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 end,
26 #ifdef CONFIG_HUGETLB_PAGE
31 unsigned char present;
34 * Huge pages are always in RAM for now, but
35 * theoretically it needs to be checked.
37 ptep = huge_pte_offset(current->mm,
38 addr & huge_page_mask(h));
39 present = ptep && !huge_pte_none(huge_ptep_get(ptep));
46 /* check hugepage border */
47 if (!(addr & ~huge_page_mask(h)))
57 * Later we can get more picky about what "in core" means precisely.
58 * For now, simply check to see if the page is in the page cache,
59 * and is up to date; i.e. that no page-in operation would be required
60 * at this time if an application were to map and access this page.
62 static unsigned char mincore_page(struct address_space *mapping, pgoff_t pgoff)
64 unsigned char present = 0;
68 * When tmpfs swaps out a page from a file, any process mapping that
69 * file will not get a swp_entry_t in its pte, but rather it is like
70 * any other file mapping (ie. marked !present and faulted in with
71 * tmpfs's .fault). So swapped out tmpfs mappings are tested here.
73 page = find_get_page(mapping, pgoff);
75 /* shmem/tmpfs may return swap: account for swapcache page too. */
76 if (radix_tree_exceptional_entry(page)) {
77 swp_entry_t swap = radix_to_swp_entry(page);
78 page = find_get_page(swap_address_space(swap), swap.val);
82 present = PageUptodate(page);
83 page_cache_release(page);
89 static void mincore_unmapped_range(struct vm_area_struct *vma,
90 unsigned long addr, unsigned long end,
93 unsigned long nr = (end - addr) >> PAGE_SHIFT;
99 pgoff = linear_page_index(vma, addr);
100 for (i = 0; i < nr; i++, pgoff++)
101 vec[i] = mincore_page(vma->vm_file->f_mapping, pgoff);
103 for (i = 0; i < nr; i++)
108 static void mincore_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
109 unsigned long addr, unsigned long end,
116 ptep = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
121 next = addr + PAGE_SIZE;
123 mincore_unmapped_range(vma, addr, next, vec);
124 else if (pte_present(pte))
126 else if (pte_file(pte)) {
127 pgoff = pte_to_pgoff(pte);
128 *vec = mincore_page(vma->vm_file->f_mapping, pgoff);
129 } else { /* pte is a swap entry */
130 swp_entry_t entry = pte_to_swp_entry(pte);
132 if (is_migration_entry(entry)) {
133 /* migration entries are always uptodate */
138 *vec = mincore_page(swap_address_space(entry),
147 } while (ptep++, addr = next, addr != end);
148 pte_unmap_unlock(ptep - 1, ptl);
151 static void mincore_pmd_range(struct vm_area_struct *vma, pud_t *pud,
152 unsigned long addr, unsigned long end,
158 pmd = pmd_offset(pud, addr);
160 next = pmd_addr_end(addr, end);
161 if (pmd_trans_huge(*pmd)) {
162 if (mincore_huge_pmd(vma, pmd, addr, next, vec)) {
163 vec += (next - addr) >> PAGE_SHIFT;
168 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
169 mincore_unmapped_range(vma, addr, next, vec);
171 mincore_pte_range(vma, pmd, addr, next, vec);
172 vec += (next - addr) >> PAGE_SHIFT;
173 } while (pmd++, addr = next, addr != end);
176 static void mincore_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
177 unsigned long addr, unsigned long end,
183 pud = pud_offset(pgd, addr);
185 next = pud_addr_end(addr, end);
186 if (pud_none_or_clear_bad(pud))
187 mincore_unmapped_range(vma, addr, next, vec);
189 mincore_pmd_range(vma, pud, addr, next, vec);
190 vec += (next - addr) >> PAGE_SHIFT;
191 } while (pud++, addr = next, addr != end);
194 static void mincore_page_range(struct vm_area_struct *vma,
195 unsigned long addr, unsigned long end,
201 pgd = pgd_offset(vma->vm_mm, addr);
203 next = pgd_addr_end(addr, end);
204 if (pgd_none_or_clear_bad(pgd))
205 mincore_unmapped_range(vma, addr, next, vec);
207 mincore_pud_range(vma, pgd, addr, next, vec);
208 vec += (next - addr) >> PAGE_SHIFT;
209 } while (pgd++, addr = next, addr != end);
213 * Do a chunk of "sys_mincore()". We've already checked
214 * all the arguments, we hold the mmap semaphore: we should
215 * just return the amount of info we're asked for.
217 static long do_mincore(unsigned long addr, unsigned long pages, unsigned char *vec)
219 struct vm_area_struct *vma;
222 vma = find_vma(current->mm, addr);
223 if (!vma || addr < vma->vm_start)
226 end = min(vma->vm_end, addr + (pages << PAGE_SHIFT));
228 if (is_vm_hugetlb_page(vma))
229 mincore_hugetlb_page_range(vma, addr, end, vec);
231 mincore_page_range(vma, addr, end, vec);
233 return (end - addr) >> PAGE_SHIFT;
237 * The mincore(2) system call.
239 * mincore() returns the memory residency status of the pages in the
240 * current process's address space specified by [addr, addr + len).
241 * The status is returned in a vector of bytes. The least significant
242 * bit of each byte is 1 if the referenced page is in memory, otherwise
245 * Because the status of a page can change after mincore() checks it
246 * but before it returns to the application, the returned vector may
247 * contain stale information. Only locked pages are guaranteed to
252 * -EFAULT - vec points to an illegal address
253 * -EINVAL - addr is not a multiple of PAGE_CACHE_SIZE
254 * -ENOMEM - Addresses in the range [addr, addr + len] are
255 * invalid for the address space of this process, or
256 * specify one or more pages which are not currently
258 * -EAGAIN - A kernel resource was temporarily unavailable.
260 SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
261 unsigned char __user *, vec)
267 /* Check the start address: needs to be page-aligned.. */
268 if (start & ~PAGE_CACHE_MASK)
271 /* ..and we need to be passed a valid user-space range */
272 if (!access_ok(VERIFY_READ, (void __user *) start, len))
275 /* This also avoids any overflows on PAGE_CACHE_ALIGN */
276 pages = len >> PAGE_SHIFT;
277 pages += (len & ~PAGE_MASK) != 0;
279 if (!access_ok(VERIFY_WRITE, vec, pages))
282 tmp = (void *) __get_free_page(GFP_USER);
289 * Do at most PAGE_SIZE entries per iteration, due to
290 * the temporary buffer size.
292 down_read(¤t->mm->mmap_sem);
293 retval = do_mincore(start, min(pages, PAGE_SIZE), tmp);
294 up_read(¤t->mm->mmap_sem);
298 if (copy_to_user(vec, tmp, retval)) {
304 start += retval << PAGE_SHIFT;
307 free_page((unsigned long) tmp);