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));
252 * Indicate if the VMA is a stack for the given task; for
253 * /proc/PID/maps that is the stack of the main task.
255 static int is_stack(struct proc_maps_private *priv,
256 struct vm_area_struct *vma, int is_pid)
261 stack = vma->vm_start <= vma->vm_mm->start_stack &&
262 vma->vm_end >= vma->vm_mm->start_stack;
264 struct inode *inode = priv->inode;
265 struct task_struct *task;
268 task = pid_task(proc_pid(inode), PIDTYPE_PID);
270 stack = vma_is_stack_for_task(vma, task);
277 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
279 struct mm_struct *mm = vma->vm_mm;
280 struct file *file = vma->vm_file;
281 struct proc_maps_private *priv = m->private;
282 vm_flags_t flags = vma->vm_flags;
283 unsigned long ino = 0;
284 unsigned long long pgoff = 0;
285 unsigned long start, end;
287 const char *name = NULL;
290 struct inode *inode = file_inode(vma->vm_file);
291 dev = inode->i_sb->s_dev;
293 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
296 /* We don't show the stack guard page in /proc/maps */
297 start = vma->vm_start;
298 if (stack_guard_page_start(vma, start))
301 if (stack_guard_page_end(vma, end))
304 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
305 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
308 flags & VM_READ ? 'r' : '-',
309 flags & VM_WRITE ? 'w' : '-',
310 flags & VM_EXEC ? 'x' : '-',
311 flags & VM_MAYSHARE ? 's' : 'p',
313 MAJOR(dev), MINOR(dev), ino);
316 * Print the dentry name for named mappings, and a
317 * special [heap] marker for the heap:
321 seq_file_path(m, file, "\n");
325 if (vma->vm_ops && vma->vm_ops->name) {
326 name = vma->vm_ops->name(vma);
331 name = arch_vma_name(vma);
338 if (vma->vm_start <= mm->brk &&
339 vma->vm_end >= mm->start_brk) {
344 if (is_stack(priv, vma, is_pid))
356 static int show_map(struct seq_file *m, void *v, int is_pid)
358 show_map_vma(m, v, is_pid);
363 static int show_pid_map(struct seq_file *m, void *v)
365 return show_map(m, v, 1);
368 static int show_tid_map(struct seq_file *m, void *v)
370 return show_map(m, v, 0);
373 static const struct seq_operations proc_pid_maps_op = {
380 static const struct seq_operations proc_tid_maps_op = {
387 static int pid_maps_open(struct inode *inode, struct file *file)
389 return do_maps_open(inode, file, &proc_pid_maps_op);
392 static int tid_maps_open(struct inode *inode, struct file *file)
394 return do_maps_open(inode, file, &proc_tid_maps_op);
397 const struct file_operations proc_pid_maps_operations = {
398 .open = pid_maps_open,
401 .release = proc_map_release,
404 const struct file_operations proc_tid_maps_operations = {
405 .open = tid_maps_open,
408 .release = proc_map_release,
412 * Proportional Set Size(PSS): my share of RSS.
414 * PSS of a process is the count of pages it has in memory, where each
415 * page is divided by the number of processes sharing it. So if a
416 * process has 1000 pages all to itself, and 1000 shared with one other
417 * process, its PSS will be 1500.
419 * To keep (accumulated) division errors low, we adopt a 64bit
420 * fixed-point pss counter to minimize division errors. So (pss >>
421 * PSS_SHIFT) would be the real byte count.
423 * A shift of 12 before division means (assuming 4K page size):
424 * - 1M 3-user-pages add up to 8KB errors;
425 * - supports mapcount up to 2^24, or 16M;
426 * - supports PSS up to 2^52 bytes, or 4PB.
430 #ifdef CONFIG_PROC_PAGE_MONITOR
431 struct mem_size_stats {
432 unsigned long resident;
433 unsigned long shared_clean;
434 unsigned long shared_dirty;
435 unsigned long private_clean;
436 unsigned long private_dirty;
437 unsigned long referenced;
438 unsigned long anonymous;
439 unsigned long anonymous_thp;
441 unsigned long shared_hugetlb;
442 unsigned long private_hugetlb;
447 static void smaps_account(struct mem_size_stats *mss, struct page *page,
448 unsigned long size, bool young, bool dirty)
453 mss->anonymous += size;
455 mss->resident += size;
456 /* Accumulate the size in pages that have been accessed. */
457 if (young || page_is_young(page) || PageReferenced(page))
458 mss->referenced += size;
459 mapcount = page_mapcount(page);
463 if (dirty || PageDirty(page))
464 mss->shared_dirty += size;
466 mss->shared_clean += size;
467 pss_delta = (u64)size << PSS_SHIFT;
468 do_div(pss_delta, mapcount);
469 mss->pss += pss_delta;
471 if (dirty || PageDirty(page))
472 mss->private_dirty += size;
474 mss->private_clean += size;
475 mss->pss += (u64)size << PSS_SHIFT;
479 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
480 struct mm_walk *walk)
482 struct mem_size_stats *mss = walk->private;
483 struct vm_area_struct *vma = walk->vma;
484 struct page *page = NULL;
486 if (pte_present(*pte)) {
487 page = vm_normal_page(vma, addr, *pte);
488 } else if (is_swap_pte(*pte)) {
489 swp_entry_t swpent = pte_to_swp_entry(*pte);
491 if (!non_swap_entry(swpent)) {
494 mss->swap += PAGE_SIZE;
495 mapcount = swp_swapcount(swpent);
497 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
499 do_div(pss_delta, mapcount);
500 mss->swap_pss += pss_delta;
502 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
504 } else if (is_migration_entry(swpent))
505 page = migration_entry_to_page(swpent);
510 smaps_account(mss, page, PAGE_SIZE, pte_young(*pte), pte_dirty(*pte));
513 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
514 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
515 struct mm_walk *walk)
517 struct mem_size_stats *mss = walk->private;
518 struct vm_area_struct *vma = walk->vma;
521 /* FOLL_DUMP will return -EFAULT on huge zero page */
522 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
523 if (IS_ERR_OR_NULL(page))
525 mss->anonymous_thp += HPAGE_PMD_SIZE;
526 smaps_account(mss, page, HPAGE_PMD_SIZE,
527 pmd_young(*pmd), pmd_dirty(*pmd));
530 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
531 struct mm_walk *walk)
536 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
537 struct mm_walk *walk)
539 struct vm_area_struct *vma = walk->vma;
543 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
544 smaps_pmd_entry(pmd, addr, walk);
549 if (pmd_trans_unstable(pmd))
552 * The mmap_sem held all the way back in m_start() is what
553 * keeps khugepaged out of here and from collapsing things
556 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
557 for (; addr != end; pte++, addr += PAGE_SIZE)
558 smaps_pte_entry(pte, addr, walk);
559 pte_unmap_unlock(pte - 1, ptl);
564 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
567 * Don't forget to update Documentation/ on changes.
569 static const char mnemonics[BITS_PER_LONG][2] = {
571 * In case if we meet a flag we don't know about.
573 [0 ... (BITS_PER_LONG-1)] = "??",
575 [ilog2(VM_READ)] = "rd",
576 [ilog2(VM_WRITE)] = "wr",
577 [ilog2(VM_EXEC)] = "ex",
578 [ilog2(VM_SHARED)] = "sh",
579 [ilog2(VM_MAYREAD)] = "mr",
580 [ilog2(VM_MAYWRITE)] = "mw",
581 [ilog2(VM_MAYEXEC)] = "me",
582 [ilog2(VM_MAYSHARE)] = "ms",
583 [ilog2(VM_GROWSDOWN)] = "gd",
584 [ilog2(VM_PFNMAP)] = "pf",
585 [ilog2(VM_DENYWRITE)] = "dw",
586 #ifdef CONFIG_X86_INTEL_MPX
587 [ilog2(VM_MPX)] = "mp",
589 [ilog2(VM_LOCKED)] = "lo",
590 [ilog2(VM_IO)] = "io",
591 [ilog2(VM_SEQ_READ)] = "sr",
592 [ilog2(VM_RAND_READ)] = "rr",
593 [ilog2(VM_DONTCOPY)] = "dc",
594 [ilog2(VM_DONTEXPAND)] = "de",
595 [ilog2(VM_ACCOUNT)] = "ac",
596 [ilog2(VM_NORESERVE)] = "nr",
597 [ilog2(VM_HUGETLB)] = "ht",
598 [ilog2(VM_ARCH_1)] = "ar",
599 [ilog2(VM_DONTDUMP)] = "dd",
600 #ifdef CONFIG_MEM_SOFT_DIRTY
601 [ilog2(VM_SOFTDIRTY)] = "sd",
603 [ilog2(VM_MIXEDMAP)] = "mm",
604 [ilog2(VM_HUGEPAGE)] = "hg",
605 [ilog2(VM_NOHUGEPAGE)] = "nh",
606 [ilog2(VM_MERGEABLE)] = "mg",
607 [ilog2(VM_UFFD_MISSING)]= "um",
608 [ilog2(VM_UFFD_WP)] = "uw",
612 seq_puts(m, "VmFlags: ");
613 for (i = 0; i < BITS_PER_LONG; i++) {
614 if (vma->vm_flags & (1UL << i)) {
615 seq_printf(m, "%c%c ",
616 mnemonics[i][0], mnemonics[i][1]);
622 #ifdef CONFIG_HUGETLB_PAGE
623 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
624 unsigned long addr, unsigned long end,
625 struct mm_walk *walk)
627 struct mem_size_stats *mss = walk->private;
628 struct vm_area_struct *vma = walk->vma;
629 struct page *page = NULL;
631 if (pte_present(*pte)) {
632 page = vm_normal_page(vma, addr, *pte);
633 } else if (is_swap_pte(*pte)) {
634 swp_entry_t swpent = pte_to_swp_entry(*pte);
636 if (is_migration_entry(swpent))
637 page = migration_entry_to_page(swpent);
640 int mapcount = page_mapcount(page);
643 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
645 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
649 #endif /* HUGETLB_PAGE */
651 static int show_smap(struct seq_file *m, void *v, int is_pid)
653 struct vm_area_struct *vma = v;
654 struct mem_size_stats mss;
655 struct mm_walk smaps_walk = {
656 .pmd_entry = smaps_pte_range,
657 #ifdef CONFIG_HUGETLB_PAGE
658 .hugetlb_entry = smaps_hugetlb_range,
664 memset(&mss, 0, sizeof mss);
665 /* mmap_sem is held in m_start */
666 walk_page_vma(vma, &smaps_walk);
668 show_map_vma(m, vma, is_pid);
674 "Shared_Clean: %8lu kB\n"
675 "Shared_Dirty: %8lu kB\n"
676 "Private_Clean: %8lu kB\n"
677 "Private_Dirty: %8lu kB\n"
678 "Referenced: %8lu kB\n"
679 "Anonymous: %8lu kB\n"
680 "AnonHugePages: %8lu kB\n"
681 "Shared_Hugetlb: %8lu kB\n"
682 "Private_Hugetlb: %7lu kB\n"
685 "KernelPageSize: %8lu kB\n"
686 "MMUPageSize: %8lu kB\n"
688 (vma->vm_end - vma->vm_start) >> 10,
690 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
691 mss.shared_clean >> 10,
692 mss.shared_dirty >> 10,
693 mss.private_clean >> 10,
694 mss.private_dirty >> 10,
695 mss.referenced >> 10,
697 mss.anonymous_thp >> 10,
698 mss.shared_hugetlb >> 10,
699 mss.private_hugetlb >> 10,
701 (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
702 vma_kernel_pagesize(vma) >> 10,
703 vma_mmu_pagesize(vma) >> 10,
704 (vma->vm_flags & VM_LOCKED) ?
705 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
707 show_smap_vma_flags(m, vma);
712 static int show_pid_smap(struct seq_file *m, void *v)
714 return show_smap(m, v, 1);
717 static int show_tid_smap(struct seq_file *m, void *v)
719 return show_smap(m, v, 0);
722 static const struct seq_operations proc_pid_smaps_op = {
726 .show = show_pid_smap
729 static const struct seq_operations proc_tid_smaps_op = {
733 .show = show_tid_smap
736 static int pid_smaps_open(struct inode *inode, struct file *file)
738 return do_maps_open(inode, file, &proc_pid_smaps_op);
741 static int tid_smaps_open(struct inode *inode, struct file *file)
743 return do_maps_open(inode, file, &proc_tid_smaps_op);
746 const struct file_operations proc_pid_smaps_operations = {
747 .open = pid_smaps_open,
750 .release = proc_map_release,
753 const struct file_operations proc_tid_smaps_operations = {
754 .open = tid_smaps_open,
757 .release = proc_map_release,
760 enum clear_refs_types {
764 CLEAR_REFS_SOFT_DIRTY,
765 CLEAR_REFS_MM_HIWATER_RSS,
769 struct clear_refs_private {
770 enum clear_refs_types type;
773 #ifdef CONFIG_MEM_SOFT_DIRTY
774 static inline void clear_soft_dirty(struct vm_area_struct *vma,
775 unsigned long addr, pte_t *pte)
778 * The soft-dirty tracker uses #PF-s to catch writes
779 * to pages, so write-protect the pte as well. See the
780 * Documentation/vm/soft-dirty.txt for full description
781 * of how soft-dirty works.
785 if (pte_present(ptent)) {
786 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
787 ptent = pte_wrprotect(ptent);
788 ptent = pte_clear_soft_dirty(ptent);
789 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
790 } else if (is_swap_pte(ptent)) {
791 ptent = pte_swp_clear_soft_dirty(ptent);
792 set_pte_at(vma->vm_mm, addr, pte, ptent);
796 static inline void clear_soft_dirty(struct vm_area_struct *vma,
797 unsigned long addr, pte_t *pte)
802 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
803 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
804 unsigned long addr, pmd_t *pmdp)
806 pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, 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);
817 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
818 unsigned long addr, pmd_t *pmdp)
823 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
824 unsigned long end, struct mm_walk *walk)
826 struct clear_refs_private *cp = walk->private;
827 struct vm_area_struct *vma = walk->vma;
832 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
833 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
834 clear_soft_dirty_pmd(vma, addr, pmd);
838 page = pmd_page(*pmd);
840 /* Clear accessed and referenced bits. */
841 pmdp_test_and_clear_young(vma, addr, pmd);
842 test_and_clear_page_young(page);
843 ClearPageReferenced(page);
849 if (pmd_trans_unstable(pmd))
852 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
853 for (; addr != end; pte++, addr += PAGE_SIZE) {
856 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
857 clear_soft_dirty(vma, addr, pte);
861 if (!pte_present(ptent))
864 page = vm_normal_page(vma, addr, ptent);
868 /* Clear accessed and referenced bits. */
869 ptep_test_and_clear_young(vma, addr, pte);
870 test_and_clear_page_young(page);
871 ClearPageReferenced(page);
873 pte_unmap_unlock(pte - 1, ptl);
878 static int clear_refs_test_walk(unsigned long start, unsigned long end,
879 struct mm_walk *walk)
881 struct clear_refs_private *cp = walk->private;
882 struct vm_area_struct *vma = walk->vma;
884 if (vma->vm_flags & VM_PFNMAP)
888 * Writing 1 to /proc/pid/clear_refs affects all pages.
889 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
890 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
891 * Writing 4 to /proc/pid/clear_refs affects all pages.
893 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
895 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
900 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
901 size_t count, loff_t *ppos)
903 struct task_struct *task;
904 char buffer[PROC_NUMBUF];
905 struct mm_struct *mm;
906 struct vm_area_struct *vma;
907 enum clear_refs_types type;
911 memset(buffer, 0, sizeof(buffer));
912 if (count > sizeof(buffer) - 1)
913 count = sizeof(buffer) - 1;
914 if (copy_from_user(buffer, buf, count))
916 rv = kstrtoint(strstrip(buffer), 10, &itype);
919 type = (enum clear_refs_types)itype;
920 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
923 task = get_proc_task(file_inode(file));
926 mm = get_task_mm(task);
928 struct clear_refs_private cp = {
931 struct mm_walk clear_refs_walk = {
932 .pmd_entry = clear_refs_pte_range,
933 .test_walk = clear_refs_test_walk,
938 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
940 * Writing 5 to /proc/pid/clear_refs resets the peak
941 * resident set size to this mm's current rss value.
943 down_write(&mm->mmap_sem);
944 reset_mm_hiwater_rss(mm);
945 up_write(&mm->mmap_sem);
949 down_read(&mm->mmap_sem);
950 if (type == CLEAR_REFS_SOFT_DIRTY) {
951 for (vma = mm->mmap; vma; vma = vma->vm_next) {
952 if (!(vma->vm_flags & VM_SOFTDIRTY))
954 up_read(&mm->mmap_sem);
955 down_write(&mm->mmap_sem);
956 for (vma = mm->mmap; vma; vma = vma->vm_next) {
957 vma->vm_flags &= ~VM_SOFTDIRTY;
958 vma_set_page_prot(vma);
960 downgrade_write(&mm->mmap_sem);
963 mmu_notifier_invalidate_range_start(mm, 0, -1);
965 walk_page_range(0, ~0UL, &clear_refs_walk);
966 if (type == CLEAR_REFS_SOFT_DIRTY)
967 mmu_notifier_invalidate_range_end(mm, 0, -1);
969 up_read(&mm->mmap_sem);
973 put_task_struct(task);
978 const struct file_operations proc_clear_refs_operations = {
979 .write = clear_refs_write,
980 .llseek = noop_llseek,
988 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
989 pagemap_entry_t *buffer;
993 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
994 #define PAGEMAP_WALK_MASK (PMD_MASK)
996 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
997 #define PM_PFRAME_BITS 55
998 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
999 #define PM_SOFT_DIRTY BIT_ULL(55)
1000 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1001 #define PM_FILE BIT_ULL(61)
1002 #define PM_SWAP BIT_ULL(62)
1003 #define PM_PRESENT BIT_ULL(63)
1005 #define PM_END_OF_BUFFER 1
1007 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1009 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1012 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1013 struct pagemapread *pm)
1015 pm->buffer[pm->pos++] = *pme;
1016 if (pm->pos >= pm->len)
1017 return PM_END_OF_BUFFER;
1021 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1022 struct mm_walk *walk)
1024 struct pagemapread *pm = walk->private;
1025 unsigned long addr = start;
1028 while (addr < end) {
1029 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1030 pagemap_entry_t pme = make_pme(0, 0);
1031 /* End of address space hole, which we mark as non-present. */
1032 unsigned long hole_end;
1035 hole_end = min(end, vma->vm_start);
1039 for (; addr < hole_end; addr += PAGE_SIZE) {
1040 err = add_to_pagemap(addr, &pme, pm);
1048 /* Addresses in the VMA. */
1049 if (vma->vm_flags & VM_SOFTDIRTY)
1050 pme = make_pme(0, PM_SOFT_DIRTY);
1051 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1052 err = add_to_pagemap(addr, &pme, pm);
1061 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1062 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1064 u64 frame = 0, flags = 0;
1065 struct page *page = NULL;
1067 if (pte_present(pte)) {
1069 frame = pte_pfn(pte);
1070 flags |= PM_PRESENT;
1071 page = vm_normal_page(vma, addr, pte);
1072 if (pte_soft_dirty(pte))
1073 flags |= PM_SOFT_DIRTY;
1074 } else if (is_swap_pte(pte)) {
1076 if (pte_swp_soft_dirty(pte))
1077 flags |= PM_SOFT_DIRTY;
1078 entry = pte_to_swp_entry(pte);
1079 frame = swp_type(entry) |
1080 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1082 if (is_migration_entry(entry))
1083 page = migration_entry_to_page(entry);
1086 if (page && !PageAnon(page))
1088 if (page && page_mapcount(page) == 1)
1089 flags |= PM_MMAP_EXCLUSIVE;
1090 if (vma->vm_flags & VM_SOFTDIRTY)
1091 flags |= PM_SOFT_DIRTY;
1093 return make_pme(frame, flags);
1096 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1097 struct mm_walk *walk)
1099 struct vm_area_struct *vma = walk->vma;
1100 struct pagemapread *pm = walk->private;
1102 pte_t *pte, *orig_pte;
1105 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1106 if (pmd_trans_huge_lock(pmdp, vma, &ptl) == 1) {
1107 u64 flags = 0, frame = 0;
1110 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1111 flags |= PM_SOFT_DIRTY;
1114 * Currently pmd for thp is always present because thp
1115 * can not be swapped-out, migrated, or HWPOISONed
1116 * (split in such cases instead.)
1117 * This if-check is just to prepare for future implementation.
1119 if (pmd_present(pmd)) {
1120 struct page *page = pmd_page(pmd);
1122 if (page_mapcount(page) == 1)
1123 flags |= PM_MMAP_EXCLUSIVE;
1125 flags |= PM_PRESENT;
1127 frame = pmd_pfn(pmd) +
1128 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1131 for (; addr != end; addr += PAGE_SIZE) {
1132 pagemap_entry_t pme = make_pme(frame, flags);
1134 err = add_to_pagemap(addr, &pme, pm);
1137 if (pm->show_pfn && (flags & PM_PRESENT))
1144 if (pmd_trans_unstable(pmdp))
1146 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1149 * We can assume that @vma always points to a valid one and @end never
1150 * goes beyond vma->vm_end.
1152 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1153 for (; addr < end; pte++, addr += PAGE_SIZE) {
1154 pagemap_entry_t pme;
1156 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1157 err = add_to_pagemap(addr, &pme, pm);
1161 pte_unmap_unlock(orig_pte, ptl);
1168 #ifdef CONFIG_HUGETLB_PAGE
1169 /* This function walks within one hugetlb entry in the single call */
1170 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1171 unsigned long addr, unsigned long end,
1172 struct mm_walk *walk)
1174 struct pagemapread *pm = walk->private;
1175 struct vm_area_struct *vma = walk->vma;
1176 u64 flags = 0, frame = 0;
1180 if (vma->vm_flags & VM_SOFTDIRTY)
1181 flags |= PM_SOFT_DIRTY;
1183 pte = huge_ptep_get(ptep);
1184 if (pte_present(pte)) {
1185 struct page *page = pte_page(pte);
1187 if (!PageAnon(page))
1190 if (page_mapcount(page) == 1)
1191 flags |= PM_MMAP_EXCLUSIVE;
1193 flags |= PM_PRESENT;
1195 frame = pte_pfn(pte) +
1196 ((addr & ~hmask) >> PAGE_SHIFT);
1199 for (; addr != end; addr += PAGE_SIZE) {
1200 pagemap_entry_t pme = make_pme(frame, flags);
1202 err = add_to_pagemap(addr, &pme, pm);
1205 if (pm->show_pfn && (flags & PM_PRESENT))
1213 #endif /* HUGETLB_PAGE */
1216 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1218 * For each page in the address space, this file contains one 64-bit entry
1219 * consisting of the following:
1221 * Bits 0-54 page frame number (PFN) if present
1222 * Bits 0-4 swap type if swapped
1223 * Bits 5-54 swap offset if swapped
1224 * Bit 55 pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1225 * Bit 56 page exclusively mapped
1227 * Bit 61 page is file-page or shared-anon
1228 * Bit 62 page swapped
1229 * Bit 63 page present
1231 * If the page is not present but in swap, then the PFN contains an
1232 * encoding of the swap file number and the page's offset into the
1233 * swap. Unmapped pages return a null PFN. This allows determining
1234 * precisely which pages are mapped (or in swap) and comparing mapped
1235 * pages between processes.
1237 * Efficient users of this interface will use /proc/pid/maps to
1238 * determine which areas of memory are actually mapped and llseek to
1239 * skip over unmapped regions.
1241 static ssize_t pagemap_read(struct file *file, char __user *buf,
1242 size_t count, loff_t *ppos)
1244 struct mm_struct *mm = file->private_data;
1245 struct pagemapread pm;
1246 struct mm_walk pagemap_walk = {};
1248 unsigned long svpfn;
1249 unsigned long start_vaddr;
1250 unsigned long end_vaddr;
1251 int ret = 0, copied = 0;
1253 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1257 /* file position must be aligned */
1258 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1265 /* do not disclose physical addresses: attack vector */
1266 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1268 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1269 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1274 pagemap_walk.pmd_entry = pagemap_pmd_range;
1275 pagemap_walk.pte_hole = pagemap_pte_hole;
1276 #ifdef CONFIG_HUGETLB_PAGE
1277 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1279 pagemap_walk.mm = mm;
1280 pagemap_walk.private = ±
1283 svpfn = src / PM_ENTRY_BYTES;
1284 start_vaddr = svpfn << PAGE_SHIFT;
1285 end_vaddr = mm->task_size;
1287 /* watch out for wraparound */
1288 if (svpfn > mm->task_size >> PAGE_SHIFT)
1289 start_vaddr = end_vaddr;
1292 * The odds are that this will stop walking way
1293 * before end_vaddr, because the length of the
1294 * user buffer is tracked in "pm", and the walk
1295 * will stop when we hit the end of the buffer.
1298 while (count && (start_vaddr < end_vaddr)) {
1303 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1305 if (end < start_vaddr || end > end_vaddr)
1307 down_read(&mm->mmap_sem);
1308 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1309 up_read(&mm->mmap_sem);
1312 len = min(count, PM_ENTRY_BYTES * pm.pos);
1313 if (copy_to_user(buf, pm.buffer, len)) {
1322 if (!ret || ret == PM_END_OF_BUFFER)
1333 static int pagemap_open(struct inode *inode, struct file *file)
1335 struct mm_struct *mm;
1337 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1340 file->private_data = mm;
1344 static int pagemap_release(struct inode *inode, struct file *file)
1346 struct mm_struct *mm = file->private_data;
1353 const struct file_operations proc_pagemap_operations = {
1354 .llseek = mem_lseek, /* borrow this */
1355 .read = pagemap_read,
1356 .open = pagemap_open,
1357 .release = pagemap_release,
1359 #endif /* CONFIG_PROC_PAGE_MONITOR */
1364 unsigned long pages;
1366 unsigned long active;
1367 unsigned long writeback;
1368 unsigned long mapcount_max;
1369 unsigned long dirty;
1370 unsigned long swapcache;
1371 unsigned long node[MAX_NUMNODES];
1374 struct numa_maps_private {
1375 struct proc_maps_private proc_maps;
1376 struct numa_maps md;
1379 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1380 unsigned long nr_pages)
1382 int count = page_mapcount(page);
1384 md->pages += nr_pages;
1385 if (pte_dirty || PageDirty(page))
1386 md->dirty += nr_pages;
1388 if (PageSwapCache(page))
1389 md->swapcache += nr_pages;
1391 if (PageActive(page) || PageUnevictable(page))
1392 md->active += nr_pages;
1394 if (PageWriteback(page))
1395 md->writeback += nr_pages;
1398 md->anon += nr_pages;
1400 if (count > md->mapcount_max)
1401 md->mapcount_max = count;
1403 md->node[page_to_nid(page)] += nr_pages;
1406 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1412 if (!pte_present(pte))
1415 page = vm_normal_page(vma, addr, pte);
1419 if (PageReserved(page))
1422 nid = page_to_nid(page);
1423 if (!node_isset(nid, node_states[N_MEMORY]))
1429 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1430 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1431 struct vm_area_struct *vma,
1437 if (!pmd_present(pmd))
1440 page = vm_normal_page_pmd(vma, addr, pmd);
1444 if (PageReserved(page))
1447 nid = page_to_nid(page);
1448 if (!node_isset(nid, node_states[N_MEMORY]))
1455 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1456 unsigned long end, struct mm_walk *walk)
1458 struct numa_maps *md = walk->private;
1459 struct vm_area_struct *vma = walk->vma;
1464 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1465 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1468 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1470 gather_stats(page, md, pmd_dirty(*pmd),
1471 HPAGE_PMD_SIZE/PAGE_SIZE);
1476 if (pmd_trans_unstable(pmd))
1479 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1481 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1484 gather_stats(page, md, pte_dirty(*pte), 1);
1486 } while (pte++, addr += PAGE_SIZE, addr != end);
1487 pte_unmap_unlock(orig_pte, ptl);
1490 #ifdef CONFIG_HUGETLB_PAGE
1491 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1492 unsigned long addr, unsigned long end, struct mm_walk *walk)
1494 pte_t huge_pte = huge_ptep_get(pte);
1495 struct numa_maps *md;
1498 if (!pte_present(huge_pte))
1501 page = pte_page(huge_pte);
1506 gather_stats(page, md, pte_dirty(huge_pte), 1);
1511 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1512 unsigned long addr, unsigned long end, struct mm_walk *walk)
1519 * Display pages allocated per node and memory policy via /proc.
1521 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1523 struct numa_maps_private *numa_priv = m->private;
1524 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1525 struct vm_area_struct *vma = v;
1526 struct numa_maps *md = &numa_priv->md;
1527 struct file *file = vma->vm_file;
1528 struct mm_struct *mm = vma->vm_mm;
1529 struct mm_walk walk = {
1530 .hugetlb_entry = gather_hugetlb_stats,
1531 .pmd_entry = gather_pte_stats,
1535 struct mempolicy *pol;
1542 /* Ensure we start with an empty set of numa_maps statistics. */
1543 memset(md, 0, sizeof(*md));
1545 pol = __get_vma_policy(vma, vma->vm_start);
1547 mpol_to_str(buffer, sizeof(buffer), pol);
1550 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1553 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1556 seq_puts(m, " file=");
1557 seq_file_path(m, file, "\n\t= ");
1558 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1559 seq_puts(m, " heap");
1560 } else if (is_stack(proc_priv, vma, is_pid)) {
1561 seq_puts(m, " stack");
1564 if (is_vm_hugetlb_page(vma))
1565 seq_puts(m, " huge");
1567 /* mmap_sem is held by m_start */
1568 walk_page_vma(vma, &walk);
1574 seq_printf(m, " anon=%lu", md->anon);
1577 seq_printf(m, " dirty=%lu", md->dirty);
1579 if (md->pages != md->anon && md->pages != md->dirty)
1580 seq_printf(m, " mapped=%lu", md->pages);
1582 if (md->mapcount_max > 1)
1583 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1586 seq_printf(m, " swapcache=%lu", md->swapcache);
1588 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1589 seq_printf(m, " active=%lu", md->active);
1592 seq_printf(m, " writeback=%lu", md->writeback);
1594 for_each_node_state(nid, N_MEMORY)
1596 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1598 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1601 m_cache_vma(m, vma);
1605 static int show_pid_numa_map(struct seq_file *m, void *v)
1607 return show_numa_map(m, v, 1);
1610 static int show_tid_numa_map(struct seq_file *m, void *v)
1612 return show_numa_map(m, v, 0);
1615 static const struct seq_operations proc_pid_numa_maps_op = {
1619 .show = show_pid_numa_map,
1622 static const struct seq_operations proc_tid_numa_maps_op = {
1626 .show = show_tid_numa_map,
1629 static int numa_maps_open(struct inode *inode, struct file *file,
1630 const struct seq_operations *ops)
1632 return proc_maps_open(inode, file, ops,
1633 sizeof(struct numa_maps_private));
1636 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1638 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1641 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1643 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1646 const struct file_operations proc_pid_numa_maps_operations = {
1647 .open = pid_numa_maps_open,
1649 .llseek = seq_lseek,
1650 .release = proc_map_release,
1653 const struct file_operations proc_tid_numa_maps_operations = {
1654 .open = tid_numa_maps_open,
1656 .llseek = seq_lseek,
1657 .release = proc_map_release,
1659 #endif /* CONFIG_NUMA */