf, copy);
277 if (likely(!bytes)) {
278 kunmap_atomic(kaddr);
283 kunmap_atomic(kaddr);
284 copy = min(bytes, iov->iov_len - skip);
286 /* Too bad - revert to non-atomic kmap */
290 left = copyin(to, buf, copy);
295 while (unlikely(!left && bytes)) {
298 copy = min(bytes, iov->iov_len);
299 left = copyin(to, buf, copy);
308 if (skip == iov->iov_len) {
312 i->count -= wanted - bytes;
313 i->nr_segs -= iov - i->iov;
315 i->iov_offset = skip;
316 return wanted - bytes;
320 static bool sanity(const struct iov_iter *i)
322 struct pipe_inode_info *pipe = i->pipe;
324 int next = pipe->curbuf + pipe->nrbufs;
326 struct pipe_buffer *p;
327 if (unlikely(!pipe->nrbufs))
328 goto Bad; // pipe must be non-empty
329 if (unlikely(idx != ((next - 1) & (pipe->buffers - 1))))
330 goto Bad; // must be at the last buffer...
332 p = &pipe->bufs[idx];
333 if (unlikely(p->offset + p->len != i->iov_offset))
334 goto Bad; // ... at the end of segment
336 if (idx != (next & (pipe->buffers - 1)))
337 goto Bad; // must be right after the last buffer
341 printk(KERN_ERR "idx = %d, offset = %zd\n", i->idx, i->iov_offset);
342 printk(KERN_ERR "curbuf = %d, nrbufs = %d, buffers = %d\n",
343 pipe->curbuf, pipe->nrbufs, pipe->buffers);
344 for (idx = 0; idx < pipe->buffers; idx++)
345 printk(KERN_ERR "[%p %p %d %d]\n",
347 pipe->bufs[idx].page,
348 pipe->bufs[idx].offset,
349 pipe->bufs[idx].len);
354 #define sanity(i) true
357 static inline int next_idx(int idx, struct pipe_inode_info *pipe)
359 return (idx + 1) & (pipe->buffers - 1);
362 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
365 struct pipe_inode_info *pipe = i->pipe;
366 struct pipe_buffer *buf;
370 if (unlikely(bytes > i->count))
373 if (unlikely(!bytes))
381 buf = &pipe->bufs[idx];
383 if (offset == off && buf->page == page) {
384 /* merge with the last one */
386 i->iov_offset += bytes;
389 idx = next_idx(idx, pipe);
390 buf = &pipe->bufs[idx];
392 if (idx == pipe->curbuf && pipe->nrbufs)
395 buf->ops = &page_cache_pipe_buf_ops;
397 get_page(buf->page = page);
398 buf->offset = offset;
400 i->iov_offset = offset + bytes;
408 * Fault in one or more iovecs of the given iov_iter, to a maximum length of
409 * bytes. For each iovec, fault in each page that constitutes the iovec.
411 * Return 0 on success, or non-zero if the memory could not be accessed (i.e.
412 * because it is an invalid address).
414 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
416 size_t skip = i->iov_offset;
417 const struct iovec *iov;
421 if (iter_is_iovec(i)) {
422 iterate_iovec(i, bytes, v, iov, skip, ({
423 err = fault_in_pages_readable(v.iov_base, v.iov_len);
430 EXPORT_SYMBOL(iov_iter_fault_in_readable);
432 void iov_iter_init(struct iov_iter *i, int direction,
433 const struct iovec *iov, unsigned long nr_segs,
436 /* It will get better. Eventually... */
437 if (uaccess_kernel()) {
438 direction |= ITER_KVEC;
440 i->kvec = (struct kvec *)iov;
445 i->nr_segs = nr_segs;
449 EXPORT_SYMBOL(iov_iter_init);
451 static void memcpy_from_page(char *to, struct page *page, size_t offset, size_t len)
453 char *from = kmap_atomic(page);
454 memcpy(to, from + offset, len);
458 static void memcpy_to_page(struct page *page, size_t offset, const char *from, size_t len)
460 char *to = kmap_atomic(page);
461 memcpy(to + offset, from, len);
465 static void memzero_page(struct page *page, size_t offset, size_t len)
467 char *addr = kmap_atomic(page);
468 memset(addr + offset, 0, len);
472 static inline bool allocated(struct pipe_buffer *buf)
474 return buf->ops == &default_pipe_buf_ops;
477 static inline void data_start(const struct iov_iter *i, int *idxp, size_t *offp)
479 size_t off = i->iov_offset;
481 if (off && (!allocated(&i->pipe->bufs[idx]) || off == PAGE_SIZE)) {
482 idx = next_idx(idx, i->pipe);
489 static size_t push_pipe(struct iov_iter *i, size_t size,
490 int *idxp, size_t *offp)
492 struct pipe_inode_info *pipe = i->pipe;
497 if (unlikely(size > i->count))
503 data_start(i, &idx, &off);
507 left -= PAGE_SIZE - off;
509 pipe->bufs[idx].len += size;
512 pipe->bufs[idx].len = PAGE_SIZE;
513 idx = next_idx(idx, pipe);
515 while (idx != pipe->curbuf || !pipe->nrbufs) {
516 struct page *page = alloc_page(GFP_USER);
520 pipe->bufs[idx].ops = &default_pipe_buf_ops;
521 pipe->bufs[idx].flags = 0;
522 pipe->bufs[idx].page = page;
523 pipe->bufs[idx].offset = 0;
524 if (left <= PAGE_SIZE) {
525 pipe->bufs[idx].len = left;
528 pipe->bufs[idx].len = PAGE_SIZE;
530 idx = next_idx(idx, pipe);
535 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
538 struct pipe_inode_info *pipe = i->pipe;
545 bytes = n = push_pipe(i, bytes, &idx, &off);
548 for ( ; n; idx = next_idx(idx, pipe), off = 0) {
549 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
550 memcpy_to_page(pipe->bufs[idx].page, off, addr, chunk);
552 i->iov_offset = off + chunk;
560 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
562 const char *from = addr;
563 if (unlikely(i->type & ITER_PIPE))
564 return copy_pipe_to_iter(addr, bytes, i);
565 if (iter_is_iovec(i))
567 iterate_and_advance(i, bytes, v,
568 copyout(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
569 memcpy_to_page(v.bv_page, v.bv_offset,
570 (from += v.bv_len) - v.bv_len, v.bv_len),
571 memcpy(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len)
576 EXPORT_SYMBOL(_copy_to_iter);
578 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
581 if (unlikely(i->type & ITER_PIPE)) {
585 if (iter_is_iovec(i))
587 iterate_and_advance(i, bytes, v,
588 copyin((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
589 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
590 v.bv_offset, v.bv_len),
591 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
596 EXPORT_SYMBOL(_copy_from_iter);
598 bool _copy_from_iter_full(void *addr, size_t bytes, struct iov_iter *i)
601 if (unlikely(i->type & ITER_PIPE)) {
605 if (unlikely(i->count < bytes))
608 if (iter_is_iovec(i))
610 iterate_all_kinds(i, bytes, v, ({
611 if (copyin((to += v.iov_len) - v.iov_len,
612 v.iov_base, v.iov_len))
615 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
616 v.bv_offset, v.bv_len),
617 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
620 iov_iter_advance(i, bytes);
623 EXPORT_SYMBOL(_copy_from_iter_full);
625 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
628 if (unlikely(i->type & ITER_PIPE)) {
632 iterate_and_advance(i, bytes, v,
633 __copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
634 v.iov_base, v.iov_len),
635 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
636 v.bv_offset, v.bv_len),
637 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
642 EXPORT_SYMBOL(_copy_from_iter_nocache);
644 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
645 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
648 if (unlikely(i->type & ITER_PIPE)) {
652 iterate_and_advance(i, bytes, v,
653 __copy_from_user_flushcache((to += v.iov_len) - v.iov_len,
654 v.iov_base, v.iov_len),
655 memcpy_page_flushcache((to += v.bv_len) - v.bv_len, v.bv_page,
656 v.bv_offset, v.bv_len),
657 memcpy_flushcache((to += v.iov_len) - v.iov_len, v.iov_base,
663 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
666 bool _copy_from_iter_full_nocache(void *addr, size_t bytes, struct iov_iter *i)
669 if (unlikely(i->type & ITER_PIPE)) {
673 if (unlikely(i->count < bytes))
675 iterate_all_kinds(i, bytes, v, ({
676 if (__copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
677 v.iov_base, v.iov_len))
680 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
681 v.bv_offset, v.bv_len),
682 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
685 iov_iter_advance(i, bytes);
688 EXPORT_SYMBOL(_copy_from_iter_full_nocache);
690 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
693 size_t v = n + offset;
696 * The general case needs to access the page order in order
697 * to compute the page size.
698 * However, we mostly deal with order-0 pages and thus can
699 * avoid a possible cache line miss for requests that fit all
702 if (n <= v && v <= PAGE_SIZE)
705 head = compound_head(page);
706 v += (page - head) << PAGE_SHIFT;
708 if (likely(n <= v && v <= (PAGE_SIZE << compound_order(head))))
714 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
717 if (unlikely(!page_copy_sane(page, offset, bytes)))
719 if (i->type & (ITER_BVEC|ITER_KVEC)) {
720 void *kaddr = kmap_atomic(page);
721 size_t wanted = copy_to_iter(kaddr + offset, bytes, i);
722 kunmap_atomic(kaddr);
724 } else if (likely(!(i->type & ITER_PIPE)))
725 return copy_page_to_iter_iovec(page, offset, bytes, i);
727 return copy_page_to_iter_pipe(page, offset, bytes, i);
729 EXPORT_SYMBOL(copy_page_to_iter);
731 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
734 if (unlikely(!page_copy_sane(page, offset, bytes)))
736 if (unlikely(i->type & ITER_PIPE)) {
740 if (i->type & (ITER_BVEC|ITER_KVEC)) {
741 void *kaddr = kmap_atomic(page);
742 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
743 kunmap_atomic(kaddr);
746 return copy_page_from_iter_iovec(page, offset, bytes, i);
748 EXPORT_SYMBOL(copy_page_from_iter);
750 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
752 struct pipe_inode_info *pipe = i->pipe;
759 bytes = n = push_pipe(i, bytes, &idx, &off);
763 for ( ; n; idx = next_idx(idx, pipe), off = 0) {
764 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
765 memzero_page(pipe->bufs[idx].page, off, chunk);
767 i->iov_offset = off + chunk;
774 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
776 if (unlikely(i->type & ITER_PIPE))
777 return pipe_zero(bytes, i);
778 iterate_and_advance(i, bytes, v,
779 clear_user(v.iov_base, v.iov_len),
780 memzero_page(v.bv_page, v.bv_offset, v.bv_len),
781 memset(v.iov_base, 0, v.iov_len)
786 EXPORT_SYMBOL(iov_iter_zero);
788 size_t iov_iter_copy_from_user_atomic(struct page *page,
789 struct iov_iter *i, unsigned long offset, size_t bytes)
791 char *kaddr = kmap_atomic(page), *p = kaddr + offset;
792 if (unlikely(!page_copy_sane(page, offset, bytes))) {
793 kunmap_atomic(kaddr);
796 if (unlikely(i->type & ITER_PIPE)) {
797 kunmap_atomic(kaddr);
801 iterate_all_kinds(i, bytes, v,
802 copyin((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
803 memcpy_from_page((p += v.bv_len) - v.bv_len, v.bv_page,
804 v.bv_offset, v.bv_len),
805 memcpy((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
807 kunmap_atomic(kaddr);
810 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
812 static inline void pipe_truncate(struct iov_iter *i)
814 struct pipe_inode_info *pipe = i->pipe;
816 size_t off = i->iov_offset;
818 int nrbufs = (idx - pipe->curbuf) & (pipe->buffers - 1);
820 pipe->bufs[idx].len = off - pipe->bufs[idx].offset;
821 idx = next_idx(idx, pipe);
824 while (pipe->nrbufs > nrbufs) {
825 pipe_buf_release(pipe, &pipe->bufs[idx]);
826 idx = next_idx(idx, pipe);
832 static void pipe_advance(struct iov_iter *i, size_t size)
834 struct pipe_inode_info *pipe = i->pipe;
835 if (unlikely(i->count < size))
838 struct pipe_buffer *buf;
839 size_t off = i->iov_offset, left = size;
841 if (off) /* make it relative to the beginning of buffer */
842 left += off - pipe->bufs[idx].offset;
844 buf = &pipe->bufs[idx];
845 if (left <= buf->len)
848 idx = next_idx(idx, pipe);
851 i->iov_offset = buf->offset + left;
854 /* ... and discard everything past that point */
858 void iov_iter_advance(struct iov_iter *i, size_t size)
860 if (unlikely(i->type & ITER_PIPE)) {
861 pipe_advance(i, size);
864 iterate_and_advance(i, size, v, 0, 0, 0)
866 EXPORT_SYMBOL(iov_iter_advance);
868 void iov_iter_revert(struct iov_iter *i, size_t unroll)
872 if (WARN_ON(unroll > MAX_RW_COUNT))
875 if (unlikely(i->type & ITER_PIPE)) {
876 struct pipe_inode_info *pipe = i->pipe;
878 size_t off = i->iov_offset;
880 size_t n = off - pipe->bufs[idx].offset;
886 if (!unroll && idx == i->start_idx) {
891 idx = pipe->buffers - 1;
892 off = pipe->bufs[idx].offset + pipe->bufs[idx].len;
899 if (unroll <= i->iov_offset) {
900 i->iov_offset -= unroll;
903 unroll -= i->iov_offset;
904 if (i->type & ITER_BVEC) {
905 const struct bio_vec *bvec = i->bvec;
907 size_t n = (--bvec)->bv_len;
911 i->iov_offset = n - unroll;
916 } else { /* same logics for iovec and kvec */
917 const struct iovec *iov = i->iov;
919 size_t n = (--iov)->iov_len;
923 i->iov_offset = n - unroll;
930 EXPORT_SYMBOL(iov_iter_revert);
933 * Return the count of just the current iov_iter segment.
935 size_t iov_iter_single_seg_count(const struct iov_iter *i)
937 if (unlikely(i->type & ITER_PIPE))
938 return i->count; // it is a silly place, anyway
941 else if (i->type & ITER_BVEC)
942 return min(i->count, i->bvec->bv_len - i->iov_offset);
944 return min(i->count, i->iov->iov_len - i->iov_offset);
946 EXPORT_SYMBOL(iov_iter_single_seg_count);
948 void iov_iter_kvec(struct iov_iter *i, int direction,
949 const struct kvec *kvec, unsigned long nr_segs,
952 BUG_ON(!(direction & ITER_KVEC));
955 i->nr_segs = nr_segs;
959 EXPORT_SYMBOL(iov_iter_kvec);
961 void iov_iter_bvec(struct iov_iter *i, int direction,
962 const struct bio_vec *bvec, unsigned long nr_segs,
965 BUG_ON(!(direction & ITER_BVEC));
968 i->nr_segs = nr_segs;
972 EXPORT_SYMBOL(iov_iter_bvec);
974 void iov_iter_pipe(struct iov_iter *i, int direction,
975 struct pipe_inode_info *pipe,
978 BUG_ON(direction != ITER_PIPE);
979 WARN_ON(pipe->nrbufs == pipe->buffers);
982 i->idx = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
985 i->start_idx = i->idx;
987 EXPORT_SYMBOL(iov_iter_pipe);
989 unsigned long iov_iter_alignment(const struct iov_iter *i)
991 unsigned long res = 0;
992 size_t size = i->count;
994 if (unlikely(i->type & ITER_PIPE)) {
995 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->idx]))
996 return size | i->iov_offset;
999 iterate_all_kinds(i, size, v,
1000 (res |= (unsigned long)v.iov_base | v.iov_len, 0),
1001 res |= v.bv_offset | v.bv_len,
1002 res |= (unsigned long)v.iov_base | v.iov_len
1006 EXPORT_SYMBOL(iov_iter_alignment);
1008 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1010 unsigned long res = 0;
1011 size_t size = i->count;
1013 if (unlikely(i->type & ITER_PIPE)) {
1018 iterate_all_kinds(i, size, v,
1019 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1020 (size != v.iov_len ? size : 0), 0),
1021 (res |= (!res ? 0 : (unsigned long)v.bv_offset) |
1022 (size != v.bv_len ? size : 0)),
1023 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1024 (size != v.iov_len ? size : 0))
1028 EXPORT_SYMBOL(iov_iter_gap_alignment);
1030 static inline size_t __pipe_get_pages(struct iov_iter *i,
1032 struct page **pages,
1036 struct pipe_inode_info *pipe = i->pipe;
1037 ssize_t n = push_pipe(i, maxsize, &idx, start);
1044 get_page(*pages++ = pipe->bufs[idx].page);
1045 idx = next_idx(idx, pipe);
1052 static ssize_t pipe_get_pages(struct iov_iter *i,
1053 struct page **pages, size_t maxsize, unsigned maxpages,
1066 data_start(i, &idx, start);
1067 /* some of this one + all after this one */
1068 npages = ((i->pipe->curbuf - idx - 1) & (i->pipe->buffers - 1)) + 1;
1069 capacity = min(npages,maxpages) * PAGE_SIZE - *start;
1071 return __pipe_get_pages(i, min(maxsize, capacity), pages, idx, start);
1074 ssize_t iov_iter_get_pages(struct iov_iter *i,
1075 struct page **pages, size_t maxsize, unsigned maxpages,
1078 if (maxsize > i->count)
1081 if (unlikely(i->type & ITER_PIPE))
1082 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1083 iterate_all_kinds(i, maxsize, v, ({
1084 unsigned long addr = (unsigned long)v.iov_base;
1085 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1089 if (len > maxpages * PAGE_SIZE)
1090 len = maxpages * PAGE_SIZE;
1091 addr &= ~(PAGE_SIZE - 1);
1092 n = DIV_ROUND_UP(len, PAGE_SIZE);
1093 res = get_user_pages_fast(addr, n, (i->type & WRITE) != WRITE, pages);
1094 if (unlikely(res < 0))
1096 return (res == n ? len : res * PAGE_SIZE) - *start;
1098 /* can't be more than PAGE_SIZE */
1099 *start = v.bv_offset;
1100 get_page(*pages = v.bv_page);
1108 EXPORT_SYMBOL(iov_iter_get_pages);
1110 static struct page **get_pages_array(size_t n)
1112 return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1115 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1116 struct page ***pages, size_t maxsize,
1130 data_start(i, &idx, start);
1131 /* some of this one + all after this one */
1132 npages = ((i->pipe->curbuf - idx - 1) & (i->pipe->buffers - 1)) + 1;
1133 n = npages * PAGE_SIZE - *start;
1137 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1138 p = get_pages_array(npages);
1141 n = __pipe_get_pages(i, maxsize, p, idx, start);
1149 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1150 struct page ***pages, size_t maxsize,
1155 if (maxsize > i->count)
1158 if (unlikely(i->type & ITER_PIPE))
1159 return pipe_get_pages_alloc(i, pages, maxsize, start);
1160 iterate_all_kinds(i, maxsize, v, ({
1161 unsigned long addr = (unsigned long)v.iov_base;
1162 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1166 addr &= ~(PAGE_SIZE - 1);
1167 n = DIV_ROUND_UP(len, PAGE_SIZE);
1168 p = get_pages_array(n);
1171 res = get_user_pages_fast(addr, n, (i->type & WRITE) != WRITE, p);
1172 if (unlikely(res < 0)) {
1177 return (res == n ? len : res * PAGE_SIZE) - *start;
1179 /* can't be more than PAGE_SIZE */
1180 *start = v.bv_offset;
1181 *pages = p = get_pages_array(1);
1184 get_page(*p = v.bv_page);
1192 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1194 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1201 if (unlikely(i->type & ITER_PIPE)) {
1205 iterate_and_advance(i, bytes, v, ({
1207 next = csum_and_copy_from_user(v.iov_base,
1208 (to += v.iov_len) - v.iov_len,
1209 v.iov_len, 0, &err);
1211 sum = csum_block_add(sum, next, off);
1214 err ? v.iov_len : 0;
1216 char *p = kmap_atomic(v.bv_page);
1217 next = csum_partial_copy_nocheck(p + v.bv_offset,
1218 (to += v.bv_len) - v.bv_len,
1221 sum = csum_block_add(sum, next, off);
1224 next = csum_partial_copy_nocheck(v.iov_base,
1225 (to += v.iov_len) - v.iov_len,
1227 sum = csum_block_add(sum, next, off);
1234 EXPORT_SYMBOL(csum_and_copy_from_iter);
1236 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum,
1243 if (unlikely(i->type & ITER_PIPE)) {
1247 if (unlikely(i->count < bytes))
1249 iterate_all_kinds(i, bytes, v, ({
1251 next = csum_and_copy_from_user(v.iov_base,
1252 (to += v.iov_len) - v.iov_len,
1253 v.iov_len, 0, &err);
1256 sum = csum_block_add(sum, next, off);
1260 char *p = kmap_atomic(v.bv_page);
1261 next = csum_partial_copy_nocheck(p + v.bv_offset,
1262 (to += v.bv_len) - v.bv_len,
1265 sum = csum_block_add(sum, next, off);
1268 next = csum_partial_copy_nocheck(v.iov_base,
1269 (to += v.iov_len) - v.iov_len,
1271 sum = csum_block_add(sum, next, off);
1276 iov_iter_advance(i, bytes);
1279 EXPORT_SYMBOL(csum_and_copy_from_iter_full);
1281 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, __wsum *csum,
1284 const char *from = addr;
1288 if (unlikely(i->type & ITER_PIPE)) {
1289 WARN_ON(1); /* for now */
1292 iterate_and_advance(i, bytes, v, ({
1294 next = csum_and_copy_to_user((from += v.iov_len) - v.iov_len,
1296 v.iov_len, 0, &err);
1298 sum = csum_block_add(sum, next, off);
1301 err ? v.iov_len : 0;
1303 char *p = kmap_atomic(v.bv_page);
1304 next = csum_partial_copy_nocheck((from += v.bv_len) - v.bv_len,
1308 sum = csum_block_add(sum, next, off);
1311 next = csum_partial_copy_nocheck((from += v.iov_len) - v.iov_len,
1314 sum = csum_block_add(sum, next, off);
1321 EXPORT_SYMBOL(csum_and_copy_to_iter);
1323 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1325 size_t size = i->count;
1331 if (unlikely(i->type & ITER_PIPE)) {
1332 struct pipe_inode_info *pipe = i->pipe;
1339 data_start(i, &idx, &off);
1340 /* some of this one + all after this one */
1341 npages = ((pipe->curbuf - idx - 1) & (pipe->buffers - 1)) + 1;
1342 if (npages >= maxpages)
1344 } else iterate_all_kinds(i, size, v, ({
1345 unsigned long p = (unsigned long)v.iov_base;
1346 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1348 if (npages >= maxpages)
1352 if (npages >= maxpages)
1355 unsigned long p = (unsigned long)v.iov_base;
1356 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1358 if (npages >= maxpages)
1364 EXPORT_SYMBOL(iov_iter_npages);
1366 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1369 if (unlikely(new->type & ITER_PIPE)) {
1373 if (new->type & ITER_BVEC)
1374 return new->bvec = kmemdup(new->bvec,
1375 new->nr_segs * sizeof(struct bio_vec),
1378 /* iovec and kvec have identical layout */
1379 return new->iov = kmemdup(new->iov,
1380 new->nr_segs * sizeof(struct iovec),
1383 EXPORT_SYMBOL(dup_iter);
1386 * import_iovec() - Copy an array of &struct iovec from userspace
1387 * into the kernel, check that it is valid, and initialize a new
1388 * &struct iov_iter iterator to access it.
1390 * @type: One of %READ or %WRITE.
1391 * @uvector: Pointer to the userspace array.
1392 * @nr_segs: Number of elements in userspace array.
1393 * @fast_segs: Number of elements in @iov.
1394 * @iov: (input and output parameter) Pointer to pointer to (usually small
1395 * on-stack) kernel array.
1396 * @i: Pointer to iterator that will be initialized on success.
1398 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1399 * then this function places %NULL in *@iov on return. Otherwise, a new
1400 * array will be allocated and the result placed in *@iov. This means that
1401 * the caller may call kfree() on *@iov regardless of whether the small
1402 * on-stack array was used or not (and regardless of whether this function
1403 * returns an error or not).
1405 * Return: 0 on success or negative error code on error.
1407 int import_iovec(int type, const struct iovec __user * uvector,
1408 unsigned nr_segs, unsigned fast_segs,
1409 struct iovec **iov, struct iov_iter *i)
1413 n = rw_copy_check_uvector(type, uvector, nr_segs, fast_segs,
1421 iov_iter_init(i, type, p, nr_segs, n);
1422 *iov = p == *iov ? NULL : p;
1425 EXPORT_SYMBOL(import_iovec);
1427 #ifdef CONFIG_COMPAT
1428 #include <linux/compat.h>
1430 int compat_import_iovec(int type, const struct compat_iovec __user * uvector,
1431 unsigned nr_segs, unsigned fast_segs,
1432 struct iovec **iov, struct iov_iter *i)
1436 n = compat_rw_copy_check_uvector(type, uvector, nr_segs, fast_segs,
1444 iov_iter_init(i, type, p, nr_segs, n);
1445 *iov = p == *iov ? NULL : p;
1450 int import_single_range(int rw, void __user *buf, size_t len,
1451 struct iovec *iov, struct iov_iter *i)
1453 if (len > MAX_RW_COUNT)
1455 if (unlikely(!access_ok(!rw, buf, len)))
1458 iov->iov_base = buf;
1460 iov_iter_init(i, rw, iov, 1, len);
1463 EXPORT_SYMBOL(import_single_range);