1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _LINUX_HIGHMEM_H
3#define _LINUX_HIGHMEM_H
4
5#include <linux/fs.h>
6#include <linux/kernel.h>
7#include <linux/bug.h>
8#include <linux/cacheflush.h>
9#include <linux/kmsan.h>
10#include <linux/mm.h>
11#include <linux/uaccess.h>
12#include <linux/hardirq.h>
13
14#include "highmem-internal.h"
15
16/**
17 * kmap - Map a page for long term usage
18 * @page: Pointer to the page to be mapped
19 *
20 * Returns: The virtual address of the mapping
21 *
22 * Can only be invoked from preemptible task context because on 32bit
23 * systems with CONFIG_HIGHMEM enabled this function might sleep.
24 *
25 * For systems with CONFIG_HIGHMEM=n and for pages in the low memory area
26 * this returns the virtual address of the direct kernel mapping.
27 *
28 * The returned virtual address is globally visible and valid up to the
29 * point where it is unmapped via kunmap(). The pointer can be handed to
30 * other contexts.
31 *
32 * For highmem pages on 32bit systems this can be slow as the mapping space
33 * is limited and protected by a global lock. In case that there is no
34 * mapping slot available the function blocks until a slot is released via
35 * kunmap().
36 */
37static inline void *kmap(struct page *page);
38
39/**
40 * kunmap - Unmap the virtual address mapped by kmap()
41 * @page: Pointer to the page which was mapped by kmap()
42 *
43 * Counterpart to kmap(). A NOOP for CONFIG_HIGHMEM=n and for mappings of
44 * pages in the low memory area.
45 */
46static inline void kunmap(const struct page *page);
47
48/**
49 * kmap_to_page - Get the page for a kmap'ed address
50 * @addr: The address to look up
51 *
52 * Returns: The page which is mapped to @addr.
53 */
54static inline struct page *kmap_to_page(void *addr);
55
56/**
57 * kmap_flush_unused - Flush all unused kmap mappings in order to
58 * remove stray mappings
59 */
60static inline void kmap_flush_unused(void);
61
62/**
63 * kmap_local_page - Map a page for temporary usage
64 * @page: Pointer to the page to be mapped
65 *
66 * Returns: The virtual address of the mapping
67 *
68 * Can be invoked from any context, including interrupts.
69 *
70 * Requires careful handling when nesting multiple mappings because the map
71 * management is stack based. The unmap has to be in the reverse order of
72 * the map operation:
73 *
74 * addr1 = kmap_local_page(page1);
75 * addr2 = kmap_local_page(page2);
76 * ...
77 * kunmap_local(addr2);
78 * kunmap_local(addr1);
79 *
80 * Unmapping addr1 before addr2 is invalid and causes malfunction.
81 *
82 * Contrary to kmap() mappings the mapping is only valid in the context of
83 * the caller and cannot be handed to other contexts.
84 *
85 * On CONFIG_HIGHMEM=n kernels and for low memory pages this returns the
86 * virtual address of the direct mapping. Only real highmem pages are
87 * temporarily mapped.
88 *
89 * While kmap_local_page() is significantly faster than kmap() for the highmem
90 * case it comes with restrictions about the pointer validity.
91 *
92 * On HIGHMEM enabled systems mapping a highmem page has the side effect of
93 * disabling migration in order to keep the virtual address stable across
94 * preemption. No caller of kmap_local_page() can rely on this side effect.
95 */
96static inline void *kmap_local_page(const struct page *page);
97
98/**
99 * kmap_local_folio - Map a page in this folio for temporary usage
100 * @folio: The folio containing the page.
101 * @offset: The byte offset within the folio which identifies the page.
102 *
103 * Requires careful handling when nesting multiple mappings because the map
104 * management is stack based. The unmap has to be in the reverse order of
105 * the map operation::
106 *
107 * addr1 = kmap_local_folio(folio1, offset1);
108 * addr2 = kmap_local_folio(folio2, offset2);
109 * ...
110 * kunmap_local(addr2);
111 * kunmap_local(addr1);
112 *
113 * Unmapping addr1 before addr2 is invalid and causes malfunction.
114 *
115 * Contrary to kmap() mappings the mapping is only valid in the context of
116 * the caller and cannot be handed to other contexts.
117 *
118 * On CONFIG_HIGHMEM=n kernels and for low memory pages this returns the
119 * virtual address of the direct mapping. Only real highmem pages are
120 * temporarily mapped.
121 *
122 * While it is significantly faster than kmap() for the highmem case it
123 * comes with restrictions about the pointer validity.
124 *
125 * On HIGHMEM enabled systems mapping a highmem page has the side effect of
126 * disabling migration in order to keep the virtual address stable across
127 * preemption. No caller of kmap_local_folio() can rely on this side effect.
128 *
129 * Context: Can be invoked from any context.
130 * Return: The virtual address of @offset.
131 */
132static inline void *kmap_local_folio(const struct folio *folio, size_t offset);
133
134/**
135 * kmap_atomic - Atomically map a page for temporary usage - Deprecated!
136 * @page: Pointer to the page to be mapped
137 *
138 * Returns: The virtual address of the mapping
139 *
140 * In fact a wrapper around kmap_local_page() which also disables pagefaults
141 * and, depending on PREEMPT_RT configuration, also CPU migration and
142 * preemption. Therefore users should not count on the latter two side effects.
143 *
144 * Mappings should always be released by kunmap_atomic().
145 *
146 * Do not use in new code. Use kmap_local_page() instead.
147 *
148 * It is used in atomic context when code wants to access the contents of a
149 * page that might be allocated from high memory (see __GFP_HIGHMEM), for
150 * example a page in the pagecache. The API has two functions, and they
151 * can be used in a manner similar to the following::
152 *
153 * // Find the page of interest.
154 * struct page *page = find_get_page(mapping, offset);
155 *
156 * // Gain access to the contents of that page.
157 * void *vaddr = kmap_atomic(page);
158 *
159 * // Do something to the contents of that page.
160 * memset(vaddr, 0, PAGE_SIZE);
161 *
162 * // Unmap that page.
163 * kunmap_atomic(vaddr);
164 *
165 * Note that the kunmap_atomic() call takes the result of the kmap_atomic()
166 * call, not the argument.
167 *
168 * If you need to map two pages because you want to copy from one page to
169 * another you need to keep the kmap_atomic calls strictly nested, like:
170 *
171 * vaddr1 = kmap_atomic(page1);
172 * vaddr2 = kmap_atomic(page2);
173 *
174 * memcpy(vaddr1, vaddr2, PAGE_SIZE);
175 *
176 * kunmap_atomic(vaddr2);
177 * kunmap_atomic(vaddr1);
178 */
179static inline void *kmap_atomic(const struct page *page);
180
181/* Highmem related interfaces for management code */
182static inline unsigned long nr_free_highpages(void);
183static inline unsigned long totalhigh_pages(void);
184
185#ifndef ARCH_HAS_FLUSH_ANON_PAGE
186static inline void flush_anon_page(struct vm_area_struct *vma, struct page *page, unsigned long vmaddr)
187{
188}
189#endif
190
191#ifndef ARCH_IMPLEMENTS_FLUSH_KERNEL_VMAP_RANGE
192static inline void flush_kernel_vmap_range(void *vaddr, int size)
193{
194}
195static inline void invalidate_kernel_vmap_range(void *vaddr, int size)
196{
197}
198#endif
199
200/* when CONFIG_HIGHMEM is not set these will be plain clear/copy_page */
201#ifndef clear_user_highpage
202static inline void clear_user_highpage(struct page *page, unsigned long vaddr)
203{
204 void *addr = kmap_local_page(page);
205 clear_user_page(page: addr, vaddr, pg: page);
206 kunmap_local(addr);
207}
208#endif
209
210#ifndef vma_alloc_zeroed_movable_folio
211/**
212 * vma_alloc_zeroed_movable_folio - Allocate a zeroed page for a VMA.
213 * @vma: The VMA the page is to be allocated for.
214 * @vaddr: The virtual address the page will be inserted into.
215 *
216 * This function will allocate a page suitable for inserting into this
217 * VMA at this virtual address. It may be allocated from highmem or
218 * the movable zone. An architecture may provide its own implementation.
219 *
220 * Return: A folio containing one allocated and zeroed page or NULL if
221 * we are out of memory.
222 */
223static inline
224struct folio *vma_alloc_zeroed_movable_folio(struct vm_area_struct *vma,
225 unsigned long vaddr)
226{
227 struct folio *folio;
228
229 folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, vma, vaddr);
230 if (folio && user_alloc_needs_zeroing())
231 clear_user_highpage(&folio->page, vaddr);
232
233 return folio;
234}
235#endif
236
237static inline void clear_highpage(struct page *page)
238{
239 void *kaddr = kmap_local_page(page);
240 clear_page(page: kaddr);
241 kunmap_local(kaddr);
242}
243
244static inline void clear_highpage_kasan_tagged(struct page *page)
245{
246 void *kaddr = kmap_local_page(page);
247
248 clear_page(page: kasan_reset_tag(addr: kaddr));
249 kunmap_local(kaddr);
250}
251
252#ifndef __HAVE_ARCH_TAG_CLEAR_HIGHPAGE
253
254static inline void tag_clear_highpage(struct page *page)
255{
256}
257
258#endif
259
260/*
261 * If we pass in a base or tail page, we can zero up to PAGE_SIZE.
262 * If we pass in a head page, we can zero up to the size of the compound page.
263 */
264#ifdef CONFIG_HIGHMEM
265void zero_user_segments(struct page *page, unsigned start1, unsigned end1,
266 unsigned start2, unsigned end2);
267#else
268static inline void zero_user_segments(struct page *page,
269 unsigned start1, unsigned end1,
270 unsigned start2, unsigned end2)
271{
272 void *kaddr = kmap_local_page(page);
273 unsigned int i;
274
275 BUG_ON(end1 > page_size(page) || end2 > page_size(page));
276
277 if (end1 > start1)
278 memset(s: kaddr + start1, c: 0, n: end1 - start1);
279
280 if (end2 > start2)
281 memset(s: kaddr + start2, c: 0, n: end2 - start2);
282
283 kunmap_local(kaddr);
284 for (i = 0; i < compound_nr(page); i++)
285 flush_dcache_page(page: page + i);
286}
287#endif
288
289static inline void zero_user_segment(struct page *page,
290 unsigned start, unsigned end)
291{
292 zero_user_segments(page, start1: start, end1: end, start2: 0, end2: 0);
293}
294
295#ifndef __HAVE_ARCH_COPY_USER_HIGHPAGE
296
297static inline void copy_user_highpage(struct page *to, struct page *from,
298 unsigned long vaddr, struct vm_area_struct *vma)
299{
300 char *vfrom, *vto;
301
302 vfrom = kmap_local_page(page: from);
303 vto = kmap_local_page(page: to);
304 copy_user_page(to: vto, from: vfrom, vaddr, topage: to);
305 kmsan_unpoison_memory(page_address(to), PAGE_SIZE);
306 kunmap_local(vto);
307 kunmap_local(vfrom);
308}
309
310#endif
311
312#ifndef __HAVE_ARCH_COPY_HIGHPAGE
313
314static inline void copy_highpage(struct page *to, struct page *from)
315{
316 char *vfrom, *vto;
317
318 vfrom = kmap_local_page(page: from);
319 vto = kmap_local_page(page: to);
320 copy_page(to: vto, from: vfrom);
321 kmsan_copy_page_meta(dst: to, src: from);
322 kunmap_local(vto);
323 kunmap_local(vfrom);
324}
325
326#endif
327
328#ifdef copy_mc_to_kernel
329/*
330 * If architecture supports machine check exception handling, define the
331 * #MC versions of copy_user_highpage and copy_highpage. They copy a memory
332 * page with #MC in source page (@from) handled, and return the number
333 * of bytes not copied if there was a #MC, otherwise 0 for success.
334 */
335static inline int copy_mc_user_highpage(struct page *to, struct page *from,
336 unsigned long vaddr, struct vm_area_struct *vma)
337{
338 unsigned long ret;
339 char *vfrom, *vto;
340
341 vfrom = kmap_local_page(page: from);
342 vto = kmap_local_page(page: to);
343 ret = copy_mc_to_kernel(to: vto, from: vfrom, PAGE_SIZE);
344 if (!ret)
345 kmsan_unpoison_memory(page_address(to), PAGE_SIZE);
346 kunmap_local(vto);
347 kunmap_local(vfrom);
348
349 if (ret)
350 memory_failure_queue(page_to_pfn(from), flags: 0);
351
352 return ret;
353}
354
355static inline int copy_mc_highpage(struct page *to, struct page *from)
356{
357 unsigned long ret;
358 char *vfrom, *vto;
359
360 vfrom = kmap_local_page(page: from);
361 vto = kmap_local_page(page: to);
362 ret = copy_mc_to_kernel(to: vto, from: vfrom, PAGE_SIZE);
363 if (!ret)
364 kmsan_copy_page_meta(dst: to, src: from);
365 kunmap_local(vto);
366 kunmap_local(vfrom);
367
368 if (ret)
369 memory_failure_queue(page_to_pfn(from), flags: 0);
370
371 return ret;
372}
373#else
374static inline int copy_mc_user_highpage(struct page *to, struct page *from,
375 unsigned long vaddr, struct vm_area_struct *vma)
376{
377 copy_user_highpage(to, from, vaddr, vma);
378 return 0;
379}
380
381static inline int copy_mc_highpage(struct page *to, struct page *from)
382{
383 copy_highpage(to, from);
384 return 0;
385}
386#endif
387
388static inline void memcpy_page(struct page *dst_page, size_t dst_off,
389 struct page *src_page, size_t src_off,
390 size_t len)
391{
392 char *dst = kmap_local_page(page: dst_page);
393 char *src = kmap_local_page(page: src_page);
394
395 VM_BUG_ON(dst_off + len > PAGE_SIZE || src_off + len > PAGE_SIZE);
396 memcpy(to: dst + dst_off, from: src + src_off, len);
397 kunmap_local(src);
398 kunmap_local(dst);
399}
400
401static inline void memcpy_folio(struct folio *dst_folio, size_t dst_off,
402 struct folio *src_folio, size_t src_off, size_t len)
403{
404 VM_BUG_ON(dst_off + len > folio_size(dst_folio));
405 VM_BUG_ON(src_off + len > folio_size(src_folio));
406
407 do {
408 char *dst = kmap_local_folio(folio: dst_folio, offset: dst_off);
409 const char *src = kmap_local_folio(folio: src_folio, offset: src_off);
410 size_t chunk = len;
411
412 if (folio_test_highmem(folio: dst_folio) &&
413 chunk > PAGE_SIZE - offset_in_page(dst_off))
414 chunk = PAGE_SIZE - offset_in_page(dst_off);
415 if (folio_test_highmem(folio: src_folio) &&
416 chunk > PAGE_SIZE - offset_in_page(src_off))
417 chunk = PAGE_SIZE - offset_in_page(src_off);
418 memcpy(to: dst, from: src, len: chunk);
419 kunmap_local(src);
420 kunmap_local(dst);
421
422 dst_off += chunk;
423 src_off += chunk;
424 len -= chunk;
425 } while (len > 0);
426}
427
428static inline void memset_page(struct page *page, size_t offset, int val,
429 size_t len)
430{
431 char *addr = kmap_local_page(page);
432
433 VM_BUG_ON(offset + len > PAGE_SIZE);
434 memset(s: addr + offset, c: val, n: len);
435 kunmap_local(addr);
436}
437
438static inline void memcpy_from_page(char *to, struct page *page,
439 size_t offset, size_t len)
440{
441 char *from = kmap_local_page(page);
442
443 VM_BUG_ON(offset + len > PAGE_SIZE);
444 memcpy(to, from: from + offset, len);
445 kunmap_local(from);
446}
447
448static inline void memcpy_to_page(struct page *page, size_t offset,
449 const char *from, size_t len)
450{
451 char *to = kmap_local_page(page);
452
453 VM_BUG_ON(offset + len > PAGE_SIZE);
454 memcpy(to: to + offset, from, len);
455 flush_dcache_page(page);
456 kunmap_local(to);
457}
458
459static inline void memzero_page(struct page *page, size_t offset, size_t len)
460{
461 char *addr = kmap_local_page(page);
462
463 VM_BUG_ON(offset + len > PAGE_SIZE);
464 memset(s: addr + offset, c: 0, n: len);
465 flush_dcache_page(page);
466 kunmap_local(addr);
467}
468
469/**
470 * memcpy_from_folio - Copy a range of bytes from a folio.
471 * @to: The memory to copy to.
472 * @folio: The folio to read from.
473 * @offset: The first byte in the folio to read.
474 * @len: The number of bytes to copy.
475 */
476static inline void memcpy_from_folio(char *to, struct folio *folio,
477 size_t offset, size_t len)
478{
479 VM_BUG_ON(offset + len > folio_size(folio));
480
481 do {
482 const char *from = kmap_local_folio(folio, offset);
483 size_t chunk = len;
484
485 if (folio_test_partial_kmap(folio) &&
486 chunk > PAGE_SIZE - offset_in_page(offset))
487 chunk = PAGE_SIZE - offset_in_page(offset);
488 memcpy(to, from, len: chunk);
489 kunmap_local(from);
490
491 to += chunk;
492 offset += chunk;
493 len -= chunk;
494 } while (len > 0);
495}
496
497/**
498 * memcpy_to_folio - Copy a range of bytes to a folio.
499 * @folio: The folio to write to.
500 * @offset: The first byte in the folio to store to.
501 * @from: The memory to copy from.
502 * @len: The number of bytes to copy.
503 */
504static inline void memcpy_to_folio(struct folio *folio, size_t offset,
505 const char *from, size_t len)
506{
507 VM_BUG_ON(offset + len > folio_size(folio));
508
509 do {
510 char *to = kmap_local_folio(folio, offset);
511 size_t chunk = len;
512
513 if (folio_test_partial_kmap(folio) &&
514 chunk > PAGE_SIZE - offset_in_page(offset))
515 chunk = PAGE_SIZE - offset_in_page(offset);
516 memcpy(to, from, len: chunk);
517 kunmap_local(to);
518
519 from += chunk;
520 offset += chunk;
521 len -= chunk;
522 } while (len > 0);
523
524 flush_dcache_folio(folio);
525}
526
527/**
528 * folio_zero_tail - Zero the tail of a folio.
529 * @folio: The folio to zero.
530 * @offset: The byte offset in the folio to start zeroing at.
531 * @kaddr: The address the folio is currently mapped to.
532 *
533 * If you have already used kmap_local_folio() to map a folio, written
534 * some data to it and now need to zero the end of the folio (and flush
535 * the dcache), you can use this function. If you do not have the
536 * folio kmapped (eg the folio has been partially populated by DMA),
537 * use folio_zero_range() or folio_zero_segment() instead.
538 *
539 * Return: An address which can be passed to kunmap_local().
540 */
541static inline __must_check void *folio_zero_tail(struct folio *folio,
542 size_t offset, void *kaddr)
543{
544 size_t len = folio_size(folio) - offset;
545
546 if (folio_test_partial_kmap(folio)) {
547 size_t max = PAGE_SIZE - offset_in_page(offset);
548
549 while (len > max) {
550 memset(s: kaddr, c: 0, n: max);
551 kunmap_local(kaddr);
552 len -= max;
553 offset += max;
554 max = PAGE_SIZE;
555 kaddr = kmap_local_folio(folio, offset);
556 }
557 }
558
559 memset(s: kaddr, c: 0, n: len);
560 flush_dcache_folio(folio);
561
562 return kaddr;
563}
564
565/**
566 * folio_fill_tail - Copy some data to a folio and pad with zeroes.
567 * @folio: The destination folio.
568 * @offset: The offset into @folio at which to start copying.
569 * @from: The data to copy.
570 * @len: How many bytes of data to copy.
571 *
572 * This function is most useful for filesystems which support inline data.
573 * When they want to copy data from the inode into the page cache, this
574 * function does everything for them. It supports large folios even on
575 * HIGHMEM configurations.
576 */
577static inline void folio_fill_tail(struct folio *folio, size_t offset,
578 const char *from, size_t len)
579{
580 char *to = kmap_local_folio(folio, offset);
581
582 VM_BUG_ON(offset + len > folio_size(folio));
583
584 if (folio_test_partial_kmap(folio)) {
585 size_t max = PAGE_SIZE - offset_in_page(offset);
586
587 while (len > max) {
588 memcpy(to, from, len: max);
589 kunmap_local(to);
590 len -= max;
591 from += max;
592 offset += max;
593 max = PAGE_SIZE;
594 to = kmap_local_folio(folio, offset);
595 }
596 }
597
598 memcpy(to, from, len);
599 to = folio_zero_tail(folio, offset: offset + len, kaddr: to + len);
600 kunmap_local(to);
601}
602
603/**
604 * memcpy_from_file_folio - Copy some bytes from a file folio.
605 * @to: The destination buffer.
606 * @folio: The folio to copy from.
607 * @pos: The position in the file.
608 * @len: The maximum number of bytes to copy.
609 *
610 * Copy up to @len bytes from this folio. This may be limited by PAGE_SIZE
611 * if the folio comes from HIGHMEM, and by the size of the folio.
612 *
613 * Return: The number of bytes copied from the folio.
614 */
615static inline size_t memcpy_from_file_folio(char *to, struct folio *folio,
616 loff_t pos, size_t len)
617{
618 size_t offset = offset_in_folio(folio, pos);
619 char *from = kmap_local_folio(folio, offset);
620
621 if (folio_test_partial_kmap(folio)) {
622 offset = offset_in_page(offset);
623 len = min_t(size_t, len, PAGE_SIZE - offset);
624 } else
625 len = min(len, folio_size(folio) - offset);
626
627 memcpy(to, from, len);
628 kunmap_local(from);
629
630 return len;
631}
632
633/**
634 * folio_zero_segments() - Zero two byte ranges in a folio.
635 * @folio: The folio to write to.
636 * @start1: The first byte to zero.
637 * @xend1: One more than the last byte in the first range.
638 * @start2: The first byte to zero in the second range.
639 * @xend2: One more than the last byte in the second range.
640 */
641static inline void folio_zero_segments(struct folio *folio,
642 size_t start1, size_t xend1, size_t start2, size_t xend2)
643{
644 zero_user_segments(page: &folio->page, start1, end1: xend1, start2, end2: xend2);
645}
646
647/**
648 * folio_zero_segment() - Zero a byte range in a folio.
649 * @folio: The folio to write to.
650 * @start: The first byte to zero.
651 * @xend: One more than the last byte to zero.
652 */
653static inline void folio_zero_segment(struct folio *folio,
654 size_t start, size_t xend)
655{
656 zero_user_segments(page: &folio->page, start1: start, end1: xend, start2: 0, end2: 0);
657}
658
659/**
660 * folio_zero_range() - Zero a byte range in a folio.
661 * @folio: The folio to write to.
662 * @start: The first byte to zero.
663 * @length: The number of bytes to zero.
664 */
665static inline void folio_zero_range(struct folio *folio,
666 size_t start, size_t length)
667{
668 zero_user_segments(page: &folio->page, start1: start, end1: start + length, start2: 0, end2: 0);
669}
670
671/**
672 * folio_release_kmap - Unmap a folio and drop a refcount.
673 * @folio: The folio to release.
674 * @addr: The address previously returned by a call to kmap_local_folio().
675 *
676 * It is common, eg in directory handling to kmap a folio. This function
677 * unmaps the folio and drops the refcount that was being held to keep the
678 * folio alive while we accessed it.
679 */
680static inline void folio_release_kmap(struct folio *folio, void *addr)
681{
682 kunmap_local(addr);
683 folio_put(folio);
684}
685#endif /* _LINUX_HIGHMEM_H */
686