1 #include <linux/export.h>
3 #include <linux/pagemap.h>
4 #include <linux/slab.h>
5 #include <linux/vmalloc.h>
7 static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes,
10 size_t skip, copy, left, wanted;
11 const struct iovec *iov;
15 if (unlikely(bytes > i->count))
24 buf = iov->iov_base + skip;
25 copy = min(bytes, iov->iov_len - skip);
27 if (!fault_in_pages_writeable(buf, copy)) {
28 kaddr = kmap_atomic(page);
29 from = kaddr + offset;
31 /* first chunk, usually the only one */
32 left = __copy_to_user_inatomic(buf, from, copy);
38 while (unlikely(!left && bytes)) {
41 copy = min(bytes, iov->iov_len);
42 left = __copy_to_user_inatomic(buf, from, copy);
52 offset = from - kaddr;
55 copy = min(bytes, iov->iov_len - skip);
57 /* Too bad - revert to non-atomic kmap */
59 from = kaddr + offset;
60 left = __copy_to_user(buf, from, copy);
65 while (unlikely(!left && bytes)) {
68 copy = min(bytes, iov->iov_len);
69 left = __copy_to_user(buf, from, copy);
77 if (skip == iov->iov_len) {
81 i->count -= wanted - bytes;
82 i->nr_segs -= iov - i->iov;
85 return wanted - bytes;
88 static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes,
91 size_t skip, copy, left, wanted;
92 const struct iovec *iov;
96 if (unlikely(bytes > i->count))
104 skip = i->iov_offset;
105 buf = iov->iov_base + skip;
106 copy = min(bytes, iov->iov_len - skip);
108 if (!fault_in_pages_readable(buf, copy)) {
109 kaddr = kmap_atomic(page);
112 /* first chunk, usually the only one */
113 left = __copy_from_user_inatomic(to, buf, copy);
119 while (unlikely(!left && bytes)) {
122 copy = min(bytes, iov->iov_len);
123 left = __copy_from_user_inatomic(to, buf, copy);
129 if (likely(!bytes)) {
130 kunmap_atomic(kaddr);
135 kunmap_atomic(kaddr);
136 copy = min(bytes, iov->iov_len - skip);
138 /* Too bad - revert to non-atomic kmap */
141 left = __copy_from_user(to, buf, copy);
146 while (unlikely(!left && bytes)) {
149 copy = min(bytes, iov->iov_len);
150 left = __copy_from_user(to, buf, copy);
158 if (skip == iov->iov_len) {
162 i->count -= wanted - bytes;
163 i->nr_segs -= iov - i->iov;
165 i->iov_offset = skip;
166 return wanted - bytes;
169 static size_t __iovec_copy_from_user_inatomic(char *vaddr,
170 const struct iovec *iov, size_t base, size_t bytes)
172 size_t copied = 0, left = 0;
175 char __user *buf = iov->iov_base + base;
176 int copy = min(bytes, iov->iov_len - base);
179 left = __copy_from_user_inatomic(vaddr, buf, copy);
188 return copied - left;
192 * Copy as much as we can into the page and return the number of bytes which
193 * were successfully copied. If a fault is encountered then return the number of
194 * bytes which were copied.
196 static size_t copy_from_user_atomic_iovec(struct page *page,
197 struct iov_iter *i, unsigned long offset, size_t bytes)
202 kaddr = kmap_atomic(page);
203 if (likely(i->nr_segs == 1)) {
205 char __user *buf = i->iov->iov_base + i->iov_offset;
206 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
207 copied = bytes - left;
209 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
210 i->iov, i->iov_offset, bytes);
212 kunmap_atomic(kaddr);
217 static void advance_iovec(struct iov_iter *i, size_t bytes)
219 BUG_ON(i->count < bytes);
221 if (likely(i->nr_segs == 1)) {
222 i->iov_offset += bytes;
225 const struct iovec *iov = i->iov;
226 size_t base = i->iov_offset;
227 unsigned long nr_segs = i->nr_segs;
230 * The !iov->iov_len check ensures we skip over unlikely
231 * zero-length segments (without overruning the iovec).
233 while (bytes || unlikely(i->count && !iov->iov_len)) {
236 copy = min(bytes, iov->iov_len - base);
237 BUG_ON(!i->count || i->count < copy);
241 if (iov->iov_len == base) {
248 i->iov_offset = base;
249 i->nr_segs = nr_segs;
254 * Fault in the first iovec of the given iov_iter, to a maximum length
255 * of bytes. Returns 0 on success, or non-zero if the memory could not be
256 * accessed (ie. because it is an invalid address).
258 * writev-intensive code may want this to prefault several iovecs -- that
259 * would be possible (callers must not rely on the fact that _only_ the
260 * first iovec will be faulted with the current implementation).
262 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
264 if (!(i->type & ITER_BVEC)) {
265 char __user *buf = i->iov->iov_base + i->iov_offset;
266 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
267 return fault_in_pages_readable(buf, bytes);
271 EXPORT_SYMBOL(iov_iter_fault_in_readable);
273 static unsigned long alignment_iovec(const struct iov_iter *i)
275 const struct iovec *iov = i->iov;
277 size_t size = i->count;
283 res = (unsigned long)iov->iov_base + i->iov_offset;
284 n = iov->iov_len - i->iov_offset;
289 while (size > (++iov)->iov_len) {
290 res |= (unsigned long)iov->iov_base | iov->iov_len;
291 size -= iov->iov_len;
293 res |= (unsigned long)iov->iov_base | size;
297 void iov_iter_init(struct iov_iter *i, int direction,
298 const struct iovec *iov, unsigned long nr_segs,
301 /* It will get better. Eventually... */
302 if (segment_eq(get_fs(), KERNEL_DS))
303 direction |= ITER_KVEC;
306 i->nr_segs = nr_segs;
310 EXPORT_SYMBOL(iov_iter_init);
312 static ssize_t get_pages_iovec(struct iov_iter *i,
313 struct page **pages, unsigned maxpages,
316 size_t offset = i->iov_offset;
317 const struct iovec *iov = i->iov;
323 len = iov->iov_len - offset;
326 addr = (unsigned long)iov->iov_base + offset;
327 len += *start = addr & (PAGE_SIZE - 1);
328 if (len > maxpages * PAGE_SIZE)
329 len = maxpages * PAGE_SIZE;
330 addr &= ~(PAGE_SIZE - 1);
331 n = (len + PAGE_SIZE - 1) / PAGE_SIZE;
332 res = get_user_pages_fast(addr, n, (i->type & WRITE) != WRITE, pages);
333 if (unlikely(res < 0))
335 return (res == n ? len : res * PAGE_SIZE) - *start;
338 static ssize_t get_pages_alloc_iovec(struct iov_iter *i,
339 struct page ***pages, size_t maxsize,
342 size_t offset = i->iov_offset;
343 const struct iovec *iov = i->iov;
350 len = iov->iov_len - offset;
355 addr = (unsigned long)iov->iov_base + offset;
356 len += *start = addr & (PAGE_SIZE - 1);
357 addr &= ~(PAGE_SIZE - 1);
358 n = (len + PAGE_SIZE - 1) / PAGE_SIZE;
360 p = kmalloc(n * sizeof(struct page *), GFP_KERNEL);
362 p = vmalloc(n * sizeof(struct page *));
366 res = get_user_pages_fast(addr, n, (i->type & WRITE) != WRITE, p);
367 if (unlikely(res < 0)) {
372 return (res == n ? len : res * PAGE_SIZE) - *start;
375 static int iov_iter_npages_iovec(const struct iov_iter *i, int maxpages)
377 size_t offset = i->iov_offset;
378 size_t size = i->count;
379 const struct iovec *iov = i->iov;
383 for (n = 0; size && n < i->nr_segs; n++, iov++) {
384 unsigned long addr = (unsigned long)iov->iov_base + offset;
385 size_t len = iov->iov_len - offset;
387 if (unlikely(!len)) /* empty segment */
391 npages += (addr + len + PAGE_SIZE - 1) / PAGE_SIZE
393 if (npages >= maxpages) /* don't bother going further */
398 return min(npages, maxpages);
401 static void memcpy_from_page(char *to, struct page *page, size_t offset, size_t len)
403 char *from = kmap_atomic(page);
404 memcpy(to, from + offset, len);
408 static void memcpy_to_page(struct page *page, size_t offset, char *from, size_t len)
410 char *to = kmap_atomic(page);
411 memcpy(to + offset, from, len);
415 static size_t copy_page_to_iter_bvec(struct page *page, size_t offset, size_t bytes,
418 size_t skip, copy, wanted;
419 const struct bio_vec *bvec;
422 if (unlikely(bytes > i->count))
425 if (unlikely(!bytes))
430 skip = i->iov_offset;
431 copy = min_t(size_t, bytes, bvec->bv_len - skip);
433 kaddr = kmap_atomic(page);
434 from = kaddr + offset;
435 memcpy_to_page(bvec->bv_page, skip + bvec->bv_offset, from, copy);
441 copy = min(bytes, (size_t)bvec->bv_len);
442 memcpy_to_page(bvec->bv_page, bvec->bv_offset, from, copy);
447 kunmap_atomic(kaddr);
448 if (skip == bvec->bv_len) {
452 i->count -= wanted - bytes;
453 i->nr_segs -= bvec - i->bvec;
455 i->iov_offset = skip;
456 return wanted - bytes;
459 static size_t copy_page_from_iter_bvec(struct page *page, size_t offset, size_t bytes,
462 size_t skip, copy, wanted;
463 const struct bio_vec *bvec;
466 if (unlikely(bytes > i->count))
469 if (unlikely(!bytes))
474 skip = i->iov_offset;
476 kaddr = kmap_atomic(page);
480 copy = min(bytes, bvec->bv_len - skip);
482 memcpy_from_page(to, bvec->bv_page, bvec->bv_offset + skip, copy);
490 copy = min(bytes, (size_t)bvec->bv_len);
491 memcpy_from_page(to, bvec->bv_page, bvec->bv_offset, copy);
496 kunmap_atomic(kaddr);
497 if (skip == bvec->bv_len) {
502 i->nr_segs -= bvec - i->bvec;
504 i->iov_offset = skip;
508 static size_t copy_from_user_bvec(struct page *page,
509 struct iov_iter *i, unsigned long offset, size_t bytes)
513 const struct bio_vec *bvec;
514 size_t base = i->iov_offset;
516 kaddr = kmap_atomic(page);
517 for (left = bytes, bvec = i->bvec; left; bvec++, base = 0) {
518 size_t copy = min(left, bvec->bv_len - base);
521 memcpy_from_page(kaddr + offset, bvec->bv_page,
522 bvec->bv_offset + base, copy);
526 kunmap_atomic(kaddr);
530 static void advance_bvec(struct iov_iter *i, size_t bytes)
532 BUG_ON(i->count < bytes);
534 if (likely(i->nr_segs == 1)) {
535 i->iov_offset += bytes;
538 const struct bio_vec *bvec = i->bvec;
539 size_t base = i->iov_offset;
540 unsigned long nr_segs = i->nr_segs;
543 * The !iov->iov_len check ensures we skip over unlikely
544 * zero-length segments (without overruning the iovec).
546 while (bytes || unlikely(i->count && !bvec->bv_len)) {
549 copy = min(bytes, bvec->bv_len - base);
550 BUG_ON(!i->count || i->count < copy);
554 if (bvec->bv_len == base) {
561 i->iov_offset = base;
562 i->nr_segs = nr_segs;
566 static unsigned long alignment_bvec(const struct iov_iter *i)
568 const struct bio_vec *bvec = i->bvec;
570 size_t size = i->count;
576 res = bvec->bv_offset + i->iov_offset;
577 n = bvec->bv_len - i->iov_offset;
582 while (size > (++bvec)->bv_len) {
583 res |= bvec->bv_offset | bvec->bv_len;
584 size -= bvec->bv_len;
586 res |= bvec->bv_offset | size;
590 static ssize_t get_pages_bvec(struct iov_iter *i,
591 struct page **pages, unsigned maxpages,
594 const struct bio_vec *bvec = i->bvec;
595 size_t len = bvec->bv_len - i->iov_offset;
598 /* can't be more than PAGE_SIZE */
599 *start = bvec->bv_offset + i->iov_offset;
601 get_page(*pages = bvec->bv_page);
606 static ssize_t get_pages_alloc_bvec(struct iov_iter *i,
607 struct page ***pages, size_t maxsize,
610 const struct bio_vec *bvec = i->bvec;
611 size_t len = bvec->bv_len - i->iov_offset;
616 *start = bvec->bv_offset + i->iov_offset;
618 *pages = kmalloc(sizeof(struct page *), GFP_KERNEL);
622 get_page(**pages = bvec->bv_page);
627 static int iov_iter_npages_bvec(const struct iov_iter *i, int maxpages)
629 size_t offset = i->iov_offset;
630 size_t size = i->count;
631 const struct bio_vec *bvec = i->bvec;
635 for (n = 0; size && n < i->nr_segs; n++, bvec++) {
636 size_t len = bvec->bv_len - offset;
638 if (unlikely(!len)) /* empty segment */
643 if (npages >= maxpages) /* don't bother going further */
648 return min(npages, maxpages);
651 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
654 if (i->type & ITER_BVEC)
655 return copy_page_to_iter_bvec(page, offset, bytes, i);
657 return copy_page_to_iter_iovec(page, offset, bytes, i);
659 EXPORT_SYMBOL(copy_page_to_iter);
661 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
664 if (i->type & ITER_BVEC)
665 return copy_page_from_iter_bvec(page, offset, bytes, i);
667 return copy_page_from_iter_iovec(page, offset, bytes, i);
669 EXPORT_SYMBOL(copy_page_from_iter);
671 size_t iov_iter_copy_from_user_atomic(struct page *page,
672 struct iov_iter *i, unsigned long offset, size_t bytes)
674 if (i->type & ITER_BVEC)
675 return copy_from_user_bvec(page, i, offset, bytes);
677 return copy_from_user_atomic_iovec(page, i, offset, bytes);
679 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
681 void iov_iter_advance(struct iov_iter *i, size_t size)
683 if (i->type & ITER_BVEC)
684 advance_bvec(i, size);
686 advance_iovec(i, size);
688 EXPORT_SYMBOL(iov_iter_advance);
691 * Return the count of just the current iov_iter segment.
693 size_t iov_iter_single_seg_count(const struct iov_iter *i)
697 else if (i->type & ITER_BVEC)
698 return min(i->count, i->iov->iov_len - i->iov_offset);
700 return min(i->count, i->bvec->bv_len - i->iov_offset);
702 EXPORT_SYMBOL(iov_iter_single_seg_count);
704 unsigned long iov_iter_alignment(const struct iov_iter *i)
706 if (i->type & ITER_BVEC)
707 return alignment_bvec(i);
709 return alignment_iovec(i);
711 EXPORT_SYMBOL(iov_iter_alignment);
713 ssize_t iov_iter_get_pages(struct iov_iter *i,
714 struct page **pages, unsigned maxpages,
717 if (i->type & ITER_BVEC)
718 return get_pages_bvec(i, pages, maxpages, start);
720 return get_pages_iovec(i, pages, maxpages, start);
722 EXPORT_SYMBOL(iov_iter_get_pages);
724 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
725 struct page ***pages, size_t maxsize,
728 if (i->type & ITER_BVEC)
729 return get_pages_alloc_bvec(i, pages, maxsize, start);
731 return get_pages_alloc_iovec(i, pages, maxsize, start);
733 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
735 int iov_iter_npages(const struct iov_iter *i, int maxpages)
737 if (i->type & ITER_BVEC)
738 return iov_iter_npages_bvec(i, maxpages);
740 return iov_iter_npages_iovec(i, maxpages);
742 EXPORT_SYMBOL(iov_iter_npages);