2 #include <linux/vmacache.h>
3 #include <linux/hugetlb.h>
4 #include <linux/huge_mm.h>
5 #include <linux/mount.h>
6 #include <linux/seq_file.h>
7 #include <linux/highmem.h>
8 #include <linux/ptrace.h>
9 #include <linux/slab.h>
10 #include <linux/pagemap.h>
11 #include <linux/mempolicy.h>
12 #include <linux/rmap.h>
13 #include <linux/swap.h>
14 #include <linux/swapops.h>
15 #include <linux/mmu_notifier.h>
16 #include <linux/page_idle.h>
19 #include <asm/uaccess.h>
20 #include <asm/tlbflush.h>
23 void task_mem(struct seq_file *m, struct mm_struct *mm)
25 unsigned long data, text, lib, swap, ptes, pmds;
26 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
29 * Note: to minimize their overhead, mm maintains hiwater_vm and
30 * hiwater_rss only when about to *lower* total_vm or rss. Any
31 * collector of these hiwater stats must therefore get total_vm
32 * and rss too, which will usually be the higher. Barriers? not
33 * worth the effort, such snapshots can always be inconsistent.
35 hiwater_vm = total_vm = mm->total_vm;
36 if (hiwater_vm < mm->hiwater_vm)
37 hiwater_vm = mm->hiwater_vm;
38 hiwater_rss = total_rss = get_mm_rss(mm);
39 if (hiwater_rss < mm->hiwater_rss)
40 hiwater_rss = mm->hiwater_rss;
42 data = mm->total_vm - mm->shared_vm - mm->stack_vm;
43 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
44 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
45 swap = get_mm_counter(mm, MM_SWAPENTS);
46 ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
47 pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
62 hiwater_vm << (PAGE_SHIFT-10),
63 total_vm << (PAGE_SHIFT-10),
64 mm->locked_vm << (PAGE_SHIFT-10),
65 mm->pinned_vm << (PAGE_SHIFT-10),
66 hiwater_rss << (PAGE_SHIFT-10),
67 total_rss << (PAGE_SHIFT-10),
68 data << (PAGE_SHIFT-10),
69 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
72 swap << (PAGE_SHIFT-10));
75 unsigned long task_vsize(struct mm_struct *mm)
77 return PAGE_SIZE * mm->total_vm;
80 unsigned long task_statm(struct mm_struct *mm,
81 unsigned long *shared, unsigned long *text,
82 unsigned long *data, unsigned long *resident)
84 *shared = get_mm_counter(mm, MM_FILEPAGES);
85 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
87 *data = mm->total_vm - mm->shared_vm;
88 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
94 * Save get_task_policy() for show_numa_map().
96 static void hold_task_mempolicy(struct proc_maps_private *priv)
98 struct task_struct *task = priv->task;
101 priv->task_mempolicy = get_task_policy(task);
102 mpol_get(priv->task_mempolicy);
105 static void release_task_mempolicy(struct proc_maps_private *priv)
107 mpol_put(priv->task_mempolicy);
110 static void hold_task_mempolicy(struct proc_maps_private *priv)
113 static void release_task_mempolicy(struct proc_maps_private *priv)
118 static void vma_stop(struct proc_maps_private *priv)
120 struct mm_struct *mm = priv->mm;
122 release_task_mempolicy(priv);
123 up_read(&mm->mmap_sem);
127 static struct vm_area_struct *
128 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
130 if (vma == priv->tail_vma)
132 return vma->vm_next ?: priv->tail_vma;
135 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
137 if (m->count < m->size) /* vma is copied successfully */
138 m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL;
141 static void *m_start(struct seq_file *m, loff_t *ppos)
143 struct proc_maps_private *priv = m->private;
144 unsigned long last_addr = m->version;
145 struct mm_struct *mm;
146 struct vm_area_struct *vma;
147 unsigned int pos = *ppos;
149 /* See m_cache_vma(). Zero at the start or after lseek. */
150 if (last_addr == -1UL)
153 priv->task = get_proc_task(priv->inode);
155 return ERR_PTR(-ESRCH);
158 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
161 down_read(&mm->mmap_sem);
162 hold_task_mempolicy(priv);
163 priv->tail_vma = get_gate_vma(mm);
166 vma = find_vma(mm, last_addr);
167 if (vma && (vma = m_next_vma(priv, vma)))
172 if (pos < mm->map_count) {
173 for (vma = mm->mmap; pos; pos--) {
174 m->version = vma->vm_start;
180 /* we do not bother to update m->version in this case */
181 if (pos == mm->map_count && priv->tail_vma)
182 return priv->tail_vma;
188 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
190 struct proc_maps_private *priv = m->private;
191 struct vm_area_struct *next;
194 next = m_next_vma(priv, v);
200 static void m_stop(struct seq_file *m, void *v)
202 struct proc_maps_private *priv = m->private;
204 if (!IS_ERR_OR_NULL(v))
207 put_task_struct(priv->task);
212 static int proc_maps_open(struct inode *inode, struct file *file,
213 const struct seq_operations *ops, int psize)
215 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
221 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
222 if (IS_ERR(priv->mm)) {
223 int err = PTR_ERR(priv->mm);
225 seq_release_private(inode, file);
232 static int proc_map_release(struct inode *inode, struct file *file)
234 struct seq_file *seq = file->private_data;
235 struct proc_maps_private *priv = seq->private;
240 return seq_release_private(inode, file);
243 static int do_maps_open(struct inode *inode, struct file *file,
244 const struct seq_operations *ops)
246 return proc_maps_open(inode, file, ops,
247 sizeof(struct proc_maps_private));
250 static pid_t pid_of_stack(struct proc_maps_private *priv,
251 struct vm_area_struct *vma, bool is_pid)
253 struct inode *inode = priv->inode;
254 struct task_struct *task;
258 task = pid_task(proc_pid(inode), PIDTYPE_PID);
260 task = task_of_stack(task, vma, is_pid);
262 ret = task_pid_nr_ns(task, inode->i_sb->s_fs_info);
270 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
272 struct mm_struct *mm = vma->vm_mm;
273 struct file *file = vma->vm_file;
274 struct proc_maps_private *priv = m->private;
275 vm_flags_t flags = vma->vm_flags;
276 unsigned long ino = 0;
277 unsigned long long pgoff = 0;
278 unsigned long start, end;
280 const char *name = NULL;
283 struct inode *inode = file_inode(vma->vm_file);
284 dev = inode->i_sb->s_dev;
286 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
289 /* We don't show the stack guard page in /proc/maps */
290 start = vma->vm_start;
291 if (stack_guard_page_start(vma, start))
294 if (stack_guard_page_end(vma, end))
297 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
298 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
301 flags & VM_READ ? 'r' : '-',
302 flags & VM_WRITE ? 'w' : '-',
303 flags & VM_EXEC ? 'x' : '-',
304 flags & VM_MAYSHARE ? 's' : 'p',
306 MAJOR(dev), MINOR(dev), ino);
309 * Print the dentry name for named mappings, and a
310 * special [heap] marker for the heap:
314 seq_file_path(m, file, "\n");
318 if (vma->vm_ops && vma->vm_ops->name) {
319 name = vma->vm_ops->name(vma);
324 name = arch_vma_name(vma);
333 if (vma->vm_start <= mm->brk &&
334 vma->vm_end >= mm->start_brk) {
339 tid = pid_of_stack(priv, vma, is_pid);
342 * Thread stack in /proc/PID/task/TID/maps or
343 * the main process stack.
345 if (!is_pid || (vma->vm_start <= mm->start_stack &&
346 vma->vm_end >= mm->start_stack)) {
349 /* Thread stack in /proc/PID/maps */
351 seq_printf(m, "[stack:%d]", tid);
364 static int show_map(struct seq_file *m, void *v, int is_pid)
366 show_map_vma(m, v, is_pid);
371 static int show_pid_map(struct seq_file *m, void *v)
373 return show_map(m, v, 1);
376 static int show_tid_map(struct seq_file *m, void *v)
378 return show_map(m, v, 0);
381 static const struct seq_operations proc_pid_maps_op = {
388 static const struct seq_operations proc_tid_maps_op = {
395 static int pid_maps_open(struct inode *inode, struct file *file)
397 return do_maps_open(inode, file, &proc_pid_maps_op);
400 static int tid_maps_open(struct inode *inode, struct file *file)
402 return do_maps_open(inode, file, &proc_tid_maps_op);
405 const struct file_operations proc_pid_maps_operations = {
406 .open = pid_maps_open,
409 .release = proc_map_release,
412 const struct file_operations proc_tid_maps_operations = {
413 .open = tid_maps_open,
416 .release = proc_map_release,
420 * Proportional Set Size(PSS): my share of RSS.
422 * PSS of a process is the count of pages it has in memory, where each
423 * page is divided by the number of processes sharing it. So if a
424 * process has 1000 pages all to itself, and 1000 shared with one other
425 * process, its PSS will be 1500.
427 * To keep (accumulated) division errors low, we adopt a 64bit
428 * fixed-point pss counter to minimize division errors. So (pss >>
429 * PSS_SHIFT) would be the real byte count.
431 * A shift of 12 before division means (assuming 4K page size):
432 * - 1M 3-user-pages add up to 8KB errors;
433 * - supports mapcount up to 2^24, or 16M;
434 * - supports PSS up to 2^52 bytes, or 4PB.
438 #ifdef CONFIG_PROC_PAGE_MONITOR
439 struct mem_size_stats {
440 unsigned long resident;
441 unsigned long shared_clean;
442 unsigned long shared_dirty;
443 unsigned long private_clean;
444 unsigned long private_dirty;
445 unsigned long referenced;
446 unsigned long anonymous;
447 unsigned long anonymous_thp;
449 unsigned long shared_hugetlb;
450 unsigned long private_hugetlb;
455 static void smaps_account(struct mem_size_stats *mss, struct page *page,
456 unsigned long size, bool young, bool dirty)
461 mss->anonymous += size;
463 mss->resident += size;
464 /* Accumulate the size in pages that have been accessed. */
465 if (young || page_is_young(page) || PageReferenced(page))
466 mss->referenced += size;
467 mapcount = page_mapcount(page);
471 if (dirty || PageDirty(page))
472 mss->shared_dirty += size;
474 mss->shared_clean += size;
475 pss_delta = (u64)size << PSS_SHIFT;
476 do_div(pss_delta, mapcount);
477 mss->pss += pss_delta;
479 if (dirty || PageDirty(page))
480 mss->private_dirty += size;
482 mss->private_clean += size;
483 mss->pss += (u64)size << PSS_SHIFT;
487 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
488 struct mm_walk *walk)
490 struct mem_size_stats *mss = walk->private;
491 struct vm_area_struct *vma = walk->vma;
492 struct page *page = NULL;
494 if (pte_present(*pte)) {
495 page = vm_normal_page(vma, addr, *pte);
496 } else if (is_swap_pte(*pte)) {
497 swp_entry_t swpent = pte_to_swp_entry(*pte);
499 if (!non_swap_entry(swpent)) {
502 mss->swap += PAGE_SIZE;
503 mapcount = swp_swapcount(swpent);
505 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
507 do_div(pss_delta, mapcount);
508 mss->swap_pss += pss_delta;
510 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
512 } else if (is_migration_entry(swpent))
513 page = migration_entry_to_page(swpent);
518 smaps_account(mss, page, PAGE_SIZE, pte_young(*pte), pte_dirty(*pte));
521 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
522 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
523 struct mm_walk *walk)
525 struct mem_size_stats *mss = walk->private;
526 struct vm_area_struct *vma = walk->vma;
529 /* FOLL_DUMP will return -EFAULT on huge zero page */
530 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
531 if (IS_ERR_OR_NULL(page))
533 mss->anonymous_thp += HPAGE_PMD_SIZE;
534 smaps_account(mss, page, HPAGE_PMD_SIZE,
535 pmd_young(*pmd), pmd_dirty(*pmd));
538 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
539 struct mm_walk *walk)
544 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
545 struct mm_walk *walk)
547 struct vm_area_struct *vma = walk->vma;
551 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
552 smaps_pmd_entry(pmd, addr, walk);
557 if (pmd_trans_unstable(pmd))
560 * The mmap_sem held all the way back in m_start() is what
561 * keeps khugepaged out of here and from collapsing things
564 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
565 for (; addr != end; pte++, addr += PAGE_SIZE)
566 smaps_pte_entry(pte, addr, walk);
567 pte_unmap_unlock(pte - 1, ptl);
572 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
575 * Don't forget to update Documentation/ on changes.
577 static const char mnemonics[BITS_PER_LONG][2] = {
579 * In case if we meet a flag we don't know about.
581 [0 ... (BITS_PER_LONG-1)] = "??",
583 [ilog2(VM_READ)] = "rd",
584 [ilog2(VM_WRITE)] = "wr",
585 [ilog2(VM_EXEC)] = "ex",
586 [ilog2(VM_SHARED)] = "sh",
587 [ilog2(VM_MAYREAD)] = "mr",
588 [ilog2(VM_MAYWRITE)] = "mw",
589 [ilog2(VM_MAYEXEC)] = "me",
590 [ilog2(VM_MAYSHARE)] = "ms",
591 [ilog2(VM_GROWSDOWN)] = "gd",
592 [ilog2(VM_PFNMAP)] = "pf",
593 [ilog2(VM_DENYWRITE)] = "dw",
594 #ifdef CONFIG_X86_INTEL_MPX
595 [ilog2(VM_MPX)] = "mp",
597 [ilog2(VM_LOCKED)] = "lo",
598 [ilog2(VM_IO)] = "io",
599 [ilog2(VM_SEQ_READ)] = "sr",
600 [ilog2(VM_RAND_READ)] = "rr",
601 [ilog2(VM_DONTCOPY)] = "dc",
602 [ilog2(VM_DONTEXPAND)] = "de",
603 [ilog2(VM_ACCOUNT)] = "ac",
604 [ilog2(VM_NORESERVE)] = "nr",
605 [ilog2(VM_HUGETLB)] = "ht",
606 [ilog2(VM_ARCH_1)] = "ar",
607 [ilog2(VM_DONTDUMP)] = "dd",
608 #ifdef CONFIG_MEM_SOFT_DIRTY
609 [ilog2(VM_SOFTDIRTY)] = "sd",
611 [ilog2(VM_MIXEDMAP)] = "mm",
612 [ilog2(VM_HUGEPAGE)] = "hg",
613 [ilog2(VM_NOHUGEPAGE)] = "nh",
614 [ilog2(VM_MERGEABLE)] = "mg",
615 [ilog2(VM_UFFD_MISSING)]= "um",
616 [ilog2(VM_UFFD_WP)] = "uw",
620 seq_puts(m, "VmFlags: ");
621 for (i = 0; i < BITS_PER_LONG; i++) {
622 if (vma->vm_flags & (1UL << i)) {
623 seq_printf(m, "%c%c ",
624 mnemonics[i][0], mnemonics[i][1]);
630 #ifdef CONFIG_HUGETLB_PAGE
631 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
632 unsigned long addr, unsigned long end,
633 struct mm_walk *walk)
635 struct mem_size_stats *mss = walk->private;
636 struct vm_area_struct *vma = walk->vma;
637 struct page *page = NULL;
639 if (pte_present(*pte)) {
640 page = vm_normal_page(vma, addr, *pte);
641 } else if (is_swap_pte(*pte)) {
642 swp_entry_t swpent = pte_to_swp_entry(*pte);
644 if (is_migration_entry(swpent))
645 page = migration_entry_to_page(swpent);
648 int mapcount = page_mapcount(page);
651 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
653 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
657 #endif /* HUGETLB_PAGE */
659 static int show_smap(struct seq_file *m, void *v, int is_pid)
661 struct vm_area_struct *vma = v;
662 struct mem_size_stats mss;
663 struct mm_walk smaps_walk = {
664 .pmd_entry = smaps_pte_range,
665 #ifdef CONFIG_HUGETLB_PAGE
666 .hugetlb_entry = smaps_hugetlb_range,
672 memset(&mss, 0, sizeof mss);
673 /* mmap_sem is held in m_start */
674 walk_page_vma(vma, &smaps_walk);
676 show_map_vma(m, vma, is_pid);
682 "Shared_Clean: %8lu kB\n"
683 "Shared_Dirty: %8lu kB\n"
684 "Private_Clean: %8lu kB\n"
685 "Private_Dirty: %8lu kB\n"
686 "Referenced: %8lu kB\n"
687 "Anonymous: %8lu kB\n"
688 "AnonHugePages: %8lu kB\n"
689 "Shared_Hugetlb: %8lu kB\n"
690 "Private_Hugetlb: %7lu kB\n"
693 "KernelPageSize: %8lu kB\n"
694 "MMUPageSize: %8lu kB\n"
696 (vma->vm_end - vma->vm_start) >> 10,
698 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
699 mss.shared_clean >> 10,
700 mss.shared_dirty >> 10,
701 mss.private_clean >> 10,
702 mss.private_dirty >> 10,
703 mss.referenced >> 10,
705 mss.anonymous_thp >> 10,
706 mss.shared_hugetlb >> 10,
707 mss.private_hugetlb >> 10,
709 (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
710 vma_kernel_pagesize(vma) >> 10,
711 vma_mmu_pagesize(vma) >> 10,
712 (vma->vm_flags & VM_LOCKED) ?
713 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
715 show_smap_vma_flags(m, vma);
720 static int show_pid_smap(struct seq_file *m, void *v)
722 return show_smap(m, v, 1);
725 static int show_tid_smap(struct seq_file *m, void *v)
727 return show_smap(m, v, 0);
730 static const struct seq_operations proc_pid_smaps_op = {
734 .show = show_pid_smap
737 static const struct seq_operations proc_tid_smaps_op = {
741 .show = show_tid_smap
744 static int pid_smaps_open(struct inode *inode, struct file *file)
746 return do_maps_open(inode, file, &proc_pid_smaps_op);
749 static int tid_smaps_open(struct inode *inode, struct file *file)
751 return do_maps_open(inode, file, &proc_tid_smaps_op);
754 const struct file_operations proc_pid_smaps_operations = {
755 .open = pid_smaps_open,
758 .release = proc_map_release,
761 const struct file_operations proc_tid_smaps_operations = {
762 .open = tid_smaps_open,
765 .release = proc_map_release,
768 enum clear_refs_types {
772 CLEAR_REFS_SOFT_DIRTY,
773 CLEAR_REFS_MM_HIWATER_RSS,
777 struct clear_refs_private {
778 enum clear_refs_types type;
781 #ifdef CONFIG_MEM_SOFT_DIRTY
782 static inline void clear_soft_dirty(struct vm_area_struct *vma,
783 unsigned long addr, pte_t *pte)
786 * The soft-dirty tracker uses #PF-s to catch writes
787 * to pages, so write-protect the pte as well. See the
788 * Documentation/vm/soft-dirty.txt for full description
789 * of how soft-dirty works.
793 if (pte_present(ptent)) {
794 ptent = pte_wrprotect(ptent);
795 ptent = pte_clear_soft_dirty(ptent);
796 } else if (is_swap_pte(ptent)) {
797 ptent = pte_swp_clear_soft_dirty(ptent);
800 set_pte_at(vma->vm_mm, addr, pte, ptent);
803 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
804 unsigned long addr, pmd_t *pmdp)
808 pmd = pmd_wrprotect(pmd);
809 pmd = pmd_clear_soft_dirty(pmd);
811 if (vma->vm_flags & VM_SOFTDIRTY)
812 vma->vm_flags &= ~VM_SOFTDIRTY;
814 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
819 static inline void clear_soft_dirty(struct vm_area_struct *vma,
820 unsigned long addr, pte_t *pte)
824 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
825 unsigned long addr, pmd_t *pmdp)
830 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
831 unsigned long end, struct mm_walk *walk)
833 struct clear_refs_private *cp = walk->private;
834 struct vm_area_struct *vma = walk->vma;
839 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
840 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
841 clear_soft_dirty_pmd(vma, addr, pmd);
845 page = pmd_page(*pmd);
847 /* Clear accessed and referenced bits. */
848 pmdp_test_and_clear_young(vma, addr, pmd);
849 test_and_clear_page_young(page);
850 ClearPageReferenced(page);
856 if (pmd_trans_unstable(pmd))
859 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
860 for (; addr != end; pte++, addr += PAGE_SIZE) {
863 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
864 clear_soft_dirty(vma, addr, pte);
868 if (!pte_present(ptent))
871 page = vm_normal_page(vma, addr, ptent);
875 /* Clear accessed and referenced bits. */
876 ptep_test_and_clear_young(vma, addr, pte);
877 test_and_clear_page_young(page);
878 ClearPageReferenced(page);
880 pte_unmap_unlock(pte - 1, ptl);
885 static int clear_refs_test_walk(unsigned long start, unsigned long end,
886 struct mm_walk *walk)
888 struct clear_refs_private *cp = walk->private;
889 struct vm_area_struct *vma = walk->vma;
891 if (vma->vm_flags & VM_PFNMAP)
895 * Writing 1 to /proc/pid/clear_refs affects all pages.
896 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
897 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
898 * Writing 4 to /proc/pid/clear_refs affects all pages.
900 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
902 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
907 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
908 size_t count, loff_t *ppos)
910 struct task_struct *task;
911 char buffer[PROC_NUMBUF];
912 struct mm_struct *mm;
913 struct vm_area_struct *vma;
914 enum clear_refs_types type;
918 memset(buffer, 0, sizeof(buffer));
919 if (count > sizeof(buffer) - 1)
920 count = sizeof(buffer) - 1;
921 if (copy_from_user(buffer, buf, count))
923 rv = kstrtoint(strstrip(buffer), 10, &itype);
926 type = (enum clear_refs_types)itype;
927 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
930 task = get_proc_task(file_inode(file));
933 mm = get_task_mm(task);
935 struct clear_refs_private cp = {
938 struct mm_walk clear_refs_walk = {
939 .pmd_entry = clear_refs_pte_range,
940 .test_walk = clear_refs_test_walk,
945 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
947 * Writing 5 to /proc/pid/clear_refs resets the peak
948 * resident set size to this mm's current rss value.
950 down_write(&mm->mmap_sem);
951 reset_mm_hiwater_rss(mm);
952 up_write(&mm->mmap_sem);
956 down_read(&mm->mmap_sem);
957 if (type == CLEAR_REFS_SOFT_DIRTY) {
958 for (vma = mm->mmap; vma; vma = vma->vm_next) {
959 if (!(vma->vm_flags & VM_SOFTDIRTY))
961 up_read(&mm->mmap_sem);
962 down_write(&mm->mmap_sem);
963 for (vma = mm->mmap; vma; vma = vma->vm_next) {
964 vma->vm_flags &= ~VM_SOFTDIRTY;
965 vma_set_page_prot(vma);
967 downgrade_write(&mm->mmap_sem);
970 mmu_notifier_invalidate_range_start(mm, 0, -1);
972 walk_page_range(0, ~0UL, &clear_refs_walk);
973 if (type == CLEAR_REFS_SOFT_DIRTY)
974 mmu_notifier_invalidate_range_end(mm, 0, -1);
976 up_read(&mm->mmap_sem);
980 put_task_struct(task);
985 const struct file_operations proc_clear_refs_operations = {
986 .write = clear_refs_write,
987 .llseek = noop_llseek,
995 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
996 pagemap_entry_t *buffer;
1000 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
1001 #define PAGEMAP_WALK_MASK (PMD_MASK)
1003 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1004 #define PM_PFRAME_BITS 55
1005 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1006 #define PM_SOFT_DIRTY BIT_ULL(55)
1007 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1008 #define PM_FILE BIT_ULL(61)
1009 #define PM_SWAP BIT_ULL(62)
1010 #define PM_PRESENT BIT_ULL(63)
1012 #define PM_END_OF_BUFFER 1
1014 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1016 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1019 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1020 struct pagemapread *pm)
1022 pm->buffer[pm->pos++] = *pme;
1023 if (pm->pos >= pm->len)
1024 return PM_END_OF_BUFFER;
1028 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1029 struct mm_walk *walk)
1031 struct pagemapread *pm = walk->private;
1032 unsigned long addr = start;
1035 while (addr < end) {
1036 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1037 pagemap_entry_t pme = make_pme(0, 0);
1038 /* End of address space hole, which we mark as non-present. */
1039 unsigned long hole_end;
1042 hole_end = min(end, vma->vm_start);
1046 for (; addr < hole_end; addr += PAGE_SIZE) {
1047 err = add_to_pagemap(addr, &pme, pm);
1055 /* Addresses in the VMA. */
1056 if (vma->vm_flags & VM_SOFTDIRTY)
1057 pme = make_pme(0, PM_SOFT_DIRTY);
1058 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1059 err = add_to_pagemap(addr, &pme, pm);
1068 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1069 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1071 u64 frame = 0, flags = 0;
1072 struct page *page = NULL;
1074 if (pte_present(pte)) {
1076 frame = pte_pfn(pte);
1077 flags |= PM_PRESENT;
1078 page = vm_normal_page(vma, addr, pte);
1079 if (pte_soft_dirty(pte))
1080 flags |= PM_SOFT_DIRTY;
1081 } else if (is_swap_pte(pte)) {
1083 if (pte_swp_soft_dirty(pte))
1084 flags |= PM_SOFT_DIRTY;
1085 entry = pte_to_swp_entry(pte);
1086 frame = swp_type(entry) |
1087 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1089 if (is_migration_entry(entry))
1090 page = migration_entry_to_page(entry);
1093 if (page && !PageAnon(page))
1095 if (page && page_mapcount(page) == 1)
1096 flags |= PM_MMAP_EXCLUSIVE;
1097 if (vma->vm_flags & VM_SOFTDIRTY)
1098 flags |= PM_SOFT_DIRTY;
1100 return make_pme(frame, flags);
1103 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1104 struct mm_walk *walk)
1106 struct vm_area_struct *vma = walk->vma;
1107 struct pagemapread *pm = walk->private;
1109 pte_t *pte, *orig_pte;
1112 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1113 if (pmd_trans_huge_lock(pmdp, vma, &ptl) == 1) {
1114 u64 flags = 0, frame = 0;
1117 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1118 flags |= PM_SOFT_DIRTY;
1121 * Currently pmd for thp is always present because thp
1122 * can not be swapped-out, migrated, or HWPOISONed
1123 * (split in such cases instead.)
1124 * This if-check is just to prepare for future implementation.
1126 if (pmd_present(pmd)) {
1127 struct page *page = pmd_page(pmd);
1129 if (page_mapcount(page) == 1)
1130 flags |= PM_MMAP_EXCLUSIVE;
1132 flags |= PM_PRESENT;
1134 frame = pmd_pfn(pmd) +
1135 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1138 for (; addr != end; addr += PAGE_SIZE) {
1139 pagemap_entry_t pme = make_pme(frame, flags);
1141 err = add_to_pagemap(addr, &pme, pm);
1144 if (pm->show_pfn && (flags & PM_PRESENT))
1151 if (pmd_trans_unstable(pmdp))
1153 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1156 * We can assume that @vma always points to a valid one and @end never
1157 * goes beyond vma->vm_end.
1159 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1160 for (; addr < end; pte++, addr += PAGE_SIZE) {
1161 pagemap_entry_t pme;
1163 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1164 err = add_to_pagemap(addr, &pme, pm);
1168 pte_unmap_unlock(orig_pte, ptl);
1175 #ifdef CONFIG_HUGETLB_PAGE
1176 /* This function walks within one hugetlb entry in the single call */
1177 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1178 unsigned long addr, unsigned long end,
1179 struct mm_walk *walk)
1181 struct pagemapread *pm = walk->private;
1182 struct vm_area_struct *vma = walk->vma;
1183 u64 flags = 0, frame = 0;
1187 if (vma->vm_flags & VM_SOFTDIRTY)
1188 flags |= PM_SOFT_DIRTY;
1190 pte = huge_ptep_get(ptep);
1191 if (pte_present(pte)) {
1192 struct page *page = pte_page(pte);
1194 if (!PageAnon(page))
1197 if (page_mapcount(page) == 1)
1198 flags |= PM_MMAP_EXCLUSIVE;
1200 flags |= PM_PRESENT;
1202 frame = pte_pfn(pte) +
1203 ((addr & ~hmask) >> PAGE_SHIFT);
1206 for (; addr != end; addr += PAGE_SIZE) {
1207 pagemap_entry_t pme = make_pme(frame, flags);
1209 err = add_to_pagemap(addr, &pme, pm);
1212 if (pm->show_pfn && (flags & PM_PRESENT))
1220 #endif /* HUGETLB_PAGE */
1223 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1225 * For each page in the address space, this file contains one 64-bit entry
1226 * consisting of the following:
1228 * Bits 0-54 page frame number (PFN) if present
1229 * Bits 0-4 swap type if swapped
1230 * Bits 5-54 swap offset if swapped
1231 * Bit 55 pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1232 * Bit 56 page exclusively mapped
1234 * Bit 61 page is file-page or shared-anon
1235 * Bit 62 page swapped
1236 * Bit 63 page present
1238 * If the page is not present but in swap, then the PFN contains an
1239 * encoding of the swap file number and the page's offset into the
1240 * swap. Unmapped pages return a null PFN. This allows determining
1241 * precisely which pages are mapped (or in swap) and comparing mapped
1242 * pages between processes.
1244 * Efficient users of this interface will use /proc/pid/maps to
1245 * determine which areas of memory are actually mapped and llseek to
1246 * skip over unmapped regions.
1248 static ssize_t pagemap_read(struct file *file, char __user *buf,
1249 size_t count, loff_t *ppos)
1251 struct mm_struct *mm = file->private_data;
1252 struct pagemapread pm;
1253 struct mm_walk pagemap_walk = {};
1255 unsigned long svpfn;
1256 unsigned long start_vaddr;
1257 unsigned long end_vaddr;
1258 int ret = 0, copied = 0;
1260 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1264 /* file position must be aligned */
1265 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1272 /* do not disclose physical addresses: attack vector */
1273 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1275 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1276 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1281 pagemap_walk.pmd_entry = pagemap_pmd_range;
1282 pagemap_walk.pte_hole = pagemap_pte_hole;
1283 #ifdef CONFIG_HUGETLB_PAGE
1284 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1286 pagemap_walk.mm = mm;
1287 pagemap_walk.private = ±
1290 svpfn = src / PM_ENTRY_BYTES;
1291 start_vaddr = svpfn << PAGE_SHIFT;
1292 end_vaddr = mm->task_size;
1294 /* watch out for wraparound */
1295 if (svpfn > mm->task_size >> PAGE_SHIFT)
1296 start_vaddr = end_vaddr;
1299 * The odds are that this will stop walking way
1300 * before end_vaddr, because the length of the
1301 * user buffer is tracked in "pm", and the walk
1302 * will stop when we hit the end of the buffer.
1305 while (count && (start_vaddr < end_vaddr)) {
1310 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1312 if (end < start_vaddr || end > end_vaddr)
1314 down_read(&mm->mmap_sem);
1315 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1316 up_read(&mm->mmap_sem);
1319 len = min(count, PM_ENTRY_BYTES * pm.pos);
1320 if (copy_to_user(buf, pm.buffer, len)) {
1329 if (!ret || ret == PM_END_OF_BUFFER)
1340 static int pagemap_open(struct inode *inode, struct file *file)
1342 struct mm_struct *mm;
1344 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1347 file->private_data = mm;
1351 static int pagemap_release(struct inode *inode, struct file *file)
1353 struct mm_struct *mm = file->private_data;
1360 const struct file_operations proc_pagemap_operations = {
1361 .llseek = mem_lseek, /* borrow this */
1362 .read = pagemap_read,
1363 .open = pagemap_open,
1364 .release = pagemap_release,
1366 #endif /* CONFIG_PROC_PAGE_MONITOR */
1371 unsigned long pages;
1373 unsigned long active;
1374 unsigned long writeback;
1375 unsigned long mapcount_max;
1376 unsigned long dirty;
1377 unsigned long swapcache;
1378 unsigned long node[MAX_NUMNODES];
1381 struct numa_maps_private {
1382 struct proc_maps_private proc_maps;
1383 struct numa_maps md;
1386 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1387 unsigned long nr_pages)
1389 int count = page_mapcount(page);
1391 md->pages += nr_pages;
1392 if (pte_dirty || PageDirty(page))
1393 md->dirty += nr_pages;
1395 if (PageSwapCache(page))
1396 md->swapcache += nr_pages;
1398 if (PageActive(page) || PageUnevictable(page))
1399 md->active += nr_pages;
1401 if (PageWriteback(page))
1402 md->writeback += nr_pages;
1405 md->anon += nr_pages;
1407 if (count > md->mapcount_max)
1408 md->mapcount_max = count;
1410 md->node[page_to_nid(page)] += nr_pages;
1413 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1419 if (!pte_present(pte))
1422 page = vm_normal_page(vma, addr, pte);
1426 if (PageReserved(page))
1429 nid = page_to_nid(page);
1430 if (!node_isset(nid, node_states[N_MEMORY]))
1436 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1437 unsigned long end, struct mm_walk *walk)
1439 struct numa_maps *md = walk->private;
1440 struct vm_area_struct *vma = walk->vma;
1445 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1446 pte_t huge_pte = *(pte_t *)pmd;
1449 page = can_gather_numa_stats(huge_pte, vma, addr);
1451 gather_stats(page, md, pte_dirty(huge_pte),
1452 HPAGE_PMD_SIZE/PAGE_SIZE);
1457 if (pmd_trans_unstable(pmd))
1459 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1461 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1464 gather_stats(page, md, pte_dirty(*pte), 1);
1466 } while (pte++, addr += PAGE_SIZE, addr != end);
1467 pte_unmap_unlock(orig_pte, ptl);
1470 #ifdef CONFIG_HUGETLB_PAGE
1471 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1472 unsigned long addr, unsigned long end, struct mm_walk *walk)
1474 struct numa_maps *md;
1477 if (!pte_present(*pte))
1480 page = pte_page(*pte);
1485 gather_stats(page, md, pte_dirty(*pte), 1);
1490 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1491 unsigned long addr, unsigned long end, struct mm_walk *walk)
1498 * Display pages allocated per node and memory policy via /proc.
1500 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1502 struct numa_maps_private *numa_priv = m->private;
1503 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1504 struct vm_area_struct *vma = v;
1505 struct numa_maps *md = &numa_priv->md;
1506 struct file *file = vma->vm_file;
1507 struct mm_struct *mm = vma->vm_mm;
1508 struct mm_walk walk = {
1509 .hugetlb_entry = gather_hugetlb_stats,
1510 .pmd_entry = gather_pte_stats,
1514 struct mempolicy *pol;
1521 /* Ensure we start with an empty set of numa_maps statistics. */
1522 memset(md, 0, sizeof(*md));
1524 pol = __get_vma_policy(vma, vma->vm_start);
1526 mpol_to_str(buffer, sizeof(buffer), pol);
1529 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1532 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1535 seq_puts(m, " file=");
1536 seq_file_path(m, file, "\n\t= ");
1537 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1538 seq_puts(m, " heap");
1540 pid_t tid = pid_of_stack(proc_priv, vma, is_pid);
1543 * Thread stack in /proc/PID/task/TID/maps or
1544 * the main process stack.
1546 if (!is_pid || (vma->vm_start <= mm->start_stack &&
1547 vma->vm_end >= mm->start_stack))
1548 seq_puts(m, " stack");
1550 seq_printf(m, " stack:%d", tid);
1554 if (is_vm_hugetlb_page(vma))
1555 seq_puts(m, " huge");
1557 /* mmap_sem is held by m_start */
1558 walk_page_vma(vma, &walk);
1564 seq_printf(m, " anon=%lu", md->anon);
1567 seq_printf(m, " dirty=%lu", md->dirty);
1569 if (md->pages != md->anon && md->pages != md->dirty)
1570 seq_printf(m, " mapped=%lu", md->pages);
1572 if (md->mapcount_max > 1)
1573 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1576 seq_printf(m, " swapcache=%lu", md->swapcache);
1578 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1579 seq_printf(m, " active=%lu", md->active);
1582 seq_printf(m, " writeback=%lu", md->writeback);
1584 for_each_node_state(nid, N_MEMORY)
1586 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1588 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1591 m_cache_vma(m, vma);
1595 static int show_pid_numa_map(struct seq_file *m, void *v)
1597 return show_numa_map(m, v, 1);
1600 static int show_tid_numa_map(struct seq_file *m, void *v)
1602 return show_numa_map(m, v, 0);
1605 static const struct seq_operations proc_pid_numa_maps_op = {
1609 .show = show_pid_numa_map,
1612 static const struct seq_operations proc_tid_numa_maps_op = {
1616 .show = show_tid_numa_map,
1619 static int numa_maps_open(struct inode *inode, struct file *file,
1620 const struct seq_operations *ops)
1622 return proc_maps_open(inode, file, ops,
1623 sizeof(struct numa_maps_private));
1626 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1628 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1631 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1633 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1636 const struct file_operations proc_pid_numa_maps_operations = {
1637 .open = pid_numa_maps_open,
1639 .llseek = seq_lseek,
1640 .release = proc_map_release,
1643 const struct file_operations proc_tid_numa_maps_operations = {
1644 .open = tid_numa_maps_open,
1646 .llseek = seq_lseek,
1647 .release = proc_map_release,
1649 #endif /* CONFIG_NUMA */