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) { \
76 size_t skip = i->iov_offset; \
77 if (unlikely(i->type & ITER_BVEC)) { \
79 struct bvec_iter __bi; \
80 iterate_bvec(i, n, v, __bi, skip, (B)) \
81 } else if (unlikely(i->type & ITER_KVEC)) { \
82 const struct kvec *kvec; \
84 iterate_kvec(i, n, v, kvec, skip, (K)) \
86 const struct iovec *iov; \
88 iterate_iovec(i, n, v, iov, skip, (I)) \
92 #define iterate_and_advance(i, n, v, I, B, K) { \
93 if (unlikely(i->count < n)) \
96 size_t skip = i->iov_offset; \
97 if (unlikely(i->type & ITER_BVEC)) { \
98 const struct bio_vec *bvec = i->bvec; \
100 struct bvec_iter __bi; \
101 iterate_bvec(i, n, v, __bi, skip, (B)) \
102 i->bvec = __bvec_iter_bvec(i->bvec, __bi); \
103 i->nr_segs -= i->bvec - bvec; \
104 skip = __bi.bi_bvec_done; \
105 } else if (unlikely(i->type & ITER_KVEC)) { \
106 const struct kvec *kvec; \
108 iterate_kvec(i, n, v, kvec, skip, (K)) \
109 if (skip == kvec->iov_len) { \
113 i->nr_segs -= kvec - i->kvec; \
116 const struct iovec *iov; \
118 iterate_iovec(i, n, v, iov, skip, (I)) \
119 if (skip == iov->iov_len) { \
123 i->nr_segs -= iov - i->iov; \
127 i->iov_offset = skip; \
131 static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes,
134 size_t skip, copy, left, wanted;
135 const struct iovec *iov;
139 if (unlikely(bytes > i->count))
142 if (unlikely(!bytes))
147 skip = i->iov_offset;
148 buf = iov->iov_base + skip;
149 copy = min(bytes, iov->iov_len - skip);
151 if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_writeable(buf, copy)) {
152 kaddr = kmap_atomic(page);
153 from = kaddr + offset;
155 /* first chunk, usually the only one */
156 left = __copy_to_user_inatomic(buf, from, copy);
162 while (unlikely(!left && bytes)) {
165 copy = min(bytes, iov->iov_len);
166 left = __copy_to_user_inatomic(buf, from, copy);
172 if (likely(!bytes)) {
173 kunmap_atomic(kaddr);
176 offset = from - kaddr;
178 kunmap_atomic(kaddr);
179 copy = min(bytes, iov->iov_len - skip);
181 /* Too bad - revert to non-atomic kmap */
184 from = kaddr + offset;
185 left = __copy_to_user(buf, from, copy);
190 while (unlikely(!left && bytes)) {
193 copy = min(bytes, iov->iov_len);
194 left = __copy_to_user(buf, from, copy);
203 if (skip == iov->iov_len) {
207 i->count -= wanted - bytes;
208 i->nr_segs -= iov - i->iov;
210 i->iov_offset = skip;
211 return wanted - bytes;
214 static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes,
217 size_t skip, copy, left, wanted;
218 const struct iovec *iov;
222 if (unlikely(bytes > i->count))
225 if (unlikely(!bytes))
230 skip = i->iov_offset;
231 buf = iov->iov_base + skip;
232 copy = min(bytes, iov->iov_len - skip);
234 if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_readable(buf, copy)) {
235 kaddr = kmap_atomic(page);
238 /* first chunk, usually the only one */
239 left = __copy_from_user_inatomic(to, buf, copy);
245 while (unlikely(!left && bytes)) {
248 copy = min(bytes, iov->iov_len);
249 left = __copy_from_user_inatomic(to, buf, copy);
255 if (likely(!bytes)) {
256 kunmap_atomic(kaddr);
261 kunmap_atomic(kaddr);
262 copy = min(bytes, iov->iov_len - skip);
264 /* Too bad - revert to non-atomic kmap */
268 left = __copy_from_user(to, buf, copy);
273 while (unlikely(!left && bytes)) {
276 copy = min(bytes, iov->iov_len);
277 left = __copy_from_user(to, buf, copy);
286 if (skip == iov->iov_len) {
290 i->count -= wanted - bytes;
291 i->nr_segs -= iov - i->iov;
293 i->iov_offset = skip;
294 return wanted - bytes;
298 static bool sanity(const struct iov_iter *i)
300 struct pipe_inode_info *pipe = i->pipe;
302 int next = pipe->curbuf + pipe->nrbufs;
304 struct pipe_buffer *p;
305 if (unlikely(!pipe->nrbufs))
306 goto Bad; // pipe must be non-empty
307 if (unlikely(idx != ((next - 1) & (pipe->buffers - 1))))
308 goto Bad; // must be at the last buffer...
310 p = &pipe->bufs[idx];
311 if (unlikely(p->offset + p->len != i->iov_offset))
312 goto Bad; // ... at the end of segment
314 if (idx != (next & (pipe->buffers - 1)))
315 goto Bad; // must be right after the last buffer
319 printk(KERN_ERR "idx = %d, offset = %zd\n", i->idx, i->iov_offset);
320 printk(KERN_ERR "curbuf = %d, nrbufs = %d, buffers = %d\n",
321 pipe->curbuf, pipe->nrbufs, pipe->buffers);
322 for (idx = 0; idx < pipe->buffers; idx++)
323 printk(KERN_ERR "[%p %p %d %d]\n",
325 pipe->bufs[idx].page,
326 pipe->bufs[idx].offset,
327 pipe->bufs[idx].len);
332 #define sanity(i) true
335 static inline int next_idx(int idx, struct pipe_inode_info *pipe)
337 return (idx + 1) & (pipe->buffers - 1);
340 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
343 struct pipe_inode_info *pipe = i->pipe;
344 struct pipe_buffer *buf;
348 if (unlikely(bytes > i->count))
351 if (unlikely(!bytes))
359 buf = &pipe->bufs[idx];
361 if (offset == off && buf->page == page) {
362 /* merge with the last one */
364 i->iov_offset += bytes;
367 idx = next_idx(idx, pipe);
368 buf = &pipe->bufs[idx];
370 if (idx == pipe->curbuf && pipe->nrbufs)
373 buf->ops = &page_cache_pipe_buf_ops;
374 get_page(buf->page = page);
375 buf->offset = offset;
377 i->iov_offset = offset + bytes;
385 * Fault in one or more iovecs of the given iov_iter, to a maximum length of
386 * bytes. For each iovec, fault in each page that constitutes the iovec.
388 * Return 0 on success, or non-zero if the memory could not be accessed (i.e.
389 * because it is an invalid address).
391 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
393 size_t skip = i->iov_offset;
394 const struct iovec *iov;
398 if (!(i->type & (ITER_BVEC|ITER_KVEC))) {
399 iterate_iovec(i, bytes, v, iov, skip, ({
400 err = fault_in_pages_readable(v.iov_base, v.iov_len);
407 EXPORT_SYMBOL(iov_iter_fault_in_readable);
409 void iov_iter_init(struct iov_iter *i, int direction,
410 const struct iovec *iov, unsigned long nr_segs,
413 /* It will get better. Eventually... */
414 if (segment_eq(get_fs(), KERNEL_DS)) {
415 direction |= ITER_KVEC;
417 i->kvec = (struct kvec *)iov;
422 i->nr_segs = nr_segs;
426 EXPORT_SYMBOL(iov_iter_init);
428 static void memcpy_from_page(char *to, struct page *page, size_t offset, size_t len)
430 char *from = kmap_atomic(page);
431 memcpy(to, from + offset, len);
435 static void memcpy_to_page(struct page *page, size_t offset, const char *from, size_t len)
437 char *to = kmap_atomic(page);
438 memcpy(to + offset, from, len);
442 static void memzero_page(struct page *page, size_t offset, size_t len)
444 char *addr = kmap_atomic(page);
445 memset(addr + offset, 0, len);
449 static inline bool allocated(struct pipe_buffer *buf)
451 return buf->ops == &default_pipe_buf_ops;
454 static inline void data_start(const struct iov_iter *i, int *idxp, size_t *offp)
456 size_t off = i->iov_offset;
458 if (off && (!allocated(&i->pipe->bufs[idx]) || off == PAGE_SIZE)) {
459 idx = next_idx(idx, i->pipe);
466 static size_t push_pipe(struct iov_iter *i, size_t size,
467 int *idxp, size_t *offp)
469 struct pipe_inode_info *pipe = i->pipe;
474 if (unlikely(size > i->count))
480 data_start(i, &idx, &off);
484 left -= PAGE_SIZE - off;
486 pipe->bufs[idx].len += size;
489 pipe->bufs[idx].len = PAGE_SIZE;
490 idx = next_idx(idx, pipe);
492 while (idx != pipe->curbuf || !pipe->nrbufs) {
493 struct page *page = alloc_page(GFP_USER);
497 pipe->bufs[idx].ops = &default_pipe_buf_ops;
498 pipe->bufs[idx].page = page;
499 pipe->bufs[idx].offset = 0;
500 if (left <= PAGE_SIZE) {
501 pipe->bufs[idx].len = left;
504 pipe->bufs[idx].len = PAGE_SIZE;
506 idx = next_idx(idx, pipe);
511 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
514 struct pipe_inode_info *pipe = i->pipe;
521 bytes = n = push_pipe(i, bytes, &idx, &off);
524 for ( ; n; idx = next_idx(idx, pipe), off = 0) {
525 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
526 memcpy_to_page(pipe->bufs[idx].page, off, addr, chunk);
528 i->iov_offset = off + chunk;
536 size_t copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
538 const char *from = addr;
539 if (unlikely(i->type & ITER_PIPE))
540 return copy_pipe_to_iter(addr, bytes, i);
541 iterate_and_advance(i, bytes, v,
542 __copy_to_user(v.iov_base, (from += v.iov_len) - v.iov_len,
544 memcpy_to_page(v.bv_page, v.bv_offset,
545 (from += v.bv_len) - v.bv_len, v.bv_len),
546 memcpy(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len)
551 EXPORT_SYMBOL(copy_to_iter);
553 size_t copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
556 if (unlikely(i->type & ITER_PIPE)) {
560 iterate_and_advance(i, bytes, v,
561 __copy_from_user((to += v.iov_len) - v.iov_len, v.iov_base,
563 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
564 v.bv_offset, v.bv_len),
565 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
570 EXPORT_SYMBOL(copy_from_iter);
572 bool copy_from_iter_full(void *addr, size_t bytes, struct iov_iter *i)
575 if (unlikely(i->type & ITER_PIPE)) {
579 if (unlikely(i->count < bytes)) \
582 iterate_all_kinds(i, bytes, v, ({
583 if (__copy_from_user((to += v.iov_len) - v.iov_len,
584 v.iov_base, v.iov_len))
587 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
588 v.bv_offset, v.bv_len),
589 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
592 iov_iter_advance(i, bytes);
595 EXPORT_SYMBOL(copy_from_iter_full);
597 size_t copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
600 if (unlikely(i->type & ITER_PIPE)) {
604 iterate_and_advance(i, bytes, v,
605 __copy_from_user_nocache((to += v.iov_len) - v.iov_len,
606 v.iov_base, v.iov_len),
607 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
608 v.bv_offset, v.bv_len),
609 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
614 EXPORT_SYMBOL(copy_from_iter_nocache);
616 bool copy_from_iter_full_nocache(void *addr, size_t bytes, struct iov_iter *i)
619 if (unlikely(i->type & ITER_PIPE)) {
623 if (unlikely(i->count < bytes)) \
625 iterate_all_kinds(i, bytes, v, ({
626 if (__copy_from_user_nocache((to += v.iov_len) - v.iov_len,
627 v.iov_base, v.iov_len))
630 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
631 v.bv_offset, v.bv_len),
632 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
635 iov_iter_advance(i, bytes);
638 EXPORT_SYMBOL(copy_from_iter_full_nocache);
640 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
643 if (i->type & (ITER_BVEC|ITER_KVEC)) {
644 void *kaddr = kmap_atomic(page);
645 size_t wanted = copy_to_iter(kaddr + offset, bytes, i);
646 kunmap_atomic(kaddr);
648 } else if (likely(!(i->type & ITER_PIPE)))
649 return copy_page_to_iter_iovec(page, offset, bytes, i);
651 return copy_page_to_iter_pipe(page, offset, bytes, i);
653 EXPORT_SYMBOL(copy_page_to_iter);
655 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
658 if (unlikely(i->type & ITER_PIPE)) {
662 if (i->type & (ITER_BVEC|ITER_KVEC)) {
663 void *kaddr = kmap_atomic(page);
664 size_t wanted = copy_from_iter(kaddr + offset, bytes, i);
665 kunmap_atomic(kaddr);
668 return copy_page_from_iter_iovec(page, offset, bytes, i);
670 EXPORT_SYMBOL(copy_page_from_iter);
672 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
674 struct pipe_inode_info *pipe = i->pipe;
681 bytes = n = push_pipe(i, bytes, &idx, &off);
685 for ( ; n; idx = next_idx(idx, pipe), off = 0) {
686 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
687 memzero_page(pipe->bufs[idx].page, off, chunk);
689 i->iov_offset = off + chunk;
696 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
698 if (unlikely(i->type & ITER_PIPE))
699 return pipe_zero(bytes, i);
700 iterate_and_advance(i, bytes, v,
701 __clear_user(v.iov_base, v.iov_len),
702 memzero_page(v.bv_page, v.bv_offset, v.bv_len),
703 memset(v.iov_base, 0, v.iov_len)
708 EXPORT_SYMBOL(iov_iter_zero);
710 size_t iov_iter_copy_from_user_atomic(struct page *page,
711 struct iov_iter *i, unsigned long offset, size_t bytes)
713 char *kaddr = kmap_atomic(page), *p = kaddr + offset;
714 if (unlikely(i->type & ITER_PIPE)) {
715 kunmap_atomic(kaddr);
719 iterate_all_kinds(i, bytes, v,
720 __copy_from_user_inatomic((p += v.iov_len) - v.iov_len,
721 v.iov_base, v.iov_len),
722 memcpy_from_page((p += v.bv_len) - v.bv_len, v.bv_page,
723 v.bv_offset, v.bv_len),
724 memcpy((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
726 kunmap_atomic(kaddr);
729 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
731 static void pipe_advance(struct iov_iter *i, size_t size)
733 struct pipe_inode_info *pipe = i->pipe;
734 struct pipe_buffer *buf;
736 size_t off = i->iov_offset, orig_sz;
738 if (unlikely(i->count < size))
743 if (off) /* make it relative to the beginning of buffer */
744 size += off - pipe->bufs[idx].offset;
746 buf = &pipe->bufs[idx];
747 if (size <= buf->len)
750 idx = next_idx(idx, pipe);
754 off = i->iov_offset = buf->offset + size;
757 idx = next_idx(idx, pipe);
759 int unused = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
760 /* [curbuf,unused) is in use. Free [idx,unused) */
761 while (idx != unused) {
762 pipe_buf_release(pipe, &pipe->bufs[idx]);
763 idx = next_idx(idx, pipe);
770 void iov_iter_advance(struct iov_iter *i, size_t size)
772 if (unlikely(i->type & ITER_PIPE)) {
773 pipe_advance(i, size);
776 iterate_and_advance(i, size, v, 0, 0, 0)
778 EXPORT_SYMBOL(iov_iter_advance);
781 * Return the count of just the current iov_iter segment.
783 size_t iov_iter_single_seg_count(const struct iov_iter *i)
785 if (unlikely(i->type & ITER_PIPE))
786 return i->count; // it is a silly place, anyway
789 else if (i->type & ITER_BVEC)
790 return min(i->count, i->bvec->bv_len - i->iov_offset);
792 return min(i->count, i->iov->iov_len - i->iov_offset);
794 EXPORT_SYMBOL(iov_iter_single_seg_count);
796 void iov_iter_kvec(struct iov_iter *i, int direction,
797 const struct kvec *kvec, unsigned long nr_segs,
800 BUG_ON(!(direction & ITER_KVEC));
803 i->nr_segs = nr_segs;
807 EXPORT_SYMBOL(iov_iter_kvec);
809 void iov_iter_bvec(struct iov_iter *i, int direction,
810 const struct bio_vec *bvec, unsigned long nr_segs,
813 BUG_ON(!(direction & ITER_BVEC));
816 i->nr_segs = nr_segs;
820 EXPORT_SYMBOL(iov_iter_bvec);
822 void iov_iter_pipe(struct iov_iter *i, int direction,
823 struct pipe_inode_info *pipe,
826 BUG_ON(direction != ITER_PIPE);
829 i->idx = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
833 EXPORT_SYMBOL(iov_iter_pipe);
835 unsigned long iov_iter_alignment(const struct iov_iter *i)
837 unsigned long res = 0;
838 size_t size = i->count;
843 if (unlikely(i->type & ITER_PIPE)) {
844 if (i->iov_offset && allocated(&i->pipe->bufs[i->idx]))
845 return size | i->iov_offset;
848 iterate_all_kinds(i, size, v,
849 (res |= (unsigned long)v.iov_base | v.iov_len, 0),
850 res |= v.bv_offset | v.bv_len,
851 res |= (unsigned long)v.iov_base | v.iov_len
855 EXPORT_SYMBOL(iov_iter_alignment);
857 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
859 unsigned long res = 0;
860 size_t size = i->count;
864 if (unlikely(i->type & ITER_PIPE)) {
869 iterate_all_kinds(i, size, v,
870 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
871 (size != v.iov_len ? size : 0), 0),
872 (res |= (!res ? 0 : (unsigned long)v.bv_offset) |
873 (size != v.bv_len ? size : 0)),
874 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
875 (size != v.iov_len ? size : 0))
879 EXPORT_SYMBOL(iov_iter_gap_alignment);
881 static inline size_t __pipe_get_pages(struct iov_iter *i,
887 struct pipe_inode_info *pipe = i->pipe;
888 ssize_t n = push_pipe(i, maxsize, &idx, start);
895 get_page(*pages++ = pipe->bufs[idx].page);
896 idx = next_idx(idx, pipe);
903 static ssize_t pipe_get_pages(struct iov_iter *i,
904 struct page **pages, size_t maxsize, unsigned maxpages,
914 data_start(i, &idx, start);
915 /* some of this one + all after this one */
916 npages = ((i->pipe->curbuf - idx - 1) & (i->pipe->buffers - 1)) + 1;
917 capacity = min(npages,maxpages) * PAGE_SIZE - *start;
919 return __pipe_get_pages(i, min(maxsize, capacity), pages, idx, start);
922 ssize_t iov_iter_get_pages(struct iov_iter *i,
923 struct page **pages, size_t maxsize, unsigned maxpages,
926 if (maxsize > i->count)
932 if (unlikely(i->type & ITER_PIPE))
933 return pipe_get_pages(i, pages, maxsize, maxpages, start);
934 iterate_all_kinds(i, maxsize, v, ({
935 unsigned long addr = (unsigned long)v.iov_base;
936 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
940 if (len > maxpages * PAGE_SIZE)
941 len = maxpages * PAGE_SIZE;
942 addr &= ~(PAGE_SIZE - 1);
943 n = DIV_ROUND_UP(len, PAGE_SIZE);
944 res = get_user_pages_fast(addr, n, (i->type & WRITE) != WRITE, pages);
945 if (unlikely(res < 0))
947 return (res == n ? len : res * PAGE_SIZE) - *start;
949 /* can't be more than PAGE_SIZE */
950 *start = v.bv_offset;
951 get_page(*pages = v.bv_page);
959 EXPORT_SYMBOL(iov_iter_get_pages);
961 static struct page **get_pages_array(size_t n)
963 struct page **p = kmalloc(n * sizeof(struct page *), GFP_KERNEL);
965 p = vmalloc(n * sizeof(struct page *));
969 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
970 struct page ***pages, size_t maxsize,
981 data_start(i, &idx, start);
982 /* some of this one + all after this one */
983 npages = ((i->pipe->curbuf - idx - 1) & (i->pipe->buffers - 1)) + 1;
984 n = npages * PAGE_SIZE - *start;
988 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
989 p = get_pages_array(npages);
992 n = __pipe_get_pages(i, maxsize, p, idx, start);
1000 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1001 struct page ***pages, size_t maxsize,
1006 if (maxsize > i->count)
1012 if (unlikely(i->type & ITER_PIPE))
1013 return pipe_get_pages_alloc(i, pages, maxsize, start);
1014 iterate_all_kinds(i, maxsize, v, ({
1015 unsigned long addr = (unsigned long)v.iov_base;
1016 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1020 addr &= ~(PAGE_SIZE - 1);
1021 n = DIV_ROUND_UP(len, PAGE_SIZE);
1022 p = get_pages_array(n);
1025 res = get_user_pages_fast(addr, n, (i->type & WRITE) != WRITE, p);
1026 if (unlikely(res < 0)) {
1031 return (res == n ? len : res * PAGE_SIZE) - *start;
1033 /* can't be more than PAGE_SIZE */
1034 *start = v.bv_offset;
1035 *pages = p = get_pages_array(1);
1038 get_page(*p = v.bv_page);
1046 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1048 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1055 if (unlikely(i->type & ITER_PIPE)) {
1059 iterate_and_advance(i, bytes, v, ({
1061 next = csum_and_copy_from_user(v.iov_base,
1062 (to += v.iov_len) - v.iov_len,
1063 v.iov_len, 0, &err);
1065 sum = csum_block_add(sum, next, off);
1068 err ? v.iov_len : 0;
1070 char *p = kmap_atomic(v.bv_page);
1071 next = csum_partial_copy_nocheck(p + v.bv_offset,
1072 (to += v.bv_len) - v.bv_len,
1075 sum = csum_block_add(sum, next, off);
1078 next = csum_partial_copy_nocheck(v.iov_base,
1079 (to += v.iov_len) - v.iov_len,
1081 sum = csum_block_add(sum, next, off);
1088 EXPORT_SYMBOL(csum_and_copy_from_iter);
1090 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum,
1097 if (unlikely(i->type & ITER_PIPE)) {
1101 if (unlikely(i->count < bytes))
1103 iterate_all_kinds(i, bytes, v, ({
1105 next = csum_and_copy_from_user(v.iov_base,
1106 (to += v.iov_len) - v.iov_len,
1107 v.iov_len, 0, &err);
1110 sum = csum_block_add(sum, next, off);
1114 char *p = kmap_atomic(v.bv_page);
1115 next = csum_partial_copy_nocheck(p + v.bv_offset,
1116 (to += v.bv_len) - v.bv_len,
1119 sum = csum_block_add(sum, next, off);
1122 next = csum_partial_copy_nocheck(v.iov_base,
1123 (to += v.iov_len) - v.iov_len,
1125 sum = csum_block_add(sum, next, off);
1130 iov_iter_advance(i, bytes);
1133 EXPORT_SYMBOL(csum_and_copy_from_iter_full);
1135 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, __wsum *csum,
1138 const char *from = addr;
1142 if (unlikely(i->type & ITER_PIPE)) {
1143 WARN_ON(1); /* for now */
1146 iterate_and_advance(i, bytes, v, ({
1148 next = csum_and_copy_to_user((from += v.iov_len) - v.iov_len,
1150 v.iov_len, 0, &err);
1152 sum = csum_block_add(sum, next, off);
1155 err ? v.iov_len : 0;
1157 char *p = kmap_atomic(v.bv_page);
1158 next = csum_partial_copy_nocheck((from += v.bv_len) - v.bv_len,
1162 sum = csum_block_add(sum, next, off);
1165 next = csum_partial_copy_nocheck((from += v.iov_len) - v.iov_len,
1168 sum = csum_block_add(sum, next, off);
1175 EXPORT_SYMBOL(csum_and_copy_to_iter);
1177 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1179 size_t size = i->count;
1185 if (unlikely(i->type & ITER_PIPE)) {
1186 struct pipe_inode_info *pipe = i->pipe;
1193 data_start(i, &idx, &off);
1194 /* some of this one + all after this one */
1195 npages = ((pipe->curbuf - idx - 1) & (pipe->buffers - 1)) + 1;
1196 if (npages >= maxpages)
1198 } else iterate_all_kinds(i, size, v, ({
1199 unsigned long p = (unsigned long)v.iov_base;
1200 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1202 if (npages >= maxpages)
1206 if (npages >= maxpages)
1209 unsigned long p = (unsigned long)v.iov_base;
1210 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1212 if (npages >= maxpages)
1218 EXPORT_SYMBOL(iov_iter_npages);
1220 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1223 if (unlikely(new->type & ITER_PIPE)) {
1227 if (new->type & ITER_BVEC)
1228 return new->bvec = kmemdup(new->bvec,
1229 new->nr_segs * sizeof(struct bio_vec),
1232 /* iovec and kvec have identical layout */
1233 return new->iov = kmemdup(new->iov,
1234 new->nr_segs * sizeof(struct iovec),
1237 EXPORT_SYMBOL(dup_iter);
1240 * import_iovec() - Copy an array of &struct iovec from userspace
1241 * into the kernel, check that it is valid, and initialize a new
1242 * &struct iov_iter iterator to access it.
1244 * @type: One of %READ or %WRITE.
1245 * @uvector: Pointer to the userspace array.
1246 * @nr_segs: Number of elements in userspace array.
1247 * @fast_segs: Number of elements in @iov.
1248 * @iov: (input and output parameter) Pointer to pointer to (usually small
1249 * on-stack) kernel array.
1250 * @i: Pointer to iterator that will be initialized on success.
1252 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1253 * then this function places %NULL in *@iov on return. Otherwise, a new
1254 * array will be allocated and the result placed in *@iov. This means that
1255 * the caller may call kfree() on *@iov regardless of whether the small
1256 * on-stack array was used or not (and regardless of whether this function
1257 * returns an error or not).
1259 * Return: 0 on success or negative error code on error.
1261 int import_iovec(int type, const struct iovec __user * uvector,
1262 unsigned nr_segs, unsigned fast_segs,
1263 struct iovec **iov, struct iov_iter *i)
1267 n = rw_copy_check_uvector(type, uvector, nr_segs, fast_segs,
1275 iov_iter_init(i, type, p, nr_segs, n);
1276 *iov = p == *iov ? NULL : p;
1279 EXPORT_SYMBOL(import_iovec);
1281 #ifdef CONFIG_COMPAT
1282 #include <linux/compat.h>
1284 int compat_import_iovec(int type, const struct compat_iovec __user * uvector,
1285 unsigned nr_segs, unsigned fast_segs,
1286 struct iovec **iov, struct iov_iter *i)
1290 n = compat_rw_copy_check_uvector(type, uvector, nr_segs, fast_segs,
1298 iov_iter_init(i, type, p, nr_segs, n);
1299 *iov = p == *iov ? NULL : p;
1304 int import_single_range(int rw, void __user *buf, size_t len,
1305 struct iovec *iov, struct iov_iter *i)
1307 if (len > MAX_RW_COUNT)
1309 if (unlikely(!access_ok(!rw, buf, len)))
1312 iov->iov_base = buf;
1314 iov_iter_init(i, rw, iov, 1, len);
1317 EXPORT_SYMBOL(import_single_range);