Loading...
Note: File does not exist in v3.15.
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
16static __always_inline
17size_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
30static __always_inline
31size_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
44static __always_inline
45size_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
61static __always_inline
62size_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
69static __always_inline
70size_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 */
90size_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}
117EXPORT_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 */
133size_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}
160EXPORT_SYMBOL(fault_in_iov_iter_writeable);
161
162void 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}
177EXPORT_SYMBOL(iov_iter_init);
178
179size_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}
188EXPORT_SYMBOL(_copy_to_iter);
189
190#ifdef CONFIG_ARCH_HAS_COPY_MC
191static __always_inline
192size_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
203static __always_inline
204size_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 */
234size_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}
243EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
244#endif /* CONFIG_ARCH_HAS_COPY_MC */
245
246static __always_inline
247size_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
253size_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}
262EXPORT_SYMBOL(_copy_from_iter);
263
264static __always_inline
265size_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
271size_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}
280EXPORT_SYMBOL(_copy_from_iter_nocache);
281
282#ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
283static __always_inline
284size_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
290static __always_inline
291size_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 */
314size_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}
323EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
324#endif
325
326static 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
349size_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}
376EXPORT_SYMBOL(copy_page_to_iter);
377
378size_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}
409EXPORT_SYMBOL(copy_page_to_iter_nofault);
410
411size_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}
436EXPORT_SYMBOL(copy_page_from_iter);
437
438static __always_inline
439size_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
445static __always_inline
446size_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
453size_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}
458EXPORT_SYMBOL(iov_iter_zero);
459
460size_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 bool uses_kmap = IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) ||
465 PageHighMem(page);
466
467 if (!page_copy_sane(page, offset, bytes))
468 return 0;
469 if (WARN_ON_ONCE(!i->data_source))
470 return 0;
471
472 do {
473 char *p;
474
475 n = bytes - copied;
476 if (uses_kmap) {
477 page += offset / PAGE_SIZE;
478 offset %= PAGE_SIZE;
479 n = min_t(size_t, n, PAGE_SIZE - offset);
480 }
481
482 p = kmap_atomic(page) + offset;
483 n = __copy_from_iter(p, n, i);
484 kunmap_atomic(p);
485 copied += n;
486 offset += n;
487 } while (uses_kmap && copied != bytes && n > 0);
488
489 return copied;
490}
491EXPORT_SYMBOL(copy_page_from_iter_atomic);
492
493static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
494{
495 const struct bio_vec *bvec, *end;
496
497 if (!i->count)
498 return;
499 i->count -= size;
500
501 size += i->iov_offset;
502
503 for (bvec = i->bvec, end = bvec + i->nr_segs; bvec < end; bvec++) {
504 if (likely(size < bvec->bv_len))
505 break;
506 size -= bvec->bv_len;
507 }
508 i->iov_offset = size;
509 i->nr_segs -= bvec - i->bvec;
510 i->bvec = bvec;
511}
512
513static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
514{
515 const struct iovec *iov, *end;
516
517 if (!i->count)
518 return;
519 i->count -= size;
520
521 size += i->iov_offset; // from beginning of current segment
522 for (iov = iter_iov(i), end = iov + i->nr_segs; iov < end; iov++) {
523 if (likely(size < iov->iov_len))
524 break;
525 size -= iov->iov_len;
526 }
527 i->iov_offset = size;
528 i->nr_segs -= iov - iter_iov(i);
529 i->__iov = iov;
530}
531
532static void iov_iter_folioq_advance(struct iov_iter *i, size_t size)
533{
534 const struct folio_queue *folioq = i->folioq;
535 unsigned int slot = i->folioq_slot;
536
537 if (!i->count)
538 return;
539 i->count -= size;
540
541 if (slot >= folioq_nr_slots(folioq)) {
542 folioq = folioq->next;
543 slot = 0;
544 }
545
546 size += i->iov_offset; /* From beginning of current segment. */
547 do {
548 size_t fsize = folioq_folio_size(folioq, slot);
549
550 if (likely(size < fsize))
551 break;
552 size -= fsize;
553 slot++;
554 if (slot >= folioq_nr_slots(folioq) && folioq->next) {
555 folioq = folioq->next;
556 slot = 0;
557 }
558 } while (size);
559
560 i->iov_offset = size;
561 i->folioq_slot = slot;
562 i->folioq = folioq;
563}
564
565void iov_iter_advance(struct iov_iter *i, size_t size)
566{
567 if (unlikely(i->count < size))
568 size = i->count;
569 if (likely(iter_is_ubuf(i)) || unlikely(iov_iter_is_xarray(i))) {
570 i->iov_offset += size;
571 i->count -= size;
572 } else if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
573 /* iovec and kvec have identical layouts */
574 iov_iter_iovec_advance(i, size);
575 } else if (iov_iter_is_bvec(i)) {
576 iov_iter_bvec_advance(i, size);
577 } else if (iov_iter_is_folioq(i)) {
578 iov_iter_folioq_advance(i, size);
579 } else if (iov_iter_is_discard(i)) {
580 i->count -= size;
581 }
582}
583EXPORT_SYMBOL(iov_iter_advance);
584
585static void iov_iter_folioq_revert(struct iov_iter *i, size_t unroll)
586{
587 const struct folio_queue *folioq = i->folioq;
588 unsigned int slot = i->folioq_slot;
589
590 for (;;) {
591 size_t fsize;
592
593 if (slot == 0) {
594 folioq = folioq->prev;
595 slot = folioq_nr_slots(folioq);
596 }
597 slot--;
598
599 fsize = folioq_folio_size(folioq, slot);
600 if (unroll <= fsize) {
601 i->iov_offset = fsize - unroll;
602 break;
603 }
604 unroll -= fsize;
605 }
606
607 i->folioq_slot = slot;
608 i->folioq = folioq;
609}
610
611void iov_iter_revert(struct iov_iter *i, size_t unroll)
612{
613 if (!unroll)
614 return;
615 if (WARN_ON(unroll > MAX_RW_COUNT))
616 return;
617 i->count += unroll;
618 if (unlikely(iov_iter_is_discard(i)))
619 return;
620 if (unroll <= i->iov_offset) {
621 i->iov_offset -= unroll;
622 return;
623 }
624 unroll -= i->iov_offset;
625 if (iov_iter_is_xarray(i) || iter_is_ubuf(i)) {
626 BUG(); /* We should never go beyond the start of the specified
627 * range since we might then be straying into pages that
628 * aren't pinned.
629 */
630 } else if (iov_iter_is_bvec(i)) {
631 const struct bio_vec *bvec = i->bvec;
632 while (1) {
633 size_t n = (--bvec)->bv_len;
634 i->nr_segs++;
635 if (unroll <= n) {
636 i->bvec = bvec;
637 i->iov_offset = n - unroll;
638 return;
639 }
640 unroll -= n;
641 }
642 } else if (iov_iter_is_folioq(i)) {
643 i->iov_offset = 0;
644 iov_iter_folioq_revert(i, unroll);
645 } else { /* same logics for iovec and kvec */
646 const struct iovec *iov = iter_iov(i);
647 while (1) {
648 size_t n = (--iov)->iov_len;
649 i->nr_segs++;
650 if (unroll <= n) {
651 i->__iov = iov;
652 i->iov_offset = n - unroll;
653 return;
654 }
655 unroll -= n;
656 }
657 }
658}
659EXPORT_SYMBOL(iov_iter_revert);
660
661/*
662 * Return the count of just the current iov_iter segment.
663 */
664size_t iov_iter_single_seg_count(const struct iov_iter *i)
665{
666 if (i->nr_segs > 1) {
667 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
668 return min(i->count, iter_iov(i)->iov_len - i->iov_offset);
669 if (iov_iter_is_bvec(i))
670 return min(i->count, i->bvec->bv_len - i->iov_offset);
671 }
672 if (unlikely(iov_iter_is_folioq(i)))
673 return !i->count ? 0 :
674 umin(folioq_folio_size(i->folioq, i->folioq_slot), i->count);
675 return i->count;
676}
677EXPORT_SYMBOL(iov_iter_single_seg_count);
678
679void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
680 const struct kvec *kvec, unsigned long nr_segs,
681 size_t count)
682{
683 WARN_ON(direction & ~(READ | WRITE));
684 *i = (struct iov_iter){
685 .iter_type = ITER_KVEC,
686 .data_source = direction,
687 .kvec = kvec,
688 .nr_segs = nr_segs,
689 .iov_offset = 0,
690 .count = count
691 };
692}
693EXPORT_SYMBOL(iov_iter_kvec);
694
695void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
696 const struct bio_vec *bvec, unsigned long nr_segs,
697 size_t count)
698{
699 WARN_ON(direction & ~(READ | WRITE));
700 *i = (struct iov_iter){
701 .iter_type = ITER_BVEC,
702 .data_source = direction,
703 .bvec = bvec,
704 .nr_segs = nr_segs,
705 .iov_offset = 0,
706 .count = count
707 };
708}
709EXPORT_SYMBOL(iov_iter_bvec);
710
711/**
712 * iov_iter_folio_queue - Initialise an I/O iterator to use the folios in a folio queue
713 * @i: The iterator to initialise.
714 * @direction: The direction of the transfer.
715 * @folioq: The starting point in the folio queue.
716 * @first_slot: The first slot in the folio queue to use
717 * @offset: The offset into the folio in the first slot to start at
718 * @count: The size of the I/O buffer in bytes.
719 *
720 * Set up an I/O iterator to either draw data out of the pages attached to an
721 * inode or to inject data into those pages. The pages *must* be prevented
722 * from evaporation, either by taking a ref on them or locking them by the
723 * caller.
724 */
725void iov_iter_folio_queue(struct iov_iter *i, unsigned int direction,
726 const struct folio_queue *folioq, unsigned int first_slot,
727 unsigned int offset, size_t count)
728{
729 BUG_ON(direction & ~1);
730 *i = (struct iov_iter) {
731 .iter_type = ITER_FOLIOQ,
732 .data_source = direction,
733 .folioq = folioq,
734 .folioq_slot = first_slot,
735 .count = count,
736 .iov_offset = offset,
737 };
738}
739EXPORT_SYMBOL(iov_iter_folio_queue);
740
741/**
742 * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
743 * @i: The iterator to initialise.
744 * @direction: The direction of the transfer.
745 * @xarray: The xarray to access.
746 * @start: The start file position.
747 * @count: The size of the I/O buffer in bytes.
748 *
749 * Set up an I/O iterator to either draw data out of the pages attached to an
750 * inode or to inject data into those pages. The pages *must* be prevented
751 * from evaporation, either by taking a ref on them or locking them by the
752 * caller.
753 */
754void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
755 struct xarray *xarray, loff_t start, size_t count)
756{
757 BUG_ON(direction & ~1);
758 *i = (struct iov_iter) {
759 .iter_type = ITER_XARRAY,
760 .data_source = direction,
761 .xarray = xarray,
762 .xarray_start = start,
763 .count = count,
764 .iov_offset = 0
765 };
766}
767EXPORT_SYMBOL(iov_iter_xarray);
768
769/**
770 * iov_iter_discard - Initialise an I/O iterator that discards data
771 * @i: The iterator to initialise.
772 * @direction: The direction of the transfer.
773 * @count: The size of the I/O buffer in bytes.
774 *
775 * Set up an I/O iterator that just discards everything that's written to it.
776 * It's only available as a READ iterator.
777 */
778void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
779{
780 BUG_ON(direction != READ);
781 *i = (struct iov_iter){
782 .iter_type = ITER_DISCARD,
783 .data_source = false,
784 .count = count,
785 .iov_offset = 0
786 };
787}
788EXPORT_SYMBOL(iov_iter_discard);
789
790static bool iov_iter_aligned_iovec(const struct iov_iter *i, unsigned addr_mask,
791 unsigned len_mask)
792{
793 const struct iovec *iov = iter_iov(i);
794 size_t size = i->count;
795 size_t skip = i->iov_offset;
796
797 do {
798 size_t len = iov->iov_len - skip;
799
800 if (len > size)
801 len = size;
802 if (len & len_mask)
803 return false;
804 if ((unsigned long)(iov->iov_base + skip) & addr_mask)
805 return false;
806
807 iov++;
808 size -= len;
809 skip = 0;
810 } while (size);
811
812 return true;
813}
814
815static bool iov_iter_aligned_bvec(const struct iov_iter *i, unsigned addr_mask,
816 unsigned len_mask)
817{
818 const struct bio_vec *bvec = i->bvec;
819 unsigned skip = i->iov_offset;
820 size_t size = i->count;
821
822 do {
823 size_t len = bvec->bv_len;
824
825 if (len > size)
826 len = size;
827 if (len & len_mask)
828 return false;
829 if ((unsigned long)(bvec->bv_offset + skip) & addr_mask)
830 return false;
831
832 bvec++;
833 size -= len;
834 skip = 0;
835 } while (size);
836
837 return true;
838}
839
840/**
841 * iov_iter_is_aligned() - Check if the addresses and lengths of each segments
842 * are aligned to the parameters.
843 *
844 * @i: &struct iov_iter to restore
845 * @addr_mask: bit mask to check against the iov element's addresses
846 * @len_mask: bit mask to check against the iov element's lengths
847 *
848 * Return: false if any addresses or lengths intersect with the provided masks
849 */
850bool iov_iter_is_aligned(const struct iov_iter *i, unsigned addr_mask,
851 unsigned len_mask)
852{
853 if (likely(iter_is_ubuf(i))) {
854 if (i->count & len_mask)
855 return false;
856 if ((unsigned long)(i->ubuf + i->iov_offset) & addr_mask)
857 return false;
858 return true;
859 }
860
861 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
862 return iov_iter_aligned_iovec(i, addr_mask, len_mask);
863
864 if (iov_iter_is_bvec(i))
865 return iov_iter_aligned_bvec(i, addr_mask, len_mask);
866
867 /* With both xarray and folioq types, we're dealing with whole folios. */
868 if (iov_iter_is_xarray(i)) {
869 if (i->count & len_mask)
870 return false;
871 if ((i->xarray_start + i->iov_offset) & addr_mask)
872 return false;
873 }
874 if (iov_iter_is_folioq(i)) {
875 if (i->count & len_mask)
876 return false;
877 if (i->iov_offset & addr_mask)
878 return false;
879 }
880
881 return true;
882}
883EXPORT_SYMBOL_GPL(iov_iter_is_aligned);
884
885static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
886{
887 const struct iovec *iov = iter_iov(i);
888 unsigned long res = 0;
889 size_t size = i->count;
890 size_t skip = i->iov_offset;
891
892 do {
893 size_t len = iov->iov_len - skip;
894 if (len) {
895 res |= (unsigned long)iov->iov_base + skip;
896 if (len > size)
897 len = size;
898 res |= len;
899 size -= len;
900 }
901 iov++;
902 skip = 0;
903 } while (size);
904 return res;
905}
906
907static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
908{
909 const struct bio_vec *bvec = i->bvec;
910 unsigned res = 0;
911 size_t size = i->count;
912 unsigned skip = i->iov_offset;
913
914 do {
915 size_t len = bvec->bv_len - skip;
916 res |= (unsigned long)bvec->bv_offset + skip;
917 if (len > size)
918 len = size;
919 res |= len;
920 bvec++;
921 size -= len;
922 skip = 0;
923 } while (size);
924
925 return res;
926}
927
928unsigned long iov_iter_alignment(const struct iov_iter *i)
929{
930 if (likely(iter_is_ubuf(i))) {
931 size_t size = i->count;
932 if (size)
933 return ((unsigned long)i->ubuf + i->iov_offset) | size;
934 return 0;
935 }
936
937 /* iovec and kvec have identical layouts */
938 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
939 return iov_iter_alignment_iovec(i);
940
941 if (iov_iter_is_bvec(i))
942 return iov_iter_alignment_bvec(i);
943
944 /* With both xarray and folioq types, we're dealing with whole folios. */
945 if (iov_iter_is_folioq(i))
946 return i->iov_offset | i->count;
947 if (iov_iter_is_xarray(i))
948 return (i->xarray_start + i->iov_offset) | i->count;
949
950 return 0;
951}
952EXPORT_SYMBOL(iov_iter_alignment);
953
954unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
955{
956 unsigned long res = 0;
957 unsigned long v = 0;
958 size_t size = i->count;
959 unsigned k;
960
961 if (iter_is_ubuf(i))
962 return 0;
963
964 if (WARN_ON(!iter_is_iovec(i)))
965 return ~0U;
966
967 for (k = 0; k < i->nr_segs; k++) {
968 const struct iovec *iov = iter_iov(i) + k;
969 if (iov->iov_len) {
970 unsigned long base = (unsigned long)iov->iov_base;
971 if (v) // if not the first one
972 res |= base | v; // this start | previous end
973 v = base + iov->iov_len;
974 if (size <= iov->iov_len)
975 break;
976 size -= iov->iov_len;
977 }
978 }
979 return res;
980}
981EXPORT_SYMBOL(iov_iter_gap_alignment);
982
983static int want_pages_array(struct page ***res, size_t size,
984 size_t start, unsigned int maxpages)
985{
986 unsigned int count = DIV_ROUND_UP(size + start, PAGE_SIZE);
987
988 if (count > maxpages)
989 count = maxpages;
990 WARN_ON(!count); // caller should've prevented that
991 if (!*res) {
992 *res = kvmalloc_array(count, sizeof(struct page *), GFP_KERNEL);
993 if (!*res)
994 return 0;
995 }
996 return count;
997}
998
999static ssize_t iter_folioq_get_pages(struct iov_iter *iter,
1000 struct page ***ppages, size_t maxsize,
1001 unsigned maxpages, size_t *_start_offset)
1002{
1003 const struct folio_queue *folioq = iter->folioq;
1004 struct page **pages;
1005 unsigned int slot = iter->folioq_slot;
1006 size_t extracted = 0, count = iter->count, iov_offset = iter->iov_offset;
1007
1008 if (slot >= folioq_nr_slots(folioq)) {
1009 folioq = folioq->next;
1010 slot = 0;
1011 if (WARN_ON(iov_offset != 0))
1012 return -EIO;
1013 }
1014
1015 maxpages = want_pages_array(ppages, maxsize, iov_offset & ~PAGE_MASK, maxpages);
1016 if (!maxpages)
1017 return -ENOMEM;
1018 *_start_offset = iov_offset & ~PAGE_MASK;
1019 pages = *ppages;
1020
1021 for (;;) {
1022 struct folio *folio = folioq_folio(folioq, slot);
1023 size_t offset = iov_offset, fsize = folioq_folio_size(folioq, slot);
1024 size_t part = PAGE_SIZE - offset % PAGE_SIZE;
1025
1026 if (offset < fsize) {
1027 part = umin(part, umin(maxsize - extracted, fsize - offset));
1028 count -= part;
1029 iov_offset += part;
1030 extracted += part;
1031
1032 *pages = folio_page(folio, offset / PAGE_SIZE);
1033 get_page(*pages);
1034 pages++;
1035 maxpages--;
1036 }
1037
1038 if (maxpages == 0 || extracted >= maxsize)
1039 break;
1040
1041 if (iov_offset >= fsize) {
1042 iov_offset = 0;
1043 slot++;
1044 if (slot == folioq_nr_slots(folioq) && folioq->next) {
1045 folioq = folioq->next;
1046 slot = 0;
1047 }
1048 }
1049 }
1050
1051 iter->count = count;
1052 iter->iov_offset = iov_offset;
1053 iter->folioq = folioq;
1054 iter->folioq_slot = slot;
1055 return extracted;
1056}
1057
1058static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1059 pgoff_t index, unsigned int nr_pages)
1060{
1061 XA_STATE(xas, xa, index);
1062 struct page *page;
1063 unsigned int ret = 0;
1064
1065 rcu_read_lock();
1066 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1067 if (xas_retry(&xas, page))
1068 continue;
1069
1070 /* Has the page moved or been split? */
1071 if (unlikely(page != xas_reload(&xas))) {
1072 xas_reset(&xas);
1073 continue;
1074 }
1075
1076 pages[ret] = find_subpage(page, xas.xa_index);
1077 get_page(pages[ret]);
1078 if (++ret == nr_pages)
1079 break;
1080 }
1081 rcu_read_unlock();
1082 return ret;
1083}
1084
1085static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1086 struct page ***pages, size_t maxsize,
1087 unsigned maxpages, size_t *_start_offset)
1088{
1089 unsigned nr, offset, count;
1090 pgoff_t index;
1091 loff_t pos;
1092
1093 pos = i->xarray_start + i->iov_offset;
1094 index = pos >> PAGE_SHIFT;
1095 offset = pos & ~PAGE_MASK;
1096 *_start_offset = offset;
1097
1098 count = want_pages_array(pages, maxsize, offset, maxpages);
1099 if (!count)
1100 return -ENOMEM;
1101 nr = iter_xarray_populate_pages(*pages, i->xarray, index, count);
1102 if (nr == 0)
1103 return 0;
1104
1105 maxsize = min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1106 i->iov_offset += maxsize;
1107 i->count -= maxsize;
1108 return maxsize;
1109}
1110
1111/* must be done on non-empty ITER_UBUF or ITER_IOVEC one */
1112static unsigned long first_iovec_segment(const struct iov_iter *i, size_t *size)
1113{
1114 size_t skip;
1115 long k;
1116
1117 if (iter_is_ubuf(i))
1118 return (unsigned long)i->ubuf + i->iov_offset;
1119
1120 for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
1121 const struct iovec *iov = iter_iov(i) + k;
1122 size_t len = iov->iov_len - skip;
1123
1124 if (unlikely(!len))
1125 continue;
1126 if (*size > len)
1127 *size = len;
1128 return (unsigned long)iov->iov_base + skip;
1129 }
1130 BUG(); // if it had been empty, we wouldn't get called
1131}
1132
1133/* must be done on non-empty ITER_BVEC one */
1134static struct page *first_bvec_segment(const struct iov_iter *i,
1135 size_t *size, size_t *start)
1136{
1137 struct page *page;
1138 size_t skip = i->iov_offset, len;
1139
1140 len = i->bvec->bv_len - skip;
1141 if (*size > len)
1142 *size = len;
1143 skip += i->bvec->bv_offset;
1144 page = i->bvec->bv_page + skip / PAGE_SIZE;
1145 *start = skip % PAGE_SIZE;
1146 return page;
1147}
1148
1149static ssize_t __iov_iter_get_pages_alloc(struct iov_iter *i,
1150 struct page ***pages, size_t maxsize,
1151 unsigned int maxpages, size_t *start)
1152{
1153 unsigned int n, gup_flags = 0;
1154
1155 if (maxsize > i->count)
1156 maxsize = i->count;
1157 if (!maxsize)
1158 return 0;
1159 if (maxsize > MAX_RW_COUNT)
1160 maxsize = MAX_RW_COUNT;
1161
1162 if (likely(user_backed_iter(i))) {
1163 unsigned long addr;
1164 int res;
1165
1166 if (iov_iter_rw(i) != WRITE)
1167 gup_flags |= FOLL_WRITE;
1168 if (i->nofault)
1169 gup_flags |= FOLL_NOFAULT;
1170
1171 addr = first_iovec_segment(i, &maxsize);
1172 *start = addr % PAGE_SIZE;
1173 addr &= PAGE_MASK;
1174 n = want_pages_array(pages, maxsize, *start, maxpages);
1175 if (!n)
1176 return -ENOMEM;
1177 res = get_user_pages_fast(addr, n, gup_flags, *pages);
1178 if (unlikely(res <= 0))
1179 return res;
1180 maxsize = min_t(size_t, maxsize, res * PAGE_SIZE - *start);
1181 iov_iter_advance(i, maxsize);
1182 return maxsize;
1183 }
1184 if (iov_iter_is_bvec(i)) {
1185 struct page **p;
1186 struct page *page;
1187
1188 page = first_bvec_segment(i, &maxsize, start);
1189 n = want_pages_array(pages, maxsize, *start, maxpages);
1190 if (!n)
1191 return -ENOMEM;
1192 p = *pages;
1193 for (int k = 0; k < n; k++)
1194 get_page(p[k] = page + k);
1195 maxsize = min_t(size_t, maxsize, n * PAGE_SIZE - *start);
1196 i->count -= maxsize;
1197 i->iov_offset += maxsize;
1198 if (i->iov_offset == i->bvec->bv_len) {
1199 i->iov_offset = 0;
1200 i->bvec++;
1201 i->nr_segs--;
1202 }
1203 return maxsize;
1204 }
1205 if (iov_iter_is_folioq(i))
1206 return iter_folioq_get_pages(i, pages, maxsize, maxpages, start);
1207 if (iov_iter_is_xarray(i))
1208 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1209 return -EFAULT;
1210}
1211
1212ssize_t iov_iter_get_pages2(struct iov_iter *i, struct page **pages,
1213 size_t maxsize, unsigned maxpages, size_t *start)
1214{
1215 if (!maxpages)
1216 return 0;
1217 BUG_ON(!pages);
1218
1219 return __iov_iter_get_pages_alloc(i, &pages, maxsize, maxpages, start);
1220}
1221EXPORT_SYMBOL(iov_iter_get_pages2);
1222
1223ssize_t iov_iter_get_pages_alloc2(struct iov_iter *i,
1224 struct page ***pages, size_t maxsize, size_t *start)
1225{
1226 ssize_t len;
1227
1228 *pages = NULL;
1229
1230 len = __iov_iter_get_pages_alloc(i, pages, maxsize, ~0U, start);
1231 if (len <= 0) {
1232 kvfree(*pages);
1233 *pages = NULL;
1234 }
1235 return len;
1236}
1237EXPORT_SYMBOL(iov_iter_get_pages_alloc2);
1238
1239static int iov_npages(const struct iov_iter *i, int maxpages)
1240{
1241 size_t skip = i->iov_offset, size = i->count;
1242 const struct iovec *p;
1243 int npages = 0;
1244
1245 for (p = iter_iov(i); size; skip = 0, p++) {
1246 unsigned offs = offset_in_page(p->iov_base + skip);
1247 size_t len = min(p->iov_len - skip, size);
1248
1249 if (len) {
1250 size -= len;
1251 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1252 if (unlikely(npages > maxpages))
1253 return maxpages;
1254 }
1255 }
1256 return npages;
1257}
1258
1259static int bvec_npages(const struct iov_iter *i, int maxpages)
1260{
1261 size_t skip = i->iov_offset, size = i->count;
1262 const struct bio_vec *p;
1263 int npages = 0;
1264
1265 for (p = i->bvec; size; skip = 0, p++) {
1266 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1267 size_t len = min(p->bv_len - skip, size);
1268
1269 size -= len;
1270 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1271 if (unlikely(npages > maxpages))
1272 return maxpages;
1273 }
1274 return npages;
1275}
1276
1277int iov_iter_npages(const struct iov_iter *i, int maxpages)
1278{
1279 if (unlikely(!i->count))
1280 return 0;
1281 if (likely(iter_is_ubuf(i))) {
1282 unsigned offs = offset_in_page(i->ubuf + i->iov_offset);
1283 int npages = DIV_ROUND_UP(offs + i->count, PAGE_SIZE);
1284 return min(npages, maxpages);
1285 }
1286 /* iovec and kvec have identical layouts */
1287 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1288 return iov_npages(i, maxpages);
1289 if (iov_iter_is_bvec(i))
1290 return bvec_npages(i, maxpages);
1291 if (iov_iter_is_folioq(i)) {
1292 unsigned offset = i->iov_offset % PAGE_SIZE;
1293 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1294 return min(npages, maxpages);
1295 }
1296 if (iov_iter_is_xarray(i)) {
1297 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1298 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1299 return min(npages, maxpages);
1300 }
1301 return 0;
1302}
1303EXPORT_SYMBOL(iov_iter_npages);
1304
1305const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1306{
1307 *new = *old;
1308 if (iov_iter_is_bvec(new))
1309 return new->bvec = kmemdup(new->bvec,
1310 new->nr_segs * sizeof(struct bio_vec),
1311 flags);
1312 else if (iov_iter_is_kvec(new) || iter_is_iovec(new))
1313 /* iovec and kvec have identical layout */
1314 return new->__iov = kmemdup(new->__iov,
1315 new->nr_segs * sizeof(struct iovec),
1316 flags);
1317 return NULL;
1318}
1319EXPORT_SYMBOL(dup_iter);
1320
1321static __noclone int copy_compat_iovec_from_user(struct iovec *iov,
1322 const struct iovec __user *uvec, u32 nr_segs)
1323{
1324 const struct compat_iovec __user *uiov =
1325 (const struct compat_iovec __user *)uvec;
1326 int ret = -EFAULT;
1327 u32 i;
1328
1329 if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1330 return -EFAULT;
1331
1332 for (i = 0; i < nr_segs; i++) {
1333 compat_uptr_t buf;
1334 compat_ssize_t len;
1335
1336 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1337 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1338
1339 /* check for compat_size_t not fitting in compat_ssize_t .. */
1340 if (len < 0) {
1341 ret = -EINVAL;
1342 goto uaccess_end;
1343 }
1344 iov[i].iov_base = compat_ptr(buf);
1345 iov[i].iov_len = len;
1346 }
1347
1348 ret = 0;
1349uaccess_end:
1350 user_access_end();
1351 return ret;
1352}
1353
1354static __noclone int copy_iovec_from_user(struct iovec *iov,
1355 const struct iovec __user *uiov, unsigned long nr_segs)
1356{
1357 int ret = -EFAULT;
1358
1359 if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1360 return -EFAULT;
1361
1362 do {
1363 void __user *buf;
1364 ssize_t len;
1365
1366 unsafe_get_user(len, &uiov->iov_len, uaccess_end);
1367 unsafe_get_user(buf, &uiov->iov_base, uaccess_end);
1368
1369 /* check for size_t not fitting in ssize_t .. */
1370 if (unlikely(len < 0)) {
1371 ret = -EINVAL;
1372 goto uaccess_end;
1373 }
1374 iov->iov_base = buf;
1375 iov->iov_len = len;
1376
1377 uiov++; iov++;
1378 } while (--nr_segs);
1379
1380 ret = 0;
1381uaccess_end:
1382 user_access_end();
1383 return ret;
1384}
1385
1386struct iovec *iovec_from_user(const struct iovec __user *uvec,
1387 unsigned long nr_segs, unsigned long fast_segs,
1388 struct iovec *fast_iov, bool compat)
1389{
1390 struct iovec *iov = fast_iov;
1391 int ret;
1392
1393 /*
1394 * SuS says "The readv() function *may* fail if the iovcnt argument was
1395 * less than or equal to 0, or greater than {IOV_MAX}. Linux has
1396 * traditionally returned zero for zero segments, so...
1397 */
1398 if (nr_segs == 0)
1399 return iov;
1400 if (nr_segs > UIO_MAXIOV)
1401 return ERR_PTR(-EINVAL);
1402 if (nr_segs > fast_segs) {
1403 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1404 if (!iov)
1405 return ERR_PTR(-ENOMEM);
1406 }
1407
1408 if (unlikely(compat))
1409 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1410 else
1411 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1412 if (ret) {
1413 if (iov != fast_iov)
1414 kfree(iov);
1415 return ERR_PTR(ret);
1416 }
1417
1418 return iov;
1419}
1420
1421/*
1422 * Single segment iovec supplied by the user, import it as ITER_UBUF.
1423 */
1424static ssize_t __import_iovec_ubuf(int type, const struct iovec __user *uvec,
1425 struct iovec **iovp, struct iov_iter *i,
1426 bool compat)
1427{
1428 struct iovec *iov = *iovp;
1429 ssize_t ret;
1430
1431 *iovp = NULL;
1432
1433 if (compat)
1434 ret = copy_compat_iovec_from_user(iov, uvec, 1);
1435 else
1436 ret = copy_iovec_from_user(iov, uvec, 1);
1437 if (unlikely(ret))
1438 return ret;
1439
1440 ret = import_ubuf(type, iov->iov_base, iov->iov_len, i);
1441 if (unlikely(ret))
1442 return ret;
1443 return i->count;
1444}
1445
1446ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1447 unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1448 struct iov_iter *i, bool compat)
1449{
1450 ssize_t total_len = 0;
1451 unsigned long seg;
1452 struct iovec *iov;
1453
1454 if (nr_segs == 1)
1455 return __import_iovec_ubuf(type, uvec, iovp, i, compat);
1456
1457 iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1458 if (IS_ERR(iov)) {
1459 *iovp = NULL;
1460 return PTR_ERR(iov);
1461 }
1462
1463 /*
1464 * According to the Single Unix Specification we should return EINVAL if
1465 * an element length is < 0 when cast to ssize_t or if the total length
1466 * would overflow the ssize_t return value of the system call.
1467 *
1468 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1469 * overflow case.
1470 */
1471 for (seg = 0; seg < nr_segs; seg++) {
1472 ssize_t len = (ssize_t)iov[seg].iov_len;
1473
1474 if (!access_ok(iov[seg].iov_base, len)) {
1475 if (iov != *iovp)
1476 kfree(iov);
1477 *iovp = NULL;
1478 return -EFAULT;
1479 }
1480
1481 if (len > MAX_RW_COUNT - total_len) {
1482 len = MAX_RW_COUNT - total_len;
1483 iov[seg].iov_len = len;
1484 }
1485 total_len += len;
1486 }
1487
1488 iov_iter_init(i, type, iov, nr_segs, total_len);
1489 if (iov == *iovp)
1490 *iovp = NULL;
1491 else
1492 *iovp = iov;
1493 return total_len;
1494}
1495
1496/**
1497 * import_iovec() - Copy an array of &struct iovec from userspace
1498 * into the kernel, check that it is valid, and initialize a new
1499 * &struct iov_iter iterator to access it.
1500 *
1501 * @type: One of %READ or %WRITE.
1502 * @uvec: Pointer to the userspace array.
1503 * @nr_segs: Number of elements in userspace array.
1504 * @fast_segs: Number of elements in @iov.
1505 * @iovp: (input and output parameter) Pointer to pointer to (usually small
1506 * on-stack) kernel array.
1507 * @i: Pointer to iterator that will be initialized on success.
1508 *
1509 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1510 * then this function places %NULL in *@iov on return. Otherwise, a new
1511 * array will be allocated and the result placed in *@iov. This means that
1512 * the caller may call kfree() on *@iov regardless of whether the small
1513 * on-stack array was used or not (and regardless of whether this function
1514 * returns an error or not).
1515 *
1516 * Return: Negative error code on error, bytes imported on success
1517 */
1518ssize_t import_iovec(int type, const struct iovec __user *uvec,
1519 unsigned nr_segs, unsigned fast_segs,
1520 struct iovec **iovp, struct iov_iter *i)
1521{
1522 return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
1523 in_compat_syscall());
1524}
1525EXPORT_SYMBOL(import_iovec);
1526
1527int import_ubuf(int rw, void __user *buf, size_t len, struct iov_iter *i)
1528{
1529 if (len > MAX_RW_COUNT)
1530 len = MAX_RW_COUNT;
1531 if (unlikely(!access_ok(buf, len)))
1532 return -EFAULT;
1533
1534 iov_iter_ubuf(i, rw, buf, len);
1535 return 0;
1536}
1537EXPORT_SYMBOL_GPL(import_ubuf);
1538
1539/**
1540 * iov_iter_restore() - Restore a &struct iov_iter to the same state as when
1541 * iov_iter_save_state() was called.
1542 *
1543 * @i: &struct iov_iter to restore
1544 * @state: state to restore from
1545 *
1546 * Used after iov_iter_save_state() to bring restore @i, if operations may
1547 * have advanced it.
1548 *
1549 * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC
1550 */
1551void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state)
1552{
1553 if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i) &&
1554 !iter_is_ubuf(i)) && !iov_iter_is_kvec(i))
1555 return;
1556 i->iov_offset = state->iov_offset;
1557 i->count = state->count;
1558 if (iter_is_ubuf(i))
1559 return;
1560 /*
1561 * For the *vec iters, nr_segs + iov is constant - if we increment
1562 * the vec, then we also decrement the nr_segs count. Hence we don't
1563 * need to track both of these, just one is enough and we can deduct
1564 * the other from that. ITER_KVEC and ITER_IOVEC are the same struct
1565 * size, so we can just increment the iov pointer as they are unionzed.
1566 * ITER_BVEC _may_ be the same size on some archs, but on others it is
1567 * not. Be safe and handle it separately.
1568 */
1569 BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
1570 if (iov_iter_is_bvec(i))
1571 i->bvec -= state->nr_segs - i->nr_segs;
1572 else
1573 i->__iov -= state->nr_segs - i->nr_segs;
1574 i->nr_segs = state->nr_segs;
1575}
1576
1577/*
1578 * Extract a list of contiguous pages from an ITER_FOLIOQ iterator. This does
1579 * not get references on the pages, nor does it get a pin on them.
1580 */
1581static ssize_t iov_iter_extract_folioq_pages(struct iov_iter *i,
1582 struct page ***pages, size_t maxsize,
1583 unsigned int maxpages,
1584 iov_iter_extraction_t extraction_flags,
1585 size_t *offset0)
1586{
1587 const struct folio_queue *folioq = i->folioq;
1588 struct page **p;
1589 unsigned int nr = 0;
1590 size_t extracted = 0, offset, slot = i->folioq_slot;
1591
1592 if (slot >= folioq_nr_slots(folioq)) {
1593 folioq = folioq->next;
1594 slot = 0;
1595 if (WARN_ON(i->iov_offset != 0))
1596 return -EIO;
1597 }
1598
1599 offset = i->iov_offset & ~PAGE_MASK;
1600 *offset0 = offset;
1601
1602 maxpages = want_pages_array(pages, maxsize, offset, maxpages);
1603 if (!maxpages)
1604 return -ENOMEM;
1605 p = *pages;
1606
1607 for (;;) {
1608 struct folio *folio = folioq_folio(folioq, slot);
1609 size_t offset = i->iov_offset, fsize = folioq_folio_size(folioq, slot);
1610 size_t part = PAGE_SIZE - offset % PAGE_SIZE;
1611
1612 if (offset < fsize) {
1613 part = umin(part, umin(maxsize - extracted, fsize - offset));
1614 i->count -= part;
1615 i->iov_offset += part;
1616 extracted += part;
1617
1618 p[nr++] = folio_page(folio, offset / PAGE_SIZE);
1619 }
1620
1621 if (nr >= maxpages || extracted >= maxsize)
1622 break;
1623
1624 if (i->iov_offset >= fsize) {
1625 i->iov_offset = 0;
1626 slot++;
1627 if (slot == folioq_nr_slots(folioq) && folioq->next) {
1628 folioq = folioq->next;
1629 slot = 0;
1630 }
1631 }
1632 }
1633
1634 i->folioq = folioq;
1635 i->folioq_slot = slot;
1636 return extracted;
1637}
1638
1639/*
1640 * Extract a list of contiguous pages from an ITER_XARRAY iterator. This does not
1641 * get references on the pages, nor does it get a pin on them.
1642 */
1643static ssize_t iov_iter_extract_xarray_pages(struct iov_iter *i,
1644 struct page ***pages, size_t maxsize,
1645 unsigned int maxpages,
1646 iov_iter_extraction_t extraction_flags,
1647 size_t *offset0)
1648{
1649 struct page *page, **p;
1650 unsigned int nr = 0, offset;
1651 loff_t pos = i->xarray_start + i->iov_offset;
1652 pgoff_t index = pos >> PAGE_SHIFT;
1653 XA_STATE(xas, i->xarray, index);
1654
1655 offset = pos & ~PAGE_MASK;
1656 *offset0 = offset;
1657
1658 maxpages = want_pages_array(pages, maxsize, offset, maxpages);
1659 if (!maxpages)
1660 return -ENOMEM;
1661 p = *pages;
1662
1663 rcu_read_lock();
1664 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1665 if (xas_retry(&xas, page))
1666 continue;
1667
1668 /* Has the page moved or been split? */
1669 if (unlikely(page != xas_reload(&xas))) {
1670 xas_reset(&xas);
1671 continue;
1672 }
1673
1674 p[nr++] = find_subpage(page, xas.xa_index);
1675 if (nr == maxpages)
1676 break;
1677 }
1678 rcu_read_unlock();
1679
1680 maxsize = min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1681 iov_iter_advance(i, maxsize);
1682 return maxsize;
1683}
1684
1685/*
1686 * Extract a list of virtually contiguous pages from an ITER_BVEC iterator.
1687 * This does not get references on the pages, nor does it get a pin on them.
1688 */
1689static ssize_t iov_iter_extract_bvec_pages(struct iov_iter *i,
1690 struct page ***pages, size_t maxsize,
1691 unsigned int maxpages,
1692 iov_iter_extraction_t extraction_flags,
1693 size_t *offset0)
1694{
1695 size_t skip = i->iov_offset, size = 0;
1696 struct bvec_iter bi;
1697 int k = 0;
1698
1699 if (i->nr_segs == 0)
1700 return 0;
1701
1702 if (i->iov_offset == i->bvec->bv_len) {
1703 i->iov_offset = 0;
1704 i->nr_segs--;
1705 i->bvec++;
1706 skip = 0;
1707 }
1708 bi.bi_idx = 0;
1709 bi.bi_size = maxsize;
1710 bi.bi_bvec_done = skip;
1711
1712 maxpages = want_pages_array(pages, maxsize, skip, maxpages);
1713
1714 while (bi.bi_size && bi.bi_idx < i->nr_segs) {
1715 struct bio_vec bv = bvec_iter_bvec(i->bvec, bi);
1716
1717 /*
1718 * The iov_iter_extract_pages interface only allows an offset
1719 * into the first page. Break out of the loop if we see an
1720 * offset into subsequent pages, the caller will have to call
1721 * iov_iter_extract_pages again for the reminder.
1722 */
1723 if (k) {
1724 if (bv.bv_offset)
1725 break;
1726 } else {
1727 *offset0 = bv.bv_offset;
1728 }
1729
1730 (*pages)[k++] = bv.bv_page;
1731 size += bv.bv_len;
1732
1733 if (k >= maxpages)
1734 break;
1735
1736 /*
1737 * We are done when the end of the bvec doesn't align to a page
1738 * boundary as that would create a hole in the returned space.
1739 * The caller will handle this with another call to
1740 * iov_iter_extract_pages.
1741 */
1742 if (bv.bv_offset + bv.bv_len != PAGE_SIZE)
1743 break;
1744
1745 bvec_iter_advance_single(i->bvec, &bi, bv.bv_len);
1746 }
1747
1748 iov_iter_advance(i, size);
1749 return size;
1750}
1751
1752/*
1753 * Extract a list of virtually contiguous pages from an ITER_KVEC iterator.
1754 * This does not get references on the pages, nor does it get a pin on them.
1755 */
1756static ssize_t iov_iter_extract_kvec_pages(struct iov_iter *i,
1757 struct page ***pages, size_t maxsize,
1758 unsigned int maxpages,
1759 iov_iter_extraction_t extraction_flags,
1760 size_t *offset0)
1761{
1762 struct page **p, *page;
1763 const void *kaddr;
1764 size_t skip = i->iov_offset, offset, len, size;
1765 int k;
1766
1767 for (;;) {
1768 if (i->nr_segs == 0)
1769 return 0;
1770 size = min(maxsize, i->kvec->iov_len - skip);
1771 if (size)
1772 break;
1773 i->iov_offset = 0;
1774 i->nr_segs--;
1775 i->kvec++;
1776 skip = 0;
1777 }
1778
1779 kaddr = i->kvec->iov_base + skip;
1780 offset = (unsigned long)kaddr & ~PAGE_MASK;
1781 *offset0 = offset;
1782
1783 maxpages = want_pages_array(pages, size, offset, maxpages);
1784 if (!maxpages)
1785 return -ENOMEM;
1786 p = *pages;
1787
1788 kaddr -= offset;
1789 len = offset + size;
1790 for (k = 0; k < maxpages; k++) {
1791 size_t seg = min_t(size_t, len, PAGE_SIZE);
1792
1793 if (is_vmalloc_or_module_addr(kaddr))
1794 page = vmalloc_to_page(kaddr);
1795 else
1796 page = virt_to_page(kaddr);
1797
1798 p[k] = page;
1799 len -= seg;
1800 kaddr += PAGE_SIZE;
1801 }
1802
1803 size = min_t(size_t, size, maxpages * PAGE_SIZE - offset);
1804 iov_iter_advance(i, size);
1805 return size;
1806}
1807
1808/*
1809 * Extract a list of contiguous pages from a user iterator and get a pin on
1810 * each of them. This should only be used if the iterator is user-backed
1811 * (IOBUF/UBUF).
1812 *
1813 * It does not get refs on the pages, but the pages must be unpinned by the
1814 * caller once the transfer is complete.
1815 *
1816 * This is safe to be used where background IO/DMA *is* going to be modifying
1817 * the buffer; using a pin rather than a ref makes forces fork() to give the
1818 * child a copy of the page.
1819 */
1820static ssize_t iov_iter_extract_user_pages(struct iov_iter *i,
1821 struct page ***pages,
1822 size_t maxsize,
1823 unsigned int maxpages,
1824 iov_iter_extraction_t extraction_flags,
1825 size_t *offset0)
1826{
1827 unsigned long addr;
1828 unsigned int gup_flags = 0;
1829 size_t offset;
1830 int res;
1831
1832 if (i->data_source == ITER_DEST)
1833 gup_flags |= FOLL_WRITE;
1834 if (extraction_flags & ITER_ALLOW_P2PDMA)
1835 gup_flags |= FOLL_PCI_P2PDMA;
1836 if (i->nofault)
1837 gup_flags |= FOLL_NOFAULT;
1838
1839 addr = first_iovec_segment(i, &maxsize);
1840 *offset0 = offset = addr % PAGE_SIZE;
1841 addr &= PAGE_MASK;
1842 maxpages = want_pages_array(pages, maxsize, offset, maxpages);
1843 if (!maxpages)
1844 return -ENOMEM;
1845 res = pin_user_pages_fast(addr, maxpages, gup_flags, *pages);
1846 if (unlikely(res <= 0))
1847 return res;
1848 maxsize = min_t(size_t, maxsize, res * PAGE_SIZE - offset);
1849 iov_iter_advance(i, maxsize);
1850 return maxsize;
1851}
1852
1853/**
1854 * iov_iter_extract_pages - Extract a list of contiguous pages from an iterator
1855 * @i: The iterator to extract from
1856 * @pages: Where to return the list of pages
1857 * @maxsize: The maximum amount of iterator to extract
1858 * @maxpages: The maximum size of the list of pages
1859 * @extraction_flags: Flags to qualify request
1860 * @offset0: Where to return the starting offset into (*@pages)[0]
1861 *
1862 * Extract a list of contiguous pages from the current point of the iterator,
1863 * advancing the iterator. The maximum number of pages and the maximum amount
1864 * of page contents can be set.
1865 *
1866 * If *@pages is NULL, a page list will be allocated to the required size and
1867 * *@pages will be set to its base. If *@pages is not NULL, it will be assumed
1868 * that the caller allocated a page list at least @maxpages in size and this
1869 * will be filled in.
1870 *
1871 * @extraction_flags can have ITER_ALLOW_P2PDMA set to request peer-to-peer DMA
1872 * be allowed on the pages extracted.
1873 *
1874 * The iov_iter_extract_will_pin() function can be used to query how cleanup
1875 * should be performed.
1876 *
1877 * Extra refs or pins on the pages may be obtained as follows:
1878 *
1879 * (*) If the iterator is user-backed (ITER_IOVEC/ITER_UBUF), pins will be
1880 * added to the pages, but refs will not be taken.
1881 * iov_iter_extract_will_pin() will return true.
1882 *
1883 * (*) If the iterator is ITER_KVEC, ITER_BVEC, ITER_FOLIOQ or ITER_XARRAY, the
1884 * pages are merely listed; no extra refs or pins are obtained.
1885 * iov_iter_extract_will_pin() will return 0.
1886 *
1887 * Note also:
1888 *
1889 * (*) Use with ITER_DISCARD is not supported as that has no content.
1890 *
1891 * On success, the function sets *@pages to the new pagelist, if allocated, and
1892 * sets *offset0 to the offset into the first page.
1893 *
1894 * It may also return -ENOMEM and -EFAULT.
1895 */
1896ssize_t iov_iter_extract_pages(struct iov_iter *i,
1897 struct page ***pages,
1898 size_t maxsize,
1899 unsigned int maxpages,
1900 iov_iter_extraction_t extraction_flags,
1901 size_t *offset0)
1902{
1903 maxsize = min_t(size_t, min_t(size_t, maxsize, i->count), MAX_RW_COUNT);
1904 if (!maxsize)
1905 return 0;
1906
1907 if (likely(user_backed_iter(i)))
1908 return iov_iter_extract_user_pages(i, pages, maxsize,
1909 maxpages, extraction_flags,
1910 offset0);
1911 if (iov_iter_is_kvec(i))
1912 return iov_iter_extract_kvec_pages(i, pages, maxsize,
1913 maxpages, extraction_flags,
1914 offset0);
1915 if (iov_iter_is_bvec(i))
1916 return iov_iter_extract_bvec_pages(i, pages, maxsize,
1917 maxpages, extraction_flags,
1918 offset0);
1919 if (iov_iter_is_folioq(i))
1920 return iov_iter_extract_folioq_pages(i, pages, maxsize,
1921 maxpages, extraction_flags,
1922 offset0);
1923 if (iov_iter_is_xarray(i))
1924 return iov_iter_extract_xarray_pages(i, pages, maxsize,
1925 maxpages, extraction_flags,
1926 offset0);
1927 return -EFAULT;
1928}
1929EXPORT_SYMBOL_GPL(iov_iter_extract_pages);