1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <linux/export.h>
3 #include <linux/bvec.h>
5 #include <linux/pagemap.h>
6 #include <linux/slab.h>
7 #include <linux/vmalloc.h>
8 #include <linux/splice.h>
9 #include <net/checksum.h>
10 #include <linux/scatterlist.h>
12 #define PIPE_PARANOIA /* for now */
14 #define iterate_iovec(i, n, __v, __p, skip, STEP) { \
18 __v.iov_len = min(n, __p->iov_len - skip); \
19 if (likely(__v.iov_len)) { \
20 __v.iov_base = __p->iov_base + skip; \
22 __v.iov_len -= left; \
23 skip += __v.iov_len; \
28 while (unlikely(!left && n)) { \
30 __v.iov_len = min(n, __p->iov_len); \
31 if (unlikely(!__v.iov_len)) \
33 __v.iov_base = __p->iov_base; \
35 __v.iov_len -= left; \
42 #define iterate_kvec(i, n, __v, __p, skip, STEP) { \
45 __v.iov_len = min(n, __p->iov_len - skip); \
46 if (likely(__v.iov_len)) { \
47 __v.iov_base = __p->iov_base + skip; \
49 skip += __v.iov_len; \
52 while (unlikely(n)) { \
54 __v.iov_len = min(n, __p->iov_len); \
55 if (unlikely(!__v.iov_len)) \
57 __v.iov_base = __p->iov_base; \
65 #define iterate_bvec(i, n, __v, __bi, skip, STEP) { \
66 struct bvec_iter __start; \
67 __start.bi_size = n; \
68 __start.bi_bvec_done = skip; \
70 for_each_bvec(__v, i->bvec, __bi, __start) { \
77 #define iterate_all_kinds(i, n, v, I, B, K) { \
79 size_t skip = i->iov_offset; \
80 if (unlikely(i->type & ITER_BVEC)) { \
82 struct bvec_iter __bi; \
83 iterate_bvec(i, n, v, __bi, skip, (B)) \
84 } else if (unlikely(i->type & ITER_KVEC)) { \
85 const struct kvec *kvec; \
87 iterate_kvec(i, n, v, kvec, skip, (K)) \
88 } else if (unlikely(i->type & ITER_DISCARD)) { \
90 const struct iovec *iov; \
92 iterate_iovec(i, n, v, iov, skip, (I)) \
97 #define iterate_and_advance(i, n, v, I, B, K) { \
98 if (unlikely(i->count < n)) \
101 size_t skip = i->iov_offset; \
102 if (unlikely(i->type & ITER_BVEC)) { \
103 const struct bio_vec *bvec = i->bvec; \
105 struct bvec_iter __bi; \
106 iterate_bvec(i, n, v, __bi, skip, (B)) \
107 i->bvec = __bvec_iter_bvec(i->bvec, __bi); \
108 i->nr_segs -= i->bvec - bvec; \
109 skip = __bi.bi_bvec_done; \
110 } else if (unlikely(i->type & ITER_KVEC)) { \
111 const struct kvec *kvec; \
113 iterate_kvec(i, n, v, kvec, skip, (K)) \
114 if (skip == kvec->iov_len) { \
118 i->nr_segs -= kvec - i->kvec; \
120 } else if (unlikely(i->type & ITER_DISCARD)) { \
123 const struct iovec *iov; \
125 iterate_iovec(i, n, v, iov, skip, (I)) \
126 if (skip == iov->iov_len) { \
130 i->nr_segs -= iov - i->iov; \
134 i->iov_offset = skip; \
138 static int copyout(void __user *to, const void *from, size_t n)
140 if (access_ok(to, n)) {
141 kasan_check_read(from, n);
142 n = raw_copy_to_user(to, from, n);
147 static int copyin(void *to, const void __user *from, size_t n)
149 if (access_ok(from, n)) {
150 kasan_check_write(to, n);
151 n = raw_copy_from_user(to, from, n);
156 static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes,
159 size_t skip, copy, left, wanted;
160 const struct iovec *iov;
164 if (unlikely(bytes > i->count))
167 if (unlikely(!bytes))
173 skip = i->iov_offset;
174 buf = iov->iov_base + skip;
175 copy = min(bytes, iov->iov_len - skip);
177 if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_writeable(buf, copy)) {
178 kaddr = kmap_atomic(page);
179 from = kaddr + offset;
181 /* first chunk, usually the only one */
182 left = copyout(buf, from, copy);
188 while (unlikely(!left && bytes)) {
191 copy = min(bytes, iov->iov_len);
192 left = copyout(buf, from, copy);
198 if (likely(!bytes)) {
199 kunmap_atomic(kaddr);
202 offset = from - kaddr;
204 kunmap_atomic(kaddr);
205 copy = min(bytes, iov->iov_len - skip);
207 /* Too bad - revert to non-atomic kmap */
210 from = kaddr + offset;
211 left = copyout(buf, from, copy);
216 while (unlikely(!left && bytes)) {
219 copy = min(bytes, iov->iov_len);
220 left = copyout(buf, from, copy);
229 if (skip == iov->iov_len) {
233 i->count -= wanted - bytes;
234 i->nr_segs -= iov - i->iov;
236 i->iov_offset = skip;
237 return wanted - bytes;
240 static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes,
243 size_t skip, copy, left, wanted;
244 const struct iovec *iov;
248 if (unlikely(bytes > i->count))
251 if (unlikely(!bytes))
257 skip = i->iov_offset;
258 buf = iov->iov_base + skip;
259 copy = min(bytes, iov->iov_len - skip);
261 if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_readable(buf, copy)) {
262 kaddr = kmap_atomic(page);
265 /* first chunk, usually the only one */
266 left = copyin(to, buf, copy);
272 while (unlikely(!left && bytes)) {
275 copy = min(bytes, iov->iov_len);
276 left = copyin(to, buf, copy);
282 if (likely(!bytes)) {
283 kunmap_atomic(kaddr);
288 kunmap_atomic(kaddr);
289 copy = min(bytes, iov->iov_len - skip);
291 /* Too bad - revert to non-atomic kmap */
295 left = copyin(to, buf, copy);
300 while (unlikely(!left && bytes)) {
303 copy = min(bytes, iov->iov_len);
304 left = copyin(to, buf, copy);
313 if (skip == iov->iov_len) {
317 i->count -= wanted - bytes;
318 i->nr_segs -= iov - i->iov;
320 i->iov_offset = skip;
321 return wanted - bytes;
325 static bool sanity(const struct iov_iter *i)
327 struct pipe_inode_info *pipe = i->pipe;
329 int next = pipe->curbuf + pipe->nrbufs;
331 struct pipe_buffer *p;
332 if (unlikely(!pipe->nrbufs))
333 goto Bad; // pipe must be non-empty
334 if (unlikely(idx != ((next - 1) & (pipe->buffers - 1))))
335 goto Bad; // must be at the last buffer...
337 p = &pipe->bufs[idx];
338 if (unlikely(p->offset + p->len != i->iov_offset))
339 goto Bad; // ... at the end of segment
341 if (idx != (next & (pipe->buffers - 1)))
342 goto Bad; // must be right after the last buffer
346 printk(KERN_ERR "idx = %d, offset = %zd\n", i->idx, i->iov_offset);
347 printk(KERN_ERR "curbuf = %d, nrbufs = %d, buffers = %d\n",
348 pipe->curbuf, pipe->nrbufs, pipe->buffers);
349 for (idx = 0; idx < pipe->buffers; idx++)
350 printk(KERN_ERR "[%p %p %d %d]\n",
352 pipe->bufs[idx].page,
353 pipe->bufs[idx].offset,
354 pipe->bufs[idx].len);
359 #define sanity(i) true
362 static inline int next_idx(int idx, struct pipe_inode_info *pipe)
364 return (idx + 1) & (pipe->buffers - 1);
367 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
370 struct pipe_inode_info *pipe = i->pipe;
371 struct pipe_buffer *buf;
375 if (unlikely(bytes > i->count))
378 if (unlikely(!bytes))
386 buf = &pipe->bufs[idx];
388 if (offset == off && buf->page == page) {
389 /* merge with the last one */
391 i->iov_offset += bytes;
394 idx = next_idx(idx, pipe);
395 buf = &pipe->bufs[idx];
397 if (idx == pipe->curbuf && pipe->nrbufs)
400 buf->ops = &page_cache_pipe_buf_ops;
402 get_page(buf->page = page);
403 buf->offset = offset;
405 i->iov_offset = offset + bytes;
413 * Fault in one or more iovecs of the given iov_iter, to a maximum length of
414 * bytes. For each iovec, fault in each page that constitutes the iovec.
416 * Return 0 on success, or non-zero if the memory could not be accessed (i.e.
417 * because it is an invalid address).
419 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
421 size_t skip = i->iov_offset;
422 const struct iovec *iov;
426 if (iter_is_iovec(i)) {
427 iterate_iovec(i, bytes, v, iov, skip, ({
428 err = fault_in_pages_readable(v.iov_base, v.iov_len);
435 EXPORT_SYMBOL(iov_iter_fault_in_readable);
437 void iov_iter_init(struct iov_iter *i, unsigned int direction,
438 const struct iovec *iov, unsigned long nr_segs,
441 WARN_ON(direction & ~(READ | WRITE));
442 direction &= READ | WRITE;
444 /* It will get better. Eventually... */
445 if (uaccess_kernel()) {
446 i->type = ITER_KVEC | direction;
447 i->kvec = (struct kvec *)iov;
449 i->type = ITER_IOVEC | direction;
452 i->nr_segs = nr_segs;
456 EXPORT_SYMBOL(iov_iter_init);
458 static void memzero_page(struct page *page, size_t offset, size_t len)
460 char *addr = kmap_atomic(page);
461 memset(addr + offset, 0, len);
465 static inline bool allocated(struct pipe_buffer *buf)
467 return buf->ops == &default_pipe_buf_ops;
470 static inline void data_start(const struct iov_iter *i, int *idxp, size_t *offp)
472 size_t off = i->iov_offset;
474 if (off && (!allocated(&i->pipe->bufs[idx]) || off == PAGE_SIZE)) {
475 idx = next_idx(idx, i->pipe);
482 static size_t push_pipe(struct iov_iter *i, size_t size,
483 int *idxp, size_t *offp)
485 struct pipe_inode_info *pipe = i->pipe;
490 if (unlikely(size > i->count))
496 data_start(i, &idx, &off);
500 left -= PAGE_SIZE - off;
502 pipe->bufs[idx].len += size;
505 pipe->bufs[idx].len = PAGE_SIZE;
506 idx = next_idx(idx, pipe);
508 while (idx != pipe->curbuf || !pipe->nrbufs) {
509 struct page *page = alloc_page(GFP_USER);
513 pipe->bufs[idx].ops = &default_pipe_buf_ops;
514 pipe->bufs[idx].flags = 0;
515 pipe->bufs[idx].page = page;
516 pipe->bufs[idx].offset = 0;
517 if (left <= PAGE_SIZE) {
518 pipe->bufs[idx].len = left;
521 pipe->bufs[idx].len = PAGE_SIZE;
523 idx = next_idx(idx, pipe);
528 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
531 struct pipe_inode_info *pipe = i->pipe;
538 bytes = n = push_pipe(i, bytes, &idx, &off);
541 for ( ; n; idx = next_idx(idx, pipe), off = 0) {
542 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
543 memcpy_to_page(pipe->bufs[idx].page, off, addr, chunk);
545 i->iov_offset = off + chunk;
553 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
554 __wsum sum, size_t off)
556 __wsum next = csum_partial_copy_nocheck(from, to, len, 0);
557 return csum_block_add(sum, next, off);
560 static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
561 struct csum_state *csstate,
564 struct pipe_inode_info *pipe = i->pipe;
565 __wsum sum = csstate->csum;
566 size_t off = csstate->off;
573 bytes = n = push_pipe(i, bytes, &idx, &r);
576 for ( ; n; idx = next_idx(idx, pipe), r = 0) {
577 size_t chunk = min_t(size_t, n, PAGE_SIZE - r);
578 char *p = kmap_atomic(pipe->bufs[idx].page);
579 sum = csum_and_memcpy(p + r, addr, chunk, sum, off);
582 i->iov_offset = r + chunk;
593 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
595 const char *from = addr;
596 if (unlikely(iov_iter_is_pipe(i)))
597 return copy_pipe_to_iter(addr, bytes, i);
598 if (iter_is_iovec(i))
600 iterate_and_advance(i, bytes, v,
601 copyout(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
602 memcpy_to_page(v.bv_page, v.bv_offset,
603 (from += v.bv_len) - v.bv_len, v.bv_len),
604 memcpy(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len)
609 EXPORT_SYMBOL(_copy_to_iter);
611 #ifdef CONFIG_ARCH_HAS_UACCESS_MCSAFE
612 static int copyout_mcsafe(void __user *to, const void *from, size_t n)
614 if (access_ok(to, n)) {
615 kasan_check_read(from, n);
616 n = copy_to_user_mcsafe((__force void *) to, from, n);
621 static unsigned long memcpy_mcsafe_to_page(struct page *page, size_t offset,
622 const char *from, size_t len)
627 to = kmap_atomic(page);
628 ret = memcpy_mcsafe(to + offset, from, len);
634 static size_t copy_pipe_to_iter_mcsafe(const void *addr, size_t bytes,
637 struct pipe_inode_info *pipe = i->pipe;
638 size_t n, off, xfer = 0;
644 bytes = n = push_pipe(i, bytes, &idx, &off);
647 for ( ; n; idx = next_idx(idx, pipe), off = 0) {
648 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
651 rem = memcpy_mcsafe_to_page(pipe->bufs[idx].page, off, addr,
654 i->iov_offset = off + chunk - rem;
666 * _copy_to_iter_mcsafe - copy to user with source-read error exception handling
667 * @addr: source kernel address
668 * @bytes: total transfer length
669 * @iter: destination iterator
671 * The pmem driver arranges for filesystem-dax to use this facility via
672 * dax_copy_to_iter() for protecting read/write to persistent memory.
673 * Unless / until an architecture can guarantee identical performance
674 * between _copy_to_iter_mcsafe() and _copy_to_iter() it would be a
675 * performance regression to switch more users to the mcsafe version.
677 * Otherwise, the main differences between this and typical _copy_to_iter().
679 * * Typical tail/residue handling after a fault retries the copy
680 * byte-by-byte until the fault happens again. Re-triggering machine
681 * checks is potentially fatal so the implementation uses source
682 * alignment and poison alignment assumptions to avoid re-triggering
683 * hardware exceptions.
685 * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
686 * Compare to copy_to_iter() where only ITER_IOVEC attempts might return
689 * See MCSAFE_TEST for self-test.
691 size_t _copy_to_iter_mcsafe(const void *addr, size_t bytes, struct iov_iter *i)
693 const char *from = addr;
694 unsigned long rem, curr_addr, s_addr = (unsigned long) addr;
696 if (unlikely(iov_iter_is_pipe(i)))
697 return copy_pipe_to_iter_mcsafe(addr, bytes, i);
698 if (iter_is_iovec(i))
700 iterate_and_advance(i, bytes, v,
701 copyout_mcsafe(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
703 rem = memcpy_mcsafe_to_page(v.bv_page, v.bv_offset,
704 (from += v.bv_len) - v.bv_len, v.bv_len);
706 curr_addr = (unsigned long) from;
707 bytes = curr_addr - s_addr - rem;
712 rem = memcpy_mcsafe(v.iov_base, (from += v.iov_len) - v.iov_len,
715 curr_addr = (unsigned long) from;
716 bytes = curr_addr - s_addr - rem;
724 EXPORT_SYMBOL_GPL(_copy_to_iter_mcsafe);
725 #endif /* CONFIG_ARCH_HAS_UACCESS_MCSAFE */
727 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
730 if (unlikely(iov_iter_is_pipe(i))) {
734 if (iter_is_iovec(i))
736 iterate_and_advance(i, bytes, v,
737 copyin((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
738 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
739 v.bv_offset, v.bv_len),
740 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
745 EXPORT_SYMBOL(_copy_from_iter);
747 bool _copy_from_iter_full(void *addr, size_t bytes, struct iov_iter *i)
750 if (unlikely(iov_iter_is_pipe(i))) {
754 if (unlikely(i->count < bytes))
757 if (iter_is_iovec(i))
759 iterate_all_kinds(i, bytes, v, ({
760 if (copyin((to += v.iov_len) - v.iov_len,
761 v.iov_base, v.iov_len))
764 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
765 v.bv_offset, v.bv_len),
766 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
769 iov_iter_advance(i, bytes);
772 EXPORT_SYMBOL(_copy_from_iter_full);
774 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
777 if (unlikely(iov_iter_is_pipe(i))) {
781 iterate_and_advance(i, bytes, v,
782 __copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
783 v.iov_base, v.iov_len),
784 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
785 v.bv_offset, v.bv_len),
786 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
791 EXPORT_SYMBOL(_copy_from_iter_nocache);
793 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
795 * _copy_from_iter_flushcache - write destination through cpu cache
796 * @addr: destination kernel address
797 * @bytes: total transfer length
798 * @iter: source iterator
800 * The pmem driver arranges for filesystem-dax to use this facility via
801 * dax_copy_from_iter() for ensuring that writes to persistent memory
802 * are flushed through the CPU cache. It is differentiated from
803 * _copy_from_iter_nocache() in that guarantees all data is flushed for
804 * all iterator types. The _copy_from_iter_nocache() only attempts to
805 * bypass the cache for the ITER_IOVEC case, and on some archs may use
806 * instructions that strand dirty-data in the cache.
808 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
811 if (unlikely(iov_iter_is_pipe(i))) {
815 iterate_and_advance(i, bytes, v,
816 __copy_from_user_flushcache((to += v.iov_len) - v.iov_len,
817 v.iov_base, v.iov_len),
818 memcpy_page_flushcache((to += v.bv_len) - v.bv_len, v.bv_page,
819 v.bv_offset, v.bv_len),
820 memcpy_flushcache((to += v.iov_len) - v.iov_len, v.iov_base,
826 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
829 bool _copy_from_iter_full_nocache(void *addr, size_t bytes, struct iov_iter *i)
832 if (unlikely(iov_iter_is_pipe(i))) {
836 if (unlikely(i->count < bytes))
838 iterate_all_kinds(i, bytes, v, ({
839 if (__copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
840 v.iov_base, v.iov_len))
843 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
844 v.bv_offset, v.bv_len),
845 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
848 iov_iter_advance(i, bytes);
851 EXPORT_SYMBOL(_copy_from_iter_full_nocache);
853 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
856 size_t v = n + offset;
859 * The general case needs to access the page order in order
860 * to compute the page size.
861 * However, we mostly deal with order-0 pages and thus can
862 * avoid a possible cache line miss for requests that fit all
865 if (n <= v && v <= PAGE_SIZE)
868 head = compound_head(page);
869 v += (page - head) << PAGE_SHIFT;
871 if (likely(n <= v && v <= (page_size(head))))
877 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
880 if (unlikely(!page_copy_sane(page, offset, bytes)))
882 if (i->type & (ITER_BVEC|ITER_KVEC)) {
883 void *kaddr = kmap_atomic(page);
884 size_t wanted = copy_to_iter(kaddr + offset, bytes, i);
885 kunmap_atomic(kaddr);
887 } else if (unlikely(iov_iter_is_discard(i))) {
888 if (unlikely(i->count < bytes))
892 } else if (likely(!iov_iter_is_pipe(i)))
893 return copy_page_to_iter_iovec(page, offset, bytes, i);
895 return copy_page_to_iter_pipe(page, offset, bytes, i);
897 EXPORT_SYMBOL(copy_page_to_iter);
899 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
902 if (unlikely(!page_copy_sane(page, offset, bytes)))
904 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
908 if (i->type & (ITER_BVEC|ITER_KVEC)) {
909 void *kaddr = kmap_atomic(page);
910 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
911 kunmap_atomic(kaddr);
914 return copy_page_from_iter_iovec(page, offset, bytes, i);
916 EXPORT_SYMBOL(copy_page_from_iter);
918 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
920 struct pipe_inode_info *pipe = i->pipe;
927 bytes = n = push_pipe(i, bytes, &idx, &off);
931 for ( ; n; idx = next_idx(idx, pipe), off = 0) {
932 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
933 memzero_page(pipe->bufs[idx].page, off, chunk);
935 i->iov_offset = off + chunk;
942 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
944 if (unlikely(iov_iter_is_pipe(i)))
945 return pipe_zero(bytes, i);
946 iterate_and_advance(i, bytes, v,
947 clear_user(v.iov_base, v.iov_len),
948 memzero_page(v.bv_page, v.bv_offset, v.bv_len),
949 memset(v.iov_base, 0, v.iov_len)
954 EXPORT_SYMBOL(iov_iter_zero);
956 size_t iov_iter_copy_from_user_atomic(struct page *page,
957 struct iov_iter *i, unsigned long offset, size_t bytes)
959 char *kaddr = kmap_atomic(page), *p = kaddr + offset;
960 if (unlikely(!page_copy_sane(page, offset, bytes))) {
961 kunmap_atomic(kaddr);
964 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
965 kunmap_atomic(kaddr);
969 iterate_all_kinds(i, bytes, v,
970 copyin((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
971 memcpy_from_page((p += v.bv_len) - v.bv_len, v.bv_page,
972 v.bv_offset, v.bv_len),
973 memcpy((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
975 kunmap_atomic(kaddr);
978 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
980 static inline void pipe_truncate(struct iov_iter *i)
982 struct pipe_inode_info *pipe = i->pipe;
984 size_t off = i->iov_offset;
986 int nrbufs = (idx - pipe->curbuf) & (pipe->buffers - 1);
988 pipe->bufs[idx].len = off - pipe->bufs[idx].offset;
989 idx = next_idx(idx, pipe);
992 while (pipe->nrbufs > nrbufs) {
993 pipe_buf_release(pipe, &pipe->bufs[idx]);
994 idx = next_idx(idx, pipe);
1000 static void pipe_advance(struct iov_iter *i, size_t size)
1002 struct pipe_inode_info *pipe = i->pipe;
1003 if (unlikely(i->count < size))
1006 struct pipe_buffer *buf;
1007 size_t off = i->iov_offset, left = size;
1009 if (off) /* make it relative to the beginning of buffer */
1010 left += off - pipe->bufs[idx].offset;
1012 buf = &pipe->bufs[idx];
1013 if (left <= buf->len)
1016 idx = next_idx(idx, pipe);
1019 i->iov_offset = buf->offset + left;
1022 /* ... and discard everything past that point */
1026 void iov_iter_advance(struct iov_iter *i, size_t size)
1028 if (unlikely(iov_iter_is_pipe(i))) {
1029 pipe_advance(i, size);
1032 if (unlikely(iov_iter_is_discard(i))) {
1036 iterate_and_advance(i, size, v, 0, 0, 0)
1038 EXPORT_SYMBOL(iov_iter_advance);
1040 void iov_iter_revert(struct iov_iter *i, size_t unroll)
1044 if (WARN_ON(unroll > MAX_RW_COUNT))
1047 if (unlikely(iov_iter_is_pipe(i))) {
1048 struct pipe_inode_info *pipe = i->pipe;
1050 size_t off = i->iov_offset;
1052 size_t n = off - pipe->bufs[idx].offset;
1058 if (!unroll && idx == i->start_idx) {
1063 idx = pipe->buffers - 1;
1064 off = pipe->bufs[idx].offset + pipe->bufs[idx].len;
1066 i->iov_offset = off;
1071 if (unlikely(iov_iter_is_discard(i)))
1073 if (unroll <= i->iov_offset) {
1074 i->iov_offset -= unroll;
1077 unroll -= i->iov_offset;
1078 if (iov_iter_is_bvec(i)) {
1079 const struct bio_vec *bvec = i->bvec;
1081 size_t n = (--bvec)->bv_len;
1085 i->iov_offset = n - unroll;
1090 } else { /* same logics for iovec and kvec */
1091 const struct iovec *iov = i->iov;
1093 size_t n = (--iov)->iov_len;
1097 i->iov_offset = n - unroll;
1104 EXPORT_SYMBOL(iov_iter_revert);
1107 * Return the count of just the current iov_iter segment.
1109 size_t iov_iter_single_seg_count(const struct iov_iter *i)
1111 if (unlikely(iov_iter_is_pipe(i)))
1112 return i->count; // it is a silly place, anyway
1113 if (i->nr_segs == 1)
1115 if (unlikely(iov_iter_is_discard(i)))
1117 else if (iov_iter_is_bvec(i))
1118 return min(i->count, i->bvec->bv_len - i->iov_offset);
1120 return min(i->count, i->iov->iov_len - i->iov_offset);
1122 EXPORT_SYMBOL(iov_iter_single_seg_count);
1124 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
1125 const struct kvec *kvec, unsigned long nr_segs,
1128 WARN_ON(direction & ~(READ | WRITE));
1129 i->type = ITER_KVEC | (direction & (READ | WRITE));
1131 i->nr_segs = nr_segs;
1135 EXPORT_SYMBOL(iov_iter_kvec);
1137 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1138 const struct bio_vec *bvec, unsigned long nr_segs,
1141 WARN_ON(direction & ~(READ | WRITE));
1142 i->type = ITER_BVEC | (direction & (READ | WRITE));
1144 i->nr_segs = nr_segs;
1148 EXPORT_SYMBOL(iov_iter_bvec);
1150 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1151 struct pipe_inode_info *pipe,
1154 BUG_ON(direction != READ);
1155 WARN_ON(pipe->nrbufs == pipe->buffers);
1156 i->type = ITER_PIPE | READ;
1158 i->idx = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1161 i->start_idx = i->idx;
1163 EXPORT_SYMBOL(iov_iter_pipe);
1166 * iov_iter_discard - Initialise an I/O iterator that discards data
1167 * @i: The iterator to initialise.
1168 * @direction: The direction of the transfer.
1169 * @count: The size of the I/O buffer in bytes.
1171 * Set up an I/O iterator that just discards everything that's written to it.
1172 * It's only available as a READ iterator.
1174 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1176 BUG_ON(direction != READ);
1177 i->type = ITER_DISCARD | READ;
1181 EXPORT_SYMBOL(iov_iter_discard);
1183 unsigned long iov_iter_alignment(const struct iov_iter *i)
1185 unsigned long res = 0;
1186 size_t size = i->count;
1188 if (unlikely(iov_iter_is_pipe(i))) {
1189 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->idx]))
1190 return size | i->iov_offset;
1193 iterate_all_kinds(i, size, v,
1194 (res |= (unsigned long)v.iov_base | v.iov_len, 0),
1195 res |= v.bv_offset | v.bv_len,
1196 res |= (unsigned long)v.iov_base | v.iov_len
1200 EXPORT_SYMBOL(iov_iter_alignment);
1202 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1204 unsigned long res = 0;
1205 size_t size = i->count;
1207 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1212 iterate_all_kinds(i, size, v,
1213 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1214 (size != v.iov_len ? size : 0), 0),
1215 (res |= (!res ? 0 : (unsigned long)v.bv_offset) |
1216 (size != v.bv_len ? size : 0)),
1217 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1218 (size != v.iov_len ? size : 0))
1222 EXPORT_SYMBOL(iov_iter_gap_alignment);
1224 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1226 struct page **pages,
1230 struct pipe_inode_info *pipe = i->pipe;
1231 ssize_t n = push_pipe(i, maxsize, &idx, start);
1238 get_page(*pages++ = pipe->bufs[idx].page);
1239 idx = next_idx(idx, pipe);
1246 static ssize_t pipe_get_pages(struct iov_iter *i,
1247 struct page **pages, size_t maxsize, unsigned maxpages,
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 capacity = min(npages,maxpages) * PAGE_SIZE - *start;
1265 return __pipe_get_pages(i, min(maxsize, capacity), pages, idx, start);
1268 ssize_t iov_iter_get_pages(struct iov_iter *i,
1269 struct page **pages, size_t maxsize, unsigned maxpages,
1272 if (maxsize > i->count)
1275 if (unlikely(iov_iter_is_pipe(i)))
1276 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1277 if (unlikely(iov_iter_is_discard(i)))
1280 iterate_all_kinds(i, maxsize, v, ({
1281 unsigned long addr = (unsigned long)v.iov_base;
1282 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1286 if (len > maxpages * PAGE_SIZE)
1287 len = maxpages * PAGE_SIZE;
1288 addr &= ~(PAGE_SIZE - 1);
1289 n = DIV_ROUND_UP(len, PAGE_SIZE);
1290 res = get_user_pages_fast(addr, n,
1291 iov_iter_rw(i) != WRITE ? FOLL_WRITE : 0,
1293 if (unlikely(res <= 0))
1295 return (res == n ? len : res * PAGE_SIZE) - *start;
1297 /* can't be more than PAGE_SIZE */
1298 *start = v.bv_offset;
1299 get_page(*pages = v.bv_page);
1307 EXPORT_SYMBOL(iov_iter_get_pages);
1309 static struct page **get_pages_array(size_t n)
1311 return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1314 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1315 struct page ***pages, size_t maxsize,
1329 data_start(i, &idx, start);
1330 /* some of this one + all after this one */
1331 npages = ((i->pipe->curbuf - idx - 1) & (i->pipe->buffers - 1)) + 1;
1332 n = npages * PAGE_SIZE - *start;
1336 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1337 p = get_pages_array(npages);
1340 n = __pipe_get_pages(i, maxsize, p, idx, start);
1348 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1349 struct page ***pages, size_t maxsize,
1354 if (maxsize > i->count)
1357 if (unlikely(iov_iter_is_pipe(i)))
1358 return pipe_get_pages_alloc(i, pages, maxsize, start);
1359 if (unlikely(iov_iter_is_discard(i)))
1362 iterate_all_kinds(i, maxsize, v, ({
1363 unsigned long addr = (unsigned long)v.iov_base;
1364 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1368 addr &= ~(PAGE_SIZE - 1);
1369 n = DIV_ROUND_UP(len, PAGE_SIZE);
1370 p = get_pages_array(n);
1373 res = get_user_pages_fast(addr, n,
1374 iov_iter_rw(i) != WRITE ? FOLL_WRITE : 0, p);
1375 if (unlikely(res <= 0)) {
1381 return (res == n ? len : res * PAGE_SIZE) - *start;
1383 /* can't be more than PAGE_SIZE */
1384 *start = v.bv_offset;
1385 *pages = p = get_pages_array(1);
1388 get_page(*p = v.bv_page);
1396 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1398 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1405 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1409 iterate_and_advance(i, bytes, v, ({
1411 next = csum_and_copy_from_user(v.iov_base,
1412 (to += v.iov_len) - v.iov_len,
1413 v.iov_len, 0, &err);
1415 sum = csum_block_add(sum, next, off);
1418 err ? v.iov_len : 0;
1420 char *p = kmap_atomic(v.bv_page);
1421 sum = csum_and_memcpy((to += v.bv_len) - v.bv_len,
1422 p + v.bv_offset, v.bv_len,
1427 sum = csum_and_memcpy((to += v.iov_len) - v.iov_len,
1428 v.iov_base, v.iov_len,
1436 EXPORT_SYMBOL(csum_and_copy_from_iter);
1438 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum,
1445 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1449 if (unlikely(i->count < bytes))
1451 iterate_all_kinds(i, bytes, v, ({
1453 next = csum_and_copy_from_user(v.iov_base,
1454 (to += v.iov_len) - v.iov_len,
1455 v.iov_len, 0, &err);
1458 sum = csum_block_add(sum, next, off);
1462 char *p = kmap_atomic(v.bv_page);
1463 sum = csum_and_memcpy((to += v.bv_len) - v.bv_len,
1464 p + v.bv_offset, v.bv_len,
1469 sum = csum_and_memcpy((to += v.iov_len) - v.iov_len,
1470 v.iov_base, v.iov_len,
1476 iov_iter_advance(i, bytes);
1479 EXPORT_SYMBOL(csum_and_copy_from_iter_full);
1481 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1484 struct csum_state *csstate = _csstate;
1485 const char *from = addr;
1489 if (unlikely(iov_iter_is_pipe(i)))
1490 return csum_and_copy_to_pipe_iter(addr, bytes, _csstate, i);
1492 sum = csstate->csum;
1494 if (unlikely(iov_iter_is_discard(i))) {
1495 WARN_ON(1); /* for now */
1498 iterate_and_advance(i, bytes, v, ({
1500 next = csum_and_copy_to_user((from += v.iov_len) - v.iov_len,
1502 v.iov_len, 0, &err);
1504 sum = csum_block_add(sum, next, off);
1507 err ? v.iov_len : 0;
1509 char *p = kmap_atomic(v.bv_page);
1510 sum = csum_and_memcpy(p + v.bv_offset,
1511 (from += v.bv_len) - v.bv_len,
1512 v.bv_len, sum, off);
1516 sum = csum_and_memcpy(v.iov_base,
1517 (from += v.iov_len) - v.iov_len,
1518 v.iov_len, sum, off);
1522 csstate->csum = sum;
1526 EXPORT_SYMBOL(csum_and_copy_to_iter);
1528 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1531 #ifdef CONFIG_CRYPTO
1532 struct ahash_request *hash = hashp;
1533 struct scatterlist sg;
1536 copied = copy_to_iter(addr, bytes, i);
1537 sg_init_one(&sg, addr, copied);
1538 ahash_request_set_crypt(hash, &sg, NULL, copied);
1539 crypto_ahash_update(hash);
1545 EXPORT_SYMBOL(hash_and_copy_to_iter);
1547 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1549 size_t size = i->count;
1554 if (unlikely(iov_iter_is_discard(i)))
1557 if (unlikely(iov_iter_is_pipe(i))) {
1558 struct pipe_inode_info *pipe = i->pipe;
1565 data_start(i, &idx, &off);
1566 /* some of this one + all after this one */
1567 npages = ((pipe->curbuf - idx - 1) & (pipe->buffers - 1)) + 1;
1568 if (npages >= maxpages)
1570 } else iterate_all_kinds(i, size, v, ({
1571 unsigned long p = (unsigned long)v.iov_base;
1572 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1574 if (npages >= maxpages)
1578 if (npages >= maxpages)
1581 unsigned long p = (unsigned long)v.iov_base;
1582 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1584 if (npages >= maxpages)
1590 EXPORT_SYMBOL(iov_iter_npages);
1592 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1595 if (unlikely(iov_iter_is_pipe(new))) {
1599 if (unlikely(iov_iter_is_discard(new)))
1601 if (iov_iter_is_bvec(new))
1602 return new->bvec = kmemdup(new->bvec,
1603 new->nr_segs * sizeof(struct bio_vec),
1606 /* iovec and kvec have identical layout */
1607 return new->iov = kmemdup(new->iov,
1608 new->nr_segs * sizeof(struct iovec),
1611 EXPORT_SYMBOL(dup_iter);
1614 * import_iovec() - Copy an array of &struct iovec from userspace
1615 * into the kernel, check that it is valid, and initialize a new
1616 * &struct iov_iter iterator to access it.
1618 * @type: One of %READ or %WRITE.
1619 * @uvector: Pointer to the userspace array.
1620 * @nr_segs: Number of elements in userspace array.
1621 * @fast_segs: Number of elements in @iov.
1622 * @iov: (input and output parameter) Pointer to pointer to (usually small
1623 * on-stack) kernel array.
1624 * @i: Pointer to iterator that will be initialized on success.
1626 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1627 * then this function places %NULL in *@iov on return. Otherwise, a new
1628 * array will be allocated and the result placed in *@iov. This means that
1629 * the caller may call kfree() on *@iov regardless of whether the small
1630 * on-stack array was used or not (and regardless of whether this function
1631 * returns an error or not).
1633 * Return: Negative error code on error, bytes imported on success
1635 ssize_t import_iovec(int type, const struct iovec __user * uvector,
1636 unsigned nr_segs, unsigned fast_segs,
1637 struct iovec **iov, struct iov_iter *i)
1641 n = rw_copy_check_uvector(type, uvector, nr_segs, fast_segs,
1649 iov_iter_init(i, type, p, nr_segs, n);
1650 *iov = p == *iov ? NULL : p;
1653 EXPORT_SYMBOL(import_iovec);
1655 #ifdef CONFIG_COMPAT
1656 #include <linux/compat.h>
1658 ssize_t compat_import_iovec(int type,
1659 const struct compat_iovec __user * uvector,
1660 unsigned nr_segs, unsigned fast_segs,
1661 struct iovec **iov, struct iov_iter *i)
1665 n = compat_rw_copy_check_uvector(type, uvector, nr_segs, fast_segs,
1673 iov_iter_init(i, type, p, nr_segs, n);
1674 *iov = p == *iov ? NULL : p;
1679 int import_single_range(int rw, void __user *buf, size_t len,
1680 struct iovec *iov, struct iov_iter *i)
1682 if (len > MAX_RW_COUNT)
1684 if (unlikely(!access_ok(buf, len)))
1687 iov->iov_base = buf;
1689 iov_iter_init(i, rw, iov, 1, len);
1692 EXPORT_SYMBOL(import_single_range);
1694 int iov_iter_for_each_range(struct iov_iter *i, size_t bytes,
1695 int (*f)(struct kvec *vec, void *context),
1703 iterate_all_kinds(i, bytes, v, -EINVAL, ({
1704 w.iov_base = kmap(v.bv_page) + v.bv_offset;
1705 w.iov_len = v.bv_len;
1706 err = f(&w, context);
1710 err = f(&w, context);})
1714 EXPORT_SYMBOL(iov_iter_for_each_range);