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>
18 #include <asm/uaccess.h>
19 #include <asm/tlbflush.h>
22 void task_mem(struct seq_file *m, struct mm_struct *mm)
24 unsigned long data, text, lib, swap;
25 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
28 * Note: to minimize their overhead, mm maintains hiwater_vm and
29 * hiwater_rss only when about to *lower* total_vm or rss. Any
30 * collector of these hiwater stats must therefore get total_vm
31 * and rss too, which will usually be the higher. Barriers? not
32 * worth the effort, such snapshots can always be inconsistent.
34 hiwater_vm = total_vm = mm->total_vm;
35 if (hiwater_vm < mm->hiwater_vm)
36 hiwater_vm = mm->hiwater_vm;
37 hiwater_rss = total_rss = get_mm_rss(mm);
38 if (hiwater_rss < mm->hiwater_rss)
39 hiwater_rss = mm->hiwater_rss;
41 data = mm->total_vm - mm->shared_vm - mm->stack_vm;
42 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
43 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
44 swap = get_mm_counter(mm, MM_SWAPENTS);
58 hiwater_vm << (PAGE_SHIFT-10),
59 total_vm << (PAGE_SHIFT-10),
60 mm->locked_vm << (PAGE_SHIFT-10),
61 mm->pinned_vm << (PAGE_SHIFT-10),
62 hiwater_rss << (PAGE_SHIFT-10),
63 total_rss << (PAGE_SHIFT-10),
64 data << (PAGE_SHIFT-10),
65 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
66 (PTRS_PER_PTE * sizeof(pte_t) *
67 atomic_long_read(&mm->nr_ptes)) >> 10,
68 swap << (PAGE_SHIFT-10));
71 unsigned long task_vsize(struct mm_struct *mm)
73 return PAGE_SIZE * mm->total_vm;
76 unsigned long task_statm(struct mm_struct *mm,
77 unsigned long *shared, unsigned long *text,
78 unsigned long *data, unsigned long *resident)
80 *shared = get_mm_counter(mm, MM_FILEPAGES);
81 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
83 *data = mm->total_vm - mm->shared_vm;
84 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
90 * Save get_task_policy() for show_numa_map().
92 static void hold_task_mempolicy(struct proc_maps_private *priv)
94 struct task_struct *task = priv->task;
97 priv->task_mempolicy = get_task_policy(task);
98 mpol_get(priv->task_mempolicy);
101 static void release_task_mempolicy(struct proc_maps_private *priv)
103 mpol_put(priv->task_mempolicy);
106 static void hold_task_mempolicy(struct proc_maps_private *priv)
109 static void release_task_mempolicy(struct proc_maps_private *priv)
114 static void vma_stop(struct proc_maps_private *priv)
116 struct mm_struct *mm = priv->mm;
118 release_task_mempolicy(priv);
119 up_read(&mm->mmap_sem);
123 static struct vm_area_struct *
124 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
126 if (vma == priv->tail_vma)
128 return vma->vm_next ?: priv->tail_vma;
131 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
133 if (m->count < m->size) /* vma is copied successfully */
134 m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL;
137 static void *m_start(struct seq_file *m, loff_t *ppos)
139 struct proc_maps_private *priv = m->private;
140 unsigned long last_addr = m->version;
141 struct mm_struct *mm;
142 struct vm_area_struct *vma;
143 unsigned int pos = *ppos;
145 /* See m_cache_vma(). Zero at the start or after lseek. */
146 if (last_addr == -1UL)
149 priv->task = get_proc_task(priv->inode);
151 return ERR_PTR(-ESRCH);
154 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
157 down_read(&mm->mmap_sem);
158 hold_task_mempolicy(priv);
159 priv->tail_vma = get_gate_vma(mm);
162 vma = find_vma(mm, last_addr);
163 if (vma && (vma = m_next_vma(priv, vma)))
168 if (pos < mm->map_count) {
169 for (vma = mm->mmap; pos; pos--) {
170 m->version = vma->vm_start;
176 /* we do not bother to update m->version in this case */
177 if (pos == mm->map_count && priv->tail_vma)
178 return priv->tail_vma;
184 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
186 struct proc_maps_private *priv = m->private;
187 struct vm_area_struct *next;
190 next = m_next_vma(priv, v);
196 static void m_stop(struct seq_file *m, void *v)
198 struct proc_maps_private *priv = m->private;
200 if (!IS_ERR_OR_NULL(v))
203 put_task_struct(priv->task);
208 static int proc_maps_open(struct inode *inode, struct file *file,
209 const struct seq_operations *ops, int psize)
211 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
217 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
218 if (IS_ERR(priv->mm)) {
219 int err = PTR_ERR(priv->mm);
221 seq_release_private(inode, file);
228 static int proc_map_release(struct inode *inode, struct file *file)
230 struct seq_file *seq = file->private_data;
231 struct proc_maps_private *priv = seq->private;
236 return seq_release_private(inode, file);
239 static int do_maps_open(struct inode *inode, struct file *file,
240 const struct seq_operations *ops)
242 return proc_maps_open(inode, file, ops,
243 sizeof(struct proc_maps_private));
246 static pid_t pid_of_stack(struct proc_maps_private *priv,
247 struct vm_area_struct *vma, bool is_pid)
249 struct inode *inode = priv->inode;
250 struct task_struct *task;
254 task = pid_task(proc_pid(inode), PIDTYPE_PID);
256 task = task_of_stack(task, vma, is_pid);
258 ret = task_pid_nr_ns(task, inode->i_sb->s_fs_info);
266 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
268 struct mm_struct *mm = vma->vm_mm;
269 struct file *file = vma->vm_file;
270 struct proc_maps_private *priv = m->private;
271 vm_flags_t flags = vma->vm_flags;
272 unsigned long ino = 0;
273 unsigned long long pgoff = 0;
274 unsigned long start, end;
276 const char *name = NULL;
279 struct inode *inode = file_inode(vma->vm_file);
280 dev = inode->i_sb->s_dev;
282 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
285 /* We don't show the stack guard page in /proc/maps */
286 start = vma->vm_start;
287 if (stack_guard_page_start(vma, start))
290 if (stack_guard_page_end(vma, end))
293 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
294 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
297 flags & VM_READ ? 'r' : '-',
298 flags & VM_WRITE ? 'w' : '-',
299 flags & VM_EXEC ? 'x' : '-',
300 flags & VM_MAYSHARE ? 's' : 'p',
302 MAJOR(dev), MINOR(dev), ino);
305 * Print the dentry name for named mappings, and a
306 * special [heap] marker for the heap:
310 seq_path(m, &file->f_path, "\n");
314 if (vma->vm_ops && vma->vm_ops->name) {
315 name = vma->vm_ops->name(vma);
320 name = arch_vma_name(vma);
329 if (vma->vm_start <= mm->brk &&
330 vma->vm_end >= mm->start_brk) {
335 tid = pid_of_stack(priv, vma, is_pid);
338 * Thread stack in /proc/PID/task/TID/maps or
339 * the main process stack.
341 if (!is_pid || (vma->vm_start <= mm->start_stack &&
342 vma->vm_end >= mm->start_stack)) {
345 /* Thread stack in /proc/PID/maps */
347 seq_printf(m, "[stack:%d]", tid);
360 static int show_map(struct seq_file *m, void *v, int is_pid)
362 show_map_vma(m, v, is_pid);
367 static int show_pid_map(struct seq_file *m, void *v)
369 return show_map(m, v, 1);
372 static int show_tid_map(struct seq_file *m, void *v)
374 return show_map(m, v, 0);
377 static const struct seq_operations proc_pid_maps_op = {
384 static const struct seq_operations proc_tid_maps_op = {
391 static int pid_maps_open(struct inode *inode, struct file *file)
393 return do_maps_open(inode, file, &proc_pid_maps_op);
396 static int tid_maps_open(struct inode *inode, struct file *file)
398 return do_maps_open(inode, file, &proc_tid_maps_op);
401 const struct file_operations proc_pid_maps_operations = {
402 .open = pid_maps_open,
405 .release = proc_map_release,
408 const struct file_operations proc_tid_maps_operations = {
409 .open = tid_maps_open,
412 .release = proc_map_release,
416 * Proportional Set Size(PSS): my share of RSS.
418 * PSS of a process is the count of pages it has in memory, where each
419 * page is divided by the number of processes sharing it. So if a
420 * process has 1000 pages all to itself, and 1000 shared with one other
421 * process, its PSS will be 1500.
423 * To keep (accumulated) division errors low, we adopt a 64bit
424 * fixed-point pss counter to minimize division errors. So (pss >>
425 * PSS_SHIFT) would be the real byte count.
427 * A shift of 12 before division means (assuming 4K page size):
428 * - 1M 3-user-pages add up to 8KB errors;
429 * - supports mapcount up to 2^24, or 16M;
430 * - supports PSS up to 2^52 bytes, or 4PB.
434 #ifdef CONFIG_PROC_PAGE_MONITOR
435 struct mem_size_stats {
436 struct vm_area_struct *vma;
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 anonymous_thp;
446 unsigned long nonlinear;
450 static void smaps_account(struct mem_size_stats *mss, struct page *page,
451 unsigned long size, bool young, bool dirty)
456 mss->anonymous += size;
458 mss->resident += size;
459 /* Accumulate the size in pages that have been accessed. */
460 if (young || PageReferenced(page))
461 mss->referenced += size;
462 mapcount = page_mapcount(page);
466 if (dirty || PageDirty(page))
467 mss->shared_dirty += size;
469 mss->shared_clean += size;
470 pss_delta = (u64)size << PSS_SHIFT;
471 do_div(pss_delta, mapcount);
472 mss->pss += pss_delta;
474 if (dirty || PageDirty(page))
475 mss->private_dirty += size;
477 mss->private_clean += size;
478 mss->pss += (u64)size << PSS_SHIFT;
482 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
483 struct mm_walk *walk)
485 struct mem_size_stats *mss = walk->private;
486 struct vm_area_struct *vma = mss->vma;
487 pgoff_t pgoff = linear_page_index(vma, addr);
488 struct page *page = NULL;
490 if (pte_present(*pte)) {
491 page = vm_normal_page(vma, addr, *pte);
492 } else if (is_swap_pte(*pte)) {
493 swp_entry_t swpent = pte_to_swp_entry(*pte);
495 if (!non_swap_entry(swpent))
496 mss->swap += PAGE_SIZE;
497 else if (is_migration_entry(swpent))
498 page = migration_entry_to_page(swpent);
499 } else if (pte_file(*pte)) {
500 if (pte_to_pgoff(*pte) != pgoff)
501 mss->nonlinear += PAGE_SIZE;
507 if (page->index != pgoff)
508 mss->nonlinear += PAGE_SIZE;
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 = mss->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 mem_size_stats *mss = walk->private;
540 struct vm_area_struct *vma = mss->vma;
544 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
545 smaps_pmd_entry(pmd, addr, walk);
550 if (pmd_trans_unstable(pmd))
553 * The mmap_sem held all the way back in m_start() is what
554 * keeps khugepaged out of here and from collapsing things
557 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
558 for (; addr != end; pte++, addr += PAGE_SIZE)
559 smaps_pte_entry(pte, addr, walk);
560 pte_unmap_unlock(pte - 1, ptl);
565 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
568 * Don't forget to update Documentation/ on changes.
570 static const char mnemonics[BITS_PER_LONG][2] = {
572 * In case if we meet a flag we don't know about.
574 [0 ... (BITS_PER_LONG-1)] = "??",
576 [ilog2(VM_READ)] = "rd",
577 [ilog2(VM_WRITE)] = "wr",
578 [ilog2(VM_EXEC)] = "ex",
579 [ilog2(VM_SHARED)] = "sh",
580 [ilog2(VM_MAYREAD)] = "mr",
581 [ilog2(VM_MAYWRITE)] = "mw",
582 [ilog2(VM_MAYEXEC)] = "me",
583 [ilog2(VM_MAYSHARE)] = "ms",
584 [ilog2(VM_GROWSDOWN)] = "gd",
585 [ilog2(VM_PFNMAP)] = "pf",
586 [ilog2(VM_DENYWRITE)] = "dw",
587 [ilog2(VM_LOCKED)] = "lo",
588 [ilog2(VM_IO)] = "io",
589 [ilog2(VM_SEQ_READ)] = "sr",
590 [ilog2(VM_RAND_READ)] = "rr",
591 [ilog2(VM_DONTCOPY)] = "dc",
592 [ilog2(VM_DONTEXPAND)] = "de",
593 [ilog2(VM_ACCOUNT)] = "ac",
594 [ilog2(VM_NORESERVE)] = "nr",
595 [ilog2(VM_HUGETLB)] = "ht",
596 [ilog2(VM_NONLINEAR)] = "nl",
597 [ilog2(VM_ARCH_1)] = "ar",
598 [ilog2(VM_DONTDUMP)] = "dd",
599 #ifdef CONFIG_MEM_SOFT_DIRTY
600 [ilog2(VM_SOFTDIRTY)] = "sd",
602 [ilog2(VM_MIXEDMAP)] = "mm",
603 [ilog2(VM_HUGEPAGE)] = "hg",
604 [ilog2(VM_NOHUGEPAGE)] = "nh",
605 [ilog2(VM_MERGEABLE)] = "mg",
609 seq_puts(m, "VmFlags: ");
610 for (i = 0; i < BITS_PER_LONG; i++) {
611 if (vma->vm_flags & (1UL << i)) {
612 seq_printf(m, "%c%c ",
613 mnemonics[i][0], mnemonics[i][1]);
619 static int show_smap(struct seq_file *m, void *v, int is_pid)
621 struct vm_area_struct *vma = v;
622 struct mem_size_stats mss;
623 struct mm_walk smaps_walk = {
624 .pmd_entry = smaps_pte_range,
629 memset(&mss, 0, sizeof mss);
631 /* mmap_sem is held in m_start */
632 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
633 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
635 show_map_vma(m, vma, is_pid);
641 "Shared_Clean: %8lu kB\n"
642 "Shared_Dirty: %8lu kB\n"
643 "Private_Clean: %8lu kB\n"
644 "Private_Dirty: %8lu kB\n"
645 "Referenced: %8lu kB\n"
646 "Anonymous: %8lu kB\n"
647 "AnonHugePages: %8lu kB\n"
649 "KernelPageSize: %8lu kB\n"
650 "MMUPageSize: %8lu kB\n"
652 (vma->vm_end - vma->vm_start) >> 10,
654 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
655 mss.shared_clean >> 10,
656 mss.shared_dirty >> 10,
657 mss.private_clean >> 10,
658 mss.private_dirty >> 10,
659 mss.referenced >> 10,
661 mss.anonymous_thp >> 10,
663 vma_kernel_pagesize(vma) >> 10,
664 vma_mmu_pagesize(vma) >> 10,
665 (vma->vm_flags & VM_LOCKED) ?
666 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
668 if (vma->vm_flags & VM_NONLINEAR)
669 seq_printf(m, "Nonlinear: %8lu kB\n",
670 mss.nonlinear >> 10);
672 show_smap_vma_flags(m, vma);
677 static int show_pid_smap(struct seq_file *m, void *v)
679 return show_smap(m, v, 1);
682 static int show_tid_smap(struct seq_file *m, void *v)
684 return show_smap(m, v, 0);
687 static const struct seq_operations proc_pid_smaps_op = {
691 .show = show_pid_smap
694 static const struct seq_operations proc_tid_smaps_op = {
698 .show = show_tid_smap
701 static int pid_smaps_open(struct inode *inode, struct file *file)
703 return do_maps_open(inode, file, &proc_pid_smaps_op);
706 static int tid_smaps_open(struct inode *inode, struct file *file)
708 return do_maps_open(inode, file, &proc_tid_smaps_op);
711 const struct file_operations proc_pid_smaps_operations = {
712 .open = pid_smaps_open,
715 .release = proc_map_release,
718 const struct file_operations proc_tid_smaps_operations = {
719 .open = tid_smaps_open,
722 .release = proc_map_release,
726 * We do not want to have constant page-shift bits sitting in
727 * pagemap entries and are about to reuse them some time soon.
729 * Here's the "migration strategy":
730 * 1. when the system boots these bits remain what they are,
731 * but a warning about future change is printed in log;
732 * 2. once anyone clears soft-dirty bits via clear_refs file,
733 * these flag is set to denote, that user is aware of the
734 * new API and those page-shift bits change their meaning.
735 * The respective warning is printed in dmesg;
736 * 3. In a couple of releases we will remove all the mentions
737 * of page-shift in pagemap entries.
740 static bool soft_dirty_cleared __read_mostly;
742 enum clear_refs_types {
746 CLEAR_REFS_SOFT_DIRTY,
750 struct clear_refs_private {
751 struct vm_area_struct *vma;
752 enum clear_refs_types type;
755 static inline void clear_soft_dirty(struct vm_area_struct *vma,
756 unsigned long addr, pte_t *pte)
758 #ifdef CONFIG_MEM_SOFT_DIRTY
760 * The soft-dirty tracker uses #PF-s to catch writes
761 * to pages, so write-protect the pte as well. See the
762 * Documentation/vm/soft-dirty.txt for full description
763 * of how soft-dirty works.
767 if (pte_present(ptent)) {
768 ptent = pte_wrprotect(ptent);
769 ptent = pte_clear_flags(ptent, _PAGE_SOFT_DIRTY);
770 } else if (is_swap_pte(ptent)) {
771 ptent = pte_swp_clear_soft_dirty(ptent);
772 } else if (pte_file(ptent)) {
773 ptent = pte_file_clear_soft_dirty(ptent);
776 set_pte_at(vma->vm_mm, addr, pte, ptent);
780 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
781 unsigned long end, struct mm_walk *walk)
783 struct clear_refs_private *cp = walk->private;
784 struct vm_area_struct *vma = cp->vma;
789 split_huge_page_pmd(vma, addr, pmd);
790 if (pmd_trans_unstable(pmd))
793 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
794 for (; addr != end; pte++, addr += PAGE_SIZE) {
797 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
798 clear_soft_dirty(vma, addr, pte);
802 if (!pte_present(ptent))
805 page = vm_normal_page(vma, addr, ptent);
809 /* Clear accessed and referenced bits. */
810 ptep_test_and_clear_young(vma, addr, pte);
811 ClearPageReferenced(page);
813 pte_unmap_unlock(pte - 1, ptl);
818 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
819 size_t count, loff_t *ppos)
821 struct task_struct *task;
822 char buffer[PROC_NUMBUF];
823 struct mm_struct *mm;
824 struct vm_area_struct *vma;
825 enum clear_refs_types type;
829 memset(buffer, 0, sizeof(buffer));
830 if (count > sizeof(buffer) - 1)
831 count = sizeof(buffer) - 1;
832 if (copy_from_user(buffer, buf, count))
834 rv = kstrtoint(strstrip(buffer), 10, &itype);
837 type = (enum clear_refs_types)itype;
838 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
841 if (type == CLEAR_REFS_SOFT_DIRTY) {
842 soft_dirty_cleared = true;
843 pr_warn_once("The pagemap bits 55-60 has changed their meaning!"
844 " See the linux/Documentation/vm/pagemap.txt for "
848 task = get_proc_task(file_inode(file));
851 mm = get_task_mm(task);
853 struct clear_refs_private cp = {
856 struct mm_walk clear_refs_walk = {
857 .pmd_entry = clear_refs_pte_range,
861 down_read(&mm->mmap_sem);
862 if (type == CLEAR_REFS_SOFT_DIRTY) {
863 for (vma = mm->mmap; vma; vma = vma->vm_next) {
864 if (!(vma->vm_flags & VM_SOFTDIRTY))
866 up_read(&mm->mmap_sem);
867 down_write(&mm->mmap_sem);
868 for (vma = mm->mmap; vma; vma = vma->vm_next) {
869 vma->vm_flags &= ~VM_SOFTDIRTY;
870 vma_set_page_prot(vma);
872 downgrade_write(&mm->mmap_sem);
875 mmu_notifier_invalidate_range_start(mm, 0, -1);
877 for (vma = mm->mmap; vma; vma = vma->vm_next) {
879 if (is_vm_hugetlb_page(vma))
882 * Writing 1 to /proc/pid/clear_refs affects all pages.
884 * Writing 2 to /proc/pid/clear_refs only affects
887 * Writing 3 to /proc/pid/clear_refs only affects file
890 * Writing 4 to /proc/pid/clear_refs affects all pages.
892 if (type == CLEAR_REFS_ANON && vma->vm_file)
894 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
896 walk_page_range(vma->vm_start, vma->vm_end,
899 if (type == CLEAR_REFS_SOFT_DIRTY)
900 mmu_notifier_invalidate_range_end(mm, 0, -1);
902 up_read(&mm->mmap_sem);
905 put_task_struct(task);
910 const struct file_operations proc_clear_refs_operations = {
911 .write = clear_refs_write,
912 .llseek = noop_llseek,
920 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
921 pagemap_entry_t *buffer;
925 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
926 #define PAGEMAP_WALK_MASK (PMD_MASK)
928 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
929 #define PM_STATUS_BITS 3
930 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
931 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
932 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
933 #define PM_PSHIFT_BITS 6
934 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
935 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
936 #define __PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
937 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
938 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
939 /* in "new" pagemap pshift bits are occupied with more status bits */
940 #define PM_STATUS2(v2, x) (__PM_PSHIFT(v2 ? x : PAGE_SHIFT))
942 #define __PM_SOFT_DIRTY (1LL)
943 #define PM_PRESENT PM_STATUS(4LL)
944 #define PM_SWAP PM_STATUS(2LL)
945 #define PM_FILE PM_STATUS(1LL)
946 #define PM_NOT_PRESENT(v2) PM_STATUS2(v2, 0)
947 #define PM_END_OF_BUFFER 1
949 static inline pagemap_entry_t make_pme(u64 val)
951 return (pagemap_entry_t) { .pme = val };
954 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
955 struct pagemapread *pm)
957 pm->buffer[pm->pos++] = *pme;
958 if (pm->pos >= pm->len)
959 return PM_END_OF_BUFFER;
963 static int pagemap_pte_hole(unsigned long start, unsigned long end,
964 struct mm_walk *walk)
966 struct pagemapread *pm = walk->private;
967 unsigned long addr = start;
971 struct vm_area_struct *vma = find_vma(walk->mm, addr);
972 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2));
973 /* End of address space hole, which we mark as non-present. */
974 unsigned long hole_end;
977 hole_end = min(end, vma->vm_start);
981 for (; addr < hole_end; addr += PAGE_SIZE) {
982 err = add_to_pagemap(addr, &pme, pm);
990 /* Addresses in the VMA. */
991 if (vma->vm_flags & VM_SOFTDIRTY)
992 pme.pme |= PM_STATUS2(pm->v2, __PM_SOFT_DIRTY);
993 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
994 err = add_to_pagemap(addr, &pme, pm);
1003 static void pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1004 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1007 struct page *page = NULL;
1010 if (pte_present(pte)) {
1011 frame = pte_pfn(pte);
1013 page = vm_normal_page(vma, addr, pte);
1014 if (pte_soft_dirty(pte))
1015 flags2 |= __PM_SOFT_DIRTY;
1016 } else if (is_swap_pte(pte)) {
1018 if (pte_swp_soft_dirty(pte))
1019 flags2 |= __PM_SOFT_DIRTY;
1020 entry = pte_to_swp_entry(pte);
1021 frame = swp_type(entry) |
1022 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1024 if (is_migration_entry(entry))
1025 page = migration_entry_to_page(entry);
1027 if (vma->vm_flags & VM_SOFTDIRTY)
1028 flags2 |= __PM_SOFT_DIRTY;
1029 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2));
1033 if (page && !PageAnon(page))
1035 if ((vma->vm_flags & VM_SOFTDIRTY))
1036 flags2 |= __PM_SOFT_DIRTY;
1038 *pme = make_pme(PM_PFRAME(frame) | PM_STATUS2(pm->v2, flags2) | flags);
1041 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1042 static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1043 pmd_t pmd, int offset, int pmd_flags2)
1046 * Currently pmd for thp is always present because thp can not be
1047 * swapped-out, migrated, or HWPOISONed (split in such cases instead.)
1048 * This if-check is just to prepare for future implementation.
1050 if (pmd_present(pmd))
1051 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset)
1052 | PM_STATUS2(pm->v2, pmd_flags2) | PM_PRESENT);
1054 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, pmd_flags2));
1057 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1058 pmd_t pmd, int offset, int pmd_flags2)
1063 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
1064 struct mm_walk *walk)
1066 struct vm_area_struct *vma;
1067 struct pagemapread *pm = walk->private;
1069 pte_t *pte, *orig_pte;
1072 /* find the first VMA at or above 'addr' */
1073 vma = find_vma(walk->mm, addr);
1074 if (vma && pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1077 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(*pmd))
1078 pmd_flags2 = __PM_SOFT_DIRTY;
1082 for (; addr != end; addr += PAGE_SIZE) {
1083 unsigned long offset;
1084 pagemap_entry_t pme;
1086 offset = (addr & ~PAGEMAP_WALK_MASK) >>
1088 thp_pmd_to_pagemap_entry(&pme, pm, *pmd, offset, pmd_flags2);
1089 err = add_to_pagemap(addr, &pme, pm);
1097 if (pmd_trans_unstable(pmd))
1101 /* End of address space hole, which we mark as non-present. */
1102 unsigned long hole_end;
1105 hole_end = min(end, vma->vm_start);
1109 for (; addr < hole_end; addr += PAGE_SIZE) {
1110 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2));
1112 err = add_to_pagemap(addr, &pme, pm);
1117 if (!vma || vma->vm_start >= end)
1120 * We can't possibly be in a hugetlb VMA. In general,
1121 * for a mm_walk with a pmd_entry and a hugetlb_entry,
1122 * the pmd_entry can only be called on addresses in a
1123 * hugetlb if the walk starts in a non-hugetlb VMA and
1124 * spans a hugepage VMA. Since pagemap_read walks are
1125 * PMD-sized and PMD-aligned, this will never be true.
1127 BUG_ON(is_vm_hugetlb_page(vma));
1129 /* Addresses in the VMA. */
1130 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1131 for (; addr < min(end, vma->vm_end); pte++, addr += PAGE_SIZE) {
1132 pagemap_entry_t pme;
1134 pte_to_pagemap_entry(&pme, pm, vma, addr, *pte);
1135 err = add_to_pagemap(addr, &pme, pm);
1139 pte_unmap_unlock(orig_pte, ptl);
1147 vma = find_vma(walk->mm, addr);
1155 #ifdef CONFIG_HUGETLB_PAGE
1156 static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1157 pte_t pte, int offset, int flags2)
1159 if (pte_present(pte))
1160 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset) |
1161 PM_STATUS2(pm->v2, flags2) |
1164 *pme = make_pme(PM_NOT_PRESENT(pm->v2) |
1165 PM_STATUS2(pm->v2, flags2));
1168 /* This function walks within one hugetlb entry in the single call */
1169 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
1170 unsigned long addr, unsigned long end,
1171 struct mm_walk *walk)
1173 struct pagemapread *pm = walk->private;
1174 struct vm_area_struct *vma;
1177 pagemap_entry_t pme;
1179 vma = find_vma(walk->mm, addr);
1182 if (vma && (vma->vm_flags & VM_SOFTDIRTY))
1183 flags2 = __PM_SOFT_DIRTY;
1187 for (; addr != end; addr += PAGE_SIZE) {
1188 int offset = (addr & ~hmask) >> PAGE_SHIFT;
1189 huge_pte_to_pagemap_entry(&pme, pm, *pte, offset, flags2);
1190 err = add_to_pagemap(addr, &pme, pm);
1199 #endif /* HUGETLB_PAGE */
1202 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1204 * For each page in the address space, this file contains one 64-bit entry
1205 * consisting of the following:
1207 * Bits 0-54 page frame number (PFN) if present
1208 * Bits 0-4 swap type if swapped
1209 * Bits 5-54 swap offset if swapped
1210 * Bits 55-60 page shift (page size = 1<<page shift)
1211 * Bit 61 page is file-page or shared-anon
1212 * Bit 62 page swapped
1213 * Bit 63 page present
1215 * If the page is not present but in swap, then the PFN contains an
1216 * encoding of the swap file number and the page's offset into the
1217 * swap. Unmapped pages return a null PFN. This allows determining
1218 * precisely which pages are mapped (or in swap) and comparing mapped
1219 * pages between processes.
1221 * Efficient users of this interface will use /proc/pid/maps to
1222 * determine which areas of memory are actually mapped and llseek to
1223 * skip over unmapped regions.
1225 static ssize_t pagemap_read(struct file *file, char __user *buf,
1226 size_t count, loff_t *ppos)
1228 struct task_struct *task = get_proc_task(file_inode(file));
1229 struct mm_struct *mm;
1230 struct pagemapread pm;
1232 struct mm_walk pagemap_walk = {};
1234 unsigned long svpfn;
1235 unsigned long start_vaddr;
1236 unsigned long end_vaddr;
1243 /* file position must be aligned */
1244 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1251 pm.v2 = soft_dirty_cleared;
1252 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1253 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1258 mm = mm_access(task, PTRACE_MODE_READ);
1260 if (!mm || IS_ERR(mm))
1263 pagemap_walk.pmd_entry = pagemap_pte_range;
1264 pagemap_walk.pte_hole = pagemap_pte_hole;
1265 #ifdef CONFIG_HUGETLB_PAGE
1266 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1268 pagemap_walk.mm = mm;
1269 pagemap_walk.private = ±
1272 svpfn = src / PM_ENTRY_BYTES;
1273 start_vaddr = svpfn << PAGE_SHIFT;
1274 end_vaddr = TASK_SIZE_OF(task);
1276 /* watch out for wraparound */
1277 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
1278 start_vaddr = end_vaddr;
1281 * The odds are that this will stop walking way
1282 * before end_vaddr, because the length of the
1283 * user buffer is tracked in "pm", and the walk
1284 * will stop when we hit the end of the buffer.
1287 while (count && (start_vaddr < end_vaddr)) {
1292 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1294 if (end < start_vaddr || end > end_vaddr)
1296 down_read(&mm->mmap_sem);
1297 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1298 up_read(&mm->mmap_sem);
1301 len = min(count, PM_ENTRY_BYTES * pm.pos);
1302 if (copy_to_user(buf, pm.buffer, len)) {
1311 if (!ret || ret == PM_END_OF_BUFFER)
1319 put_task_struct(task);
1324 static int pagemap_open(struct inode *inode, struct file *file)
1326 /* do not disclose physical addresses: attack vector */
1327 if (!capable(CAP_SYS_ADMIN))
1329 pr_warn_once("Bits 55-60 of /proc/PID/pagemap entries are about "
1330 "to stop being page-shift some time soon. See the "
1331 "linux/Documentation/vm/pagemap.txt for details.\n");
1335 const struct file_operations proc_pagemap_operations = {
1336 .llseek = mem_lseek, /* borrow this */
1337 .read = pagemap_read,
1338 .open = pagemap_open,
1340 #endif /* CONFIG_PROC_PAGE_MONITOR */
1345 struct vm_area_struct *vma;
1346 unsigned long pages;
1348 unsigned long active;
1349 unsigned long writeback;
1350 unsigned long mapcount_max;
1351 unsigned long dirty;
1352 unsigned long swapcache;
1353 unsigned long node[MAX_NUMNODES];
1356 struct numa_maps_private {
1357 struct proc_maps_private proc_maps;
1358 struct numa_maps md;
1361 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1362 unsigned long nr_pages)
1364 int count = page_mapcount(page);
1366 md->pages += nr_pages;
1367 if (pte_dirty || PageDirty(page))
1368 md->dirty += nr_pages;
1370 if (PageSwapCache(page))
1371 md->swapcache += nr_pages;
1373 if (PageActive(page) || PageUnevictable(page))
1374 md->active += nr_pages;
1376 if (PageWriteback(page))
1377 md->writeback += nr_pages;
1380 md->anon += nr_pages;
1382 if (count > md->mapcount_max)
1383 md->mapcount_max = count;
1385 md->node[page_to_nid(page)] += nr_pages;
1388 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1394 if (!pte_present(pte))
1397 page = vm_normal_page(vma, addr, pte);
1401 if (PageReserved(page))
1404 nid = page_to_nid(page);
1405 if (!node_isset(nid, node_states[N_MEMORY]))
1411 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1412 unsigned long end, struct mm_walk *walk)
1414 struct numa_maps *md;
1421 if (pmd_trans_huge_lock(pmd, md->vma, &ptl) == 1) {
1422 pte_t huge_pte = *(pte_t *)pmd;
1425 page = can_gather_numa_stats(huge_pte, md->vma, addr);
1427 gather_stats(page, md, pte_dirty(huge_pte),
1428 HPAGE_PMD_SIZE/PAGE_SIZE);
1433 if (pmd_trans_unstable(pmd))
1435 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1437 struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
1440 gather_stats(page, md, pte_dirty(*pte), 1);
1442 } while (pte++, addr += PAGE_SIZE, addr != end);
1443 pte_unmap_unlock(orig_pte, ptl);
1446 #ifdef CONFIG_HUGETLB_PAGE
1447 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1448 unsigned long addr, unsigned long end, struct mm_walk *walk)
1450 struct numa_maps *md;
1453 if (!pte_present(*pte))
1456 page = pte_page(*pte);
1461 gather_stats(page, md, pte_dirty(*pte), 1);
1466 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1467 unsigned long addr, unsigned long end, struct mm_walk *walk)
1474 * Display pages allocated per node and memory policy via /proc.
1476 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1478 struct numa_maps_private *numa_priv = m->private;
1479 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1480 struct vm_area_struct *vma = v;
1481 struct numa_maps *md = &numa_priv->md;
1482 struct file *file = vma->vm_file;
1483 struct mm_struct *mm = vma->vm_mm;
1484 struct mm_walk walk = {};
1485 struct mempolicy *pol;
1492 /* Ensure we start with an empty set of numa_maps statistics. */
1493 memset(md, 0, sizeof(*md));
1497 walk.hugetlb_entry = gather_hugetbl_stats;
1498 walk.pmd_entry = gather_pte_stats;
1502 pol = __get_vma_policy(vma, vma->vm_start);
1504 mpol_to_str(buffer, sizeof(buffer), pol);
1507 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1510 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1513 seq_puts(m, " file=");
1514 seq_path(m, &file->f_path, "\n\t= ");
1515 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1516 seq_puts(m, " heap");
1518 pid_t tid = pid_of_stack(proc_priv, vma, is_pid);
1521 * Thread stack in /proc/PID/task/TID/maps or
1522 * the main process stack.
1524 if (!is_pid || (vma->vm_start <= mm->start_stack &&
1525 vma->vm_end >= mm->start_stack))
1526 seq_puts(m, " stack");
1528 seq_printf(m, " stack:%d", tid);
1532 if (is_vm_hugetlb_page(vma))
1533 seq_puts(m, " huge");
1535 walk_page_range(vma->vm_start, vma->vm_end, &walk);
1541 seq_printf(m, " anon=%lu", md->anon);
1544 seq_printf(m, " dirty=%lu", md->dirty);
1546 if (md->pages != md->anon && md->pages != md->dirty)
1547 seq_printf(m, " mapped=%lu", md->pages);
1549 if (md->mapcount_max > 1)
1550 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1553 seq_printf(m, " swapcache=%lu", md->swapcache);
1555 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1556 seq_printf(m, " active=%lu", md->active);
1559 seq_printf(m, " writeback=%lu", md->writeback);
1561 for_each_node_state(nid, N_MEMORY)
1563 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1566 m_cache_vma(m, vma);
1570 static int show_pid_numa_map(struct seq_file *m, void *v)
1572 return show_numa_map(m, v, 1);
1575 static int show_tid_numa_map(struct seq_file *m, void *v)
1577 return show_numa_map(m, v, 0);
1580 static const struct seq_operations proc_pid_numa_maps_op = {
1584 .show = show_pid_numa_map,
1587 static const struct seq_operations proc_tid_numa_maps_op = {
1591 .show = show_tid_numa_map,
1594 static int numa_maps_open(struct inode *inode, struct file *file,
1595 const struct seq_operations *ops)
1597 return proc_maps_open(inode, file, ops,
1598 sizeof(struct numa_maps_private));
1601 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1603 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1606 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1608 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1611 const struct file_operations proc_pid_numa_maps_operations = {
1612 .open = pid_numa_maps_open,
1614 .llseek = seq_lseek,
1615 .release = proc_map_release,
1618 const struct file_operations proc_tid_numa_maps_operations = {
1619 .open = tid_numa_maps_open,
1621 .llseek = seq_lseek,
1622 .release = proc_map_release,
1624 #endif /* CONFIG_NUMA */