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));
73 hugetlb_report_usage(m, mm);
76 unsigned long task_vsize(struct mm_struct *mm)
78 return PAGE_SIZE * mm->total_vm;
81 unsigned long task_statm(struct mm_struct *mm,
82 unsigned long *shared, unsigned long *text,
83 unsigned long *data, unsigned long *resident)
85 *shared = get_mm_counter(mm, MM_FILEPAGES);
86 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
88 *data = mm->total_vm - mm->shared_vm;
89 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
95 * Save get_task_policy() for show_numa_map().
97 static void hold_task_mempolicy(struct proc_maps_private *priv)
99 struct task_struct *task = priv->task;
102 priv->task_mempolicy = get_task_policy(task);
103 mpol_get(priv->task_mempolicy);
106 static void release_task_mempolicy(struct proc_maps_private *priv)
108 mpol_put(priv->task_mempolicy);
111 static void hold_task_mempolicy(struct proc_maps_private *priv)
114 static void release_task_mempolicy(struct proc_maps_private *priv)
119 static void vma_stop(struct proc_maps_private *priv)
121 struct mm_struct *mm = priv->mm;
123 release_task_mempolicy(priv);
124 up_read(&mm->mmap_sem);
128 static struct vm_area_struct *
129 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
131 if (vma == priv->tail_vma)
133 return vma->vm_next ?: priv->tail_vma;
136 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
138 if (m->count < m->size) /* vma is copied successfully */
139 m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL;
142 static void *m_start(struct seq_file *m, loff_t *ppos)
144 struct proc_maps_private *priv = m->private;
145 unsigned long last_addr = m->version;
146 struct mm_struct *mm;
147 struct vm_area_struct *vma;
148 unsigned int pos = *ppos;
150 /* See m_cache_vma(). Zero at the start or after lseek. */
151 if (last_addr == -1UL)
154 priv->task = get_proc_task(priv->inode);
156 return ERR_PTR(-ESRCH);
159 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
162 down_read(&mm->mmap_sem);
163 hold_task_mempolicy(priv);
164 priv->tail_vma = get_gate_vma(mm);
167 vma = find_vma(mm, last_addr);
168 if (vma && (vma = m_next_vma(priv, vma)))
173 if (pos < mm->map_count) {
174 for (vma = mm->mmap; pos; pos--) {
175 m->version = vma->vm_start;
181 /* we do not bother to update m->version in this case */
182 if (pos == mm->map_count && priv->tail_vma)
183 return priv->tail_vma;
189 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
191 struct proc_maps_private *priv = m->private;
192 struct vm_area_struct *next;
195 next = m_next_vma(priv, v);
201 static void m_stop(struct seq_file *m, void *v)
203 struct proc_maps_private *priv = m->private;
205 if (!IS_ERR_OR_NULL(v))
208 put_task_struct(priv->task);
213 static int proc_maps_open(struct inode *inode, struct file *file,
214 const struct seq_operations *ops, int psize)
216 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
222 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
223 if (IS_ERR(priv->mm)) {
224 int err = PTR_ERR(priv->mm);
226 seq_release_private(inode, file);
233 static int proc_map_release(struct inode *inode, struct file *file)
235 struct seq_file *seq = file->private_data;
236 struct proc_maps_private *priv = seq->private;
241 return seq_release_private(inode, file);
244 static int do_maps_open(struct inode *inode, struct file *file,
245 const struct seq_operations *ops)
247 return proc_maps_open(inode, file, ops,
248 sizeof(struct proc_maps_private));
251 static pid_t pid_of_stack(struct proc_maps_private *priv,
252 struct vm_area_struct *vma, bool is_pid)
254 struct inode *inode = priv->inode;
255 struct task_struct *task;
259 task = pid_task(proc_pid(inode), PIDTYPE_PID);
261 task = task_of_stack(task, vma, is_pid);
263 ret = task_pid_nr_ns(task, inode->i_sb->s_fs_info);
271 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
273 struct mm_struct *mm = vma->vm_mm;
274 struct file *file = vma->vm_file;
275 struct proc_maps_private *priv = m->private;
276 vm_flags_t flags = vma->vm_flags;
277 unsigned long ino = 0;
278 unsigned long long pgoff = 0;
279 unsigned long start, end;
281 const char *name = NULL;
284 struct inode *inode = file_inode(vma->vm_file);
285 dev = inode->i_sb->s_dev;
287 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
290 /* We don't show the stack guard page in /proc/maps */
291 start = vma->vm_start;
292 if (stack_guard_page_start(vma, start))
295 if (stack_guard_page_end(vma, end))
298 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
299 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
302 flags & VM_READ ? 'r' : '-',
303 flags & VM_WRITE ? 'w' : '-',
304 flags & VM_EXEC ? 'x' : '-',
305 flags & VM_MAYSHARE ? 's' : 'p',
307 MAJOR(dev), MINOR(dev), ino);
310 * Print the dentry name for named mappings, and a
311 * special [heap] marker for the heap:
315 seq_file_path(m, file, "\n");
319 if (vma->vm_ops && vma->vm_ops->name) {
320 name = vma->vm_ops->name(vma);
325 name = arch_vma_name(vma);
334 if (vma->vm_start <= mm->brk &&
335 vma->vm_end >= mm->start_brk) {
340 tid = pid_of_stack(priv, vma, is_pid);
343 * Thread stack in /proc/PID/task/TID/maps or
344 * the main process stack.
346 if (!is_pid || (vma->vm_start <= mm->start_stack &&
347 vma->vm_end >= mm->start_stack)) {
350 /* Thread stack in /proc/PID/maps */
352 seq_printf(m, "[stack:%d]", tid);
365 static int show_map(struct seq_file *m, void *v, int is_pid)
367 show_map_vma(m, v, is_pid);
372 static int show_pid_map(struct seq_file *m, void *v)
374 return show_map(m, v, 1);
377 static int show_tid_map(struct seq_file *m, void *v)
379 return show_map(m, v, 0);
382 static const struct seq_operations proc_pid_maps_op = {
389 static const struct seq_operations proc_tid_maps_op = {
396 static int pid_maps_open(struct inode *inode, struct file *file)
398 return do_maps_open(inode, file, &proc_pid_maps_op);
401 static int tid_maps_open(struct inode *inode, struct file *file)
403 return do_maps_open(inode, file, &proc_tid_maps_op);
406 const struct file_operations proc_pid_maps_operations = {
407 .open = pid_maps_open,
410 .release = proc_map_release,
413 const struct file_operations proc_tid_maps_operations = {
414 .open = tid_maps_open,
417 .release = proc_map_release,
421 * Proportional Set Size(PSS): my share of RSS.
423 * PSS of a process is the count of pages it has in memory, where each
424 * page is divided by the number of processes sharing it. So if a
425 * process has 1000 pages all to itself, and 1000 shared with one other
426 * process, its PSS will be 1500.
428 * To keep (accumulated) division errors low, we adopt a 64bit
429 * fixed-point pss counter to minimize division errors. So (pss >>
430 * PSS_SHIFT) would be the real byte count.
432 * A shift of 12 before division means (assuming 4K page size):
433 * - 1M 3-user-pages add up to 8KB errors;
434 * - supports mapcount up to 2^24, or 16M;
435 * - supports PSS up to 2^52 bytes, or 4PB.
439 #ifdef CONFIG_PROC_PAGE_MONITOR
440 struct mem_size_stats {
441 unsigned long resident;
442 unsigned long shared_clean;
443 unsigned long shared_dirty;
444 unsigned long private_clean;
445 unsigned long private_dirty;
446 unsigned long referenced;
447 unsigned long anonymous;
448 unsigned long anonymous_thp;
450 unsigned long shared_hugetlb;
451 unsigned long private_hugetlb;
456 static void smaps_account(struct mem_size_stats *mss, struct page *page,
457 unsigned long size, bool young, bool dirty)
462 mss->anonymous += size;
464 mss->resident += size;
465 /* Accumulate the size in pages that have been accessed. */
466 if (young || page_is_young(page) || PageReferenced(page))
467 mss->referenced += size;
468 mapcount = page_mapcount(page);
472 if (dirty || PageDirty(page))
473 mss->shared_dirty += size;
475 mss->shared_clean += size;
476 pss_delta = (u64)size << PSS_SHIFT;
477 do_div(pss_delta, mapcount);
478 mss->pss += pss_delta;
480 if (dirty || PageDirty(page))
481 mss->private_dirty += size;
483 mss->private_clean += size;
484 mss->pss += (u64)size << PSS_SHIFT;
488 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
489 struct mm_walk *walk)
491 struct mem_size_stats *mss = walk->private;
492 struct vm_area_struct *vma = walk->vma;
493 struct page *page = NULL;
495 if (pte_present(*pte)) {
496 page = vm_normal_page(vma, addr, *pte);
497 } else if (is_swap_pte(*pte)) {
498 swp_entry_t swpent = pte_to_swp_entry(*pte);
500 if (!non_swap_entry(swpent)) {
503 mss->swap += PAGE_SIZE;
504 mapcount = swp_swapcount(swpent);
506 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
508 do_div(pss_delta, mapcount);
509 mss->swap_pss += pss_delta;
511 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
513 } else if (is_migration_entry(swpent))
514 page = migration_entry_to_page(swpent);
519 smaps_account(mss, page, PAGE_SIZE, pte_young(*pte), pte_dirty(*pte));
522 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
523 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
524 struct mm_walk *walk)
526 struct mem_size_stats *mss = walk->private;
527 struct vm_area_struct *vma = walk->vma;
530 /* FOLL_DUMP will return -EFAULT on huge zero page */
531 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
532 if (IS_ERR_OR_NULL(page))
534 mss->anonymous_thp += HPAGE_PMD_SIZE;
535 smaps_account(mss, page, HPAGE_PMD_SIZE,
536 pmd_young(*pmd), pmd_dirty(*pmd));
539 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
540 struct mm_walk *walk)
545 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
546 struct mm_walk *walk)
548 struct vm_area_struct *vma = walk->vma;
552 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
553 smaps_pmd_entry(pmd, addr, walk);
558 if (pmd_trans_unstable(pmd))
561 * The mmap_sem held all the way back in m_start() is what
562 * keeps khugepaged out of here and from collapsing things
565 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
566 for (; addr != end; pte++, addr += PAGE_SIZE)
567 smaps_pte_entry(pte, addr, walk);
568 pte_unmap_unlock(pte - 1, ptl);
573 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
576 * Don't forget to update Documentation/ on changes.
578 static const char mnemonics[BITS_PER_LONG][2] = {
580 * In case if we meet a flag we don't know about.
582 [0 ... (BITS_PER_LONG-1)] = "??",
584 [ilog2(VM_READ)] = "rd",
585 [ilog2(VM_WRITE)] = "wr",
586 [ilog2(VM_EXEC)] = "ex",
587 [ilog2(VM_SHARED)] = "sh",
588 [ilog2(VM_MAYREAD)] = "mr",
589 [ilog2(VM_MAYWRITE)] = "mw",
590 [ilog2(VM_MAYEXEC)] = "me",
591 [ilog2(VM_MAYSHARE)] = "ms",
592 [ilog2(VM_GROWSDOWN)] = "gd",
593 [ilog2(VM_PFNMAP)] = "pf",
594 [ilog2(VM_DENYWRITE)] = "dw",
595 #ifdef CONFIG_X86_INTEL_MPX
596 [ilog2(VM_MPX)] = "mp",
598 [ilog2(VM_LOCKED)] = "lo",
599 [ilog2(VM_IO)] = "io",
600 [ilog2(VM_SEQ_READ)] = "sr",
601 [ilog2(VM_RAND_READ)] = "rr",
602 [ilog2(VM_DONTCOPY)] = "dc",
603 [ilog2(VM_DONTEXPAND)] = "de",
604 [ilog2(VM_ACCOUNT)] = "ac",
605 [ilog2(VM_NORESERVE)] = "nr",
606 [ilog2(VM_HUGETLB)] = "ht",
607 [ilog2(VM_ARCH_1)] = "ar",
608 [ilog2(VM_DONTDUMP)] = "dd",
609 #ifdef CONFIG_MEM_SOFT_DIRTY
610 [ilog2(VM_SOFTDIRTY)] = "sd",
612 [ilog2(VM_MIXEDMAP)] = "mm",
613 [ilog2(VM_HUGEPAGE)] = "hg",
614 [ilog2(VM_NOHUGEPAGE)] = "nh",
615 [ilog2(VM_MERGEABLE)] = "mg",
616 [ilog2(VM_UFFD_MISSING)]= "um",
617 [ilog2(VM_UFFD_WP)] = "uw",
621 seq_puts(m, "VmFlags: ");
622 for (i = 0; i < BITS_PER_LONG; i++) {
623 if (vma->vm_flags & (1UL << i)) {
624 seq_printf(m, "%c%c ",
625 mnemonics[i][0], mnemonics[i][1]);
631 #ifdef CONFIG_HUGETLB_PAGE
632 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
633 unsigned long addr, unsigned long end,
634 struct mm_walk *walk)
636 struct mem_size_stats *mss = walk->private;
637 struct vm_area_struct *vma = walk->vma;
638 struct page *page = NULL;
640 if (pte_present(*pte)) {
641 page = vm_normal_page(vma, addr, *pte);
642 } else if (is_swap_pte(*pte)) {
643 swp_entry_t swpent = pte_to_swp_entry(*pte);
645 if (is_migration_entry(swpent))
646 page = migration_entry_to_page(swpent);
649 int mapcount = page_mapcount(page);
652 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
654 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
658 #endif /* HUGETLB_PAGE */
660 static int show_smap(struct seq_file *m, void *v, int is_pid)
662 struct vm_area_struct *vma = v;
663 struct mem_size_stats mss;
664 struct mm_walk smaps_walk = {
665 .pmd_entry = smaps_pte_range,
666 #ifdef CONFIG_HUGETLB_PAGE
667 .hugetlb_entry = smaps_hugetlb_range,
673 memset(&mss, 0, sizeof mss);
674 /* mmap_sem is held in m_start */
675 walk_page_vma(vma, &smaps_walk);
677 show_map_vma(m, vma, is_pid);
683 "Shared_Clean: %8lu kB\n"
684 "Shared_Dirty: %8lu kB\n"
685 "Private_Clean: %8lu kB\n"
686 "Private_Dirty: %8lu kB\n"
687 "Referenced: %8lu kB\n"
688 "Anonymous: %8lu kB\n"
689 "AnonHugePages: %8lu kB\n"
690 "Shared_Hugetlb: %8lu kB\n"
691 "Private_Hugetlb: %7lu kB\n"
694 "KernelPageSize: %8lu kB\n"
695 "MMUPageSize: %8lu kB\n"
697 (vma->vm_end - vma->vm_start) >> 10,
699 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
700 mss.shared_clean >> 10,
701 mss.shared_dirty >> 10,
702 mss.private_clean >> 10,
703 mss.private_dirty >> 10,
704 mss.referenced >> 10,
706 mss.anonymous_thp >> 10,
707 mss.shared_hugetlb >> 10,
708 mss.private_hugetlb >> 10,
710 (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
711 vma_kernel_pagesize(vma) >> 10,
712 vma_mmu_pagesize(vma) >> 10,
713 (vma->vm_flags & VM_LOCKED) ?
714 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
716 show_smap_vma_flags(m, vma);
721 static int show_pid_smap(struct seq_file *m, void *v)
723 return show_smap(m, v, 1);
726 static int show_tid_smap(struct seq_file *m, void *v)
728 return show_smap(m, v, 0);
731 static const struct seq_operations proc_pid_smaps_op = {
735 .show = show_pid_smap
738 static const struct seq_operations proc_tid_smaps_op = {
742 .show = show_tid_smap
745 static int pid_smaps_open(struct inode *inode, struct file *file)
747 return do_maps_open(inode, file, &proc_pid_smaps_op);
750 static int tid_smaps_open(struct inode *inode, struct file *file)
752 return do_maps_open(inode, file, &proc_tid_smaps_op);
755 const struct file_operations proc_pid_smaps_operations = {
756 .open = pid_smaps_open,
759 .release = proc_map_release,
762 const struct file_operations proc_tid_smaps_operations = {
763 .open = tid_smaps_open,
766 .release = proc_map_release,
769 enum clear_refs_types {
773 CLEAR_REFS_SOFT_DIRTY,
774 CLEAR_REFS_MM_HIWATER_RSS,
778 struct clear_refs_private {
779 enum clear_refs_types type;
782 #ifdef CONFIG_MEM_SOFT_DIRTY
783 static inline void clear_soft_dirty(struct vm_area_struct *vma,
784 unsigned long addr, pte_t *pte)
787 * The soft-dirty tracker uses #PF-s to catch writes
788 * to pages, so write-protect the pte as well. See the
789 * Documentation/vm/soft-dirty.txt for full description
790 * of how soft-dirty works.
794 if (pte_present(ptent)) {
795 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
796 ptent = pte_wrprotect(ptent);
797 ptent = pte_clear_soft_dirty(ptent);
798 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
799 } else if (is_swap_pte(ptent)) {
800 ptent = pte_swp_clear_soft_dirty(ptent);
801 set_pte_at(vma->vm_mm, addr, pte, ptent);
805 static inline void clear_soft_dirty(struct vm_area_struct *vma,
806 unsigned long addr, pte_t *pte)
811 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
812 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
813 unsigned long addr, pmd_t *pmdp)
815 pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp);
817 pmd = pmd_wrprotect(pmd);
818 pmd = pmd_clear_soft_dirty(pmd);
820 if (vma->vm_flags & VM_SOFTDIRTY)
821 vma->vm_flags &= ~VM_SOFTDIRTY;
823 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
826 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
827 unsigned long addr, pmd_t *pmdp)
832 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
833 unsigned long end, struct mm_walk *walk)
835 struct clear_refs_private *cp = walk->private;
836 struct vm_area_struct *vma = walk->vma;
841 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
842 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
843 clear_soft_dirty_pmd(vma, addr, pmd);
847 page = pmd_page(*pmd);
849 /* Clear accessed and referenced bits. */
850 pmdp_test_and_clear_young(vma, addr, pmd);
851 test_and_clear_page_young(page);
852 ClearPageReferenced(page);
858 if (pmd_trans_unstable(pmd))
861 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
862 for (; addr != end; pte++, addr += PAGE_SIZE) {
865 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
866 clear_soft_dirty(vma, addr, pte);
870 if (!pte_present(ptent))
873 page = vm_normal_page(vma, addr, ptent);
877 /* Clear accessed and referenced bits. */
878 ptep_test_and_clear_young(vma, addr, pte);
879 test_and_clear_page_young(page);
880 ClearPageReferenced(page);
882 pte_unmap_unlock(pte - 1, ptl);
887 static int clear_refs_test_walk(unsigned long start, unsigned long end,
888 struct mm_walk *walk)
890 struct clear_refs_private *cp = walk->private;
891 struct vm_area_struct *vma = walk->vma;
893 if (vma->vm_flags & VM_PFNMAP)
897 * Writing 1 to /proc/pid/clear_refs affects all pages.
898 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
899 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
900 * Writing 4 to /proc/pid/clear_refs affects all pages.
902 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
904 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
909 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
910 size_t count, loff_t *ppos)
912 struct task_struct *task;
913 char buffer[PROC_NUMBUF];
914 struct mm_struct *mm;
915 struct vm_area_struct *vma;
916 enum clear_refs_types type;
920 memset(buffer, 0, sizeof(buffer));
921 if (count > sizeof(buffer) - 1)
922 count = sizeof(buffer) - 1;
923 if (copy_from_user(buffer, buf, count))
925 rv = kstrtoint(strstrip(buffer), 10, &itype);
928 type = (enum clear_refs_types)itype;
929 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
932 task = get_proc_task(file_inode(file));
935 mm = get_task_mm(task);
937 struct clear_refs_private cp = {
940 struct mm_walk clear_refs_walk = {
941 .pmd_entry = clear_refs_pte_range,
942 .test_walk = clear_refs_test_walk,
947 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
949 * Writing 5 to /proc/pid/clear_refs resets the peak
950 * resident set size to this mm's current rss value.
952 down_write(&mm->mmap_sem);
953 reset_mm_hiwater_rss(mm);
954 up_write(&mm->mmap_sem);
958 down_read(&mm->mmap_sem);
959 if (type == CLEAR_REFS_SOFT_DIRTY) {
960 for (vma = mm->mmap; vma; vma = vma->vm_next) {
961 if (!(vma->vm_flags & VM_SOFTDIRTY))
963 up_read(&mm->mmap_sem);
964 down_write(&mm->mmap_sem);
965 for (vma = mm->mmap; vma; vma = vma->vm_next) {
966 vma->vm_flags &= ~VM_SOFTDIRTY;
967 vma_set_page_prot(vma);
969 downgrade_write(&mm->mmap_sem);
972 mmu_notifier_invalidate_range_start(mm, 0, -1);
974 walk_page_range(0, ~0UL, &clear_refs_walk);
975 if (type == CLEAR_REFS_SOFT_DIRTY)
976 mmu_notifier_invalidate_range_end(mm, 0, -1);
978 up_read(&mm->mmap_sem);
982 put_task_struct(task);
987 const struct file_operations proc_clear_refs_operations = {
988 .write = clear_refs_write,
989 .llseek = noop_llseek,
997 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
998 pagemap_entry_t *buffer;
1002 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
1003 #define PAGEMAP_WALK_MASK (PMD_MASK)
1005 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1006 #define PM_PFRAME_BITS 55
1007 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1008 #define PM_SOFT_DIRTY BIT_ULL(55)
1009 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1010 #define PM_FILE BIT_ULL(61)
1011 #define PM_SWAP BIT_ULL(62)
1012 #define PM_PRESENT BIT_ULL(63)
1014 #define PM_END_OF_BUFFER 1
1016 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1018 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1021 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1022 struct pagemapread *pm)
1024 pm->buffer[pm->pos++] = *pme;
1025 if (pm->pos >= pm->len)
1026 return PM_END_OF_BUFFER;
1030 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1031 struct mm_walk *walk)
1033 struct pagemapread *pm = walk->private;
1034 unsigned long addr = start;
1037 while (addr < end) {
1038 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1039 pagemap_entry_t pme = make_pme(0, 0);
1040 /* End of address space hole, which we mark as non-present. */
1041 unsigned long hole_end;
1044 hole_end = min(end, vma->vm_start);
1048 for (; addr < hole_end; addr += PAGE_SIZE) {
1049 err = add_to_pagemap(addr, &pme, pm);
1057 /* Addresses in the VMA. */
1058 if (vma->vm_flags & VM_SOFTDIRTY)
1059 pme = make_pme(0, PM_SOFT_DIRTY);
1060 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1061 err = add_to_pagemap(addr, &pme, pm);
1070 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1071 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1073 u64 frame = 0, flags = 0;
1074 struct page *page = NULL;
1076 if (pte_present(pte)) {
1078 frame = pte_pfn(pte);
1079 flags |= PM_PRESENT;
1080 page = vm_normal_page(vma, addr, pte);
1081 if (pte_soft_dirty(pte))
1082 flags |= PM_SOFT_DIRTY;
1083 } else if (is_swap_pte(pte)) {
1085 if (pte_swp_soft_dirty(pte))
1086 flags |= PM_SOFT_DIRTY;
1087 entry = pte_to_swp_entry(pte);
1088 frame = swp_type(entry) |
1089 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1091 if (is_migration_entry(entry))
1092 page = migration_entry_to_page(entry);
1095 if (page && !PageAnon(page))
1097 if (page && page_mapcount(page) == 1)
1098 flags |= PM_MMAP_EXCLUSIVE;
1099 if (vma->vm_flags & VM_SOFTDIRTY)
1100 flags |= PM_SOFT_DIRTY;
1102 return make_pme(frame, flags);
1105 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1106 struct mm_walk *walk)
1108 struct vm_area_struct *vma = walk->vma;
1109 struct pagemapread *pm = walk->private;
1111 pte_t *pte, *orig_pte;
1114 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1115 if (pmd_trans_huge_lock(pmdp, vma, &ptl) == 1) {
1116 u64 flags = 0, frame = 0;
1119 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1120 flags |= PM_SOFT_DIRTY;
1123 * Currently pmd for thp is always present because thp
1124 * can not be swapped-out, migrated, or HWPOISONed
1125 * (split in such cases instead.)
1126 * This if-check is just to prepare for future implementation.
1128 if (pmd_present(pmd)) {
1129 struct page *page = pmd_page(pmd);
1131 if (page_mapcount(page) == 1)
1132 flags |= PM_MMAP_EXCLUSIVE;
1134 flags |= PM_PRESENT;
1136 frame = pmd_pfn(pmd) +
1137 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1140 for (; addr != end; addr += PAGE_SIZE) {
1141 pagemap_entry_t pme = make_pme(frame, flags);
1143 err = add_to_pagemap(addr, &pme, pm);
1146 if (pm->show_pfn && (flags & PM_PRESENT))
1153 if (pmd_trans_unstable(pmdp))
1155 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1158 * We can assume that @vma always points to a valid one and @end never
1159 * goes beyond vma->vm_end.
1161 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1162 for (; addr < end; pte++, addr += PAGE_SIZE) {
1163 pagemap_entry_t pme;
1165 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1166 err = add_to_pagemap(addr, &pme, pm);
1170 pte_unmap_unlock(orig_pte, ptl);
1177 #ifdef CONFIG_HUGETLB_PAGE
1178 /* This function walks within one hugetlb entry in the single call */
1179 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1180 unsigned long addr, unsigned long end,
1181 struct mm_walk *walk)
1183 struct pagemapread *pm = walk->private;
1184 struct vm_area_struct *vma = walk->vma;
1185 u64 flags = 0, frame = 0;
1189 if (vma->vm_flags & VM_SOFTDIRTY)
1190 flags |= PM_SOFT_DIRTY;
1192 pte = huge_ptep_get(ptep);
1193 if (pte_present(pte)) {
1194 struct page *page = pte_page(pte);
1196 if (!PageAnon(page))
1199 if (page_mapcount(page) == 1)
1200 flags |= PM_MMAP_EXCLUSIVE;
1202 flags |= PM_PRESENT;
1204 frame = pte_pfn(pte) +
1205 ((addr & ~hmask) >> PAGE_SHIFT);
1208 for (; addr != end; addr += PAGE_SIZE) {
1209 pagemap_entry_t pme = make_pme(frame, flags);
1211 err = add_to_pagemap(addr, &pme, pm);
1214 if (pm->show_pfn && (flags & PM_PRESENT))
1222 #endif /* HUGETLB_PAGE */
1225 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1227 * For each page in the address space, this file contains one 64-bit entry
1228 * consisting of the following:
1230 * Bits 0-54 page frame number (PFN) if present
1231 * Bits 0-4 swap type if swapped
1232 * Bits 5-54 swap offset if swapped
1233 * Bit 55 pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1234 * Bit 56 page exclusively mapped
1236 * Bit 61 page is file-page or shared-anon
1237 * Bit 62 page swapped
1238 * Bit 63 page present
1240 * If the page is not present but in swap, then the PFN contains an
1241 * encoding of the swap file number and the page's offset into the
1242 * swap. Unmapped pages return a null PFN. This allows determining
1243 * precisely which pages are mapped (or in swap) and comparing mapped
1244 * pages between processes.
1246 * Efficient users of this interface will use /proc/pid/maps to
1247 * determine which areas of memory are actually mapped and llseek to
1248 * skip over unmapped regions.
1250 static ssize_t pagemap_read(struct file *file, char __user *buf,
1251 size_t count, loff_t *ppos)
1253 struct mm_struct *mm = file->private_data;
1254 struct pagemapread pm;
1255 struct mm_walk pagemap_walk = {};
1257 unsigned long svpfn;
1258 unsigned long start_vaddr;
1259 unsigned long end_vaddr;
1260 int ret = 0, copied = 0;
1262 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1266 /* file position must be aligned */
1267 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1274 /* do not disclose physical addresses: attack vector */
1275 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1277 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1278 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1283 pagemap_walk.pmd_entry = pagemap_pmd_range;
1284 pagemap_walk.pte_hole = pagemap_pte_hole;
1285 #ifdef CONFIG_HUGETLB_PAGE
1286 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1288 pagemap_walk.mm = mm;
1289 pagemap_walk.private = ±
1292 svpfn = src / PM_ENTRY_BYTES;
1293 start_vaddr = svpfn << PAGE_SHIFT;
1294 end_vaddr = mm->task_size;
1296 /* watch out for wraparound */
1297 if (svpfn > mm->task_size >> PAGE_SHIFT)
1298 start_vaddr = end_vaddr;
1301 * The odds are that this will stop walking way
1302 * before end_vaddr, because the length of the
1303 * user buffer is tracked in "pm", and the walk
1304 * will stop when we hit the end of the buffer.
1307 while (count && (start_vaddr < end_vaddr)) {
1312 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1314 if (end < start_vaddr || end > end_vaddr)
1316 down_read(&mm->mmap_sem);
1317 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1318 up_read(&mm->mmap_sem);
1321 len = min(count, PM_ENTRY_BYTES * pm.pos);
1322 if (copy_to_user(buf, pm.buffer, len)) {
1331 if (!ret || ret == PM_END_OF_BUFFER)
1342 static int pagemap_open(struct inode *inode, struct file *file)
1344 struct mm_struct *mm;
1346 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1349 file->private_data = mm;
1353 static int pagemap_release(struct inode *inode, struct file *file)
1355 struct mm_struct *mm = file->private_data;
1362 const struct file_operations proc_pagemap_operations = {
1363 .llseek = mem_lseek, /* borrow this */
1364 .read = pagemap_read,
1365 .open = pagemap_open,
1366 .release = pagemap_release,
1368 #endif /* CONFIG_PROC_PAGE_MONITOR */
1373 unsigned long pages;
1375 unsigned long active;
1376 unsigned long writeback;
1377 unsigned long mapcount_max;
1378 unsigned long dirty;
1379 unsigned long swapcache;
1380 unsigned long node[MAX_NUMNODES];
1383 struct numa_maps_private {
1384 struct proc_maps_private proc_maps;
1385 struct numa_maps md;
1388 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1389 unsigned long nr_pages)
1391 int count = page_mapcount(page);
1393 md->pages += nr_pages;
1394 if (pte_dirty || PageDirty(page))
1395 md->dirty += nr_pages;
1397 if (PageSwapCache(page))
1398 md->swapcache += nr_pages;
1400 if (PageActive(page) || PageUnevictable(page))
1401 md->active += nr_pages;
1403 if (PageWriteback(page))
1404 md->writeback += nr_pages;
1407 md->anon += nr_pages;
1409 if (count > md->mapcount_max)
1410 md->mapcount_max = count;
1412 md->node[page_to_nid(page)] += nr_pages;
1415 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1421 if (!pte_present(pte))
1424 page = vm_normal_page(vma, addr, pte);
1428 if (PageReserved(page))
1431 nid = page_to_nid(page);
1432 if (!node_isset(nid, node_states[N_MEMORY]))
1438 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1439 unsigned long end, struct mm_walk *walk)
1441 struct numa_maps *md = walk->private;
1442 struct vm_area_struct *vma = walk->vma;
1447 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1448 pte_t huge_pte = *(pte_t *)pmd;
1451 page = can_gather_numa_stats(huge_pte, vma, addr);
1453 gather_stats(page, md, pte_dirty(huge_pte),
1454 HPAGE_PMD_SIZE/PAGE_SIZE);
1459 if (pmd_trans_unstable(pmd))
1461 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1463 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1466 gather_stats(page, md, pte_dirty(*pte), 1);
1468 } while (pte++, addr += PAGE_SIZE, addr != end);
1469 pte_unmap_unlock(orig_pte, ptl);
1472 #ifdef CONFIG_HUGETLB_PAGE
1473 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1474 unsigned long addr, unsigned long end, struct mm_walk *walk)
1476 struct numa_maps *md;
1479 if (!pte_present(*pte))
1482 page = pte_page(*pte);
1487 gather_stats(page, md, pte_dirty(*pte), 1);
1492 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1493 unsigned long addr, unsigned long end, struct mm_walk *walk)
1500 * Display pages allocated per node and memory policy via /proc.
1502 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1504 struct numa_maps_private *numa_priv = m->private;
1505 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1506 struct vm_area_struct *vma = v;
1507 struct numa_maps *md = &numa_priv->md;
1508 struct file *file = vma->vm_file;
1509 struct mm_struct *mm = vma->vm_mm;
1510 struct mm_walk walk = {
1511 .hugetlb_entry = gather_hugetlb_stats,
1512 .pmd_entry = gather_pte_stats,
1516 struct mempolicy *pol;
1523 /* Ensure we start with an empty set of numa_maps statistics. */
1524 memset(md, 0, sizeof(*md));
1526 pol = __get_vma_policy(vma, vma->vm_start);
1528 mpol_to_str(buffer, sizeof(buffer), pol);
1531 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1534 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1537 seq_puts(m, " file=");
1538 seq_file_path(m, file, "\n\t= ");
1539 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1540 seq_puts(m, " heap");
1542 pid_t tid = pid_of_stack(proc_priv, vma, is_pid);
1545 * Thread stack in /proc/PID/task/TID/maps or
1546 * the main process stack.
1548 if (!is_pid || (vma->vm_start <= mm->start_stack &&
1549 vma->vm_end >= mm->start_stack))
1550 seq_puts(m, " stack");
1552 seq_printf(m, " stack:%d", tid);
1556 if (is_vm_hugetlb_page(vma))
1557 seq_puts(m, " huge");
1559 /* mmap_sem is held by m_start */
1560 walk_page_vma(vma, &walk);
1566 seq_printf(m, " anon=%lu", md->anon);
1569 seq_printf(m, " dirty=%lu", md->dirty);
1571 if (md->pages != md->anon && md->pages != md->dirty)
1572 seq_printf(m, " mapped=%lu", md->pages);
1574 if (md->mapcount_max > 1)
1575 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1578 seq_printf(m, " swapcache=%lu", md->swapcache);
1580 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1581 seq_printf(m, " active=%lu", md->active);
1584 seq_printf(m, " writeback=%lu", md->writeback);
1586 for_each_node_state(nid, N_MEMORY)
1588 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1590 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1593 m_cache_vma(m, vma);
1597 static int show_pid_numa_map(struct seq_file *m, void *v)
1599 return show_numa_map(m, v, 1);
1602 static int show_tid_numa_map(struct seq_file *m, void *v)
1604 return show_numa_map(m, v, 0);
1607 static const struct seq_operations proc_pid_numa_maps_op = {
1611 .show = show_pid_numa_map,
1614 static const struct seq_operations proc_tid_numa_maps_op = {
1618 .show = show_tid_numa_map,
1621 static int numa_maps_open(struct inode *inode, struct file *file,
1622 const struct seq_operations *ops)
1624 return proc_maps_open(inode, file, ops,
1625 sizeof(struct numa_maps_private));
1628 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1630 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1633 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1635 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1638 const struct file_operations proc_pid_numa_maps_operations = {
1639 .open = pid_numa_maps_open,
1641 .llseek = seq_lseek,
1642 .release = proc_map_release,
1645 const struct file_operations proc_tid_numa_maps_operations = {
1646 .open = tid_numa_maps_open,
1648 .llseek = seq_lseek,
1649 .release = proc_map_release,
1651 #endif /* CONFIG_NUMA */