Linux Audio

Check our new training course

Yocto / OpenEmbedded training

Feb 10-13, 2025
Register
Loading...
v6.13.7
  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 *	linux/mm/mincore.c
  4 *
  5 * Copyright (C) 1994-2006  Linus Torvalds
  6 */
  7
  8/*
  9 * The mincore() system call.
 10 */
 11#include <linux/pagemap.h>
 12#include <linux/gfp.h>
 13#include <linux/pagewalk.h>
 14#include <linux/mman.h>
 15#include <linux/syscalls.h>
 16#include <linux/swap.h>
 17#include <linux/swapops.h>
 18#include <linux/shmem_fs.h>
 19#include <linux/hugetlb.h>
 20#include <linux/pgtable.h>
 21
 22#include <linux/uaccess.h>
 23#include "swap.h"
 24
 25static int mincore_hugetlb(pte_t *pte, unsigned long hmask, unsigned long addr,
 26			unsigned long end, struct mm_walk *walk)
 
 27{
 28#ifdef CONFIG_HUGETLB_PAGE
 29	unsigned char present;
 30	unsigned char *vec = walk->private;
 31
 32	/*
 33	 * Hugepages under user process are always in RAM and never
 34	 * swapped out, but theoretically it needs to be checked.
 35	 */
 36	present = pte && !huge_pte_none_mostly(huge_ptep_get(walk->mm, addr, pte));
 37	for (; addr != end; vec++, addr += PAGE_SIZE)
 38		*vec = present;
 39	walk->private = vec;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 40#else
 41	BUG();
 42#endif
 43	return 0;
 44}
 45
 46/*
 47 * Later we can get more picky about what "in core" means precisely.
 48 * For now, simply check to see if the page is in the page cache,
 49 * and is up to date; i.e. that no page-in operation would be required
 50 * at this time if an application were to map and access this page.
 51 */
 52static unsigned char mincore_page(struct address_space *mapping, pgoff_t index)
 53{
 54	unsigned char present = 0;
 55	struct folio *folio;
 56
 57	/*
 58	 * When tmpfs swaps out a page from a file, any process mapping that
 59	 * file will not get a swp_entry_t in its pte, but rather it is like
 60	 * any other file mapping (ie. marked !present and faulted in with
 61	 * tmpfs's .fault). So swapped out tmpfs mappings are tested here.
 62	 */
 63	folio = filemap_get_incore_folio(mapping, index);
 64	if (!IS_ERR(folio)) {
 65		present = folio_test_uptodate(folio);
 66		folio_put(folio);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 67	}
 68
 69	return present;
 70}
 71
 72static int __mincore_unmapped_range(unsigned long addr, unsigned long end,
 73				struct vm_area_struct *vma, unsigned char *vec)
 
 74{
 75	unsigned long nr = (end - addr) >> PAGE_SHIFT;
 76	int i;
 77
 78	if (vma->vm_file) {
 79		pgoff_t pgoff;
 80
 81		pgoff = linear_page_index(vma, addr);
 82		for (i = 0; i < nr; i++, pgoff++)
 83			vec[i] = mincore_page(vma->vm_file->f_mapping, pgoff);
 84	} else {
 85		for (i = 0; i < nr; i++)
 86			vec[i] = 0;
 87	}
 88	return nr;
 89}
 90
 91static int mincore_unmapped_range(unsigned long addr, unsigned long end,
 92				   __always_unused int depth,
 93				   struct mm_walk *walk)
 94{
 95	walk->private += __mincore_unmapped_range(addr, end,
 96						  walk->vma, walk->private);
 97	return 0;
 98}
 99
100static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
101			struct mm_walk *walk)
 
102{
 
103	spinlock_t *ptl;
104	struct vm_area_struct *vma = walk->vma;
105	pte_t *ptep;
106	unsigned char *vec = walk->private;
107	int nr = (end - addr) >> PAGE_SHIFT;
108
109	ptl = pmd_trans_huge_lock(pmd, vma);
110	if (ptl) {
111		memset(vec, 1, nr);
112		spin_unlock(ptl);
113		goto out;
114	}
115
116	ptep = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
117	if (!ptep) {
118		walk->action = ACTION_AGAIN;
119		return 0;
120	}
121	for (; addr != end; ptep++, addr += PAGE_SIZE) {
122		pte_t pte = ptep_get(ptep);
123
124		/* We need to do cache lookup too for pte markers */
125		if (pte_none_mostly(pte))
126			__mincore_unmapped_range(addr, addr + PAGE_SIZE,
127						 vma, vec);
128		else if (pte_present(pte))
129			*vec = 1;
130		else { /* pte is a swap entry */
 
 
 
131			swp_entry_t entry = pte_to_swp_entry(pte);
132
133			if (non_swap_entry(entry)) {
134				/*
135				 * migration or hwpoison entries are always
136				 * uptodate
137				 */
138				*vec = 1;
139			} else {
140#ifdef CONFIG_SWAP
 
141				*vec = mincore_page(swap_address_space(entry),
142						    swap_cache_index(entry));
143#else
144				WARN_ON(1);
145				*vec = 1;
146#endif
147			}
148		}
149		vec++;
150	}
151	pte_unmap_unlock(ptep - 1, ptl);
152out:
153	walk->private += nr;
154	cond_resched();
155	return 0;
156}
157
158static inline bool can_do_mincore(struct vm_area_struct *vma)
 
 
159{
160	if (vma_is_anonymous(vma))
161		return true;
162	if (!vma->vm_file)
163		return false;
164	/*
165	 * Reveal pagecache information only for non-anonymous mappings that
166	 * correspond to the files the calling process could (if tried) open
167	 * for writing; otherwise we'd be including shared non-exclusive
168	 * mappings, which opens a side channel.
169	 */
170	return inode_owner_or_capable(&nop_mnt_idmap,
171				      file_inode(vma->vm_file)) ||
172	       file_permission(vma->vm_file, MAY_WRITE) == 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
173}
174
175static const struct mm_walk_ops mincore_walk_ops = {
176	.pmd_entry		= mincore_pte_range,
177	.pte_hole		= mincore_unmapped_range,
178	.hugetlb_entry		= mincore_hugetlb,
179	.walk_lock		= PGWALK_RDLOCK,
180};
 
 
 
 
 
 
 
 
 
 
 
181
182/*
183 * Do a chunk of "sys_mincore()". We've already checked
184 * all the arguments, we hold the mmap semaphore: we should
185 * just return the amount of info we're asked for.
186 */
187static long do_mincore(unsigned long addr, unsigned long pages, unsigned char *vec)
188{
189	struct vm_area_struct *vma;
190	unsigned long end;
191	int err;
192
193	vma = vma_lookup(current->mm, addr);
194	if (!vma)
195		return -ENOMEM;
 
196	end = min(vma->vm_end, addr + (pages << PAGE_SHIFT));
197	if (!can_do_mincore(vma)) {
198		unsigned long pages = DIV_ROUND_UP(end - addr, PAGE_SIZE);
199		memset(vec, 1, pages);
200		return pages;
201	}
202	err = walk_page_range(vma->vm_mm, addr, end, &mincore_walk_ops, vec);
203	if (err < 0)
204		return err;
205	return (end - addr) >> PAGE_SHIFT;
206}
207
208/*
209 * The mincore(2) system call.
210 *
211 * mincore() returns the memory residency status of the pages in the
212 * current process's address space specified by [addr, addr + len).
213 * The status is returned in a vector of bytes.  The least significant
214 * bit of each byte is 1 if the referenced page is in memory, otherwise
215 * it is zero.
216 *
217 * Because the status of a page can change after mincore() checks it
218 * but before it returns to the application, the returned vector may
219 * contain stale information.  Only locked pages are guaranteed to
220 * remain in memory.
221 *
222 * return values:
223 *  zero    - success
224 *  -EFAULT - vec points to an illegal address
225 *  -EINVAL - addr is not a multiple of PAGE_SIZE
226 *  -ENOMEM - Addresses in the range [addr, addr + len] are
227 *		invalid for the address space of this process, or
228 *		specify one or more pages which are not currently
229 *		mapped
230 *  -EAGAIN - A kernel resource was temporarily unavailable.
231 */
232SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
233		unsigned char __user *, vec)
234{
235	long retval;
236	unsigned long pages;
237	unsigned char *tmp;
238
239	start = untagged_addr(start);
240
241	/* Check the start address: needs to be page-aligned.. */
242	if (start & ~PAGE_MASK)
243		return -EINVAL;
244
245	/* ..and we need to be passed a valid user-space range */
246	if (!access_ok((void __user *) start, len))
247		return -ENOMEM;
248
249	/* This also avoids any overflows on PAGE_ALIGN */
250	pages = len >> PAGE_SHIFT;
251	pages += (offset_in_page(len)) != 0;
252
253	if (!access_ok(vec, pages))
254		return -EFAULT;
255
256	tmp = (void *) __get_free_page(GFP_USER);
257	if (!tmp)
258		return -EAGAIN;
259
260	retval = 0;
261	while (pages) {
262		/*
263		 * Do at most PAGE_SIZE entries per iteration, due to
264		 * the temporary buffer size.
265		 */
266		mmap_read_lock(current->mm);
267		retval = do_mincore(start, min(pages, PAGE_SIZE), tmp);
268		mmap_read_unlock(current->mm);
269
270		if (retval <= 0)
271			break;
272		if (copy_to_user(vec, tmp, retval)) {
273			retval = -EFAULT;
274			break;
275		}
276		pages -= retval;
277		vec += retval;
278		start += retval << PAGE_SHIFT;
279		retval = 0;
280	}
281	free_page((unsigned long) tmp);
282	return retval;
283}
v3.15
 
  1/*
  2 *	linux/mm/mincore.c
  3 *
  4 * Copyright (C) 1994-2006  Linus Torvalds
  5 */
  6
  7/*
  8 * The mincore() system call.
  9 */
 10#include <linux/pagemap.h>
 11#include <linux/gfp.h>
 12#include <linux/mm.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>
 
 18
 19#include <asm/uaccess.h>
 20#include <asm/pgtable.h>
 21
 22static void mincore_hugetlb_page_range(struct vm_area_struct *vma,
 23				unsigned long addr, unsigned long end,
 24				unsigned char *vec)
 25{
 26#ifdef CONFIG_HUGETLB_PAGE
 27	struct hstate *h;
 
 28
 29	h = hstate_vma(vma);
 30	while (1) {
 31		unsigned char present;
 32		pte_t *ptep;
 33		/*
 34		 * Huge pages are always in RAM for now, but
 35		 * theoretically it needs to be checked.
 36		 */
 37		ptep = huge_pte_offset(current->mm,
 38				       addr & huge_page_mask(h));
 39		present = ptep && !huge_pte_none(huge_ptep_get(ptep));
 40		while (1) {
 41			*vec = present;
 42			vec++;
 43			addr += PAGE_SIZE;
 44			if (addr == end)
 45				return;
 46			/* check hugepage border */
 47			if (!(addr & ~huge_page_mask(h)))
 48				break;
 49		}
 50	}
 51#else
 52	BUG();
 53#endif
 
 54}
 55
 56/*
 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.
 61 */
 62static unsigned char mincore_page(struct address_space *mapping, pgoff_t pgoff)
 63{
 64	unsigned char present = 0;
 65	struct page *page;
 66
 67	/*
 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.
 72	 */
 73#ifdef CONFIG_SWAP
 74	if (shmem_mapping(mapping)) {
 75		page = find_get_entry(mapping, pgoff);
 76		/*
 77		 * shmem/tmpfs may return swap: account for swapcache
 78		 * page too.
 79		 */
 80		if (radix_tree_exceptional_entry(page)) {
 81			swp_entry_t swp = radix_to_swp_entry(page);
 82			page = find_get_page(swap_address_space(swp), swp.val);
 83		}
 84	} else
 85		page = find_get_page(mapping, pgoff);
 86#else
 87	page = find_get_page(mapping, pgoff);
 88#endif
 89	if (page) {
 90		present = PageUptodate(page);
 91		page_cache_release(page);
 92	}
 93
 94	return present;
 95}
 96
 97static void mincore_unmapped_range(struct vm_area_struct *vma,
 98				unsigned long addr, unsigned long end,
 99				unsigned char *vec)
100{
101	unsigned long nr = (end - addr) >> PAGE_SHIFT;
102	int i;
103
104	if (vma->vm_file) {
105		pgoff_t pgoff;
106
107		pgoff = linear_page_index(vma, addr);
108		for (i = 0; i < nr; i++, pgoff++)
109			vec[i] = mincore_page(vma->vm_file->f_mapping, pgoff);
110	} else {
111		for (i = 0; i < nr; i++)
112			vec[i] = 0;
113	}
 
 
 
 
 
 
 
 
 
 
114}
115
116static void mincore_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
117			unsigned long addr, unsigned long end,
118			unsigned char *vec)
119{
120	unsigned long next;
121	spinlock_t *ptl;
 
122	pte_t *ptep;
 
 
123
124	ptep = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
125	do {
126		pte_t pte = *ptep;
127		pgoff_t pgoff;
 
 
 
 
 
 
 
 
 
 
128
129		next = addr + PAGE_SIZE;
130		if (pte_none(pte))
131			mincore_unmapped_range(vma, addr, next, vec);
 
132		else if (pte_present(pte))
133			*vec = 1;
134		else if (pte_file(pte)) {
135			pgoff = pte_to_pgoff(pte);
136			*vec = mincore_page(vma->vm_file->f_mapping, pgoff);
137		} else { /* pte is a swap entry */
138			swp_entry_t entry = pte_to_swp_entry(pte);
139
140			if (is_migration_entry(entry)) {
141				/* migration entries are always uptodate */
 
 
 
142				*vec = 1;
143			} else {
144#ifdef CONFIG_SWAP
145				pgoff = entry.val;
146				*vec = mincore_page(swap_address_space(entry),
147					pgoff);
148#else
149				WARN_ON(1);
150				*vec = 1;
151#endif
152			}
153		}
154		vec++;
155	} while (ptep++, addr = next, addr != end);
156	pte_unmap_unlock(ptep - 1, ptl);
 
 
 
 
157}
158
159static void mincore_pmd_range(struct vm_area_struct *vma, pud_t *pud,
160			unsigned long addr, unsigned long end,
161			unsigned char *vec)
162{
163	unsigned long next;
164	pmd_t *pmd;
165
166	pmd = pmd_offset(pud, addr);
167	do {
168		next = pmd_addr_end(addr, end);
169		if (pmd_trans_huge(*pmd)) {
170			if (mincore_huge_pmd(vma, pmd, addr, next, vec)) {
171				vec += (next - addr) >> PAGE_SHIFT;
172				continue;
173			}
174			/* fall through */
175		}
176		if (pmd_none_or_trans_huge_or_clear_bad(pmd))
177			mincore_unmapped_range(vma, addr, next, vec);
178		else
179			mincore_pte_range(vma, pmd, addr, next, vec);
180		vec += (next - addr) >> PAGE_SHIFT;
181	} while (pmd++, addr = next, addr != end);
182}
183
184static void mincore_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
185			unsigned long addr, unsigned long end,
186			unsigned char *vec)
187{
188	unsigned long next;
189	pud_t *pud;
190
191	pud = pud_offset(pgd, addr);
192	do {
193		next = pud_addr_end(addr, end);
194		if (pud_none_or_clear_bad(pud))
195			mincore_unmapped_range(vma, addr, next, vec);
196		else
197			mincore_pmd_range(vma, pud, addr, next, vec);
198		vec += (next - addr) >> PAGE_SHIFT;
199	} while (pud++, addr = next, addr != end);
200}
201
202static void mincore_page_range(struct vm_area_struct *vma,
203			unsigned long addr, unsigned long end,
204			unsigned char *vec)
205{
206	unsigned long next;
207	pgd_t *pgd;
208
209	pgd = pgd_offset(vma->vm_mm, addr);
210	do {
211		next = pgd_addr_end(addr, end);
212		if (pgd_none_or_clear_bad(pgd))
213			mincore_unmapped_range(vma, addr, next, vec);
214		else
215			mincore_pud_range(vma, pgd, addr, next, vec);
216		vec += (next - addr) >> PAGE_SHIFT;
217	} while (pgd++, addr = next, addr != end);
218}
219
220/*
221 * Do a chunk of "sys_mincore()". We've already checked
222 * all the arguments, we hold the mmap semaphore: we should
223 * just return the amount of info we're asked for.
224 */
225static long do_mincore(unsigned long addr, unsigned long pages, unsigned char *vec)
226{
227	struct vm_area_struct *vma;
228	unsigned long end;
 
229
230	vma = find_vma(current->mm, addr);
231	if (!vma || addr < vma->vm_start)
232		return -ENOMEM;
233
234	end = min(vma->vm_end, addr + (pages << PAGE_SHIFT));
235
236	if (is_vm_hugetlb_page(vma))
237		mincore_hugetlb_page_range(vma, addr, end, vec);
238	else
239		mincore_page_range(vma, addr, end, vec);
240
 
 
241	return (end - addr) >> PAGE_SHIFT;
242}
243
244/*
245 * The mincore(2) system call.
246 *
247 * mincore() returns the memory residency status of the pages in the
248 * current process's address space specified by [addr, addr + len).
249 * The status is returned in a vector of bytes.  The least significant
250 * bit of each byte is 1 if the referenced page is in memory, otherwise
251 * it is zero.
252 *
253 * Because the status of a page can change after mincore() checks it
254 * but before it returns to the application, the returned vector may
255 * contain stale information.  Only locked pages are guaranteed to
256 * remain in memory.
257 *
258 * return values:
259 *  zero    - success
260 *  -EFAULT - vec points to an illegal address
261 *  -EINVAL - addr is not a multiple of PAGE_CACHE_SIZE
262 *  -ENOMEM - Addresses in the range [addr, addr + len] are
263 *		invalid for the address space of this process, or
264 *		specify one or more pages which are not currently
265 *		mapped
266 *  -EAGAIN - A kernel resource was temporarily unavailable.
267 */
268SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
269		unsigned char __user *, vec)
270{
271	long retval;
272	unsigned long pages;
273	unsigned char *tmp;
274
 
 
275	/* Check the start address: needs to be page-aligned.. */
276 	if (start & ~PAGE_CACHE_MASK)
277		return -EINVAL;
278
279	/* ..and we need to be passed a valid user-space range */
280	if (!access_ok(VERIFY_READ, (void __user *) start, len))
281		return -ENOMEM;
282
283	/* This also avoids any overflows on PAGE_CACHE_ALIGN */
284	pages = len >> PAGE_SHIFT;
285	pages += (len & ~PAGE_MASK) != 0;
286
287	if (!access_ok(VERIFY_WRITE, vec, pages))
288		return -EFAULT;
289
290	tmp = (void *) __get_free_page(GFP_USER);
291	if (!tmp)
292		return -EAGAIN;
293
294	retval = 0;
295	while (pages) {
296		/*
297		 * Do at most PAGE_SIZE entries per iteration, due to
298		 * the temporary buffer size.
299		 */
300		down_read(&current->mm->mmap_sem);
301		retval = do_mincore(start, min(pages, PAGE_SIZE), tmp);
302		up_read(&current->mm->mmap_sem);
303
304		if (retval <= 0)
305			break;
306		if (copy_to_user(vec, tmp, retval)) {
307			retval = -EFAULT;
308			break;
309		}
310		pages -= retval;
311		vec += retval;
312		start += retval << PAGE_SHIFT;
313		retval = 0;
314	}
315	free_page((unsigned long) tmp);
316	return retval;
317}