1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <linux/export.h>
3 #include <linux/bvec.h>
4 #include <linux/fault-inject-usercopy.h>
5 #include <linux/uio.h>
6 #include <linux/pagemap.h>
7 #include <linux/highmem.h>
8 #include <linux/slab.h>
9 #include <linux/vmalloc.h>
10 #include <linux/splice.h>
11 #include <linux/compat.h>
12 #include <linux/scatterlist.h>
13 #include <linux/instrumented.h>
14 #include <linux/iov_iter.h>
15
16 static __always_inline
17 size_t copy_to_user_iter(void __user *iter_to, size_t progress,
18 size_t len, void *from, void *priv2)
19 {
20 if (should_fail_usercopy())
21 return len;
22 if (access_ok(iter_to, len)) {
23 from += progress;
24 instrument_copy_to_user(iter_to, from, len);
25 len = raw_copy_to_user(iter_to, from, len);
26 }
27 return len;
28 }
29
30 static __always_inline
31 size_t copy_to_user_iter_nofault(void __user *iter_to, size_t progress,
32 size_t len, void *from, void *priv2)
33 {
34 ssize_t res;
35
36 if (should_fail_usercopy())
37 return len;
38
39 from += progress;
40 res = copy_to_user_nofault(iter_to, from, len);
41 return res < 0 ? len : res;
42 }
43
44 static __always_inline
45 size_t copy_from_user_iter(void __user *iter_from, size_t progress,
46 size_t len, void *to, void *priv2)
47 {
48 size_t res = len;
49
50 if (should_fail_usercopy())
51 return len;
52 if (access_ok(iter_from, len)) {
53 to += progress;
54 instrument_copy_from_user_before(to, iter_from, len);
55 res = raw_copy_from_user(to, iter_from, len);
56 instrument_copy_from_user_after(to, iter_from, len, res);
57 }
58 return res;
59 }
60
61 static __always_inline
62 size_t memcpy_to_iter(void *iter_to, size_t progress,
63 size_t len, void *from, void *priv2)
64 {
65 memcpy(iter_to, from + progress, len);
66 return 0;
67 }
68
69 static __always_inline
70 size_t memcpy_from_iter(void *iter_from, size_t progress,
71 size_t len, void *to, void *priv2)
72 {
73 memcpy(to + progress, iter_from, len);
74 return 0;
75 }
76
77 /*
78 * fault_in_iov_iter_readable - fault in iov iterator for reading
79 * @i: iterator
80 * @size: maximum length
81 *
82 * Fault in one or more iovecs of the given iov_iter, to a maximum length of
83 * @size. For each iovec, fault in each page that constitutes the iovec.
84 *
85 * Returns the number of bytes not faulted in (like copy_to_user() and
86 * copy_from_user()).
87 *
88 * Always returns 0 for non-userspace iterators.
89 */
90 size_t fault_in_iov_iter_readable(const struct iov_iter *i, size_t size)
91 {
92 if (iter_is_ubuf(i)) {
93 size_t n = min(size, iov_iter_count(i));
94 n -= fault_in_readable(i->ubuf + i->iov_offset, n);
95 return size - n;
96 } else if (iter_is_iovec(i)) {
97 size_t count = min(size, iov_iter_count(i));
98 const struct iovec *p;
99 size_t skip;
100
101 size -= count;
102 for (p = iter_iov(i), skip = i->iov_offset; count; p++, skip = 0) {
103 size_t len = min(count, p->iov_len - skip);
104 size_t ret;
105
106 if (unlikely(!len))
107 continue;
108 ret = fault_in_readable(p->iov_base + skip, len);
109 count -= len - ret;
110 if (ret)
111 break;
112 }
113 return count + size;
114 }
115 return 0;
116 }
117 EXPORT_SYMBOL(fault_in_iov_iter_readable);
118
119 /*
120 * fault_in_iov_iter_writeable - fault in iov iterator for writing
121 * @i: iterator
122 * @size: maximum length
123 *
124 * Faults in the iterator using get_user_pages(), i.e., without triggering
125 * hardware page faults. This is primarily useful when we already know that
126 * some or all of the pages in @i aren't in memory.
127 *
128 * Returns the number of bytes not faulted in, like copy_to_user() and
129 * copy_from_user().
130 *
131 * Always returns 0 for non-user-space iterators.
132 */
133 size_t fault_in_iov_iter_writeable(const struct iov_iter *i, size_t size)
134 {
135 if (iter_is_ubuf(i)) {
136 size_t n = min(size, iov_iter_count(i));
137 n -= fault_in_safe_writeable(i->ubuf + i->iov_offset, n);
138 return size - n;
139 } else if (iter_is_iovec(i)) {
140 size_t count = min(size, iov_iter_count(i));
141 const struct iovec *p;
142 size_t skip;
143
144 size -= count;
145 for (p = iter_iov(i), skip = i->iov_offset; count; p++, skip = 0) {
146 size_t len = min(count, p->iov_len - skip);
147 size_t ret;
148
149 if (unlikely(!len))
150 continue;
151 ret = fault_in_safe_writeable(p->iov_base + skip, len);
152 count -= len - ret;
153 if (ret)
154 break;
155 }
156 return count + size;
157 }
158 return 0;
159 }
160 EXPORT_SYMBOL(fault_in_iov_iter_writeable);
161
162 void iov_iter_init(struct iov_iter *i, unsigned int direction,
163 const struct iovec *iov, unsigned long nr_segs,
164 size_t count)
165 {
166 WARN_ON(direction & ~(READ | WRITE));
167 *i = (struct iov_iter) {
168 .iter_type = ITER_IOVEC,
169 .nofault = false,
170 .data_source = direction,
171 .__iov = iov,
172 .nr_segs = nr_segs,
173 .iov_offset = 0,
174 .count = count
175 };
176 }
177 EXPORT_SYMBOL(iov_iter_init);
178
179 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
180 {
181 if (WARN_ON_ONCE(i->data_source))
182 return 0;
183 if (user_backed_iter(i))
184 might_fault();
185 return iterate_and_advance(i, bytes, (void *)addr,
186 copy_to_user_iter, memcpy_to_iter);
187 }
188 EXPORT_SYMBOL(_copy_to_iter);
189
190 #ifdef CONFIG_ARCH_HAS_COPY_MC
191 static __always_inline
192 size_t copy_to_user_iter_mc(void __user *iter_to, size_t progress,
193 size_t len, void *from, void *priv2)
194 {
195 if (access_ok(iter_to, len)) {
196 from += progress;
197 instrument_copy_to_user(iter_to, from, len);
198 len = copy_mc_to_user(iter_to, from, len);
199 }
200 return len;
201 }
202
203 static __always_inline
204 size_t memcpy_to_iter_mc(void *iter_to, size_t progress,
205 size_t len, void *from, void *priv2)
206 {
207 return copy_mc_to_kernel(iter_to, from + progress, len);
208 }
209
210 /**
211 * _copy_mc_to_iter - copy to iter with source memory error exception handling
212 * @addr: source kernel address
213 * @bytes: total transfer length
214 * @i: destination iterator
215 *
216 * The pmem driver deploys this for the dax operation
217 * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
218 * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
219 * successfully copied.
220 *
221 * The main differences between this and typical _copy_to_iter().
222 *
223 * * Typical tail/residue handling after a fault retries the copy
224 * byte-by-byte until the fault happens again. Re-triggering machine
225 * checks is potentially fatal so the implementation uses source
226 * alignment and poison alignment assumptions to avoid re-triggering
227 * hardware exceptions.
228 *
229 * * ITER_KVEC and ITER_BVEC can return short copies. Compare to
230 * copy_to_iter() where only ITER_IOVEC attempts might return a short copy.
231 *
232 * Return: number of bytes copied (may be %0)
233 */
234 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
235 {
236 if (WARN_ON_ONCE(i->data_source))
237 return 0;
238 if (user_backed_iter(i))
239 might_fault();
240 return iterate_and_advance(i, bytes, (void *)addr,
241 copy_to_user_iter_mc, memcpy_to_iter_mc);
242 }
243 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
244 #endif /* CONFIG_ARCH_HAS_COPY_MC */
245
246 static __always_inline
247 size_t __copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
248 {
249 return iterate_and_advance(i, bytes, addr,
250 copy_from_user_iter, memcpy_from_iter);
251 }
252
253 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
254 {
255 if (WARN_ON_ONCE(!i->data_source))
256 return 0;
257
258 if (user_backed_iter(i))
259 might_fault();
260 return __copy_from_iter(addr, bytes, i);
261 }
262 EXPORT_SYMBOL(_copy_from_iter);
263
264 static __always_inline
265 size_t copy_from_user_iter_nocache(void __user *iter_from, size_t progress,
266 size_t len, void *to, void *priv2)
267 {
268 return __copy_from_user_inatomic_nocache(to + progress, iter_from, len);
269 }
270
271 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
272 {
273 if (WARN_ON_ONCE(!i->data_source))
274 return 0;
275
276 return iterate_and_advance(i, bytes, addr,
277 copy_from_user_iter_nocache,
278 memcpy_from_iter);
279 }
280 EXPORT_SYMBOL(_copy_from_iter_nocache);
281
282 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
283 static __always_inline
284 size_t copy_from_user_iter_flushcache(void __user *iter_from, size_t progress,
285 size_t len, void *to, void *priv2)
286 {
287 return __copy_from_user_flushcache(to + progress, iter_from, len);
288 }
289
290 static __always_inline
291 size_t memcpy_from_iter_flushcache(void *iter_from, size_t progress,
292 size_t len, void *to, void *priv2)
293 {
294 memcpy_flushcache(to + progress, iter_from, len);
295 return 0;
296 }
297
298 /**
299 * _copy_from_iter_flushcache - write destination through cpu cache
300 * @addr: destination kernel address
301 * @bytes: total transfer length
302 * @i: source iterator
303 *
304 * The pmem driver arranges for filesystem-dax to use this facility via
305 * dax_copy_from_iter() for ensuring that writes to persistent memory
306 * are flushed through the CPU cache. It is differentiated from
307 * _copy_from_iter_nocache() in that guarantees all data is flushed for
308 * all iterator types. The _copy_from_iter_nocache() only attempts to
309 * bypass the cache for the ITER_IOVEC case, and on some archs may use
310 * instructions that strand dirty-data in the cache.
311 *
312 * Return: number of bytes copied (may be %0)
313 */
314 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
315 {
316 if (WARN_ON_ONCE(!i->data_source))
317 return 0;
318
319 return iterate_and_advance(i, bytes, addr,
320 copy_from_user_iter_flushcache,
321 memcpy_from_iter_flushcache);
322 }
323 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
324 #endif
325
326 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
327 {
328 struct page *head;
329 size_t v = n + offset;
330
331 /*
332 * The general case needs to access the page order in order
333 * to compute the page size.
334 * However, we mostly deal with order-0 pages and thus can
335 * avoid a possible cache line miss for requests that fit all
336 * page orders.
337 */
338 if (n <= v && v <= PAGE_SIZE)
339 return true;
340
341 head = compound_head(page);
342 v += (page - head) << PAGE_SHIFT;
343
344 if (WARN_ON(n > v || v > page_size(head)))
345 return false;
346 return true;
347 }
348
349 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
350 struct iov_iter *i)
351 {
352 size_t res = 0;
353 if (!page_copy_sane(page, offset, bytes))
354 return 0;
355 if (WARN_ON_ONCE(i->data_source))
356 return 0;
357 page += offset / PAGE_SIZE; // first subpage
358 offset %= PAGE_SIZE;
359 while (1) {
360 void *kaddr = kmap_local_page(page);
361 size_t n = min(bytes, (size_t)PAGE_SIZE - offset);
362 n = _copy_to_iter(kaddr + offset, n, i);
363 kunmap_local(kaddr);
364 res += n;
365 bytes -= n;
366 if (!bytes || !n)
367 break;
368 offset += n;
369 if (offset == PAGE_SIZE) {
370 page++;
371 offset = 0;
372 }
373 }
374 return res;
375 }
376 EXPORT_SYMBOL(copy_page_to_iter);
377
378 size_t copy_page_to_iter_nofault(struct page *page, unsigned offset, size_t bytes,
379 struct iov_iter *i)
380 {
381 size_t res = 0;
382
383 if (!page_copy_sane(page, offset, bytes))
384 return 0;
385 if (WARN_ON_ONCE(i->data_source))
386 return 0;
387 page += offset / PAGE_SIZE; // first subpage
388 offset %= PAGE_SIZE;
389 while (1) {
390 void *kaddr = kmap_local_page(page);
391 size_t n = min(bytes, (size_t)PAGE_SIZE - offset);
392
393 n = iterate_and_advance(i, n, kaddr + offset,
394 copy_to_user_iter_nofault,
395 memcpy_to_iter);
396 kunmap_local(kaddr);
397 res += n;
398 bytes -= n;
399 if (!bytes || !n)
400 break;
401 offset += n;
402 if (offset == PAGE_SIZE) {
403 page++;
404 offset = 0;
405 }
406 }
407 return res;
408 }
409 EXPORT_SYMBOL(copy_page_to_iter_nofault);
410
411 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
412 struct iov_iter *i)
413 {
414 size_t res = 0;
415 if (!page_copy_sane(page, offset, bytes))
416 return 0;
417 page += offset / PAGE_SIZE; // first subpage
418 offset %= PAGE_SIZE;
419 while (1) {
420 void *kaddr = kmap_local_page(page);
421 size_t n = min(bytes, (size_t)PAGE_SIZE - offset);
422 n = _copy_from_iter(kaddr + offset, n, i);
423 kunmap_local(kaddr);
424 res += n;
425 bytes -= n;
426 if (!bytes || !n)
427 break;
428 offset += n;
429 if (offset == PAGE_SIZE) {
430 page++;
431 offset = 0;
432 }
433 }
434 return res;
435 }
436 EXPORT_SYMBOL(copy_page_from_iter);
437
438 static __always_inline
439 size_t zero_to_user_iter(void __user *iter_to, size_t progress,
440 size_t len, void *priv, void *priv2)
441 {
442 return clear_user(iter_to, len);
443 }
444
445 static __always_inline
446 size_t zero_to_iter(void *iter_to, size_t progress,
447 size_t len, void *priv, void *priv2)
448 {
449 memset(iter_to, 0, len);
450 return 0;
451 }
452
453 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
454 {
455 return iterate_and_advance(i, bytes, NULL,
456 zero_to_user_iter, zero_to_iter);
457 }
458 EXPORT_SYMBOL(iov_iter_zero);
459
460 size_t copy_page_from_iter_atomic(struct page *page, size_t offset,
461 size_t bytes, struct iov_iter *i)
462 {
463 size_t n, copied = 0;
464
465 if (!page_copy_sane(page, offset, bytes))
466 return 0;
467 if (WARN_ON_ONCE(!i->data_source))
468 return 0;
469
470 do {
471 char *p;
472
473 n = bytes - copied;
474 if (PageHighMem(page)) {
475 page += offset / PAGE_SIZE;
476 offset %= PAGE_SIZE;
477 n = min_t(size_t, n, PAGE_SIZE - offset);
478 }
479
480 p = kmap_atomic(page) + offset;
481 n = __copy_from_iter(p, n, i);
482 kunmap_atomic(p);
483 copied += n;
484 offset += n;
485 } while (PageHighMem(page) && copied != bytes && n > 0);
486
487 return copied;
488 }
489 EXPORT_SYMBOL(copy_page_from_iter_atomic);
490
491 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
492 {
493 const struct bio_vec *bvec, *end;
494
495 if (!i->count)
496 return;
497 i->count -= size;
498
499 size += i->iov_offset;
500
501 for (bvec = i->bvec, end = bvec + i->nr_segs; bvec < end; bvec++) {
502 if (likely(size < bvec->bv_len))
503 break;
504 size -= bvec->bv_len;
505 }
506 i->iov_offset = size;
507 i->nr_segs -= bvec - i->bvec;
508 i->bvec = bvec;
509 }
510
511 static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
512 {
513 const struct iovec *iov, *end;
514
515 if (!i->count)
516 return;
517 i->count -= size;
518
519 size += i->iov_offset; // from beginning of current segment
520 for (iov = iter_iov(i), end = iov + i->nr_segs; iov < end; iov++) {
521 if (likely(size < iov->iov_len))
522 break;
523 size -= iov->iov_len;
524 }
525 i->iov_offset = size;
526 i->nr_segs -= iov - iter_iov(i);
527 i->__iov = iov;
528 }
529
530 void iov_iter_advance(struct iov_iter *i, size_t size)
531 {
532 if (unlikely(i->count < size))
533 size = i->count;
534 if (likely(iter_is_ubuf(i)) || unlikely(iov_iter_is_xarray(i))) {
535 i->iov_offset += size;
536 i->count -= size;
537 } else if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
538 /* iovec and kvec have identical layouts */
539 iov_iter_iovec_advance(i, size);
540 } else if (iov_iter_is_bvec(i)) {
541 iov_iter_bvec_advance(i, size);
542 } else if (iov_iter_is_discard(i)) {
543 i->count -= size;
544 }
545 }
546 EXPORT_SYMBOL(iov_iter_advance);
547
548 void iov_iter_revert(struct iov_iter *i, size_t unroll)
549 {
550 if (!unroll)
551 return;
552 if (WARN_ON(unroll > MAX_RW_COUNT))
553 return;
554 i->count += unroll;
555 if (unlikely(iov_iter_is_discard(i)))
556 return;
557 if (unroll <= i->iov_offset) {
558 i->iov_offset -= unroll;
559 return;
560 }
561 unroll -= i->iov_offset;
562 if (iov_iter_is_xarray(i) || iter_is_ubuf(i)) {
563 BUG(); /* We should never go beyond the start of the specified
564 * range since we might then be straying into pages that
565 * aren't pinned.
566 */
567 } else if (iov_iter_is_bvec(i)) {
568 const struct bio_vec *bvec = i->bvec;
569 while (1) {
570 size_t n = (--bvec)->bv_len;
571 i->nr_segs++;
572 if (unroll <= n) {
573 i->bvec = bvec;
574 i->iov_offset = n - unroll;
575 return;
576 }
577 unroll -= n;
578 }
579 } else { /* same logics for iovec and kvec */
580 const struct iovec *iov = iter_iov(i);
581 while (1) {
582 size_t n = (--iov)->iov_len;
583 i->nr_segs++;
584 if (unroll <= n) {
585 i->__iov = iov;
586 i->iov_offset = n - unroll;
587 return;
588 }
589 unroll -= n;
590 }
591 }
592 }
593 EXPORT_SYMBOL(iov_iter_revert);
594
595 /*
596 * Return the count of just the current iov_iter segment.
597 */
598 size_t iov_iter_single_seg_count(const struct iov_iter *i)
599 {
600 if (i->nr_segs > 1) {
601 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
602 return min(i->count, iter_iov(i)->iov_len - i->iov_offset);
603 if (iov_iter_is_bvec(i))
604 return min(i->count, i->bvec->bv_len - i->iov_offset);
605 }
606 return i->count;
607 }
608 EXPORT_SYMBOL(iov_iter_single_seg_count);
609
610 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
611 const struct kvec *kvec, unsigned long nr_segs,
612 size_t count)
613 {
614 WARN_ON(direction & ~(READ | WRITE));
615 *i = (struct iov_iter){
616 .iter_type = ITER_KVEC,
617 .data_source = direction,
618 .kvec = kvec,
619 .nr_segs = nr_segs,
620 .iov_offset = 0,
621 .count = count
622 };
623 }
624 EXPORT_SYMBOL(iov_iter_kvec);
625
626 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
627 const struct bio_vec *bvec, unsigned long nr_segs,
628 size_t count)
629 {
630 WARN_ON(direction & ~(READ | WRITE));
631 *i = (struct iov_iter){
632 .iter_type = ITER_BVEC,
633 .data_source = direction,
634 .bvec = bvec,
635 .nr_segs = nr_segs,
636 .iov_offset = 0,
637 .count = count
638 };
639 }
640 EXPORT_SYMBOL(iov_iter_bvec);
641
642 /**
643 * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
644 * @i: The iterator to initialise.
645 * @direction: The direction of the transfer.
646 * @xarray: The xarray to access.
647 * @start: The start file position.
648 * @count: The size of the I/O buffer in bytes.
649 *
650 * Set up an I/O iterator to either draw data out of the pages attached to an
651 * inode or to inject data into those pages. The pages *must* be prevented
652 * from evaporation, either by taking a ref on them or locking them by the
653 * caller.
654 */
655 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
656 struct xarray *xarray, loff_t start, size_t count)
657 {
658 BUG_ON(direction & ~1);
659 *i = (struct iov_iter) {
660 .iter_type = ITER_XARRAY,
661 .data_source = direction,
662 .xarray = xarray,
663 .xarray_start = start,
664 .count = count,
665 .iov_offset = 0
666 };
667 }
668 EXPORT_SYMBOL(iov_iter_xarray);
669
670 /**
671 * iov_iter_discard - Initialise an I/O iterator that discards data
672 * @i: The iterator to initialise.
673 * @direction: The direction of the transfer.
674 * @count: The size of the I/O buffer in bytes.
675 *
676 * Set up an I/O iterator that just discards everything that's written to it.
677 * It's only available as a READ iterator.
678 */
679 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
680 {
681 BUG_ON(direction != READ);
682 *i = (struct iov_iter){
683 .iter_type = ITER_DISCARD,
684 .data_source = false,
685 .count = count,
686 .iov_offset = 0
687 };
688 }
689 EXPORT_SYMBOL(iov_iter_discard);
690
691 static bool iov_iter_aligned_iovec(const struct iov_iter *i, unsigned addr_mask,
692 unsigned len_mask)
693 {
694 const struct iovec *iov = iter_iov(i);
695 size_t size = i->count;
696 size_t skip = i->iov_offset;
697
698 do {
699 size_t len = iov->iov_len - skip;
700
701 if (len > size)
702 len = size;
703 if (len & len_mask)
704 return false;
705 if ((unsigned long)(iov->iov_base + skip) & addr_mask)
706 return false;
707
708 iov++;
709 size -= len;
710 skip = 0;
711 } while (size);
712
713 return true;
714 }
715
716 static bool iov_iter_aligned_bvec(const struct iov_iter *i, unsigned addr_mask,
717 unsigned len_mask)
718 {
719 const struct bio_vec *bvec = i->bvec;
720 unsigned skip = i->iov_offset;
721 size_t size = i->count;
722
723 do {
724 size_t len = bvec->bv_len;
725
726 if (len > size)
727 len = size;
728 if (len & len_mask)
729 return false;
730 if ((unsigned long)(bvec->bv_offset + skip) & addr_mask)
731 return false;
732
733 bvec++;
734 size -= len;
735 skip = 0;
736 } while (size);
737
738 return true;
739 }
740
741 /**
742 * iov_iter_is_aligned() - Check if the addresses and lengths of each segments
743 * are aligned to the parameters.
744 *
745 * @i: &struct iov_iter to restore
746 * @addr_mask: bit mask to check against the iov element's addresses
747 * @len_mask: bit mask to check against the iov element's lengths
748 *
749 * Return: false if any addresses or lengths intersect with the provided masks
750 */
751 bool iov_iter_is_aligned(const struct iov_iter *i, unsigned addr_mask,
752 unsigned len_mask)
753 {
754 if (likely(iter_is_ubuf(i))) {
755 if (i->count & len_mask)
756 return false;
757 if ((unsigned long)(i->ubuf + i->iov_offset) & addr_mask)
758 return false;
759 return true;
760 }
761
762 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
763 return iov_iter_aligned_iovec(i, addr_mask, len_mask);
764
765 if (iov_iter_is_bvec(i))
766 return iov_iter_aligned_bvec(i, addr_mask, len_mask);
767
768 if (iov_iter_is_xarray(i)) {
769 if (i->count & len_mask)
770 return false;
771 if ((i->xarray_start + i->iov_offset) & addr_mask)
772 return false;
773 }
774
775 return true;
776 }
777 EXPORT_SYMBOL_GPL(iov_iter_is_aligned);
778
779 static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
780 {
781 const struct iovec *iov = iter_iov(i);
782 unsigned long res = 0;
783 size_t size = i->count;
784 size_t skip = i->iov_offset;
785
786 do {
787 size_t len = iov->iov_len - skip;
788 if (len) {
789 res |= (unsigned long)iov->iov_base + skip;
790 if (len > size)
791 len = size;
792 res |= len;
793 size -= len;
794 }
795 iov++;
796 skip = 0;
797 } while (size);
798 return res;
799 }
800
801 static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
802 {
803 const struct bio_vec *bvec = i->bvec;
804 unsigned res = 0;
805 size_t size = i->count;
806 unsigned skip = i->iov_offset;
807
808 do {
809 size_t len = bvec->bv_len - skip;
810 res |= (unsigned long)bvec->bv_offset + skip;
811 if (len > size)
812 len = size;
813 res |= len;
814 bvec++;
815 size -= len;
816 skip = 0;
817 } while (size);
818
819 return res;
820 }
821
822 unsigned long iov_iter_alignment(const struct iov_iter *i)
823 {
824 if (likely(iter_is_ubuf(i))) {
825 size_t size = i->count;
826 if (size)
827 return ((unsigned long)i->ubuf + i->iov_offset) | size;
828 return 0;
829 }
830
831 /* iovec and kvec have identical layouts */
832 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
833 return iov_iter_alignment_iovec(i);
834
835 if (iov_iter_is_bvec(i))
836 return iov_iter_alignment_bvec(i);
837
838 if (iov_iter_is_xarray(i))
839 return (i->xarray_start + i->iov_offset) | i->count;
840
841 return 0;
842 }
843 EXPORT_SYMBOL(iov_iter_alignment);
844
845 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
846 {
847 unsigned long res = 0;
848 unsigned long v = 0;
849 size_t size = i->count;
850 unsigned k;
851
852 if (iter_is_ubuf(i))
853 return 0;
854
855 if (WARN_ON(!iter_is_iovec(i)))
856 return ~0U;
857
858 for (k = 0; k < i->nr_segs; k++) {
859 const struct iovec *iov = iter_iov(i) + k;
860 if (iov->iov_len) {
861 unsigned long base = (unsigned long)iov->iov_base;
862 if (v) // if not the first one
863 res |= base | v; // this start | previous end
864 v = base + iov->iov_len;
865 if (size <= iov->iov_len)
866 break;
867 size -= iov->iov_len;
868 }
869 }
870 return res;
871 }
872 EXPORT_SYMBOL(iov_iter_gap_alignment);
873
874 static int want_pages_array(struct page ***res, size_t size,
875 size_t start, unsigned int maxpages)
876 {
877 unsigned int count = DIV_ROUND_UP(size + start, PAGE_SIZE);
878
879 if (count > maxpages)
880 count = maxpages;
881 WARN_ON(!count); // caller should've prevented that
882 if (!*res) {
883 *res = kvmalloc_array(count, sizeof(struct page *), GFP_KERNEL);
884 if (!*res)
885 return 0;
886 }
887 return count;
888 }
889
890 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
891 pgoff_t index, unsigned int nr_pages)
892 {
893 XA_STATE(xas, xa, index);
894 struct page *page;
895 unsigned int ret = 0;
896
897 rcu_read_lock();
898 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
899 if (xas_retry(&xas, page))
900 continue;
901
902 /* Has the page moved or been split? */
903 if (unlikely(page != xas_reload(&xas))) {
904 xas_reset(&xas);
905 continue;
906 }
907
908 pages[ret] = find_subpage(page, xas.xa_index);
909 get_page(pages[ret]);
910 if (++ret == nr_pages)
911 break;
912 }
913 rcu_read_unlock();
914 return ret;
915 }
916
917 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
918 struct page ***pages, size_t maxsize,
919 unsigned maxpages, size_t *_start_offset)
920 {
921 unsigned nr, offset, count;
922 pgoff_t index;
923 loff_t pos;
924
925 pos = i->xarray_start + i->iov_offset;
926 index = pos >> PAGE_SHIFT;
927 offset = pos & ~PAGE_MASK;
928 *_start_offset = offset;
929
930 count = want_pages_array(pages, maxsize, offset, maxpages);
931 if (!count)
932 return -ENOMEM;
933 nr = iter_xarray_populate_pages(*pages, i->xarray, index, count);
934 if (nr == 0)
935 return 0;
936
937 maxsize = min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
938 i->iov_offset += maxsize;
939 i->count -= maxsize;
940 return maxsize;
941 }
942
943 /* must be done on non-empty ITER_UBUF or ITER_IOVEC one */
944 static unsigned long first_iovec_segment(const struct iov_iter *i, size_t *size)
945 {
946 size_t skip;
947 long k;
948
949 if (iter_is_ubuf(i))
950 return (unsigned long)i->ubuf + i->iov_offset;
951
952 for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
953 const struct iovec *iov = iter_iov(i) + k;
954 size_t len = iov->iov_len - skip;
955
956 if (unlikely(!len))
957 continue;
958 if (*size > len)
959 *size = len;
960 return (unsigned long)iov->iov_base + skip;
961 }
962 BUG(); // if it had been empty, we wouldn't get called
963 }
964
965 /* must be done on non-empty ITER_BVEC one */
966 static struct page *first_bvec_segment(const struct iov_iter *i,
967 size_t *size, size_t *start)
968 {
969 struct page *page;
970 size_t skip = i->iov_offset, len;
971
972 len = i->bvec->bv_len - skip;
973 if (*size > len)
974 *size = len;
975 skip += i->bvec->bv_offset;
976 page = i->bvec->bv_page + skip / PAGE_SIZE;
977 *start = skip % PAGE_SIZE;
978 return page;
979 }
980
981 static ssize_t __iov_iter_get_pages_alloc(struct iov_iter *i,
982 struct page ***pages, size_t maxsize,
983 unsigned int maxpages, size_t *start)
984 {
985 unsigned int n, gup_flags = 0;
986
987 if (maxsize > i->count)
988 maxsize = i->count;
989 if (!maxsize)
990 return 0;
991 if (maxsize > MAX_RW_COUNT)
992 maxsize = MAX_RW_COUNT;
993
994 if (likely(user_backed_iter(i))) {
995 unsigned long addr;
996 int res;
997
998 if (iov_iter_rw(i) != WRITE)
999 gup_flags |= FOLL_WRITE;
1000 if (i->nofault)
1001 gup_flags |= FOLL_NOFAULT;
1002
1003 addr = first_iovec_segment(i, &maxsize);
1004 *start = addr % PAGE_SIZE;
1005 addr &= PAGE_MASK;
1006 n = want_pages_array(pages, maxsize, *start, maxpages);
1007 if (!n)
1008 return -ENOMEM;
1009 res = get_user_pages_fast(addr, n, gup_flags, *pages);
1010 if (unlikely(res <= 0))
1011 return res;
1012 maxsize = min_t(size_t, maxsize, res * PAGE_SIZE - *start);
1013 iov_iter_advance(i, maxsize);
1014 return maxsize;
1015 }
1016 if (iov_iter_is_bvec(i)) {
1017 struct page **p;
1018 struct page *page;
1019
1020 page = first_bvec_segment(i, &maxsize, start);
1021 n = want_pages_array(pages, maxsize, *start, maxpages);
1022 if (!n)
1023 return -ENOMEM;
1024 p = *pages;
1025 for (int k = 0; k < n; k++)
1026 get_page(p[k] = page + k);
1027 maxsize = min_t(size_t, maxsize, n * PAGE_SIZE - *start);
1028 i->count -= maxsize;
1029 i->iov_offset += maxsize;
1030 if (i->iov_offset == i->bvec->bv_len) {
1031 i->iov_offset = 0;
1032 i->bvec++;
1033 i->nr_segs--;
1034 }
1035 return maxsize;
1036 }
1037 if (iov_iter_is_xarray(i))
1038 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1039 return -EFAULT;
1040 }
1041
1042 ssize_t iov_iter_get_pages2(struct iov_iter *i, struct page **pages,
1043 size_t maxsize, unsigned maxpages, size_t *start)
1044 {
1045 if (!maxpages)
1046 return 0;
1047 BUG_ON(!pages);
1048
1049 return __iov_iter_get_pages_alloc(i, &pages, maxsize, maxpages, start);
1050 }
1051 EXPORT_SYMBOL(iov_iter_get_pages2);
1052
1053 ssize_t iov_iter_get_pages_alloc2(struct iov_iter *i,
1054 struct page ***pages, size_t maxsize, size_t *start)
1055 {
1056 ssize_t len;
1057
1058 *pages = NULL;
1059
1060 len = __iov_iter_get_pages_alloc(i, pages, maxsize, ~0U, start);
1061 if (len <= 0) {
1062 kvfree(*pages);
1063 *pages = NULL;
1064 }
1065 return len;
1066 }
1067 EXPORT_SYMBOL(iov_iter_get_pages_alloc2);
1068
1069 static int iov_npages(const struct iov_iter *i, int maxpages)
1070 {
1071 size_t skip = i->iov_offset, size = i->count;
1072 const struct iovec *p;
1073 int npages = 0;
1074
1075 for (p = iter_iov(i); size; skip = 0, p++) {
1076 unsigned offs = offset_in_page(p->iov_base + skip);
1077 size_t len = min(p->iov_len - skip, size);
1078
1079 if (len) {
1080 size -= len;
1081 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1082 if (unlikely(npages > maxpages))
1083 return maxpages;
1084 }
1085 }
1086 return npages;
1087 }
1088
1089 static int bvec_npages(const struct iov_iter *i, int maxpages)
1090 {
1091 size_t skip = i->iov_offset, size = i->count;
1092 const struct bio_vec *p;
1093 int npages = 0;
1094
1095 for (p = i->bvec; size; skip = 0, p++) {
1096 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1097 size_t len = min(p->bv_len - skip, size);
1098
1099 size -= len;
1100 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1101 if (unlikely(npages > maxpages))
1102 return maxpages;
1103 }
1104 return npages;
1105 }
1106
1107 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1108 {
1109 if (unlikely(!i->count))
1110 return 0;
1111 if (likely(iter_is_ubuf(i))) {
1112 unsigned offs = offset_in_page(i->ubuf + i->iov_offset);
1113 int npages = DIV_ROUND_UP(offs + i->count, PAGE_SIZE);
1114 return min(npages, maxpages);
1115 }
1116 /* iovec and kvec have identical layouts */
1117 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1118 return iov_npages(i, maxpages);
1119 if (iov_iter_is_bvec(i))
1120 return bvec_npages(i, maxpages);
1121 if (iov_iter_is_xarray(i)) {
1122 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1123 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1124 return min(npages, maxpages);
1125 }
1126 return 0;
1127 }
1128 EXPORT_SYMBOL(iov_iter_npages);
1129
1130 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1131 {
1132 *new = *old;
1133 if (iov_iter_is_bvec(new))
1134 return new->bvec = kmemdup(new->bvec,
1135 new->nr_segs * sizeof(struct bio_vec),
1136 flags);
1137 else if (iov_iter_is_kvec(new) || iter_is_iovec(new))
1138 /* iovec and kvec have identical layout */
1139 return new->__iov = kmemdup(new->__iov,
1140 new->nr_segs * sizeof(struct iovec),
1141 flags);
1142 return NULL;
1143 }
1144 EXPORT_SYMBOL(dup_iter);
1145
1146 static __noclone int copy_compat_iovec_from_user(struct iovec *iov,
1147 const struct iovec __user *uvec, u32 nr_segs)
1148 {
1149 const struct compat_iovec __user *uiov =
1150 (const struct compat_iovec __user *)uvec;
1151 int ret = -EFAULT;
1152 u32 i;
1153
1154 if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1155 return -EFAULT;
1156
1157 for (i = 0; i < nr_segs; i++) {
1158 compat_uptr_t buf;
1159 compat_ssize_t len;
1160
1161 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1162 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1163
1164 /* check for compat_size_t not fitting in compat_ssize_t .. */
1165 if (len < 0) {
1166 ret = -EINVAL;
1167 goto uaccess_end;
1168 }
1169 iov[i].iov_base = compat_ptr(buf);
1170 iov[i].iov_len = len;
1171 }
1172
1173 ret = 0;
1174 uaccess_end:
1175 user_access_end();
1176 return ret;
1177 }
1178
1179 static __noclone int copy_iovec_from_user(struct iovec *iov,
1180 const struct iovec __user *uiov, unsigned long nr_segs)
1181 {
1182 int ret = -EFAULT;
1183
1184 if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1185 return -EFAULT;
1186
1187 do {
1188 void __user *buf;
1189 ssize_t len;
1190
1191 unsafe_get_user(len, &uiov->iov_len, uaccess_end);
1192 unsafe_get_user(buf, &uiov->iov_base, uaccess_end);
1193
1194 /* check for size_t not fitting in ssize_t .. */
1195 if (unlikely(len < 0)) {
1196 ret = -EINVAL;
1197 goto uaccess_end;
1198 }
1199 iov->iov_base = buf;
1200 iov->iov_len = len;
1201
1202 uiov++; iov++;
1203 } while (--nr_segs);
1204
1205 ret = 0;
1206 uaccess_end:
1207 user_access_end();
1208 return ret;
1209 }
1210
1211 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1212 unsigned long nr_segs, unsigned long fast_segs,
1213 struct iovec *fast_iov, bool compat)
1214 {
1215 struct iovec *iov = fast_iov;
1216 int ret;
1217
1218 /*
1219 * SuS says "The readv() function *may* fail if the iovcnt argument was
1220 * less than or equal to 0, or greater than {IOV_MAX}. Linux has
1221 * traditionally returned zero for zero segments, so...
1222 */
1223 if (nr_segs == 0)
1224 return iov;
1225 if (nr_segs > UIO_MAXIOV)
1226 return ERR_PTR(-EINVAL);
1227 if (nr_segs > fast_segs) {
1228 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1229 if (!iov)
1230 return ERR_PTR(-ENOMEM);
1231 }
1232
1233 if (unlikely(compat))
1234 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1235 else
1236 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1237 if (ret) {
1238 if (iov != fast_iov)
1239 kfree(iov);
1240 return ERR_PTR(ret);
1241 }
1242
1243 return iov;
1244 }
1245
1246 /*
1247 * Single segment iovec supplied by the user, import it as ITER_UBUF.
1248 */
1249 static ssize_t __import_iovec_ubuf(int type, const struct iovec __user *uvec,
1250 struct iovec **iovp, struct iov_iter *i,
1251 bool compat)
1252 {
1253 struct iovec *iov = *iovp;
1254 ssize_t ret;
1255
1256 if (compat)
1257 ret = copy_compat_iovec_from_user(iov, uvec, 1);
1258 else
1259 ret = copy_iovec_from_user(iov, uvec, 1);
1260 if (unlikely(ret))
1261 return ret;
1262
1263 ret = import_ubuf(type, iov->iov_base, iov->iov_len, i);
1264 if (unlikely(ret))
1265 return ret;
1266 *iovp = NULL;
1267 return i->count;
1268 }
1269
1270 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1271 unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1272 struct iov_iter *i, bool compat)
1273 {
1274 ssize_t total_len = 0;
1275 unsigned long seg;
1276 struct iovec *iov;
1277
1278 if (nr_segs == 1)
1279 return __import_iovec_ubuf(type, uvec, iovp, i, compat);
1280
1281 iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1282 if (IS_ERR(iov)) {
1283 *iovp = NULL;
1284 return PTR_ERR(iov);
1285 }
1286
1287 /*
1288 * According to the Single Unix Specification we should return EINVAL if
1289 * an element length is < 0 when cast to ssize_t or if the total length
1290 * would overflow the ssize_t return value of the system call.
1291 *
1292 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1293 * overflow case.
1294 */
1295 for (seg = 0; seg < nr_segs; seg++) {
1296 ssize_t len = (ssize_t)iov[seg].iov_len;
1297
1298 if (!access_ok(iov[seg].iov_base, len)) {
1299 if (iov != *iovp)
1300 kfree(iov);
1301 *iovp = NULL;
1302 return -EFAULT;
1303 }
1304
1305 if (len > MAX_RW_COUNT - total_len) {
1306 len = MAX_RW_COUNT - total_len;
1307 iov[seg].iov_len = len;
1308 }
1309 total_len += len;
1310 }
1311
1312 iov_iter_init(i, type, iov, nr_segs, total_len);
1313 if (iov == *iovp)
1314 *iovp = NULL;
1315 else
1316 *iovp = iov;
1317 return total_len;
1318 }
1319
1320 /**
1321 * import_iovec() - Copy an array of &struct iovec from userspace
1322 * into the kernel, check that it is valid, and initialize a new
1323 * &struct iov_iter iterator to access it.
1324 *
1325 * @type: One of %READ or %WRITE.
1326 * @uvec: Pointer to the userspace array.
1327 * @nr_segs: Number of elements in userspace array.
1328 * @fast_segs: Number of elements in @iov.
1329 * @iovp: (input and output parameter) Pointer to pointer to (usually small
1330 * on-stack) kernel array.
1331 * @i: Pointer to iterator that will be initialized on success.
1332 *
1333 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1334 * then this function places %NULL in *@iov on return. Otherwise, a new
1335 * array will be allocated and the result placed in *@iov. This means that
1336 * the caller may call kfree() on *@iov regardless of whether the small
1337 * on-stack array was used or not (and regardless of whether this function
1338 * returns an error or not).
1339 *
1340 * Return: Negative error code on error, bytes imported on success
1341 */
1342 ssize_t import_iovec(int type, const struct iovec __user *uvec,
1343 unsigned nr_segs, unsigned fast_segs,
1344 struct iovec **iovp, struct iov_iter *i)
1345 {
1346 return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
1347 in_compat_syscall());
1348 }
1349 EXPORT_SYMBOL(import_iovec);
1350
1351 int import_ubuf(int rw, void __user *buf, size_t len, struct iov_iter *i)
1352 {
1353 if (len > MAX_RW_COUNT)
1354 len = MAX_RW_COUNT;
1355 if (unlikely(!access_ok(buf, len)))
1356 return -EFAULT;
1357
1358 iov_iter_ubuf(i, rw, buf, len);
1359 return 0;
1360 }
1361 EXPORT_SYMBOL_GPL(import_ubuf);
1362
1363 /**
1364 * iov_iter_restore() - Restore a &struct iov_iter to the same state as when
1365 * iov_iter_save_state() was called.
1366 *
1367 * @i: &struct iov_iter to restore
1368 * @state: state to restore from
1369 *
1370 * Used after iov_iter_save_state() to bring restore @i, if operations may
1371 * have advanced it.
1372 *
1373 * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC
1374 */
1375 void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state)
1376 {
1377 if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i) &&
1378 !iter_is_ubuf(i)) && !iov_iter_is_kvec(i))
1379 return;
1380 i->iov_offset = state->iov_offset;
1381 i->count = state->count;
1382 if (iter_is_ubuf(i))
1383 return;
1384 /*
1385 * For the *vec iters, nr_segs + iov is constant - if we increment
1386 * the vec, then we also decrement the nr_segs count. Hence we don't
1387 * need to track both of these, just one is enough and we can deduct
1388 * the other from that. ITER_KVEC and ITER_IOVEC are the same struct
1389 * size, so we can just increment the iov pointer as they are unionzed.
1390 * ITER_BVEC _may_ be the same size on some archs, but on others it is
1391 * not. Be safe and handle it separately.
1392 */
1393 BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
1394 if (iov_iter_is_bvec(i))
1395 i->bvec -= state->nr_segs - i->nr_segs;
1396 else
1397 i->__iov -= state->nr_segs - i->nr_segs;
1398 i->nr_segs = state->nr_segs;
1399 }
1400
1401 /*
1402 * Extract a list of contiguous pages from an ITER_XARRAY iterator. This does not
1403 * get references on the pages, nor does it get a pin on them.
1404 */
1405 static ssize_t iov_iter_extract_xarray_pages(struct iov_iter *i,
1406 struct page ***pages, size_t maxsize,
1407 unsigned int maxpages,
1408 iov_iter_extraction_t extraction_flags,
1409 size_t *offset0)
1410 {
1411 struct page *page, **p;
1412 unsigned int nr = 0, offset;
1413 loff_t pos = i->xarray_start + i->iov_offset;
1414 pgoff_t index = pos >> PAGE_SHIFT;
1415 XA_STATE(xas, i->xarray, index);
1416
1417 offset = pos & ~PAGE_MASK;
1418 *offset0 = offset;
1419
1420 maxpages = want_pages_array(pages, maxsize, offset, maxpages);
1421 if (!maxpages)
1422 return -ENOMEM;
1423 p = *pages;
1424
1425 rcu_read_lock();
1426 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1427 if (xas_retry(&xas, page))
1428 continue;
1429
1430 /* Has the page moved or been split? */
1431 if (unlikely(page != xas_reload(&xas))) {
1432 xas_reset(&xas);
1433 continue;
1434 }
1435
1436 p[nr++] = find_subpage(page, xas.xa_index);
1437 if (nr == maxpages)
1438 break;
1439 }
1440 rcu_read_unlock();
1441
1442 maxsize = min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1443 iov_iter_advance(i, maxsize);
1444 return maxsize;
1445 }
1446
1447 /*
1448 * Extract a list of contiguous pages from an ITER_BVEC iterator. This does
1449 * not get references on the pages, nor does it get a pin on them.
1450 */
1451 static ssize_t iov_iter_extract_bvec_pages(struct iov_iter *i,
1452 struct page ***pages, size_t maxsize,
1453 unsigned int maxpages,
1454 iov_iter_extraction_t extraction_flags,
1455 size_t *offset0)
1456 {
1457 struct page **p, *page;
1458 size_t skip = i->iov_offset, offset, size;
1459 int k;
1460
1461 for (;;) {
1462 if (i->nr_segs == 0)
1463 return 0;
1464 size = min(maxsize, i->bvec->bv_len - skip);
1465 if (size)
1466 break;
1467 i->iov_offset = 0;
1468 i->nr_segs--;
1469 i->bvec++;
1470 skip = 0;
1471 }
1472
1473 skip += i->bvec->bv_offset;
1474 page = i->bvec->bv_page + skip / PAGE_SIZE;
1475 offset = skip % PAGE_SIZE;
1476 *offset0 = offset;
1477
1478 maxpages = want_pages_array(pages, size, offset, maxpages);
1479 if (!maxpages)
1480 return -ENOMEM;
1481 p = *pages;
1482 for (k = 0; k < maxpages; k++)
1483 p[k] = page + k;
1484
1485 size = min_t(size_t, size, maxpages * PAGE_SIZE - offset);
1486 iov_iter_advance(i, size);
1487 return size;
1488 }
1489
1490 /*
1491 * Extract a list of virtually contiguous pages from an ITER_KVEC iterator.
1492 * This does not get references on the pages, nor does it get a pin on them.
1493 */
1494 static ssize_t iov_iter_extract_kvec_pages(struct iov_iter *i,
1495 struct page ***pages, size_t maxsize,
1496 unsigned int maxpages,
1497 iov_iter_extraction_t extraction_flags,
1498 size_t *offset0)
1499 {
1500 struct page **p, *page;
1501 const void *kaddr;
1502 size_t skip = i->iov_offset, offset, len, size;
1503 int k;
1504
1505 for (;;) {
1506 if (i->nr_segs == 0)
1507 return 0;
1508 size = min(maxsize, i->kvec->iov_len - skip);
1509 if (size)
1510 break;
1511 i->iov_offset = 0;
1512 i->nr_segs--;
1513 i->kvec++;
1514 skip = 0;
1515 }
1516
1517 kaddr = i->kvec->iov_base + skip;
1518 offset = (unsigned long)kaddr & ~PAGE_MASK;
1519 *offset0 = offset;
1520
1521 maxpages = want_pages_array(pages, size, offset, maxpages);
1522 if (!maxpages)
1523 return -ENOMEM;
1524 p = *pages;
1525
1526 kaddr -= offset;
1527 len = offset + size;
1528 for (k = 0; k < maxpages; k++) {
1529 size_t seg = min_t(size_t, len, PAGE_SIZE);
1530
1531 if (is_vmalloc_or_module_addr(kaddr))
1532 page = vmalloc_to_page(kaddr);
1533 else
1534 page = virt_to_page(kaddr);
1535
1536 p[k] = page;
1537 len -= seg;
1538 kaddr += PAGE_SIZE;
1539 }
1540
1541 size = min_t(size_t, size, maxpages * PAGE_SIZE - offset);
1542 iov_iter_advance(i, size);
1543 return size;
1544 }
1545
1546 /*
1547 * Extract a list of contiguous pages from a user iterator and get a pin on
1548 * each of them. This should only be used if the iterator is user-backed
1549 * (IOBUF/UBUF).
1550 *
1551 * It does not get refs on the pages, but the pages must be unpinned by the
1552 * caller once the transfer is complete.
1553 *
1554 * This is safe to be used where background IO/DMA *is* going to be modifying
1555 * the buffer; using a pin rather than a ref makes forces fork() to give the
1556 * child a copy of the page.
1557 */
1558 static ssize_t iov_iter_extract_user_pages(struct iov_iter *i,
1559 struct page ***pages,
1560 size_t maxsize,
1561 unsigned int maxpages,
1562 iov_iter_extraction_t extraction_flags,
1563 size_t *offset0)
1564 {
1565 unsigned long addr;
1566 unsigned int gup_flags = 0;
1567 size_t offset;
1568 int res;
1569
1570 if (i->data_source == ITER_DEST)
1571 gup_flags |= FOLL_WRITE;
1572 if (extraction_flags & ITER_ALLOW_P2PDMA)
1573 gup_flags |= FOLL_PCI_P2PDMA;
1574 if (i->nofault)
1575 gup_flags |= FOLL_NOFAULT;
1576
1577 addr = first_iovec_segment(i, &maxsize);
1578 *offset0 = offset = addr % PAGE_SIZE;
1579 addr &= PAGE_MASK;
1580 maxpages = want_pages_array(pages, maxsize, offset, maxpages);
1581 if (!maxpages)
1582 return -ENOMEM;
1583 res = pin_user_pages_fast(addr, maxpages, gup_flags, *pages);
1584 if (unlikely(res <= 0))
1585 return res;
1586 maxsize = min_t(size_t, maxsize, res * PAGE_SIZE - offset);
1587 iov_iter_advance(i, maxsize);
1588 return maxsize;
1589 }
1590
1591 /**
1592 * iov_iter_extract_pages - Extract a list of contiguous pages from an iterator
1593 * @i: The iterator to extract from
1594 * @pages: Where to return the list of pages
1595 * @maxsize: The maximum amount of iterator to extract
1596 * @maxpages: The maximum size of the list of pages
1597 * @extraction_flags: Flags to qualify request
1598 * @offset0: Where to return the starting offset into (*@pages)[0]
1599 *
1600 * Extract a list of contiguous pages from the current point of the iterator,
1601 * advancing the iterator. The maximum number of pages and the maximum amount
1602 * of page contents can be set.
1603 *
1604 * If *@pages is NULL, a page list will be allocated to the required size and
1605 * *@pages will be set to its base. If *@pages is not NULL, it will be assumed
1606 * that the caller allocated a page list at least @maxpages in size and this
1607 * will be filled in.
1608 *
1609 * @extraction_flags can have ITER_ALLOW_P2PDMA set to request peer-to-peer DMA
1610 * be allowed on the pages extracted.
1611 *
1612 * The iov_iter_extract_will_pin() function can be used to query how cleanup
1613 * should be performed.
1614 *
1615 * Extra refs or pins on the pages may be obtained as follows:
1616 *
1617 * (*) If the iterator is user-backed (ITER_IOVEC/ITER_UBUF), pins will be
1618 * added to the pages, but refs will not be taken.
1619 * iov_iter_extract_will_pin() will return true.
1620 *
1621 * (*) If the iterator is ITER_KVEC, ITER_BVEC or ITER_XARRAY, the pages are
1622 * merely listed; no extra refs or pins are obtained.
1623 * iov_iter_extract_will_pin() will return 0.
1624 *
1625 * Note also:
1626 *
1627 * (*) Use with ITER_DISCARD is not supported as that has no content.
1628 *
1629 * On success, the function sets *@pages to the new pagelist, if allocated, and
1630 * sets *offset0 to the offset into the first page.
1631 *
1632 * It may also return -ENOMEM and -EFAULT.
1633 */
1634 ssize_t iov_iter_extract_pages(struct iov_iter *i,
1635 struct page ***pages,
1636 size_t maxsize,
1637 unsigned int maxpages,
1638 iov_iter_extraction_t extraction_flags,
1639 size_t *offset0)
1640 {
1641 maxsize = min_t(size_t, min_t(size_t, maxsize, i->count), MAX_RW_COUNT);
1642 if (!maxsize)
1643 return 0;
1644
1645 if (likely(user_backed_iter(i)))
1646 return iov_iter_extract_user_pages(i, pages, maxsize,
1647 maxpages, extraction_flags,
1648 offset0);
1649 if (iov_iter_is_kvec(i))
1650 return iov_iter_extract_kvec_pages(i, pages, maxsize,
1651 maxpages, extraction_flags,
1652 offset0);
1653 if (iov_iter_is_bvec(i))
1654 return iov_iter_extract_bvec_pages(i, pages, maxsize,
1655 maxpages, extraction_flags,
1656 offset0);
1657 if (iov_iter_is_xarray(i))
1658 return iov_iter_extract_xarray_pages(i, pages, maxsize,
1659 maxpages, extraction_flags,
1660 offset0);
1661 return -EFAULT;
1662 }
1663 EXPORT_SYMBOL_GPL(iov_iter_extract_pages);
1664
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.