the-tree/mm/mincore.c

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  *  linux/mm/mincore.c
0004  *
0005  * Copyright (C) 1994-2006  Linus Torvalds
0006  */
0007
0008 /*
0009  * The mincore() system call.
0010  */
0011 #include <linux/pagemap.h>
0012 #include <linux/gfp.h>
0013 #include <linux/pagewalk.h>
0014 #include <linux/mman.h>
0015 #include <linux/syscalls.h>
0016 #include <linux/swap.h>
0017 #include <linux/swapops.h>
0018 #include <linux/shmem_fs.h>
0019 #include <linux/hugetlb.h>
0020 #include <linux/pgtable.h>
0021
0022 #include <linux/uaccess.h>
0023 #include "swap.h"
0024
0025 static int mincore_hugetlb(pte_t *pte, unsigned long hmask, unsigned long addr,
0026             unsigned long end, struct mm_walk *walk)
0027 {
0028 #ifdef CONFIG_HUGETLB_PAGE
0029     unsigned char present;
0030     unsigned char *vec = walk->private;
0031
0032     /*
0033      * Hugepages under user process are always in RAM and never
0034      * swapped out, but theoretically it needs to be checked.
0035      */
0036     present = pte && !huge_pte_none(huge_ptep_get(pte));
0037     for (; addr != end; vec++, addr += PAGE_SIZE)
0038         *vec = present;
0039     walk->private = vec;
0040 #else
0041     BUG();
0042 #endif
0043     return 0;
0044 }
0045
0046 /*
0047  * Later we can get more picky about what "in core" means precisely.
0048  * For now, simply check to see if the page is in the page cache,
0049  * and is up to date; i.e. that no page-in operation would be required
0050  * at this time if an application were to map and access this page.
0051  */
0052 static unsigned char mincore_page(struct address_space *mapping, pgoff_t index)
0053 {
0054     unsigned char present = 0;
0055     struct page *page;
0056
0057     /*
0058      * When tmpfs swaps out a page from a file, any process mapping that
0059      * file will not get a swp_entry_t in its pte, but rather it is like
0060      * any other file mapping (ie. marked !present and faulted in with
0061      * tmpfs's .fault). So swapped out tmpfs mappings are tested here.
0062      */
0063     page = find_get_incore_page(mapping, index);
0064     if (page) {
0065         present = PageUptodate(page);
0066         put_page(page);
0067     }
0068
0069     return present;
0070 }
0071
0072 static int __mincore_unmapped_range(unsigned long addr, unsigned long end,
0073                 struct vm_area_struct *vma, unsigned char *vec)
0074 {
0075     unsigned long nr = (end - addr) >> PAGE_SHIFT;
0076     int i;
0077
0078     if (vma->vm_file) {
0079         pgoff_t pgoff;
0080
0081         pgoff = linear_page_index(vma, addr);
0082         for (i = 0; i < nr; i++, pgoff++)
0083             vec[i] = mincore_page(vma->vm_file->f_mapping, pgoff);
0084     } else {
0085         for (i = 0; i < nr; i++)
0086             vec[i] = 0;
0087     }
0088     return nr;
0089 }
0090
0091 static int mincore_unmapped_range(unsigned long addr, unsigned long end,
0092                    __always_unused int depth,
0093                    struct mm_walk *walk)
0094 {
0095     walk->private += __mincore_unmapped_range(addr, end,
0096                           walk->vma, walk->private);
0097     return 0;
0098 }
0099
0100 static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
0101             struct mm_walk *walk)
0102 {
0103     spinlock_t *ptl;
0104     struct vm_area_struct *vma = walk->vma;
0105     pte_t *ptep;
0106     unsigned char *vec = walk->private;
0107     int nr = (end - addr) >> PAGE_SHIFT;
0108
0109     ptl = pmd_trans_huge_lock(pmd, vma);
0110     if (ptl) {
0111         memset(vec, 1, nr);
0112         spin_unlock(ptl);
0113         goto out;
0114     }
0115
0116     if (pmd_trans_unstable(pmd)) {
0117         __mincore_unmapped_range(addr, end, vma, vec);
0118         goto out;
0119     }
0120
0121     ptep = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
0122     for (; addr != end; ptep++, addr += PAGE_SIZE) {
0123         pte_t pte = *ptep;
0124
0125         /* We need to do cache lookup too for pte markers */
0126         if (pte_none_mostly(pte))
0127             __mincore_unmapped_range(addr, addr + PAGE_SIZE,
0128                          vma, vec);
0129         else if (pte_present(pte))
0130             *vec = 1;
0131         else { /* pte is a swap entry */
0132             swp_entry_t entry = pte_to_swp_entry(pte);
0133
0134             if (non_swap_entry(entry)) {
0135                 /*
0136                  * migration or hwpoison entries are always
0137                  * uptodate
0138                  */
0139                 *vec = 1;
0140             } else {
0141 #ifdef CONFIG_SWAP
0142                 *vec = mincore_page(swap_address_space(entry),
0143                             swp_offset(entry));
0144 #else
0145                 WARN_ON(1);
0146                 *vec = 1;
0147 #endif
0148             }
0149         }
0150         vec++;
0151     }
0152     pte_unmap_unlock(ptep - 1, ptl);
0153 out:
0154     walk->private += nr;
0155     cond_resched();
0156     return 0;
0157 }
0158
0159 static inline bool can_do_mincore(struct vm_area_struct *vma)
0160 {
0161     if (vma_is_anonymous(vma))
0162         return true;
0163     if (!vma->vm_file)
0164         return false;
0165     /*
0166      * Reveal pagecache information only for non-anonymous mappings that
0167      * correspond to the files the calling process could (if tried) open
0168      * for writing; otherwise we'd be including shared non-exclusive
0169      * mappings, which opens a side channel.
0170      */
0171     return inode_owner_or_capable(&init_user_ns,
0172                       file_inode(vma->vm_file)) ||
0173            file_permission(vma->vm_file, MAY_WRITE) == 0;
0174 }
0175
0176 static const struct mm_walk_ops mincore_walk_ops = {
0177     .pmd_entry      = mincore_pte_range,
0178     .pte_hole       = mincore_unmapped_range,
0179     .hugetlb_entry      = mincore_hugetlb,
0180 };
0181
0182 /*
0183  * Do a chunk of "sys_mincore()". We've already checked
0184  * all the arguments, we hold the mmap semaphore: we should
0185  * just return the amount of info we're asked for.
0186  */
0187 static long do_mincore(unsigned long addr, unsigned long pages, unsigned char *vec)
0188 {
0189     struct vm_area_struct *vma;
0190     unsigned long end;
0191     int err;
0192
0193     vma = find_vma(current->mm, addr);
0194     if (!vma || addr < vma->vm_start)
0195         return -ENOMEM;
0196     end = min(vma->vm_end, addr + (pages << PAGE_SHIFT));
0197     if (!can_do_mincore(vma)) {
0198         unsigned long pages = DIV_ROUND_UP(end - addr, PAGE_SIZE);
0199         memset(vec, 1, pages);
0200         return pages;
0201     }
0202     err = walk_page_range(vma->vm_mm, addr, end, &mincore_walk_ops, vec);
0203     if (err < 0)
0204         return err;
0205     return (end - addr) >> PAGE_SHIFT;
0206 }
0207
0208 /*
0209  * The mincore(2) system call.
0210  *
0211  * mincore() returns the memory residency status of the pages in the
0212  * current process's address space specified by [addr, addr + len).
0213  * The status is returned in a vector of bytes.  The least significant
0214  * bit of each byte is 1 if the referenced page is in memory, otherwise
0215  * it is zero.
0216  *
0217  * Because the status of a page can change after mincore() checks it
0218  * but before it returns to the application, the returned vector may
0219  * contain stale information.  Only locked pages are guaranteed to
0220  * remain in memory.
0221  *
0222  * return values:
0223  *  zero    - success
0224  *  -EFAULT - vec points to an illegal address
0225  *  -EINVAL - addr is not a multiple of PAGE_SIZE
0226  *  -ENOMEM - Addresses in the range [addr, addr + len] are
0227  *      invalid for the address space of this process, or
0228  *      specify one or more pages which are not currently
0229  *      mapped
0230  *  -EAGAIN - A kernel resource was temporarily unavailable.
0231  */
0232 SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
0233         unsigned char __user *, vec)
0234 {
0235     long retval;
0236     unsigned long pages;
0237     unsigned char *tmp;
0238
0239     start = untagged_addr(start);
0240
0241     /* Check the start address: needs to be page-aligned.. */
0242     if (start & ~PAGE_MASK)
0243         return -EINVAL;
0244
0245     /* ..and we need to be passed a valid user-space range */
0246     if (!access_ok((void __user *) start, len))
0247         return -ENOMEM;
0248
0249     /* This also avoids any overflows on PAGE_ALIGN */
0250     pages = len >> PAGE_SHIFT;
0251     pages += (offset_in_page(len)) != 0;
0252
0253     if (!access_ok(vec, pages))
0254         return -EFAULT;
0255
0256     tmp = (void *) __get_free_page(GFP_USER);
0257     if (!tmp)
0258         return -EAGAIN;
0259
0260     retval = 0;
0261     while (pages) {
0262         /*
0263          * Do at most PAGE_SIZE entries per iteration, due to
0264          * the temporary buffer size.
0265          */
0266         mmap_read_lock(current->mm);
0267         retval = do_mincore(start, min(pages, PAGE_SIZE), tmp);
0268         mmap_read_unlock(current->mm);
0269
0270         if (retval <= 0)
0271             break;
0272         if (copy_to_user(vec, tmp, retval)) {
0273             retval = -EFAULT;
0274             break;
0275         }
0276         pages -= retval;
0277         vec += retval;
0278         start += retval << PAGE_SHIFT;
0279         retval = 0;
0280     }
0281     free_page((unsigned long) tmp);
0282     return retval;
0283 }