]> 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
UBUNTU: SAUCE: whitelist platforms that needs save/restore ASPM L1SS for suspend...
[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 size_t n, off, xfer = 0;
695
696 if (!sanity(i))
697 return 0;
698
699 n = push_pipe(i, bytes, &i_head, &off);
700 while (n) {
701 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
702 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
703 unsigned long rem;
704 rem = copy_mc_to_kernel(p + off, addr + xfer, chunk);
705 chunk -= rem;
706 kunmap_local(p);
707 i->head = i_head;
708 i->iov_offset = off + chunk;
709 xfer += chunk;
710 if (rem)
711 break;
712 n -= chunk;
713 off = 0;
714 i_head++;
715 }
716 i->count -= xfer;
717 return xfer;
718 }
719
720 /**
721 * _copy_mc_to_iter - copy to iter with source memory error exception handling
722 * @addr: source kernel address
723 * @bytes: total transfer length
724 * @i: destination iterator
725 *
726 * The pmem driver deploys this for the dax operation
727 * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
728 * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
729 * successfully copied.
730 *
731 * The main differences between this and typical _copy_to_iter().
732 *
733 * * Typical tail/residue handling after a fault retries the copy
734 * byte-by-byte until the fault happens again. Re-triggering machine
735 * checks is potentially fatal so the implementation uses source
736 * alignment and poison alignment assumptions to avoid re-triggering
737 * hardware exceptions.
738 *
739 * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
740 * Compare to copy_to_iter() where only ITER_IOVEC attempts might return
741 * a short copy.
742 *
743 * Return: number of bytes copied (may be %0)
744 */
745 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
746 {
747 if (unlikely(iov_iter_is_pipe(i)))
748 return copy_mc_pipe_to_iter(addr, bytes, i);
749 if (iter_is_iovec(i))
750 might_fault();
751 __iterate_and_advance(i, bytes, base, len, off,
752 copyout_mc(base, addr + off, len),
753 copy_mc_to_kernel(base, addr + off, len)
754 )
755
756 return bytes;
757 }
758 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
759 #endif /* CONFIG_ARCH_HAS_COPY_MC */
760
761 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
762 {
763 if (unlikely(iov_iter_is_pipe(i))) {
764 WARN_ON(1);
765 return 0;
766 }
767 if (iter_is_iovec(i))
768 might_fault();
769 iterate_and_advance(i, bytes, base, len, off,
770 copyin(addr + off, base, len),
771 memcpy(addr + off, base, len)
772 )
773
774 return bytes;
775 }
776 EXPORT_SYMBOL(_copy_from_iter);
777
778 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
779 {
780 if (unlikely(iov_iter_is_pipe(i))) {
781 WARN_ON(1);
782 return 0;
783 }
784 iterate_and_advance(i, bytes, base, len, off,
785 __copy_from_user_inatomic_nocache(addr + off, base, len),
786 memcpy(addr + off, base, len)
787 )
788
789 return bytes;
790 }
791 EXPORT_SYMBOL(_copy_from_iter_nocache);
792
793 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
794 /**
795 * _copy_from_iter_flushcache - write destination through cpu cache
796 * @addr: destination kernel address
797 * @bytes: total transfer length
798 * @i: source iterator
799 *
800 * The pmem driver arranges for filesystem-dax to use this facility via
801 * dax_copy_from_iter() for ensuring that writes to persistent memory
802 * are flushed through the CPU cache. It is differentiated from
803 * _copy_from_iter_nocache() in that guarantees all data is flushed for
804 * all iterator types. The _copy_from_iter_nocache() only attempts to
805 * bypass the cache for the ITER_IOVEC case, and on some archs may use
806 * instructions that strand dirty-data in the cache.
807 *
808 * Return: number of bytes copied (may be %0)
809 */
810 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
811 {
812 if (unlikely(iov_iter_is_pipe(i))) {
813 WARN_ON(1);
814 return 0;
815 }
816 iterate_and_advance(i, bytes, base, len, off,
817 __copy_from_user_flushcache(addr + off, base, len),
818 memcpy_flushcache(addr + off, base, len)
819 )
820
821 return bytes;
822 }
823 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
824 #endif
825
826 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
827 {
828 struct page *head;
829 size_t v = n + offset;
830
831 /*
832 * The general case needs to access the page order in order
833 * to compute the page size.
834 * However, we mostly deal with order-0 pages and thus can
835 * avoid a possible cache line miss for requests that fit all
836 * page orders.
837 */
838 if (n <= v && v <= PAGE_SIZE)
839 return true;
840
841 head = compound_head(page);
842 v += (page - head) << PAGE_SHIFT;
843
844 if (likely(n <= v && v <= (page_size(head))))
845 return true;
846 WARN_ON(1);
847 return false;
848 }
849
850 static size_t __copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
851 struct iov_iter *i)
852 {
853 if (likely(iter_is_iovec(i)))
854 return copy_page_to_iter_iovec(page, offset, bytes, i);
855 if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
856 void *kaddr = kmap_local_page(page);
857 size_t wanted = _copy_to_iter(kaddr + offset, bytes, i);
858 kunmap_local(kaddr);
859 return wanted;
860 }
861 if (iov_iter_is_pipe(i))
862 return copy_page_to_iter_pipe(page, offset, bytes, i);
863 if (unlikely(iov_iter_is_discard(i))) {
864 if (unlikely(i->count < bytes))
865 bytes = i->count;
866 i->count -= bytes;
867 return bytes;
868 }
869 WARN_ON(1);
870 return 0;
871 }
872
873 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
874 struct iov_iter *i)
875 {
876 size_t res = 0;
877 if (unlikely(!page_copy_sane(page, offset, bytes)))
878 return 0;
879 page += offset / PAGE_SIZE; // first subpage
880 offset %= PAGE_SIZE;
881 while (1) {
882 size_t n = __copy_page_to_iter(page, offset,
883 min(bytes, (size_t)PAGE_SIZE - offset), i);
884 res += n;
885 bytes -= n;
886 if (!bytes || !n)
887 break;
888 offset += n;
889 if (offset == PAGE_SIZE) {
890 page++;
891 offset = 0;
892 }
893 }
894 return res;
895 }
896 EXPORT_SYMBOL(copy_page_to_iter);
897
898 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
899 struct iov_iter *i)
900 {
901 if (unlikely(!page_copy_sane(page, offset, bytes)))
902 return 0;
903 if (likely(iter_is_iovec(i)))
904 return copy_page_from_iter_iovec(page, offset, bytes, i);
905 if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) {
906 void *kaddr = kmap_local_page(page);
907 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
908 kunmap_local(kaddr);
909 return wanted;
910 }
911 WARN_ON(1);
912 return 0;
913 }
914 EXPORT_SYMBOL(copy_page_from_iter);
915
916 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
917 {
918 struct pipe_inode_info *pipe = i->pipe;
919 unsigned int p_mask = pipe->ring_size - 1;
920 unsigned int i_head;
921 size_t n, off;
922
923 if (!sanity(i))
924 return 0;
925
926 bytes = n = push_pipe(i, bytes, &i_head, &off);
927 if (unlikely(!n))
928 return 0;
929
930 do {
931 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
932 char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page);
933 memset(p + off, 0, chunk);
934 kunmap_local(p);
935 i->head = i_head;
936 i->iov_offset = off + chunk;
937 n -= chunk;
938 off = 0;
939 i_head++;
940 } while (n);
941 i->count -= bytes;
942 return bytes;
943 }
944
945 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
946 {
947 if (unlikely(iov_iter_is_pipe(i)))
948 return pipe_zero(bytes, i);
949 iterate_and_advance(i, bytes, base, len, count,
950 clear_user(base, len),
951 memset(base, 0, len)
952 )
953
954 return bytes;
955 }
956 EXPORT_SYMBOL(iov_iter_zero);
957
958 size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes,
959 struct iov_iter *i)
960 {
961 char *kaddr = kmap_atomic(page), *p = kaddr + offset;
962 if (unlikely(!page_copy_sane(page, offset, bytes))) {
963 kunmap_atomic(kaddr);
964 return 0;
965 }
966 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
967 kunmap_atomic(kaddr);
968 WARN_ON(1);
969 return 0;
970 }
971 iterate_and_advance(i, bytes, base, len, off,
972 copyin(p + off, base, len),
973 memcpy(p + off, base, len)
974 )
975 kunmap_atomic(kaddr);
976 return bytes;
977 }
978 EXPORT_SYMBOL(copy_page_from_iter_atomic);
979
980 static inline void pipe_truncate(struct iov_iter *i)
981 {
982 struct pipe_inode_info *pipe = i->pipe;
983 unsigned int p_tail = pipe->tail;
984 unsigned int p_head = pipe->head;
985 unsigned int p_mask = pipe->ring_size - 1;
986
987 if (!pipe_empty(p_head, p_tail)) {
988 struct pipe_buffer *buf;
989 unsigned int i_head = i->head;
990 size_t off = i->iov_offset;
991
992 if (off) {
993 buf = &pipe->bufs[i_head & p_mask];
994 buf->len = off - buf->offset;
995 i_head++;
996 }
997 while (p_head != i_head) {
998 p_head--;
999 pipe_buf_release(pipe, &pipe->bufs[p_head & p_mask]);
1000 }
1001
1002 pipe->head = p_head;
1003 }
1004 }
1005
1006 static void pipe_advance(struct iov_iter *i, size_t size)
1007 {
1008 struct pipe_inode_info *pipe = i->pipe;
1009 if (size) {
1010 struct pipe_buffer *buf;
1011 unsigned int p_mask = pipe->ring_size - 1;
1012 unsigned int i_head = i->head;
1013 size_t off = i->iov_offset, left = size;
1014
1015 if (off) /* make it relative to the beginning of buffer */
1016 left += off - pipe->bufs[i_head & p_mask].offset;
1017 while (1) {
1018 buf = &pipe->bufs[i_head & p_mask];
1019 if (left <= buf->len)
1020 break;
1021 left -= buf->len;
1022 i_head++;
1023 }
1024 i->head = i_head;
1025 i->iov_offset = buf->offset + left;
1026 }
1027 i->count -= size;
1028 /* ... and discard everything past that point */
1029 pipe_truncate(i);
1030 }
1031
1032 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
1033 {
1034 struct bvec_iter bi;
1035
1036 bi.bi_size = i->count;
1037 bi.bi_bvec_done = i->iov_offset;
1038 bi.bi_idx = 0;
1039 bvec_iter_advance(i->bvec, &bi, size);
1040
1041 i->bvec += bi.bi_idx;
1042 i->nr_segs -= bi.bi_idx;
1043 i->count = bi.bi_size;
1044 i->iov_offset = bi.bi_bvec_done;
1045 }
1046
1047 static void iov_iter_iovec_advance(struct iov_iter *i, size_t size)
1048 {
1049 const struct iovec *iov, *end;
1050
1051 if (!i->count)
1052 return;
1053 i->count -= size;
1054
1055 size += i->iov_offset; // from beginning of current segment
1056 for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) {
1057 if (likely(size < iov->iov_len))
1058 break;
1059 size -= iov->iov_len;
1060 }
1061 i->iov_offset = size;
1062 i->nr_segs -= iov - i->iov;
1063 i->iov = iov;
1064 }
1065
1066 void iov_iter_advance(struct iov_iter *i, size_t size)
1067 {
1068 if (unlikely(i->count < size))
1069 size = i->count;
1070 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) {
1071 /* iovec and kvec have identical layouts */
1072 iov_iter_iovec_advance(i, size);
1073 } else if (iov_iter_is_bvec(i)) {
1074 iov_iter_bvec_advance(i, size);
1075 } else if (iov_iter_is_pipe(i)) {
1076 pipe_advance(i, size);
1077 } else if (unlikely(iov_iter_is_xarray(i))) {
1078 i->iov_offset += size;
1079 i->count -= size;
1080 } else if (iov_iter_is_discard(i)) {
1081 i->count -= size;
1082 }
1083 }
1084 EXPORT_SYMBOL(iov_iter_advance);
1085
1086 void iov_iter_revert(struct iov_iter *i, size_t unroll)
1087 {
1088 if (!unroll)
1089 return;
1090 if (WARN_ON(unroll > MAX_RW_COUNT))
1091 return;
1092 i->count += unroll;
1093 if (unlikely(iov_iter_is_pipe(i))) {
1094 struct pipe_inode_info *pipe = i->pipe;
1095 unsigned int p_mask = pipe->ring_size - 1;
1096 unsigned int i_head = i->head;
1097 size_t off = i->iov_offset;
1098 while (1) {
1099 struct pipe_buffer *b = &pipe->bufs[i_head & p_mask];
1100 size_t n = off - b->offset;
1101 if (unroll < n) {
1102 off -= unroll;
1103 break;
1104 }
1105 unroll -= n;
1106 if (!unroll && i_head == i->start_head) {
1107 off = 0;
1108 break;
1109 }
1110 i_head--;
1111 b = &pipe->bufs[i_head & p_mask];
1112 off = b->offset + b->len;
1113 }
1114 i->iov_offset = off;
1115 i->head = i_head;
1116 pipe_truncate(i);
1117 return;
1118 }
1119 if (unlikely(iov_iter_is_discard(i)))
1120 return;
1121 if (unroll <= i->iov_offset) {
1122 i->iov_offset -= unroll;
1123 return;
1124 }
1125 unroll -= i->iov_offset;
1126 if (iov_iter_is_xarray(i)) {
1127 BUG(); /* We should never go beyond the start of the specified
1128 * range since we might then be straying into pages that
1129 * aren't pinned.
1130 */
1131 } else if (iov_iter_is_bvec(i)) {
1132 const struct bio_vec *bvec = i->bvec;
1133 while (1) {
1134 size_t n = (--bvec)->bv_len;
1135 i->nr_segs++;
1136 if (unroll <= n) {
1137 i->bvec = bvec;
1138 i->iov_offset = n - unroll;
1139 return;
1140 }
1141 unroll -= n;
1142 }
1143 } else { /* same logics for iovec and kvec */
1144 const struct iovec *iov = i->iov;
1145 while (1) {
1146 size_t n = (--iov)->iov_len;
1147 i->nr_segs++;
1148 if (unroll <= n) {
1149 i->iov = iov;
1150 i->iov_offset = n - unroll;
1151 return;
1152 }
1153 unroll -= n;
1154 }
1155 }
1156 }
1157 EXPORT_SYMBOL(iov_iter_revert);
1158
1159 /*
1160 * Return the count of just the current iov_iter segment.
1161 */
1162 size_t iov_iter_single_seg_count(const struct iov_iter *i)
1163 {
1164 if (i->nr_segs > 1) {
1165 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1166 return min(i->count, i->iov->iov_len - i->iov_offset);
1167 if (iov_iter_is_bvec(i))
1168 return min(i->count, i->bvec->bv_len - i->iov_offset);
1169 }
1170 return i->count;
1171 }
1172 EXPORT_SYMBOL(iov_iter_single_seg_count);
1173
1174 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
1175 const struct kvec *kvec, unsigned long nr_segs,
1176 size_t count)
1177 {
1178 WARN_ON(direction & ~(READ | WRITE));
1179 *i = (struct iov_iter){
1180 .iter_type = ITER_KVEC,
1181 .data_source = direction,
1182 .kvec = kvec,
1183 .nr_segs = nr_segs,
1184 .iov_offset = 0,
1185 .count = count
1186 };
1187 }
1188 EXPORT_SYMBOL(iov_iter_kvec);
1189
1190 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1191 const struct bio_vec *bvec, unsigned long nr_segs,
1192 size_t count)
1193 {
1194 WARN_ON(direction & ~(READ | WRITE));
1195 *i = (struct iov_iter){
1196 .iter_type = ITER_BVEC,
1197 .data_source = direction,
1198 .bvec = bvec,
1199 .nr_segs = nr_segs,
1200 .iov_offset = 0,
1201 .count = count
1202 };
1203 }
1204 EXPORT_SYMBOL(iov_iter_bvec);
1205
1206 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1207 struct pipe_inode_info *pipe,
1208 size_t count)
1209 {
1210 BUG_ON(direction != READ);
1211 WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1212 *i = (struct iov_iter){
1213 .iter_type = ITER_PIPE,
1214 .data_source = false,
1215 .pipe = pipe,
1216 .head = pipe->head,
1217 .start_head = pipe->head,
1218 .iov_offset = 0,
1219 .count = count
1220 };
1221 }
1222 EXPORT_SYMBOL(iov_iter_pipe);
1223
1224 /**
1225 * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray
1226 * @i: The iterator to initialise.
1227 * @direction: The direction of the transfer.
1228 * @xarray: The xarray to access.
1229 * @start: The start file position.
1230 * @count: The size of the I/O buffer in bytes.
1231 *
1232 * Set up an I/O iterator to either draw data out of the pages attached to an
1233 * inode or to inject data into those pages. The pages *must* be prevented
1234 * from evaporation, either by taking a ref on them or locking them by the
1235 * caller.
1236 */
1237 void iov_iter_xarray(struct iov_iter *i, unsigned int direction,
1238 struct xarray *xarray, loff_t start, size_t count)
1239 {
1240 BUG_ON(direction & ~1);
1241 *i = (struct iov_iter) {
1242 .iter_type = ITER_XARRAY,
1243 .data_source = direction,
1244 .xarray = xarray,
1245 .xarray_start = start,
1246 .count = count,
1247 .iov_offset = 0
1248 };
1249 }
1250 EXPORT_SYMBOL(iov_iter_xarray);
1251
1252 /**
1253 * iov_iter_discard - Initialise an I/O iterator that discards data
1254 * @i: The iterator to initialise.
1255 * @direction: The direction of the transfer.
1256 * @count: The size of the I/O buffer in bytes.
1257 *
1258 * Set up an I/O iterator that just discards everything that's written to it.
1259 * It's only available as a READ iterator.
1260 */
1261 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1262 {
1263 BUG_ON(direction != READ);
1264 *i = (struct iov_iter){
1265 .iter_type = ITER_DISCARD,
1266 .data_source = false,
1267 .count = count,
1268 .iov_offset = 0
1269 };
1270 }
1271 EXPORT_SYMBOL(iov_iter_discard);
1272
1273 static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i)
1274 {
1275 unsigned long res = 0;
1276 size_t size = i->count;
1277 size_t skip = i->iov_offset;
1278 unsigned k;
1279
1280 for (k = 0; k < i->nr_segs; k++, skip = 0) {
1281 size_t len = i->iov[k].iov_len - skip;
1282 if (len) {
1283 res |= (unsigned long)i->iov[k].iov_base + skip;
1284 if (len > size)
1285 len = size;
1286 res |= len;
1287 size -= len;
1288 if (!size)
1289 break;
1290 }
1291 }
1292 return res;
1293 }
1294
1295 static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i)
1296 {
1297 unsigned res = 0;
1298 size_t size = i->count;
1299 unsigned skip = i->iov_offset;
1300 unsigned k;
1301
1302 for (k = 0; k < i->nr_segs; k++, skip = 0) {
1303 size_t len = i->bvec[k].bv_len - skip;
1304 res |= (unsigned long)i->bvec[k].bv_offset + skip;
1305 if (len > size)
1306 len = size;
1307 res |= len;
1308 size -= len;
1309 if (!size)
1310 break;
1311 }
1312 return res;
1313 }
1314
1315 unsigned long iov_iter_alignment(const struct iov_iter *i)
1316 {
1317 /* iovec and kvec have identical layouts */
1318 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1319 return iov_iter_alignment_iovec(i);
1320
1321 if (iov_iter_is_bvec(i))
1322 return iov_iter_alignment_bvec(i);
1323
1324 if (iov_iter_is_pipe(i)) {
1325 unsigned int p_mask = i->pipe->ring_size - 1;
1326 size_t size = i->count;
1327
1328 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->head & p_mask]))
1329 return size | i->iov_offset;
1330 return size;
1331 }
1332
1333 if (iov_iter_is_xarray(i))
1334 return (i->xarray_start + i->iov_offset) | i->count;
1335
1336 return 0;
1337 }
1338 EXPORT_SYMBOL(iov_iter_alignment);
1339
1340 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1341 {
1342 unsigned long res = 0;
1343 unsigned long v = 0;
1344 size_t size = i->count;
1345 unsigned k;
1346
1347 if (WARN_ON(!iter_is_iovec(i)))
1348 return ~0U;
1349
1350 for (k = 0; k < i->nr_segs; k++) {
1351 if (i->iov[k].iov_len) {
1352 unsigned long base = (unsigned long)i->iov[k].iov_base;
1353 if (v) // if not the first one
1354 res |= base | v; // this start | previous end
1355 v = base + i->iov[k].iov_len;
1356 if (size <= i->iov[k].iov_len)
1357 break;
1358 size -= i->iov[k].iov_len;
1359 }
1360 }
1361 return res;
1362 }
1363 EXPORT_SYMBOL(iov_iter_gap_alignment);
1364
1365 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1366 size_t maxsize,
1367 struct page **pages,
1368 int iter_head,
1369 size_t *start)
1370 {
1371 struct pipe_inode_info *pipe = i->pipe;
1372 unsigned int p_mask = pipe->ring_size - 1;
1373 ssize_t n = push_pipe(i, maxsize, &iter_head, start);
1374 if (!n)
1375 return -EFAULT;
1376
1377 maxsize = n;
1378 n += *start;
1379 while (n > 0) {
1380 get_page(*pages++ = pipe->bufs[iter_head & p_mask].page);
1381 iter_head++;
1382 n -= PAGE_SIZE;
1383 }
1384
1385 return maxsize;
1386 }
1387
1388 static ssize_t pipe_get_pages(struct iov_iter *i,
1389 struct page **pages, size_t maxsize, unsigned maxpages,
1390 size_t *start)
1391 {
1392 unsigned int iter_head, npages;
1393 size_t capacity;
1394
1395 if (!sanity(i))
1396 return -EFAULT;
1397
1398 data_start(i, &iter_head, start);
1399 /* Amount of free space: some of this one + all after this one */
1400 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1401 capacity = min(npages, maxpages) * PAGE_SIZE - *start;
1402
1403 return __pipe_get_pages(i, min(maxsize, capacity), pages, iter_head, start);
1404 }
1405
1406 static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa,
1407 pgoff_t index, unsigned int nr_pages)
1408 {
1409 XA_STATE(xas, xa, index);
1410 struct page *page;
1411 unsigned int ret = 0;
1412
1413 rcu_read_lock();
1414 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1415 if (xas_retry(&xas, page))
1416 continue;
1417
1418 /* Has the page moved or been split? */
1419 if (unlikely(page != xas_reload(&xas))) {
1420 xas_reset(&xas);
1421 continue;
1422 }
1423
1424 pages[ret] = find_subpage(page, xas.xa_index);
1425 get_page(pages[ret]);
1426 if (++ret == nr_pages)
1427 break;
1428 }
1429 rcu_read_unlock();
1430 return ret;
1431 }
1432
1433 static ssize_t iter_xarray_get_pages(struct iov_iter *i,
1434 struct page **pages, size_t maxsize,
1435 unsigned maxpages, size_t *_start_offset)
1436 {
1437 unsigned nr, offset;
1438 pgoff_t index, count;
1439 size_t size = maxsize;
1440 loff_t pos;
1441
1442 if (!size || !maxpages)
1443 return 0;
1444
1445 pos = i->xarray_start + i->iov_offset;
1446 index = pos >> PAGE_SHIFT;
1447 offset = pos & ~PAGE_MASK;
1448 *_start_offset = offset;
1449
1450 count = 1;
1451 if (size > PAGE_SIZE - offset) {
1452 size -= PAGE_SIZE - offset;
1453 count += size >> PAGE_SHIFT;
1454 size &= ~PAGE_MASK;
1455 if (size)
1456 count++;
1457 }
1458
1459 if (count > maxpages)
1460 count = maxpages;
1461
1462 nr = iter_xarray_populate_pages(pages, i->xarray, index, count);
1463 if (nr == 0)
1464 return 0;
1465
1466 return min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1467 }
1468
1469 /* must be done on non-empty ITER_IOVEC one */
1470 static unsigned long first_iovec_segment(const struct iov_iter *i,
1471 size_t *size, size_t *start,
1472 size_t maxsize, unsigned maxpages)
1473 {
1474 size_t skip;
1475 long k;
1476
1477 for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) {
1478 unsigned long addr = (unsigned long)i->iov[k].iov_base + skip;
1479 size_t len = i->iov[k].iov_len - skip;
1480
1481 if (unlikely(!len))
1482 continue;
1483 if (len > maxsize)
1484 len = maxsize;
1485 len += (*start = addr % PAGE_SIZE);
1486 if (len > maxpages * PAGE_SIZE)
1487 len = maxpages * PAGE_SIZE;
1488 *size = len;
1489 return addr & PAGE_MASK;
1490 }
1491 BUG(); // if it had been empty, we wouldn't get called
1492 }
1493
1494 /* must be done on non-empty ITER_BVEC one */
1495 static struct page *first_bvec_segment(const struct iov_iter *i,
1496 size_t *size, size_t *start,
1497 size_t maxsize, unsigned maxpages)
1498 {
1499 struct page *page;
1500 size_t skip = i->iov_offset, len;
1501
1502 len = i->bvec->bv_len - skip;
1503 if (len > maxsize)
1504 len = maxsize;
1505 skip += i->bvec->bv_offset;
1506 page = i->bvec->bv_page + skip / PAGE_SIZE;
1507 len += (*start = skip % PAGE_SIZE);
1508 if (len > maxpages * PAGE_SIZE)
1509 len = maxpages * PAGE_SIZE;
1510 *size = len;
1511 return page;
1512 }
1513
1514 ssize_t iov_iter_get_pages(struct iov_iter *i,
1515 struct page **pages, size_t maxsize, unsigned maxpages,
1516 size_t *start)
1517 {
1518 size_t len;
1519 int n, res;
1520
1521 if (maxsize > i->count)
1522 maxsize = i->count;
1523 if (!maxsize)
1524 return 0;
1525
1526 if (likely(iter_is_iovec(i))) {
1527 unsigned int gup_flags = 0;
1528 unsigned long addr;
1529
1530 if (iov_iter_rw(i) != WRITE)
1531 gup_flags |= FOLL_WRITE;
1532 if (i->nofault)
1533 gup_flags |= FOLL_NOFAULT;
1534
1535 addr = first_iovec_segment(i, &len, start, maxsize, maxpages);
1536 n = DIV_ROUND_UP(len, PAGE_SIZE);
1537 res = get_user_pages_fast(addr, n, gup_flags, pages);
1538 if (unlikely(res <= 0))
1539 return res;
1540 return (res == n ? len : res * PAGE_SIZE) - *start;
1541 }
1542 if (iov_iter_is_bvec(i)) {
1543 struct page *page;
1544
1545 page = first_bvec_segment(i, &len, start, maxsize, maxpages);
1546 n = DIV_ROUND_UP(len, PAGE_SIZE);
1547 while (n--)
1548 get_page(*pages++ = page++);
1549 return len - *start;
1550 }
1551 if (iov_iter_is_pipe(i))
1552 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1553 if (iov_iter_is_xarray(i))
1554 return iter_xarray_get_pages(i, pages, maxsize, maxpages, start);
1555 return -EFAULT;
1556 }
1557 EXPORT_SYMBOL(iov_iter_get_pages);
1558
1559 static struct page **get_pages_array(size_t n)
1560 {
1561 return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1562 }
1563
1564 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1565 struct page ***pages, size_t maxsize,
1566 size_t *start)
1567 {
1568 struct page **p;
1569 unsigned int iter_head, npages;
1570 ssize_t n;
1571
1572 if (!sanity(i))
1573 return -EFAULT;
1574
1575 data_start(i, &iter_head, start);
1576 /* Amount of free space: some of this one + all after this one */
1577 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1578 n = npages * PAGE_SIZE - *start;
1579 if (maxsize > n)
1580 maxsize = n;
1581 else
1582 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1583 p = get_pages_array(npages);
1584 if (!p)
1585 return -ENOMEM;
1586 n = __pipe_get_pages(i, maxsize, p, iter_head, start);
1587 if (n > 0)
1588 *pages = p;
1589 else
1590 kvfree(p);
1591 return n;
1592 }
1593
1594 static ssize_t iter_xarray_get_pages_alloc(struct iov_iter *i,
1595 struct page ***pages, size_t maxsize,
1596 size_t *_start_offset)
1597 {
1598 struct page **p;
1599 unsigned nr, offset;
1600 pgoff_t index, count;
1601 size_t size = maxsize;
1602 loff_t pos;
1603
1604 if (!size)
1605 return 0;
1606
1607 pos = i->xarray_start + i->iov_offset;
1608 index = pos >> PAGE_SHIFT;
1609 offset = pos & ~PAGE_MASK;
1610 *_start_offset = offset;
1611
1612 count = 1;
1613 if (size > PAGE_SIZE - offset) {
1614 size -= PAGE_SIZE - offset;
1615 count += size >> PAGE_SHIFT;
1616 size &= ~PAGE_MASK;
1617 if (size)
1618 count++;
1619 }
1620
1621 p = get_pages_array(count);
1622 if (!p)
1623 return -ENOMEM;
1624 *pages = p;
1625
1626 nr = iter_xarray_populate_pages(p, i->xarray, index, count);
1627 if (nr == 0)
1628 return 0;
1629
1630 return min_t(size_t, nr * PAGE_SIZE - offset, maxsize);
1631 }
1632
1633 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1634 struct page ***pages, size_t maxsize,
1635 size_t *start)
1636 {
1637 struct page **p;
1638 size_t len;
1639 int n, res;
1640
1641 if (maxsize > i->count)
1642 maxsize = i->count;
1643 if (!maxsize)
1644 return 0;
1645
1646 if (likely(iter_is_iovec(i))) {
1647 unsigned int gup_flags = 0;
1648 unsigned long addr;
1649
1650 if (iov_iter_rw(i) != WRITE)
1651 gup_flags |= FOLL_WRITE;
1652 if (i->nofault)
1653 gup_flags |= FOLL_NOFAULT;
1654
1655 addr = first_iovec_segment(i, &len, start, maxsize, ~0U);
1656 n = DIV_ROUND_UP(len, PAGE_SIZE);
1657 p = get_pages_array(n);
1658 if (!p)
1659 return -ENOMEM;
1660 res = get_user_pages_fast(addr, n, gup_flags, p);
1661 if (unlikely(res <= 0)) {
1662 kvfree(p);
1663 *pages = NULL;
1664 return res;
1665 }
1666 *pages = p;
1667 return (res == n ? len : res * PAGE_SIZE) - *start;
1668 }
1669 if (iov_iter_is_bvec(i)) {
1670 struct page *page;
1671
1672 page = first_bvec_segment(i, &len, start, maxsize, ~0U);
1673 n = DIV_ROUND_UP(len, PAGE_SIZE);
1674 *pages = p = get_pages_array(n);
1675 if (!p)
1676 return -ENOMEM;
1677 while (n--)
1678 get_page(*p++ = page++);
1679 return len - *start;
1680 }
1681 if (iov_iter_is_pipe(i))
1682 return pipe_get_pages_alloc(i, pages, maxsize, start);
1683 if (iov_iter_is_xarray(i))
1684 return iter_xarray_get_pages_alloc(i, pages, maxsize, start);
1685 return -EFAULT;
1686 }
1687 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1688
1689 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1690 struct iov_iter *i)
1691 {
1692 __wsum sum, next;
1693 sum = *csum;
1694 if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1695 WARN_ON(1);
1696 return 0;
1697 }
1698 iterate_and_advance(i, bytes, base, len, off, ({
1699 next = csum_and_copy_from_user(base, addr + off, len);
1700 sum = csum_block_add(sum, next, off);
1701 next ? 0 : len;
1702 }), ({
1703 sum = csum_and_memcpy(addr + off, base, len, sum, off);
1704 })
1705 )
1706 *csum = sum;
1707 return bytes;
1708 }
1709 EXPORT_SYMBOL(csum_and_copy_from_iter);
1710
1711 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1712 struct iov_iter *i)
1713 {
1714 struct csum_state *csstate = _csstate;
1715 __wsum sum, next;
1716
1717 if (unlikely(iov_iter_is_discard(i))) {
1718 WARN_ON(1); /* for now */
1719 return 0;
1720 }
1721
1722 sum = csum_shift(csstate->csum, csstate->off);
1723 if (unlikely(iov_iter_is_pipe(i)))
1724 bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum);
1725 else iterate_and_advance(i, bytes, base, len, off, ({
1726 next = csum_and_copy_to_user(addr + off, base, len);
1727 sum = csum_block_add(sum, next, off);
1728 next ? 0 : len;
1729 }), ({
1730 sum = csum_and_memcpy(base, addr + off, len, sum, off);
1731 })
1732 )
1733 csstate->csum = csum_shift(sum, csstate->off);
1734 csstate->off += bytes;
1735 return bytes;
1736 }
1737 EXPORT_SYMBOL(csum_and_copy_to_iter);
1738
1739 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1740 struct iov_iter *i)
1741 {
1742 #ifdef CONFIG_CRYPTO_HASH
1743 struct ahash_request *hash = hashp;
1744 struct scatterlist sg;
1745 size_t copied;
1746
1747 copied = copy_to_iter(addr, bytes, i);
1748 sg_init_one(&sg, addr, copied);
1749 ahash_request_set_crypt(hash, &sg, NULL, copied);
1750 crypto_ahash_update(hash);
1751 return copied;
1752 #else
1753 return 0;
1754 #endif
1755 }
1756 EXPORT_SYMBOL(hash_and_copy_to_iter);
1757
1758 static int iov_npages(const struct iov_iter *i, int maxpages)
1759 {
1760 size_t skip = i->iov_offset, size = i->count;
1761 const struct iovec *p;
1762 int npages = 0;
1763
1764 for (p = i->iov; size; skip = 0, p++) {
1765 unsigned offs = offset_in_page(p->iov_base + skip);
1766 size_t len = min(p->iov_len - skip, size);
1767
1768 if (len) {
1769 size -= len;
1770 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1771 if (unlikely(npages > maxpages))
1772 return maxpages;
1773 }
1774 }
1775 return npages;
1776 }
1777
1778 static int bvec_npages(const struct iov_iter *i, int maxpages)
1779 {
1780 size_t skip = i->iov_offset, size = i->count;
1781 const struct bio_vec *p;
1782 int npages = 0;
1783
1784 for (p = i->bvec; size; skip = 0, p++) {
1785 unsigned offs = (p->bv_offset + skip) % PAGE_SIZE;
1786 size_t len = min(p->bv_len - skip, size);
1787
1788 size -= len;
1789 npages += DIV_ROUND_UP(offs + len, PAGE_SIZE);
1790 if (unlikely(npages > maxpages))
1791 return maxpages;
1792 }
1793 return npages;
1794 }
1795
1796 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1797 {
1798 if (unlikely(!i->count))
1799 return 0;
1800 /* iovec and kvec have identical layouts */
1801 if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i)))
1802 return iov_npages(i, maxpages);
1803 if (iov_iter_is_bvec(i))
1804 return bvec_npages(i, maxpages);
1805 if (iov_iter_is_pipe(i)) {
1806 unsigned int iter_head;
1807 int npages;
1808 size_t off;
1809
1810 if (!sanity(i))
1811 return 0;
1812
1813 data_start(i, &iter_head, &off);
1814 /* some of this one + all after this one */
1815 npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1816 return min(npages, maxpages);
1817 }
1818 if (iov_iter_is_xarray(i)) {
1819 unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE;
1820 int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE);
1821 return min(npages, maxpages);
1822 }
1823 return 0;
1824 }
1825 EXPORT_SYMBOL(iov_iter_npages);
1826
1827 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1828 {
1829 *new = *old;
1830 if (unlikely(iov_iter_is_pipe(new))) {
1831 WARN_ON(1);
1832 return NULL;
1833 }
1834 if (unlikely(iov_iter_is_discard(new) || iov_iter_is_xarray(new)))
1835 return NULL;
1836 if (iov_iter_is_bvec(new))
1837 return new->bvec = kmemdup(new->bvec,
1838 new->nr_segs * sizeof(struct bio_vec),
1839 flags);
1840 else
1841 /* iovec and kvec have identical layout */
1842 return new->iov = kmemdup(new->iov,
1843 new->nr_segs * sizeof(struct iovec),
1844 flags);
1845 }
1846 EXPORT_SYMBOL(dup_iter);
1847
1848 static int copy_compat_iovec_from_user(struct iovec *iov,
1849 const struct iovec __user *uvec, unsigned long nr_segs)
1850 {
1851 const struct compat_iovec __user *uiov =
1852 (const struct compat_iovec __user *)uvec;
1853 int ret = -EFAULT, i;
1854
1855 if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1856 return -EFAULT;
1857
1858 for (i = 0; i < nr_segs; i++) {
1859 compat_uptr_t buf;
1860 compat_ssize_t len;
1861
1862 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1863 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1864
1865 /* check for compat_size_t not fitting in compat_ssize_t .. */
1866 if (len < 0) {
1867 ret = -EINVAL;
1868 goto uaccess_end;
1869 }
1870 iov[i].iov_base = compat_ptr(buf);
1871 iov[i].iov_len = len;
1872 }
1873
1874 ret = 0;
1875 uaccess_end:
1876 user_access_end();
1877 return ret;
1878 }
1879
1880 static int copy_iovec_from_user(struct iovec *iov,
1881 const struct iovec __user *uvec, unsigned long nr_segs)
1882 {
1883 unsigned long seg;
1884
1885 if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1886 return -EFAULT;
1887 for (seg = 0; seg < nr_segs; seg++) {
1888 if ((ssize_t)iov[seg].iov_len < 0)
1889 return -EINVAL;
1890 }
1891
1892 return 0;
1893 }
1894
1895 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1896 unsigned long nr_segs, unsigned long fast_segs,
1897 struct iovec *fast_iov, bool compat)
1898 {
1899 struct iovec *iov = fast_iov;
1900 int ret;
1901
1902 /*
1903 * SuS says "The readv() function *may* fail if the iovcnt argument was
1904 * less than or equal to 0, or greater than {IOV_MAX}. Linux has
1905 * traditionally returned zero for zero segments, so...
1906 */
1907 if (nr_segs == 0)
1908 return iov;
1909 if (nr_segs > UIO_MAXIOV)
1910 return ERR_PTR(-EINVAL);
1911 if (nr_segs > fast_segs) {
1912 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1913 if (!iov)
1914 return ERR_PTR(-ENOMEM);
1915 }
1916
1917 if (compat)
1918 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1919 else
1920 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1921 if (ret) {
1922 if (iov != fast_iov)
1923 kfree(iov);
1924 return ERR_PTR(ret);
1925 }
1926
1927 return iov;
1928 }
1929
1930 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1931 unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1932 struct iov_iter *i, bool compat)
1933 {
1934 ssize_t total_len = 0;
1935 unsigned long seg;
1936 struct iovec *iov;
1937
1938 iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1939 if (IS_ERR(iov)) {
1940 *iovp = NULL;
1941 return PTR_ERR(iov);
1942 }
1943
1944 /*
1945 * According to the Single Unix Specification we should return EINVAL if
1946 * an element length is < 0 when cast to ssize_t or if the total length
1947 * would overflow the ssize_t return value of the system call.
1948 *
1949 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1950 * overflow case.
1951 */
1952 for (seg = 0; seg < nr_segs; seg++) {
1953 ssize_t len = (ssize_t)iov[seg].iov_len;
1954
1955 if (!access_ok(iov[seg].iov_base, len)) {
1956 if (iov != *iovp)
1957 kfree(iov);
1958 *iovp = NULL;
1959 return -EFAULT;
1960 }
1961
1962 if (len > MAX_RW_COUNT - total_len) {
1963 len = MAX_RW_COUNT - total_len;
1964 iov[seg].iov_len = len;
1965 }
1966 total_len += len;
1967 }
1968
1969 iov_iter_init(i, type, iov, nr_segs, total_len);
1970 if (iov == *iovp)
1971 *iovp = NULL;
1972 else
1973 *iovp = iov;
1974 return total_len;
1975 }
1976
1977 /**
1978 * import_iovec() - Copy an array of &struct iovec from userspace
1979 * into the kernel, check that it is valid, and initialize a new
1980 * &struct iov_iter iterator to access it.
1981 *
1982 * @type: One of %READ or %WRITE.
1983 * @uvec: Pointer to the userspace array.
1984 * @nr_segs: Number of elements in userspace array.
1985 * @fast_segs: Number of elements in @iov.
1986 * @iovp: (input and output parameter) Pointer to pointer to (usually small
1987 * on-stack) kernel array.
1988 * @i: Pointer to iterator that will be initialized on success.
1989 *
1990 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1991 * then this function places %NULL in *@iov on return. Otherwise, a new
1992 * array will be allocated and the result placed in *@iov. This means that
1993 * the caller may call kfree() on *@iov regardless of whether the small
1994 * on-stack array was used or not (and regardless of whether this function
1995 * returns an error or not).
1996 *
1997 * Return: Negative error code on error, bytes imported on success
1998 */
1999 ssize_t import_iovec(int type, const struct iovec __user *uvec,
2000 unsigned nr_segs, unsigned fast_segs,
2001 struct iovec **iovp, struct iov_iter *i)
2002 {
2003 return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
2004 in_compat_syscall());
2005 }
2006 EXPORT_SYMBOL(import_iovec);
2007
2008 int import_single_range(int rw, void __user *buf, size_t len,
2009 struct iovec *iov, struct iov_iter *i)
2010 {
2011 if (len > MAX_RW_COUNT)
2012 len = MAX_RW_COUNT;
2013 if (unlikely(!access_ok(buf, len)))
2014 return -EFAULT;
2015
2016 iov->iov_base = buf;
2017 iov->iov_len = len;
2018 iov_iter_init(i, rw, iov, 1, len);
2019 return 0;
2020 }
2021 EXPORT_SYMBOL(import_single_range);
2022
2023 /**
2024 * iov_iter_restore() - Restore a &struct iov_iter to the same state as when
2025 * iov_iter_save_state() was called.
2026 *
2027 * @i: &struct iov_iter to restore
2028 * @state: state to restore from
2029 *
2030 * Used after iov_iter_save_state() to bring restore @i, if operations may
2031 * have advanced it.
2032 *
2033 * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC
2034 */
2035 void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state)
2036 {
2037 if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i)) &&
2038 !iov_iter_is_kvec(i))
2039 return;
2040 i->iov_offset = state->iov_offset;
2041 i->count = state->count;
2042 /*
2043 * For the *vec iters, nr_segs + iov is constant - if we increment
2044 * the vec, then we also decrement the nr_segs count. Hence we don't
2045 * need to track both of these, just one is enough and we can deduct
2046 * the other from that. ITER_KVEC and ITER_IOVEC are the same struct
2047 * size, so we can just increment the iov pointer as they are unionzed.
2048 * ITER_BVEC _may_ be the same size on some archs, but on others it is
2049 * not. Be safe and handle it separately.
2050 */
2051 BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
2052 if (iov_iter_is_bvec(i))
2053 i->bvec -= state->nr_segs - i->nr_segs;
2054 else
2055 i->iov -= state->nr_segs - i->nr_segs;
2056 i->nr_segs = state->nr_segs;
2057 }
Ubuntu Jammy Linux kernel mirror