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 #ifdef CONFIG_ARCH_HAS_UACCESS_MCSAFE
579 static int copyout_mcsafe(void __user *to, const void *from, size_t n)
581 if (access_ok(VERIFY_WRITE, to, n)) {
582 kasan_check_read(from, n);
583 n = copy_to_user_mcsafe((__force void *) to, from, n);
588 static unsigned long memcpy_mcsafe_to_page(struct page *page, size_t offset,
589 const char *from, size_t len)
594 to = kmap_atomic(page);
595 ret = memcpy_mcsafe(to + offset, from, len);
601 static size_t copy_pipe_to_iter_mcsafe(const void *addr, size_t bytes,
604 struct pipe_inode_info *pipe = i->pipe;
605 size_t n, off, xfer = 0;
611 bytes = n = push_pipe(i, bytes, &idx, &off);
614 for ( ; n; idx = next_idx(idx, pipe), off = 0) {
615 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
618 rem = memcpy_mcsafe_to_page(pipe->bufs[idx].page, off, addr,
621 i->iov_offset = off + chunk - rem;
633 * _copy_to_iter_mcsafe - copy to user with source-read error exception handling
634 * @addr: source kernel address
635 * @bytes: total transfer length
636 * @iter: destination iterator
638 * The pmem driver arranges for filesystem-dax to use this facility via
639 * dax_copy_to_iter() for protecting read/write to persistent memory.
640 * Unless / until an architecture can guarantee identical performance
641 * between _copy_to_iter_mcsafe() and _copy_to_iter() it would be a
642 * performance regression to switch more users to the mcsafe version.
644 * Otherwise, the main differences between this and typical _copy_to_iter().
646 * * Typical tail/residue handling after a fault retries the copy
647 * byte-by-byte until the fault happens again. Re-triggering machine
648 * checks is potentially fatal so the implementation uses source
649 * alignment and poison alignment assumptions to avoid re-triggering
650 * hardware exceptions.
652 * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
653 * Compare to copy_to_iter() where only ITER_IOVEC attempts might return
656 * See MCSAFE_TEST for self-test.
658 size_t _copy_to_iter_mcsafe(const void *addr, size_t bytes, struct iov_iter *i)
660 const char *from = addr;
661 unsigned long rem, curr_addr, s_addr = (unsigned long) addr;
663 if (unlikely(i->type & ITER_PIPE))
664 return copy_pipe_to_iter_mcsafe(addr, bytes, i);
665 if (iter_is_iovec(i))
667 iterate_and_advance(i, bytes, v,
668 copyout_mcsafe(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
670 rem = memcpy_mcsafe_to_page(v.bv_page, v.bv_offset,
671 (from += v.bv_len) - v.bv_len, v.bv_len);
673 curr_addr = (unsigned long) from;
674 bytes = curr_addr - s_addr - rem;
679 rem = memcpy_mcsafe(v.iov_base, (from += v.iov_len) - v.iov_len,
682 curr_addr = (unsigned long) from;
683 bytes = curr_addr - s_addr - rem;
691 EXPORT_SYMBOL_GPL(_copy_to_iter_mcsafe);
692 #endif /* CONFIG_ARCH_HAS_UACCESS_MCSAFE */
694 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
697 if (unlikely(i->type & ITER_PIPE)) {
701 if (iter_is_iovec(i))
703 iterate_and_advance(i, bytes, v,
704 copyin((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
705 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
706 v.bv_offset, v.bv_len),
707 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
712 EXPORT_SYMBOL(_copy_from_iter);
714 bool _copy_from_iter_full(void *addr, size_t bytes, struct iov_iter *i)
717 if (unlikely(i->type & ITER_PIPE)) {
721 if (unlikely(i->count < bytes))
724 if (iter_is_iovec(i))
726 iterate_all_kinds(i, bytes, v, ({
727 if (copyin((to += v.iov_len) - v.iov_len,
728 v.iov_base, v.iov_len))
731 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
732 v.bv_offset, v.bv_len),
733 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
736 iov_iter_advance(i, bytes);
739 EXPORT_SYMBOL(_copy_from_iter_full);
741 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
744 if (unlikely(i->type & ITER_PIPE)) {
748 iterate_and_advance(i, bytes, v,
749 __copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
750 v.iov_base, v.iov_len),
751 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
752 v.bv_offset, v.bv_len),
753 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
758 EXPORT_SYMBOL(_copy_from_iter_nocache);
760 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
762 * _copy_from_iter_flushcache - write destination through cpu cache
763 * @addr: destination kernel address
764 * @bytes: total transfer length
765 * @iter: source iterator
767 * The pmem driver arranges for filesystem-dax to use this facility via
768 * dax_copy_from_iter() for ensuring that writes to persistent memory
769 * are flushed through the CPU cache. It is differentiated from
770 * _copy_from_iter_nocache() in that guarantees all data is flushed for
771 * all iterator types. The _copy_from_iter_nocache() only attempts to
772 * bypass the cache for the ITER_IOVEC case, and on some archs may use
773 * instructions that strand dirty-data in the cache.
775 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
778 if (unlikely(i->type & ITER_PIPE)) {
782 iterate_and_advance(i, bytes, v,
783 __copy_from_user_flushcache((to += v.iov_len) - v.iov_len,
784 v.iov_base, v.iov_len),
785 memcpy_page_flushcache((to += v.bv_len) - v.bv_len, v.bv_page,
786 v.bv_offset, v.bv_len),
787 memcpy_flushcache((to += v.iov_len) - v.iov_len, v.iov_base,
793 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
796 bool _copy_from_iter_full_nocache(void *addr, size_t bytes, struct iov_iter *i)
799 if (unlikely(i->type & ITER_PIPE)) {
803 if (unlikely(i->count < bytes))
805 iterate_all_kinds(i, bytes, v, ({
806 if (__copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
807 v.iov_base, v.iov_len))
810 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
811 v.bv_offset, v.bv_len),
812 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
815 iov_iter_advance(i, bytes);
818 EXPORT_SYMBOL(_copy_from_iter_full_nocache);
820 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
823 size_t v = n + offset;
826 * The general case needs to access the page order in order
827 * to compute the page size.
828 * However, we mostly deal with order-0 pages and thus can
829 * avoid a possible cache line miss for requests that fit all
832 if (n <= v && v <= PAGE_SIZE)
835 head = compound_head(page);
836 v += (page - head) << PAGE_SHIFT;
838 if (likely(n <= v && v <= (PAGE_SIZE << compound_order(head))))
844 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
847 if (unlikely(!page_copy_sane(page, offset, bytes)))
849 if (i->type & (ITER_BVEC|ITER_KVEC)) {
850 void *kaddr = kmap_atomic(page);
851 size_t wanted = copy_to_iter(kaddr + offset, bytes, i);
852 kunmap_atomic(kaddr);
854 } else if (likely(!(i->type & ITER_PIPE)))
855 return copy_page_to_iter_iovec(page, offset, bytes, i);
857 return copy_page_to_iter_pipe(page, offset, bytes, i);
859 EXPORT_SYMBOL(copy_page_to_iter);
861 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
864 if (unlikely(!page_copy_sane(page, offset, bytes)))
866 if (unlikely(i->type & ITER_PIPE)) {
870 if (i->type & (ITER_BVEC|ITER_KVEC)) {
871 void *kaddr = kmap_atomic(page);
872 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
873 kunmap_atomic(kaddr);
876 return copy_page_from_iter_iovec(page, offset, bytes, i);
878 EXPORT_SYMBOL(copy_page_from_iter);
880 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
882 struct pipe_inode_info *pipe = i->pipe;
889 bytes = n = push_pipe(i, bytes, &idx, &off);
893 for ( ; n; idx = next_idx(idx, pipe), off = 0) {
894 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
895 memzero_page(pipe->bufs[idx].page, off, chunk);
897 i->iov_offset = off + chunk;
904 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
906 if (unlikely(i->type & ITER_PIPE))
907 return pipe_zero(bytes, i);
908 iterate_and_advance(i, bytes, v,
909 clear_user(v.iov_base, v.iov_len),
910 memzero_page(v.bv_page, v.bv_offset, v.bv_len),
911 memset(v.iov_base, 0, v.iov_len)
916 EXPORT_SYMBOL(iov_iter_zero);
918 size_t iov_iter_copy_from_user_atomic(struct page *page,
919 struct iov_iter *i, unsigned long offset, size_t bytes)
921 char *kaddr = kmap_atomic(page), *p = kaddr + offset;
922 if (unlikely(!page_copy_sane(page, offset, bytes))) {
923 kunmap_atomic(kaddr);
926 if (unlikely(i->type & ITER_PIPE)) {
927 kunmap_atomic(kaddr);
931 iterate_all_kinds(i, bytes, v,
932 copyin((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
933 memcpy_from_page((p += v.bv_len) - v.bv_len, v.bv_page,
934 v.bv_offset, v.bv_len),
935 memcpy((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
937 kunmap_atomic(kaddr);
940 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
942 static inline void pipe_truncate(struct iov_iter *i)
944 struct pipe_inode_info *pipe = i->pipe;
946 size_t off = i->iov_offset;
948 int nrbufs = (idx - pipe->curbuf) & (pipe->buffers - 1);
950 pipe->bufs[idx].len = off - pipe->bufs[idx].offset;
951 idx = next_idx(idx, pipe);
954 while (pipe->nrbufs > nrbufs) {
955 pipe_buf_release(pipe, &pipe->bufs[idx]);
956 idx = next_idx(idx, pipe);
962 static void pipe_advance(struct iov_iter *i, size_t size)
964 struct pipe_inode_info *pipe = i->pipe;
965 if (unlikely(i->count < size))
968 struct pipe_buffer *buf;
969 size_t off = i->iov_offset, left = size;
971 if (off) /* make it relative to the beginning of buffer */
972 left += off - pipe->bufs[idx].offset;
974 buf = &pipe->bufs[idx];
975 if (left <= buf->len)
978 idx = next_idx(idx, pipe);
981 i->iov_offset = buf->offset + left;
984 /* ... and discard everything past that point */
988 void iov_iter_advance(struct iov_iter *i, size_t size)
990 if (unlikely(i->type & ITER_PIPE)) {
991 pipe_advance(i, size);
994 iterate_and_advance(i, size, v, 0, 0, 0)
996 EXPORT_SYMBOL(iov_iter_advance);
998 void iov_iter_revert(struct iov_iter *i, size_t unroll)
1002 if (WARN_ON(unroll > MAX_RW_COUNT))
1005 if (unlikely(i->type & ITER_PIPE)) {
1006 struct pipe_inode_info *pipe = i->pipe;
1008 size_t off = i->iov_offset;
1010 size_t n = off - pipe->bufs[idx].offset;
1016 if (!unroll && idx == i->start_idx) {
1021 idx = pipe->buffers - 1;
1022 off = pipe->bufs[idx].offset + pipe->bufs[idx].len;
1024 i->iov_offset = off;
1029 if (unroll <= i->iov_offset) {
1030 i->iov_offset -= unroll;
1033 unroll -= i->iov_offset;
1034 if (i->type & ITER_BVEC) {
1035 const struct bio_vec *bvec = i->bvec;
1037 size_t n = (--bvec)->bv_len;
1041 i->iov_offset = n - unroll;
1046 } else { /* same logics for iovec and kvec */
1047 const struct iovec *iov = i->iov;
1049 size_t n = (--iov)->iov_len;
1053 i->iov_offset = n - unroll;
1060 EXPORT_SYMBOL(iov_iter_revert);
1063 * Return the count of just the current iov_iter segment.
1065 size_t iov_iter_single_seg_count(const struct iov_iter *i)
1067 if (unlikely(i->type & ITER_PIPE))
1068 return i->count; // it is a silly place, anyway
1069 if (i->nr_segs == 1)
1071 else if (i->type & ITER_BVEC)
1072 return min(i->count, i->bvec->bv_len - i->iov_offset);
1074 return min(i->count, i->iov->iov_len - i->iov_offset);
1076 EXPORT_SYMBOL(iov_iter_single_seg_count);
1078 void iov_iter_kvec(struct iov_iter *i, int direction,
1079 const struct kvec *kvec, unsigned long nr_segs,
1082 BUG_ON(!(direction & ITER_KVEC));
1083 i->type = direction;
1085 i->nr_segs = nr_segs;
1089 EXPORT_SYMBOL(iov_iter_kvec);
1091 void iov_iter_bvec(struct iov_iter *i, int direction,
1092 const struct bio_vec *bvec, unsigned long nr_segs,
1095 BUG_ON(!(direction & ITER_BVEC));
1096 i->type = direction;
1098 i->nr_segs = nr_segs;
1102 EXPORT_SYMBOL(iov_iter_bvec);
1104 void iov_iter_pipe(struct iov_iter *i, int direction,
1105 struct pipe_inode_info *pipe,
1108 BUG_ON(direction != ITER_PIPE);
1109 WARN_ON(pipe->nrbufs == pipe->buffers);
1110 i->type = direction;
1112 i->idx = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1115 i->start_idx = i->idx;
1117 EXPORT_SYMBOL(iov_iter_pipe);
1119 unsigned long iov_iter_alignment(const struct iov_iter *i)
1121 unsigned long res = 0;
1122 size_t size = i->count;
1124 if (unlikely(i->type & ITER_PIPE)) {
1125 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->idx]))
1126 return size | i->iov_offset;
1129 iterate_all_kinds(i, size, v,
1130 (res |= (unsigned long)v.iov_base | v.iov_len, 0),
1131 res |= v.bv_offset | v.bv_len,
1132 res |= (unsigned long)v.iov_base | v.iov_len
1136 EXPORT_SYMBOL(iov_iter_alignment);
1138 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1140 unsigned long res = 0;
1141 size_t size = i->count;
1143 if (unlikely(i->type & ITER_PIPE)) {
1148 iterate_all_kinds(i, size, v,
1149 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1150 (size != v.iov_len ? size : 0), 0),
1151 (res |= (!res ? 0 : (unsigned long)v.bv_offset) |
1152 (size != v.bv_len ? size : 0)),
1153 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1154 (size != v.iov_len ? size : 0))
1158 EXPORT_SYMBOL(iov_iter_gap_alignment);
1160 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1162 struct page **pages,
1166 struct pipe_inode_info *pipe = i->pipe;
1167 ssize_t n = push_pipe(i, maxsize, &idx, start);
1174 get_page(*pages++ = pipe->bufs[idx].page);
1175 idx = next_idx(idx, pipe);
1182 static ssize_t pipe_get_pages(struct iov_iter *i,
1183 struct page **pages, size_t maxsize, unsigned maxpages,
1196 data_start(i, &idx, start);
1197 /* some of this one + all after this one */
1198 npages = ((i->pipe->curbuf - idx - 1) & (i->pipe->buffers - 1)) + 1;
1199 capacity = min(npages,maxpages) * PAGE_SIZE - *start;
1201 return __pipe_get_pages(i, min(maxsize, capacity), pages, idx, start);
1204 ssize_t iov_iter_get_pages(struct iov_iter *i,
1205 struct page **pages, size_t maxsize, unsigned maxpages,
1208 if (maxsize > i->count)
1211 if (unlikely(i->type & ITER_PIPE))
1212 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1213 iterate_all_kinds(i, maxsize, v, ({
1214 unsigned long addr = (unsigned long)v.iov_base;
1215 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1219 if (len > maxpages * PAGE_SIZE)
1220 len = maxpages * PAGE_SIZE;
1221 addr &= ~(PAGE_SIZE - 1);
1222 n = DIV_ROUND_UP(len, PAGE_SIZE);
1223 res = get_user_pages_fast(addr, n, (i->type & WRITE) != WRITE, pages);
1224 if (unlikely(res < 0))
1226 return (res == n ? len : res * PAGE_SIZE) - *start;
1228 /* can't be more than PAGE_SIZE */
1229 *start = v.bv_offset;
1230 get_page(*pages = v.bv_page);
1238 EXPORT_SYMBOL(iov_iter_get_pages);
1240 static struct page **get_pages_array(size_t n)
1242 return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1245 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1246 struct page ***pages, size_t maxsize,
1260 data_start(i, &idx, start);
1261 /* some of this one + all after this one */
1262 npages = ((i->pipe->curbuf - idx - 1) & (i->pipe->buffers - 1)) + 1;
1263 n = npages * PAGE_SIZE - *start;
1267 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1268 p = get_pages_array(npages);
1271 n = __pipe_get_pages(i, maxsize, p, idx, start);
1279 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1280 struct page ***pages, size_t maxsize,
1285 if (maxsize > i->count)
1288 if (unlikely(i->type & ITER_PIPE))
1289 return pipe_get_pages_alloc(i, pages, maxsize, start);
1290 iterate_all_kinds(i, maxsize, v, ({
1291 unsigned long addr = (unsigned long)v.iov_base;
1292 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1296 addr &= ~(PAGE_SIZE - 1);
1297 n = DIV_ROUND_UP(len, PAGE_SIZE);
1298 p = get_pages_array(n);
1301 res = get_user_pages_fast(addr, n, (i->type & WRITE) != WRITE, p);
1302 if (unlikely(res < 0)) {
1307 return (res == n ? len : res * PAGE_SIZE) - *start;
1309 /* can't be more than PAGE_SIZE */
1310 *start = v.bv_offset;
1311 *pages = p = get_pages_array(1);
1314 get_page(*p = v.bv_page);
1322 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1324 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1331 if (unlikely(i->type & ITER_PIPE)) {
1335 iterate_and_advance(i, bytes, v, ({
1337 next = csum_and_copy_from_user(v.iov_base,
1338 (to += v.iov_len) - v.iov_len,
1339 v.iov_len, 0, &err);
1341 sum = csum_block_add(sum, next, off);
1344 err ? v.iov_len : 0;
1346 char *p = kmap_atomic(v.bv_page);
1347 next = csum_partial_copy_nocheck(p + v.bv_offset,
1348 (to += v.bv_len) - v.bv_len,
1351 sum = csum_block_add(sum, next, off);
1354 next = csum_partial_copy_nocheck(v.iov_base,
1355 (to += v.iov_len) - v.iov_len,
1357 sum = csum_block_add(sum, next, off);
1364 EXPORT_SYMBOL(csum_and_copy_from_iter);
1366 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum,
1373 if (unlikely(i->type & ITER_PIPE)) {
1377 if (unlikely(i->count < bytes))
1379 iterate_all_kinds(i, bytes, v, ({
1381 next = csum_and_copy_from_user(v.iov_base,
1382 (to += v.iov_len) - v.iov_len,
1383 v.iov_len, 0, &err);
1386 sum = csum_block_add(sum, next, off);
1390 char *p = kmap_atomic(v.bv_page);
1391 next = csum_partial_copy_nocheck(p + v.bv_offset,
1392 (to += v.bv_len) - v.bv_len,
1395 sum = csum_block_add(sum, next, off);
1398 next = csum_partial_copy_nocheck(v.iov_base,
1399 (to += v.iov_len) - v.iov_len,
1401 sum = csum_block_add(sum, next, off);
1406 iov_iter_advance(i, bytes);
1409 EXPORT_SYMBOL(csum_and_copy_from_iter_full);
1411 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, __wsum *csum,
1414 const char *from = addr;
1418 if (unlikely(i->type & ITER_PIPE)) {
1419 WARN_ON(1); /* for now */
1422 iterate_and_advance(i, bytes, v, ({
1424 next = csum_and_copy_to_user((from += v.iov_len) - v.iov_len,
1426 v.iov_len, 0, &err);
1428 sum = csum_block_add(sum, next, off);
1431 err ? v.iov_len : 0;
1433 char *p = kmap_atomic(v.bv_page);
1434 next = csum_partial_copy_nocheck((from += v.bv_len) - v.bv_len,
1438 sum = csum_block_add(sum, next, off);
1441 next = csum_partial_copy_nocheck((from += v.iov_len) - v.iov_len,
1444 sum = csum_block_add(sum, next, off);
1451 EXPORT_SYMBOL(csum_and_copy_to_iter);
1453 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1455 size_t size = i->count;
1461 if (unlikely(i->type & ITER_PIPE)) {
1462 struct pipe_inode_info *pipe = i->pipe;
1469 data_start(i, &idx, &off);
1470 /* some of this one + all after this one */
1471 npages = ((pipe->curbuf - idx - 1) & (pipe->buffers - 1)) + 1;
1472 if (npages >= maxpages)
1474 } else iterate_all_kinds(i, size, v, ({
1475 unsigned long p = (unsigned long)v.iov_base;
1476 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1478 if (npages >= maxpages)
1482 if (npages >= maxpages)
1485 unsigned long p = (unsigned long)v.iov_base;
1486 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1488 if (npages >= maxpages)
1494 EXPORT_SYMBOL(iov_iter_npages);
1496 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1499 if (unlikely(new->type & ITER_PIPE)) {
1503 if (new->type & ITER_BVEC)
1504 return new->bvec = kmemdup(new->bvec,
1505 new->nr_segs * sizeof(struct bio_vec),
1508 /* iovec and kvec have identical layout */
1509 return new->iov = kmemdup(new->iov,
1510 new->nr_segs * sizeof(struct iovec),
1513 EXPORT_SYMBOL(dup_iter);
1516 * import_iovec() - Copy an array of &struct iovec from userspace
1517 * into the kernel, check that it is valid, and initialize a new
1518 * &struct iov_iter iterator to access it.
1520 * @type: One of %READ or %WRITE.
1521 * @uvector: Pointer to the userspace array.
1522 * @nr_segs: Number of elements in userspace array.
1523 * @fast_segs: Number of elements in @iov.
1524 * @iov: (input and output parameter) Pointer to pointer to (usually small
1525 * on-stack) kernel array.
1526 * @i: Pointer to iterator that will be initialized on success.
1528 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1529 * then this function places %NULL in *@iov on return. Otherwise, a new
1530 * array will be allocated and the result placed in *@iov. This means that
1531 * the caller may call kfree() on *@iov regardless of whether the small
1532 * on-stack array was used or not (and regardless of whether this function
1533 * returns an error or not).
1535 * Return: 0 on success or negative error code on error.
1537 int import_iovec(int type, const struct iovec __user * uvector,
1538 unsigned nr_segs, unsigned fast_segs,
1539 struct iovec **iov, struct iov_iter *i)
1543 n = rw_copy_check_uvector(type, uvector, nr_segs, fast_segs,
1551 iov_iter_init(i, type, p, nr_segs, n);
1552 *iov = p == *iov ? NULL : p;
1555 EXPORT_SYMBOL(import_iovec);
1557 #ifdef CONFIG_COMPAT
1558 #include <linux/compat.h>
1560 int compat_import_iovec(int type, const struct compat_iovec __user * uvector,
1561 unsigned nr_segs, unsigned fast_segs,
1562 struct iovec **iov, struct iov_iter *i)
1566 n = compat_rw_copy_check_uvector(type, uvector, nr_segs, fast_segs,
1574 iov_iter_init(i, type, p, nr_segs, n);
1575 *iov = p == *iov ? NULL : p;
1580 int import_single_range(int rw, void __user *buf, size_t len,
1581 struct iovec *iov, struct iov_iter *i)
1583 if (len > MAX_RW_COUNT)
1585 if (unlikely(!access_ok(!rw, buf, len)))
1588 iov->iov_base = buf;
1590 iov_iter_init(i, rw, iov, 1, len);
1593 EXPORT_SYMBOL(import_single_range);
1595 int iov_iter_for_each_range(struct iov_iter *i, size_t bytes,
1596 int (*f)(struct kvec *vec, void *context),
1604 iterate_all_kinds(i, bytes, v, -EINVAL, ({
1605 w.iov_base = kmap(v.bv_page) + v.bv_offset;
1606 w.iov_len = v.bv_len;
1607 err = f(&w, context);
1611 err = f(&w, context);})
1615 EXPORT_SYMBOL(iov_iter_for_each_range);