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/sched/mm.h>
15 #include <linux/swapops.h>
16 #include <linux/mmu_notifier.h>
17 #include <linux/page_idle.h>
18 #include <linux/shmem_fs.h>
21 #include <linux/uaccess.h>
22 #include <asm/tlbflush.h>
25 void task_mem(struct seq_file *m, struct mm_struct *mm)
27 unsigned long text, lib, swap, ptes, pmds, anon, file, shmem;
28 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
30 anon = get_mm_counter(mm, MM_ANONPAGES);
31 file = get_mm_counter(mm, MM_FILEPAGES);
32 shmem = get_mm_counter(mm, MM_SHMEMPAGES);
35 * Note: to minimize their overhead, mm maintains hiwater_vm and
36 * hiwater_rss only when about to *lower* total_vm or rss. Any
37 * collector of these hiwater stats must therefore get total_vm
38 * and rss too, which will usually be the higher. Barriers? not
39 * worth the effort, such snapshots can always be inconsistent.
41 hiwater_vm = total_vm = mm->total_vm;
42 if (hiwater_vm < mm->hiwater_vm)
43 hiwater_vm = mm->hiwater_vm;
44 hiwater_rss = total_rss = anon + file + shmem;
45 if (hiwater_rss < mm->hiwater_rss)
46 hiwater_rss = mm->hiwater_rss;
48 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
49 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
50 swap = get_mm_counter(mm, MM_SWAPENTS);
51 ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
52 pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
62 "RssShmem:\t%8lu kB\n"
70 hiwater_vm << (PAGE_SHIFT-10),
71 total_vm << (PAGE_SHIFT-10),
72 mm->locked_vm << (PAGE_SHIFT-10),
73 mm->pinned_vm << (PAGE_SHIFT-10),
74 hiwater_rss << (PAGE_SHIFT-10),
75 total_rss << (PAGE_SHIFT-10),
76 anon << (PAGE_SHIFT-10),
77 file << (PAGE_SHIFT-10),
78 shmem << (PAGE_SHIFT-10),
79 mm->data_vm << (PAGE_SHIFT-10),
80 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
83 swap << (PAGE_SHIFT-10));
84 hugetlb_report_usage(m, mm);
87 unsigned long task_vsize(struct mm_struct *mm)
89 return PAGE_SIZE * mm->total_vm;
92 unsigned long task_statm(struct mm_struct *mm,
93 unsigned long *shared, unsigned long *text,
94 unsigned long *data, unsigned long *resident)
96 *shared = get_mm_counter(mm, MM_FILEPAGES) +
97 get_mm_counter(mm, MM_SHMEMPAGES);
98 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
100 *data = mm->data_vm + mm->stack_vm;
101 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
107 * Save get_task_policy() for show_numa_map().
109 static void hold_task_mempolicy(struct proc_maps_private *priv)
111 struct task_struct *task = priv->task;
114 priv->task_mempolicy = get_task_policy(task);
115 mpol_get(priv->task_mempolicy);
118 static void release_task_mempolicy(struct proc_maps_private *priv)
120 mpol_put(priv->task_mempolicy);
123 static void hold_task_mempolicy(struct proc_maps_private *priv)
126 static void release_task_mempolicy(struct proc_maps_private *priv)
131 static void vma_stop(struct proc_maps_private *priv)
133 struct mm_struct *mm = priv->mm;
135 release_task_mempolicy(priv);
136 up_read(&mm->mmap_sem);
140 static struct vm_area_struct *
141 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
143 if (vma == priv->tail_vma)
145 return vma->vm_next ?: priv->tail_vma;
148 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
150 if (m->count < m->size) /* vma is copied successfully */
151 m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL;
154 static void *m_start(struct seq_file *m, loff_t *ppos)
156 struct proc_maps_private *priv = m->private;
157 unsigned long last_addr = m->version;
158 struct mm_struct *mm;
159 struct vm_area_struct *vma;
160 unsigned int pos = *ppos;
162 /* See m_cache_vma(). Zero at the start or after lseek. */
163 if (last_addr == -1UL)
166 priv->task = get_proc_task(priv->inode);
168 return ERR_PTR(-ESRCH);
171 if (!mm || !mmget_not_zero(mm))
174 down_read(&mm->mmap_sem);
175 hold_task_mempolicy(priv);
176 priv->tail_vma = get_gate_vma(mm);
179 vma = find_vma(mm, last_addr - 1);
180 if (vma && vma->vm_start <= last_addr)
181 vma = m_next_vma(priv, vma);
187 if (pos < mm->map_count) {
188 for (vma = mm->mmap; pos; pos--) {
189 m->version = vma->vm_start;
195 /* we do not bother to update m->version in this case */
196 if (pos == mm->map_count && priv->tail_vma)
197 return priv->tail_vma;
203 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
205 struct proc_maps_private *priv = m->private;
206 struct vm_area_struct *next;
209 next = m_next_vma(priv, v);
215 static void m_stop(struct seq_file *m, void *v)
217 struct proc_maps_private *priv = m->private;
219 if (!IS_ERR_OR_NULL(v))
222 put_task_struct(priv->task);
227 static int proc_maps_open(struct inode *inode, struct file *file,
228 const struct seq_operations *ops, int psize)
230 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
236 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
237 if (IS_ERR(priv->mm)) {
238 int err = PTR_ERR(priv->mm);
240 seq_release_private(inode, file);
247 static int proc_map_release(struct inode *inode, struct file *file)
249 struct seq_file *seq = file->private_data;
250 struct proc_maps_private *priv = seq->private;
255 return seq_release_private(inode, file);
258 static int do_maps_open(struct inode *inode, struct file *file,
259 const struct seq_operations *ops)
261 return proc_maps_open(inode, file, ops,
262 sizeof(struct proc_maps_private));
266 * Indicate if the VMA is a stack for the given task; for
267 * /proc/PID/maps that is the stack of the main task.
269 static int is_stack(struct proc_maps_private *priv,
270 struct vm_area_struct *vma)
273 * We make no effort to guess what a given thread considers to be
274 * its "stack". It's not even well-defined for programs written
277 return vma->vm_start <= vma->vm_mm->start_stack &&
278 vma->vm_end >= vma->vm_mm->start_stack;
282 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
284 struct mm_struct *mm = vma->vm_mm;
285 struct file *file = vma->vm_file;
286 struct proc_maps_private *priv = m->private;
287 vm_flags_t flags = vma->vm_flags;
288 unsigned long ino = 0;
289 unsigned long long pgoff = 0;
290 unsigned long start, end;
292 const char *name = NULL;
295 struct inode *inode = file_inode(vma->vm_file);
296 dev = inode->i_sb->s_dev;
298 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
301 /* We don't show the stack guard page in /proc/maps */
302 start = vma->vm_start;
303 if (stack_guard_page_start(vma, start))
306 if (stack_guard_page_end(vma, end))
309 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
310 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
313 flags & VM_READ ? 'r' : '-',
314 flags & VM_WRITE ? 'w' : '-',
315 flags & VM_EXEC ? 'x' : '-',
316 flags & VM_MAYSHARE ? 's' : 'p',
318 MAJOR(dev), MINOR(dev), ino);
321 * Print the dentry name for named mappings, and a
322 * special [heap] marker for the heap:
326 seq_file_path(m, file, "\n");
330 if (vma->vm_ops && vma->vm_ops->name) {
331 name = vma->vm_ops->name(vma);
336 name = arch_vma_name(vma);
343 if (vma->vm_start <= mm->brk &&
344 vma->vm_end >= mm->start_brk) {
349 if (is_stack(priv, vma))
361 static int show_map(struct seq_file *m, void *v, int is_pid)
363 show_map_vma(m, v, is_pid);
368 static int show_pid_map(struct seq_file *m, void *v)
370 return show_map(m, v, 1);
373 static int show_tid_map(struct seq_file *m, void *v)
375 return show_map(m, v, 0);
378 static const struct seq_operations proc_pid_maps_op = {
385 static const struct seq_operations proc_tid_maps_op = {
392 static int pid_maps_open(struct inode *inode, struct file *file)
394 return do_maps_open(inode, file, &proc_pid_maps_op);
397 static int tid_maps_open(struct inode *inode, struct file *file)
399 return do_maps_open(inode, file, &proc_tid_maps_op);
402 const struct file_operations proc_pid_maps_operations = {
403 .open = pid_maps_open,
406 .release = proc_map_release,
409 const struct file_operations proc_tid_maps_operations = {
410 .open = tid_maps_open,
413 .release = proc_map_release,
417 * Proportional Set Size(PSS): my share of RSS.
419 * PSS of a process is the count of pages it has in memory, where each
420 * page is divided by the number of processes sharing it. So if a
421 * process has 1000 pages all to itself, and 1000 shared with one other
422 * process, its PSS will be 1500.
424 * To keep (accumulated) division errors low, we adopt a 64bit
425 * fixed-point pss counter to minimize division errors. So (pss >>
426 * PSS_SHIFT) would be the real byte count.
428 * A shift of 12 before division means (assuming 4K page size):
429 * - 1M 3-user-pages add up to 8KB errors;
430 * - supports mapcount up to 2^24, or 16M;
431 * - supports PSS up to 2^52 bytes, or 4PB.
435 #ifdef CONFIG_PROC_PAGE_MONITOR
436 struct mem_size_stats {
437 unsigned long resident;
438 unsigned long shared_clean;
439 unsigned long shared_dirty;
440 unsigned long private_clean;
441 unsigned long private_dirty;
442 unsigned long referenced;
443 unsigned long anonymous;
444 unsigned long lazyfree;
445 unsigned long anonymous_thp;
446 unsigned long shmem_thp;
448 unsigned long shared_hugetlb;
449 unsigned long private_hugetlb;
452 bool check_shmem_swap;
455 static void smaps_account(struct mem_size_stats *mss, struct page *page,
456 bool compound, bool young, bool dirty)
458 int i, nr = compound ? 1 << compound_order(page) : 1;
459 unsigned long size = nr * PAGE_SIZE;
461 if (PageAnon(page)) {
462 mss->anonymous += size;
463 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
464 mss->lazyfree += size;
467 mss->resident += size;
468 /* Accumulate the size in pages that have been accessed. */
469 if (young || page_is_young(page) || PageReferenced(page))
470 mss->referenced += size;
473 * page_count(page) == 1 guarantees the page is mapped exactly once.
474 * If any subpage of the compound page mapped with PTE it would elevate
477 if (page_count(page) == 1) {
478 if (dirty || PageDirty(page))
479 mss->private_dirty += size;
481 mss->private_clean += size;
482 mss->pss += (u64)size << PSS_SHIFT;
486 for (i = 0; i < nr; i++, page++) {
487 int mapcount = page_mapcount(page);
490 if (dirty || PageDirty(page))
491 mss->shared_dirty += PAGE_SIZE;
493 mss->shared_clean += PAGE_SIZE;
494 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
496 if (dirty || PageDirty(page))
497 mss->private_dirty += PAGE_SIZE;
499 mss->private_clean += PAGE_SIZE;
500 mss->pss += PAGE_SIZE << PSS_SHIFT;
506 static int smaps_pte_hole(unsigned long addr, unsigned long end,
507 struct mm_walk *walk)
509 struct mem_size_stats *mss = walk->private;
511 mss->swap += shmem_partial_swap_usage(
512 walk->vma->vm_file->f_mapping, addr, end);
518 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
519 struct mm_walk *walk)
521 struct mem_size_stats *mss = walk->private;
522 struct vm_area_struct *vma = walk->vma;
523 struct page *page = NULL;
525 if (pte_present(*pte)) {
526 page = vm_normal_page(vma, addr, *pte);
527 } else if (is_swap_pte(*pte)) {
528 swp_entry_t swpent = pte_to_swp_entry(*pte);
530 if (!non_swap_entry(swpent)) {
533 mss->swap += PAGE_SIZE;
534 mapcount = swp_swapcount(swpent);
536 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
538 do_div(pss_delta, mapcount);
539 mss->swap_pss += pss_delta;
541 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
543 } else if (is_migration_entry(swpent))
544 page = migration_entry_to_page(swpent);
545 } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
546 && pte_none(*pte))) {
547 page = find_get_entry(vma->vm_file->f_mapping,
548 linear_page_index(vma, addr));
552 if (radix_tree_exceptional_entry(page))
553 mss->swap += PAGE_SIZE;
563 smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
566 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
567 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
568 struct mm_walk *walk)
570 struct mem_size_stats *mss = walk->private;
571 struct vm_area_struct *vma = walk->vma;
574 /* FOLL_DUMP will return -EFAULT on huge zero page */
575 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
576 if (IS_ERR_OR_NULL(page))
579 mss->anonymous_thp += HPAGE_PMD_SIZE;
580 else if (PageSwapBacked(page))
581 mss->shmem_thp += HPAGE_PMD_SIZE;
582 else if (is_zone_device_page(page))
585 VM_BUG_ON_PAGE(1, page);
586 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
589 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
590 struct mm_walk *walk)
595 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
596 struct mm_walk *walk)
598 struct vm_area_struct *vma = walk->vma;
602 ptl = pmd_trans_huge_lock(pmd, vma);
604 smaps_pmd_entry(pmd, addr, walk);
609 if (pmd_trans_unstable(pmd))
612 * The mmap_sem held all the way back in m_start() is what
613 * keeps khugepaged out of here and from collapsing things
616 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
617 for (; addr != end; pte++, addr += PAGE_SIZE)
618 smaps_pte_entry(pte, addr, walk);
619 pte_unmap_unlock(pte - 1, ptl);
624 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
627 * Don't forget to update Documentation/ on changes.
629 static const char mnemonics[BITS_PER_LONG][2] = {
631 * In case if we meet a flag we don't know about.
633 [0 ... (BITS_PER_LONG-1)] = "??",
635 [ilog2(VM_READ)] = "rd",
636 [ilog2(VM_WRITE)] = "wr",
637 [ilog2(VM_EXEC)] = "ex",
638 [ilog2(VM_SHARED)] = "sh",
639 [ilog2(VM_MAYREAD)] = "mr",
640 [ilog2(VM_MAYWRITE)] = "mw",
641 [ilog2(VM_MAYEXEC)] = "me",
642 [ilog2(VM_MAYSHARE)] = "ms",
643 [ilog2(VM_GROWSDOWN)] = "gd",
644 [ilog2(VM_PFNMAP)] = "pf",
645 [ilog2(VM_DENYWRITE)] = "dw",
646 #ifdef CONFIG_X86_INTEL_MPX
647 [ilog2(VM_MPX)] = "mp",
649 [ilog2(VM_LOCKED)] = "lo",
650 [ilog2(VM_IO)] = "io",
651 [ilog2(VM_SEQ_READ)] = "sr",
652 [ilog2(VM_RAND_READ)] = "rr",
653 [ilog2(VM_DONTCOPY)] = "dc",
654 [ilog2(VM_DONTEXPAND)] = "de",
655 [ilog2(VM_ACCOUNT)] = "ac",
656 [ilog2(VM_NORESERVE)] = "nr",
657 [ilog2(VM_HUGETLB)] = "ht",
658 [ilog2(VM_ARCH_1)] = "ar",
659 [ilog2(VM_DONTDUMP)] = "dd",
660 #ifdef CONFIG_MEM_SOFT_DIRTY
661 [ilog2(VM_SOFTDIRTY)] = "sd",
663 [ilog2(VM_MIXEDMAP)] = "mm",
664 [ilog2(VM_HUGEPAGE)] = "hg",
665 [ilog2(VM_NOHUGEPAGE)] = "nh",
666 [ilog2(VM_MERGEABLE)] = "mg",
667 [ilog2(VM_UFFD_MISSING)]= "um",
668 [ilog2(VM_UFFD_WP)] = "uw",
669 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
670 /* These come out via ProtectionKey: */
671 [ilog2(VM_PKEY_BIT0)] = "",
672 [ilog2(VM_PKEY_BIT1)] = "",
673 [ilog2(VM_PKEY_BIT2)] = "",
674 [ilog2(VM_PKEY_BIT3)] = "",
679 seq_puts(m, "VmFlags: ");
680 for (i = 0; i < BITS_PER_LONG; i++) {
681 if (!mnemonics[i][0])
683 if (vma->vm_flags & (1UL << i)) {
684 seq_printf(m, "%c%c ",
685 mnemonics[i][0], mnemonics[i][1]);
691 #ifdef CONFIG_HUGETLB_PAGE
692 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
693 unsigned long addr, unsigned long end,
694 struct mm_walk *walk)
696 struct mem_size_stats *mss = walk->private;
697 struct vm_area_struct *vma = walk->vma;
698 struct page *page = NULL;
700 if (pte_present(*pte)) {
701 page = vm_normal_page(vma, addr, *pte);
702 } else if (is_swap_pte(*pte)) {
703 swp_entry_t swpent = pte_to_swp_entry(*pte);
705 if (is_migration_entry(swpent))
706 page = migration_entry_to_page(swpent);
709 int mapcount = page_mapcount(page);
712 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
714 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
718 #endif /* HUGETLB_PAGE */
720 void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
724 static int show_smap(struct seq_file *m, void *v, int is_pid)
726 struct vm_area_struct *vma = v;
727 struct mem_size_stats mss;
728 struct mm_walk smaps_walk = {
729 .pmd_entry = smaps_pte_range,
730 #ifdef CONFIG_HUGETLB_PAGE
731 .hugetlb_entry = smaps_hugetlb_range,
737 memset(&mss, 0, sizeof mss);
740 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
742 * For shared or readonly shmem mappings we know that all
743 * swapped out pages belong to the shmem object, and we can
744 * obtain the swap value much more efficiently. For private
745 * writable mappings, we might have COW pages that are
746 * not affected by the parent swapped out pages of the shmem
747 * object, so we have to distinguish them during the page walk.
748 * Unless we know that the shmem object (or the part mapped by
749 * our VMA) has no swapped out pages at all.
751 unsigned long shmem_swapped = shmem_swap_usage(vma);
753 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
754 !(vma->vm_flags & VM_WRITE)) {
755 mss.swap = shmem_swapped;
757 mss.check_shmem_swap = true;
758 smaps_walk.pte_hole = smaps_pte_hole;
763 /* mmap_sem is held in m_start */
764 walk_page_vma(vma, &smaps_walk);
766 show_map_vma(m, vma, is_pid);
772 "Shared_Clean: %8lu kB\n"
773 "Shared_Dirty: %8lu kB\n"
774 "Private_Clean: %8lu kB\n"
775 "Private_Dirty: %8lu kB\n"
776 "Referenced: %8lu kB\n"
777 "Anonymous: %8lu kB\n"
778 "LazyFree: %8lu kB\n"
779 "AnonHugePages: %8lu kB\n"
780 "ShmemPmdMapped: %8lu kB\n"
781 "Shared_Hugetlb: %8lu kB\n"
782 "Private_Hugetlb: %7lu kB\n"
785 "KernelPageSize: %8lu kB\n"
786 "MMUPageSize: %8lu kB\n"
788 (vma->vm_end - vma->vm_start) >> 10,
790 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
791 mss.shared_clean >> 10,
792 mss.shared_dirty >> 10,
793 mss.private_clean >> 10,
794 mss.private_dirty >> 10,
795 mss.referenced >> 10,
798 mss.anonymous_thp >> 10,
800 mss.shared_hugetlb >> 10,
801 mss.private_hugetlb >> 10,
803 (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
804 vma_kernel_pagesize(vma) >> 10,
805 vma_mmu_pagesize(vma) >> 10,
806 (vma->vm_flags & VM_LOCKED) ?
807 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
809 arch_show_smap(m, vma);
810 show_smap_vma_flags(m, vma);
815 static int show_pid_smap(struct seq_file *m, void *v)
817 return show_smap(m, v, 1);
820 static int show_tid_smap(struct seq_file *m, void *v)
822 return show_smap(m, v, 0);
825 static const struct seq_operations proc_pid_smaps_op = {
829 .show = show_pid_smap
832 static const struct seq_operations proc_tid_smaps_op = {
836 .show = show_tid_smap
839 static int pid_smaps_open(struct inode *inode, struct file *file)
841 return do_maps_open(inode, file, &proc_pid_smaps_op);
844 static int tid_smaps_open(struct inode *inode, struct file *file)
846 return do_maps_open(inode, file, &proc_tid_smaps_op);
849 const struct file_operations proc_pid_smaps_operations = {
850 .open = pid_smaps_open,
853 .release = proc_map_release,
856 const struct file_operations proc_tid_smaps_operations = {
857 .open = tid_smaps_open,
860 .release = proc_map_release,
863 enum clear_refs_types {
867 CLEAR_REFS_SOFT_DIRTY,
868 CLEAR_REFS_MM_HIWATER_RSS,
872 struct clear_refs_private {
873 enum clear_refs_types type;
876 #ifdef CONFIG_MEM_SOFT_DIRTY
877 static inline void clear_soft_dirty(struct vm_area_struct *vma,
878 unsigned long addr, pte_t *pte)
881 * The soft-dirty tracker uses #PF-s to catch writes
882 * to pages, so write-protect the pte as well. See the
883 * Documentation/vm/soft-dirty.txt for full description
884 * of how soft-dirty works.
888 if (pte_present(ptent)) {
889 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
890 ptent = pte_wrprotect(ptent);
891 ptent = pte_clear_soft_dirty(ptent);
892 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
893 } else if (is_swap_pte(ptent)) {
894 ptent = pte_swp_clear_soft_dirty(ptent);
895 set_pte_at(vma->vm_mm, addr, pte, ptent);
899 static inline void clear_soft_dirty(struct vm_area_struct *vma,
900 unsigned long addr, pte_t *pte)
905 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
906 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
907 unsigned long addr, pmd_t *pmdp)
911 /* See comment in change_huge_pmd() */
912 pmdp_invalidate(vma, addr, pmdp);
913 if (pmd_dirty(*pmdp))
914 pmd = pmd_mkdirty(pmd);
915 if (pmd_young(*pmdp))
916 pmd = pmd_mkyoung(pmd);
918 pmd = pmd_wrprotect(pmd);
919 pmd = pmd_clear_soft_dirty(pmd);
921 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
924 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
925 unsigned long addr, pmd_t *pmdp)
930 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
931 unsigned long end, struct mm_walk *walk)
933 struct clear_refs_private *cp = walk->private;
934 struct vm_area_struct *vma = walk->vma;
939 ptl = pmd_trans_huge_lock(pmd, vma);
941 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
942 clear_soft_dirty_pmd(vma, addr, pmd);
946 page = pmd_page(*pmd);
948 /* Clear accessed and referenced bits. */
949 pmdp_test_and_clear_young(vma, addr, pmd);
950 test_and_clear_page_young(page);
951 ClearPageReferenced(page);
957 if (pmd_trans_unstable(pmd))
960 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
961 for (; addr != end; pte++, addr += PAGE_SIZE) {
964 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
965 clear_soft_dirty(vma, addr, pte);
969 if (!pte_present(ptent))
972 page = vm_normal_page(vma, addr, ptent);
976 /* Clear accessed and referenced bits. */
977 ptep_test_and_clear_young(vma, addr, pte);
978 test_and_clear_page_young(page);
979 ClearPageReferenced(page);
981 pte_unmap_unlock(pte - 1, ptl);
986 static int clear_refs_test_walk(unsigned long start, unsigned long end,
987 struct mm_walk *walk)
989 struct clear_refs_private *cp = walk->private;
990 struct vm_area_struct *vma = walk->vma;
992 if (vma->vm_flags & VM_PFNMAP)
996 * Writing 1 to /proc/pid/clear_refs affects all pages.
997 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
998 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
999 * Writing 4 to /proc/pid/clear_refs affects all pages.
1001 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1003 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1008 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1009 size_t count, loff_t *ppos)
1011 struct task_struct *task;
1012 char buffer[PROC_NUMBUF];
1013 struct mm_struct *mm;
1014 struct vm_area_struct *vma;
1015 enum clear_refs_types type;
1019 memset(buffer, 0, sizeof(buffer));
1020 if (count > sizeof(buffer) - 1)
1021 count = sizeof(buffer) - 1;
1022 if (copy_from_user(buffer, buf, count))
1024 rv = kstrtoint(strstrip(buffer), 10, &itype);
1027 type = (enum clear_refs_types)itype;
1028 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1031 task = get_proc_task(file_inode(file));
1034 mm = get_task_mm(task);
1036 struct clear_refs_private cp = {
1039 struct mm_walk clear_refs_walk = {
1040 .pmd_entry = clear_refs_pte_range,
1041 .test_walk = clear_refs_test_walk,
1046 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1047 if (down_write_killable(&mm->mmap_sem)) {
1053 * Writing 5 to /proc/pid/clear_refs resets the peak
1054 * resident set size to this mm's current rss value.
1056 reset_mm_hiwater_rss(mm);
1057 up_write(&mm->mmap_sem);
1061 down_read(&mm->mmap_sem);
1062 if (type == CLEAR_REFS_SOFT_DIRTY) {
1063 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1064 if (!(vma->vm_flags & VM_SOFTDIRTY))
1066 up_read(&mm->mmap_sem);
1067 if (down_write_killable(&mm->mmap_sem)) {
1071 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1072 vma->vm_flags &= ~VM_SOFTDIRTY;
1073 vma_set_page_prot(vma);
1075 downgrade_write(&mm->mmap_sem);
1078 mmu_notifier_invalidate_range_start(mm, 0, -1);
1080 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1081 if (type == CLEAR_REFS_SOFT_DIRTY)
1082 mmu_notifier_invalidate_range_end(mm, 0, -1);
1084 up_read(&mm->mmap_sem);
1088 put_task_struct(task);
1093 const struct file_operations proc_clear_refs_operations = {
1094 .write = clear_refs_write,
1095 .llseek = noop_llseek,
1102 struct pagemapread {
1103 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1104 pagemap_entry_t *buffer;
1108 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
1109 #define PAGEMAP_WALK_MASK (PMD_MASK)
1111 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1112 #define PM_PFRAME_BITS 55
1113 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1114 #define PM_SOFT_DIRTY BIT_ULL(55)
1115 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1116 #define PM_FILE BIT_ULL(61)
1117 #define PM_SWAP BIT_ULL(62)
1118 #define PM_PRESENT BIT_ULL(63)
1120 #define PM_END_OF_BUFFER 1
1122 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1124 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1127 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1128 struct pagemapread *pm)
1130 pm->buffer[pm->pos++] = *pme;
1131 if (pm->pos >= pm->len)
1132 return PM_END_OF_BUFFER;
1136 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1137 struct mm_walk *walk)
1139 struct pagemapread *pm = walk->private;
1140 unsigned long addr = start;
1143 while (addr < end) {
1144 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1145 pagemap_entry_t pme = make_pme(0, 0);
1146 /* End of address space hole, which we mark as non-present. */
1147 unsigned long hole_end;
1150 hole_end = min(end, vma->vm_start);
1154 for (; addr < hole_end; addr += PAGE_SIZE) {
1155 err = add_to_pagemap(addr, &pme, pm);
1163 /* Addresses in the VMA. */
1164 if (vma->vm_flags & VM_SOFTDIRTY)
1165 pme = make_pme(0, PM_SOFT_DIRTY);
1166 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1167 err = add_to_pagemap(addr, &pme, pm);
1176 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1177 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1179 u64 frame = 0, flags = 0;
1180 struct page *page = NULL;
1182 if (pte_present(pte)) {
1184 frame = pte_pfn(pte);
1185 flags |= PM_PRESENT;
1186 page = vm_normal_page(vma, addr, pte);
1187 if (pte_soft_dirty(pte))
1188 flags |= PM_SOFT_DIRTY;
1189 } else if (is_swap_pte(pte)) {
1191 if (pte_swp_soft_dirty(pte))
1192 flags |= PM_SOFT_DIRTY;
1193 entry = pte_to_swp_entry(pte);
1194 frame = swp_type(entry) |
1195 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1197 if (is_migration_entry(entry))
1198 page = migration_entry_to_page(entry);
1201 if (page && !PageAnon(page))
1203 if (page && page_mapcount(page) == 1)
1204 flags |= PM_MMAP_EXCLUSIVE;
1205 if (vma->vm_flags & VM_SOFTDIRTY)
1206 flags |= PM_SOFT_DIRTY;
1208 return make_pme(frame, flags);
1211 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1212 struct mm_walk *walk)
1214 struct vm_area_struct *vma = walk->vma;
1215 struct pagemapread *pm = walk->private;
1217 pte_t *pte, *orig_pte;
1220 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1221 ptl = pmd_trans_huge_lock(pmdp, vma);
1223 u64 flags = 0, frame = 0;
1226 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1227 flags |= PM_SOFT_DIRTY;
1230 * Currently pmd for thp is always present because thp
1231 * can not be swapped-out, migrated, or HWPOISONed
1232 * (split in such cases instead.)
1233 * This if-check is just to prepare for future implementation.
1235 if (pmd_present(pmd)) {
1236 struct page *page = pmd_page(pmd);
1238 if (page_mapcount(page) == 1)
1239 flags |= PM_MMAP_EXCLUSIVE;
1241 flags |= PM_PRESENT;
1243 frame = pmd_pfn(pmd) +
1244 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1247 for (; addr != end; addr += PAGE_SIZE) {
1248 pagemap_entry_t pme = make_pme(frame, flags);
1250 err = add_to_pagemap(addr, &pme, pm);
1253 if (pm->show_pfn && (flags & PM_PRESENT))
1260 if (pmd_trans_unstable(pmdp))
1262 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1265 * We can assume that @vma always points to a valid one and @end never
1266 * goes beyond vma->vm_end.
1268 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1269 for (; addr < end; pte++, addr += PAGE_SIZE) {
1270 pagemap_entry_t pme;
1272 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1273 err = add_to_pagemap(addr, &pme, pm);
1277 pte_unmap_unlock(orig_pte, ptl);
1284 #ifdef CONFIG_HUGETLB_PAGE
1285 /* This function walks within one hugetlb entry in the single call */
1286 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1287 unsigned long addr, unsigned long end,
1288 struct mm_walk *walk)
1290 struct pagemapread *pm = walk->private;
1291 struct vm_area_struct *vma = walk->vma;
1292 u64 flags = 0, frame = 0;
1296 if (vma->vm_flags & VM_SOFTDIRTY)
1297 flags |= PM_SOFT_DIRTY;
1299 pte = huge_ptep_get(ptep);
1300 if (pte_present(pte)) {
1301 struct page *page = pte_page(pte);
1303 if (!PageAnon(page))
1306 if (page_mapcount(page) == 1)
1307 flags |= PM_MMAP_EXCLUSIVE;
1309 flags |= PM_PRESENT;
1311 frame = pte_pfn(pte) +
1312 ((addr & ~hmask) >> PAGE_SHIFT);
1315 for (; addr != end; addr += PAGE_SIZE) {
1316 pagemap_entry_t pme = make_pme(frame, flags);
1318 err = add_to_pagemap(addr, &pme, pm);
1321 if (pm->show_pfn && (flags & PM_PRESENT))
1329 #endif /* HUGETLB_PAGE */
1332 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1334 * For each page in the address space, this file contains one 64-bit entry
1335 * consisting of the following:
1337 * Bits 0-54 page frame number (PFN) if present
1338 * Bits 0-4 swap type if swapped
1339 * Bits 5-54 swap offset if swapped
1340 * Bit 55 pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1341 * Bit 56 page exclusively mapped
1343 * Bit 61 page is file-page or shared-anon
1344 * Bit 62 page swapped
1345 * Bit 63 page present
1347 * If the page is not present but in swap, then the PFN contains an
1348 * encoding of the swap file number and the page's offset into the
1349 * swap. Unmapped pages return a null PFN. This allows determining
1350 * precisely which pages are mapped (or in swap) and comparing mapped
1351 * pages between processes.
1353 * Efficient users of this interface will use /proc/pid/maps to
1354 * determine which areas of memory are actually mapped and llseek to
1355 * skip over unmapped regions.
1357 static ssize_t pagemap_read(struct file *file, char __user *buf,
1358 size_t count, loff_t *ppos)
1360 struct mm_struct *mm = file->private_data;
1361 struct pagemapread pm;
1362 struct mm_walk pagemap_walk = {};
1364 unsigned long svpfn;
1365 unsigned long start_vaddr;
1366 unsigned long end_vaddr;
1367 int ret = 0, copied = 0;
1369 if (!mm || !mmget_not_zero(mm))
1373 /* file position must be aligned */
1374 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1381 /* do not disclose physical addresses: attack vector */
1382 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1384 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1385 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1390 pagemap_walk.pmd_entry = pagemap_pmd_range;
1391 pagemap_walk.pte_hole = pagemap_pte_hole;
1392 #ifdef CONFIG_HUGETLB_PAGE
1393 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1395 pagemap_walk.mm = mm;
1396 pagemap_walk.private = ±
1399 svpfn = src / PM_ENTRY_BYTES;
1400 start_vaddr = svpfn << PAGE_SHIFT;
1401 end_vaddr = mm->task_size;
1403 /* watch out for wraparound */
1404 if (svpfn > mm->task_size >> PAGE_SHIFT)
1405 start_vaddr = end_vaddr;
1408 * The odds are that this will stop walking way
1409 * before end_vaddr, because the length of the
1410 * user buffer is tracked in "pm", and the walk
1411 * will stop when we hit the end of the buffer.
1414 while (count && (start_vaddr < end_vaddr)) {
1419 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1421 if (end < start_vaddr || end > end_vaddr)
1423 down_read(&mm->mmap_sem);
1424 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1425 up_read(&mm->mmap_sem);
1428 len = min(count, PM_ENTRY_BYTES * pm.pos);
1429 if (copy_to_user(buf, pm.buffer, len)) {
1438 if (!ret || ret == PM_END_OF_BUFFER)
1449 static int pagemap_open(struct inode *inode, struct file *file)
1451 struct mm_struct *mm;
1453 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1456 file->private_data = mm;
1460 static int pagemap_release(struct inode *inode, struct file *file)
1462 struct mm_struct *mm = file->private_data;
1469 const struct file_operations proc_pagemap_operations = {
1470 .llseek = mem_lseek, /* borrow this */
1471 .read = pagemap_read,
1472 .open = pagemap_open,
1473 .release = pagemap_release,
1475 #endif /* CONFIG_PROC_PAGE_MONITOR */
1480 unsigned long pages;
1482 unsigned long active;
1483 unsigned long writeback;
1484 unsigned long mapcount_max;
1485 unsigned long dirty;
1486 unsigned long swapcache;
1487 unsigned long node[MAX_NUMNODES];
1490 struct numa_maps_private {
1491 struct proc_maps_private proc_maps;
1492 struct numa_maps md;
1495 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1496 unsigned long nr_pages)
1498 int count = page_mapcount(page);
1500 md->pages += nr_pages;
1501 if (pte_dirty || PageDirty(page))
1502 md->dirty += nr_pages;
1504 if (PageSwapCache(page))
1505 md->swapcache += nr_pages;
1507 if (PageActive(page) || PageUnevictable(page))
1508 md->active += nr_pages;
1510 if (PageWriteback(page))
1511 md->writeback += nr_pages;
1514 md->anon += nr_pages;
1516 if (count > md->mapcount_max)
1517 md->mapcount_max = count;
1519 md->node[page_to_nid(page)] += nr_pages;
1522 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1528 if (!pte_present(pte))
1531 page = vm_normal_page(vma, addr, pte);
1535 if (PageReserved(page))
1538 nid = page_to_nid(page);
1539 if (!node_isset(nid, node_states[N_MEMORY]))
1545 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1546 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1547 struct vm_area_struct *vma,
1553 if (!pmd_present(pmd))
1556 page = vm_normal_page_pmd(vma, addr, pmd);
1560 if (PageReserved(page))
1563 nid = page_to_nid(page);
1564 if (!node_isset(nid, node_states[N_MEMORY]))
1571 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1572 unsigned long end, struct mm_walk *walk)
1574 struct numa_maps *md = walk->private;
1575 struct vm_area_struct *vma = walk->vma;
1580 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1581 ptl = pmd_trans_huge_lock(pmd, vma);
1585 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1587 gather_stats(page, md, pmd_dirty(*pmd),
1588 HPAGE_PMD_SIZE/PAGE_SIZE);
1593 if (pmd_trans_unstable(pmd))
1596 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1598 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1601 gather_stats(page, md, pte_dirty(*pte), 1);
1603 } while (pte++, addr += PAGE_SIZE, addr != end);
1604 pte_unmap_unlock(orig_pte, ptl);
1608 #ifdef CONFIG_HUGETLB_PAGE
1609 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1610 unsigned long addr, unsigned long end, struct mm_walk *walk)
1612 pte_t huge_pte = huge_ptep_get(pte);
1613 struct numa_maps *md;
1616 if (!pte_present(huge_pte))
1619 page = pte_page(huge_pte);
1624 gather_stats(page, md, pte_dirty(huge_pte), 1);
1629 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1630 unsigned long addr, unsigned long end, struct mm_walk *walk)
1637 * Display pages allocated per node and memory policy via /proc.
1639 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1641 struct numa_maps_private *numa_priv = m->private;
1642 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1643 struct vm_area_struct *vma = v;
1644 struct numa_maps *md = &numa_priv->md;
1645 struct file *file = vma->vm_file;
1646 struct mm_struct *mm = vma->vm_mm;
1647 struct mm_walk walk = {
1648 .hugetlb_entry = gather_hugetlb_stats,
1649 .pmd_entry = gather_pte_stats,
1653 struct mempolicy *pol;
1660 /* Ensure we start with an empty set of numa_maps statistics. */
1661 memset(md, 0, sizeof(*md));
1663 pol = __get_vma_policy(vma, vma->vm_start);
1665 mpol_to_str(buffer, sizeof(buffer), pol);
1668 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1671 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1674 seq_puts(m, " file=");
1675 seq_file_path(m, file, "\n\t= ");
1676 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1677 seq_puts(m, " heap");
1678 } else if (is_stack(proc_priv, vma)) {
1679 seq_puts(m, " stack");
1682 if (is_vm_hugetlb_page(vma))
1683 seq_puts(m, " huge");
1685 /* mmap_sem is held by m_start */
1686 walk_page_vma(vma, &walk);
1692 seq_printf(m, " anon=%lu", md->anon);
1695 seq_printf(m, " dirty=%lu", md->dirty);
1697 if (md->pages != md->anon && md->pages != md->dirty)
1698 seq_printf(m, " mapped=%lu", md->pages);
1700 if (md->mapcount_max > 1)
1701 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1704 seq_printf(m, " swapcache=%lu", md->swapcache);
1706 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1707 seq_printf(m, " active=%lu", md->active);
1710 seq_printf(m, " writeback=%lu", md->writeback);
1712 for_each_node_state(nid, N_MEMORY)
1714 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1716 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1719 m_cache_vma(m, vma);
1723 static int show_pid_numa_map(struct seq_file *m, void *v)
1725 return show_numa_map(m, v, 1);
1728 static int show_tid_numa_map(struct seq_file *m, void *v)
1730 return show_numa_map(m, v, 0);
1733 static const struct seq_operations proc_pid_numa_maps_op = {
1737 .show = show_pid_numa_map,
1740 static const struct seq_operations proc_tid_numa_maps_op = {
1744 .show = show_tid_numa_map,
1747 static int numa_maps_open(struct inode *inode, struct file *file,
1748 const struct seq_operations *ops)
1750 return proc_maps_open(inode, file, ops,
1751 sizeof(struct numa_maps_private));
1754 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1756 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1759 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1761 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1764 const struct file_operations proc_pid_numa_maps_operations = {
1765 .open = pid_numa_maps_open,
1767 .llseek = seq_lseek,
1768 .release = proc_map_release,
1771 const struct file_operations proc_tid_numa_maps_operations = {
1772 .open = tid_numa_maps_open,
1774 .llseek = seq_lseek,
1775 .release = proc_map_release,
1777 #endif /* CONFIG_NUMA */