]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - lib/iov_iter.c
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
vmxnet3: use correct intrConf reference when using extended queues
[mirror_ubuntu-jammy-kernel.git] / lib / iov_iter.c
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
151 static 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
162 static 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
173 static 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_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
245 done:
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
257 static 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_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
329 done:
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
342 static 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;
367 Bad:
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
384 static 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 buf->flags = 0;
420 get_page(page);
421 buf->page = page;
422 buf->offset = offset;
423 buf->len = bytes;
424
425 pipe->head = i_head + 1;
426 i->iov_offset = offset + bytes;
427 i->head = i_head;
428 out:
429 i->count -= bytes;
430 return bytes;
431 }
432
433 /*
434 * fault_in_iov_iter_readable - fault in iov iterator for reading
435 * @i: iterator
436 * @size: maximum length
437 *
438 * Fault in one or more iovecs of the given iov_iter, to a maximum length of
439 * @size. For each iovec, fault in each page that constitutes the iovec.
440 *
441 * Returns the number of bytes not faulted in (like copy_to_user() and
442 * copy_from_user()).
443 *
444 * Always returns 0 for non-userspace iterators.
445 */
446 size_t fault_in_iov_iter_readable(const struct iov_iter *i, size_t size)
447 {
448 if (iter_is_iovec(i)) {
449 size_t count = min(size, iov_iter_count(i));
450 const struct iovec *p;
451 size_t skip;
452
453 size -= count;
454 for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) {
455 size_t len = min(count, p->iov_len - skip);
456 size_t ret;
457
458 if (unlikely(!len))
459 continue;
460 ret = fault_in_readable(p->iov_base + skip, len);
461 count -= len - ret;
462 if (ret)
463 break;
464 }
465 return count + size;
466 }
467 return 0;
468 }
469 EXPORT_SYMBOL(fault_in_iov_iter_readable);
470
471 /*
472 * fault_in_iov_iter_writeable - fault in iov iterator for writing
473 * @i: iterator
474 * @size: maximum length
475 *
476 * Faults in the iterator using get_user_pages(), i.e., without triggering
477 * hardware page faults. This is primarily useful when we already know that
478 * some or all of the pages in @i aren't in memory.
479 *
480 * Returns the number of bytes not faulted in, like copy_to_user() and
481 * copy_from_user().
482 *
483 * Always returns 0 for non-user-space iterators.
484 */
485 size_t fault_in_iov_iter_writeable(const struct iov_iter *i, size_t size)
486 {
487 if (iter_is_iovec(i)) {
488 size_t count = min(size, iov_iter_count(i));
489 const struct iovec *p;
490 size_t skip;
491
492 size -= count;
493 for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) {
494 size_t len = min(count, p->iov_len - skip);
495 size_t ret;
496
497 if (unlikely(!len))
498 continue;
499 ret = fault_in_safe_writeable(p->iov_base + skip, len);
500 count -= len - ret;
501 if (ret)
502 break;
503 }
504 return count + size;
505 }
506 return 0;
507 }
508 EXPORT_SYMBOL(fault_in_iov_iter_writeable);
509
510 void iov_iter_init(struct iov_iter *i, unsigned int direction,
511 const struct iovec *iov, unsigned long nr_segs,
512 size_t count)
513 {
514 WARN_ON(direction & ~(READ | WRITE));
515 *i = (struct iov_iter) {
516 .iter_type = ITER_IOVEC,
517 .nofault = false,
518 .data_source = direction,
519 .iov = iov,
520 .nr_segs = nr_segs,
521 .iov_offset = 0,
522 .count = count
523 };
524 }
525 EXPORT_SYMBOL(iov_iter_init);
526
527 static inline bool allocated(struct pipe_buffer *buf)
528 {
529 return buf->ops == &default_pipe_buf_ops;
530 }
531
532 static inline void data_start(const struct iov_iter *i,
533 unsigned int *iter_headp, size_t *offp)
534 {
535 unsigned int p_mask = i->pipe->ring_size - 1;
536 unsigned int iter_head = i->head;
537 size_t off = i->iov_offset;
538
539 if (off && (!allocated(&i->pipe->bufs[iter_head & p_mask]) ||
540 off == PAGE_SIZE)) {
541 iter_head++;
542 off = 0;
543 }
544 *iter_headp = iter_head;
545 *offp = off;
546 }
547
548 static size_t push_pipe(struct iov_iter *i, size_t size,
549 int *iter_headp, size_t *offp)
550 {
551 struct pipe_inode_info *pipe = i->pipe;
552 unsigned int p_tail = pipe->tail;
553 unsigned int p_mask = pipe->ring_size - 1;
554 unsigned int iter_head;
555 size_t off;
556 ssize_t left;
557
558 if (unlikely(size > i->count))
559 size = i->count;
560 if (unlikely(!size))
561 return 0;
562
563 left = size;
564 data_start(i, &iter_head, &off);
565 *iter_headp = iter_head;
566 *offp = off;
567 if (off) {
568 left -= PAGE_SIZE - off;
569 if (left <= 0) {
570 pipe->bufs[iter_head & p_mask].len += size;
571 return size;
572 }
573 pipe->bufs[iter_head & p_mask].len = PAGE_SIZE;
574 iter_head++;
575 }
576 while (!pipe_full(iter_head, p_tail, pipe->max_usage)) {
577 struct pipe_buffer *buf = &pipe->bufs[iter_head & p_mask];
578 struct page *page = alloc_page(GFP_USER);
579 if (!page)
580 break;
581
582 buf->ops = &default_pipe_buf_ops;
583 buf->flags = 0;
584 buf->page = page;
585 buf->offset = 0;
586 buf->len = min_t(ssize_t, left, PAGE_SIZE);
587 left -= buf->len;
588 iter_head++;
589 pipe->head = iter_head;
590
591 if (left == 0)
592 return size;
593 }
594 return size - left;
595 }
596
597 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
598 struct iov_iter *i)
599 {
600 struct pipe_inode_info *pipe = i->pipe;
601 unsigned int p_mask = pipe->ring_size - 1;
602 unsigned int i_head;
603 size_t n, off;
604
605 if (!sanity(i))
606 return 0;
607
608 bytes = n = push_pipe(i, bytes, &i_head, &off);
609 if (unlikely(!n))
610 return 0;
611 do {
612 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
613 memcpy_to_page(pipe->bufs[i_head & p_mask].page, off, addr, chunk);
614 i->head = i_head;
615 i->iov_offset = off + chunk;
616 n -= chunk;
617 addr += chunk;
618 off = 0;
619 i_head++;
620 } while (n);
621 i->count -= bytes;
622 return bytes;
623 }
624
625 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
626 __wsum sum, size_t off)
627 {
628 __wsum next = csum_partial_copy_nocheck(from, to, len);
629 return csum_block_add(sum, next, off);
630 }
631
632 static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
633 struct iov_iter *i, __wsum *sump)
634 {
635 struct pipe_inode_info *pipe = i->pipe;
636 unsigned int p_mask = pipe->ring_size - 1;
637 __wsum sum = *sump;
638 size_t off = 0;
639 unsigned int i_head;
640 size_t r;
641
642 if (!sanity(i))
643 return 0;
644
645 bytes = push_pipe(i, bytes, &i_head, &r);
646 while (bytes) {
647 size_t chunk = min_t(size_t, bytes, PAGE_SIZE - r);
648 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
649 sum = csum_and_memcpy(p + r, addr + off, chunk, sum, off);
650 kunmap_local(p);
651 i->head = i_head;
652 i->iov_offset = r + chunk;
653 bytes -= chunk;
654 off += chunk;
655 r = 0;
656 i_head++;
657 }
658 *sump = sum;
659 i->count -= off;
660 return off;
661 }
662
663 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
664 {
665 if (unlikely(iov_iter_is_pipe(i)))
666 return copy_pipe_to_iter(addr, bytes, i);
667 if (iter_is_iovec(i))
668 might_fault();
669 iterate_and_advance(i, bytes, base, len, off,
670 copyout(base, addr + off, len),
671 memcpy(base, addr + off, len)
672 )
673
674 return bytes;
675 }
676 EXPORT_SYMBOL(_copy_to_iter);
677
678 #ifdef CONFIG_ARCH_HAS_COPY_MC
679 static int copyout_mc(void __user *to, const void *from, size_t n)
680 {
681 if (access_ok(to, n)) {
682 instrument_copy_to_user(to, from, n);
683 n = copy_mc_to_user((__force void *) to, from, n);
684 }
685 return n;
686 }
687
688 static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes,
689 struct iov_iter *i)
690 {
691 struct pipe_inode_info *pipe = i->pipe;
692 unsigned int p_mask = pipe->ring_size - 1;
693 unsigned int i_head;
694 unsigned int valid = pipe->head;
695 size_t n, off, xfer = 0;
696
697 if (!sanity(i))
698 return 0;
699
700 n = push_pipe(i, bytes, &i_head, &off);
701 while (n) {
702 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
703 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
704 unsigned long rem;
705 rem = copy_mc_to_kernel(p + off, addr + xfer, chunk);
706 chunk -= rem;
707 kunmap_local(p);
708 if (chunk) {
709 i->head = i_head;
710 i->iov_offset = off + chunk;
711 xfer += chunk;
712 valid = i_head + 1;
713 }
714 if (rem) {
715 pipe->bufs[i_head & p_mask].len -= rem;
716 pipe_discard_from(pipe, valid);
717 break;
718 }
719 n -= chunk;
720 off = 0;
721 i_head++;
722 }
723 i->count -= xfer;
724 return xfer;
725 }
726
727 /**
728 * _copy_mc_to_iter - copy to iter with source memory error exception handling
729 * @addr: source kernel address
730 * @bytes: total transfer length
731 * @i: destination iterator
732 *
733 * The pmem driver deploys this for the dax operation
734 * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
735 * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
736 * successfully copied.
737 *
738 * The main differences between this and typical _copy_to_iter().
739 *
740 * * Typical tail/residue handling after a fault retries the copy
741 * byte-by-byte until the fault happens again. Re-triggering machine
742 * checks is potentially fatal so the implementation uses source
743 * alignment and poison alignment assumptions to avoid re-triggering
744 * hardware exceptions.
745 *
746 * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
747 * Compare to copy_to_iter() where only ITER_IOVEC attempts might return
748 * a short copy.
749 *
750 * Return: number of bytes copied (may be %0)
751 */
752 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
753 {
754 if (unlikely(iov_iter_is_pipe(i)))
755 return copy_mc_pipe_to_iter(addr, bytes, i);
756 if (iter_is_iovec(i))
757 might_fault();
758 __iterate_and_advance(i, bytes, base, len, off,
759 copyout_mc(base, addr + off, len),
760 copy_mc_to_kernel(base, addr + off, len)
761 )
762
763 return bytes;
764 }
765 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
766 #endif /* CONFIG_ARCH_HAS_COPY_MC */
767
768 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
769 {
770 if (unlikely(iov_iter_is_pipe(i))) {
771 WARN_ON(1);
772 return 0;
773 }
774 if (iter_is_iovec(i))
775 might_fault();
776 iterate_and_advance(i, bytes, base, len, off,
777 copyin(addr + off, base, len),
778 memcpy(addr + off, base, len)
779 )
780
781 return bytes;
782 }
783 EXPORT_SYMBOL(_copy_from_iter);
784
785 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
786 {
787 if (unlikely(iov_iter_is_pipe(i))) {
788 WARN_ON(1);
789 return 0;
790 }
791 iterate_and_advance(i, bytes, base, len, off,
792 __copy_from_user_inatomic_nocache(addr + off, base, len),
793 memcpy(addr + off, base, len)
794 )
795
796 return bytes;
797 }
798 EXPORT_SYMBOL(_copy_from_iter_nocache);
799
800 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
801 /**
802 * _copy_from_iter_flushcache - write destination through cpu cache
803 * @addr: destination kernel address
804 * @bytes: total transfer length
805 * @i: source iterator
806 *
807 * The pmem driver arranges for filesystem-dax to use this facility via
808 * dax_copy_from_iter() for ensuring that writes to persistent memory
809 * are flushed through the CPU cache. It is differentiated from
810 * _copy_from_iter_nocache() in that guarantees all data is flushed for
811 * all iterator types. The _copy_from_iter_nocache() only attempts to
812 * bypass the cache for the ITER_IOVEC case, and on some archs may use
813 * instructions that strand dirty-data in the cache.
814 *
815 * Return: number of bytes copied (may be %0)
816 */
817 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
818 {
819 if (unlikely(iov_iter_is_pipe(i))) {
820 WARN_ON(1);
821 return 0;
822 }
823 iterate_and_advance(i, bytes, base, len, off,
824 __copy_from_user_flushcache(addr + off, base, len),
825 memcpy_flushcache(addr + off, base, len)
826 )
827
828 return bytes;
829 }
830 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
831 #endif
832
833 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
834 {
835 struct page *head;
836 size_t v = n + offset;
837
838 /*
839 * The general case needs to access the page order in order
840 * to compute the page size.
841 * However, we mostly deal with order-0 pages and thus can
842 * avoid a possible cache line miss for requests that fit all
843 * page orders.
844 */
845 if (n <= v && v <= PAGE_SIZE)
846 return true;
847
848 head = compound_head(page);
849 v += (page - head) << PAGE_SHIFT;
850
851 if (likely(n <= v && v <= (page_size(head))))
852 return true;
853 WARN_ON(1);
854 return false;
855 }
856
857 static size_t __copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
858 struct iov_iter *i)
859 {
860 if (likely(iter_is_iovec(i)))
861 return copy_page_to_iter_iovec(page, offset, bytes, i);
862 if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
863 void *kaddr = kmap_local_page(page);
864 size_t wanted = _copy_to_iter(kaddr + offset, bytes, i);
865 kunmap_local(kaddr);
866 return wanted;
867 }
868 if (iov_iter_is_pipe(i))
869 return copy_page_to_iter_pipe(page, offset, bytes, i);
870 if (unlikely(iov_iter_is_discard(i))) {
871 if (unlikely(i->count < bytes))
872 bytes = i->count;
873 i->count -= bytes;
874 return bytes;
875 }
876 WARN_ON(1);
877 return 0;
878 }
879
880 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
881 struct iov_iter *i)
882 {
883 size_t res = 0;
884 if (unlikely(!page_copy_sane(page, offset, bytes)))
885 return 0;
886 page += offset / PAGE_SIZE; // first subpage
887 offset %= PAGE_SIZE;
888 while (1) {
889 size_t n = __copy_page_to_iter(page, offset,
890 min(bytes, (size_t)PAGE_SIZE - offset), i);
891 res += n;
892 bytes -= n;
893 if (!bytes || !n)
894 break;
895 offset += n;
896 if (offset == PAGE_SIZE) {
897 page++;
898 offset = 0;
899 }
900 }
901 return res;
902 }
903 EXPORT_SYMBOL(copy_page_to_iter);
904
905 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
906 struct iov_iter *i)
907 {
908 if (unlikely(!page_copy_sane(page, offset, bytes)))
909 return 0;
910 if (likely(iter_is_iovec(i)))
911 return copy_page_from_iter_iovec(page, offset, bytes, i);
912 if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
913 void *kaddr = kmap_local_page(page);
914 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
915 kunmap_local(kaddr);
916 return wanted;
917 }
918 WARN_ON(1);
919 return 0;
920 }
921 EXPORT_SYMBOL(copy_page_from_iter);
922
923 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
924 {
925 struct pipe_inode_info *pipe = i->pipe;
926 unsigned int p_mask = pipe->ring_size - 1;
927 unsigned int i_head;
928 size_t n, off;
929
930 if (!sanity(i))
931 return 0;
932
933 bytes = n = push_pipe(i, bytes, &i_head, &off);
934 if (unlikely(!n))
935 return 0;
936
937 do {
938 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
939 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
940 memset(p + off, 0, chunk);
941 kunmap_local(p);
942 i->head = i_head;
943 i->iov_offset = off + chunk;
944 n -= chunk;
945 off = 0;
946 i_head++;
947 } while (n);
948 i->count -= bytes;
949 return bytes;
950 }
951
952 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
953 {
954 if (unlikely(iov_iter_is_pipe(i)))
955 return pipe_zero(bytes, i);
956 iterate_and_advance(i, bytes, base, len, count,
957 clear_user(base, len),
958 memset(base, 0, len)
959 )
960
961 return bytes;
962 }
963 EXPORT_SYMBOL(iov_iter_zero);
964
965 size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes,
966 struct iov_iter *i)
967 {
968 char *kaddr = kmap_atomic(page), *p = kaddr + offset;
969 if (unlikely(!page_copy_sane(page, offset, bytes))) {
970 kunmap_atomic(kaddr);
971 return 0;
972 }
973 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
974 kunmap_atomic(kaddr);
975 WARN_ON(1);
976 return 0;
977 }
978 iterate_and_advance(i, bytes, base, len, off,
979 copyin(p + off, base, len),
980 memcpy(p + off, base, len)
981 )
982 kunmap_atomic(kaddr);
983 return bytes;
984 }
985 EXPORT_SYMBOL(copy_page_from_iter_atomic);
986
987 static inline void pipe_truncate(struct iov_iter *i)
988 {
989 struct pipe_inode_info *pipe = i->pipe;
990 unsigned int p_tail = pipe->tail;
991 unsigned int p_head = pipe->head;
992 unsigned int p_mask = pipe->ring_size - 1;
993
994 if (!pipe_empty(p_head, p_tail)) {
995 struct pipe_buffer *buf;
996 unsigned int i_head = i->head;
997 size_t off = i->iov_offset;
998
999 if (off) {
1000 buf = &pipe->bufs[i_head & p_mask];
1001 buf->len = off - buf->offset;
1002 i_head++;
1003 }
1004 while (p_head != i_head) {
1005 p_head--;
1006 pipe_buf_release(pipe, &pipe->bufs[p_head & p_mask]);
1007 }
1008
1009 pipe->head = p_head;
1010 }
1011 }
1012
1013 static void pipe_advance(struct iov_iter *i, size_t size)
1014 {
1015 struct pipe_inode_info *pipe = i->pipe;
1016 if (size) {
1017 struct pipe_buffer *buf;
1018 unsigned int p_mask = pipe->ring_size - 1;
1019 unsigned int i_head = i->head;
1020 size_t off = i->iov_offset, left = size;
1021
1022 if (off) /* make it relative to the beginning of buffer */
1023 left += off - pipe->bufs[i_head & p_mask].offset;
1024 while (1) {
1025 buf = &pipe->bufs[i_head & p_mask];
1026 if (left <= buf->len)
1027 break;
1028 left -= buf->len;
1029 i_head++;
1030 }
1031 i->head = i_head;
1032 i->iov_offset = buf->offset + left;
1033 }
1034 i->count -= size;
1035 /* ... and discard everything past that point */
1036 pipe_truncate(i);
1037 }
1038
1039 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
1040 {
1041 struct bvec_iter bi;
1042
1043 bi.bi_size = i->count;
1044 bi.bi_bvec_done = i->iov_offset;
1045 bi.bi_idx = 0;
1046 bvec_iter_advance(i->bvec, &bi, size);
1047
1048 i->bvec += bi.bi_idx;
1049 i->nr_segs -= bi.bi_idx;
1050 i->count = bi.bi_size;
1051 i->iov_offset = bi.bi_bvec_done;
1052 }
1053
1054 static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
1055 {
1056 const struct iovec *iov, *end;
1057
1058 if (!i->count)
1059 return;
1060 i->count -= size;
1061
1062 size += i->iov_offset; // from beginning of current segment
1063 for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) {
1064 if (likely(size < iov->iov_len))
1065 break;
1066 size -= iov->iov_len;
1067 }
1068 i->iov_offset = size;
1069 i->nr_segs -= iov - i->iov;
1070 i->iov = iov;
1071 }
1072
1073 void iov_iter_advance(struct iov_iter *i, size_t size)
1074 {
1075 if (unlikely(i->count < size))
1076 size = i->count;
1077 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
1078 /* iovec and kvec have identical layouts */
1079 iov_iter_iovec_advance(i, size);
1080 } else if (iov_iter_is_bvec(i)) {
1081 iov_iter_bvec_advance(i, size);
1082 } else if (iov_iter_is_pipe(i)) {
1083 pipe_advance(i, size);
1084 } else if (unlikely(iov_iter_is_xarray(i))) {
1085 i->iov_offset += size;
1086 i->count -= size;
1087 } else if (iov_iter_is_discard(i)) {
1088 i->count -= size;
1089 }
1090 }
1091 EXPORT_SYMBOL(iov_iter_advance);
1092
1093 void iov_iter_revert(struct iov_iter *i, size_t unroll)
1094 {
1095 if (!unroll)
1096 return;
1097 if (WARN_ON(unroll > MAX_RW_COUNT))
1098 return;
1099 i->count += unroll;
1100 if (unlikely(iov_iter_is_pipe(i))) {
1101 struct pipe_inode_info *pipe = i->pipe;
1102 unsigned int p_mask = pipe->ring_size - 1;
1103 unsigned int i_head = i->head;
1104 size_t off = i->iov_offset;
1105 while (1) {
1106 struct pipe_buffer *b = &pipe->bufs[i_head & p_mask];
1107 size_t n = off - b->offset;
1108 if (unroll < n) {
1109 off -= unroll;
1110 break;
1111 }
1112 unroll -= n;
1113 if (!unroll && i_head == i->start_head) {
1114 off = 0;
1115 break;
1116 }
1117 i_head--;
1118 b = &pipe->bufs[i_head & p_mask];
1119 off = b->offset + b->len;
1120 }
1121 i->iov_offset = off;
1122 i->head = i_head;
1123 pipe_truncate(i);
1124 return;
1125 }
1126 if (unlikely(iov_iter_is_discard(i)))
1127 return;
1128 if (unroll <= i->iov_offset) {
1129 i->iov_offset -= unroll;
1130 return;
1131 }
1132 unroll -= i->iov_offset;
1133 if (iov_iter_is_xarray(i)) {
1134 BUG(); /* We should never go beyond the start of the specified
1135 * range since we might then be straying into pages that
1136 * aren't pinned.
1137 */
1138 } else if (iov_iter_is_bvec(i)) {
1139 const struct bio_vec *bvec = i->bvec;
1140 while (1) {
1141 size_t n = (--bvec)->bv_len;
1142 i->nr_segs++;
1143 if (unroll <= n) {
1144 i->bvec = bvec;
1145 i->iov_offset = n - unroll;
1146 return;
1147 }
1148 unroll -= n;
1149 }
1150 } else { /* same logics for iovec and kvec */
1151 const struct iovec *iov = i->iov;
1152 while (1) {
1153 size_t n = (--iov)->iov_len;
1154 i->nr_segs++;
1155 if (unroll <= n) {
1156 i->iov = iov;
1157 i->iov_offset = n - unroll;
1158 return;
1159 }
1160 unroll -= n;
1161 }
1162 }
1163 }
1164 EXPORT_SYMBOL(iov_iter_revert);
1165
1166 /*
1167 * Return the count of just the current iov_iter segment.
1168 */
1169 size_t iov_iter_single_seg_count(const struct iov_iter *i)
1170 {
1171 if (i->nr_segs > 1) {
1172 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1173 return min(i->count, i->iov->iov_len - i->iov_offset);
1174 if (iov_iter_is_bvec(i))
1175 return min(i->count, i->bvec->bv_len - i->iov_offset);
1176 }
1177 return i->count;
1178 }
1179 EXPORT_SYMBOL(iov_iter_single_seg_count);
1180
1181 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
1182 const struct kvec *kvec, unsigned long nr_segs,
1183 size_t count)
1184 {
1185 WARN_ON(direction & ~(READ | WRITE));
1186 *i = (struct iov_iter){
1187 .iter_type = ITER_KVEC,
1188 .data_source = direction,
1189 .kvec = kvec,
1190 .nr_segs = nr_segs,
1191 .iov_offset = 0,
1192 .count = count
1193 };
1194 }
1195 EXPORT_SYMBOL(iov_iter_kvec);
1196
1197 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1198 const struct bio_vec *bvec, unsigned long nr_segs,
1199 size_t count)
1200 {
1201 WARN_ON(direction & ~(READ | WRITE));
1202 *i = (struct iov_iter){
1203 .iter_type = ITER_BVEC,
1204 .data_source = direction,
1205 .bvec = bvec,
1206 .nr_segs = nr_segs,
1207 .iov_offset = 0,
1208 .count = count
1209 };
1210 }
1211 EXPORT_SYMBOL(iov_iter_bvec);
1212
1213 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1214 struct pipe_inode_info *pipe,
1215 size_t count)
1216 {
1217 BUG_ON(direction != READ);
1218 WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1219 *i = (struct iov_iter){
1220 .iter_type = ITER_PIPE,
1221 .data_source = false,
1222 .pipe = pipe,
1223 .head = pipe->head,
1224 .start_head = pipe->head,
1225 .iov_offset = 0,
1226 .count = count
1227 };
1228 }
1229 EXPORT_SYMBOL(iov_iter_pipe);
1230
1231 /**
1232 * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1233 * @i: The iterator to initialise.
1234 * @direction: The direction of the transfer.
1235 * @xarray: The xarray to access.
1236 * @start: The start file position.
1237 * @count: The size of the I/O buffer in bytes.
1238 *
1239 * Set up an I/O iterator to either draw data out of the pages attached to an
1240 * inode or to inject data into those pages. The pages *must* be prevented
1241 * from evaporation, either by taking a ref on them or locking them by the
1242 * caller.
1243 */
1244 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
1245 struct xarray *xarray, loff_t start, size_t count)
1246 {
1247 BUG_ON(direction & ~1);
1248 *i = (struct iov_iter) {
1249 .iter_type = ITER_XARRAY,
1250 .data_source = direction,
1251 .xarray = xarray,
1252 .xarray_start = start,
1253 .count = count,
1254 .iov_offset = 0
1255 };
1256 }
1257 EXPORT_SYMBOL(iov_iter_xarray);
1258
1259 /**
1260 * iov_iter_discard - Initialise an I/O iterator that discards data
1261 * @i: The iterator to initialise.
1262 * @direction: The direction of the transfer.
1263 * @count: The size of the I/O buffer in bytes.
1264 *
1265 * Set up an I/O iterator that just discards everything that's written to it.
1266 * It's only available as a READ iterator.
1267 */
1268 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1269 {
1270 BUG_ON(direction != READ);
1271 *i = (struct iov_iter){
1272 .iter_type = ITER_DISCARD,
1273 .data_source = false,
1274 .count = count,
1275 .iov_offset = 0
1276 };
1277 }
1278 EXPORT_SYMBOL(iov_iter_discard);
1279
1280 static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
1281 {
1282 unsigned long res = 0;
1283 size_t size = i->count;
1284 size_t skip = i->iov_offset;
1285 unsigned k;
1286
1287 for (k = 0; k < i->nr_segs; k++, skip = 0) {
1288 size_t len = i->iov[k].iov_len - skip;
1289 if (len) {
1290 res |= (unsigned long)i->iov[k].iov_base + skip;
1291 if (len > size)
1292 len = size;
1293 res |= len;
1294 size -= len;
1295 if (!size)
1296 break;
1297 }
1298 }
1299 return res;
1300 }
1301
1302 static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
1303 {
1304 unsigned res = 0;
1305 size_t size = i->count;
1306 unsigned skip = i->iov_offset;
1307 unsigned k;
1308
1309 for (k = 0; k < i->nr_segs; k++, skip = 0) {
1310 size_t len = i->bvec[k].bv_len - skip;
1311 res |= (unsigned long)i->bvec[k].bv_offset + skip;
1312 if (len > size)
1313 len = size;
1314 res |= len;
1315 size -= len;
1316 if (!size)
1317 break;
1318 }
1319 return res;
1320 }
1321
1322 unsigned long iov_iter_alignment(const struct iov_iter *i)
1323 {
1324 /* iovec and kvec have identical layouts */
1325 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1326 return iov_iter_alignment_iovec(i);
1327
1328 if (iov_iter_is_bvec(i))
1329 return iov_iter_alignment_bvec(i);
1330
1331 if (iov_iter_is_pipe(i)) {
1332 unsigned int p_mask = i->pipe->ring_size - 1;
1333 size_t size = i->count;
1334
1335 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->head & p_mask]))
1336 return size | i->iov_offset;
1337 return size;
1338 }
1339
1340 if (iov_iter_is_xarray(i))
1341 return (i->xarray_start + i->iov_offset) | i->count;
1342
1343 return 0;
1344 }
1345 EXPORT_SYMBOL(iov_iter_alignment);
1346
1347 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1348 {
1349 unsigned long res = 0;
1350 unsigned long v = 0;
1351 size_t size = i->count;
1352 unsigned k;
1353
1354 if (WARN_ON(!iter_is_iovec(i)))
1355 return ~0U;
1356
1357 for (k = 0; k < i->nr_segs; k++) {
1358 if (i->iov[k].iov_len) {
1359 unsigned long base = (unsigned long)i->iov[k].iov_base;
1360 if (v) // if not the first one
1361 res |= base | v; // this start | previous end
1362 v = base + i->iov[k].iov_len;
1363 if (size <= i->iov[k].iov_len)
1364 break;
1365 size -= i->iov[k].iov_len;
1366 }
1367 }
1368 return res;
1369 }
1370 EXPORT_SYMBOL(iov_iter_gap_alignment);
1371
1372 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1373 size_t maxsize,
1374 struct page **pages,
1375 int iter_head,
1376 size_t *start)
1377 {
1378 struct pipe_inode_info *pipe = i->pipe;
1379 unsigned int p_mask = pipe->ring_size - 1;
1380 ssize_t n = push_pipe(i, maxsize, &iter_head, start);
1381 if (!n)
1382 return -EFAULT;
1383
1384 maxsize = n;
1385 n += *start;
1386 while (n > 0) {
1387 get_page(*pages++ = pipe->bufs[iter_head & p_mask].page);
1388 iter_head++;
1389 n -= PAGE_SIZE;
1390 }
1391
1392 return maxsize;
1393 }
1394
1395 static ssize_t pipe_get_pages(struct iov_iter *i,
1396 struct page **pages, size_t maxsize, unsigned maxpages,
1397 size_t *start)
1398 {
1399 unsigned int iter_head, npages;
1400 size_t capacity;
1401
1402 if (!sanity(i))
1403 return -EFAULT;
1404
1405 data_start(i, &iter_head, start);
1406 /* Amount of free space: some of this one + all after this one */
1407 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1408 capacity = min(npages, maxpages) * PAGE_SIZE - *start;
1409
1410 return __pipe_get_pages(i, min(maxsize, capacity), pages, iter_head, start);
1411 }
1412
1413 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1414 pgoff_t index, unsigned int nr_pages)
1415 {
1416 XA_STATE(xas, xa, index);
1417 struct page *page;
1418 unsigned int ret = 0;
1419
1420 rcu_read_lock();
1421 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1422 if (xas_retry(&xas, page))
1423 continue;
1424
1425 /* Has the page moved or been split? */
1426 if (unlikely(page != xas_reload(&xas))) {
1427 xas_reset(&xas);
1428 continue;
1429 }
1430
1431 pages[ret] = find_subpage(page, xas.xa_index);
1432 get_page(pages[ret]);
1433 if (++ret == nr_pages)
1434 break;
1435 }
1436 rcu_read_unlock();
1437 return ret;
1438 }
1439
1440 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1441 struct page **pages, size_t maxsize,
1442 unsigned maxpages, size_t *_start_offset)
1443 {
1444 unsigned nr, offset;
1445 pgoff_t index, count;
1446 size_t size = maxsize;
1447 loff_t pos;
1448
1449 if (!size || !maxpages)
1450 return 0;
1451
1452 pos = i->xarray_start + i->iov_offset;
1453 index = pos >> PAGE_SHIFT;
1454 offset = pos & ~PAGE_MASK;
1455 *_start_offset = offset;
1456
1457 count = 1;
1458 if (size > PAGE_SIZE - offset) {
1459 size -= PAGE_SIZE - offset;
1460 count += size >> PAGE_SHIFT;
1461 size &= ~PAGE_MASK;
1462 if (size)
1463 count++;
1464 }
1465
1466 if (count > maxpages)
1467 count = maxpages;
1468
1469 nr = iter_xarray_populate_pages(pages, i->xarray, index, count);
1470 if (nr == 0)
1471 return 0;
1472
1473 return min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1474 }
1475
1476 /* must be done on non-empty ITER_IOVEC one */
1477 static unsigned long first_iovec_segment(const struct iov_iter *i,
1478 size_t *size, size_t *start,
1479 size_t maxsize, unsigned maxpages)
1480 {
1481 size_t skip;
1482 long k;
1483
1484 for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
1485 unsigned long addr = (unsigned long)i->iov[k].iov_base + skip;
1486 size_t len = i->iov[k].iov_len - skip;
1487
1488 if (unlikely(!len))
1489 continue;
1490 if (len > maxsize)
1491 len = maxsize;
1492 len += (*start = addr % PAGE_SIZE);
1493 if (len > maxpages * PAGE_SIZE)
1494 len = maxpages * PAGE_SIZE;
1495 *size = len;
1496 return addr & PAGE_MASK;
1497 }
1498 BUG(); // if it had been empty, we wouldn't get called
1499 }
1500
1501 /* must be done on non-empty ITER_BVEC one */
1502 static struct page *first_bvec_segment(const struct iov_iter *i,
1503 size_t *size, size_t *start,
1504 size_t maxsize, unsigned maxpages)
1505 {
1506 struct page *page;
1507 size_t skip = i->iov_offset, len;
1508
1509 len = i->bvec->bv_len - skip;
1510 if (len > maxsize)
1511 len = maxsize;
1512 skip += i->bvec->bv_offset;
1513 page = i->bvec->bv_page + skip / PAGE_SIZE;
1514 len += (*start = skip % PAGE_SIZE);
1515 if (len > maxpages * PAGE_SIZE)
1516 len = maxpages * PAGE_SIZE;
1517 *size = len;
1518 return page;
1519 }
1520
1521 ssize_t iov_iter_get_pages(struct iov_iter *i,
1522 struct page **pages, size_t maxsize, unsigned maxpages,
1523 size_t *start)
1524 {
1525 size_t len;
1526 int n, res;
1527
1528 if (maxsize > i->count)
1529 maxsize = i->count;
1530 if (!maxsize)
1531 return 0;
1532
1533 if (likely(iter_is_iovec(i))) {
1534 unsigned int gup_flags = 0;
1535 unsigned long addr;
1536
1537 if (iov_iter_rw(i) != WRITE)
1538 gup_flags |= FOLL_WRITE;
1539 if (i->nofault)
1540 gup_flags |= FOLL_NOFAULT;
1541
1542 addr = first_iovec_segment(i, &len, start, maxsize, maxpages);
1543 n = DIV_ROUND_UP(len, PAGE_SIZE);
1544 res = get_user_pages_fast(addr, n, gup_flags, pages);
1545 if (unlikely(res <= 0))
1546 return res;
1547 return (res == n ? len : res * PAGE_SIZE) - *start;
1548 }
1549 if (iov_iter_is_bvec(i)) {
1550 struct page *page;
1551
1552 page = first_bvec_segment(i, &len, start, maxsize, maxpages);
1553 n = DIV_ROUND_UP(len, PAGE_SIZE);
1554 while (n--)
1555 get_page(*pages++ = page++);
1556 return len - *start;
1557 }
1558 if (iov_iter_is_pipe(i))
1559 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1560 if (iov_iter_is_xarray(i))
1561 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1562 return -EFAULT;
1563 }
1564 EXPORT_SYMBOL(iov_iter_get_pages);
1565
1566 static struct page **get_pages_array(size_t n)
1567 {
1568 return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1569 }
1570
1571 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1572 struct page ***pages, size_t maxsize,
1573 size_t *start)
1574 {
1575 struct page **p;
1576 unsigned int iter_head, npages;
1577 ssize_t n;
1578
1579 if (!sanity(i))
1580 return -EFAULT;
1581
1582 data_start(i, &iter_head, start);
1583 /* Amount of free space: some of this one + all after this one */
1584 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1585 n = npages * PAGE_SIZE - *start;
1586 if (maxsize > n)
1587 maxsize = n;
1588 else
1589 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1590 p = get_pages_array(npages);
1591 if (!p)
1592 return -ENOMEM;
1593 n = __pipe_get_pages(i, maxsize, p, iter_head, start);
1594 if (n > 0)
1595 *pages = p;
1596 else
1597 kvfree(p);
1598 return n;
1599 }
1600
1601 static ssize_t iter_xarray_get_pages_alloc(struct iov_iter *i,
1602 struct page ***pages, size_t maxsize,
1603 size_t *_start_offset)
1604 {
1605 struct page **p;
1606 unsigned nr, offset;
1607 pgoff_t index, count;
1608 size_t size = maxsize;
1609 loff_t pos;
1610
1611 if (!size)
1612 return 0;
1613
1614 pos = i->xarray_start + i->iov_offset;
1615 index = pos >> PAGE_SHIFT;
1616 offset = pos & ~PAGE_MASK;
1617 *_start_offset = offset;
1618
1619 count = 1;
1620 if (size > PAGE_SIZE - offset) {
1621 size -= PAGE_SIZE - offset;
1622 count += size >> PAGE_SHIFT;
1623 size &= ~PAGE_MASK;
1624 if (size)
1625 count++;
1626 }
1627
1628 p = get_pages_array(count);
1629 if (!p)
1630 return -ENOMEM;
1631 *pages = p;
1632
1633 nr = iter_xarray_populate_pages(p, i->xarray, index, count);
1634 if (nr == 0)
1635 return 0;
1636
1637 return min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1638 }
1639
1640 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1641 struct page ***pages, size_t maxsize,
1642 size_t *start)
1643 {
1644 struct page **p;
1645 size_t len;
1646 int n, res;
1647
1648 if (maxsize > i->count)
1649 maxsize = i->count;
1650 if (!maxsize)
1651 return 0;
1652
1653 if (likely(iter_is_iovec(i))) {
1654 unsigned int gup_flags = 0;
1655 unsigned long addr;
1656
1657 if (iov_iter_rw(i) != WRITE)
1658 gup_flags |= FOLL_WRITE;
1659 if (i->nofault)
1660 gup_flags |= FOLL_NOFAULT;
1661
1662 addr = first_iovec_segment(i, &len, start, maxsize, ~0U);
1663 n = DIV_ROUND_UP(len, PAGE_SIZE);
1664 p = get_pages_array(n);
1665 if (!p)
1666 return -ENOMEM;
1667 res = get_user_pages_fast(addr, n, gup_flags, p);
1668 if (unlikely(res <= 0)) {
1669 kvfree(p);
1670 *pages = NULL;
1671 return res;
1672 }
1673 *pages = p;
1674 return (res == n ? len : res * PAGE_SIZE) - *start;
1675 }
1676 if (iov_iter_is_bvec(i)) {
1677 struct page *page;
1678
1679 page = first_bvec_segment(i, &len, start, maxsize, ~0U);
1680 n = DIV_ROUND_UP(len, PAGE_SIZE);
1681 *pages = p = get_pages_array(n);
1682 if (!p)
1683 return -ENOMEM;
1684 while (n--)
1685 get_page(*p++ = page++);
1686 return len - *start;
1687 }
1688 if (iov_iter_is_pipe(i))
1689 return pipe_get_pages_alloc(i, pages, maxsize, start);
1690 if (iov_iter_is_xarray(i))
1691 return iter_xarray_get_pages_alloc(i, pages, maxsize, start);
1692 return -EFAULT;
1693 }
1694 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1695
1696 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1697 struct iov_iter *i)
1698 {
1699 __wsum sum, next;
1700 sum = *csum;
1701 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1702 WARN_ON(1);
1703 return 0;
1704 }
1705 iterate_and_advance(i, bytes, base, len, off, ({
1706 next = csum_and_copy_from_user(base, addr + off, len);
1707 sum = csum_block_add(sum, next, off);
1708 next ? 0 : len;
1709 }), ({
1710 sum = csum_and_memcpy(addr + off, base, len, sum, off);
1711 })
1712 )
1713 *csum = sum;
1714 return bytes;
1715 }
1716 EXPORT_SYMBOL(csum_and_copy_from_iter);
1717
1718 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1719 struct iov_iter *i)
1720 {
1721 struct csum_state *csstate = _csstate;
1722 __wsum sum, next;
1723
1724 if (unlikely(iov_iter_is_discard(i))) {
1725 WARN_ON(1); /* for now */
1726 return 0;
1727 }
1728
1729 sum = csum_shift(csstate->csum, csstate->off);
1730 if (unlikely(iov_iter_is_pipe(i)))
1731 bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum);
1732 else iterate_and_advance(i, bytes, base, len, off, ({
1733 next = csum_and_copy_to_user(addr + off, base, len);
1734 sum = csum_block_add(sum, next, off);
1735 next ? 0 : len;
1736 }), ({
1737 sum = csum_and_memcpy(base, addr + off, len, sum, off);
1738 })
1739 )
1740 csstate->csum = csum_shift(sum, csstate->off);
1741 csstate->off += bytes;
1742 return bytes;
1743 }
1744 EXPORT_SYMBOL(csum_and_copy_to_iter);
1745
1746 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1747 struct iov_iter *i)
1748 {
1749 #ifdef CONFIG_CRYPTO_HASH
1750 struct ahash_request *hash = hashp;
1751 struct scatterlist sg;
1752 size_t copied;
1753
1754 copied = copy_to_iter(addr, bytes, i);
1755 sg_init_one(&sg, addr, copied);
1756 ahash_request_set_crypt(hash, &sg, NULL, copied);
1757 crypto_ahash_update(hash);
1758 return copied;
1759 #else
1760 return 0;
1761 #endif
1762 }
1763 EXPORT_SYMBOL(hash_and_copy_to_iter);
1764
1765 static int iov_npages(const struct iov_iter *i, int maxpages)
1766 {
1767 size_t skip = i->iov_offset, size = i->count;
1768 const struct iovec *p;
1769 int npages = 0;
1770
1771 for (p = i->iov; size; skip = 0, p++) {
1772 unsigned offs = offset_in_page(p->iov_base + skip);
1773 size_t len = min(p->iov_len - skip, size);
1774
1775 if (len) {
1776 size -= len;
1777 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1778 if (unlikely(npages > maxpages))
1779 return maxpages;
1780 }
1781 }
1782 return npages;
1783 }
1784
1785 static int bvec_npages(const struct iov_iter *i, int maxpages)
1786 {
1787 size_t skip = i->iov_offset, size = i->count;
1788 const struct bio_vec *p;
1789 int npages = 0;
1790
1791 for (p = i->bvec; size; skip = 0, p++) {
1792 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1793 size_t len = min(p->bv_len - skip, size);
1794
1795 size -= len;
1796 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1797 if (unlikely(npages > maxpages))
1798 return maxpages;
1799 }
1800 return npages;
1801 }
1802
1803 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1804 {
1805 if (unlikely(!i->count))
1806 return 0;
1807 /* iovec and kvec have identical layouts */
1808 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1809 return iov_npages(i, maxpages);
1810 if (iov_iter_is_bvec(i))
1811 return bvec_npages(i, maxpages);
1812 if (iov_iter_is_pipe(i)) {
1813 unsigned int iter_head;
1814 int npages;
1815 size_t off;
1816
1817 if (!sanity(i))
1818 return 0;
1819
1820 data_start(i, &iter_head, &off);
1821 /* some of this one + all after this one */
1822 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1823 return min(npages, maxpages);
1824 }
1825 if (iov_iter_is_xarray(i)) {
1826 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1827 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1828 return min(npages, maxpages);
1829 }
1830 return 0;
1831 }
1832 EXPORT_SYMBOL(iov_iter_npages);
1833
1834 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1835 {
1836 *new = *old;
1837 if (unlikely(iov_iter_is_pipe(new))) {
1838 WARN_ON(1);
1839 return NULL;
1840 }
1841 if (unlikely(iov_iter_is_discard(new) || iov_iter_is_xarray(new)))
1842 return NULL;
1843 if (iov_iter_is_bvec(new))
1844 return new->bvec = kmemdup(new->bvec,
1845 new->nr_segs * sizeof(struct bio_vec),
1846 flags);
1847 else
1848 /* iovec and kvec have identical layout */
1849 return new->iov = kmemdup(new->iov,
1850 new->nr_segs * sizeof(struct iovec),
1851 flags);
1852 }
1853 EXPORT_SYMBOL(dup_iter);
1854
1855 static int copy_compat_iovec_from_user(struct iovec *iov,
1856 const struct iovec __user *uvec, unsigned long nr_segs)
1857 {
1858 const struct compat_iovec __user *uiov =
1859 (const struct compat_iovec __user *)uvec;
1860 int ret = -EFAULT, i;
1861
1862 if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1863 return -EFAULT;
1864
1865 for (i = 0; i < nr_segs; i++) {
1866 compat_uptr_t buf;
1867 compat_ssize_t len;
1868
1869 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1870 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1871
1872 /* check for compat_size_t not fitting in compat_ssize_t .. */
1873 if (len < 0) {
1874 ret = -EINVAL;
1875 goto uaccess_end;
1876 }
1877 iov[i].iov_base = compat_ptr(buf);
1878 iov[i].iov_len = len;
1879 }
1880
1881 ret = 0;
1882 uaccess_end:
1883 user_access_end();
1884 return ret;
1885 }
1886
1887 static int copy_iovec_from_user(struct iovec *iov,
1888 const struct iovec __user *uvec, unsigned long nr_segs)
1889 {
1890 unsigned long seg;
1891
1892 if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1893 return -EFAULT;
1894 for (seg = 0; seg < nr_segs; seg++) {
1895 if ((ssize_t)iov[seg].iov_len < 0)
1896 return -EINVAL;
1897 }
1898
1899 return 0;
1900 }
1901
1902 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1903 unsigned long nr_segs, unsigned long fast_segs,
1904 struct iovec *fast_iov, bool compat)
1905 {
1906 struct iovec *iov = fast_iov;
1907 int ret;
1908
1909 /*
1910 * SuS says "The readv() function *may* fail if the iovcnt argument was
1911 * less than or equal to 0, or greater than {IOV_MAX}. Linux has
1912 * traditionally returned zero for zero segments, so...
1913 */
1914 if (nr_segs == 0)
1915 return iov;
1916 if (nr_segs > UIO_MAXIOV)
1917 return ERR_PTR(-EINVAL);
1918 if (nr_segs > fast_segs) {
1919 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1920 if (!iov)
1921 return ERR_PTR(-ENOMEM);
1922 }
1923
1924 if (compat)
1925 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1926 else
1927 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1928 if (ret) {
1929 if (iov != fast_iov)
1930 kfree(iov);
1931 return ERR_PTR(ret);
1932 }
1933
1934 return iov;
1935 }
1936
1937 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1938 unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1939 struct iov_iter *i, bool compat)
1940 {
1941 ssize_t total_len = 0;
1942 unsigned long seg;
1943 struct iovec *iov;
1944
1945 iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1946 if (IS_ERR(iov)) {
1947 *iovp = NULL;
1948 return PTR_ERR(iov);
1949 }
1950
1951 /*
1952 * According to the Single Unix Specification we should return EINVAL if
1953 * an element length is < 0 when cast to ssize_t or if the total length
1954 * would overflow the ssize_t return value of the system call.
1955 *
1956 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1957 * overflow case.
1958 */
1959 for (seg = 0; seg < nr_segs; seg++) {
1960 ssize_t len = (ssize_t)iov[seg].iov_len;
1961
1962 if (!access_ok(iov[seg].iov_base, len)) {
1963 if (iov != *iovp)
1964 kfree(iov);
1965 *iovp = NULL;
1966 return -EFAULT;
1967 }
1968
1969 if (len > MAX_RW_COUNT - total_len) {
1970 len = MAX_RW_COUNT - total_len;
1971 iov[seg].iov_len = len;
1972 }
1973 total_len += len;
1974 }
1975
1976 iov_iter_init(i, type, iov, nr_segs, total_len);
1977 if (iov == *iovp)
1978 *iovp = NULL;
1979 else
1980 *iovp = iov;
1981 return total_len;
1982 }
1983
1984 /**
1985 * import_iovec() - Copy an array of &struct iovec from userspace
1986 * into the kernel, check that it is valid, and initialize a new
1987 * &struct iov_iter iterator to access it.
1988 *
1989 * @type: One of %READ or %WRITE.
1990 * @uvec: Pointer to the userspace array.
1991 * @nr_segs: Number of elements in userspace array.
1992 * @fast_segs: Number of elements in @iov.
1993 * @iovp: (input and output parameter) Pointer to pointer to (usually small
1994 * on-stack) kernel array.
1995 * @i: Pointer to iterator that will be initialized on success.
1996 *
1997 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1998 * then this function places %NULL in *@iov on return. Otherwise, a new
1999 * array will be allocated and the result placed in *@iov. This means that
2000 * the caller may call kfree() on *@iov regardless of whether the small
2001 * on-stack array was used or not (and regardless of whether this function
2002 * returns an error or not).
2003 *
2004 * Return: Negative error code on error, bytes imported on success
2005 */
2006 ssize_t import_iovec(int type, const struct iovec __user *uvec,
2007 unsigned nr_segs, unsigned fast_segs,
2008 struct iovec **iovp, struct iov_iter *i)
2009 {
2010 return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
2011 in_compat_syscall());
2012 }
2013 EXPORT_SYMBOL(import_iovec);
2014
2015 int import_single_range(int rw, void __user *buf, size_t len,
2016 struct iovec *iov, struct iov_iter *i)
2017 {
2018 if (len > MAX_RW_COUNT)
2019 len = MAX_RW_COUNT;
2020 if (unlikely(!access_ok(buf, len)))
2021 return -EFAULT;
2022
2023 iov->iov_base = buf;
2024 iov->iov_len = len;
2025 iov_iter_init(i, rw, iov, 1, len);
2026 return 0;
2027 }
2028 EXPORT_SYMBOL(import_single_range);
2029
2030 /**
2031 * iov_iter_restore() - Restore a &struct iov_iter to the same state as when
2032 * iov_iter_save_state() was called.
2033 *
2034 * @i: &struct iov_iter to restore
2035 * @state: state to restore from
2036 *
2037 * Used after iov_iter_save_state() to bring restore @i, if operations may
2038 * have advanced it.
2039 *
2040 * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC
2041 */
2042 void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state)
2043 {
2044 if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i)) &&
2045 !iov_iter_is_kvec(i))
2046 return;
2047 i->iov_offset = state->iov_offset;
2048 i->count = state->count;
2049 /*
2050 * For the *vec iters, nr_segs + iov is constant - if we increment
2051 * the vec, then we also decrement the nr_segs count. Hence we don't
2052 * need to track both of these, just one is enough and we can deduct
2053 * the other from that. ITER_KVEC and ITER_IOVEC are the same struct
2054 * size, so we can just increment the iov pointer as they are unionzed.
2055 * ITER_BVEC _may_ be the same size on some archs, but on others it is
2056 * not. Be safe and handle it separately.
2057 */
2058 BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
2059 if (iov_iter_is_bvec(i))
2060 i->bvec -= state->nr_segs - i->nr_segs;
2061 else
2062 i->iov -= state->nr_segs - i->nr_segs;
2063 i->nr_segs = state->nr_segs;
2064 }
Ubuntu Jammy Linux kernel mirror