1 #include <linux/kernel.h>
2 #include <linux/errno.h>
4 #include <linux/spinlock.h>
6 #include <linux/hugetlb.h>
8 #include <linux/pagemap.h>
9 #include <linux/rmap.h>
10 #include <linux/swap.h>
11 #include <linux/swapops.h>
15 static struct page
*no_page_table(struct vm_area_struct
*vma
,
19 * When core dumping an enormous anonymous area that nobody
20 * has touched so far, we don't want to allocate unnecessary pages or
21 * page tables. Return error instead of NULL to skip handle_mm_fault,
22 * then get_dump_page() will return NULL to leave a hole in the dump.
23 * But we can only make this optimization where a hole would surely
24 * be zero-filled if handle_mm_fault() actually did handle it.
26 if ((flags
& FOLL_DUMP
) && (!vma
->vm_ops
|| !vma
->vm_ops
->fault
))
27 return ERR_PTR(-EFAULT
);
31 static struct page
*follow_page_pte(struct vm_area_struct
*vma
,
32 unsigned long address
, pmd_t
*pmd
, unsigned int flags
)
34 struct mm_struct
*mm
= vma
->vm_mm
;
40 if (unlikely(pmd_bad(*pmd
)))
41 return no_page_table(vma
, flags
);
43 ptep
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
45 if (!pte_present(pte
)) {
48 * KSM's break_ksm() relies upon recognizing a ksm page
49 * even while it is being migrated, so for that case we
50 * need migration_entry_wait().
52 if (likely(!(flags
& FOLL_MIGRATION
)))
54 if (pte_none(pte
) || pte_file(pte
))
56 entry
= pte_to_swp_entry(pte
);
57 if (!is_migration_entry(entry
))
59 pte_unmap_unlock(ptep
, ptl
);
60 migration_entry_wait(mm
, pmd
, address
);
63 if ((flags
& FOLL_NUMA
) && pte_numa(pte
))
65 if ((flags
& FOLL_WRITE
) && !pte_write(pte
)) {
66 pte_unmap_unlock(ptep
, ptl
);
70 page
= vm_normal_page(vma
, address
, pte
);
71 if (unlikely(!page
)) {
72 if ((flags
& FOLL_DUMP
) ||
73 !is_zero_pfn(pte_pfn(pte
)))
80 if (flags
& FOLL_TOUCH
) {
81 if ((flags
& FOLL_WRITE
) &&
82 !pte_dirty(pte
) && !PageDirty(page
))
85 * pte_mkyoung() would be more correct here, but atomic care
86 * is needed to avoid losing the dirty bit: it is easier to use
87 * mark_page_accessed().
89 mark_page_accessed(page
);
91 if ((flags
& FOLL_MLOCK
) && (vma
->vm_flags
& VM_LOCKED
)) {
93 * The preliminary mapping check is mainly to avoid the
94 * pointless overhead of lock_page on the ZERO_PAGE
95 * which might bounce very badly if there is contention.
97 * If the page is already locked, we don't need to
98 * handle it now - vmscan will handle it later if and
99 * when it attempts to reclaim the page.
101 if (page
->mapping
&& trylock_page(page
)) {
102 lru_add_drain(); /* push cached pages to LRU */
104 * Because we lock page here, and migration is
105 * blocked by the pte's page reference, and we
106 * know the page is still mapped, we don't even
107 * need to check for file-cache page truncation.
109 mlock_vma_page(page
);
113 pte_unmap_unlock(ptep
, ptl
);
116 pte_unmap_unlock(ptep
, ptl
);
117 return ERR_PTR(-EFAULT
);
120 pte_unmap_unlock(ptep
, ptl
);
123 return no_page_table(vma
, flags
);
127 * follow_page_mask - look up a page descriptor from a user-virtual address
128 * @vma: vm_area_struct mapping @address
129 * @address: virtual address to look up
130 * @flags: flags modifying lookup behaviour
131 * @page_mask: on output, *page_mask is set according to the size of the page
133 * @flags can have FOLL_ flags set, defined in <linux/mm.h>
135 * Returns the mapped (struct page *), %NULL if no mapping exists, or
136 * an error pointer if there is a mapping to something not represented
137 * by a page descriptor (see also vm_normal_page()).
139 struct page
*follow_page_mask(struct vm_area_struct
*vma
,
140 unsigned long address
, unsigned int flags
,
141 unsigned int *page_mask
)
148 struct mm_struct
*mm
= vma
->vm_mm
;
152 page
= follow_huge_addr(mm
, address
, flags
& FOLL_WRITE
);
154 BUG_ON(flags
& FOLL_GET
);
158 pgd
= pgd_offset(mm
, address
);
159 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
160 return no_page_table(vma
, flags
);
162 pud
= pud_offset(pgd
, address
);
164 return no_page_table(vma
, flags
);
165 if (pud_huge(*pud
) && vma
->vm_flags
& VM_HUGETLB
) {
166 if (flags
& FOLL_GET
)
168 page
= follow_huge_pud(mm
, address
, pud
, flags
& FOLL_WRITE
);
171 if (unlikely(pud_bad(*pud
)))
172 return no_page_table(vma
, flags
);
174 pmd
= pmd_offset(pud
, address
);
176 return no_page_table(vma
, flags
);
177 if (pmd_huge(*pmd
) && vma
->vm_flags
& VM_HUGETLB
) {
178 page
= follow_huge_pmd(mm
, address
, pmd
, flags
& FOLL_WRITE
);
179 if (flags
& FOLL_GET
) {
181 * Refcount on tail pages are not well-defined and
182 * shouldn't be taken. The caller should handle a NULL
183 * return when trying to follow tail pages.
192 if ((flags
& FOLL_NUMA
) && pmd_numa(*pmd
))
193 return no_page_table(vma
, flags
);
194 if (pmd_trans_huge(*pmd
)) {
195 if (flags
& FOLL_SPLIT
) {
196 split_huge_page_pmd(vma
, address
, pmd
);
197 return follow_page_pte(vma
, address
, pmd
, flags
);
199 ptl
= pmd_lock(mm
, pmd
);
200 if (likely(pmd_trans_huge(*pmd
))) {
201 if (unlikely(pmd_trans_splitting(*pmd
))) {
203 wait_split_huge_page(vma
->anon_vma
, pmd
);
205 page
= follow_trans_huge_pmd(vma
, address
,
208 *page_mask
= HPAGE_PMD_NR
- 1;
214 return follow_page_pte(vma
, address
, pmd
, flags
);
217 static int get_gate_page(struct mm_struct
*mm
, unsigned long address
,
218 unsigned int gup_flags
, struct vm_area_struct
**vma
,
227 /* user gate pages are read-only */
228 if (gup_flags
& FOLL_WRITE
)
230 if (address
> TASK_SIZE
)
231 pgd
= pgd_offset_k(address
);
233 pgd
= pgd_offset_gate(mm
, address
);
234 BUG_ON(pgd_none(*pgd
));
235 pud
= pud_offset(pgd
, address
);
236 BUG_ON(pud_none(*pud
));
237 pmd
= pmd_offset(pud
, address
);
240 VM_BUG_ON(pmd_trans_huge(*pmd
));
241 pte
= pte_offset_map(pmd
, address
);
244 *vma
= get_gate_vma(mm
);
247 *page
= vm_normal_page(*vma
, address
, *pte
);
249 if ((gup_flags
& FOLL_DUMP
) || !is_zero_pfn(pte_pfn(*pte
)))
251 *page
= pte_page(*pte
);
262 * mmap_sem must be held on entry. If @nonblocking != NULL and
263 * *@flags does not include FOLL_NOWAIT, the mmap_sem may be released.
264 * If it is, *@nonblocking will be set to 0 and -EBUSY returned.
266 static int faultin_page(struct task_struct
*tsk
, struct vm_area_struct
*vma
,
267 unsigned long address
, unsigned int *flags
, int *nonblocking
)
269 struct mm_struct
*mm
= vma
->vm_mm
;
270 unsigned int fault_flags
= 0;
273 /* For mlock, just skip the stack guard page. */
274 if ((*flags
& FOLL_MLOCK
) &&
275 (stack_guard_page_start(vma
, address
) ||
276 stack_guard_page_end(vma
, address
+ PAGE_SIZE
)))
278 if (*flags
& FOLL_WRITE
)
279 fault_flags
|= FAULT_FLAG_WRITE
;
281 fault_flags
|= FAULT_FLAG_ALLOW_RETRY
;
282 if (*flags
& FOLL_NOWAIT
)
283 fault_flags
|= FAULT_FLAG_ALLOW_RETRY
| FAULT_FLAG_RETRY_NOWAIT
;
284 if (*flags
& FOLL_TRIED
) {
285 VM_WARN_ON_ONCE(fault_flags
& FAULT_FLAG_ALLOW_RETRY
);
286 fault_flags
|= FAULT_FLAG_TRIED
;
289 ret
= handle_mm_fault(mm
, vma
, address
, fault_flags
);
290 if (ret
& VM_FAULT_ERROR
) {
291 if (ret
& VM_FAULT_OOM
)
293 if (ret
& (VM_FAULT_HWPOISON
| VM_FAULT_HWPOISON_LARGE
))
294 return *flags
& FOLL_HWPOISON
? -EHWPOISON
: -EFAULT
;
295 if (ret
& VM_FAULT_SIGBUS
)
301 if (ret
& VM_FAULT_MAJOR
)
307 if (ret
& VM_FAULT_RETRY
) {
314 * The VM_FAULT_WRITE bit tells us that do_wp_page has broken COW when
315 * necessary, even if maybe_mkwrite decided not to set pte_write. We
316 * can thus safely do subsequent page lookups as if they were reads.
317 * But only do so when looping for pte_write is futile: in some cases
318 * userspace may also be wanting to write to the gotten user page,
319 * which a read fault here might prevent (a readonly page might get
320 * reCOWed by userspace write).
322 if ((ret
& VM_FAULT_WRITE
) && !(vma
->vm_flags
& VM_WRITE
))
323 *flags
&= ~FOLL_WRITE
;
327 static int check_vma_flags(struct vm_area_struct
*vma
, unsigned long gup_flags
)
329 vm_flags_t vm_flags
= vma
->vm_flags
;
331 if (vm_flags
& (VM_IO
| VM_PFNMAP
))
334 if (gup_flags
& FOLL_WRITE
) {
335 if (!(vm_flags
& VM_WRITE
)) {
336 if (!(gup_flags
& FOLL_FORCE
))
339 * We used to let the write,force case do COW in a
340 * VM_MAYWRITE VM_SHARED !VM_WRITE vma, so ptrace could
341 * set a breakpoint in a read-only mapping of an
342 * executable, without corrupting the file (yet only
343 * when that file had been opened for writing!).
344 * Anon pages in shared mappings are surprising: now
347 if (!is_cow_mapping(vm_flags
)) {
348 WARN_ON_ONCE(vm_flags
& VM_MAYWRITE
);
352 } else if (!(vm_flags
& VM_READ
)) {
353 if (!(gup_flags
& FOLL_FORCE
))
356 * Is there actually any vma we can reach here which does not
357 * have VM_MAYREAD set?
359 if (!(vm_flags
& VM_MAYREAD
))
366 * __get_user_pages() - pin user pages in memory
367 * @tsk: task_struct of target task
368 * @mm: mm_struct of target mm
369 * @start: starting user address
370 * @nr_pages: number of pages from start to pin
371 * @gup_flags: flags modifying pin behaviour
372 * @pages: array that receives pointers to the pages pinned.
373 * Should be at least nr_pages long. Or NULL, if caller
374 * only intends to ensure the pages are faulted in.
375 * @vmas: array of pointers to vmas corresponding to each page.
376 * Or NULL if the caller does not require them.
377 * @nonblocking: whether waiting for disk IO or mmap_sem contention
379 * Returns number of pages pinned. This may be fewer than the number
380 * requested. If nr_pages is 0 or negative, returns 0. If no pages
381 * were pinned, returns -errno. Each page returned must be released
382 * with a put_page() call when it is finished with. vmas will only
383 * remain valid while mmap_sem is held.
385 * Must be called with mmap_sem held. It may be released. See below.
387 * __get_user_pages walks a process's page tables and takes a reference to
388 * each struct page that each user address corresponds to at a given
389 * instant. That is, it takes the page that would be accessed if a user
390 * thread accesses the given user virtual address at that instant.
392 * This does not guarantee that the page exists in the user mappings when
393 * __get_user_pages returns, and there may even be a completely different
394 * page there in some cases (eg. if mmapped pagecache has been invalidated
395 * and subsequently re faulted). However it does guarantee that the page
396 * won't be freed completely. And mostly callers simply care that the page
397 * contains data that was valid *at some point in time*. Typically, an IO
398 * or similar operation cannot guarantee anything stronger anyway because
399 * locks can't be held over the syscall boundary.
401 * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If
402 * the page is written to, set_page_dirty (or set_page_dirty_lock, as
403 * appropriate) must be called after the page is finished with, and
404 * before put_page is called.
406 * If @nonblocking != NULL, __get_user_pages will not wait for disk IO
407 * or mmap_sem contention, and if waiting is needed to pin all pages,
408 * *@nonblocking will be set to 0. Further, if @gup_flags does not
409 * include FOLL_NOWAIT, the mmap_sem will be released via up_read() in
412 * A caller using such a combination of @nonblocking and @gup_flags
413 * must therefore hold the mmap_sem for reading only, and recognize
414 * when it's been released. Otherwise, it must be held for either
415 * reading or writing and will not be released.
417 * In most cases, get_user_pages or get_user_pages_fast should be used
418 * instead of __get_user_pages. __get_user_pages should be used only if
419 * you need some special @gup_flags.
421 long __get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
422 unsigned long start
, unsigned long nr_pages
,
423 unsigned int gup_flags
, struct page
**pages
,
424 struct vm_area_struct
**vmas
, int *nonblocking
)
427 unsigned int page_mask
;
428 struct vm_area_struct
*vma
= NULL
;
433 VM_BUG_ON(!!pages
!= !!(gup_flags
& FOLL_GET
));
436 * If FOLL_FORCE is set then do not force a full fault as the hinting
437 * fault information is unrelated to the reference behaviour of a task
438 * using the address space
440 if (!(gup_flags
& FOLL_FORCE
))
441 gup_flags
|= FOLL_NUMA
;
445 unsigned int foll_flags
= gup_flags
;
446 unsigned int page_increm
;
448 /* first iteration or cross vma bound */
449 if (!vma
|| start
>= vma
->vm_end
) {
450 vma
= find_extend_vma(mm
, start
);
451 if (!vma
&& in_gate_area(mm
, start
)) {
453 ret
= get_gate_page(mm
, start
& PAGE_MASK
,
455 pages
? &pages
[i
] : NULL
);
462 if (!vma
|| check_vma_flags(vma
, gup_flags
))
463 return i
? : -EFAULT
;
464 if (is_vm_hugetlb_page(vma
)) {
465 i
= follow_hugetlb_page(mm
, vma
, pages
, vmas
,
466 &start
, &nr_pages
, i
,
473 * If we have a pending SIGKILL, don't keep faulting pages and
474 * potentially allocating memory.
476 if (unlikely(fatal_signal_pending(current
)))
477 return i
? i
: -ERESTARTSYS
;
479 page
= follow_page_mask(vma
, start
, foll_flags
, &page_mask
);
482 ret
= faultin_page(tsk
, vma
, start
, &foll_flags
,
499 return i
? i
: PTR_ERR(page
);
502 flush_anon_page(vma
, page
, start
);
503 flush_dcache_page(page
);
511 page_increm
= 1 + (~(start
>> PAGE_SHIFT
) & page_mask
);
512 if (page_increm
> nr_pages
)
513 page_increm
= nr_pages
;
515 start
+= page_increm
* PAGE_SIZE
;
516 nr_pages
-= page_increm
;
520 EXPORT_SYMBOL(__get_user_pages
);
523 * fixup_user_fault() - manually resolve a user page fault
524 * @tsk: the task_struct to use for page fault accounting, or
525 * NULL if faults are not to be recorded.
526 * @mm: mm_struct of target mm
527 * @address: user address
528 * @fault_flags:flags to pass down to handle_mm_fault()
530 * This is meant to be called in the specific scenario where for locking reasons
531 * we try to access user memory in atomic context (within a pagefault_disable()
532 * section), this returns -EFAULT, and we want to resolve the user fault before
535 * Typically this is meant to be used by the futex code.
537 * The main difference with get_user_pages() is that this function will
538 * unconditionally call handle_mm_fault() which will in turn perform all the
539 * necessary SW fixup of the dirty and young bits in the PTE, while
540 * handle_mm_fault() only guarantees to update these in the struct page.
542 * This is important for some architectures where those bits also gate the
543 * access permission to the page because they are maintained in software. On
544 * such architectures, gup() will not be enough to make a subsequent access
547 * This has the same semantics wrt the @mm->mmap_sem as does filemap_fault().
549 int fixup_user_fault(struct task_struct
*tsk
, struct mm_struct
*mm
,
550 unsigned long address
, unsigned int fault_flags
)
552 struct vm_area_struct
*vma
;
556 vma
= find_extend_vma(mm
, address
);
557 if (!vma
|| address
< vma
->vm_start
)
560 vm_flags
= (fault_flags
& FAULT_FLAG_WRITE
) ? VM_WRITE
: VM_READ
;
561 if (!(vm_flags
& vma
->vm_flags
))
564 ret
= handle_mm_fault(mm
, vma
, address
, fault_flags
);
565 if (ret
& VM_FAULT_ERROR
) {
566 if (ret
& VM_FAULT_OOM
)
568 if (ret
& (VM_FAULT_HWPOISON
| VM_FAULT_HWPOISON_LARGE
))
570 if (ret
& VM_FAULT_SIGBUS
)
575 if (ret
& VM_FAULT_MAJOR
)
584 * get_user_pages() - pin user pages in memory
585 * @tsk: the task_struct to use for page fault accounting, or
586 * NULL if faults are not to be recorded.
587 * @mm: mm_struct of target mm
588 * @start: starting user address
589 * @nr_pages: number of pages from start to pin
590 * @write: whether pages will be written to by the caller
591 * @force: whether to force access even when user mapping is currently
592 * protected (but never forces write access to shared mapping).
593 * @pages: array that receives pointers to the pages pinned.
594 * Should be at least nr_pages long. Or NULL, if caller
595 * only intends to ensure the pages are faulted in.
596 * @vmas: array of pointers to vmas corresponding to each page.
597 * Or NULL if the caller does not require them.
599 * Returns number of pages pinned. This may be fewer than the number
600 * requested. If nr_pages is 0 or negative, returns 0. If no pages
601 * were pinned, returns -errno. Each page returned must be released
602 * with a put_page() call when it is finished with. vmas will only
603 * remain valid while mmap_sem is held.
605 * Must be called with mmap_sem held for read or write.
607 * get_user_pages walks a process's page tables and takes a reference to
608 * each struct page that each user address corresponds to at a given
609 * instant. That is, it takes the page that would be accessed if a user
610 * thread accesses the given user virtual address at that instant.
612 * This does not guarantee that the page exists in the user mappings when
613 * get_user_pages returns, and there may even be a completely different
614 * page there in some cases (eg. if mmapped pagecache has been invalidated
615 * and subsequently re faulted). However it does guarantee that the page
616 * won't be freed completely. And mostly callers simply care that the page
617 * contains data that was valid *at some point in time*. Typically, an IO
618 * or similar operation cannot guarantee anything stronger anyway because
619 * locks can't be held over the syscall boundary.
621 * If write=0, the page must not be written to. If the page is written to,
622 * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called
623 * after the page is finished with, and before put_page is called.
625 * get_user_pages is typically used for fewer-copy IO operations, to get a
626 * handle on the memory by some means other than accesses via the user virtual
627 * addresses. The pages may be submitted for DMA to devices or accessed via
628 * their kernel linear mapping (via the kmap APIs). Care should be taken to
629 * use the correct cache flushing APIs.
631 * See also get_user_pages_fast, for performance critical applications.
633 long get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
634 unsigned long start
, unsigned long nr_pages
, int write
,
635 int force
, struct page
**pages
, struct vm_area_struct
**vmas
)
637 int flags
= FOLL_TOUCH
;
646 return __get_user_pages(tsk
, mm
, start
, nr_pages
, flags
, pages
, vmas
,
649 EXPORT_SYMBOL(get_user_pages
);
652 * get_dump_page() - pin user page in memory while writing it to core dump
653 * @addr: user address
655 * Returns struct page pointer of user page pinned for dump,
656 * to be freed afterwards by page_cache_release() or put_page().
658 * Returns NULL on any kind of failure - a hole must then be inserted into
659 * the corefile, to preserve alignment with its headers; and also returns
660 * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -
661 * allowing a hole to be left in the corefile to save diskspace.
663 * Called without mmap_sem, but after all other threads have been killed.
665 #ifdef CONFIG_ELF_CORE
666 struct page
*get_dump_page(unsigned long addr
)
668 struct vm_area_struct
*vma
;
671 if (__get_user_pages(current
, current
->mm
, addr
, 1,
672 FOLL_FORCE
| FOLL_DUMP
| FOLL_GET
, &page
, &vma
,
675 flush_cache_page(vma
, addr
, page_to_pfn(page
));
678 #endif /* CONFIG_ELF_CORE */
This page took 0.070019 seconds and 6 git commands to generate.