Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2#include <crypto/hash.h>
3#include <linux/export.h>
4#include <linux/bvec.h>
5#include <linux/fault-inject-usercopy.h>
6#include <linux/uio.h>
7#include <linux/pagemap.h>
8#include <linux/highmem.h>
9#include <linux/slab.h>
10#include <linux/vmalloc.h>
11#include <linux/splice.h>
12#include <linux/compat.h>
13#include <net/checksum.h>
14#include <linux/scatterlist.h>
15#include <linux/instrumented.h>
16
17#define PIPE_PARANOIA /* for now */
18
19/* covers iovec and kvec alike */
20#define iterate_iovec(i, n, base, len, off, __p, STEP) { \
21 size_t off = 0; \
22 size_t skip = i->iov_offset; \
23 do { \
24 len = min(n, __p->iov_len - skip); \
25 if (likely(len)) { \
26 base = __p->iov_base + skip; \
27 len -= (STEP); \
28 off += len; \
29 skip += len; \
30 n -= len; \
31 if (skip < __p->iov_len) \
32 break; \
33 } \
34 __p++; \
35 skip = 0; \
36 } while (n); \
37 i->iov_offset = skip; \
38 n = off; \
39}
40
41#define iterate_bvec(i, n, base, len, off, p, STEP) { \
42 size_t off = 0; \
43 unsigned skip = i->iov_offset; \
44 while (n) { \
45 unsigned offset = p->bv_offset + skip; \
46 unsigned left; \
47 void *kaddr = kmap_local_page(p->bv_page + \
48 offset / PAGE_SIZE); \
49 base = kaddr + offset % PAGE_SIZE; \
50 len = min(min(n, (size_t)(p->bv_len - skip)), \
51 (size_t)(PAGE_SIZE - offset % PAGE_SIZE)); \
52 left = (STEP); \
53 kunmap_local(kaddr); \
54 len -= left; \
55 off += len; \
56 skip += len; \
57 if (skip == p->bv_len) { \
58 skip = 0; \
59 p++; \
60 } \
61 n -= len; \
62 if (left) \
63 break; \
64 } \
65 i->iov_offset = skip; \
66 n = off; \
67}
68
69#define iterate_xarray(i, n, base, len, __off, STEP) { \
70 __label__ __out; \
71 size_t __off = 0; \
72 struct page *head = NULL; \
73 loff_t start = i->xarray_start + i->iov_offset; \
74 unsigned offset = start % PAGE_SIZE; \
75 pgoff_t index = start / PAGE_SIZE; \
76 int j; \
77 \
78 XA_STATE(xas, i->xarray, index); \
79 \
80 rcu_read_lock(); \
81 xas_for_each(&xas, head, ULONG_MAX) { \
82 unsigned left; \
83 if (xas_retry(&xas, head)) \
84 continue; \
85 if (WARN_ON(xa_is_value(head))) \
86 break; \
87 if (WARN_ON(PageHuge(head))) \
88 break; \
89 for (j = (head->index < index) ? index - head->index : 0; \
90 j < thp_nr_pages(head); j++) { \
91 void *kaddr = kmap_local_page(head + j); \
92 base = kaddr + offset; \
93 len = PAGE_SIZE - offset; \
94 len = min(n, len); \
95 left = (STEP); \
96 kunmap_local(kaddr); \
97 len -= left; \
98 __off += len; \
99 n -= len; \
100 if (left || n == 0) \
101 goto __out; \
102 offset = 0; \
103 } \
104 } \
105__out: \
106 rcu_read_unlock(); \
107 i->iov_offset += __off; \
108 n = __off; \
109}
110
111#define __iterate_and_advance(i, n, base, len, off, I, K) { \
112 if (unlikely(i->count < n)) \
113 n = i->count; \
114 if (likely(n)) { \
115 if (likely(iter_is_iovec(i))) { \
116 const struct iovec *iov = i->iov; \
117 void __user *base; \
118 size_t len; \
119 iterate_iovec(i, n, base, len, off, \
120 iov, (I)) \
121 i->nr_segs -= iov - i->iov; \
122 i->iov = iov; \
123 } else if (iov_iter_is_bvec(i)) { \
124 const struct bio_vec *bvec = i->bvec; \
125 void *base; \
126 size_t len; \
127 iterate_bvec(i, n, base, len, off, \
128 bvec, (K)) \
129 i->nr_segs -= bvec - i->bvec; \
130 i->bvec = bvec; \
131 } else if (iov_iter_is_kvec(i)) { \
132 const struct kvec *kvec = i->kvec; \
133 void *base; \
134 size_t len; \
135 iterate_iovec(i, n, base, len, off, \
136 kvec, (K)) \
137 i->nr_segs -= kvec - i->kvec; \
138 i->kvec = kvec; \
139 } else if (iov_iter_is_xarray(i)) { \
140 void *base; \
141 size_t len; \
142 iterate_xarray(i, n, base, len, off, \
143 (K)) \
144 } \
145 i->count -= n; \
146 } \
147}
148#define iterate_and_advance(i, n, base, len, off, I, K) \
149 __iterate_and_advance(i, n, base, len, off, I, ((void)(K),0))
150
151static int copyout(void __user *to, const void *from, size_t n)
152{
153 if (should_fail_usercopy())
154 return n;
155 if (access_ok(to, n)) {
156 instrument_copy_to_user(to, from, n);
157 n = raw_copy_to_user(to, from, n);
158 }
159 return n;
160}
161
162static int copyin(void *to, const void __user *from, size_t n)
163{
164 if (should_fail_usercopy())
165 return n;
166 if (access_ok(from, n)) {
167 instrument_copy_from_user(to, from, n);
168 n = raw_copy_from_user(to, from, n);
169 }
170 return n;
171}
172
173static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes,
174 struct iov_iter *i)
175{
176 size_t skip, copy, left, wanted;
177 const struct iovec *iov;
178 char __user *buf;
179 void *kaddr, *from;
180
181 if (unlikely(bytes > i->count))
182 bytes = i->count;
183
184 if (unlikely(!bytes))
185 return 0;
186
187 might_fault();
188 wanted = bytes;
189 iov = i->iov;
190 skip = i->iov_offset;
191 buf = iov->iov_base + skip;
192 copy = min(bytes, iov->iov_len - skip);
193
194 if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_writeable(buf, copy)) {
195 kaddr = kmap_atomic(page);
196 from = kaddr + offset;
197
198 /* first chunk, usually the only one */
199 left = copyout(buf, from, copy);
200 copy -= left;
201 skip += copy;
202 from += copy;
203 bytes -= copy;
204
205 while (unlikely(!left && bytes)) {
206 iov++;
207 buf = iov->iov_base;
208 copy = min(bytes, iov->iov_len);
209 left = copyout(buf, from, copy);
210 copy -= left;
211 skip = copy;
212 from += copy;
213 bytes -= copy;
214 }
215 if (likely(!bytes)) {
216 kunmap_atomic(kaddr);
217 goto done;
218 }
219 offset = from - kaddr;
220 buf += copy;
221 kunmap_atomic(kaddr);
222 copy = min(bytes, iov->iov_len - skip);
223 }
224 /* Too bad - revert to non-atomic kmap */
225
226 kaddr = kmap(page);
227 from = kaddr + offset;
228 left = copyout(buf, from, copy);
229 copy -= left;
230 skip += copy;
231 from += copy;
232 bytes -= copy;
233 while (unlikely(!left && bytes)) {
234 iov++;
235 buf = iov->iov_base;
236 copy = min(bytes, iov->iov_len);
237 left = copyout(buf, from, copy);
238 copy -= left;
239 skip = copy;
240 from += copy;
241 bytes -= copy;
242 }
243 kunmap(page);
244
245done:
246 if (skip == iov->iov_len) {
247 iov++;
248 skip = 0;
249 }
250 i->count -= wanted - bytes;
251 i->nr_segs -= iov - i->iov;
252 i->iov = iov;
253 i->iov_offset = skip;
254 return wanted - bytes;
255}
256
257static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes,
258 struct iov_iter *i)
259{
260 size_t skip, copy, left, wanted;
261 const struct iovec *iov;
262 char __user *buf;
263 void *kaddr, *to;
264
265 if (unlikely(bytes > i->count))
266 bytes = i->count;
267
268 if (unlikely(!bytes))
269 return 0;
270
271 might_fault();
272 wanted = bytes;
273 iov = i->iov;
274 skip = i->iov_offset;
275 buf = iov->iov_base + skip;
276 copy = min(bytes, iov->iov_len - skip);
277
278 if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_readable(buf, copy)) {
279 kaddr = kmap_atomic(page);
280 to = kaddr + offset;
281
282 /* first chunk, usually the only one */
283 left = copyin(to, buf, copy);
284 copy -= left;
285 skip += copy;
286 to += copy;
287 bytes -= copy;
288
289 while (unlikely(!left && bytes)) {
290 iov++;
291 buf = iov->iov_base;
292 copy = min(bytes, iov->iov_len);
293 left = copyin(to, buf, copy);
294 copy -= left;
295 skip = copy;
296 to += copy;
297 bytes -= copy;
298 }
299 if (likely(!bytes)) {
300 kunmap_atomic(kaddr);
301 goto done;
302 }
303 offset = to - kaddr;
304 buf += copy;
305 kunmap_atomic(kaddr);
306 copy = min(bytes, iov->iov_len - skip);
307 }
308 /* Too bad - revert to non-atomic kmap */
309
310 kaddr = kmap(page);
311 to = kaddr + offset;
312 left = copyin(to, buf, copy);
313 copy -= left;
314 skip += copy;
315 to += copy;
316 bytes -= copy;
317 while (unlikely(!left && bytes)) {
318 iov++;
319 buf = iov->iov_base;
320 copy = min(bytes, iov->iov_len);
321 left = copyin(to, buf, copy);
322 copy -= left;
323 skip = copy;
324 to += copy;
325 bytes -= copy;
326 }
327 kunmap(page);
328
329done:
330 if (skip == iov->iov_len) {
331 iov++;
332 skip = 0;
333 }
334 i->count -= wanted - bytes;
335 i->nr_segs -= iov - i->iov;
336 i->iov = iov;
337 i->iov_offset = skip;
338 return wanted - bytes;
339}
340
341#ifdef PIPE_PARANOIA
342static bool sanity(const struct iov_iter *i)
343{
344 struct pipe_inode_info *pipe = i->pipe;
345 unsigned int p_head = pipe->head;
346 unsigned int p_tail = pipe->tail;
347 unsigned int p_mask = pipe->ring_size - 1;
348 unsigned int p_occupancy = pipe_occupancy(p_head, p_tail);
349 unsigned int i_head = i->head;
350 unsigned int idx;
351
352 if (i->iov_offset) {
353 struct pipe_buffer *p;
354 if (unlikely(p_occupancy == 0))
355 goto Bad; // pipe must be non-empty
356 if (unlikely(i_head != p_head - 1))
357 goto Bad; // must be at the last buffer...
358
359 p = &pipe->bufs[i_head & p_mask];
360 if (unlikely(p->offset + p->len != i->iov_offset))
361 goto Bad; // ... at the end of segment
362 } else {
363 if (i_head != p_head)
364 goto Bad; // must be right after the last buffer
365 }
366 return true;
367Bad:
368 printk(KERN_ERR "idx = %d, offset = %zd\n", i_head, i->iov_offset);
369 printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n",
370 p_head, p_tail, pipe->ring_size);
371 for (idx = 0; idx < pipe->ring_size; idx++)
372 printk(KERN_ERR "[%p %p %d %d]\n",
373 pipe->bufs[idx].ops,
374 pipe->bufs[idx].page,
375 pipe->bufs[idx].offset,
376 pipe->bufs[idx].len);
377 WARN_ON(1);
378 return false;
379}
380#else
381#define sanity(i) true
382#endif
383
384static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
385 struct iov_iter *i)
386{
387 struct pipe_inode_info *pipe = i->pipe;
388 struct pipe_buffer *buf;
389 unsigned int p_tail = pipe->tail;
390 unsigned int p_mask = pipe->ring_size - 1;
391 unsigned int i_head = i->head;
392 size_t off;
393
394 if (unlikely(bytes > i->count))
395 bytes = i->count;
396
397 if (unlikely(!bytes))
398 return 0;
399
400 if (!sanity(i))
401 return 0;
402
403 off = i->iov_offset;
404 buf = &pipe->bufs[i_head & p_mask];
405 if (off) {
406 if (offset == off && buf->page == page) {
407 /* merge with the last one */
408 buf->len += bytes;
409 i->iov_offset += bytes;
410 goto out;
411 }
412 i_head++;
413 buf = &pipe->bufs[i_head & p_mask];
414 }
415 if (pipe_full(i_head, p_tail, pipe->max_usage))
416 return 0;
417
418 buf->ops = &page_cache_pipe_buf_ops;
419 get_page(page);
420 buf->page = page;
421 buf->offset = offset;
422 buf->len = bytes;
423
424 pipe->head = i_head + 1;
425 i->iov_offset = offset + bytes;
426 i->head = i_head;
427out:
428 i->count -= bytes;
429 return bytes;
430}
431
432/*
433 * Fault in one or more iovecs of the given iov_iter, to a maximum length of
434 * bytes. For each iovec, fault in each page that constitutes the iovec.
435 *
436 * Return 0 on success, or non-zero if the memory could not be accessed (i.e.
437 * because it is an invalid address).
438 */
439int iov_iter_fault_in_readable(const struct iov_iter *i, size_t bytes)
440{
441 if (iter_is_iovec(i)) {
442 const struct iovec *p;
443 size_t skip;
444
445 if (bytes > i->count)
446 bytes = i->count;
447 for (p = i->iov, skip = i->iov_offset; bytes; p++, skip = 0) {
448 size_t len = min(bytes, p->iov_len - skip);
449 int err;
450
451 if (unlikely(!len))
452 continue;
453 err = fault_in_pages_readable(p->iov_base + skip, len);
454 if (unlikely(err))
455 return err;
456 bytes -= len;
457 }
458 }
459 return 0;
460}
461EXPORT_SYMBOL(iov_iter_fault_in_readable);
462
463void iov_iter_init(struct iov_iter *i, unsigned int direction,
464 const struct iovec *iov, unsigned long nr_segs,
465 size_t count)
466{
467 WARN_ON(direction & ~(READ | WRITE));
468 *i = (struct iov_iter) {
469 .iter_type = ITER_IOVEC,
470 .data_source = direction,
471 .iov = iov,
472 .nr_segs = nr_segs,
473 .iov_offset = 0,
474 .count = count
475 };
476}
477EXPORT_SYMBOL(iov_iter_init);
478
479static inline bool allocated(struct pipe_buffer *buf)
480{
481 return buf->ops == &default_pipe_buf_ops;
482}
483
484static inline void data_start(const struct iov_iter *i,
485 unsigned int *iter_headp, size_t *offp)
486{
487 unsigned int p_mask = i->pipe->ring_size - 1;
488 unsigned int iter_head = i->head;
489 size_t off = i->iov_offset;
490
491 if (off && (!allocated(&i->pipe->bufs[iter_head & p_mask]) ||
492 off == PAGE_SIZE)) {
493 iter_head++;
494 off = 0;
495 }
496 *iter_headp = iter_head;
497 *offp = off;
498}
499
500static size_t push_pipe(struct iov_iter *i, size_t size,
501 int *iter_headp, size_t *offp)
502{
503 struct pipe_inode_info *pipe = i->pipe;
504 unsigned int p_tail = pipe->tail;
505 unsigned int p_mask = pipe->ring_size - 1;
506 unsigned int iter_head;
507 size_t off;
508 ssize_t left;
509
510 if (unlikely(size > i->count))
511 size = i->count;
512 if (unlikely(!size))
513 return 0;
514
515 left = size;
516 data_start(i, &iter_head, &off);
517 *iter_headp = iter_head;
518 *offp = off;
519 if (off) {
520 left -= PAGE_SIZE - off;
521 if (left <= 0) {
522 pipe->bufs[iter_head & p_mask].len += size;
523 return size;
524 }
525 pipe->bufs[iter_head & p_mask].len = PAGE_SIZE;
526 iter_head++;
527 }
528 while (!pipe_full(iter_head, p_tail, pipe->max_usage)) {
529 struct pipe_buffer *buf = &pipe->bufs[iter_head & p_mask];
530 struct page *page = alloc_page(GFP_USER);
531 if (!page)
532 break;
533
534 buf->ops = &default_pipe_buf_ops;
535 buf->page = page;
536 buf->offset = 0;
537 buf->len = min_t(ssize_t, left, PAGE_SIZE);
538 left -= buf->len;
539 iter_head++;
540 pipe->head = iter_head;
541
542 if (left == 0)
543 return size;
544 }
545 return size - left;
546}
547
548static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
549 struct iov_iter *i)
550{
551 struct pipe_inode_info *pipe = i->pipe;
552 unsigned int p_mask = pipe->ring_size - 1;
553 unsigned int i_head;
554 size_t n, off;
555
556 if (!sanity(i))
557 return 0;
558
559 bytes = n = push_pipe(i, bytes, &i_head, &off);
560 if (unlikely(!n))
561 return 0;
562 do {
563 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
564 memcpy_to_page(pipe->bufs[i_head & p_mask].page, off, addr, chunk);
565 i->head = i_head;
566 i->iov_offset = off + chunk;
567 n -= chunk;
568 addr += chunk;
569 off = 0;
570 i_head++;
571 } while (n);
572 i->count -= bytes;
573 return bytes;
574}
575
576static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
577 __wsum sum, size_t off)
578{
579 __wsum next = csum_partial_copy_nocheck(from, to, len);
580 return csum_block_add(sum, next, off);
581}
582
583static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
584 struct iov_iter *i, __wsum *sump)
585{
586 struct pipe_inode_info *pipe = i->pipe;
587 unsigned int p_mask = pipe->ring_size - 1;
588 __wsum sum = *sump;
589 size_t off = 0;
590 unsigned int i_head;
591 size_t r;
592
593 if (!sanity(i))
594 return 0;
595
596 bytes = push_pipe(i, bytes, &i_head, &r);
597 while (bytes) {
598 size_t chunk = min_t(size_t, bytes, PAGE_SIZE - r);
599 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
600 sum = csum_and_memcpy(p + r, addr + off, chunk, sum, off);
601 kunmap_local(p);
602 i->head = i_head;
603 i->iov_offset = r + chunk;
604 bytes -= chunk;
605 off += chunk;
606 r = 0;
607 i_head++;
608 }
609 *sump = sum;
610 i->count -= off;
611 return off;
612}
613
614size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
615{
616 if (unlikely(iov_iter_is_pipe(i)))
617 return copy_pipe_to_iter(addr, bytes, i);
618 if (iter_is_iovec(i))
619 might_fault();
620 iterate_and_advance(i, bytes, base, len, off,
621 copyout(base, addr + off, len),
622 memcpy(base, addr + off, len)
623 )
624
625 return bytes;
626}
627EXPORT_SYMBOL(_copy_to_iter);
628
629#ifdef CONFIG_ARCH_HAS_COPY_MC
630static int copyout_mc(void __user *to, const void *from, size_t n)
631{
632 if (access_ok(to, n)) {
633 instrument_copy_to_user(to, from, n);
634 n = copy_mc_to_user((__force void *) to, from, n);
635 }
636 return n;
637}
638
639static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes,
640 struct iov_iter *i)
641{
642 struct pipe_inode_info *pipe = i->pipe;
643 unsigned int p_mask = pipe->ring_size - 1;
644 unsigned int i_head;
645 size_t n, off, xfer = 0;
646
647 if (!sanity(i))
648 return 0;
649
650 n = push_pipe(i, bytes, &i_head, &off);
651 while (n) {
652 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
653 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
654 unsigned long rem;
655 rem = copy_mc_to_kernel(p + off, addr + xfer, chunk);
656 chunk -= rem;
657 kunmap_local(p);
658 i->head = i_head;
659 i->iov_offset = off + chunk;
660 xfer += chunk;
661 if (rem)
662 break;
663 n -= chunk;
664 off = 0;
665 i_head++;
666 }
667 i->count -= xfer;
668 return xfer;
669}
670
671/**
672 * _copy_mc_to_iter - copy to iter with source memory error exception handling
673 * @addr: source kernel address
674 * @bytes: total transfer length
675 * @iter: destination iterator
676 *
677 * The pmem driver deploys this for the dax operation
678 * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
679 * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
680 * successfully copied.
681 *
682 * The main differences between this and typical _copy_to_iter().
683 *
684 * * Typical tail/residue handling after a fault retries the copy
685 * byte-by-byte until the fault happens again. Re-triggering machine
686 * checks is potentially fatal so the implementation uses source
687 * alignment and poison alignment assumptions to avoid re-triggering
688 * hardware exceptions.
689 *
690 * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
691 * Compare to copy_to_iter() where only ITER_IOVEC attempts might return
692 * a short copy.
693 */
694size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
695{
696 if (unlikely(iov_iter_is_pipe(i)))
697 return copy_mc_pipe_to_iter(addr, bytes, i);
698 if (iter_is_iovec(i))
699 might_fault();
700 __iterate_and_advance(i, bytes, base, len, off,
701 copyout_mc(base, addr + off, len),
702 copy_mc_to_kernel(base, addr + off, len)
703 )
704
705 return bytes;
706}
707EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
708#endif /* CONFIG_ARCH_HAS_COPY_MC */
709
710size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
711{
712 if (unlikely(iov_iter_is_pipe(i))) {
713 WARN_ON(1);
714 return 0;
715 }
716 if (iter_is_iovec(i))
717 might_fault();
718 iterate_and_advance(i, bytes, base, len, off,
719 copyin(addr + off, base, len),
720 memcpy(addr + off, base, len)
721 )
722
723 return bytes;
724}
725EXPORT_SYMBOL(_copy_from_iter);
726
727size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
728{
729 if (unlikely(iov_iter_is_pipe(i))) {
730 WARN_ON(1);
731 return 0;
732 }
733 iterate_and_advance(i, bytes, base, len, off,
734 __copy_from_user_inatomic_nocache(addr + off, base, len),
735 memcpy(addr + off, base, len)
736 )
737
738 return bytes;
739}
740EXPORT_SYMBOL(_copy_from_iter_nocache);
741
742#ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
743/**
744 * _copy_from_iter_flushcache - write destination through cpu cache
745 * @addr: destination kernel address
746 * @bytes: total transfer length
747 * @iter: source iterator
748 *
749 * The pmem driver arranges for filesystem-dax to use this facility via
750 * dax_copy_from_iter() for ensuring that writes to persistent memory
751 * are flushed through the CPU cache. It is differentiated from
752 * _copy_from_iter_nocache() in that guarantees all data is flushed for
753 * all iterator types. The _copy_from_iter_nocache() only attempts to
754 * bypass the cache for the ITER_IOVEC case, and on some archs may use
755 * instructions that strand dirty-data in the cache.
756 */
757size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
758{
759 if (unlikely(iov_iter_is_pipe(i))) {
760 WARN_ON(1);
761 return 0;
762 }
763 iterate_and_advance(i, bytes, base, len, off,
764 __copy_from_user_flushcache(addr + off, base, len),
765 memcpy_flushcache(addr + off, base, len)
766 )
767
768 return bytes;
769}
770EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
771#endif
772
773static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
774{
775 struct page *head;
776 size_t v = n + offset;
777
778 /*
779 * The general case needs to access the page order in order
780 * to compute the page size.
781 * However, we mostly deal with order-0 pages and thus can
782 * avoid a possible cache line miss for requests that fit all
783 * page orders.
784 */
785 if (n <= v && v <= PAGE_SIZE)
786 return true;
787
788 head = compound_head(page);
789 v += (page - head) << PAGE_SHIFT;
790
791 if (likely(n <= v && v <= (page_size(head))))
792 return true;
793 WARN_ON(1);
794 return false;
795}
796
797static size_t __copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
798 struct iov_iter *i)
799{
800 if (likely(iter_is_iovec(i)))
801 return copy_page_to_iter_iovec(page, offset, bytes, i);
802 if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
803 void *kaddr = kmap_local_page(page);
804 size_t wanted = _copy_to_iter(kaddr + offset, bytes, i);
805 kunmap_local(kaddr);
806 return wanted;
807 }
808 if (iov_iter_is_pipe(i))
809 return copy_page_to_iter_pipe(page, offset, bytes, i);
810 if (unlikely(iov_iter_is_discard(i))) {
811 if (unlikely(i->count < bytes))
812 bytes = i->count;
813 i->count -= bytes;
814 return bytes;
815 }
816 WARN_ON(1);
817 return 0;
818}
819
820size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
821 struct iov_iter *i)
822{
823 size_t res = 0;
824 if (unlikely(!page_copy_sane(page, offset, bytes)))
825 return 0;
826 page += offset / PAGE_SIZE; // first subpage
827 offset %= PAGE_SIZE;
828 while (1) {
829 size_t n = __copy_page_to_iter(page, offset,
830 min(bytes, (size_t)PAGE_SIZE - offset), i);
831 res += n;
832 bytes -= n;
833 if (!bytes || !n)
834 break;
835 offset += n;
836 if (offset == PAGE_SIZE) {
837 page++;
838 offset = 0;
839 }
840 }
841 return res;
842}
843EXPORT_SYMBOL(copy_page_to_iter);
844
845size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
846 struct iov_iter *i)
847{
848 if (unlikely(!page_copy_sane(page, offset, bytes)))
849 return 0;
850 if (likely(iter_is_iovec(i)))
851 return copy_page_from_iter_iovec(page, offset, bytes, i);
852 if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
853 void *kaddr = kmap_local_page(page);
854 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
855 kunmap_local(kaddr);
856 return wanted;
857 }
858 WARN_ON(1);
859 return 0;
860}
861EXPORT_SYMBOL(copy_page_from_iter);
862
863static size_t pipe_zero(size_t bytes, struct iov_iter *i)
864{
865 struct pipe_inode_info *pipe = i->pipe;
866 unsigned int p_mask = pipe->ring_size - 1;
867 unsigned int i_head;
868 size_t n, off;
869
870 if (!sanity(i))
871 return 0;
872
873 bytes = n = push_pipe(i, bytes, &i_head, &off);
874 if (unlikely(!n))
875 return 0;
876
877 do {
878 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
879 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
880 memset(p + off, 0, chunk);
881 kunmap_local(p);
882 i->head = i_head;
883 i->iov_offset = off + chunk;
884 n -= chunk;
885 off = 0;
886 i_head++;
887 } while (n);
888 i->count -= bytes;
889 return bytes;
890}
891
892size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
893{
894 if (unlikely(iov_iter_is_pipe(i)))
895 return pipe_zero(bytes, i);
896 iterate_and_advance(i, bytes, base, len, count,
897 clear_user(base, len),
898 memset(base, 0, len)
899 )
900
901 return bytes;
902}
903EXPORT_SYMBOL(iov_iter_zero);
904
905size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes,
906 struct iov_iter *i)
907{
908 char *kaddr = kmap_atomic(page), *p = kaddr + offset;
909 if (unlikely(!page_copy_sane(page, offset, bytes))) {
910 kunmap_atomic(kaddr);
911 return 0;
912 }
913 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
914 kunmap_atomic(kaddr);
915 WARN_ON(1);
916 return 0;
917 }
918 iterate_and_advance(i, bytes, base, len, off,
919 copyin(p + off, base, len),
920 memcpy(p + off, base, len)
921 )
922 kunmap_atomic(kaddr);
923 return bytes;
924}
925EXPORT_SYMBOL(copy_page_from_iter_atomic);
926
927static inline void pipe_truncate(struct iov_iter *i)
928{
929 struct pipe_inode_info *pipe = i->pipe;
930 unsigned int p_tail = pipe->tail;
931 unsigned int p_head = pipe->head;
932 unsigned int p_mask = pipe->ring_size - 1;
933
934 if (!pipe_empty(p_head, p_tail)) {
935 struct pipe_buffer *buf;
936 unsigned int i_head = i->head;
937 size_t off = i->iov_offset;
938
939 if (off) {
940 buf = &pipe->bufs[i_head & p_mask];
941 buf->len = off - buf->offset;
942 i_head++;
943 }
944 while (p_head != i_head) {
945 p_head--;
946 pipe_buf_release(pipe, &pipe->bufs[p_head & p_mask]);
947 }
948
949 pipe->head = p_head;
950 }
951}
952
953static void pipe_advance(struct iov_iter *i, size_t size)
954{
955 struct pipe_inode_info *pipe = i->pipe;
956 if (size) {
957 struct pipe_buffer *buf;
958 unsigned int p_mask = pipe->ring_size - 1;
959 unsigned int i_head = i->head;
960 size_t off = i->iov_offset, left = size;
961
962 if (off) /* make it relative to the beginning of buffer */
963 left += off - pipe->bufs[i_head & p_mask].offset;
964 while (1) {
965 buf = &pipe->bufs[i_head & p_mask];
966 if (left <= buf->len)
967 break;
968 left -= buf->len;
969 i_head++;
970 }
971 i->head = i_head;
972 i->iov_offset = buf->offset + left;
973 }
974 i->count -= size;
975 /* ... and discard everything past that point */
976 pipe_truncate(i);
977}
978
979static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
980{
981 struct bvec_iter bi;
982
983 bi.bi_size = i->count;
984 bi.bi_bvec_done = i->iov_offset;
985 bi.bi_idx = 0;
986 bvec_iter_advance(i->bvec, &bi, size);
987
988 i->bvec += bi.bi_idx;
989 i->nr_segs -= bi.bi_idx;
990 i->count = bi.bi_size;
991 i->iov_offset = bi.bi_bvec_done;
992}
993
994static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
995{
996 const struct iovec *iov, *end;
997
998 if (!i->count)
999 return;
1000 i->count -= size;
1001
1002 size += i->iov_offset; // from beginning of current segment
1003 for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) {
1004 if (likely(size < iov->iov_len))
1005 break;
1006 size -= iov->iov_len;
1007 }
1008 i->iov_offset = size;
1009 i->nr_segs -= iov - i->iov;
1010 i->iov = iov;
1011}
1012
1013void iov_iter_advance(struct iov_iter *i, size_t size)
1014{
1015 if (unlikely(i->count < size))
1016 size = i->count;
1017 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
1018 /* iovec and kvec have identical layouts */
1019 iov_iter_iovec_advance(i, size);
1020 } else if (iov_iter_is_bvec(i)) {
1021 iov_iter_bvec_advance(i, size);
1022 } else if (iov_iter_is_pipe(i)) {
1023 pipe_advance(i, size);
1024 } else if (unlikely(iov_iter_is_xarray(i))) {
1025 i->iov_offset += size;
1026 i->count -= size;
1027 } else if (iov_iter_is_discard(i)) {
1028 i->count -= size;
1029 }
1030}
1031EXPORT_SYMBOL(iov_iter_advance);
1032
1033void iov_iter_revert(struct iov_iter *i, size_t unroll)
1034{
1035 if (!unroll)
1036 return;
1037 if (WARN_ON(unroll > MAX_RW_COUNT))
1038 return;
1039 i->count += unroll;
1040 if (unlikely(iov_iter_is_pipe(i))) {
1041 struct pipe_inode_info *pipe = i->pipe;
1042 unsigned int p_mask = pipe->ring_size - 1;
1043 unsigned int i_head = i->head;
1044 size_t off = i->iov_offset;
1045 while (1) {
1046 struct pipe_buffer *b = &pipe->bufs[i_head & p_mask];
1047 size_t n = off - b->offset;
1048 if (unroll < n) {
1049 off -= unroll;
1050 break;
1051 }
1052 unroll -= n;
1053 if (!unroll && i_head == i->start_head) {
1054 off = 0;
1055 break;
1056 }
1057 i_head--;
1058 b = &pipe->bufs[i_head & p_mask];
1059 off = b->offset + b->len;
1060 }
1061 i->iov_offset = off;
1062 i->head = i_head;
1063 pipe_truncate(i);
1064 return;
1065 }
1066 if (unlikely(iov_iter_is_discard(i)))
1067 return;
1068 if (unroll <= i->iov_offset) {
1069 i->iov_offset -= unroll;
1070 return;
1071 }
1072 unroll -= i->iov_offset;
1073 if (iov_iter_is_xarray(i)) {
1074 BUG(); /* We should never go beyond the start of the specified
1075 * range since we might then be straying into pages that
1076 * aren't pinned.
1077 */
1078 } else if (iov_iter_is_bvec(i)) {
1079 const struct bio_vec *bvec = i->bvec;
1080 while (1) {
1081 size_t n = (--bvec)->bv_len;
1082 i->nr_segs++;
1083 if (unroll <= n) {
1084 i->bvec = bvec;
1085 i->iov_offset = n - unroll;
1086 return;
1087 }
1088 unroll -= n;
1089 }
1090 } else { /* same logics for iovec and kvec */
1091 const struct iovec *iov = i->iov;
1092 while (1) {
1093 size_t n = (--iov)->iov_len;
1094 i->nr_segs++;
1095 if (unroll <= n) {
1096 i->iov = iov;
1097 i->iov_offset = n - unroll;
1098 return;
1099 }
1100 unroll -= n;
1101 }
1102 }
1103}
1104EXPORT_SYMBOL(iov_iter_revert);
1105
1106/*
1107 * Return the count of just the current iov_iter segment.
1108 */
1109size_t iov_iter_single_seg_count(const struct iov_iter *i)
1110{
1111 if (i->nr_segs > 1) {
1112 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1113 return min(i->count, i->iov->iov_len - i->iov_offset);
1114 if (iov_iter_is_bvec(i))
1115 return min(i->count, i->bvec->bv_len - i->iov_offset);
1116 }
1117 return i->count;
1118}
1119EXPORT_SYMBOL(iov_iter_single_seg_count);
1120
1121void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
1122 const struct kvec *kvec, unsigned long nr_segs,
1123 size_t count)
1124{
1125 WARN_ON(direction & ~(READ | WRITE));
1126 *i = (struct iov_iter){
1127 .iter_type = ITER_KVEC,
1128 .data_source = direction,
1129 .kvec = kvec,
1130 .nr_segs = nr_segs,
1131 .iov_offset = 0,
1132 .count = count
1133 };
1134}
1135EXPORT_SYMBOL(iov_iter_kvec);
1136
1137void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1138 const struct bio_vec *bvec, unsigned long nr_segs,
1139 size_t count)
1140{
1141 WARN_ON(direction & ~(READ | WRITE));
1142 *i = (struct iov_iter){
1143 .iter_type = ITER_BVEC,
1144 .data_source = direction,
1145 .bvec = bvec,
1146 .nr_segs = nr_segs,
1147 .iov_offset = 0,
1148 .count = count
1149 };
1150}
1151EXPORT_SYMBOL(iov_iter_bvec);
1152
1153void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1154 struct pipe_inode_info *pipe,
1155 size_t count)
1156{
1157 BUG_ON(direction != READ);
1158 WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1159 *i = (struct iov_iter){
1160 .iter_type = ITER_PIPE,
1161 .data_source = false,
1162 .pipe = pipe,
1163 .head = pipe->head,
1164 .start_head = pipe->head,
1165 .iov_offset = 0,
1166 .count = count
1167 };
1168}
1169EXPORT_SYMBOL(iov_iter_pipe);
1170
1171/**
1172 * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1173 * @i: The iterator to initialise.
1174 * @direction: The direction of the transfer.
1175 * @xarray: The xarray to access.
1176 * @start: The start file position.
1177 * @count: The size of the I/O buffer in bytes.
1178 *
1179 * Set up an I/O iterator to either draw data out of the pages attached to an
1180 * inode or to inject data into those pages. The pages *must* be prevented
1181 * from evaporation, either by taking a ref on them or locking them by the
1182 * caller.
1183 */
1184void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
1185 struct xarray *xarray, loff_t start, size_t count)
1186{
1187 BUG_ON(direction & ~1);
1188 *i = (struct iov_iter) {
1189 .iter_type = ITER_XARRAY,
1190 .data_source = direction,
1191 .xarray = xarray,
1192 .xarray_start = start,
1193 .count = count,
1194 .iov_offset = 0
1195 };
1196}
1197EXPORT_SYMBOL(iov_iter_xarray);
1198
1199/**
1200 * iov_iter_discard - Initialise an I/O iterator that discards data
1201 * @i: The iterator to initialise.
1202 * @direction: The direction of the transfer.
1203 * @count: The size of the I/O buffer in bytes.
1204 *
1205 * Set up an I/O iterator that just discards everything that's written to it.
1206 * It's only available as a READ iterator.
1207 */
1208void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1209{
1210 BUG_ON(direction != READ);
1211 *i = (struct iov_iter){
1212 .iter_type = ITER_DISCARD,
1213 .data_source = false,
1214 .count = count,
1215 .iov_offset = 0
1216 };
1217}
1218EXPORT_SYMBOL(iov_iter_discard);
1219
1220static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
1221{
1222 unsigned long res = 0;
1223 size_t size = i->count;
1224 size_t skip = i->iov_offset;
1225 unsigned k;
1226
1227 for (k = 0; k < i->nr_segs; k++, skip = 0) {
1228 size_t len = i->iov[k].iov_len - skip;
1229 if (len) {
1230 res |= (unsigned long)i->iov[k].iov_base + skip;
1231 if (len > size)
1232 len = size;
1233 res |= len;
1234 size -= len;
1235 if (!size)
1236 break;
1237 }
1238 }
1239 return res;
1240}
1241
1242static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
1243{
1244 unsigned res = 0;
1245 size_t size = i->count;
1246 unsigned skip = i->iov_offset;
1247 unsigned k;
1248
1249 for (k = 0; k < i->nr_segs; k++, skip = 0) {
1250 size_t len = i->bvec[k].bv_len - skip;
1251 res |= (unsigned long)i->bvec[k].bv_offset + skip;
1252 if (len > size)
1253 len = size;
1254 res |= len;
1255 size -= len;
1256 if (!size)
1257 break;
1258 }
1259 return res;
1260}
1261
1262unsigned long iov_iter_alignment(const struct iov_iter *i)
1263{
1264 /* iovec and kvec have identical layouts */
1265 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1266 return iov_iter_alignment_iovec(i);
1267
1268 if (iov_iter_is_bvec(i))
1269 return iov_iter_alignment_bvec(i);
1270
1271 if (iov_iter_is_pipe(i)) {
1272 unsigned int p_mask = i->pipe->ring_size - 1;
1273 size_t size = i->count;
1274
1275 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->head & p_mask]))
1276 return size | i->iov_offset;
1277 return size;
1278 }
1279
1280 if (iov_iter_is_xarray(i))
1281 return (i->xarray_start + i->iov_offset) | i->count;
1282
1283 return 0;
1284}
1285EXPORT_SYMBOL(iov_iter_alignment);
1286
1287unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1288{
1289 unsigned long res = 0;
1290 unsigned long v = 0;
1291 size_t size = i->count;
1292 unsigned k;
1293
1294 if (WARN_ON(!iter_is_iovec(i)))
1295 return ~0U;
1296
1297 for (k = 0; k < i->nr_segs; k++) {
1298 if (i->iov[k].iov_len) {
1299 unsigned long base = (unsigned long)i->iov[k].iov_base;
1300 if (v) // if not the first one
1301 res |= base | v; // this start | previous end
1302 v = base + i->iov[k].iov_len;
1303 if (size <= i->iov[k].iov_len)
1304 break;
1305 size -= i->iov[k].iov_len;
1306 }
1307 }
1308 return res;
1309}
1310EXPORT_SYMBOL(iov_iter_gap_alignment);
1311
1312static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1313 size_t maxsize,
1314 struct page **pages,
1315 int iter_head,
1316 size_t *start)
1317{
1318 struct pipe_inode_info *pipe = i->pipe;
1319 unsigned int p_mask = pipe->ring_size - 1;
1320 ssize_t n = push_pipe(i, maxsize, &iter_head, start);
1321 if (!n)
1322 return -EFAULT;
1323
1324 maxsize = n;
1325 n += *start;
1326 while (n > 0) {
1327 get_page(*pages++ = pipe->bufs[iter_head & p_mask].page);
1328 iter_head++;
1329 n -= PAGE_SIZE;
1330 }
1331
1332 return maxsize;
1333}
1334
1335static ssize_t pipe_get_pages(struct iov_iter *i,
1336 struct page **pages, size_t maxsize, unsigned maxpages,
1337 size_t *start)
1338{
1339 unsigned int iter_head, npages;
1340 size_t capacity;
1341
1342 if (!sanity(i))
1343 return -EFAULT;
1344
1345 data_start(i, &iter_head, start);
1346 /* Amount of free space: some of this one + all after this one */
1347 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1348 capacity = min(npages, maxpages) * PAGE_SIZE - *start;
1349
1350 return __pipe_get_pages(i, min(maxsize, capacity), pages, iter_head, start);
1351}
1352
1353static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1354 pgoff_t index, unsigned int nr_pages)
1355{
1356 XA_STATE(xas, xa, index);
1357 struct page *page;
1358 unsigned int ret = 0;
1359
1360 rcu_read_lock();
1361 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1362 if (xas_retry(&xas, page))
1363 continue;
1364
1365 /* Has the page moved or been split? */
1366 if (unlikely(page != xas_reload(&xas))) {
1367 xas_reset(&xas);
1368 continue;
1369 }
1370
1371 pages[ret] = find_subpage(page, xas.xa_index);
1372 get_page(pages[ret]);
1373 if (++ret == nr_pages)
1374 break;
1375 }
1376 rcu_read_unlock();
1377 return ret;
1378}
1379
1380static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1381 struct page **pages, size_t maxsize,
1382 unsigned maxpages, size_t *_start_offset)
1383{
1384 unsigned nr, offset;
1385 pgoff_t index, count;
1386 size_t size = maxsize, actual;
1387 loff_t pos;
1388
1389 if (!size || !maxpages)
1390 return 0;
1391
1392 pos = i->xarray_start + i->iov_offset;
1393 index = pos >> PAGE_SHIFT;
1394 offset = pos & ~PAGE_MASK;
1395 *_start_offset = offset;
1396
1397 count = 1;
1398 if (size > PAGE_SIZE - offset) {
1399 size -= PAGE_SIZE - offset;
1400 count += size >> PAGE_SHIFT;
1401 size &= ~PAGE_MASK;
1402 if (size)
1403 count++;
1404 }
1405
1406 if (count > maxpages)
1407 count = maxpages;
1408
1409 nr = iter_xarray_populate_pages(pages, i->xarray, index, count);
1410 if (nr == 0)
1411 return 0;
1412
1413 actual = PAGE_SIZE * nr;
1414 actual -= offset;
1415 if (nr == count && size > 0) {
1416 unsigned last_offset = (nr > 1) ? 0 : offset;
1417 actual -= PAGE_SIZE - (last_offset + size);
1418 }
1419 return actual;
1420}
1421
1422/* must be done on non-empty ITER_IOVEC one */
1423static unsigned long first_iovec_segment(const struct iov_iter *i,
1424 size_t *size, size_t *start,
1425 size_t maxsize, unsigned maxpages)
1426{
1427 size_t skip;
1428 long k;
1429
1430 for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
1431 unsigned long addr = (unsigned long)i->iov[k].iov_base + skip;
1432 size_t len = i->iov[k].iov_len - skip;
1433
1434 if (unlikely(!len))
1435 continue;
1436 if (len > maxsize)
1437 len = maxsize;
1438 len += (*start = addr % PAGE_SIZE);
1439 if (len > maxpages * PAGE_SIZE)
1440 len = maxpages * PAGE_SIZE;
1441 *size = len;
1442 return addr & PAGE_MASK;
1443 }
1444 BUG(); // if it had been empty, we wouldn't get called
1445}
1446
1447/* must be done on non-empty ITER_BVEC one */
1448static struct page *first_bvec_segment(const struct iov_iter *i,
1449 size_t *size, size_t *start,
1450 size_t maxsize, unsigned maxpages)
1451{
1452 struct page *page;
1453 size_t skip = i->iov_offset, len;
1454
1455 len = i->bvec->bv_len - skip;
1456 if (len > maxsize)
1457 len = maxsize;
1458 skip += i->bvec->bv_offset;
1459 page = i->bvec->bv_page + skip / PAGE_SIZE;
1460 len += (*start = skip % PAGE_SIZE);
1461 if (len > maxpages * PAGE_SIZE)
1462 len = maxpages * PAGE_SIZE;
1463 *size = len;
1464 return page;
1465}
1466
1467ssize_t iov_iter_get_pages(struct iov_iter *i,
1468 struct page **pages, size_t maxsize, unsigned maxpages,
1469 size_t *start)
1470{
1471 size_t len;
1472 int n, res;
1473
1474 if (maxsize > i->count)
1475 maxsize = i->count;
1476 if (!maxsize)
1477 return 0;
1478
1479 if (likely(iter_is_iovec(i))) {
1480 unsigned long addr;
1481
1482 addr = first_iovec_segment(i, &len, start, maxsize, maxpages);
1483 n = DIV_ROUND_UP(len, PAGE_SIZE);
1484 res = get_user_pages_fast(addr, n,
1485 iov_iter_rw(i) != WRITE ? FOLL_WRITE : 0,
1486 pages);
1487 if (unlikely(res < 0))
1488 return res;
1489 return (res == n ? len : res * PAGE_SIZE) - *start;
1490 }
1491 if (iov_iter_is_bvec(i)) {
1492 struct page *page;
1493
1494 page = first_bvec_segment(i, &len, start, maxsize, maxpages);
1495 n = DIV_ROUND_UP(len, PAGE_SIZE);
1496 while (n--)
1497 get_page(*pages++ = page++);
1498 return len - *start;
1499 }
1500 if (iov_iter_is_pipe(i))
1501 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1502 if (iov_iter_is_xarray(i))
1503 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1504 return -EFAULT;
1505}
1506EXPORT_SYMBOL(iov_iter_get_pages);
1507
1508static struct page **get_pages_array(size_t n)
1509{
1510 return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1511}
1512
1513static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1514 struct page ***pages, size_t maxsize,
1515 size_t *start)
1516{
1517 struct page **p;
1518 unsigned int iter_head, npages;
1519 ssize_t n;
1520
1521 if (!sanity(i))
1522 return -EFAULT;
1523
1524 data_start(i, &iter_head, start);
1525 /* Amount of free space: some of this one + all after this one */
1526 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1527 n = npages * PAGE_SIZE - *start;
1528 if (maxsize > n)
1529 maxsize = n;
1530 else
1531 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1532 p = get_pages_array(npages);
1533 if (!p)
1534 return -ENOMEM;
1535 n = __pipe_get_pages(i, maxsize, p, iter_head, start);
1536 if (n > 0)
1537 *pages = p;
1538 else
1539 kvfree(p);
1540 return n;
1541}
1542
1543static ssize_t iter_xarray_get_pages_alloc(struct iov_iter *i,
1544 struct page ***pages, size_t maxsize,
1545 size_t *_start_offset)
1546{
1547 struct page **p;
1548 unsigned nr, offset;
1549 pgoff_t index, count;
1550 size_t size = maxsize, actual;
1551 loff_t pos;
1552
1553 if (!size)
1554 return 0;
1555
1556 pos = i->xarray_start + i->iov_offset;
1557 index = pos >> PAGE_SHIFT;
1558 offset = pos & ~PAGE_MASK;
1559 *_start_offset = offset;
1560
1561 count = 1;
1562 if (size > PAGE_SIZE - offset) {
1563 size -= PAGE_SIZE - offset;
1564 count += size >> PAGE_SHIFT;
1565 size &= ~PAGE_MASK;
1566 if (size)
1567 count++;
1568 }
1569
1570 p = get_pages_array(count);
1571 if (!p)
1572 return -ENOMEM;
1573 *pages = p;
1574
1575 nr = iter_xarray_populate_pages(p, i->xarray, index, count);
1576 if (nr == 0)
1577 return 0;
1578
1579 actual = PAGE_SIZE * nr;
1580 actual -= offset;
1581 if (nr == count && size > 0) {
1582 unsigned last_offset = (nr > 1) ? 0 : offset;
1583 actual -= PAGE_SIZE - (last_offset + size);
1584 }
1585 return actual;
1586}
1587
1588ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1589 struct page ***pages, size_t maxsize,
1590 size_t *start)
1591{
1592 struct page **p;
1593 size_t len;
1594 int n, res;
1595
1596 if (maxsize > i->count)
1597 maxsize = i->count;
1598 if (!maxsize)
1599 return 0;
1600
1601 if (likely(iter_is_iovec(i))) {
1602 unsigned long addr;
1603
1604 addr = first_iovec_segment(i, &len, start, maxsize, ~0U);
1605 n = DIV_ROUND_UP(len, PAGE_SIZE);
1606 p = get_pages_array(n);
1607 if (!p)
1608 return -ENOMEM;
1609 res = get_user_pages_fast(addr, n,
1610 iov_iter_rw(i) != WRITE ? FOLL_WRITE : 0, p);
1611 if (unlikely(res < 0)) {
1612 kvfree(p);
1613 return res;
1614 }
1615 *pages = p;
1616 return (res == n ? len : res * PAGE_SIZE) - *start;
1617 }
1618 if (iov_iter_is_bvec(i)) {
1619 struct page *page;
1620
1621 page = first_bvec_segment(i, &len, start, maxsize, ~0U);
1622 n = DIV_ROUND_UP(len, PAGE_SIZE);
1623 *pages = p = get_pages_array(n);
1624 if (!p)
1625 return -ENOMEM;
1626 while (n--)
1627 get_page(*p++ = page++);
1628 return len - *start;
1629 }
1630 if (iov_iter_is_pipe(i))
1631 return pipe_get_pages_alloc(i, pages, maxsize, start);
1632 if (iov_iter_is_xarray(i))
1633 return iter_xarray_get_pages_alloc(i, pages, maxsize, start);
1634 return -EFAULT;
1635}
1636EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1637
1638size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1639 struct iov_iter *i)
1640{
1641 __wsum sum, next;
1642 sum = *csum;
1643 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1644 WARN_ON(1);
1645 return 0;
1646 }
1647 iterate_and_advance(i, bytes, base, len, off, ({
1648 next = csum_and_copy_from_user(base, addr + off, len);
1649 sum = csum_block_add(sum, next, off);
1650 next ? 0 : len;
1651 }), ({
1652 sum = csum_and_memcpy(addr + off, base, len, sum, off);
1653 })
1654 )
1655 *csum = sum;
1656 return bytes;
1657}
1658EXPORT_SYMBOL(csum_and_copy_from_iter);
1659
1660size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1661 struct iov_iter *i)
1662{
1663 struct csum_state *csstate = _csstate;
1664 __wsum sum, next;
1665
1666 if (unlikely(iov_iter_is_discard(i))) {
1667 WARN_ON(1); /* for now */
1668 return 0;
1669 }
1670
1671 sum = csum_shift(csstate->csum, csstate->off);
1672 if (unlikely(iov_iter_is_pipe(i)))
1673 bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum);
1674 else iterate_and_advance(i, bytes, base, len, off, ({
1675 next = csum_and_copy_to_user(addr + off, base, len);
1676 sum = csum_block_add(sum, next, off);
1677 next ? 0 : len;
1678 }), ({
1679 sum = csum_and_memcpy(base, addr + off, len, sum, off);
1680 })
1681 )
1682 csstate->csum = csum_shift(sum, csstate->off);
1683 csstate->off += bytes;
1684 return bytes;
1685}
1686EXPORT_SYMBOL(csum_and_copy_to_iter);
1687
1688size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1689 struct iov_iter *i)
1690{
1691#ifdef CONFIG_CRYPTO_HASH
1692 struct ahash_request *hash = hashp;
1693 struct scatterlist sg;
1694 size_t copied;
1695
1696 copied = copy_to_iter(addr, bytes, i);
1697 sg_init_one(&sg, addr, copied);
1698 ahash_request_set_crypt(hash, &sg, NULL, copied);
1699 crypto_ahash_update(hash);
1700 return copied;
1701#else
1702 return 0;
1703#endif
1704}
1705EXPORT_SYMBOL(hash_and_copy_to_iter);
1706
1707static int iov_npages(const struct iov_iter *i, int maxpages)
1708{
1709 size_t skip = i->iov_offset, size = i->count;
1710 const struct iovec *p;
1711 int npages = 0;
1712
1713 for (p = i->iov; size; skip = 0, p++) {
1714 unsigned offs = offset_in_page(p->iov_base + skip);
1715 size_t len = min(p->iov_len - skip, size);
1716
1717 if (len) {
1718 size -= len;
1719 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1720 if (unlikely(npages > maxpages))
1721 return maxpages;
1722 }
1723 }
1724 return npages;
1725}
1726
1727static int bvec_npages(const struct iov_iter *i, int maxpages)
1728{
1729 size_t skip = i->iov_offset, size = i->count;
1730 const struct bio_vec *p;
1731 int npages = 0;
1732
1733 for (p = i->bvec; size; skip = 0, p++) {
1734 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1735 size_t len = min(p->bv_len - skip, size);
1736
1737 size -= len;
1738 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1739 if (unlikely(npages > maxpages))
1740 return maxpages;
1741 }
1742 return npages;
1743}
1744
1745int iov_iter_npages(const struct iov_iter *i, int maxpages)
1746{
1747 if (unlikely(!i->count))
1748 return 0;
1749 /* iovec and kvec have identical layouts */
1750 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1751 return iov_npages(i, maxpages);
1752 if (iov_iter_is_bvec(i))
1753 return bvec_npages(i, maxpages);
1754 if (iov_iter_is_pipe(i)) {
1755 unsigned int iter_head;
1756 int npages;
1757 size_t off;
1758
1759 if (!sanity(i))
1760 return 0;
1761
1762 data_start(i, &iter_head, &off);
1763 /* some of this one + all after this one */
1764 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1765 return min(npages, maxpages);
1766 }
1767 if (iov_iter_is_xarray(i)) {
1768 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1769 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1770 return min(npages, maxpages);
1771 }
1772 return 0;
1773}
1774EXPORT_SYMBOL(iov_iter_npages);
1775
1776const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1777{
1778 *new = *old;
1779 if (unlikely(iov_iter_is_pipe(new))) {
1780 WARN_ON(1);
1781 return NULL;
1782 }
1783 if (unlikely(iov_iter_is_discard(new) || iov_iter_is_xarray(new)))
1784 return NULL;
1785 if (iov_iter_is_bvec(new))
1786 return new->bvec = kmemdup(new->bvec,
1787 new->nr_segs * sizeof(struct bio_vec),
1788 flags);
1789 else
1790 /* iovec and kvec have identical layout */
1791 return new->iov = kmemdup(new->iov,
1792 new->nr_segs * sizeof(struct iovec),
1793 flags);
1794}
1795EXPORT_SYMBOL(dup_iter);
1796
1797static int copy_compat_iovec_from_user(struct iovec *iov,
1798 const struct iovec __user *uvec, unsigned long nr_segs)
1799{
1800 const struct compat_iovec __user *uiov =
1801 (const struct compat_iovec __user *)uvec;
1802 int ret = -EFAULT, i;
1803
1804 if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1805 return -EFAULT;
1806
1807 for (i = 0; i < nr_segs; i++) {
1808 compat_uptr_t buf;
1809 compat_ssize_t len;
1810
1811 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1812 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1813
1814 /* check for compat_size_t not fitting in compat_ssize_t .. */
1815 if (len < 0) {
1816 ret = -EINVAL;
1817 goto uaccess_end;
1818 }
1819 iov[i].iov_base = compat_ptr(buf);
1820 iov[i].iov_len = len;
1821 }
1822
1823 ret = 0;
1824uaccess_end:
1825 user_access_end();
1826 return ret;
1827}
1828
1829static int copy_iovec_from_user(struct iovec *iov,
1830 const struct iovec __user *uvec, unsigned long nr_segs)
1831{
1832 unsigned long seg;
1833
1834 if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1835 return -EFAULT;
1836 for (seg = 0; seg < nr_segs; seg++) {
1837 if ((ssize_t)iov[seg].iov_len < 0)
1838 return -EINVAL;
1839 }
1840
1841 return 0;
1842}
1843
1844struct iovec *iovec_from_user(const struct iovec __user *uvec,
1845 unsigned long nr_segs, unsigned long fast_segs,
1846 struct iovec *fast_iov, bool compat)
1847{
1848 struct iovec *iov = fast_iov;
1849 int ret;
1850
1851 /*
1852 * SuS says "The readv() function *may* fail if the iovcnt argument was
1853 * less than or equal to 0, or greater than {IOV_MAX}. Linux has
1854 * traditionally returned zero for zero segments, so...
1855 */
1856 if (nr_segs == 0)
1857 return iov;
1858 if (nr_segs > UIO_MAXIOV)
1859 return ERR_PTR(-EINVAL);
1860 if (nr_segs > fast_segs) {
1861 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1862 if (!iov)
1863 return ERR_PTR(-ENOMEM);
1864 }
1865
1866 if (compat)
1867 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1868 else
1869 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1870 if (ret) {
1871 if (iov != fast_iov)
1872 kfree(iov);
1873 return ERR_PTR(ret);
1874 }
1875
1876 return iov;
1877}
1878
1879ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1880 unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1881 struct iov_iter *i, bool compat)
1882{
1883 ssize_t total_len = 0;
1884 unsigned long seg;
1885 struct iovec *iov;
1886
1887 iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1888 if (IS_ERR(iov)) {
1889 *iovp = NULL;
1890 return PTR_ERR(iov);
1891 }
1892
1893 /*
1894 * According to the Single Unix Specification we should return EINVAL if
1895 * an element length is < 0 when cast to ssize_t or if the total length
1896 * would overflow the ssize_t return value of the system call.
1897 *
1898 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1899 * overflow case.
1900 */
1901 for (seg = 0; seg < nr_segs; seg++) {
1902 ssize_t len = (ssize_t)iov[seg].iov_len;
1903
1904 if (!access_ok(iov[seg].iov_base, len)) {
1905 if (iov != *iovp)
1906 kfree(iov);
1907 *iovp = NULL;
1908 return -EFAULT;
1909 }
1910
1911 if (len > MAX_RW_COUNT - total_len) {
1912 len = MAX_RW_COUNT - total_len;
1913 iov[seg].iov_len = len;
1914 }
1915 total_len += len;
1916 }
1917
1918 iov_iter_init(i, type, iov, nr_segs, total_len);
1919 if (iov == *iovp)
1920 *iovp = NULL;
1921 else
1922 *iovp = iov;
1923 return total_len;
1924}
1925
1926/**
1927 * import_iovec() - Copy an array of &struct iovec from userspace
1928 * into the kernel, check that it is valid, and initialize a new
1929 * &struct iov_iter iterator to access it.
1930 *
1931 * @type: One of %READ or %WRITE.
1932 * @uvec: Pointer to the userspace array.
1933 * @nr_segs: Number of elements in userspace array.
1934 * @fast_segs: Number of elements in @iov.
1935 * @iovp: (input and output parameter) Pointer to pointer to (usually small
1936 * on-stack) kernel array.
1937 * @i: Pointer to iterator that will be initialized on success.
1938 *
1939 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1940 * then this function places %NULL in *@iov on return. Otherwise, a new
1941 * array will be allocated and the result placed in *@iov. This means that
1942 * the caller may call kfree() on *@iov regardless of whether the small
1943 * on-stack array was used or not (and regardless of whether this function
1944 * returns an error or not).
1945 *
1946 * Return: Negative error code on error, bytes imported on success
1947 */
1948ssize_t import_iovec(int type, const struct iovec __user *uvec,
1949 unsigned nr_segs, unsigned fast_segs,
1950 struct iovec **iovp, struct iov_iter *i)
1951{
1952 return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
1953 in_compat_syscall());
1954}
1955EXPORT_SYMBOL(import_iovec);
1956
1957int import_single_range(int rw, void __user *buf, size_t len,
1958 struct iovec *iov, struct iov_iter *i)
1959{
1960 if (len > MAX_RW_COUNT)
1961 len = MAX_RW_COUNT;
1962 if (unlikely(!access_ok(buf, len)))
1963 return -EFAULT;
1964
1965 iov->iov_base = buf;
1966 iov->iov_len = len;
1967 iov_iter_init(i, rw, iov, 1, len);
1968 return 0;
1969}
1970EXPORT_SYMBOL(import_single_range);
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);