1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <crypto/hash.h>
3 #include <linux/export.h>
4 #include <linux/bvec.h>
5 #include <linux/fault-inject-usercopy.h>
7 #include <linux/pagemap.h>
8 #include <linux/slab.h>
9 #include <linux/vmalloc.h>
10 #include <linux/splice.h>
11 #include <linux/compat.h>
12 #include <net/checksum.h>
13 #include <linux/scatterlist.h>
14 #include <linux/instrumented.h>
16 #define PIPE_PARANOIA /* for now */
18 #define iterate_iovec(i, n, __v, __p, skip, STEP) { \
22 __v.iov_len = min(n, __p->iov_len - skip); \
23 if (likely(__v.iov_len)) { \
24 __v.iov_base = __p->iov_base + skip; \
26 __v.iov_len -= left; \
27 skip += __v.iov_len; \
32 while (unlikely(!left && n)) { \
34 __v.iov_len = min(n, __p->iov_len); \
35 if (unlikely(!__v.iov_len)) \
37 __v.iov_base = __p->iov_base; \
39 __v.iov_len -= left; \
46 #define iterate_kvec(i, n, __v, __p, skip, STEP) { \
49 __v.iov_len = min(n, __p->iov_len - skip); \
50 if (likely(__v.iov_len)) { \
51 __v.iov_base = __p->iov_base + skip; \
53 skip += __v.iov_len; \
56 while (unlikely(n)) { \
58 __v.iov_len = min(n, __p->iov_len); \
59 if (unlikely(!__v.iov_len)) \
61 __v.iov_base = __p->iov_base; \
69 #define iterate_bvec(i, n, __v, __bi, skip, STEP) { \
70 struct bvec_iter __start; \
71 __start.bi_size = n; \
72 __start.bi_bvec_done = skip; \
74 for_each_bvec(__v, i->bvec, __bi, __start) { \
81 #define iterate_all_kinds(i, n, v, I, B, K) { \
83 size_t skip = i->iov_offset; \
84 if (unlikely(i->type & ITER_BVEC)) { \
86 struct bvec_iter __bi; \
87 iterate_bvec(i, n, v, __bi, skip, (B)) \
88 } else if (unlikely(i->type & ITER_KVEC)) { \
89 const struct kvec *kvec; \
91 iterate_kvec(i, n, v, kvec, skip, (K)) \
92 } else if (unlikely(i->type & ITER_DISCARD)) { \
94 const struct iovec *iov; \
96 iterate_iovec(i, n, v, iov, skip, (I)) \
101 #define iterate_and_advance(i, n, v, I, B, K) { \
102 if (unlikely(i->count < n)) \
105 size_t skip = i->iov_offset; \
106 if (unlikely(i->type & ITER_BVEC)) { \
107 const struct bio_vec *bvec = i->bvec; \
109 struct bvec_iter __bi; \
110 iterate_bvec(i, n, v, __bi, skip, (B)) \
111 i->bvec = __bvec_iter_bvec(i->bvec, __bi); \
112 i->nr_segs -= i->bvec - bvec; \
113 skip = __bi.bi_bvec_done; \
114 } else if (unlikely(i->type & ITER_KVEC)) { \
115 const struct kvec *kvec; \
117 iterate_kvec(i, n, v, kvec, skip, (K)) \
118 if (skip == kvec->iov_len) { \
122 i->nr_segs -= kvec - i->kvec; \
124 } else if (unlikely(i->type & ITER_DISCARD)) { \
127 const struct iovec *iov; \
129 iterate_iovec(i, n, v, iov, skip, (I)) \
130 if (skip == iov->iov_len) { \
134 i->nr_segs -= iov - i->iov; \
138 i->iov_offset = skip; \
142 static int copyout(void __user *to, const void *from, size_t n)
144 if (should_fail_usercopy())
146 if (access_ok(to, n)) {
147 instrument_copy_to_user(to, from, n);
148 n = raw_copy_to_user(to, from, n);
153 static int copyin(void *to, const void __user *from, size_t n)
155 if (should_fail_usercopy())
157 if (access_ok(from, n)) {
158 instrument_copy_from_user(to, from, n);
159 n = raw_copy_from_user(to, from, n);
164 static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes,
167 size_t skip, copy, left, wanted;
168 const struct iovec *iov;
172 if (unlikely(bytes > i->count))
175 if (unlikely(!bytes))
181 skip = i->iov_offset;
182 buf = iov->iov_base + skip;
183 copy = min(bytes, iov->iov_len - skip);
185 if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_writeable(buf, copy)) {
186 kaddr = kmap_atomic(page);
187 from = kaddr + offset;
189 /* first chunk, usually the only one */
190 left = copyout(buf, from, copy);
196 while (unlikely(!left && bytes)) {
199 copy = min(bytes, iov->iov_len);
200 left = copyout(buf, from, copy);
206 if (likely(!bytes)) {
207 kunmap_atomic(kaddr);
210 offset = from - kaddr;
212 kunmap_atomic(kaddr);
213 copy = min(bytes, iov->iov_len - skip);
215 /* Too bad - revert to non-atomic kmap */
218 from = kaddr + offset;
219 left = copyout(buf, from, copy);
224 while (unlikely(!left && bytes)) {
227 copy = min(bytes, iov->iov_len);
228 left = copyout(buf, from, copy);
237 if (skip == iov->iov_len) {
241 i->count -= wanted - bytes;
242 i->nr_segs -= iov - i->iov;
244 i->iov_offset = skip;
245 return wanted - bytes;
248 static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes,
251 size_t skip, copy, left, wanted;
252 const struct iovec *iov;
256 if (unlikely(bytes > i->count))
259 if (unlikely(!bytes))
265 skip = i->iov_offset;
266 buf = iov->iov_base + skip;
267 copy = min(bytes, iov->iov_len - skip);
269 if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_readable(buf, copy)) {
270 kaddr = kmap_atomic(page);
273 /* first chunk, usually the only one */
274 left = copyin(to, buf, copy);
280 while (unlikely(!left && bytes)) {
283 copy = min(bytes, iov->iov_len);
284 left = copyin(to, buf, copy);
290 if (likely(!bytes)) {
291 kunmap_atomic(kaddr);
296 kunmap_atomic(kaddr);
297 copy = min(bytes, iov->iov_len - skip);
299 /* Too bad - revert to non-atomic kmap */
303 left = copyin(to, buf, copy);
308 while (unlikely(!left && bytes)) {
311 copy = min(bytes, iov->iov_len);
312 left = copyin(to, buf, copy);
321 if (skip == iov->iov_len) {
325 i->count -= wanted - bytes;
326 i->nr_segs -= iov - i->iov;
328 i->iov_offset = skip;
329 return wanted - bytes;
333 static bool sanity(const struct iov_iter *i)
335 struct pipe_inode_info *pipe = i->pipe;
336 unsigned int p_head = pipe->head;
337 unsigned int p_tail = pipe->tail;
338 unsigned int p_mask = pipe->ring_size - 1;
339 unsigned int p_occupancy = pipe_occupancy(p_head, p_tail);
340 unsigned int i_head = i->head;
344 struct pipe_buffer *p;
345 if (unlikely(p_occupancy == 0))
346 goto Bad; // pipe must be non-empty
347 if (unlikely(i_head != p_head - 1))
348 goto Bad; // must be at the last buffer...
350 p = &pipe->bufs[i_head & p_mask];
351 if (unlikely(p->offset + p->len != i->iov_offset))
352 goto Bad; // ... at the end of segment
354 if (i_head != p_head)
355 goto Bad; // must be right after the last buffer
359 printk(KERN_ERR "idx = %d, offset = %zd\n", i_head, i->iov_offset);
360 printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n",
361 p_head, p_tail, pipe->ring_size);
362 for (idx = 0; idx < pipe->ring_size; idx++)
363 printk(KERN_ERR "[%p %p %d %d]\n",
365 pipe->bufs[idx].page,
366 pipe->bufs[idx].offset,
367 pipe->bufs[idx].len);
372 #define sanity(i) true
375 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
378 struct pipe_inode_info *pipe = i->pipe;
379 struct pipe_buffer *buf;
380 unsigned int p_tail = pipe->tail;
381 unsigned int p_mask = pipe->ring_size - 1;
382 unsigned int i_head = i->head;
385 if (unlikely(bytes > i->count))
388 if (unlikely(!bytes))
395 buf = &pipe->bufs[i_head & p_mask];
397 if (offset == off && buf->page == page) {
398 /* merge with the last one */
400 i->iov_offset += bytes;
404 buf = &pipe->bufs[i_head & p_mask];
406 if (pipe_full(i_head, p_tail, pipe->max_usage))
409 buf->ops = &page_cache_pipe_buf_ops;
413 buf->offset = offset;
416 pipe->head = i_head + 1;
417 i->iov_offset = offset + bytes;
425 * Fault in one or more iovecs of the given iov_iter, to a maximum length of
426 * bytes. For each iovec, fault in each page that constitutes the iovec.
428 * Return 0 on success, or non-zero if the memory could not be accessed (i.e.
429 * because it is an invalid address).
431 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
433 size_t skip = i->iov_offset;
434 const struct iovec *iov;
438 if (iter_is_iovec(i)) {
439 iterate_iovec(i, bytes, v, iov, skip, ({
440 err = fault_in_pages_readable(v.iov_base, v.iov_len);
447 EXPORT_SYMBOL(iov_iter_fault_in_readable);
449 void iov_iter_init(struct iov_iter *i, unsigned int direction,
450 const struct iovec *iov, unsigned long nr_segs,
453 WARN_ON(direction & ~(READ | WRITE));
454 direction &= READ | WRITE;
456 /* It will get better. Eventually... */
457 if (uaccess_kernel()) {
458 i->type = ITER_KVEC | direction;
459 i->kvec = (struct kvec *)iov;
461 i->type = ITER_IOVEC | direction;
464 i->nr_segs = nr_segs;
468 EXPORT_SYMBOL(iov_iter_init);
470 static void memcpy_from_page(char *to, struct page *page, size_t offset, size_t len)
472 char *from = kmap_atomic(page);
473 memcpy(to, from + offset, len);
477 static void memcpy_to_page(struct page *page, size_t offset, const char *from, size_t len)
479 char *to = kmap_atomic(page);
480 memcpy(to + offset, from, len);
484 static void memzero_page(struct page *page, size_t offset, size_t len)
486 char *addr = kmap_atomic(page);
487 memset(addr + offset, 0, len);
491 static inline bool allocated(struct pipe_buffer *buf)
493 return buf->ops == &default_pipe_buf_ops;
496 static inline void data_start(const struct iov_iter *i,
497 unsigned int *iter_headp, size_t *offp)
499 unsigned int p_mask = i->pipe->ring_size - 1;
500 unsigned int iter_head = i->head;
501 size_t off = i->iov_offset;
503 if (off && (!allocated(&i->pipe->bufs[iter_head & p_mask]) ||
508 *iter_headp = iter_head;
512 static size_t push_pipe(struct iov_iter *i, size_t size,
513 int *iter_headp, size_t *offp)
515 struct pipe_inode_info *pipe = i->pipe;
516 unsigned int p_tail = pipe->tail;
517 unsigned int p_mask = pipe->ring_size - 1;
518 unsigned int iter_head;
522 if (unlikely(size > i->count))
528 data_start(i, &iter_head, &off);
529 *iter_headp = iter_head;
532 left -= PAGE_SIZE - off;
534 pipe->bufs[iter_head & p_mask].len += size;
537 pipe->bufs[iter_head & p_mask].len = PAGE_SIZE;
540 while (!pipe_full(iter_head, p_tail, pipe->max_usage)) {
541 struct pipe_buffer *buf = &pipe->bufs[iter_head & p_mask];
542 struct page *page = alloc_page(GFP_USER);
546 buf->ops = &default_pipe_buf_ops;
550 buf->len = min_t(ssize_t, left, PAGE_SIZE);
553 pipe->head = iter_head;
561 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
564 struct pipe_inode_info *pipe = i->pipe;
565 unsigned int p_mask = pipe->ring_size - 1;
572 bytes = n = push_pipe(i, bytes, &i_head, &off);
576 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
577 memcpy_to_page(pipe->bufs[i_head & p_mask].page, off, addr, chunk);
579 i->iov_offset = off + chunk;
589 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
590 __wsum sum, size_t off)
592 __wsum next = csum_partial_copy_nocheck(from, to, len);
593 return csum_block_add(sum, next, off);
596 static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
597 struct csum_state *csstate,
600 struct pipe_inode_info *pipe = i->pipe;
601 unsigned int p_mask = pipe->ring_size - 1;
602 __wsum sum = csstate->csum;
603 size_t off = csstate->off;
610 bytes = n = push_pipe(i, bytes, &i_head, &r);
614 size_t chunk = min_t(size_t, n, PAGE_SIZE - r);
615 char *p = kmap_atomic(pipe->bufs[i_head & p_mask].page);
616 sum = csum_and_memcpy(p + r, addr, chunk, sum, off);
619 i->iov_offset = r + chunk;
632 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
634 const char *from = addr;
635 if (unlikely(iov_iter_is_pipe(i)))
636 return copy_pipe_to_iter(addr, bytes, i);
637 if (iter_is_iovec(i))
639 iterate_and_advance(i, bytes, v,
640 copyout(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
641 memcpy_to_page(v.bv_page, v.bv_offset,
642 (from += v.bv_len) - v.bv_len, v.bv_len),
643 memcpy(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len)
648 EXPORT_SYMBOL(_copy_to_iter);
650 #ifdef CONFIG_ARCH_HAS_COPY_MC
651 static int copyout_mc(void __user *to, const void *from, size_t n)
653 if (access_ok(to, n)) {
654 instrument_copy_to_user(to, from, n);
655 n = copy_mc_to_user((__force void *) to, from, n);
660 static unsigned long copy_mc_to_page(struct page *page, size_t offset,
661 const char *from, size_t len)
666 to = kmap_atomic(page);
667 ret = copy_mc_to_kernel(to + offset, from, len);
673 static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes,
676 struct pipe_inode_info *pipe = i->pipe;
677 unsigned int p_mask = pipe->ring_size - 1;
679 size_t n, off, xfer = 0;
684 bytes = n = push_pipe(i, bytes, &i_head, &off);
688 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
691 rem = copy_mc_to_page(pipe->bufs[i_head & p_mask].page,
694 i->iov_offset = off + chunk - rem;
708 * _copy_mc_to_iter - copy to iter with source memory error exception handling
709 * @addr: source kernel address
710 * @bytes: total transfer length
711 * @iter: destination iterator
713 * The pmem driver deploys this for the dax operation
714 * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
715 * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
716 * successfully copied.
718 * The main differences between this and typical _copy_to_iter().
720 * * Typical tail/residue handling after a fault retries the copy
721 * byte-by-byte until the fault happens again. Re-triggering machine
722 * checks is potentially fatal so the implementation uses source
723 * alignment and poison alignment assumptions to avoid re-triggering
724 * hardware exceptions.
726 * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
727 * Compare to copy_to_iter() where only ITER_IOVEC attempts might return
730 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
732 const char *from = addr;
733 unsigned long rem, curr_addr, s_addr = (unsigned long) addr;
735 if (unlikely(iov_iter_is_pipe(i)))
736 return copy_mc_pipe_to_iter(addr, bytes, i);
737 if (iter_is_iovec(i))
739 iterate_and_advance(i, bytes, v,
740 copyout_mc(v.iov_base, (from += v.iov_len) - v.iov_len,
743 rem = copy_mc_to_page(v.bv_page, v.bv_offset,
744 (from += v.bv_len) - v.bv_len, v.bv_len);
746 curr_addr = (unsigned long) from;
747 bytes = curr_addr - s_addr - rem;
752 rem = copy_mc_to_kernel(v.iov_base, (from += v.iov_len)
753 - v.iov_len, v.iov_len);
755 curr_addr = (unsigned long) from;
756 bytes = curr_addr - s_addr - rem;
764 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
765 #endif /* CONFIG_ARCH_HAS_COPY_MC */
767 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
770 if (unlikely(iov_iter_is_pipe(i))) {
774 if (iter_is_iovec(i))
776 iterate_and_advance(i, bytes, v,
777 copyin((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
778 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
779 v.bv_offset, v.bv_len),
780 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
785 EXPORT_SYMBOL(_copy_from_iter);
787 bool _copy_from_iter_full(void *addr, size_t bytes, struct iov_iter *i)
790 if (unlikely(iov_iter_is_pipe(i))) {
794 if (unlikely(i->count < bytes))
797 if (iter_is_iovec(i))
799 iterate_all_kinds(i, bytes, v, ({
800 if (copyin((to += v.iov_len) - v.iov_len,
801 v.iov_base, v.iov_len))
804 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
805 v.bv_offset, v.bv_len),
806 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
809 iov_iter_advance(i, bytes);
812 EXPORT_SYMBOL(_copy_from_iter_full);
814 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
817 if (unlikely(iov_iter_is_pipe(i))) {
821 iterate_and_advance(i, bytes, v,
822 __copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
823 v.iov_base, v.iov_len),
824 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
825 v.bv_offset, v.bv_len),
826 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
831 EXPORT_SYMBOL(_copy_from_iter_nocache);
833 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
835 * _copy_from_iter_flushcache - write destination through cpu cache
836 * @addr: destination kernel address
837 * @bytes: total transfer length
838 * @iter: source iterator
840 * The pmem driver arranges for filesystem-dax to use this facility via
841 * dax_copy_from_iter() for ensuring that writes to persistent memory
842 * are flushed through the CPU cache. It is differentiated from
843 * _copy_from_iter_nocache() in that guarantees all data is flushed for
844 * all iterator types. The _copy_from_iter_nocache() only attempts to
845 * bypass the cache for the ITER_IOVEC case, and on some archs may use
846 * instructions that strand dirty-data in the cache.
848 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
851 if (unlikely(iov_iter_is_pipe(i))) {
855 iterate_and_advance(i, bytes, v,
856 __copy_from_user_flushcache((to += v.iov_len) - v.iov_len,
857 v.iov_base, v.iov_len),
858 memcpy_page_flushcache((to += v.bv_len) - v.bv_len, v.bv_page,
859 v.bv_offset, v.bv_len),
860 memcpy_flushcache((to += v.iov_len) - v.iov_len, v.iov_base,
866 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
869 bool _copy_from_iter_full_nocache(void *addr, size_t bytes, struct iov_iter *i)
872 if (unlikely(iov_iter_is_pipe(i))) {
876 if (unlikely(i->count < bytes))
878 iterate_all_kinds(i, bytes, v, ({
879 if (__copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
880 v.iov_base, v.iov_len))
883 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
884 v.bv_offset, v.bv_len),
885 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
888 iov_iter_advance(i, bytes);
891 EXPORT_SYMBOL(_copy_from_iter_full_nocache);
893 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
896 size_t v = n + offset;
899 * The general case needs to access the page order in order
900 * to compute the page size.
901 * However, we mostly deal with order-0 pages and thus can
902 * avoid a possible cache line miss for requests that fit all
905 if (n <= v && v <= PAGE_SIZE)
908 head = compound_head(page);
909 v += (page - head) << PAGE_SHIFT;
911 if (likely(n <= v && v <= (page_size(head))))
917 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
920 if (unlikely(!page_copy_sane(page, offset, bytes)))
922 if (i->type & (ITER_BVEC|ITER_KVEC)) {
923 void *kaddr = kmap_atomic(page);
924 size_t wanted = copy_to_iter(kaddr + offset, bytes, i);
925 kunmap_atomic(kaddr);
927 } else if (unlikely(iov_iter_is_discard(i))) {
928 if (unlikely(i->count < bytes))
932 } else if (likely(!iov_iter_is_pipe(i)))
933 return copy_page_to_iter_iovec(page, offset, bytes, i);
935 return copy_page_to_iter_pipe(page, offset, bytes, i);
937 EXPORT_SYMBOL(copy_page_to_iter);
939 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
942 if (unlikely(!page_copy_sane(page, offset, bytes)))
944 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
948 if (i->type & (ITER_BVEC|ITER_KVEC)) {
949 void *kaddr = kmap_atomic(page);
950 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
951 kunmap_atomic(kaddr);
954 return copy_page_from_iter_iovec(page, offset, bytes, i);
956 EXPORT_SYMBOL(copy_page_from_iter);
958 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
960 struct pipe_inode_info *pipe = i->pipe;
961 unsigned int p_mask = pipe->ring_size - 1;
968 bytes = n = push_pipe(i, bytes, &i_head, &off);
973 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
974 memzero_page(pipe->bufs[i_head & p_mask].page, off, chunk);
976 i->iov_offset = off + chunk;
985 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
987 if (unlikely(iov_iter_is_pipe(i)))
988 return pipe_zero(bytes, i);
989 iterate_and_advance(i, bytes, v,
990 clear_user(v.iov_base, v.iov_len),
991 memzero_page(v.bv_page, v.bv_offset, v.bv_len),
992 memset(v.iov_base, 0, v.iov_len)
997 EXPORT_SYMBOL(iov_iter_zero);
999 size_t iov_iter_copy_from_user_atomic(struct page *page,
1000 struct iov_iter *i, unsigned long offset, size_t bytes)
1002 char *kaddr = kmap_atomic(page), *p = kaddr + offset;
1003 if (unlikely(!page_copy_sane(page, offset, bytes))) {
1004 kunmap_atomic(kaddr);
1007 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1008 kunmap_atomic(kaddr);
1012 iterate_all_kinds(i, bytes, v,
1013 copyin((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
1014 memcpy_from_page((p += v.bv_len) - v.bv_len, v.bv_page,
1015 v.bv_offset, v.bv_len),
1016 memcpy((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
1018 kunmap_atomic(kaddr);
1021 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
1023 static inline void pipe_truncate(struct iov_iter *i)
1025 struct pipe_inode_info *pipe = i->pipe;
1026 unsigned int p_tail = pipe->tail;
1027 unsigned int p_head = pipe->head;
1028 unsigned int p_mask = pipe->ring_size - 1;
1030 if (!pipe_empty(p_head, p_tail)) {
1031 struct pipe_buffer *buf;
1032 unsigned int i_head = i->head;
1033 size_t off = i->iov_offset;
1036 buf = &pipe->bufs[i_head & p_mask];
1037 buf->len = off - buf->offset;
1040 while (p_head != i_head) {
1042 pipe_buf_release(pipe, &pipe->bufs[p_head & p_mask]);
1045 pipe->head = p_head;
1049 static void pipe_advance(struct iov_iter *i, size_t size)
1051 struct pipe_inode_info *pipe = i->pipe;
1052 if (unlikely(i->count < size))
1055 struct pipe_buffer *buf;
1056 unsigned int p_mask = pipe->ring_size - 1;
1057 unsigned int i_head = i->head;
1058 size_t off = i->iov_offset, left = size;
1060 if (off) /* make it relative to the beginning of buffer */
1061 left += off - pipe->bufs[i_head & p_mask].offset;
1063 buf = &pipe->bufs[i_head & p_mask];
1064 if (left <= buf->len)
1070 i->iov_offset = buf->offset + left;
1073 /* ... and discard everything past that point */
1077 void iov_iter_advance(struct iov_iter *i, size_t size)
1079 if (unlikely(iov_iter_is_pipe(i))) {
1080 pipe_advance(i, size);
1083 if (unlikely(iov_iter_is_discard(i))) {
1087 iterate_and_advance(i, size, v, 0, 0, 0)
1089 EXPORT_SYMBOL(iov_iter_advance);
1091 void iov_iter_revert(struct iov_iter *i, size_t unroll)
1095 if (WARN_ON(unroll > MAX_RW_COUNT))
1098 if (unlikely(iov_iter_is_pipe(i))) {
1099 struct pipe_inode_info *pipe = i->pipe;
1100 unsigned int p_mask = pipe->ring_size - 1;
1101 unsigned int i_head = i->head;
1102 size_t off = i->iov_offset;
1104 struct pipe_buffer *b = &pipe->bufs[i_head & p_mask];
1105 size_t n = off - b->offset;
1111 if (!unroll && i_head == i->start_head) {
1116 b = &pipe->bufs[i_head & p_mask];
1117 off = b->offset + b->len;
1119 i->iov_offset = off;
1124 if (unlikely(iov_iter_is_discard(i)))
1126 if (unroll <= i->iov_offset) {
1127 i->iov_offset -= unroll;
1130 unroll -= i->iov_offset;
1131 if (iov_iter_is_bvec(i)) {
1132 const struct bio_vec *bvec = i->bvec;
1134 size_t n = (--bvec)->bv_len;
1138 i->iov_offset = n - unroll;
1143 } else { /* same logics for iovec and kvec */
1144 const struct iovec *iov = i->iov;
1146 size_t n = (--iov)->iov_len;
1150 i->iov_offset = n - unroll;
1157 EXPORT_SYMBOL(iov_iter_revert);
1160 * Return the count of just the current iov_iter segment.
1162 size_t iov_iter_single_seg_count(const struct iov_iter *i)
1164 if (unlikely(iov_iter_is_pipe(i)))
1165 return i->count; // it is a silly place, anyway
1166 if (i->nr_segs == 1)
1168 if (unlikely(iov_iter_is_discard(i)))
1170 else if (iov_iter_is_bvec(i))
1171 return min(i->count, i->bvec->bv_len - i->iov_offset);
1173 return min(i->count, i->iov->iov_len - i->iov_offset);
1175 EXPORT_SYMBOL(iov_iter_single_seg_count);
1177 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
1178 const struct kvec *kvec, unsigned long nr_segs,
1181 WARN_ON(direction & ~(READ | WRITE));
1182 i->type = ITER_KVEC | (direction & (READ | WRITE));
1184 i->nr_segs = nr_segs;
1188 EXPORT_SYMBOL(iov_iter_kvec);
1190 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1191 const struct bio_vec *bvec, unsigned long nr_segs,
1194 WARN_ON(direction & ~(READ | WRITE));
1195 i->type = ITER_BVEC | (direction & (READ | WRITE));
1197 i->nr_segs = nr_segs;
1201 EXPORT_SYMBOL(iov_iter_bvec);
1203 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1204 struct pipe_inode_info *pipe,
1207 BUG_ON(direction != READ);
1208 WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1209 i->type = ITER_PIPE | READ;
1211 i->head = pipe->head;
1214 i->start_head = i->head;
1216 EXPORT_SYMBOL(iov_iter_pipe);
1219 * iov_iter_discard - Initialise an I/O iterator that discards data
1220 * @i: The iterator to initialise.
1221 * @direction: The direction of the transfer.
1222 * @count: The size of the I/O buffer in bytes.
1224 * Set up an I/O iterator that just discards everything that's written to it.
1225 * It's only available as a READ iterator.
1227 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1229 BUG_ON(direction != READ);
1230 i->type = ITER_DISCARD | READ;
1234 EXPORT_SYMBOL(iov_iter_discard);
1236 unsigned long iov_iter_alignment(const struct iov_iter *i)
1238 unsigned long res = 0;
1239 size_t size = i->count;
1241 if (unlikely(iov_iter_is_pipe(i))) {
1242 unsigned int p_mask = i->pipe->ring_size - 1;
1244 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->head & p_mask]))
1245 return size | i->iov_offset;
1248 iterate_all_kinds(i, size, v,
1249 (res |= (unsigned long)v.iov_base | v.iov_len, 0),
1250 res |= v.bv_offset | v.bv_len,
1251 res |= (unsigned long)v.iov_base | v.iov_len
1255 EXPORT_SYMBOL(iov_iter_alignment);
1257 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1259 unsigned long res = 0;
1260 size_t size = i->count;
1262 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1267 iterate_all_kinds(i, size, v,
1268 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1269 (size != v.iov_len ? size : 0), 0),
1270 (res |= (!res ? 0 : (unsigned long)v.bv_offset) |
1271 (size != v.bv_len ? size : 0)),
1272 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1273 (size != v.iov_len ? size : 0))
1277 EXPORT_SYMBOL(iov_iter_gap_alignment);
1279 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1281 struct page **pages,
1285 struct pipe_inode_info *pipe = i->pipe;
1286 unsigned int p_mask = pipe->ring_size - 1;
1287 ssize_t n = push_pipe(i, maxsize, &iter_head, start);
1294 get_page(*pages++ = pipe->bufs[iter_head & p_mask].page);
1302 static ssize_t pipe_get_pages(struct iov_iter *i,
1303 struct page **pages, size_t maxsize, unsigned maxpages,
1306 unsigned int iter_head, npages;
1315 data_start(i, &iter_head, start);
1316 /* Amount of free space: some of this one + all after this one */
1317 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1318 capacity = min(npages, maxpages) * PAGE_SIZE - *start;
1320 return __pipe_get_pages(i, min(maxsize, capacity), pages, iter_head, start);
1323 ssize_t iov_iter_get_pages(struct iov_iter *i,
1324 struct page **pages, size_t maxsize, unsigned maxpages,
1327 if (maxsize > i->count)
1330 if (unlikely(iov_iter_is_pipe(i)))
1331 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1332 if (unlikely(iov_iter_is_discard(i)))
1335 iterate_all_kinds(i, maxsize, v, ({
1336 unsigned long addr = (unsigned long)v.iov_base;
1337 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1341 if (len > maxpages * PAGE_SIZE)
1342 len = maxpages * PAGE_SIZE;
1343 addr &= ~(PAGE_SIZE - 1);
1344 n = DIV_ROUND_UP(len, PAGE_SIZE);
1345 res = get_user_pages_fast(addr, n,
1346 iov_iter_rw(i) != WRITE ? FOLL_WRITE : 0,
1348 if (unlikely(res <= 0))
1350 return (res == n ? len : res * PAGE_SIZE) - *start;
1352 /* can't be more than PAGE_SIZE */
1353 *start = v.bv_offset;
1354 get_page(*pages = v.bv_page);
1362 EXPORT_SYMBOL(iov_iter_get_pages);
1364 static struct page **get_pages_array(size_t n)
1366 return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1369 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1370 struct page ***pages, size_t maxsize,
1374 unsigned int iter_head, npages;
1383 data_start(i, &iter_head, start);
1384 /* Amount of free space: some of this one + all after this one */
1385 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1386 n = npages * PAGE_SIZE - *start;
1390 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1391 p = get_pages_array(npages);
1394 n = __pipe_get_pages(i, maxsize, p, iter_head, start);
1402 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1403 struct page ***pages, size_t maxsize,
1408 if (maxsize > i->count)
1411 if (unlikely(iov_iter_is_pipe(i)))
1412 return pipe_get_pages_alloc(i, pages, maxsize, start);
1413 if (unlikely(iov_iter_is_discard(i)))
1416 iterate_all_kinds(i, maxsize, v, ({
1417 unsigned long addr = (unsigned long)v.iov_base;
1418 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1422 addr &= ~(PAGE_SIZE - 1);
1423 n = DIV_ROUND_UP(len, PAGE_SIZE);
1424 p = get_pages_array(n);
1427 res = get_user_pages_fast(addr, n,
1428 iov_iter_rw(i) != WRITE ? FOLL_WRITE : 0, p);
1429 if (unlikely(res <= 0)) {
1435 return (res == n ? len : res * PAGE_SIZE) - *start;
1437 /* can't be more than PAGE_SIZE */
1438 *start = v.bv_offset;
1439 *pages = p = get_pages_array(1);
1442 get_page(*p = v.bv_page);
1450 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1452 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1459 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1463 iterate_and_advance(i, bytes, v, ({
1464 next = csum_and_copy_from_user(v.iov_base,
1465 (to += v.iov_len) - v.iov_len,
1468 sum = csum_block_add(sum, next, off);
1471 next ? 0 : v.iov_len;
1473 char *p = kmap_atomic(v.bv_page);
1474 sum = csum_and_memcpy((to += v.bv_len) - v.bv_len,
1475 p + v.bv_offset, v.bv_len,
1480 sum = csum_and_memcpy((to += v.iov_len) - v.iov_len,
1481 v.iov_base, v.iov_len,
1489 EXPORT_SYMBOL(csum_and_copy_from_iter);
1491 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum,
1498 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1502 if (unlikely(i->count < bytes))
1504 iterate_all_kinds(i, bytes, v, ({
1505 next = csum_and_copy_from_user(v.iov_base,
1506 (to += v.iov_len) - v.iov_len,
1510 sum = csum_block_add(sum, next, off);
1514 char *p = kmap_atomic(v.bv_page);
1515 sum = csum_and_memcpy((to += v.bv_len) - v.bv_len,
1516 p + v.bv_offset, v.bv_len,
1521 sum = csum_and_memcpy((to += v.iov_len) - v.iov_len,
1522 v.iov_base, v.iov_len,
1528 iov_iter_advance(i, bytes);
1531 EXPORT_SYMBOL(csum_and_copy_from_iter_full);
1533 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1536 struct csum_state *csstate = _csstate;
1537 const char *from = addr;
1541 if (unlikely(iov_iter_is_pipe(i)))
1542 return csum_and_copy_to_pipe_iter(addr, bytes, _csstate, i);
1544 sum = csstate->csum;
1546 if (unlikely(iov_iter_is_discard(i))) {
1547 WARN_ON(1); /* for now */
1550 iterate_and_advance(i, bytes, v, ({
1551 next = csum_and_copy_to_user((from += v.iov_len) - v.iov_len,
1555 sum = csum_block_add(sum, next, off);
1558 next ? 0 : v.iov_len;
1560 char *p = kmap_atomic(v.bv_page);
1561 sum = csum_and_memcpy(p + v.bv_offset,
1562 (from += v.bv_len) - v.bv_len,
1563 v.bv_len, sum, off);
1567 sum = csum_and_memcpy(v.iov_base,
1568 (from += v.iov_len) - v.iov_len,
1569 v.iov_len, sum, off);
1573 csstate->csum = sum;
1577 EXPORT_SYMBOL(csum_and_copy_to_iter);
1579 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1582 #ifdef CONFIG_CRYPTO_HASH
1583 struct ahash_request *hash = hashp;
1584 struct scatterlist sg;
1587 copied = copy_to_iter(addr, bytes, i);
1588 sg_init_one(&sg, addr, copied);
1589 ahash_request_set_crypt(hash, &sg, NULL, copied);
1590 crypto_ahash_update(hash);
1596 EXPORT_SYMBOL(hash_and_copy_to_iter);
1598 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1600 size_t size = i->count;
1605 if (unlikely(iov_iter_is_discard(i)))
1608 if (unlikely(iov_iter_is_pipe(i))) {
1609 struct pipe_inode_info *pipe = i->pipe;
1610 unsigned int iter_head;
1616 data_start(i, &iter_head, &off);
1617 /* some of this one + all after this one */
1618 npages = pipe_space_for_user(iter_head, pipe->tail, pipe);
1619 if (npages >= maxpages)
1621 } else iterate_all_kinds(i, size, v, ({
1622 unsigned long p = (unsigned long)v.iov_base;
1623 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1625 if (npages >= maxpages)
1629 if (npages >= maxpages)
1632 unsigned long p = (unsigned long)v.iov_base;
1633 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1635 if (npages >= maxpages)
1641 EXPORT_SYMBOL(iov_iter_npages);
1643 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1646 if (unlikely(iov_iter_is_pipe(new))) {
1650 if (unlikely(iov_iter_is_discard(new)))
1652 if (iov_iter_is_bvec(new))
1653 return new->bvec = kmemdup(new->bvec,
1654 new->nr_segs * sizeof(struct bio_vec),
1657 /* iovec and kvec have identical layout */
1658 return new->iov = kmemdup(new->iov,
1659 new->nr_segs * sizeof(struct iovec),
1662 EXPORT_SYMBOL(dup_iter);
1664 static int copy_compat_iovec_from_user(struct iovec *iov,
1665 const struct iovec __user *uvec, unsigned long nr_segs)
1667 const struct compat_iovec __user *uiov =
1668 (const struct compat_iovec __user *)uvec;
1669 int ret = -EFAULT, i;
1671 if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1674 for (i = 0; i < nr_segs; i++) {
1678 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1679 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1681 /* check for compat_size_t not fitting in compat_ssize_t .. */
1686 iov[i].iov_base = compat_ptr(buf);
1687 iov[i].iov_len = len;
1696 static int copy_iovec_from_user(struct iovec *iov,
1697 const struct iovec __user *uvec, unsigned long nr_segs)
1701 if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1703 for (seg = 0; seg < nr_segs; seg++) {
1704 if ((ssize_t)iov[seg].iov_len < 0)
1711 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1712 unsigned long nr_segs, unsigned long fast_segs,
1713 struct iovec *fast_iov, bool compat)
1715 struct iovec *iov = fast_iov;
1719 * SuS says "The readv() function *may* fail if the iovcnt argument was
1720 * less than or equal to 0, or greater than {IOV_MAX}. Linux has
1721 * traditionally returned zero for zero segments, so...
1725 if (nr_segs > UIO_MAXIOV)
1726 return ERR_PTR(-EINVAL);
1727 if (nr_segs > fast_segs) {
1728 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1730 return ERR_PTR(-ENOMEM);
1734 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1736 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1738 if (iov != fast_iov)
1740 return ERR_PTR(ret);
1746 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1747 unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1748 struct iov_iter *i, bool compat)
1750 ssize_t total_len = 0;
1754 iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1757 return PTR_ERR(iov);
1761 * According to the Single Unix Specification we should return EINVAL if
1762 * an element length is < 0 when cast to ssize_t or if the total length
1763 * would overflow the ssize_t return value of the system call.
1765 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1768 for (seg = 0; seg < nr_segs; seg++) {
1769 ssize_t len = (ssize_t)iov[seg].iov_len;
1771 if (!access_ok(iov[seg].iov_base, len)) {
1778 if (len > MAX_RW_COUNT - total_len) {
1779 len = MAX_RW_COUNT - total_len;
1780 iov[seg].iov_len = len;
1785 iov_iter_init(i, type, iov, nr_segs, total_len);
1794 * import_iovec() - Copy an array of &struct iovec from userspace
1795 * into the kernel, check that it is valid, and initialize a new
1796 * &struct iov_iter iterator to access it.
1798 * @type: One of %READ or %WRITE.
1799 * @uvec: Pointer to the userspace array.
1800 * @nr_segs: Number of elements in userspace array.
1801 * @fast_segs: Number of elements in @iov.
1802 * @iovp: (input and output parameter) Pointer to pointer to (usually small
1803 * on-stack) kernel array.
1804 * @i: Pointer to iterator that will be initialized on success.
1806 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1807 * then this function places %NULL in *@iov on return. Otherwise, a new
1808 * array will be allocated and the result placed in *@iov. This means that
1809 * the caller may call kfree() on *@iov regardless of whether the small
1810 * on-stack array was used or not (and regardless of whether this function
1811 * returns an error or not).
1813 * Return: Negative error code on error, bytes imported on success
1815 ssize_t import_iovec(int type, const struct iovec __user *uvec,
1816 unsigned nr_segs, unsigned fast_segs,
1817 struct iovec **iovp, struct iov_iter *i)
1819 return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
1820 in_compat_syscall());
1822 EXPORT_SYMBOL(import_iovec);
1824 int import_single_range(int rw, void __user *buf, size_t len,
1825 struct iovec *iov, struct iov_iter *i)
1827 if (len > MAX_RW_COUNT)
1829 if (unlikely(!access_ok(buf, len)))
1832 iov->iov_base = buf;
1834 iov_iter_init(i, rw, iov, 1, len);
1837 EXPORT_SYMBOL(import_single_range);
1839 int iov_iter_for_each_range(struct iov_iter *i, size_t bytes,
1840 int (*f)(struct kvec *vec, void *context),
1848 iterate_all_kinds(i, bytes, v, -EINVAL, ({
1849 w.iov_base = kmap(v.bv_page) + v.bv_offset;
1850 w.iov_len = v.bv_len;
1851 err = f(&w, context);
1855 err = f(&w, context);})
1859 EXPORT_SYMBOL(iov_iter_for_each_range);