4 * Copyright (C) 1994-2006 Linus Torvalds
8 * The mincore() system call.
10 #include <linux/pagemap.h>
11 #include <linux/gfp.h>
13 #include <linux/mman.h>
14 #include <linux/syscalls.h>
15 #include <linux/swap.h>
16 #include <linux/swapops.h>
17 #include <linux/hugetlb.h>
19 #include <asm/uaccess.h>
20 #include <asm/pgtable.h>
22 static void mincore_hugetlb_page_range(struct vm_area_struct *vma,
23 unsigned long addr, unsigned long nr,
26 #ifdef CONFIG_HUGETLB_PAGE
33 unsigned char present;
36 * Huge pages are always in RAM for now, but
37 * theoretically it needs to be checked.
39 ptep = huge_pte_offset(current->mm,
40 addr & huge_page_mask(h));
41 present = ptep && !huge_pte_none(huge_ptep_get(ptep));
45 /* reach buffer limit */
48 /* check hugepage border */
49 if (!(addr & ~huge_page_mask(h)))
59 * Later we can get more picky about what "in core" means precisely.
60 * For now, simply check to see if the page is in the page cache,
61 * and is up to date; i.e. that no page-in operation would be required
62 * at this time if an application were to map and access this page.
64 static unsigned char mincore_page(struct address_space *mapping, pgoff_t pgoff)
66 unsigned char present = 0;
70 * When tmpfs swaps out a page from a file, any process mapping that
71 * file will not get a swp_entry_t in its pte, but rather it is like
72 * any other file mapping (ie. marked !present and faulted in with
73 * tmpfs's .fault). So swapped out tmpfs mappings are tested here.
75 * However when tmpfs moves the page from pagecache and into swapcache,
76 * it is still in core, but the find_get_page below won't find it.
77 * No big deal, but make a note of it.
79 page = find_get_page(mapping, pgoff);
81 present = PageUptodate(page);
82 page_cache_release(page);
88 static void mincore_unmapped_range(struct vm_area_struct *vma,
89 unsigned long addr, unsigned long nr,
97 pgoff = linear_page_index(vma, addr);
98 for (i = 0; i < nr; i++, pgoff++)
99 vec[i] = mincore_page(vma->vm_file->f_mapping, pgoff);
101 for (i = 0; i < nr; i++)
106 static void mincore_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
107 unsigned long addr, unsigned long nr,
114 ptep = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
115 for (i = 0; i < nr; i++, ptep++, addr += PAGE_SIZE) {
120 mincore_unmapped_range(vma, addr, 1, vec);
121 else if (pte_present(pte))
123 else if (pte_file(pte)) {
124 pgoff = pte_to_pgoff(pte);
125 vec[i] = mincore_page(vma->vm_file->f_mapping, pgoff);
126 } else { /* pte is a swap entry */
127 swp_entry_t entry = pte_to_swp_entry(pte);
129 if (is_migration_entry(entry)) {
130 /* migration entries are always uptodate */
135 vec[i] = mincore_page(&swapper_space, pgoff);
143 pte_unmap_unlock(ptep - 1, ptl);
147 * Do a chunk of "sys_mincore()". We've already checked
148 * all the arguments, we hold the mmap semaphore: we should
149 * just return the amount of info we're asked for.
151 static long do_mincore(unsigned long addr, unsigned long pages, unsigned char *vec)
157 struct vm_area_struct *vma;
159 vma = find_vma(current->mm, addr);
160 if (!vma || addr < vma->vm_start)
163 nr = min(pages, (vma->vm_end - addr) >> PAGE_SHIFT);
165 if (is_vm_hugetlb_page(vma)) {
166 mincore_hugetlb_page_range(vma, addr, nr, vec);
171 * Calculate how many pages there are left in the last level of the
172 * PTE array for our address.
174 nr = min(nr, PTRS_PER_PTE - ((addr >> PAGE_SHIFT) & (PTRS_PER_PTE-1)));
176 pgd = pgd_offset(vma->vm_mm, addr);
177 if (pgd_none_or_clear_bad(pgd))
179 pud = pud_offset(pgd, addr);
180 if (pud_none_or_clear_bad(pud))
182 pmd = pmd_offset(pud, addr);
183 if (pmd_none_or_clear_bad(pmd))
186 mincore_pte_range(vma, pmd, addr, nr, vec);
190 mincore_unmapped_range(vma, addr, nr, vec);
195 * The mincore(2) system call.
197 * mincore() returns the memory residency status of the pages in the
198 * current process's address space specified by [addr, addr + len).
199 * The status is returned in a vector of bytes. The least significant
200 * bit of each byte is 1 if the referenced page is in memory, otherwise
203 * Because the status of a page can change after mincore() checks it
204 * but before it returns to the application, the returned vector may
205 * contain stale information. Only locked pages are guaranteed to
210 * -EFAULT - vec points to an illegal address
211 * -EINVAL - addr is not a multiple of PAGE_CACHE_SIZE
212 * -ENOMEM - Addresses in the range [addr, addr + len] are
213 * invalid for the address space of this process, or
214 * specify one or more pages which are not currently
216 * -EAGAIN - A kernel resource was temporarily unavailable.
218 SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
219 unsigned char __user *, vec)
225 /* Check the start address: needs to be page-aligned.. */
226 if (start & ~PAGE_CACHE_MASK)
229 /* ..and we need to be passed a valid user-space range */
230 if (!access_ok(VERIFY_READ, (void __user *) start, len))
233 /* This also avoids any overflows on PAGE_CACHE_ALIGN */
234 pages = len >> PAGE_SHIFT;
235 pages += (len & ~PAGE_MASK) != 0;
237 if (!access_ok(VERIFY_WRITE, vec, pages))
240 tmp = (void *) __get_free_page(GFP_USER);
247 * Do at most PAGE_SIZE entries per iteration, due to
248 * the temporary buffer size.
250 down_read(¤t->mm->mmap_sem);
251 retval = do_mincore(start, min(pages, PAGE_SIZE), tmp);
252 up_read(¤t->mm->mmap_sem);
256 if (copy_to_user(vec, tmp, retval)) {
262 start += retval << PAGE_SHIFT;
265 free_page((unsigned long) tmp);