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, size_t maxsize, unsigned maxpages,
316 size_t offset = i->iov_offset;
317 const struct iovec *iov = i->iov;
323 len = iov->iov_len - offset;
328 addr = (unsigned long)iov->iov_base + offset;
329 len += *start = addr & (PAGE_SIZE - 1);
330 if (len > maxpages * PAGE_SIZE)
331 len = maxpages * PAGE_SIZE;
332 addr &= ~(PAGE_SIZE - 1);
333 n = (len + PAGE_SIZE - 1) / PAGE_SIZE;
334 res = get_user_pages_fast(addr, n, (i->type & WRITE) != WRITE, pages);
335 if (unlikely(res < 0))
337 return (res == n ? len : res * PAGE_SIZE) - *start;
340 static ssize_t get_pages_alloc_iovec(struct iov_iter *i,
341 struct page ***pages, size_t maxsize,
344 size_t offset = i->iov_offset;
345 const struct iovec *iov = i->iov;
352 len = iov->iov_len - offset;
357 addr = (unsigned long)iov->iov_base + offset;
358 len += *start = addr & (PAGE_SIZE - 1);
359 addr &= ~(PAGE_SIZE - 1);
360 n = (len + PAGE_SIZE - 1) / PAGE_SIZE;
362 p = kmalloc(n * sizeof(struct page *), GFP_KERNEL);
364 p = vmalloc(n * sizeof(struct page *));
368 res = get_user_pages_fast(addr, n, (i->type & WRITE) != WRITE, p);
369 if (unlikely(res < 0)) {
374 return (res == n ? len : res * PAGE_SIZE) - *start;
377 static int iov_iter_npages_iovec(const struct iov_iter *i, int maxpages)
379 size_t offset = i->iov_offset;
380 size_t size = i->count;
381 const struct iovec *iov = i->iov;
385 for (n = 0; size && n < i->nr_segs; n++, iov++) {
386 unsigned long addr = (unsigned long)iov->iov_base + offset;
387 size_t len = iov->iov_len - offset;
389 if (unlikely(!len)) /* empty segment */
393 npages += (addr + len + PAGE_SIZE - 1) / PAGE_SIZE
395 if (npages >= maxpages) /* don't bother going further */
400 return min(npages, maxpages);
403 static void memcpy_from_page(char *to, struct page *page, size_t offset, size_t len)
405 char *from = kmap_atomic(page);
406 memcpy(to, from + offset, len);
410 static void memcpy_to_page(struct page *page, size_t offset, char *from, size_t len)
412 char *to = kmap_atomic(page);
413 memcpy(to + offset, from, len);
417 static size_t copy_page_to_iter_bvec(struct page *page, size_t offset, size_t bytes,
420 size_t skip, copy, wanted;
421 const struct bio_vec *bvec;
424 if (unlikely(bytes > i->count))
427 if (unlikely(!bytes))
432 skip = i->iov_offset;
433 copy = min_t(size_t, bytes, bvec->bv_len - skip);
435 kaddr = kmap_atomic(page);
436 from = kaddr + offset;
437 memcpy_to_page(bvec->bv_page, skip + bvec->bv_offset, from, copy);
443 copy = min(bytes, (size_t)bvec->bv_len);
444 memcpy_to_page(bvec->bv_page, bvec->bv_offset, from, copy);
449 kunmap_atomic(kaddr);
450 if (skip == bvec->bv_len) {
454 i->count -= wanted - bytes;
455 i->nr_segs -= bvec - i->bvec;
457 i->iov_offset = skip;
458 return wanted - bytes;
461 static size_t copy_page_from_iter_bvec(struct page *page, size_t offset, size_t bytes,
464 size_t skip, copy, wanted;
465 const struct bio_vec *bvec;
468 if (unlikely(bytes > i->count))
471 if (unlikely(!bytes))
476 skip = i->iov_offset;
478 kaddr = kmap_atomic(page);
482 copy = min(bytes, bvec->bv_len - skip);
484 memcpy_from_page(to, bvec->bv_page, bvec->bv_offset + skip, copy);
492 copy = min(bytes, (size_t)bvec->bv_len);
493 memcpy_from_page(to, bvec->bv_page, bvec->bv_offset, copy);
498 kunmap_atomic(kaddr);
499 if (skip == bvec->bv_len) {
504 i->nr_segs -= bvec - i->bvec;
506 i->iov_offset = skip;
510 static size_t copy_from_user_bvec(struct page *page,
511 struct iov_iter *i, unsigned long offset, size_t bytes)
515 const struct bio_vec *bvec;
516 size_t base = i->iov_offset;
518 kaddr = kmap_atomic(page);
519 for (left = bytes, bvec = i->bvec; left; bvec++, base = 0) {
520 size_t copy = min(left, bvec->bv_len - base);
523 memcpy_from_page(kaddr + offset, bvec->bv_page,
524 bvec->bv_offset + base, copy);
528 kunmap_atomic(kaddr);
532 static void advance_bvec(struct iov_iter *i, size_t bytes)
534 BUG_ON(i->count < bytes);
536 if (likely(i->nr_segs == 1)) {
537 i->iov_offset += bytes;
540 const struct bio_vec *bvec = i->bvec;
541 size_t base = i->iov_offset;
542 unsigned long nr_segs = i->nr_segs;
545 * The !iov->iov_len check ensures we skip over unlikely
546 * zero-length segments (without overruning the iovec).
548 while (bytes || unlikely(i->count && !bvec->bv_len)) {
551 copy = min(bytes, bvec->bv_len - base);
552 BUG_ON(!i->count || i->count < copy);
556 if (bvec->bv_len == base) {
563 i->iov_offset = base;
564 i->nr_segs = nr_segs;
568 static unsigned long alignment_bvec(const struct iov_iter *i)
570 const struct bio_vec *bvec = i->bvec;
572 size_t size = i->count;
578 res = bvec->bv_offset + i->iov_offset;
579 n = bvec->bv_len - i->iov_offset;
584 while (size > (++bvec)->bv_len) {
585 res |= bvec->bv_offset | bvec->bv_len;
586 size -= bvec->bv_len;
588 res |= bvec->bv_offset | size;
592 static ssize_t get_pages_bvec(struct iov_iter *i,
593 struct page **pages, size_t maxsize, unsigned maxpages,
596 const struct bio_vec *bvec = i->bvec;
597 size_t len = bvec->bv_len - i->iov_offset;
602 /* can't be more than PAGE_SIZE */
603 *start = bvec->bv_offset + i->iov_offset;
605 get_page(*pages = bvec->bv_page);
610 static ssize_t get_pages_alloc_bvec(struct iov_iter *i,
611 struct page ***pages, size_t maxsize,
614 const struct bio_vec *bvec = i->bvec;
615 size_t len = bvec->bv_len - i->iov_offset;
620 *start = bvec->bv_offset + i->iov_offset;
622 *pages = kmalloc(sizeof(struct page *), GFP_KERNEL);
626 get_page(**pages = bvec->bv_page);
631 static int iov_iter_npages_bvec(const struct iov_iter *i, int maxpages)
633 size_t offset = i->iov_offset;
634 size_t size = i->count;
635 const struct bio_vec *bvec = i->bvec;
639 for (n = 0; size && n < i->nr_segs; n++, bvec++) {
640 size_t len = bvec->bv_len - offset;
642 if (unlikely(!len)) /* empty segment */
647 if (npages >= maxpages) /* don't bother going further */
652 return min(npages, maxpages);
655 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
658 if (i->type & ITER_BVEC)
659 return copy_page_to_iter_bvec(page, offset, bytes, i);
661 return copy_page_to_iter_iovec(page, offset, bytes, i);
663 EXPORT_SYMBOL(copy_page_to_iter);
665 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
668 if (i->type & ITER_BVEC)
669 return copy_page_from_iter_bvec(page, offset, bytes, i);
671 return copy_page_from_iter_iovec(page, offset, bytes, i);
673 EXPORT_SYMBOL(copy_page_from_iter);
675 size_t iov_iter_copy_from_user_atomic(struct page *page,
676 struct iov_iter *i, unsigned long offset, size_t bytes)
678 if (i->type & ITER_BVEC)
679 return copy_from_user_bvec(page, i, offset, bytes);
681 return copy_from_user_atomic_iovec(page, i, offset, bytes);
683 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
685 void iov_iter_advance(struct iov_iter *i, size_t size)
687 if (i->type & ITER_BVEC)
688 advance_bvec(i, size);
690 advance_iovec(i, size);
692 EXPORT_SYMBOL(iov_iter_advance);
695 * Return the count of just the current iov_iter segment.
697 size_t iov_iter_single_seg_count(const struct iov_iter *i)
701 else if (i->type & ITER_BVEC)
702 return min(i->count, i->iov->iov_len - i->iov_offset);
704 return min(i->count, i->bvec->bv_len - i->iov_offset);
706 EXPORT_SYMBOL(iov_iter_single_seg_count);
708 unsigned long iov_iter_alignment(const struct iov_iter *i)
710 if (i->type & ITER_BVEC)
711 return alignment_bvec(i);
713 return alignment_iovec(i);
715 EXPORT_SYMBOL(iov_iter_alignment);
717 ssize_t iov_iter_get_pages(struct iov_iter *i,
718 struct page **pages, size_t maxsize, unsigned maxpages,
721 if (i->type & ITER_BVEC)
722 return get_pages_bvec(i, pages, maxsize, maxpages, start);
724 return get_pages_iovec(i, pages, maxsize, maxpages, start);
726 EXPORT_SYMBOL(iov_iter_get_pages);
728 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
729 struct page ***pages, size_t maxsize,
732 if (i->type & ITER_BVEC)
733 return get_pages_alloc_bvec(i, pages, maxsize, start);
735 return get_pages_alloc_iovec(i, pages, maxsize, start);
737 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
739 int iov_iter_npages(const struct iov_iter *i, int maxpages)
741 if (i->type & ITER_BVEC)
742 return iov_iter_npages_bvec(i, maxpages);
744 return iov_iter_npages_iovec(i, maxpages);
746 EXPORT_SYMBOL(iov_iter_npages);